australian social policy journal no. 9 2010

Improving the lives of Australians

Australian Social Policy Journal No. 9 2010

The Australian Social Policy Journal publishes current research and analysis on a broad range of issues topical to Australia’s social policy and its administration. Regular features include major articles, social policy notes and book reviews.

Content is compiled by the Research and Analysis Branch of the Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA). Australian Social Policy Journal supercedes Australian Social Policy, published by FaHCSIA, and the Social Security Journal published by the former Department of Social Security.

Refereed publication

Australian Social Policy Journal is a fully refereed academic journal; all submissions of major articles and social policy notes to the journal are subject to an external blind peer review. The journal is recognised by the Australian Research Council’s (ARC) Excellence in Research for Australia (ERA) Ranked Journal List of refereed journals.

Submissions

Submissions are accepted from academic researchers, government employees and relevant practitioners. Submissions that contribute to current social policy research issues and debates are particularly encouraged. Submissions can be forwarded by email to <[email protected]>. Submission guidelines are available at the back of the journal or online at <www.fahcsia.gov.au/research>.

Copyright

This work is copyright. Apart from any use as permitted under the Copyright Act 1968, no part may be reproduced by any process without prior written permission from the Commonwealth available from the Commonwealth Copyright Administration, Attorney-General’s Department. Requests and inquiries concerning reproduction and rights should be addressed to the Commonwealth Copyright Administration, Attorney-General’s Department, Robert Garran Offices, National Circuit, Barton, ACT 2600 or posted at <http://www.ag.gov.au/ccca>.

Disclaimer

The opinions, comments and/or analysis expressed in this document are those of the authors and do not necessarily represent the views of the Minister for Families, Housing, Community Services and Indigenous Affairs, or the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs, and cannot be taken in any way as expressions of Government policy.

For more information on FaHCSIA research publications and subscription, please contact:

Research Publications UnitResearch and Analysis BranchAustralian Government Department of Families, Housing, Community Services and Indigenous AffairsPO Box 7576Canberra Business Centre ACT 2610

Phone: (02) 6244 5458 © Commonwealth of Australia 2010Fax: (02) 6133 8387 ISSN 1442-6331 (PRINT)Email: <[email protected]> ISBN 978-1-921647-37-9Web: <www.fahcsia.gov.au/research>

Australian Social Policy Journal No. 9

Major articles

Deborah A Cobb-Clark & Vincent A HildebrandThe asset portfolios of older Australian households

Jason D Brandrup & Paula L ManceChanges in household expenditure associated with the arrival of newborn children

Liana Leach, Peter Butterworth, Bryan Rodgers and Lyndall StrazdinsDeriving an evidence-based measure of job quality from the HILDA survey

Peng YuSequence matters: understanding the relationship between parental income support receipt and child mortality

Samara McPhedranRegional living and community participation: are people with disability at a disadvantage?

Social policy note

Ibolya Losoncz and Benjamin GrahamWork–life tension and its impact on the workforce participation of Australian mothers

Book review

Utopias and revolutions Gornick, J & Meyers, M (eds), Gender equality: transforming family divisions of labor and Epsing-Anderson, G, The incomplete revolution: adapting welfare states to women’s new roles (Reviewer: Julie Connolly)

AuStrAlIAN SocIAl PolIcy JourNAl No. 9

iv

v

Contents

Major articles

Deborah A Cobb-Clark & Vincent A HildebrandThe asset portfolios of older Australian households 1

Jason D Brandrup & Paula L ManceChanges in household expenditure associated with the arrival of newborn children 41

Liana Leach, Peter Butterworth, Bryan Rodgers and Lyndall StrazdinsDeriving an evidence-based measure of job quality from the HILDA survey 67

Peng YuSequence matters: understanding the relationship between parental income support receipt and child mortality 87

Samara McPhedranRegional living and community participation: are people with disability at a disadvantage? 111

Social policy note

Ibolya Losoncz and Benjamin GrahamWork–life tension and its impact on the workforce participation of Australian mothers 139

Book review

Utopias and revolutions 159Gornick, J & Meyers, M (eds), Gender equality: transforming family divisions of labor and Epsing-Anderson, G, The incomplete revolution: adapting welfare states to women’s new roles (Reviewer: Julie Connolly)

Guidelines for contributors 163

Subscription form 167


vi

vii

Major articlesAustralian Social Policy Journal No. 9


viii

1

The asset portfolios of older Australian householdsDeborah A Cobb-Clark1 & Vincent A Hildebrand2

1 Economics Program, Research School of Social Sciences, The Australian National University and Institute for the Study of Labor (IZA) Bonn2 Department of Economics, Glendon College, York University and CEPS/INSTEAD, Luxembourg

This paper uses confidentialised unit record file data from the Household, Income and Labour Dynamics in Australia (HILDA) survey. The HILDA project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services, and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (MIAESR). The findings and views reported in this paper, however, are those of the authors and should not be attributed to FaHCSIA or MIAESR.

AbstractThis paper investigates whether there is evidence that households adjust their asset portfolios just prior to retirement in order to maximise their eligibility for a means-tested public pension. To this end, we take advantage of recently available, detailed micro data for a nationally-representative sample of Australian households to estimate a system of asset equations that are constrained to add up to net worth. Our results provide little evidence that in 2006 healthy households or couples were responding to the incentives embedded in the asset and income tests used to determine Australian Age Pension eligibility by reallocating their assets. While there are some significant differences in asset portfolios associated with having an income near the income threshold, being of pensionable age and being in poor health, these differences are often only marginally significant, are not robust across time, and are not clearly consistent with the incentives inherent in the Australian Age Pension eligibility rules. Any behavioral response to the incentives inherent in the Age Pension means test in 2006 appears to be predominately concentrated among single pensioners who are in poor health. In 2002 there is also evidence that healthy households above pension age held significantly more wealth in their homes than did otherwise similar younger households, perhaps suggesting some reduction in the incentives to reallocate assets over time.

Keywords: asset portfolios; means testing; public pension; household wealth


2

1 IntroductionEconomists are increasingly using detailed comparative studies of wealth levels across different groups as a way of gaining a deeper understanding of the processes through which households accumulate and manage their wealth (Cobb-Clark & Hildebrand 2006a, 2006b, 2009). As eligibility for the Age Pension is asset and income tested, policy makers are particularly interested in understanding how households allocate their wealth across asset types in the lead-up to retirement.

This report takes advantage of recently available Household, Income and Labour Dynamics in Australia (HILDA) data on household wealth in Australia to seek answers to two questions:

�� How do the portfolio choices of pre and post-retirement period households differ?

�� Are these differences consistent with households managing their wealth in a way that maximises access to the Age Pension?

Sections 2 and 3 briefly present some key features of the conceptual framework and the means tests underlying the Age Pension. Section 4 presents the data and discusses descriptive statistics. The empirical strategy and regression results are presented and discussed in Section 5, and concluding remarks are in Section 6.

2 Conceptual frameworkHouseholds make decisions about how to allocate their wealth across asset types by comparing the relative risks and relative returns of various assets. Government policy—such as tax policy, means testing or regulation—can influence those decisions by affecting both the risk and return associated with holding a particular asset. Certain assets, such as housing, typically require minimum investment, which implies that the ability to hold particular assets may depend, in part, on a household’s overall wealth level. It is therefore reasonable to expect that the mix of assets a household maintains will depend on its overall wealth level.

Economists typically rely on the lifecycle hypothesis to understand the way in which a household’s consumption and savings decisions—and ultimately wealth accumulation—evolve as that household ages. If households had perfect foresight and faced no credit constraints, they would be able to borrow and save so as to smooth their consumption levels across the lifecycle, despite often substantially fluctuating income. In reality, however, imperfect information and credit constraints imply that a household’s current situation (for example, stage of lifecycle, composition and financial situation) is likely to be important in understanding both its net worth and asset portfolio.

This conceptual framework is a useful tool with which to assess the potential effect of the Age Pension means test on the portfolio allocations of Australian households. In such an assessment it would be important to consider the incentives inherent in the means tests associated with the Age Pension. Moreover, we need to make comparisons between households that are at the same approximate lifecycle stage, that is, immediately before and after retirement age, and are equally wealthy.

the ASSet PortfolIoS of older AuStrAlIAN houSeholdS

3

3 Australian Age PensionThe Age Pension is the first tier program of the Australian pension system, which covered about 80 per cent of the elderly population in 2004–05. Eligibility for the Age Pension is contingent on a claimant being an Australian resident, residing in Australia at the time his or her claim is lodged, and being a resident for a total of at least 10 years, including five consecutive years.1 Male claimants become eligible at age 65 years while female claimants’ eligibility is being gradually increased from age 60 years in 1995 to age 65 years by 2014.2

In June 2006, single recipients of the Age Pension received $499 every two weeks. Partnered recipients received $834 combined.3 The benefit level for individuals is explicitly set at 25 per cent of gross male average total earnings. Benefit levels (and means test thresholds) are adjusted every six months in line with changes in the consumer price index or average male earnings—whichever is greater.

Age Pension coverage is not universal. Simply meeting age and residency requirements does not guarantee benefit payments. The Age Pension targets the poorest elderly by subjecting the level of benefit payments to a broad means test based on income and assets. The level of pension benefits is determined by the test that results in the lowest payment, making (in this context) the arbitrage between the optimal levels of income and assets very complex. Furthermore, age pensioners also receive subsidies for health care, pharmaceuticals, public transport, utilities and rent. As a result, a real incentive exists at the margin to qualify for a small pension in order to take advantage of the various additional, lump-sum benefits derived from these subsidies.

In 2006, the income test implied that couples experienced a reduction in pension benefit payments of 40 cents in every dollar earned in excess of $228 per fortnight; for singles it was 20 cents in every dollar earned in excess of $128 per fortnight. Relevant sources of income include all incomes received, derived and/or earned. The most common sources of income are salaries and wages; the monetary value of non-income benefits; annuities and pensions, including superannuation and overseas pensions; real estate, estates and life interests; profits and distributions from private trusts and businesses; and deemed income from financial investments.

In deeming income from financial investments it is assumed that these investments are earning a specific, fixed rate of interest, regardless of the rate they are actually earning. The first $63,800 of financial investments for a couple or the first $38,400 of financial investments for a single person is deemed to be earning a rate of 3 per cent. Remaining financial investments are deemed to be earning 5 per cent. This particular aspect of the income test may give households an incentive to reallocate their financial wealth towards more volatile (riskier) financial assets that are expected to yield returns exceeding the deemed rate set by government rather than safer financial assets that yield returns lower than the deemed rate. Thus, it is reasonable to expect that the way in which households hold their financial wealth may be affected by the deeming rules. It is less clear how the deeming rules might affect the incentives to hold financial wealth in general.

Home ownership status is central to the asset test. However, a claimant’s principal place of residence is exempt from the asset test.4 As a result, the asset test is a function of home owner


4

status, but is independent of the value of the principal residence. More precisely, in 2006, single home owners could have up to $161,500 in assets and non-home owners could have up to $278,500 before their pension rate would be affected. Couples could have up to $229,000 if they owned their home and up to $346,000 if they did not. Assets exceeding these exemption amounts reduced pension rates by $3 per fortnight for every $1,000.5

Given this favourable treatment of housing, one might expect households to allocate more wealth towards their principal residence upon reaching pension age. This might imply increasing equity in the family home by decreasing some or all other types of assets. In particular, there may be an incentive to decrease holdings in risky (often liquid) assets with high yields. In the context of the Age Pension, the benefits associated with reducing positions in these assets are twofold. First, it might help households achieve a level of assets below the exemption amount set by the asset test. Second, it would decrease the amount of income received from their financial wealth, which factors into the income test calculation. The resulting combined effect would be to increase the probability that a claimant qualifies for the Age Pension under both rules. It is important to note that claimants only qualify for the lowest benefits level as determined by either the income test or the asset test. As a result, households could be expected to decrease holdings in riskier assets, such as stocks and mutual funds, and to increase assets in their principal residence.

More generally, households qualifying for Age Pension benefits under the income test may have an incentive to shift investments towards either less risky, non-financial assets with low returns or towards assets that do not generate additional income (such as cars, recreational vehicles and so on), hence reducing the amount of income subjected to the income test. In such cases, an increase in lifestyle assets, such as holiday homes or recreational vehicles might be observed.

Finally, if shifting assets into the family home is not possible some incentives remain to reduce overall wealth by simply purchasing expensive consumer goods; for example, cashing out superannuation to finance expensive holidays.

It seems clear that targeting Age Pension benefits towards poorer households creates incentives for Australian households to reallocate their portfolios in order to maximise the likelihood of qualifying for Age Pension benefits under the combined income and asset tests. Using cross-sectional data on the wealth levels and asset portfolios of Australian households in 2002 and 2006, we attempt to determine if there is evidence of behavioural response consistent with the various possible scenarios discussed earlier.

4 Data

the hIldA survey

The data used in this paper come from the Household, Income and Labour Dynamics in Australia (HILDA) survey, which is a longitudinal survey of Australian households encompassing approximately 13,000 individual respondents living in more than 7,000 households. The analysis in this paper relies on the 2002 and 2006 releases of HILDA (Waves 2 and 6), which provide the


5

only micro-level, longitudinal data on household wealth holding in Australia (see Headey, Marks & Wooden 2005; Wooden, Freidin & Watson 2002 for a detailed presentation of these data).6

We have necessarily made a number of sample restrictions. Because household wealth can be difficult to measure and conceptualise in households with multiple families, a small number of multi-family households, all group households, and all related family households have been excluded, as have all single or couple-headed households in which the respondent (or his or her partner) did not provide an interview. Finally, in order to maintain a sufficiently large sample of households around retirement age, the sample used was restricted to all households in which the reference person was aged between 55 years and 74 years. These restrictions resulted in a total analysis sample of 927 couple-headed households and 582 single-headed households in 2002, and 867 couple-headed households and 602 single-headed households in 2006.7

Most of HILDA’s wealth components are collected at the household level.8 In this paper, we consider the way in which wealth is distributed across five broad asset types—net financial wealth, net business equity, net equity in own home, lifestyle assets and total value of superannuation assets. These asset types are defined so as to capture possible incentives to reallocate assets embedded in the pre-2007 asset/income test rules for qualifying for the Age Pension.

Net financial wealth is calculated as the total value of interest-bearing assets held in banks and other institutions, stocks and mutual funds, life insurance funds, trust funds and collectibles minus the total value of unsecured debts (which also include car loans). The net value (equity) of own home captures households’ equity in their principal residence. Net business equity includes the net value of all business shares owned by all household members. Lifestyle assets include all non-liquid assets that do not necessarily generate a steady income stream and include all transport and recreational vehicles (such as boats or caravans) and all other real estate assets (such as holiday homes and other properties) owned by household members.9 The superannuation component of net wealth includes the total amount of superannuation capital owned by all household members.

While HILDA does not use the concept of a reference person (or household head), in this report we define the head of household to be the oldest partner in couple-headed households. We then separately account for the age of household heads and their spouses in the estimation model. Moreover, the analysis considers single and couple-headed households separately as each group faces different incentives given the asset and income test rules in place.

retirement status of older Australians

As this study’s objective was to explore whether the incentives embedded in the asset and income tests used to determine eligibility for the Age Pension induce older Australian households to revise their portfolio allocation, the analysis explicitly considered two subpopulations. The first included all households in which the reference person (or household head) was aged between 55 and 64 years. Given that the reference person was defined as the oldest partner in a couple, few household members from this group were entitled to claim the Age Pension (about 3 per cent


6

of all couple-headed households in 2006). The second subpopulation included all households in which the reference person was aged between 65 and 74 years. This implies that in the second group at least one household member was eligible to receive Age Pension benefits.

We began by considering the retirement status of individuals in these two groups of households. Information on relevant demographic characteristics and place of residence for individuals in our estimation sample is reported in Table 1 for couple-headed households and in Table 2 for single individuals.

While most household members in younger couple-headed households (that is, those in which the head is aged between 55 and 64 years) are not eligible to claim Age Pension benefits, in about 17 per cent (22 per cent) of couples both partners nonetheless reported being retired in 2006 (2002). In contrast, approximately 40 per cent of single-headed households in this younger age group had already left the labour force during the same period. Not surprisingly, the proportion of retirees rises substantially after the age of 64 years. For instance, in 2006 (2002), at least 80 (83) per cent of all couple-headed households reported at least one household member being retired while up to 87 (88) per cent of single individuals were no longer in the labour force in 2006 (2002).

health status and wealth

Recipients of the Age Pension are eligible to receive subsidies for health care and pharmaceuticals. As a result, the incentive to qualify for the Age Pension might also be affected by the health status of future claimants. Individuals in poor health may have greater incentives to reallocate their assets in order to qualify for the Age Pension. This report examines the impact of health using a measure of self-assessed (non-fatal) health commonly used in the health literature. Specifically, HILDA respondents are asked to rate their health on a five-point scale labelled: ‘excellent’, ‘very good’, ‘good’, ‘fair’ and ‘poor’. We have created an indicator variable for poor health equal to 1 whenever a respondent rated their health as either ‘fair’ or ‘poor’ and 0 otherwise.

Tables 1 and 2 reveal that the incidence of poor health does not differ substantially across household types, with about 30 per cent of reference persons reporting being in poor health. Surprisingly, being older is also not associated with significant differences in self-reported health status. For instance, approximately 27 (30) per cent of married heads of household aged 55 to 64 years report being in poor health compared to 33 (27) per cent of married household heads in the 65 to 74 years age group in 2006 (2002) respectively. These differences in self-reported health status across age groups are not statistically significant.10

Information about the relationship between net wealth, asset portfolios and self-reported health status is reported in Table 3 for couple-headed households and in Table 4 for single-headed households. Being in good health is associated with a higher incidence of owning each asset type as well as with holding more money in all asset types.11 For instance, couple-headed households in which both partners report being in good health hold over $300,000 more wealth at the median


7

(and the mean) than couple-headed households in which at least one spouse reports being in poor health (Table 3). These results align with findings from earlier US studies that demonstrate the close link between health and wealth. Given these differences in the level of net worth—and the potential incentives inherent in the Age Pension eligibility rules—it is reasonable to expect that health status may affect the portfolio choices of older households.

Wealth levels and asset portfolios of households

Descriptive statistics for household net wealth, asset portfolios and income are presented in Table 5 for couple-headed households and in Table 6 for single-headed households. These results are presented separately for the 55 to 64 year-old and 65 to 74 year-old age groups. Cross-sectional comparisons across age groups should be interpreted carefully, however, because such a comparison involves comparing individuals belonging to different birth cohorts. For example, a cross-sectional comparison of the level of assets held by younger and older households reveals that in 2002 (2006) couple-headed households aged 55 to 64 years had on average about $200,000 ($90,000) more net wealth than couple-headed households aged 65 to 74 years. This suggests that older households might be depleting their wealth as they grow older. It is difficult, however, to attribute this disparity in wealth across these two groups of households to the result of lifecycle changes. It is not possible to know whether any difference in the net wealth or asset portfolios of younger and older households stems from the fact that older household heads are approximately 10 years older (see Table 1); that is, a lifecycle change, or because they were born in an earlier decade (for example, the 1930s versus the 1940s).

The median net wealth of younger Australian households grew substantially between 2002 and 2006. In 2006, couple-headed (single-headed) households aged 55 to 64 years had a median wealth of about $137,000 ($115,000) more than the same age group in 2002.12 This is not surprising given the exceptional boom in both the equity and the real estate markets over this period. What is surprising is that the financial boom did not extend to older households. Single-headed households aged 65 to 74 years saw only a slight increase (less than $7,000) in median wealth between 2002 and 2006, while the median net wealth of couple-headed households in this age group increased substantially over the same period. These patterns of growth are consistent with the existence of a positive cohort effect, but are difficult to reconcile with simple ageing or period effects.


8


9

Tabl

e 1:

D

escr

iptiv

e st

atis

tics

by a

ge g

roup

(cou

ple-

head

ed h

ouse

hold

s)

2002

2006

Age

of re

fere

nce

pers

on55

–64

year

sSt

anda

rd

devi

atio

n65

–74

year

sSt

anda

rd

devi

atio

n55

–64

year

sSt

anda

rd

devi

atio

n65

–74

year

sSt

anda

rd

devi

atio

nD

emog

raph

ics

Age

59.0

52.

8069

.13

2.92

59.3

42.

8069

.13

2.77

Spou

se a

ge54

.40

5.56

64.9

05.

0054

.94

5.12

64.4

75.

27Ye

ars

of e

duca

tion

11.3

42.

6310

.71

2.65

11.7

52.

4911

.02

2.77

Spou

se y

ears

of e

duca

tion

11.1

92.

4510

.50

2.60

11.6

42.

4510

.82

2.49

Fem

ale

(pro

port

ion)

0.19

0.40

0.21

0.41

0.22

0.41

0.20

0.40

Hom

e ow

ners

(pr

opor

tion

)0.

890.

320.

910.

290.

910.

290.

910.

29H

ealt

h an

d re

tire

men

t (pr

opor

tion

of)

Reti

red

0.37

0.48

0.83

0.38

0.32

0.47

0.80

0.40

Spou

se re

tire

d0.

360.

480.

780.

410.

260.

440.

750.

43B

oth

reti

red

0.22

0.42

0.70

0.46

0.17

0.37

0.67

0.47

Poor

hea

lth

0.30

0.46

0.27

0.44

0.27

0.44

0.33

0.47

Spou

se p

oor h

ealt

h0.

220.

410.

250.

430.

190.

400.

240.

43Pl

ace

of re

side

nce

(pro

port

ion)

New

Sou

th W

ales

0.31

0.46

0.36

0.48

0.32

0.47

0.38

0.48

Vic

tori

a0.

250.

430.

270.

440.

260.

440.

240.

43Q

ueen

slan

d0.

190.

400.

140.

340.

190.

390.

160.

36S

outh

Aus

tral

ia0.

070.

250.

100.

290.

070.

260.

090.

29W

este

rn A

ustr

alia

0.12

0.32

0.11

0.31

0.12

0.33

0.11

0.32

Tasm

ania

0.03

0.16

0.03

0.17

0.02

0.14

0.01

0.12

Nor

ther

n Te

rrit

ory

0.01

0.08

0.00

0.00

0.01

0.11

0.00

0.00

Aus

tral

ian

Cap

ital

Ter

rito

ry0.

030.

160.

000.

060.

010.

090.

010.

11n

548

379

511

356

Not

e:

Calc

ulat

ions

are

bas

ed o

n W

ave

2 an

d W

ave

6 of

the

HIL

DA

sur

vey.


8


9

Tabl

e 2:

D

escr

iptiv

e st

atis

tics

by a

ge g

roup

(sin

gle-

head

ed h

ouse

hold

s)

2002

2006

Age

of re

fere

nce

pers

on55

–64

year

sSt

anda

rd

devi

atio

n65

–74

year

sSt

anda

rd

devi

atio

n55

–64

year

sSt

anda

rd

devi

atio

n65

–74

year

sSt

anda

rd

devi

atio

nD

emog

raph

ics

Age

59.1

62.

9969

.71

2.83

59.5

22.

7869

.33

2.92

Year

s of

edu

cati

on11

.03

2.70

10.3

62.

6011

.14

2.63

10.8

82.

51Fe

mal

e (p

ropo

rtio

n)0.

620.

490.

700.

460.

620.

490.

670.

47H

ome

owne

rs (

prop

orti

on)

0.66

0.47

0.77

0.42

0.69

0.46

0.73

0.45

Nev

er m

arri

ed (

prop

orti

on)

0.15

0.36

0.11

0.31

0.16

0.36

0.10

0.30

Wid

owed

(pr

opor

tion

)0.

230.

420.

590.

490.

250.

430.

550.

50D

ivor

ced

(pro

port

ion)

0.62

0.49

0.30

0.46

0.60

0.49

0.35

0.48

Hea

lth

and

reti

rem

ent (

prop

orti

on o

f)Re

tire

d0.

420.

500.

880.

320.

400.

490.

870.

34Po

or h

ealt

h0.

310.

460.

340.

470.

320.

470.

350.

48Pl

ace

of re

side

nce

(pro

port

ion)

New

Sou

th W

ales

0.32

0.47

0.42

0.49

0.32

0.47

0.33

0.47

Vic

tori

a0.

220.

420.

210.

410.

280.

450.

220.

42Q

ueen

slan

d0.

230.

420.

170.

370.

180.

390.

220.

41S

outh

Aus

tral

ia0.

080.

270.

080.

270.

090.

290.

080.

27W

este

rn A

ustr

alia

0.10

0.30

0.09

0.29

0.10

0.30

0.12

0.33

Tasm

ania

0.03

0.17

0.03

0.16

0.02

0.14

0.01

0.12

Nor

ther

n Te

rrit

ory

0.01

0.11

0.00

0.00

0.01

0.10

0.01

0.10

Aus

tral

ian

Cap

ital

Ter

rito

ry0.

010.

090.

010.

100.

010.

070.

010.

08n

306

276

336

266

Not

e:

Calc

ulat

ions

are

bas

ed o

n W

ave

2 an

d W

ave

6 of

the

HIL

DA

sur

vey.


10


11

Tabl

e 3:

W

ealth

hol

ding

by

subj

ectiv

e he

alth

sta

tus

(cou

ple-

head

ed h

ouse

hold

s)

2002

2006

Subj

ectiv

e he

alth

sta

tus

Poor

/fai

rSt

anda

rd

devi

atio

nG

ood

Stan

dard

de

viat

ion

Poor

/fai

rSt

anda

rd

devi

atio

nG

ood

Stan

dard

de

viat

ion

Net

wea

lth

($)

Mea

n to

tal n

et w

ealt

h61

2,93

178

6,57

694

7,68

397

0,51

176

1,87

888

5,89

31,

124,

719

940,

733

Med

ian

tota

l net

wea

lth

375,

404

157,

000

688,

543

233,

219

498,

280

199,

350

837,

482

266,

700

Mea

n as

set p

ortf

olio

($)

Tota

l fina

ncia

l wea

lth

115,

224

283,

515

185,

508

402,

832

113,

203

326,

342

169,

881

357,

838

Inte

rest

-ear

ning

ass

ets

(ban

ks)

45,3

2614

0,99

354

,734

118,

993

46,3

0793

,293

52,2

7010

5,19

8In

tere

st-e

arni

ng a

sset

s (o

ther

)5,

527

40,4

186,

312

40,2

301,

395

10,3

993,

725

48,5

72Eq

uity

in s

tock

s59

,990

183,

610

96,0

5524

1,12

965

,378

291,

329

112,

447

312,

259

Oth

er a

sset

s12

,981

110,

470

40,7

9922

8,21

16,

479

31,6

4116

,328

98,4

04U

nsec

ured

deb

ts8,

600

37,3

3212

,393

51,1

896,

356

17,6

3414

,889

64,8

16B

usin

ess

11,4

9676

,396

79,6

0638

4,69

817

,374

98,9

2849

,640

217,

258

Ow

n ho

me

284,

567

318,

254

329,

192

279,

938

346,

320

307,

547

434,

013

307,

410

Tota

l life

styl

e75

,132

221,

608

134,

980

264,

944

122,

772

309,

191

172,

453

345,

876

Oth

er re

al e

stat

e49

,353

186,

834

100,

568

253,

882

96,5

8528

7,07

213

9,28

833

7,82

7Ve

hicl

es25

,779

61,2

6934

,411

64,1

5026

,187

48,3

3933

,166

45,2

30Su

pera

nnua

tion

126,

513

257,

142

218,

398

303,

930

162,

209

293,

031

298,

732

386,

381

Curr

ent i

ncom

e ($

)52

,792

62,6

2275

,498

62,3

2357

,265

64,4

7483

,684

67,9

09Pr

opor

tion

ow

ning

Fina

ncia

l wea

lth

0.99

70.

050

0.99

90.

033

1.00

00.

000

0.99

50.

068

Bus

ines

s0.

060

0.23

80.

176

0.38

10.

077

0.26

70.

164

0.37

1O

wn

hom

e0.

867

0.34

00.

915

0.27

90.

864

0.34

30.

932

0.25

1Li

fest

yle

0.95

40.

210

0.97

40.

160

0.96

70.

180

0.99

40.

078

Supe

rann

uati

on0.

632

0.48

30.

780

0.41

50.

685

0.46

50.

863

0.34

4n

369

558

332

535

Not

es:

Aut

hors

’ cal

cula

tion

bas

ed o

n W

aves

2 a

nd 6

of H

ILD

A d

ata.

Poo

r/fa

ir h

ealt

h st

atus

if a

t lea

st o

ne p

artn

er ra

ted

thei

r hea

lth

as ‘p

oor’

or ‘f

air’

. All

figur

es

are

repo

rted

in c

onst

ant 2

006

Aus

tral

ian

dolla

rs.


10


11

Tabl

e 4:

W

ealth

hol

ding

by

subj

ectiv

e he

alth

sta

tus

(sin

gle-

head

ed h

ouse

hold

s)

2002

2006

Subj

ectiv

e he

alth

sta

tus

Poor

/fai

rSt

anda

rd

devi

atio

nG

ood

Stan

dard

de

viat

ion

Poor

/fai

rSt

anda

rd

devi

atio

nG

ood

Stan

dard

de

viat

ion

Net

wea

lth

($)

Mea

n to

tal n

et w

ealt

h29

6,53

945

5,59

147

1,53

466

8,70

333

6,52

045

5,05

360

7,61

872

7,05

5M

edia

n to

tal n

et w

ealt

h15

1,98

712

4,61

627

5,08

516

1,85

922

3,15

319

4,68

037

9,91

923

9,60

0M

ean

asse

t por

tfol

io ($

)To

tal fi

nanc

ial w

ealt

h60

,737

162,

328

112,

585

267,

428

61,6

8517

9,55

212

3,64

826

3,71

6In

tere

st-e

arni

ng a

sset

s (b

anks

)19

,977

41,3

2129

,923

69,8

6620

,366

50,5

8236

,759

86,2

22In

tere

st-e

arni

ng a

sset

s (o

ther

)38

43,

436

5,05

037

,521

2,69

417

,667

3,33

924

,550

Equi

ty in

sto

cks

32,3

8511

7,68

070

,660

237,

067

35,4

8714

8,65

980

,972

212,

972

Oth

er a

sset

s11

,533

60,7

6310

,116

43,5

945,

761

56,8

458,

267

40,1

87U

nsec

ured

deb

ts3,

542

17,2

203,

165

20,0

642,

622

7,20

65,

689

24,7

60B

usin

ess

17,8

8919

9,30

125

,131

171,

658

352

4,06

116

,506

96,6

86O

wn

hom

e14

9,07

118

1,92

521

5,08

722

8,40

618

2,41

120

5,84

728

3,08

730

6,54

7To

tal l

ifest

yle

28,5

3478

,250

53,2

0524

3,94

042

,581

133,

313

77,4

9823

9,98

9O

ther

real

est

ate

17,5

4374

,893

40,1

5924

2,94

632

,276

129,

564

61,8

9523

2,50

5Ve

hicl

es10

,991

17,4

2813

,047

18,5

3810

,305

13,4

4115

,603

22,1

97Su

pera

nnua

tion

40,3

0811

7,53

065

,527

159,

605

49,4

9113

7,32

110

6,87

923

1,47

9Cu

rren

t inc

ome

($)

21,1

0320

,914

33,1

6032

,634

23,5

4830

,987

41,0

0755

,626

Prop

orti

on o

wni

ngFi

nanc

ial w

ealt

h0.

961

0.19

40.

987

0.11

40.

978

0.14

80.

987

0.11

4B

usin

ess

0.05

00.

218

0.05

00.

219

0.01

60.

124

0.06

70.

251

Ow

n ho

me

0.61

60.

488

0.75

80.

429

0.61

00.

489

0.75

20.

433

Life

styl

e0.

778

0.41

60.

830

0.37

60.

792

0.40

70.

882

0.32

3Su

pera

nnua

tion

0.32

80.

471

0.51

30.

500

0.40

40.

492

0.62

10.

486

n18

439

819

640

6

Not

es:

Aut

hors

’ cal

cula

tion

bas

ed o

n W

aves

2 a

nd 6

of H

ILD

A d

ata.

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


12


13

Tabl

e 5:

W

ealth

hol

ding

by

age

(cou

ple-

head

ed h

ouse

hold

s) 2002

2006

Age

of re

fere

nce

pers

on55

–64

year

sSt

anda

rd

devi

atio

n65

–74

year

sSt

anda

rd

devi

atio

n55

–64

year

sSt

anda

rd

devi

atio

n65

–74

year

sSt

anda

rd

devi

atio

nN

et w

ealt

h ($

)M

ean

tota

l net

wea

lth

891,

434

969,

935

684,

294

805,

118

1,02

6,31

893

0,76

891

1,84

693

9,73

8M

edia

n to

tal n

et w

ealt

h61

7,91

822

9,35

942

2,89

617

8,50

075

5,25

025

6,95

564

7,80

025

9,52

7M

ean

asse

t por

tfol

io ($

)To

tal fi

nanc

ial w

ealt

h16

1,33

940

6,51

814

9,39

427

3,50

712

0,59

328

4,00

418

4,63

141

6,80

0In

tere

st-e

arni

ng a

sset

s (b

anks

)53

,564

137,

935

46,6

9411

2,31

444

,198

78,7

3057

,907

124,

818

Inte

rest

-ear

ning

ass

ets

(oth

er)

4,21

138

,551

8,75

142

,750

1,82

534

,063

4,15

443

,279

Equi

ty in

sto

cks

77,9

0321

4,80

486

,452

227,

940

74,8

2423

2,50

912

0,01

538

3,24

2O

ther

ass

ets

41,6

6824

0,08

210

,304

39,2

6115

,233

95,9

348,

319

44,3

49U

nsec

ured

deb

ts16

,007

55,0

782,

808

24,4

0715

,486

56,9

965,

763

41,9

82B

usin

ess

71,1

7236

4,18

121

,341

155,

847

49,5

3020

5,70

118

,482

133,

638

Ow

n ho

me

318,

443

300,

309

299,

128

291,

566

393,

808

304,

463

405,

807

318,

631

Tota

l life

styl

e12

0,82

822

9,05

794

,250

278,

274

176,

139

339,

666

119,

021

319,

151

Oth

er re

al e

stat

e84

,056

195,

016

72,5

4727

5,86

214

1,97

132

6,03

894

,047

306,

831

Vehi

cles

36,7

7178

,443

21,7

0321

,843

34,1

6949

,616

24,9

7441

,443

Supe

rann

uati

on21

9,65

230

2,60

812

0,18

125

5,55

128

6,24

738

1,80

418

3,90

631

2,12

2Cu

rren

t inc

ome

($)

78,2

8971

,526

47,3

7641

,780

90,2

5977

,347

48,7

3940

,269

Prop

orti

on o

wni

ngFi

nanc

ial w

ealt

h0.

999

0.03

30.

997

0.05

10.

997

0.05

60.

998

0.04

8B

usin

ess

0.16

50.

371

0.07

10.

258

0.17

90.

384

0.05

90.

235

Ow

n ho

me

0.88

90.

315

0.90

60.

292

0.90

40.

295

0.90

70.

291

Life

styl

e0.

968

0.17

60.

961

0.19

30.

985

0.12

20.

980

0.13

9Su

pera

nnua

tion

0.85

20.

355

0.51

30.

500

0.90

80.

289

0.62

70.

484

n54

837

951

135

6

Not

es:

Aut

hors

’ cal

cula

tion

bas

ed o

n W

aves

2 a

nd 6

of H

ILD

A d

ata.

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


12


13

Tabl

e 6:

W

ealth

hol

ding

by

age

(sin

gle-

head

ed h

ouse

hold

s)

2002

2006

Age

of re

fere

nce

pers

on55

–64

year

sSt

anda

rd

devi

atio

n65

–74

year

sSt

anda

rd

devi

atio

n55

–64

year

sSt

anda

rd

devi

atio

n65

–74

year

sSt

anda

rd

devi

atio

nN

et w

ealt

h ($

)M

ean

tota

l net

wea

lth

428,

472

615,

859

399,

912

611,

573

584,

648

753,

879

432,

331

507,

972

Med

ian

tota

l net

wea

lth

224,

619

145,

700

240,

818

146,

295

339,

925

247,

200

313,

775

194,

750

Mea

n as

set p

ortf

olio

($)

Tota

l fina

ncia

l wea

lth

96,8

0825

9,75

494

,910

214,

618

108,

532

265,

658

96,4

4120

4,63

6In

tere

st-e

arni

ng a

sset

s (b

anks

)22

,239

61,3

3832

,005

63,0

7832

,236

89,5

8530

,244

55,7

85In

tere

st-e

arni

ng a

sset

s (o

ther

)5,

200

41,0

761,

616

10,4

733,

923

27,5

262,

090

13,3

87Eq

uity

in s

tock

s61

,627

222,

024

54,5

4218

8,31

272

,412

213,

282

57,6

9516

8,92

1O

ther

ass

ets

12,9

7654

,215

7,74

643

,755

6,96

249

,778

8,07

141

,402

Uns

ecur

ed d

ebts

5,23

325

,595

999

4,71

17,

000

26,9

821,

658

5,24

9B

usin

ess

31,5

8218

1,25

112

,495

180,

173

16,4

4910

1,38

54,

352

33,6

17O

wn

hom

e17

6,31

421

9,28

521

4,53

121

2,01

926

0,53

732

5,01

423

6,26

021

1,16

0To

tal l

ifest

yle

38,1

0796

,923

53,7

1228

5,19

879

,137

200,

783

48,9

3222

3,89

4O

ther

real

est

ate

24,2

3992

,413

43,0

6628

4,88

862

,099

197,

479

39,2

0821

3,70

9Ve

hicl

es13

,869

20,3

7010

,645

15,1

1417

,038

22,1

719,

723

15,5

65Su

pera

nnua

tion

85,6

5918

2,06

124

,264

81,1

6311

9,99

423

7,46

446

,347

149,

230

Curr

ent i

ncom

e ($

)34

,952

34,6

1522

,635

21,3

9942

,133

58,9

7426

,325

31,5

30Pr

opor

tion

ow

ning

Fina

ncia

l wea

lth

0.97

10.

167

0.98

70.

113

0.98

00.

140

0.98

90.

105

Bus

ines

s0.

078

0.26

80.

017

0.13

10.

067

0.25

00.

029

0.16

8O

wn

hom

e0.

664

0.47

30.

769

0.42

20.

687

0.46

50.

729

0.44

5Li

fest

yle

0.86

70.

340

0.75

00.

434

0.86

60.

341

0.83

40.

373

Supe

rann

uati

on0.

611

0.48

80.

269

0.44

40.

696

0.46

10.

360

0.48

1n

306

276

336

266

Not

es:

Aut

hors

’ cal

cula

tion

bas

ed o

n W

aves

2 a

nd 6

of H

ILD

A d

ata.

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


14

younger households (2002) versus older households (2006)

Given the longitudinal nature of the HILDA survey, some respondents aged 55 to 64 years in 2002 would be in the older age group in 2006. In order to infer more accurately the changes in wealth components due strictly to ageing, it is best to compare the wealth and asset portfolios of households aged 55 to 64 years in 2002 with those of households aged 65 to 74 years in 2006. Such a comparison seems to suggest that overall older Australian households are not depleting their wealth as they age. For instance, couples aged 65 to 74 years in 2006 hold on average about $30,000 more net wealth at the median than couples aged 55 to 64 years in 2002. At the same time, this relatively small growth in net wealth may stem from changes in the equity and housing markets over this period.

Comparison of the asset portfolios of those households aged 65 to 74 years in 2006 with those of households aged 55 to 64 years in 2002 shows that on average couples aged 65 to 74 years in 2006 held approximately $23,000 more financial wealth than couples aged 55 to 64 years in 2002. Surprisingly, couples aged 65 to 74 years in 2006 held, on average, significantly more wealth in stocks and mutual funds (about $120,000) than their counterparts aged 55 to 64 years did in 2002 (about $78,000). At the same time, they also held significantly less wealth in life insurance, trust funds or collectables. This pattern was surprising because we typically expect households to reduce their exposure to risky assets as they age.

changes in asset portfolios over time

Asset portfolios have changed over time. Specifically, couple-headed households aged 55 to 64 years in 2006 held less financial wealth than their counterparts in 2002 ($120,593 versus $161,339) but held more net wealth on average (and at the median). From this observation, one could speculate that the lower level of financial wealth the younger cohort held in 2006 could have reflected a change in allocation across asset types. In particular, younger couple and single-headed households held substantially more assets in superannuation in 2006 than their counterparts did in 2002. For example, couple-headed (single-headed) households aged 55 to 64 years in 2006 held approximately $67,000 ($34,000) more wealth in superannuation than the same age group did in 2002. Similar growth over time is observed in the superannuation wealth of older households. As a result, the substantial gap in superannuation wealth between those aged 55 to 64 years and those aged 65 to 74 years in 2002 (and 2006) is most likely a combination of a pure age effect (superannuation is used to finance consumption in retirement) and a cohort effect (the younger cohort tends to have reallocated more assets towards superannuation).

A cross-sectional comparison suggests little difference in the average home equity of the two age groups, regardless of household type. There was a small drop in housing equity between the two age groups (approximately $19,000) in 2002 and a small increase (less than $13,000 in 2006). Over time, younger households appear to be holding more wealth in their homes. For example, couple-headed households aged 55 to 64 years in 2006 held $75,000 more wealth in the form of equity in their principal residence than did their counterparts in 2002. Moreover, both singles and couples aged 65 to 74 years in 2006 held significantly more wealth in their own home than did corresponding households aged 55 to 64 years in 2002.


15

These patterns provide some evidence that Australian households hold more wealth in their homes as they age, which is consistent with incentives created by the asset test to become eligible for or increase payment of Age Pension benefits. However, it could also merely reflect the boom in the housing market over the period covered by these data. Furthermore, home ownership rates are consistently high, regardless of age, among couple-headed households (at about 90 per cent) and increasing over time among single-headed households.

Summary

Figures 1 to 8 depict the way in which households of different ages distribute their net wealth across the five major asset categories.

Figures 1 and 2 present the asset portfolios of couple-headed households aged 55 to 64 years in 2006 and 2002 respectively. The most striking difference in the asset portfolios of younger couples over time is the disparity in financial wealth. Younger couples held 18 per cent of total net wealth in the form of financial wealth in 2002 compared to 12 per cent in financial wealth in 2006. This suggests a switch in preferences for other assets, in particular superannuation, lifestyle assets and, to some degree, housing.

Figures 3 and 4 present the asset allocation of older couple-headed households in 2006 and 2002 respectively. The asset portfolios of older couples do not seem to have changed much across survey years.13 A comparison of Figures 2 and 3—which potentially controls to some degree for the existence of cohort effects—seems to support the view that households increase the portfolio share devoted to equity in their own home as they age. The fact that the 65 to 74 years age group had, on average, more total net wealth in 2006 than did the 55 to 64 years age group in 2002 (see Table 5) provides additional evidence that households hold a higher share of their wealth in their homes as they age.14

These patterns are even more striking among single-headed households (see Figures 5, 6, 7 and 8). In particular, these figures reveal that the share of wealth held in one’s own home was 14 percentage points higher among singles aged 65 to 74 years in 2006 than among those aged 55 to 64 years in 2002 with approximately 55 per cent of total net wealth allocated to the family home. Furthermore, single-headed households appear to have allocated a smaller proportion of wealth to superannuation than did couples.


16

Figure 1: Couple-headed households, head aged 55 to 64 years, 2006

Financial

wealthBusiness

Own homeLifestyle

Superannuation12%

5%

38%17%

28%

Note: Authors’ calculation based on Wave 6 of HILDA data.


Financial wealth

Business

Own home

Lifestyle

Superannuation18%

8%

36%

14%

25%



17


Financial wealth

Business

Own home

Lifestyle

Superannuation

20%

2%

45%

13%

20%



Financial wealth

Business

Own home

Lifestyle

Superannuation

22%

3%

44%

14%

18%



18

Figure 5: Single-headed households, head aged 55 to 64 years, 2006

Financial wealth

Business

Own home

Lifestyle

Superannuation

19%

3%

45%

14%

21%



Financial wealth

Business

Own home

Lifestyle

Superannuation

23%

7%

41%

9%

20%



19


Financial wealth

Business

Own home

Lifestyle

Superannuation

22%

1%

55%

11%

11%



Financial wealth

Business

Own home

Lifestyle

24%

3%

54%

13%

6%Superannuation



20

5 Regression resultsThe descriptive results discussed above highlight the broad differences in asset portfolios across household type, age and time. However, it is often difficult to interpret these differences because the level of household wealth varies with household type, age and time. Consequently, we are often comparing households that are not equally wealthy. This is problematic because the nature of credit markets and financial institutions implies a link between total wealth and asset portfolios. We would like to know whether the changes over time in asset portfolios represent a structural change in the way Australian households allocate their wealth or are merely the result of Australians becoming wealthier. Similarly, we would like to understand whether changes in portfolios as households age can be attributed to the incentives inherent in the Age Pension eligibility rules or are merely the result of households spending down their wealth to finance consumption in retirement.

To gain a deeper understanding of these issues, a model is needed that will help estimate the effect of access to a public pension (the Age Pension) on households’ portfolios. Such an estimation strategy must recognise that the propensity to invest in a specific asset will depend on the types (and amounts) of other assets held; compare households with the same level of net wealth; and allow control for other confounding factors, such as poor health.

Therefore, we need to estimate a system of regression equations with an adding up constraint imposed to account for total net wealth (see Blau & Graham 1990). Consequently, we estimated the following reduced-form model of asset composition:

sinh–1(Aik

)=a0k

+Yib

1k+X

ib

2k+AgePension

ib

3k+W

ib

4k+µ

ik

Aik

where is the dollar value of asset k that household i holds.

We consider the five major asset categories of financial wealth, business equity, equity in own home, lifestyle assets, and superannuation funds where:

Yi includes both total family gross income and a dummy variable capturing whether household

income is within the range of being able to collect Age Pension15 and

Xi includes a measure of poor health as well as a vector of those demographic characteristics

reflecting a household’s lifecycle stage.

This specification allows households’ asset portfolios to depend on net wealth in order to account for any capital market imperfections (such as credit constraints), which might vary across households and be related to the decision to hold a particular asset.

The variable ‘Age Pension’ captures the impact of meeting the age requirement for claiming the Age Pension. For couple-headed households, we account separately for the effect of each partner being over the pension age.

An inverse hyperbolic sine transformation (sinh–1) of assets and income has been adopted to account for the potentially non-positive and highly skewed nature of the distributions of these variables (see Cobb-Clark & Hildebrand 2006a, for further discussion). Finally, equation (eq1)


21

is estimated as a system of equations and a set of cross-equation restrictions are imposed in order to satisfy the adding-up requirement that the sum of assets across asset types equals net wealth.16

We considered two model specifications. Our baseline model does not capture explicitly the effect a household’s lifecycle stage on asset allocation other than by controlling for whether its members have become eligible to claim Age Pension. Our second specification adds a quadratic term in the age of the reference person (the oldest partner in a couple) in order to better disentangle the effects due to ageing from the effect of becoming eligible to claim pension benefits. The model for single individuals is defined analogously with an added control for whether singles are divorced or never married. Being widowed constitutes the reference group. Marginal effects and t-statistics from this estimation are presented in Tables 7 to 10.17 We have estimated our models using both the 2002 and 2006 HILDA data. In this section, the results from the 2006 data are discussed (see Tables 7 to 10) and the estimation results based on the 2002 data are reported in the appendix (see Tables A1 to A4).

determinants of asset portfolios

Given the estimation framework described above, the potential impact of the Age Pension on asset portfolios is captured in two ways: first, through a measure of income eligibility and second, through measures of age eligibility. Total wealth levels are held constant through the inclusion of our measure of net wealth. In effect, the results are calculated for households with average levels of wealth.18

Income and income eligibility

By considering the effects of income first, we found that asset allocation is, unsurprisingly, related to households’ current income levels. Comparing households that are equally wealthy, but that have different incomes, we find that at higher levels of household income couples and single individuals held significantly more of their net wealth in superannuation and less in their own homes. In addition, couples allocate more wealth to lifestyle assets, a finding which is robust to model specifications. Finally, both single and couple-headed households allocate more wealth to business assets when they have higher incomes.

When considering the effect of income eligibility for the Age Pension, we found that, among couples, there were no significant effects of having household income in the range of income eligibility on asset portfolios (see Tables 7 and 9). However, among singles, we found that being within the income eligibility range was associated with holding significantly less wealth in one’s own home and more in financial wealth. This effect is robust across specifications and is net of the (linear) effect of current income on asset allocation generally. In particular, although there was no overall relationship between income levels and the amount of financial wealth that single individuals held, there was a sharp increase in the holding of financial wealth in the income range associated with eligibility for the Age Pension.


22


23

Tabl

e 7:

(A

vera

ge) d

eter

min

ants

of a

sset

por

tfol

ios:

cou

ple-

head

ed h

ouse

hold

s (m

argi

nal e

ffect

s an

d t-

stat

istic

s), W

ave

6

Fina

ncia

l wea

lthB

usin

ess

asse

tsO

wn

hom

eLi

fest

yle

Supe

rann

uatio

n

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

Inco

me

($)

Tota

l inc

ome

2.09

1.46

0.27

2.45

–11.

62–4

.83

2.37

4.26

6.88

5.99

Elig

ibili

ty r

ange

–141

,110

.05

–1.1

113

,145

.73

0.56

–12,

981.

48–0

.10

33,7

05.1

90.

9210

7,24

0.62

1.31

Dem

ogra

phic

s

Educ

atio

n–1

0,94

0.74

–0.5

365

9.13

0.33

4,62

4.86

0.22

–6,4

17.9

1–1

.00

12,0

74.6

70.

83

Hea

d el

igib

le fo

r A

ge P

ensi

on25

2,47

1.95

1.81

–9,8

94.2

6–0

.50

–13,

296.

45–0

.08

–81,

987.

10–1

.43

–147

,294

.12

–1.0

2

Spou

se e

ligib

le fo

r A

ge P

ensi

on31

7,46

5.28

2.22

–23,

241.

27–1

.11

33,1

55.3

00.

2164

,402

.85

1.05

–391

,782

.16

–2.3

2

Fem

ale

head

–62,

517.

02–0

.71

–9,8

60.0

9–0

.81

76,9

29.3

40.

8128

,347

.90

1.04

–32,

900.

13–0

.44

Prev

ious

ly m

arri

ed–1

78,6

26.3

1–2

.02

15,1

18.5

11.

09–1

0,47

9.37

–0.1

1–5

69.2

0–0

.02

174,

556.

382.

10

Poor

hea

lth

–165

,470

.17

–1.3

7–1

0,98

5.54

–0.6

323

2,18

9.88

2.08

–60,

701.

48–1

.73

4,96

7.30

0.06

Elig

ible

for A

ge

Pens

ion

x po

or h

ealt

h41

4,95

3.00

1.97

–28,

542.

13–1

.14

–213

,699

.94

–0.9

072

,849

.75

0.83

–245

,560

.67

–1.0

7

Spou

se e

ligib

le x

po

or h

ealt

h–1

69,9

24.8

1–0

.90

15,2

83.8

70.

6799

,017

.37

0.42

–73,

729.

78–0

.80

129,

353.

340.

48

Net

wor

th–0

.38

–0.6

90.

002.

840.

5320

.44

0.72

1.37

0.12

5.54

n8

618

618

618

618

61

R2

0.05

0.06

0.25

0.18

0.35

Not

es:

Elig

ible

for A

ge P

ensi

on if

at l

east

one

par

tner

is e

ligib

le. P

oor h

ealt

h if

one

mem

ber r

epor

ts b

eing

in p

oor h

ealt

h (s

ee te

xt fo

r pre

cise

defi

niti

on).

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


22


23

Tabl

e 8:

(Av

erag

e) d

eter

min

ants

of a

sset

por

tfol

ios:

sin

gle-

head

ed h

ouse

hold

s (m

argi

nal e

ffect

s an

d t-

stat

istic

s), W

ave

6

Fina

ncia

l wea

lthB

usin

ess

asse

tsO

wn

hom

eLi

fest

yle

Supe

rann

uatio

n

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

Inco

me

($)

Tota

l inc

ome

2.80

0.92

0.24

1.77

–9.0

7–2

.10

0.66

0.56

5.36

5.42

Elig

ibili

ty r

ange

348,

396.

193.

6412

,093

.36

1.05

–423

,493

.31

–3.7

449

,585

.87

1.71

13,4

17.9

10.

17

Dem

ogra

phic

s

Educ

atio

n13

,544

.75

1.08

37.6

80.

08–1

9,00

5.58

–1.3

55,

296.

711.

2812

6.45

0.02

Elig

ible

for A

ge

Pens

ion

297,

243.

563.

44–3

,427

.01

–1.0

246

,892

.15

0.53

–86,

083.

52–3

.38

–254

,625

.19

–5.1

2

Div

orce

d–2

21,7

38.0

5–2

.94

2,27

0.78

0.92

107,

974.

281.

3330

,684

.03

1.32

80,8

08.9

61.

69

Nev

er m

arri

ed–3

,678

.12

–0.0

4–2

,315

.62

–0.7

026

,563

.90

0.22

–56,

463.

84–1

.71

35,8

93.6

70.

54

Fem

ale

–194

,902

.73

–2.5

5–6

,084

.76

–1.8

719

9,37

7.11

2.48

–45,

775.

47–2

.17

47,3

85.8

71.

12

Poor

hea

lth

254,

875.

022.

28–5

,680

.73

–1.9

1–1

20,9

83.8

7–1

.02

–25,

174.

48–0

.78

–103

,035

.92

–1.6

6

Elig

ible

for A

ge

Pens

ion

x po

or h

ealt

h–3

83,9

44.7

8–2

.71

4,05

9.91

1.09

325,

834.

132.

143,

883.

980.

0850

,166

.76

0.58

Net

wor

th–0

.35

–0.8

60.

000.

910.

4819

.91

0.84

5.38

0.03

0.08

n59

559

559

559

559

5

R2

0.08

0.07

0.34

0.26

0.26

Not

es:

Elig

ible

for A

ge P

ensi

on if

at l

east

one

par

tner

is e

ligib

le. P

oor h

ealt

h if

one

mem

ber r

epor

ts b

eing

in p

oor h

ealt

h (s

ee te

xt fo

r pre

cise

defi

niti

on).

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


24


25

Tabl

e 9:

(Av

erag

e) d

eter

min

ants

of a

sset

por

tfol

ios:

cou

ple-

head

ed h

ouse

hold

s (m

argi

nal e

ffect

s an

d t-

stat

istic

s), W

ave

6

Fina

ncia

l wea

lthB

usin

ess

asse

tsO

wn

hom

eLi

fest

yle

Supe

rann

uatio

n

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

Inco

me

($)

Tota

l inc

ome

2.46

1.71

0.25

2.22

–11.

61–4

.81

2.31

4.15

6.58

5.82

Elig

ibili

ty r

ange

–149

,058

.39

–1.1

913

,962

.05

0.60

–8,1

23.2

6–0

.06

32,5

13.5

90.

9111

0,70

6.01

1.37

Dem

ogra

phic

s

Age

42,9

06.2

02.

80–3

,420

.74

–1.7

3–5

,313

.79

–0.3

0–4

,272

.77

–0.9

7–2

9,89

8.90

–2.3

4

Educ

atio

n–6

,453

.72

–0.3

326

7.35

0.13

3,17

0.34

0.16

–6,5

28.6

4–1

.06

9,54

4.66

0.63

Hea

d el

igib

le fo

r Age

Pe

nsio

n–7

1,66

5.33

–0.4

216

,460

.19

0.66

36,7

77.5

50.

20–5

3,89

7.23

–0.7

572

,324

.81

0.43

Spou

se e

ligib

le fo

r A

ge P

ensi

on21

6,20

8.08

1.43

–18,

783.

26–0

.85

–1,0

08.1

2–0

.01

97,9

21.0

91.

60–2

94,3

37.7

8–1

.65

Fem

ale

head

–59,

098.

09–0

.67

–10,

100.

45–0

.82

76,8

78.4

20.

8027

,762

.31

1.01

–35,

442.

19–0

.47

Prev

ious

ly m

arri

ed–1

22,2

33.8

0–1

.36

11,5

17.4

90.

82–5

,468

.03

–0.0

6–1

2,18

9.96

–0.4

212

8,37

4.30

1.54

Poor

hea

lth

–157

,176

.11

–1.3

2–1

1,01

3.81

–0.6

323

6,16

0.92

2.13

–63,

342.

79–1

.83

–4,6

28.2

3–0

.05

Elig

ible

for A

ge

Pens

ion

x po

or h

ealt

h38

1,06

9.34

1.81

–28,

100.

32–1

.12

–232

,626

.09

–0.9

887

,465

.66

1.05

–207

,808

.58

–0.8

9

Spou

se e

ligib

le x

po

or h

ealt

h–1

45,3

15.1

1–0

.77

14,7

43.4

90.

6411

0,44

6.47

0.47

–83,

636.

55–0

.94

103,

761.

710.

38

Net

wor

th–0

.40

–0.7

40.

003.

000.

5320

.49

0.75

1.43

0.12

5.61

n8

618

618

618

618

61

R2

0.06

0.07

0.25

0.19

0.35

Not

es:

Elig

ible

for A

ge P

ensi

on if

at l

east

one

par

tner

is e

ligib

le. P

oor h

ealt

h if

one

mem

ber r

epor

ts b

eing

in p

oor h

ealt

h (s

ee te

xt fo

r pre

cise

defi

niti

on).

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


24


25

Tabl

e 10

: (Av

erag

e) d

eter

min

ants

of a

sset

por

tfol

ios:

sin

gle-

head

ed h

ouse

hold

s (m

argi

nal e

ffect

s an

d t-

stat

istic

s), W

ave

6

Fina

ncia

l wea

lthB

usin

ess

asse

tsO

wn

hom

eLi

fest

yle

Supe

rann

uatio

n

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

Inco

me

($)

Tota

l inc

ome

3.45

1.01

0.24

1.70

–9.3

9–2

.08

0.58

0.53

5.12

5.64

Elig

ibili

ty r

ange

383,

312.

063.

9211

,885

.50

1.04

–441

,710

.00

–3.7

753

,527

.20

1.79

–7,0

14.7

6–0

.09

Dem

ogra

phic

s

Age

34,4

31.2

72.

32–1

06.4

3–0

.29

–3,8

10.3

7–0

.24

–6,2

58.9

6–1

.78

–24,

255.

51–3

.35

Educ

atio

n13

,471

.00

1.07

44.0

00.

09–1

8,49

6.23

–1.3

14,

872.

951.

2010

8.28

0.01

Elig

ible

for A

ge

Pens

ion

–19,

389.

14–0

.12

–2,4

55.9

2–0

.49

81,5

50.6

40.

54–2

8,40

4.18

–0.6

6–3

1,30

1.39

–0.3

9

Div

orce

d–1

80,0

69.2

2–2

.33

2,10

0.05

0.82

99,2

28.7

81.

2127

,090

.13

1.14

51,6

50.2

61.

07

Nev

er m

arri

ed55

,356

.40

0.51

–2,5

40.0

5–0

.77

14,5

43.7

70.

12–6

2,95

5.89

–1.8

4–4

,404

.22

–0.0

7

Fem

ale

–161

,596

.08

–2.0

6–6

,187

.17

–1.8

819

5,83

0.06

2.30

–51,

723.

57–2

.42

23,6

76.7

50.

55

Poor

hea

lth

257,

652.

022.

25–5

,728

.56

–1.8

9–1

25,3

20.0

1–1

.04

–22,

739.

51–0

.70

–103

,863

.94

–1.6

4

Elig

ible

for A

ge

Pens

ion

x po

or h

ealt

h–3

89,7

20.1

6–2

.77

4,11

6.17

1.10

330,

662.

912.

171,

926.

340.

0453

,014

.71

0.61

Net

wor

th–0

.50

–1.1

80.

000.

960.

4920

.48

0.87

5.59

0.15

0.44

n59

559

559

559

559

5

R2

0.09

0.07

0.34

0.26

0.29

Not

es:

Elig

ible

for A

ge P

ensi

on if

at l

east

one

par

tner

is e

ligib

le. P

oor h

ealt

h if

one

mem

ber r

epor

ts b

eing

in p

oor h

ealt

h (s

ee te

xt fo

r pre

cise

defi

niti

on).

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


26

demographicsCouple-headed households in which the head of household is female (those in which the female partner is older) allocate their wealth across asset types in the same way as couple-headed households in which the head of the household is male. This symmetry is robust to model specifications. Single women, however, allocate significantly more wealth than comparable single men to their homes, while holding significantly less financial wealth and lifestyle assets.

A relationship clearly exists between previous marital status and asset portfolios, even after accounting for any potential differences in wealth levels associated with marital history. Single individuals who are divorced (rather than widowed) hold substantially less financial wealth and somewhat more superannuation wealth and lifestyle assets. Singles who have never married allocate their wealth across asset types in much the same way as equally wealthy widowers who have not remarried, except that they hold relatively fewer lifestyle assets. Couple-headed households in which the reference person has been previously married hold less financial wealth and more superannuation wealth. However, once detailed controls for age are added this effect is no longer statistically significant (compare Tables 7 and 9).

The discussion in Section 2 suggests that individuals’ health status might also affect the way in which they allocate wealth. Regardless of model specification, we found that couple-headed households in which at least one member is in poor health have significantly more equity in their homes and less in lifestyle assets than similar couples in which both partners are in good health. It is possible that these differences stem from the incentives for household members with poor health to become eligible for the Age Pension in order to access the associated health care benefits. At the same time, these differences may simply reflect the effects of poor health on households’ optimal asset allocation. This seems to indicate that those in poor health may spend more time at home (and hence allocate more wealth to their homes) and have less use for lifestyle assets. Singles in poor health held more financial wealth and less business assets than singles in good health.

The model also includes an interaction term that separately identifies those singles who are both in poor health and who have reached pension eligibility age. In the case of couples, we interacted poor health status (specifically, at least one partner reporting poor health) with the pension eligibility indicator for each partner. These interactions allowed us to distinguish the asset portfolios of those who had reached pension age in good health from those who had reached pension age in poor health. This distinction provides some evidence on whether the health care benefits associated with the Age Pension seem to be leading people in poor health (and who presumably most value these additional health care benefits) to hold their wealth differently to similar people in good health.

The results indicate that couples in which at least one partner is in poor health and in which the head of the household has reached pensionable age hold significantly more financial wealth than equally wealthy couples in which both partners are in good health. Singles who are over pension age and in poor health hold significantly less financial wealth and more equity in their own homes than similar single individuals who are in good health. This latter effect is consistent with the rules that exclude home equity from the asset test when determining Age Pension eligibility. It is less clear whether the income and asset tests underlying the Age Pension give couples in poor health an incentive to hold more financial wealth.


27

Age and age eligibility

All income and demographic results discussed above are remarkably robust to model specification. Our substantive conclusions remain the same, even if they include detailed controls for age in the model. Despite this, we found that allocation of wealth across asset types is nonetheless clearly related to age. This relationship is, however, difficult to interpret. Given the cross-sectional nature of the analysis, we cannot explicitly control for birth cohorts. As a result, the estimated age effect on the level of any particular asset captures both differences across birth cohorts in allocation of assets as well as any effect due to ageing (lifecycle stages).

Consider first the baseline specification that simply controls the effects of having reached pension eligibility age (see Tables 7 and 8). Among couples, the indicator for head of household eligibility captures the effect of the oldest partner being over pension age, while the spouse eligibility indicator measures the additional effect of the spouse also being over pension age. Using this specification we found some evidence that couples may hold more financial wealth and less superannuation wealth once both partners become eligible to claim the Age Pension. Singles who have reached pension age hold more financial wealth, but less wealth in superannuation or lifestyle assets.

However, these findings are not robust to model specification (see Tables 8 and 10 for singles and Tables 7 and 9 for couples). Once we controlled for the overall effects of ageing,19 we found that, not surprisingly, there was a relationship between a household’s age and the way in which it allocates its assets.20 However, the effect of reaching pension eligibility age completely disappears. Thus, there is no additional effect of reaching pension age on portfolio allocations. These results—based on Wave 6 of HILDA—suggest that the disparity in the asset portfolios of younger and older households stem from lifecycle changes (that is, ageing) rather than from changes associated specifically with reaching pension eligibility age.

However, results based on Wave 2 of HILDA (see Tables A3 and A4) do indicate some independent effect of reaching pension age on portfolio allocations. In particular, couples in which both partners are eligible for the Age Pension hold significantly more financial wealth and home equity and significantly less in all other assets than do couples in which only the head has reached pension age. Singles who have reached pension age hold significantly more wealth in their own homes than do other singles. This is consistent with the Age Pension having had some effect on asset portfolios in 2002.

longitudinal evidence

All results discussed so far have been derived using cross-sectional variation in the data. The HILDA data are, however, longitudinal, which opens up the possibility of assessing changes in asset portfolios for the same households over time. The main limitation of such analysis is the small sample size available. There were 539 couple-headed households that did not change household type and reported wealth information in both Waves 2 and 6 of HILDA. There were 334 single-headed households that did not change household type and provided wealth data in both waves. Unfortunately, these sample sizes are too small for the type of simultaneous


28

regression analysis conducted above. However, for each household type, we separately identify those households in which at least one member had become eligible for the Age Pension and those in which there was no change in eligibility between the two waves. Comparison of those households that had become eligible with those that had not provides a crude estimate of the potential effect of reaching pension age on asset allocation.

Table 11 presents the average change in asset levels between 2006 and 2002 for those households present in both HILDA waves. Among couples, we found a real increase in all assets except business equity, irrespective of pension eligibility status. However, we did not find any statistically significant differences in the magnitude of these changes between those households that had become eligible for the Age Pension and those that had not (see p-values, Table 11). The same result held for singles, with the exception that levels of financial wealth appear to have increased more among households that had become eligible for the Age Pension.

Table 11: Changes in assets holding by change in eligibility to Age Pension

Couples SinglesChange in eligibility Change in eligibility

Yes ($) No ($) p-value Yes ($) No ($) p-valueBusiness –21,718 –12,507 0.615 –8,412 1,031 0.569Financial wealth 28,231 19,369 0.836 35,859 –10,095 0.028Lifestyle 38,135 34,237 0.886 32,712 11,912 0.557Own home 70,088 94,646 0.471 61,264 56,864 0.723Superannuation 39,918 38,079 0.943 16,675 17,370 0.968Wealth 154,653 173,824 0.708 138,098 77,083 0.146n 133 406 90 254

Notes: Authors’ calculation based on Waves 2 and 6 of HILDA data. All figures are reported in constant 2006 Australian dollars.

Finally, Figures 9, 10, 11 and 12 present the portfolio allocations for the subset of couple-headed and single-headed households that had become eligible to receive the Age Pension (between 2002 and 2006) both before and after becoming eligible. These figures do not reveal any significant changes in asset shares after becoming eligible to claim the Age Pension. In particular, the share in own home remains unchanged—40 per cent among couples and 43 per cent among singles.

Taken together, these crude longitudinal comparisons seem to corroborate the main findings from the cross-sectional analysis of 2006 HILDA data that the asset and income tests underlying the Age Pension do not seem to trigger substantial changes in the portfolio choice of Australian households.


29

Figure 9: Couple-headed households, before eligibility, 2002

Financial wealth

Business

Own home

Lifestyle

Superannuation 18%

5%

39%

13%

25%


Figure 10: Couple-headed households, after eligibility, 2006

Financial wealth

Business

Own home

Lifestyle

Superannuation18%

2%

40%

15%

25%



30

Figure 11: Single-headed households, before eligibility, 2002

Financial wealth

Business

Own home

Lifestyle

Superannuation 19%

4%

43%

10%

24%


Figure 12: Single-headed households, after eligibility, 2006

Financial wealth

Business

Own home

Lifestyle

Superannuation

21%

1%

43%

13%

21%



31

6 ConclusionsThis paper has examined the wealth levels and asset portfolios of Australian households in an effort to determine whether there is evidence that the income and asset tests underlying eligibility for the Age Pension lead households to reallocate their assets. The question has been examined using descriptive analysis (both cross-sectional and longitudinal) as well as simultaneous-equation regression analysis.

The simple descriptive statistics presented in this paper suggest that, irrespective of year, younger Australian households (that is, those in which the reference person is aged 55 to 64 years) have more wealth on average than those households that are approximately 10 years older. This is consistent with households spending-down wealth in order to finance post-retirement consumption. There was a substantial increase in the amount of wealth both younger and older households held in superannuation assets and owner-occupied housing between 2002 and 2006. The result is that assets in the form of owner-occupied housing remained high even among households over the pension age. This finding is not surprising given the preferential tax treatment of owner-occupied housing in Australia and its pivotal role in establishing eligibility to receive the Age Pension under the asset test. Moreover, we would expect that the incentives to qualify for the Age Pension benefits (and its associated health care benefits) would result in households either decreasing their holdings of non-housing assets or reallocating wealth towards assets that do not generate future income streams. As Barrett and Tseng (2008) point out, the fact that many households over the pension age still hold substantial superannuation assets—instead of converting them into a secure income stream—might be crude evidence that some behavioural responses to the incentives to qualify for the Age Pension are taking place.

The regression analysis results indicate that having an income that is within 10 per cent of the relevant income eligibility threshold for the Age Pension has no effect on the asset allocation of couples. However, single-headed households that are within the income eligibility range hold significantly less wealth in their homes and more in lifestyle assets. While the latter is consistent with the incentives inherent in the Age Pension eligibility rules, the former is not.

We also found that single-headed households that are over pension age and in poor health hold significantly less financial wealth and more equity in their own homes than similar single individuals who are in good health. This is consistent with the rules that exclude home equity from the asset test when determining Age Pension eligibility.

At the same time, using Wave 6 HILDA data, no evidence was found that reaching pension age is associated with reallocation of household assets once the effects of ageing are taken into account. In other words, the disparity in the asset portfolios of younger and older households appears to stem from lifecycle changes (that is, ageing) rather than from changes associated specifically with reaching pension eligibility age. However, parallel results based on Wave 2 of HILDA data do indicate some independent effect of reaching pension age on portfolio allocations. Specifically, couples in which both partners are eligible for the Age Pension hold significantly more financial wealth and home equity and significantly less of all other assets than do couples in


32

which only the head has reached pension age. Single individuals who have reached pension age hold significantly more wealth in their own homes than do other singles. Again, the increase in home equity is consistent with the incentive to qualify for the Age Pension, while the disparity in other assets is less clear-cut.

Finally, simple longitudinal comparisons between 2002 and 2006 do not provide any evidence that the asset and income tests underlying the Age Pension result in significant changes in the portfolio choices of Australian households.

Taken together, these results do not provide compelling evidence that on average households respond to the incentives embedded in the asset and income tests used to determine Age Pension eligibility by reallocating their assets. While there are some significant differences in asset portfolios associated with having an income near the income threshold, being of pensionable age, and being in poor health, these differences are often only marginally significant, not robust across HILDA waves, and not clearly consistent with the incentives inherent in the Age Pension eligibility rules.

It is important to note several caveats. First, some households may adjust their financial situation in order to ensure they are eligible for the Age Pension. By its very nature, however, this analysis was concerned with the behaviour of households in the aggregate (or on average). We did not find evidence showing large disparities in the asset portfolios of large numbers of Australia households that are consistent with Age Pension eligibility. Second, while we believe that the HILDA data provide the best opportunity to address the research question of interest, it is not ideal. In particular, small sample sizes make it impossible to estimate a model of asset allocation using longitudinal variation in asset portfolios. Estimation of asset determination is complex and, in this case, involved simultaneous estimation of five separate asset equations imposing cross-equation restrictions in order to ensure the sum of assets equals total net wealth. Estimating such a model in a longitudinal context would require far more data than are likely to ever be available in a panel survey like HILDA, which is representative of the entire population. More progress would be made using data, such as that from the US Health and Retirement Survey, which specifically sample older cohorts.


33

Appe

ndix

Tabl

e A1

: (A

vera

ge) d

eter

min

ants

of a

sset

por

tfol

ios:

cou

ple-

head

ed h

ouse

hold

s (m

argi

nal e

ffect

s an

d t-

stat

istic

s), W

ave

2

Fina

ncia

l wea

lthB

usin

ess

asse

tsO

wn

hom

eLi

fest

yle

Supe

rann

uatio

n

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

Inco

me

($)

Tota

l inc

ome

–7.5

9–6

.25

–0.2

8–1

.23

11.1

910

.70

–1.1

8–3

.41

–2.1

3–2

.76

Elig

ibili

ty r

ange

83,5

07.7

60.

7831

,314

.63

0.85

–66,

311.

56–0

.59

66,1

01.9

42.

28–1

14,6

12.7

7–1

.20

Dem

ogra

phic

s

Educ

atio

n15

,032

.82

1.19

6,21

9.63

2.50

–62,

646.

32–5

.35

1,60

0.19

0.34

39,7

93.6

94.

00

Hea

d el

igib

le fo

r A

ge P

ensi

on23

3,38

7.95

2.84

–40,

417.

05–1

.36

299,

604.

062.

50–1

4,47

0.28

–0.4

5–4

78,1

04.6

9–4

.83

Spou

se e

ligib

le fo

r A

ge P

ensi

on29

5,83

6.50

3.48

–65,

816.

23–2

.24

356,

209.

912.

64–6

8,63

2.11

–1.9

7–5

17,5

98.0

9–4

.25

Fem

ale

head

–135

,329

.09

–1.5

312

,614

.93

0.67

98,0

35.0

50.

9931

,827

.83

1.03

–7,1

48.7

1–0

.11

Prev

ious

ly m

arri

ed–1

56,1

53.2

8–2

.22

28,1

40.5

61.

41–1

95,2

82.7

7–2

.32

12,0

52.8

40.

4931

1,24

2.66

4.68

Poor

hea

lth

–184

,796

.83

–1.9

0–8

2,84

5.14

–3.9

061

4,62

8.25

5.91

–65,

527.

38–1

.72

–281

,458

.91

–4.4

5

Elig

ible

for A

ge

Pens

ion

x po

or h

ealt

h7,

653.

020.

0520

,308

.47

0.58

–7,8

92.8

0–0

.05

–58,

161.

18–1

.01

38,0

92.4

90.

24

Spou

se e

ligib

le x

po

or h

ealt

h19

,790

.82

0.11

62,8

95.4

51.

83–4

17,1

40.7

5–2

.16

124,

794.

232.

1020

9,66

0.23

1.06

Net

wor

th0.

205.

310.

002.

000.

6415

.34

0.06

5.40

0.09

5.54

n92

192

192

192

192

1

R2

0.09

0.06

0.24

0.05

0.25

Not

es:

Elig

ible

for A

ge P

ensi

on if

at l

east

one

par

tner

is e

ligib

le. P

oor h

ealt

h if

one

mem

ber r

epor

ts b

eing

in p

oor h

ealt

h (s

ee te

xt fo

r pre

cise

defi

niti

on).

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


34


35

Tabl

e A2

: (A

vera

ge) d

eter

min

ants

of a

sset

por

tfol

ios:

sin

gle-

head

ed h

ouse

hold

s (m

argi

nal e

ffect

s an

d t-

stat

istic

s), W

ave

2

Fina

ncia

l wea

lthB

usin

ess

asse

tsO

wn

hom

eLi

fest

yle

Supe

rann

uatio

n

dy/d

x t-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

Inco

me

($)

Tota

l inc

ome

–7.2

6–8

.73

1.11

1.26

2.69

2.13

0.45

1.15

3.01

2.85

Elig

ibili

ty r

ange

–20,

371.

17–0

.20

9,52

2.74

0.52

–61,

562.

60–0

.61

49,3

23.9

01.

8523

,087

.13

0.38

Dem

ogra

phic

s

Educ

atio

n14

,296

.42

1.61

–873

.44

–0.7

6–2

1,44

7.46

–2.3

11,

852.

110.

636,

172.

371.

24

Elig

ible

for A

ge

Pens

ion

53,5

54.4

40.

90–1

5,94

9.57

–2.0

922

7,41

3.19

3.91

–29,

681.

78–1

.66

–235

,336

.28

–7.2

1

Div

orce

d–5

7,42

3.04

–1.0

4–3

,824

.95

–0.6

2–1

4,41

3.67

–0.2

817

,024

.04

1.05

58,6

37.6

12.

02

Nev

er m

arri

ed–2

4,48

7.14

–0.3

0–1

3,50

7.88

–1.6

313

0,20

0.73

1.45

–54,

702.

34–2

.05

–37,

503.

37–0

.93

Fem

ale

–105

,873

.15

–1.7

8–1

8,71

2.71

–2.9

111

9,06

3.05

1.92

–15,

352.

01–0

.95

20,8

74.8

10.

74

Poor

hea

lth

28,6

10.6

20.

326,

698.

700.

5546

,363

.15

0.50

24,0

04.3

20.

97–1

05,6

76.7

9–2

.53

Elig

ible

for A

ge

Pens

ion

x po

or h

ealt

h–6

3,09

1.35

–0.6

0–4

,306

.57

–0.3

5–2

1,44

5.01

–0.2

0–2

2,07

5.73

–0.7

011

0,91

8.66

2.00

Net

wor

th0.

3816

.46

0.00

2.72

0.46

27.4

30.

099.

230.

077.

44

n57

757

757

757

757

7

R2

0.17

0.09

0.34

0.17

0.29

Not

e:

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


34


35

Tabl

e A3

: (A

vera

ge) d

eter

min

ants

of a

sset

por

tfol

ios:

cou

ple-

head

ed h

ouse

hold

s (m

argi

nal e

ffect

s an

d t-

stat

istic

s), W

ave

2

Fina

ncia

l wea

lthB

usin

ess

asse

tsO

wn

hom

eLi

fest

yle

Supe

rann

uatio

n

dy/d

x t-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

Inco

me

($)

Tota

l inc

ome

–6.6

4–5

.41

–0.3

0–1

.30

11.0

410

.79

–1.2

7–3

.59

–2.8

3–3

.59

Elig

ibili

ty r

ange

95,3

80.9

10.

9230

,370

.63

0.83

–60,

899.

69–0

.54

64,8

71.2

52.

21–1

29,7

23.1

0–1

.34

Dem

ogra

phic

s

Age

48,2

25.0

64.

11–2

,732

.47

–0.9

715

,896

.34

1.20

–5,6

45.6

4–1

.25

–55,

743.

30–5

.76

Educ

atio

n17

,159

.27

1.36

5,99

0.60

2.41

–60,

710.

17–5

.25

1,31

9.31

0.28

36,2

40.9

93.

70

Hea

d el

igib

le fo

r A

ge P

ensi

on–1

53,7

66.4

7–1

.31

–18,

541.

36–0

.49

169,

365.

301.

1232

,562

.81

0.65

–29,

620.

29–0

.23

Spou

se e

ligib

le fo

r A

ge P

ensi

on18

3,97

5.41

1.93

–55,

968.

68–1

.75

289,

287.

382.

02–6

3,90

9.54

–1.6

3–3

53,3

84.5

6–2

.72

Fem

ale

head

–136

,404

.89

–1.5

712

,383

.11

0.66

102,

187.

851.

0332

,444

.28

1.04

–10,

610.

35–0

.17

Prev

ious

ly m

arri

ed–7

4,29

6.80

–1.0

022

,477

.15

1.10

–159

,669

.81

–1.8

14,

802.

410.

1820

6,68

7.05

2.97

Poor

hea

lth

–181

,211

.92

–1.8

6–8

2,83

7.37

–3.9

061

3,40

3.69

5.87

–64,

593.

87–1

.69

–284

,760

.50

–4.5

3

Elig

ible

for A

ge

Pens

ion

x po

or h

ealt

h28

,047

.56

0.17

18,6

46.5

90.

535,

720.

670.

03–6

0,00

0.70

–1.0

37,

585.

880.

05

Spou

se e

ligib

le x

po

or h

ealt

h–8

,271

.02

–0.0

565

,367

.16

1.92

–435

,673

.06

–2.2

312

5,27

9.58

2.10

253,

297.

341.

29

Net

wor

th0.

205.

290.

002.

180.

6315

.65

0.06

5.51

0.10

6.06

n92

192

192

192

192

1

R2

0.13

0.06

0.24

0.05

0.25

Not

es:

Elig

ible

to A

ge P

ensi

on if

at l

east

one

par

tner

is e

ligib

le. P

oor h

ealt

h if

one

mem

ber r

epor

ts b

eing

in p

oor h

ealt

h (s

ee te

xt fo

r pre

cise

defi

niti

on).

All

figur

es a

re re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


36


37

Tabl

e A4

: (A

vera

ge) d

eter

min

ants

of a

sset

por

tfol

ios:

sin

gle-

head

ed h

ouse

hold

s (m

argi

nal e

ffect

s an

d t-

stat

istic

s), W

ave

2

Fina

ncia

l wea

lthB

usin

ess

asse

tsO

wn

hom

eLi

fest

yle

Supe

rann

uatio

n

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

dy/d

xt-

stat

Inco

me

($)

Tota

l inc

ome

–7.1

2–8

.76

0.90

1.06

3.06

2.47

0.45

1.14

2.70

2.58

Elig

ibili

ty r

ange

–13,

890.

16–0

.14

9,61

7.73

0.50

–62,

651.

11–0

.62

48,2

72.5

81.

7818

,650

.95

0.29

Dem

ogra

phic

s

Age

18,3

94.5

22.

36–2

,112

.07

–1.6

91,

269.

830.

16–2

,011

.10

–0.7

6–1

5,54

1.19

–3.3

1

Educ

atio

n15

,653

.64

1.78

–786

.10

–0.6

8–2

1,52

9.81

–2.3

11,

545.

430.

525,

116.

841.

07

Elig

ible

for A

ge

Pens

ion

–126

,801

.80

–1.2

57,

507.

860.

5621

7,51

8.66

2.16

–12,

641.

27–0

.39

–85,

583.

45–1

.47

Div

orce

d–2

8,37

6.68

–0.5

1–5

,592

.91

–0.8

2–1

1,86

1.32

–0.2

312

,549

.09

0.75

33,2

81.8

31.

10

Nev

er m

arri

ed–9

,183

.29

–0.1

1–1

5,54

3.31

–1.8

113

1,60

8.27

1.46

–56,

217.

93–2

.12

–50,

663.

73–1

.22

Fem

ale

–84,

323.

42–1

.41

–22,

310.

79–3

.00

120,

543.

961.

94–1

6,82

8.92

–0.9

92,

919.

170.

10

Poor

hea

lth

20,6

99.8

80.

238,

815.

500.

7547

,458

.00

0.51

23,6

12.1

00.

93–1

00,5

85.4

8–2

.42

Elig

ible

for A

ge

Pens

ion

x po

or h

ealt

h–6

1,22

0.58

–0.5

8–6

,611

.65

–0.5

5–2

2,36

8.69

–0.2

1–2

0,46

8.89

–0.6

411

0,66

9.81

2.01

Net

wor

th0.

3816

.39

0.00

2.76

0.46

27.3

00.

099.

260.

077.

49

n57

757

757

757

757

7

R20.

180.

110.

340.

180.

31

Not

e:

All

figur

es re

port

ed in

con

stan

t 200

6 A

ustr

alia

n do

llars

.


36


37

Endnotes1 There are exceptions to this general rule for refugees, newcomers under special programs, or

those widowed in Australia not meeting the 10-year residency requirement.

2 At the time of collection of the data used in the regression analysis discussed in this report, all females aged 63 years and over were eligible to claim the Age Pension.

3 A partnered recipient may be eligible to receive the single rate if the couple is separated because of illness (in this case, the combined assets of the couple are still used to calculate rates) or if his or her partner, who does not receive a pension, is jailed or in a psychiatric hospital.

4 If the title to the claimant’s principal residence includes less than two hectares of land, the land must be used primarily for private and domestic purposes in order for it to be included in the assets test exemption. If the title includes more than two hectares of land, the claimant must have an attachment to the land of 20 consecutive years and be using the land (if productive) to the best of their abilities so as to generate an income. Income generated from productive land is assessed under the income test if it accrues directly to the pensioner. It is not assessed if it accrues to a family member working the land.

5 Major changes to the asset test rules were introduced in September 2007. In particular, the level of pension benefits were reduced by $1.50 per fortnight for every $1,000 assets above the disregard levels.

6 Alternative recently available micro-level data on wealth include the Household Expenditure Survey and the Survey of Income and Housing. These data are not, however, longitudinal.

7 Couple-headed households include both married and cohabiting couples.

8 See Headey (2003) for a detailed discussion of wealth measurement in HILDA.

9 This paper considers the total value of all vehicles, not vehicle equity because the amount of any car loans is combined with other debts (such as other loans, hire purchase or overdraft) in the HILDA survey making it impossible to derive a measure of vehicle equity.

10 Test results are not reported in the tables but are available upon request.

11 These differences across health status are both economically meaningful (that is, relatively large) and statistically significant. Test results are not reported in the tables but are available upon request.

12 These differences are statistically significant. All 2002 figures are expressed in 2006 dollars. The ABS CPI quarterly number for September was used as deflator.

13 However, it is likely to change when all members in the 55 to 64 year-old cohort in 2006 reach pension age.


38

14 As previously mentioned, this variation could also be driven by the strong house price growth that occurred between 2002 and 2006.

15 The reported specification assumes that a household is in the range of eligibility when total household gross income is +/– 10 per cent of the relevant eligibility threshold.

16 Specifically, the estimated marginal effect of an additional dollar of wealth are required to sum to one across asset types, while the marginal effect of a change in any other independent variable is restricted to sum to zero. Note that while these constraints hold on average, they may not hold for any particular couple.

17 Marginal effects are calculated for each individual and then averaged over the relevant sub-sample using the sample weights (see Greene 1997, p. 876). Bootstrapped standard errors (with 500 replications) are used to calculate the reported t-statistics. Following standard practice, on each bootstrap sample, we first generated our measure of permanent/transitory income, and then estimated our model of net wealth.

18 The following discussion concentrates those results, which are statistically significant, that is, those which we can be reasonably confident do not result from random chance. Results are significant at the 5 per cent level (in a two-tailed test) when the t-statistic (reported in all regression tables) exceeds 1.96. Results are significant at the 10 per cent level (in a two-tailed test) when the t-statistic exceeds 1.65.

19 This is done through a quadratic in age. The marginal effect of age reported in Tables 9 and 10 accounts for both terms in the quadratic. Accounting for age through a cubic and quartic resulted in substantially the same results.

20 The exception is that couples in which both partners are eligible for the Age Pension hold less superannuation wealth than couples in which only the head is eligible. This effect is, however, only marginally significant at 10 per cent.

ReferencesBarrett, G & Tseng, YP 2008, ‘Retirement saving in Australia’, Canadian Public Policy, vol. 34, supplement 1, pp. S177–93.

Blau, FD & Graham, JW 1990, ‘Black–white differences in wealth and asset composition’, Quarterly Journal of Economics, vol. 105, no. 2, pp. 321–39.

Cobb-Clark, DA & Hildebrand, VA 2006a, ‘The wealth and asset holdings of US-born and foreign-born households: evidence from SIPP data’, Review of Income and Wealth, vol. 52, no. 1,pp. 17–42.

——2006b, ‘The portfolio choices of Hispanic couples’, Social Science Quarterly, vol. 87, no. 5, pp. 1344–63.


39

——2009, ‘The asset portfolios of native born and foreign born Australian households’, Economic Record, vol. 85, no. 268, pp. 46–59.

Greene, WH 1997, Econometric analysis, 3rd edn, Prentice-Hall Inc., Upper Saddle.

Headey, B 2003, Income and wealth—facilitating multiple approaches to measurement and permitting different levels of aggregation, HILDA Project Discussion Paper Series, 3/03.

Headey, B, Marks, G & Wooden, M 2005, ‘The structure and distribution of household wealth in Australia’, Australian Economic Review, vol. 38, no. 2, pp. 159–75.

Wooden, M, Freidin, S & Watson, N 2002, ‘The Household, Income and Labour Dynamics in Australia (HILDA) survey: Wave 1’, Australian Economic Review, vol. 35, no 3, pp. 339–48.


40

41

Changes in household expenditure associated with the arrival of newborn childrenJason D Brandrup & Paula L Mance

Research and Analysis Branch, Department of Families, Housing, Community Services and Indigenous Affairs

Acknowledgements

This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) survey. The HILDA project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (MIAESR). The opinions, comments and/or analysis expressed in this document are those of the authors and do not necessarily represent the views of the Minister for Families, Housing, Community Services and Indigenous Affairs or the MIAESR and cannot be taken in any way as expressions of Government policy.

We thank our Departmental colleagues for their comments and/or assistance and Dr Tue Gørgens (Australian National University) for his statistical advice during the early stages of this study.

AbstractAn understanding of the changed financial circumstances of families with newborn children is important to a range of current policy debates, including those surrounding the provision of family assistance, women’s attachment to the labour force and paid parental leave. Although there is a body of Australian research on the costs of raising children, in most cases this has been undertaken to enable the calculation of child support entitlement or to evaluate the effects of policy designed to reverse the effects of an ageing demographic. These studies do not report specifically on expenses associated with the arrival of newborn children.

To address this gap in the evidence base, the current study investigates changes in household expenditure associated with the arrival of newborn children for three groups of families—those experiencing the arrival of their first, second, or third and subsequent-born children. Household spending items in Waves 6 and 7 (2006 and 2007) of the Household, Income and Labour Dynamics in Australia (HILDA) survey are used to estimate whether different categories of expenditure typically increase or decrease for couple families with the arrival of newborn children. This study shows that a range of expenditure categories are influenced by


42

the arrival of a new baby. Parents of first-born children increase expenditure on health care and clothing. Parents of second-born children increase expenditure on health care, and on meals eaten out and takeaway; however, they decrease expenditure on child care. Parents of third and subsequent-born children increase expenditure on health care.

Keywords: household expenditure; child costs; newborn children, family policy, HILDA

1 IntroductionExplicit in many family policies is the recognition of additional expenses incurred by families raising children. Many family assistance and income support programs have child-specific payments and in a number of cases higher income thresholds apply to these payments to cater for the presence of one or more children. Some payments also provide different rates of assistance for children of younger and older ages, acknowledging that children place different demands on the family budget as they progress through each life stage.

There is a body of Australian and international research that has provided broad evidence to underpin policy formulation and evaluation in this field. The existing studies have mainly focused on estimating the costs of raising children in broad age ranges (that is, under school age, school age and adolescence) and examining how families allocate their household budget when children are present. In the Australian context, research has largely centred on examining the policy issues of child support and demographic ageing. For instance, studies conducted in the period leading up to the reforms of the Child Support Scheme in 2005 concentrated on estimating the costs of children to enable the calculation of child support entitlement (Gray 2007; Henman 2007; Percival & Harding 2007). Studies concerned with compensating for the effects of demographic ageing generally focused on evaluating the effects of government payments—such as the Baby Bonus1—on increasing Australia’s fertility rate (for example, see Drago et al. 2009). However, to our knowledge, no Australian studies specifically focus on analysing changes to family expenditure patterns associated with the birth of a child.

The current study exploits the rich panel data in the Household, Income and Labour Dynamics in Australia (HILDA) survey to examine family expenditure patterns for three types of couple families who experienced the birth of a child between 2006 and 2007—those experiencing the birth of a first child, second child or third and subsequent child. In the first group, the research compares expenditure for the same families before and after they had children. In the other two groups, the effect of additional children can be examined.

The strength of this study lies in the methodology applied and the longitudinal nature of information collected in the HILDA survey. Fixed effects linear regressions are used to analyse the impact of the arrival of newborn children, while controlling for a range of time-varying independent variables. In using this methodology we are able to exploit the potential of the longitudinal data set to reduce omitted variable bias and generate estimates that are closer to the true underlying effect.

Increasing our knowledge of the demands on, and choices made by, parents of newborns is particularly important in the current policy climate. In May 2009 the government announced that

chANgeS IN houSehold exPeNdIture ASSocIAted WIth the ArrIvAl of NeWBorN chIldreN

43

it will introduce a Paid Parental Leave scheme for new parents who are the primary carers of a child born or adopted on or after 1 January 2011.2 The scheme will prove financial assistance to parents of newborns with the dual aims of encouraging women to maintain their connection with the workforce to help prepare Australia for the challenges of an ageing population, and enabling more parents to stay at home to care for their baby full-time during the vital early months of social, cognitive and physical development.3 Although this study will not provide an estimate of total costs associated with the birth of a child, and consequently the level of financial assistance required by new parents, it will contribute empirical evidence on changes in parental expenditure behaviour occurring in a discrete window of time around the birth of a child.

2 What influences spending patterns for families with children?

It is reasonable to assume that when households increase in size, there will be increased costs associated with additional consumers. However, prior studies suggest that the magnitude of these costs are not objectively based but vary according to the tastes, preferences and the amount of money that parents have available to spend on their children (McDonald 1990). In most cases, expenditure on children will need to be offset by reductions in expenditure on other household items. As such, we expect that when adults become parents, they will reallocate their financial resources to provide adequate food, clothing, shelter and education to their children.

However, it is likely that the relationship between expenditure and income pre and post-children may not be as straightforward as reallocating resources. For instance, there may be changes in total household income coinciding with the birth of a child and, as a result, the amount of money available to spend. On one hand, income may decrease temporarily or permanently as mothers withdraw from the workforce to care for the new baby. On the other hand, family income may increase due to rises in fathers’ earnings due to career advancement or longer working hours, or eligibility for government payments, particularly family assistance.4 Parents may also meet additional demands for expenditure through consumer credit.

A number of studies have also found sociodemographic factors that influence how families allocate their resources, including the ages of children, marital status, geographic location, income level of parents and number of other children and adults in the family (Exter 1992; Percival & Harding 2007; Valenzuela 1999). For example, a couple whose oldest child is aged less than 6 years has been estimated to spend over 10 per cent more in annual expenditures than the average couple without children, while a couple whose oldest child is aged 6 to 17 years spends 24 per cent more (Exter 1992). While the total costs of raising children increase as the number of children increases, the average cost per child falls, noting that the relationship between family size and expenditure on children is not linear and varies according to family income (Percival & Harding 2007). When considered together, these studies indicate that, despite economies of scale being evident as the number of children increase, for most families, income necessarily constrains expenditure on children by determining the upper limit on the amount the family has available to spend (McDonald 1990; Percival & Harding 2007).


44

From these studies it follows that when the patterns of expenditure for individual items are examined, expenditure is likely to vary depending on a range of demographic and contextual factors. For example, expenditure on housing may increase prior to the birth of a first child as a formerly childless couple purchase a family home, but transport costs may increase with the addition of subsequent children due to upsizing the family car and as older children start school.

Although Australian studies have not specifically focused on demographic factors that influence families’ expenditure on particular items in the period following the birth of a child, a number of Australian and overseas studies contribute to our understanding of this issue more broadly. These studies are reviewed below, with particular attention to the relationship between demographic factors and resource allocation, and how these factors might influence expenditure on items collected in the HILDA survey for families with newborn children.

Groceries: Australian and overseas studies consistently report that food expenditure comprises the largest share of total expenditure, followed by housing and transport (Percival et al. 2007; Valenzuela 1999). Together, these three expenditures have been estimated to make up between 60 and 73 per cent of subsistence, or basic, expenditures of a typical Australian household (Valenzuela 1999). The contribution of each expenditure item depends on family composition and sociodemographic factors. For instance, housing expenditure has been reported to exceed food expenditure for Australian couple families with young children (ABS 2006), while for low-income American families, work status predicted whether families expended a greater proportion of income on food and housing as against transportation, personal insurance and retirement pensions (Passero 1996).5

Prevailing evidence suggests that children are ‘food intensive’ and economies of scale do not apply to food consumption in the same way as they might apply to other essential items. For example, on average, it has been estimated that one child will increase food requirements by 22 per cent, while two children will increase food expenditure by 44 per cent, when compared to food expenditure by families with no children (Valenzuela 1999). Existing studies also find evidence of income constraints, with food expenditure consuming a larger proportion of family income for lower-income families than high-income families (ABS 2006; Valenzuela 1999). These studies suggest that as most food expenditure is a necessity, rather than a luxury, it is difficult for families to make significant savings below a threshold level.

Meals eaten out and takeaway: Expenditure on takeaway or food consumed away from home has been examined in several overseas studies, with a number of socioeconomic factors and income constraints evident in determining expenditure. Studies from the United States (US) cite the relevance of maternal labour force participation, marital status and education, the presence of children and ethnicity (Fan & Zuiker 1998; Kaushal, Gao & Waldfogel 2007), while in a Korean study, lifecycle stage and cultural values influenced families to spend more on education, which constrained their budget and limited expenditure on food away from home (Lee & Huh 2004). In the Australian context, Australian Bureau of Statistics (ABS) (2006) statistics show increasing takeaway expenditure associated with increasing income quintile.

Clothing: Previous studies observe increased spending on clothing as a proportion of total household expenditure, as household expenditure increases overall (Percival & Harding 2007). Although a basic level of clothing may be considered essential, many clothing expenditures


45

are purchases of ‘comfort’ goods, rather than necessities (Carlucci & Zelli 1998). It follows that wealthy families have more scope to increase their expenditure on clothing over and above what could be considered as a necessity when compared to basics such as groceries, where there is an upper limit on the amount of grocery items a household can consume (Percival & Harding 2007).

Expenditure on adult and children’s clothing may also be influenced by factors apart from income. For example, increasing expenditure on adult clothing has been associated with increasing labour force participation (Kaushal, Gao & Waldfogel 2007), while decreasing expenditure on children’s clothing has been associated with increasing family size as clothing is passed down from older to younger children (Valenzuela 1999).

Child care: Expenditure on child care is influenced by a range of contextual factors, including mother’s employment hours, the number of children in care, service accessibility, the availability of relatives and partners, and cost (ABS 2008; Doiron & Kalb 2005; Walker & Reschke 2004). The cost of child care also largely depends on the child care setting. Most formal care6 involves a cost to parents whereas informal care, provided predominantly by relatives and friends, is mostly provided at low or zero cost. Expenditure on child care is also higher for children under school age and high-income families, generally reflecting longer average hours in care (ABS 2008).

Although overseas researchers have examined factors influencing child care expenditure, it is difficult to compare studies across jurisdictions due to differing levels of government support and parental preferences for informal and formal care. For example, in contrast to expectations, Kaushal, Gao and Waldfogel (2007) found there was no corresponding increase in child care costs associated with increasing labour force attachment. This was most likely to be due to the expansion of child care subsidies in the US during the 1990s, when their study was undertaken, and reliance on relatives or other informal sources of child care.

Education: Intuitively it might be expected that there would be no increased education expenditure associated with the arrival of a newborn child, although parents may reduce their own participation in education due to time pressures associated with caring for a new child. In support of this proposition, in a related study, US researchers did not find significant differences in education expenditure for couples with and without children (Exter 1992). However, expenditure differences associated with older children, family income, marital status and type of education have been observed (ABS 2006; Exter 1992).

Health: In Australia the Medicare system meets a large proportion of basic health care costs7 and as such it is not surprising that studies find health expenditure comprising a small proportion of total family expenditure overall (Valenzuela 1999). Health expenditure has been reported to increase with rising income quintile (ABS 2006) and, contrary to expectations, decrease with the presence of children (Exter 1992). However, many studies compare couples with and without children and thus the higher health expenditure for childless couples reflects their older average ages. When households with children are compared, rising health expenditures are associated with increasing income, family size and couple’s marital status (ABS 2008; Valenzuela 1999).

Transport: The costs of transport may impose a significant burden on the household budget. Transportation has been estimated as the second-largest spending category for American families (Exter 1992), and the third-largest spending category for Australian families (Valenzuela


46

1999). Among Australian households, car ownership is a major influencing factor for transport expenditure (Saunders et al. 1998), with expenditure costs reflecting the purchase price of the car, fuel, insurance, registration, servicing and repairs.

Average spending on transport tends to increase as children age; however, on average, expenditure is less for families with children than for young childless couples (ABS 2006), possibly reflecting lower workforce participation by mothers (Kaushal, Gao & Waldfogel 2007; Passero 1996). Other demographic factors have also been found to be relevant including income (ABS 2006) and lifecycle stage (Lee & Huh 2004).

Housing: Demands for housing are likely to place a major strain on the family budget as a family grows in size. In fact, housing expenditure may constitute almost one-third of the before-tax income of couple families with children aged less than 6 years (Exter 1992).

Although it has been estimated that a family of three needs a housing budget 38 per cent higher than that required by a two–adult childless household, researchers have observed economies of scale with increasing numbers of children, with no significant differences found for two-parent families with two and three children (Valenzuela 1999). Economies could be due to children sharing rooms or foresight by parents; the home being purchased or rented at the time of the birth of the first child had sufficient rooms to accommodate extra occupants (Valenzuela 1999).

Similar to other essential expenditures, demographic factors have been observed to influence housing expenditure. For example, unemployed families apportion greater expenditure to housing when compared to employed families (Passero 1996), while expenditure tends to diminish over the life cycle as housing increases in value and capital is paid off (ABS 2006; Carlucci & Zelli 1998; Percival & Harding 2007). However, studies have also identified characteristics that set housing expenditure apart from other essential expenditures. Housing expenditure can differ markedly between households that are similar in every other aspect apart from tenure type (Carlucci & Zelli 1998).

Furniture and appliances: Few studies have undertaken detailed examination of expenditure items that comprise smaller shares of total expenditure, such as furnishing and appliances. Of exception are two Australian studies. Valenzuela (1999) observed economies of scale for household furnishings with increasing numbers of children, while Percival and Harding (2007) found increased expenditure recorded in the ‘other’ category of the Household Expenditure Survey for families with young children and speculated that this might be due to expenditure on furnishings for babies.

General insurance, telephone and internet: Similar to other items that comprise a small share of total household expenditure, studies tend to group the items of insurance, telephone and internet expenditure, or not treat them as the subject of the research focus (for example, see Valenzuela 1999). Conversely, research findings are difficult to apply to the Australian context given differences in cultural or policy settings between countries. For example, in a US study, increases in spending on insurance associated with labour force participation reflected US retirement savings policy, which requires working people to contribute to their own retirement pension (Kaushal, Gao & Waldfogel 2007)


47

Electronic goods: Of particular interest in our study is expenditure on electronic goods. Although there are no Australian studies specifically examining family expenditure on electronic goods, the ABS (2006) reports differences associated with increasing income quintile and geographic location. However, various media reports suggest that some families spend the government’s baby bonus on electronic goods8 rather than child-related expenses incurred by the family. These reports are based on anecdotes and personal observation and have not been the subject of quantitative analysis. Data from the HILDA survey used in this study allows us to examine this spending item separately.

Holidays: The discretionary nature of spending on recreation may lead to a differential pattern of holiday taking behaviour through various stages of the family life cycle that is distinct from expenditure on essential items. In an Australian study by Yusuf and Naseri (2005), the authors found income, ethnicity and family life cycle (rather than family size) influenced holiday taking behaviour. Ethnicity and the presence of an older child were positively associated with overseas travel, while the presence of a pre-school child was negatively associated with travel overseas, resulting in low overall combined expenditure on holidays for families with pre-school children when compared to other household types.

Alcohol and cigarettes: The presence of children appears to move expenditure towards child-focused goods and, as a consequence, away from adult-focused goods such as alcohol and tobacco. These behavioural changes may reflect decreasing social opportunities associated with increasing demands of childrearing, reductions in consumption for health reasons relating to pregnancy or budget restraints. Few studies have examined the reasons for such reductions; however, demographic differences in expenditure on these items have been found. For example, when compared to married parent families, cohabiting parent families tend to spend more on adult goods such as alcohol and tobacco and less on potentially child-related goods such as education (DeLeire & Kalil 2005), while declining consumption of alcohol and tobacco is associated with increasing family size (Valenzuela 1999). However, separate examination of these items using the 2004–05 ABS Household Expenditure Survey (HES) (ABS 2006) indicates that expenditure on tobacco products remains relatively constant across couple families with and without dependent children, while alcohol expenditure is markedly less for couples with dependent children when compared to childless couples.

3 Theoretical frameworkA review of literature formed the basis for the following hypotheses for changes to family expenditure patterns associated with the birth of a child.

Firstly, we expect increased expenditure on groceries, clothing, health, housing, furniture and appliances, and transport. Expenditure on groceries and transport is likely to increase by a similar amount for families experiencing a first, second or higher order birth due to the addition of an extra consumer to the household. However, for grocery expenditure, families have some discretion in the choice of individual items that may offset some expenditure demands if families adjust their consumption in favour of baby-specific items.


48

For items of children’s clothing and furniture and appliances, we expect increases in expenditure with the birth of the first child, with smaller increases for subsequent children because they are often able to use items originally purchased for their older siblings. Although we might also expect decreases in adult clothing expenditure due to women’s labour force withdrawal, overall we anticipate increases as mothers purchase maternity clothes.

Consistent and sizeable increases in health expenditure associated with pregnancy, childbirth and an increasing number of health service consumers in the household are predicted. However, some health expenditure is likely to be discretionary given the choice between low-cost public and higher-cost private health systems in Australia, and thus, on average, we may observe lower increases than expected.

Secondly, we expect decreases in expenditure on meals eaten out and takeaway food, housing, child care, and alcohol and tobacco. Expenditure on meals eaten out is likely to be limited by reduced opportunities for social engagement outside the home and budgetary constraints on consumer spending. However, some of these decreases might be offset by increases in expenditure on takeaway food consumed at home. A decrease in housing expenditure associated with the birth of the first child is anticipated as home purchases, relocations to larger premises and renovations are likely to have been made in the period before the birth of the first child. Further, families are likely to decrease higher mortgage payments, which may have been affordable for a childless couple, to offset demands for expenditure on child-related goods. For second and subsequent births, we expect no change in housing expenditure.

Decreased expenditure on child care is anticipated due to reductions in mothers’ employment hours, while decreased spending on alcohol and tobacco is expected due to behavioural change. However, it is likely that some of the behavioural change associated with having children may have already occurred prior to the time of data collection, given that approximately three-quarters of the mothers in the study were already pregnant at the time of pre-birth survey.

Thirdly, we expect expenditure on education, insurance, telephone, internet, electronic goods and holidays to remain constant. It is possible that overall expenditure on education will decrease if parents place on hold or reduce their own participation in education with the arrival of a new child, but we expect, on average, these changes to fall short of statistical significance. Further, although intuitively we expect expenditure on holidays to decrease when a new child enters the household, expenditure may increase if families time a holiday to introduce a new baby to family and friends. We do not expect to observe evidence to support press reports that imply families with newborns spend government lump-sum payments on consumer items.

4 Method

Data

The data used in this study were drawn from Waves 6 (2006) and 7 (2007) of the HILDA survey, a household-based panel survey described in detail by Wooden and Watson (2007). The HILDA


49

survey is a broad social and economic survey established to support research in: household and family dynamics; income and welfare dynamics; and labour market dynamics. The survey began in 2001 with a large probability sample of Australian households occupying private dwellings (ed. Watson 2009). All members of the households providing at least one interview in 2001 form the basis of the panel that was pursued in each subsequent wave.

In this study, data from female HILDA survey respondents were selected for analysis if the respondents were: interviewed in both years; aged between 15 and 45 years in Wave 7; living in a couple household in both years without any other persons besides their own children; were in a married or cohabiting relationship with the same person in both years;9 and were not in a same-sex relationship (to avoid double-counting some households). The selection process therefore excluded women who were not living with the same male partner10 in both waves and resulted in a sample of 1,621 women. The study did not use expenditure data collected in earlier surveys because only 10 of the 25 household spending categories used in 2006 and 2007 were common across all surveys.

Household, respondent and partner data were incorporated into each respondent’s record. This enabled the household to be the unit of analysis because the selection process ensured that the household each respondent lived in at the time of the 2007 survey was essentially the same one in which they had lived in a year earlier. Eighty-one households in 2006 and 97 households in 2007 had missing data for women’s partners in relation to education, employment, income support and health status. These households were dropped from the models when these variables were used. Missing data for household disposable income data was imputed from income data provided in the HILDA survey datasets (Starick & Watson 2007).

A total of 248 households who did not provide any household spending data in at least one of the two waves and six households with top-coded income in either 2006 or 2007 were also dropped from the study sample. The final sample was therefore 1,367 households, in which 307 newborns arrived in the twelve months leading up to either of the 2006 or 2007 surveys. One hundred and twenty-six of the newborns had no older siblings in the household (first-born), 118 had one older sibling in the household (second-born) and 63 had more than one older sibling in the household (third or subsequent-born).

Dependent variables

The household spending section of the 2006 and 2007 HILDA self-completion questionnaires listed 25 types of expenses on which Australians regularly spend money. This study used the household spending items contained in the HILDA survey household file, which averaged the household spending responses if more than one person in a household provided responses (ed. Watson 2009). The 25 items were aggregated into 13 spending categories as per Table A1, with each of those spending categories being the dependent variable in 13 fixed effects linear regression models. Two other spending categories were created from household spending items found in the HILDA household questionnaire, thereby resulting in 15 dependent variables in total.11 The household spending responses in 2006 were converted to 2007 dollars by adjusting for changes in the Consumer Price Index (CPI), using CPI figures broken down into the group of goods and services most closely corresponding to the expenditure item. Table 1 presents


50

descriptive statistics on the 15 dependent variables used in this study, for the households included in this study and for other HILDA survey households in 2006 and 2007.

Table 1: Average household expenditure (2007 dollars), study sample compared with other HILDA survey households(a)

2006 2007

Expenditure category(annual)

Study sample (n=1,367)

Other HILDA households (n=4,742)

Study sample (n=1,367)

Other HILDA households (n=4,543)

Groceries 10,113.48* 7,732.00* 10,469.90* 7,819.10*

Meals eaten out and takeaway 2,660.41* 2,091.81* 2,807.92* 2,120.51*

Adults’ clothing and footwear 1,525.13* 1,176.84* 1,699.56* 1,246.42*

Children’s clothing and footwear 798.61* 282.06* 884.87* 255.28*

Child care (typical week)(b) 16.38* 1.79* 21.39* 1.54*

Education 1,314.89* 786.31* 1,507.81* 687.74*

Health 2,234.69 2,048.68 2,428.12* 2,058.50*

Transport 8,999.14* 6,682.58* 9,631.62* 6,137.82*

Housing 20,474.01* 10,230.11* 21,549.37* 10,082.32*

Furniture and appliances 1,614.87* 1,015.75* 1,558.54* 1,021.78*

General insurance 1,346.13* 1,077.11* 1,462.27* 1,099.17*

Electronic goods 1,069.20* 762.05* 1,334.40* 870.50*

Telephone and internet 1,901.60 1,724.50 2,140.06* 1,735.09*

Holidays and holiday travel 2,781.64* 2,384.70* 2,731.06 2,591.50

Alcohol and cigarettes/tobacco 2,300.00* 2,030.66* 2,371.76* 1,988.71*

(a) Excluding households that did not provide household spending responses.(b) Not annualised as data on the number of typical weeks was not available.Note: * Study sample different from other HILDA survey households at p<0.05.Source: HILDA (2006, 2007) release 7.0.

In general, there was an increase in reports of expenditure for all households between waves. Mean spending was significantly higher for sample households than it was for other HILDA survey households, except in relation to health (in 2006), telephone and internet (in 2006), and holidays and holiday travel (in 2007). The largest differences were in the spending categories of: child care; children’s clothing and footwear; and housing. Furthermore, because of increased spending by sample households between 2006 and 2007, the differences between the means increased in 2007 for all categories except for holidays and holiday travel, and for furniture and appliances.

As the household spending items were self-reported and no spending diary was kept, it can be expected that the items are subject to substantial measurement error (Gibson & Kim 2007). Some expenditure items are more likely to exhibit large measurement errors. For example, Headey


51

(2008) found that the means of several 2005 HILDA survey expenditure items (clothing and footwear; health care; holidays; hobbies; and child care) differed by more than 10 per cent from the means of comparable items in the ABS 2003–04 HES, which was based on spending diaries and thought to be generally more accurate. More encouragingly, Headey also found that the other HILDA survey expenditure items, when totalled, differed by less than 4 per cent from the HES items after adjustment for inflation.12 Respondents are also likely to have varied widely in how they estimated their usual spending and this is especially problematic in regard to the items for which they were asked to provide an estimate of annual expenditure (such as the purchase of motor vehicles or computers). Some respondents may have attempted to average their purchases over several years, others may have estimated their expenditure for the current year, while others may have used an estimation approach that falls between these two extremes.

The measurement errors described above would reduce the efficiency of the fixed effects linear regression models used in this study and also bias the model coefficients if the measurement errors are correlated with the true values (Bound, Brown & Mathiowetz 2001). Although it is unfortunate that model efficiency would be adversely affected by measurement error, this only means that false negatives are more likely (that is, the measurement errors are likely to lead to fewer statistically significant results), and any findings that some spending categories show significant change would remain valid. This validity would also be unaffected by bias in the model coefficients; however, the presence of bias does mean that more emphasis should be placed on the direction of significant change in expenditure rather than on the dollar estimates when interpreting the results.

Independent variables

A range of independent variables were used to develop the fixed effects linear regression models with time-varying controls. The variables of interest in the study were: birth13 of first-born child; birth of second-born child; and birth of third or subsequent-born child. These variables were used in preference to the option of using a single birth variable for the purpose of capturing birth order effects. Eighteen per cent of households that had reported the arrival of a second-born baby in the 2007 survey had reported the arrival of a first-born baby in the 2006 survey, and 14 per cent of households that had reported the arrival of a third or subsequent-born baby in the 2007 survey had reported the arrival of a second-born baby in the 2006 survey. Consequently, all three birth variables were included in each model to control for preceding or following births in households that had experienced births occurring in each of the two years. Other control variables included: education,14 employment, income support and health status of women and their partners; household disposable income;15 geographical remoteness; and housing tenure.

Table 2 presents descriptive statistics on the 142 households that had experienced the arrival of a singleton newborn in the 12 months leading up to the 2007 survey.16 These descriptive statistics provide potential explanations of how some variables with little or no time variance could be affecting the results. These variables could have been incorporated into fixed effects models as interaction terms; however, interaction terms were not used in this study because highly parsimonious models were preferred.


52

Table 2: Selected descriptive statistics in 2007 for households with a singleton newborn, by parity

Mean

First-born (n=62) Second-born (n=51)

Third or subsequent-born (n=29)

Household financial year disposable income ($‘000)

95.76a

93.32a

83.58a

Mother with university qualification 0.53a

0.49a

0.41a

Mother’s age (years) 29.97a

32.49b

32.83b

Age of oldest resident child (years) 0.23a

4.10b

7.83c

SEIFA(a) 6.45a

6.22a

5.66a

Mother’s employment (%)

Employed 0.58a

0.41a,b

0.24b

Unemployed 0.02a

0.02a

0.00a

Not in labour force 0.40a

0.57a,b

0.76b

Partner’s employment (%)

Employed 0.90a

0.96a

0.89a

Unemployed 0.03a

0.00a

0.07a

Not in labour force 0.07a

0.04a

0.04a

Mother currently receiving income support (%)

0.08a

0.10a

0.21a

Partner currently receiving income support (%)

0.03a

0.04a,b

0.15b

Remoteness area (%)

Major city 0.71a,b

0.75a

0.52b

Inner regional 0.16a

0.14a

0.28a

Outer regional 0.08a

0.12a

0.21a

Remote 0.05a

0.00a

0.00a

Very remote 0.00a

0.00a

0.00a

Housing tenure (%)

Own outright 0.09a

0.16a

0.07a

Paying off mortgage 0.58a

0.63a

0.55a

Rent (or pay board) 0.27a

0.18a

0.34a

Other 0.05a

0.04a

0.03a

(a) SEIFA—Socio-Economic Indicators For Areas (decile of relative socioeconomic advantage/disadvantage). (Notes continue on p. 53.)


53

Note: Values in the same row that have different subscripts are significantly different at p<0.05. For example, mother’s age is significantly different between first-born and second-born because the subscripts (a and b respectively) are different. However, there is no significant difference between mothers of second-born and third or subsequent-born children because they share a subscript (b).

Source: HILDA (2007) release 7.0.

Although there were few significant differences, statistics for households with third or subsequent-born babies generally indicated higher prevalence of characteristics that are associated with disadvantage relative to households with first or second-born babies. Note that values in the same row that do not share subscripts (a, b or c) are significantly different at p<0.05. Partners of mothers in households with third or subsequent-born babies were significantly more likely to be receiving income support than those residing in households with first-born babies. Mothers in households with third or subsequent-born babies were significantly less likely to be employed and more likely to not be in the labour force than those residing in households with first-born babies. Households with third or subsequent-born babies were also significantly less likely to be located in major cities than households with second-born babies. As expected, the ages of mothers and their oldest children were significantly higher for the later birth-orders: however, there was one notable exception—the average age of mothers in households with third or subsequent-born babies was not significantly different from the average age of mothers in households with second-born babies. This indicated that the average age of mothers in households with third or subsequent-born babies when they had given birth to their first child was approximately17 3.4 years less than the average age of mothers in households with second-born babies when they had given birth to their first child.

Although not reported in Table 2, households experiencing the arrival of newborn babies in the 12 months prior to the 2007 survey reported disposable incomes in 2006 that were comparable to other sample households. However, because households experiencing the arrival of newborns experienced real disposable income growth of 10.8 per cent between 2006 and 2007, compared to growth of 6.4 per cent for other sample households, they had significantly higher disposable incomes to other households by 2007. This difference predominantly reflects receipt of maternity payments and family tax benefits resulting from a birth. However, households that had reported the arrival of a third or subsequent-born baby in the 12 months prior to the 2007 survey experienced a real disposable income growth rate of a relatively modest 5.1 per cent (in comparison to 9.9 per cent and 14.6 per cent respectively for households with first-born babies and households with second-born babies). The slower growth rate for households with third or subsequent-born babies reflects a lack of growth in private incomes.

data analysis

Both fixed effects and random effects linear regression models were investigated for the purposes of estimating changes in the dependent variables associated with the arrival of newborn babies in a household. However, random effects linear regression models were rejected because Hausman tests demonstrated that the random effects models were producing inconsistent results. Unfortunately, the inconsistency of the random effects models made it inadvisable to use random effects tobit models, which would have controlled for left-censoring in the dependent variables. Fixed effects tobit models based on a maximum likelihood estimator


54

would produce inconsistent results caused by the ‘incidental parameters problem’ (Greene 2004) and, as such, are not available in statistical software packages. Although Honoré developed a semi-parametric estimator for fixed effects tobit models (Honoré 1992), it was tested for this study and found to be highly sensitive to the starting point of the optimisation routine, and consequently also produced inconsistent results. Consequently, standard fixed effects linear regression models were ultimately used in this study because they were the most likely to produce consistent results.

For each of the dependent variables, fixed effects models without time-varying control variables were produced for the three birth variables (first-born, second-born, and third or subsequent-born) to establish the baseline. Fixed effects models with time-varying controls were then developed for the 15 dependent variables. Not all independent variables were used in each model, with details of which variables were used in each model provided in Table A2. The three birth variables were always retained in the models with time-varying control variables as they were the variables of interest and were needed as controls for each other. With one exception, the other variables were dropped from a model if they had little relationship with the dependent variable in a fixed effects model without time-varying controls, or contributed little to a model with time-varying controls. The exception was the household disposable income variable, which was always retained in the models because it was assumed that measurement errors would be large for an income variable (Moore, Stinson & Welniak 2000) and that any lack of statistical significance was not due to a lack of association between change in income and change in consumption, but due to attenuation bias caused by the errors. Bootstrap standard errors (1,000 replications) were calculated for the models as heteroscedasticity in the error terms was nearly always present.


55

5 ResultsTable 3: Fixed effects linear regression models without time-varying controls: change in annual

household expenditure (2007 dollars) associated with newborns, by parity

Expenditure category Birth order

1st 2nd 3rd, 4th, etc.

Groceries 213.51 120.83 –523.43

Meals eaten out and takeaway –365.17 375.50* 168.83

Adults’ clothing and footwear 230.11 –186.44 135.88

Children’s clothing and footwear 424.21** 88.27 68.00

Child care (typical week)(a) –3.15 –30.82* –10.06

Education –217.74 –101.32 31.18

Health 623.66** 544.01* 589.71**

Transport –437.87 2,927.99 2,900.71

Housing –3,075.11 959.74 739.96

Furniture and appliances 138.17 –122.90 –387.94

General insurance –38.01 –103.30 –1.87

Electronic goods –78.79 96.13 541.71*

Telephone and internet –43.35 6.23 –237.82

Holidays and holiday travel –616.32 –69.94 369.40

Alcohol and cigarettes/tobacco –212.71 75.93 –129.85

(a) Not annualised as data on the number of typical weeks were not available.Notes: *p<0.05, **p<0.01. Each cell is a separate regression.Source: HILDA (2006, 2007), release 7.0.

Results for the fixed effects models without time-varying controls are presented in Table 3. Regardless of birth order, there were no changes in expenditure between 2006 and 2007 that were significant at the 95 per cent confidence level for the following categories: groceries; adults’ clothing; education; transport; housing; furniture and appliances; general insurance; telephone and internet; holidays; and alcohol and cigarettes. There was a significant increase in spending on meals eaten out and takeaway associated with the arrival of second-born babies ($375 per annum, p=0.039). A significant increase in spending on children’s clothing ($424 per annum, p<0.001) was associated with the arrival of a first-born baby; however, increases in this category of expenditure for second and subsequent-born babies were smaller and non-significant. There was a significant decrease in child care expenditure ($31 in typical weekly expenses, p=0.011) associated with the arrival of second-born babies, but no significant changes in this spending category were associated with the arrival of other babies. There were significant increases in health care expenditure for first, second and third or subsequent-born babies (p=0.002, p=0.016 and p=0.004 respectively). Furthermore, the increases in health care


56

spending were very similar across the birth orders, ranging between $544 and $624 per annum. A significant increase in spending on electronic goods ($542 per annum, p=0.040) was associated with the arrival of third or subsequent-born babies, but no significant changes were observed for other babies. Some of the items significant at the 95 per cent confidence level may be spuriously significant given that a total of 45 coefficients were estimated; however, this is unlikely for the items significant at the 99 per cent confidence level.

Table 4: Fixed effects linear regression models with time-varying controls: change in annual household expenditure (2007 dollars) associated with newborns, by birth order

Expenditure category Birth order1st 2nd 3rd, 4th, etc.

Groceries 58.23 –16.48 –147.70Meals eaten out and takeaway –285.77 434.07* 257.70Adults’ clothing and footwear 221.75 –168.82 125.58Children’s clothing and footwear 433.92** 127.91 87.26Child care (typical week)(a) –1.62 –26.97* –10.24Education –192.51 –121.19 13.47Health 662.07** 622.54** 648.09**Transport 17.04 2,744.90 3,060.33Housing –4,069.01 –71.00 645.68Furniture and appliances 186.05 –92.17 –377.35General insurance –103.64 –134.94 –45.44Electronic goods –125.86 107.94 544.39Telephone and internet –46.20 –36.38 –195.67Holidays and holiday travel –641.21 –163.99 381.31Alcohol and cigarettes/tobacco –268.49 0.62 –232.62

(a) Not annualised as data on the number of typical weeks were not available.Notes: *p<0.05, **p<0.01. Each row is a separate regression.Source: HILDA (2006, 2007), release 7.0.

The results for the fixed effects models with time-varying controls, presented in Table 4, were not substantially different from the results of the models without the controls. The only difference in terms of which estimates were significant at the 95 per cent confidence level is that the increase in spending on electronic goods ($544 per annum) associated with the arrival of third or subsequent-born babies fell short of significance (p=0.053), whereas in the analysis without controls it was significant (p=0.040).

6 DiscussionThe purpose of this study was to examine whether there were changes in expenditure on particular items for families experiencing the birth of a child by applying longitudinal analysis techniques to a sample of couples of childbearing age who reported expenditure in two waves of the HILDA survey, collected before and after the birth event.


57

Reports of actual family expenditure by couple households in this study indicate that, although expenditure on many items is comparable between couples with and without children, there are also some shifts in expenditure associated with the arrival of first or subsequent-born children.

Contrary to expectations, expenditure on groceries did not increase with the addition of a newborn child. Parents may have adjusted their expenditure to purchase lower cost grocery items to offset new baby-related expenses. However, it is also possible that, due to question design,18 expenditure reports only include food—excluding nappies and formula—and as such small increases may reflect low food consumption by babies.

In contrast, increased expenditure was observed for meals eaten out and takeaway. These results may indicate that parents may be outsourcing food purchase and preparation as time pressures increase with the addition of a second or subsequent child to the household.

Both adult and children’s clothing expenditure increased with the birth of a first child, although the change in adult clothing expenditure fell short of significance (p=0.076). This relationship persisted after controlling for income, suggesting that increasing clothing expenditure was not as a result of increasing family income and purchases of luxury items. Expenditure on children’s clothing most likely reflects purchases of clothing for the new baby, while for adults, increased expenditure may be due to purchase of maternity clothes, or shopping opportunities created by the need to shop for children’s clothing. Consistent with both hypotheses, there was strong correlation between adults’ and children’s clothing expenditure (r=0.54) in 2007. In contrast, there were no significant changes in expenditure on adult and children’s clothing for couples having second or subsequent children. Consistent with previous research, there appear to be economies of scale associated with increasing family size (Valenzuela 1999).

Although it might be expected that child care costs will increase with increasing family size, this study indicates that labour force participation rates decrease for women with young children and result in lower demand for child care services. As women’s broad employment status is included as a control variable in the model, the results for child care expenditure most likely reflect employed mothers taking up maternity leave and therefore having less need for paid child care when they have a second child. Descriptive statistics, and the results of the models with time-varying controls in relation to parents of third or subsequent births, suggest that couples going on to have a third or subsequent child, on average, spent less on child care in both waves when compared to couples who experienced a second birth, and thus the birth event did not impact on child care usage for these families.

Unlike previous research, health expenditure comprised a large proportion of the expenditure reported in the HILDA survey. This result is reflective of the study design, which captures expenditure of couples at a time when demands for maternity-related health services are likely to be great. A positive and substantial increase in expenditure on health was observed for families experiencing a birth, regardless of the birth order. Increased health expenditure is likely to represent out-of-pocket health care expenses associated with pregnancy, birth and care of a newborn. The inclusion of other births in the preceding or following year as control variables had the effect of increasing the estimated change in health expenditure. This is because the models without time-varying controls did not take into account that health spending may already be elevated in the year preceding a birth if another recent birth had occurred.


58

Contrary to expectations, there was no observed decrease in transport expenditure associated with a first birth. However, the results show a large increase—falling short of significance—in transport expenditure for couples experiencing the birth of a second or subsequent child; possibly reflecting older children starting school or upsizing of the family car. Given the average size of increases reported, a highly significant result was expected; however, large variability in reports of expenditure indicate that attenuation bias or other confounding factors may be influencing the results. It is probable that transport costs vary greatly between individuals depending on whether older children attend school, whether there are one or more cars in the household, the availability of public transport and the workforce participation of each parent (Kaushal, Gao & Waldfogel 2007; Passero 1996).

As stated previously, housing expenditure generally comprises a large proportion of total family expenditure and therefore it is likely to impose constraints on the amount of money a family has available to spend elsewhere. Consistent with ABS reports, the results suggest HILDA families with young children spend more on housing than groceries, noting percentage expenditure is not comparable between data sources due to differences in survey design. Overall, the results suggest decreasing housing expenditure associated with the birth of a first child, with economies of scale evident for second and subsequent children; noting, as in previous studies, that there was a large amount of variation in housing expenditure between study couples (Carlucci & Zelli 1998). Further analysis, after disaggregating the housing expenditure category reported in this study, reveals that the decrease in spending associated with the birth of a first child largely reflects childless couples reducing their spending on home renovations once their baby arrives.

In contrast with predictions, no significant results were observed for expenditure on furniture and appliances. This result could possibly reflect the large degree of variability in reports or the study design whereby three-quarters of mothers were already pregnant at the time of the previous year’s survey. In the latter case, purchases of baby-specific items may have occurred prior to the arrival of the newborn. Similarly, no significant differences were observed for expenditure on education, holidays, general insurance, or telephone and internet. These results are in line with expectations that expenditure on these items does not necessarily increase or decrease with the birth of a child.

Reports of changes in expenditure on electronic goods appear to depend on birth order. There were no significant changes in expenditure associated with the birth of a first or second child; however, the result approached significance for couples experiencing the birth of a third or subsequent child (p=0.053). This result provides some evidence to substantiate anecdotal reports. It is possible that these families have already purchased the baby goods they require for previous births and therefore demand for baby-related goods does not coincide with the birth of the child. Thus increases in income associated with the birth may be used to purchase consumer items that were previously unaffordable. However, given the higher prevalence of characteristics associated with disadvantage—including lower family income—in the third and subsequent child subgroup, this is a result that warrants further investigation.

Changes in expenditure on alcohol and tobacco fell short of significance. Expenditure on alcohol and tobacco decreased with the birth of a first child,19 possibly reflecting reductions in


59

consumption for health-related reasons, or movement of expenditure away from adult-centred goods to child-centred goods.

Finally, it is interesting to note that most couple households with newborn babies experienced strong growth in disposable income, partly because of maternity payments and family tax benefits, but also because they were households headed by couples of prime working age who typically experienced strong growth in private incomes in this period. As such, income constraints were unlikely to have played a significant role in changing household spending patterns for most couples, even in the analysis where no time-varying controls (such as income) were included. Of exception are the households with a third or subsequent-born baby as these households experienced relatively low growth in disposable income in the study timeframe. However, it must be recognised that this study reports total expenditure by the family unit and each birth resulted in a larger family size and therefore more consumers to share this income.

7 ConclusionInformation on the expenditure patterns of parents of newborns is important to a range of current policy debates, including those that seek to determine the level of family assistance and influence women’s attachment to the labour force, and more recently, the policy debate on paid parental leave. This study shows that a range of expenditure categories are influenced by the arrival of a new baby. Parents of first-born children increase expenditure on health care and clothing. Parents of second-born children increase expenditure on health care, and on meals eaten out and takeaway; however, they decrease expenditure on child care. Parents of third and subsequent-born children increase expenditure on health care. When considered together, these results indicate that parents adjust their spending in response to the birth of their baby.

Furthermore, the absence from the study of household expenditure data prior to 2006 is a possible explanation as to why other significant changes were not found. For example, although significant results were not observed for expenditure on housing, furniture and electrical appliances, it is likely that couples were purchasing homes with sufficient space to raise children—and purchasing the furniture and appliances for those homes—prior to 2006. Similarly, couples may have already reduced their expenditure on holidays, and on alcohol and tobacco, as most of the women were pregnant at the time of the 2006 survey. Had the study been able to analyse a longer time series of household spending data, it is possible that it would have found that parents were making more spending adjustments than those indicated by the significant results found in the 2006 to 2007 comparison. Families may have also been changing their spending patterns in response to the arrival of a newborn within each of the broad expenditure categories used in this study. Such changes would not be visible in this research, and this may hide financial pressures if families have to make savings to offset new purchases.

This study provides support for government financial assistance associated with the birth of children primarily because of large observed expenses for health care and children’s clothing. The study’s results are also not inconsistent with expectations that couples typically would have spent more money on housing, furniture and electrical appliances at least a year in advance of starting a family, although no conclusive evidence could be found because of the limited time


60

series of consistent household spending data. The study generally does not support reports in the press that the parents of newborns are often spending their Baby Bonus on plasma televisions, other electronic goods, holidays and alcohol (Haynes 2008). However, this research identifies some increase in spending on electronic goods for couples when the newborn is a third or subsequent-born child.

As a final point, we would caution against applying dollar values reported in this study to determine rates of parental leave or family assistance payments. Firstly, as stated earlier in this paper, expenditure items collected in the HILDA survey do not add to total expenditure, although many of the main items are reported. Secondly, the paper only includes reports of families with mothers of childbearing age who were coupled in both waves. Thirdly, an unknown amount of expenditure change relating to the birth of a child may have already been made, given that approximately three-quarters of women were already pregnant at the time of pre-birth data collection. Finally, we do not use equivalised income and therefore do not explicitly take account of the number of consumers in the household.

Appendix

Table A1: Aggregation of HILDA survey spending items into dependent variables

Spending item Dependent variable

Groceries (including food, cleaning products, pet food and personal care products. Does not include alcohol or tobacco)

Groceries

Meals eaten out (including restaurants, takeaway food, and bought lunches and snacks. Does not include alcohol)

Meals eaten out and takeaway

Alcohol (including alcohol consumed with meals eaten out) Alcohol and cigarettes/tobacco

Cigarettes and other tobacco products Alcohol and cigarettes/tobacco

Public transport and taxis Transport

Motor vehicle fuel (petrol, diesel, LPG) and engine oil Transport

Motor vehicle repairs and maintenance (including regular servicing) Transport

Buying brand new motor vehicles, motorbikes or other vehicles (including boats, planes, caravans, trailers and jet skis)

Transport

Buying used second-hand motor vehicles, motorbikes or vehicles (including boats, planes, caravans, trailers and jet skis)

Transport

Men’s clothing and footwear Adults’ clothing and footwear

Women’s clothing and footwear Adults’ clothing and footwear

Children’s clothing and footwear Children’s clothing and footwear


61

Table A1: Aggregation of HILDA survey spending items into dependent variables (continued)

Spending item Dependent variable

Telephone rent and calls, and internet charges (including rent and charges on mobile phones)

Telephone and internet

Holidays and holiday travel costs (including short and long holidays) Holidays and holiday travel

Private health insurance Health

Fees paid to doctors, dentists, opticians, physiotherapists, chiropractors and any other health practitioner

Health

Medicines, prescriptions and pharmaceuticals Health

Other insurance (such as home and contents and motor vehicle insurance)

General insurance

Rent or board(a) Housing

Usual home loan repayment(a) Housing

Repairs, renovations and maintenance to your home Housing

Electricity bills, gas bills and other heating fuel (such as firewood and heating oil)

Housing

Furniture Furniture and appliances

Household appliances, such as ovens, fridges, washing machines and air conditioners

Furniture and appliances

Education fees paid to schools, universities, and other education providers

Education

Computers and related devices (such as printers, digital cameras, iPods, MP3 players, electronic organisers and games consoles)

Electronic goods

Televisions, home entertainment systems and other audio visual equipment (such as DVD players and video cameras)

Electronic goods

Child care cost after any benefit for all not-yet-at-school children across all types of care, while parents at work(a)

Child care

(a) Obtained from the HILDA household questionnaire.


62


63

Tabl

e A2

: In

depe

nden

t var

iabl

e ut

iliza

tion

in fi

xed

effe

cts

linea

r reg

ress

ion

mod

els

with

tim

e-va

ryin

g co

ntro

l var

iabl

es

Expe

ndit

ure

cate

gory

Bir

th/

adop

tion

in

past

yea

r

Bir

th/

adop

tion

of

sibl

ing

in

prev

ious

or

follo

win

g ye

ar

Educ

atio

n (b

oth

pare

nts)

—

curr

ently

en

rolle

d in

a

cour

se o

f st

udy

Empl

oym

ent

stat

us (b

oth

pare

nts)

Inco

me

supp

ort (

both

pa

rent

s)—

curr

ently

re

ceiv

ing

inco

me

supp

ort

Long

-ter

m

heal

th

cond

ition

(b

oth

pare

nts)

Hou

seho

ld

finan

cial

yea

r di

spos

able

in

com

e

Rem

oten

ess

area

Hou

sing

te

nure

Gro

ceri

esx

xx

xx

x

Mea

ls e

aten

out

&

take

away

xx

xx

x

Adu

lts’

clo

thin

gx

xx

x

Child

ren’

s cl

othi

ngx

xx

x

Child

car

ex

xx

x

Educ

atio

nx

xx

xx

x

Hea

lth

xx

xx

Tran

spor

tx

xx

xx

Hou

sing

xx

xx

Furn

itur

e &

ap

plia

nces

xx

xx

Gen

eral

insu

ranc

ex

xx

x

Elec

tron

ic g

oods

xx

xx

xx

Tele

phon

e &

in

tern

etx

xx

xx

x

Hol

iday

sx

xx

x

Alc

ohol

&

ciga

rett

esx

xx

x


62


63

Endnotes1 A$5,000 as at 1 Jan 2009. Also, from this date the Baby Bonus is spread over 13 bi-weekly

payments.

2 The Paid Parental Leave scheme will replace the Baby Bonus for eligible working parents, while the Baby Bonus and Family Tax Benefits will still be available for families not eligible for Paid Parental Leave, and for those who choose not to participate in the scheme.

3 For an overview of the policy go to <www.fahcsia.gov.au>.

4 For a description of eligibility and rates of payment for family assistance see <http://www.familyassist.gov.au/Pages/default.aspx>.

5 Noting that in the United States (US), paid employees contribute to their retirement pension.

6 Approximately 25 per cent of children aged less than 1 year, and 22 per cent of children aged between 0 and 12 years, attend formal child care in Australia (ABS 2008).

7 Noting that in some instances there can be substantial additional out-of-pocket costs.

8 For example, see Haynes, R 2008, ‘Gerry Harvey knows baby bonus spent on plasma TVs’, The Daily Telegraph, 3 April.

9 It was not possible to determine if a married woman in 2007 was cohabiting with her husband-to-be in 2006. In these cases, the women were dropped from the sample.

10 Selecting and separately analysing households with women who have had other relationship histories, including those who were a lone parent in either year, was not viable due to small sample sizes.

11 Summing these dependent variables does not equal total household expenditure, as a quick glance through a telephone yellow pages directory will show that spending on many types of goods and services would not have been captured in the HILDA survey.

12 Headey did not benchmark the eight household spending categories that did not appear until the 2006 and 2007 surveys.

13 Birth or adoption.

14 Education status being measured by whether the respondent is currently enrolled in a course of study.

15 Equivalised income scales are not used in the models. These scales are conceptually inconsistent with this study because they are based on assumptions about the cost of raising children.

16 These additional independent variables were not used in the fixed effects models for any of the following reasons: they added little to the models, they had little or no time-variance, or they were perfectly correlated with time.


64

17 It was not possible to precisely calculate the age of mothers when they had given birth to their first child because date of birth data was not included in the general release of the HILDA survey datasets.

18 The HILDA survey groceries item instructs respondents to ‘include food, cleaning products, pet food and personal care products’, but provides no guidance on whether to include other items that might also be considered to be part of grocery spending under a broader definition.

19 Most of the $268 decrease was due to reduced expenditure on alcohol.

ReferencesAustralian Bureau of Statistics (ABS) 2006, Household Expenditure Survey, Australia: detailed expenditure items, 2003–04 (Reissue), cat. no. 6535.0.55.001, ABS, Canberra.

——2008, Childhood education and care, cat. no. 4402.0, ABS, Canberra.

Bound, J, Brown, C & Mathiowetz, N 2001, ‘Measurement error in survey data’, in JJ Heckman & E Leamer (eds), Handbook of econometrics, 5th edition, Elsevier Science, Amsterdam.

Carlucci, M & Zelli, R 1998, ‘Expenditure patterns and equivalence scales’, paper presented to the 25th General Conference of the International Association for Research in Income and Wealth, Cambridge, 23–29 August.

DeLeire, T & Kalil, A 2005, ‘How do cohabiting couples with children spend their money?’, Journal of Marriage and Family, vol. 67, pp. 286–95.

Doiron, D & Kalb, G 2005, ‘Demands for child care and household labour supply in Australia’, The Economic Record, vol. 81, no. 254, pp. 215–36.

Drago, R, Sawyer, K, Sheffler, K, Warren, D & Wooden, M 2009, Did Australia’s baby bonus increase the fertility rate?, Melbourne Institute of Applied Economic and Social Research Working Paper Series no. 1/09, MIAESR, Melbourne.

Exter, TG 1992, ‘Big spending on little ones’, American Demographics, vol. 14, no. 2, p. 6.

Fan, JX & Zuiker, VS 1998, ‘A comparison of household budget allocation patterns between Hispanic Americans and non-Hispanic white Americans’, Journal of Family and Economic Issues, vol. 19, no. 2, pp. 151–74.

Gibson, J & Kim, B 2007, ‘Measurement error in recall surveys and the relationship between household size and food demand’, American Journal of Agricultural Economics, vol. 89, no. 2, pp. 473–89.

Gray, M 2007, ‘Costs of children and equivalence scales: a review of methodological issues and Australian estimates’, in Costs of children: research commissioned by the Ministerial Taskforce on Child Support, Occasional Paper no. 18, Department of Families, Community Services and Indigenous Affairs, Canberra.


65

Greene, W 2004, ‘Fixed effects and bias due to the incidental parameters problem in the tobit model’, Econometric Reviews, vol. 23, no. 2, pp. 125–47.

Haynes, R 2008, ‘Gerry Harvey knows baby bonus spent on plasma TVs’, The Daily Telegraph, 3 April.

Headey, B 2008, ‘Poverty is low consumption and low wealth, not just low income’, Social Indicators Research, vol. 89, no. 1, pp. 23–39.

Henman, P 2007, ‘Updated costs of children using Australian budget standards’, in Costs of children: research commissioned by the Ministerial Taskforce on Child Support, Occasional Paper no. 18, Department of Families, Community Services and Indigenous Affairs, Canberra.

Honoré, B 1992, ‘Trimmed LAD and least squares estimation of truncated and censored regression models with fixed effects’, Econometrica, vol. 60, no. 3, pp. 533–65.

Kaushal, NK, Gao, QG & Waldfogel, JW 2007, ‘Welfare reform and family expenditures: how are single mothers adapting to the new welfare and work regime?’, Social Service Review, vol. 81, no. 3, pp. 369–96.

Lee, YG & Huh, E 2004, ‘Consumption and saving behavior of older and younger baby boomers in Korea’, Journal of Family and Economic Issues, vol. 25, no. 4, pp. 507–26.

McDonald, P 1990, ‘The costs of children: a review of methods and results’, Family Matters, no. 27, pp. 19–22.

Moore, J, Stinson, L & Welniak, E 2000, ‘Income measurement error in surveys: a review’, Journal of Official Statistics, vol. 16, no. 4, pp. 331–61.

Passero, WD 1996, ‘Spending patterns of families receiving public assistance’, Monthly Labor Review, vol. 119, pp. 21–28.

Percival, R & Harding, A 2007, ‘The estimated costs of children in Australian families in 2005–06’, in Costs of children: research commissioned by the Ministerial Taskforce on Child Support, Occasional Paper no. 18, Department of Families, Community Services and Indigenous Affairs, Canberra.

Percival, R, Payne, A, Harding, A & Abello, A 2007, Australian child costs in 2007 —‘Honey, I calculated the kids … it’s $537,000’, AMP.NATSEM Income and Wealth Report, no. 18, available at <www.amp.com/ampnatsemreports>.

Saunders, P, Chalmers, J, McHugh, M, Murray, C, Bittman, M & Bradbury, B 1998, Development of indicative budget standards for Australia, report to the Department of Social Security, Canberra.

Starick, R & Watson, N 2007, Evaluation of alternative income imputation methods for the HILDA survey, HILDA Project Discussion Paper Series no. 1/07, Melbourne Institute of Applied Economic and Social Research, Melbourne.


66

Valenzuela, MR 1999, Costs of children and living standards in Australian households, Melbourne Institute Working Paper no. 8/99, Melbourne Institute of Applied Economic and Social Research, University of Melbourne, Melbourne.

Walker, S & Reschke, K 2004, ‘Child care use by low-income families in rural areas’, Journal of Children and Poverty, vol. 10, no. 2, pp. 149–67.

Watson, N (ed.) 2009, HILDA User Manual—Release 7, Melbourne Institute of Applied Economic and Social Research, University of Melbourne.

Wooden, M & Watson, N 2007, ‘The HILDA survey and its contribution to economic and social research (so far)’, The Economic Record, vol. 83, no. 261, pp. 208–31.

Yusuf, F & Naseri, MB 2005, ‘A study of domestic and overseas holidays taken by Australian households’, Proceedings of the Australian and New Zealand Marketing Academy, Fremantle.

67

Deriving an evidence-based measure of job quality from the HILDA surveyLiana Leach1, Peter Butterworth1, Bryan Rodgers2 and Lyndall Strazdins3

1Centre for Mental Health Research, The Australian National University2Australian Demographic and Social Research Institute, The Australian National University3National Centre for Epidemiology and Population Health, The Australian National University

Acknowledgements

This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) survey. The HILDA project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (MIAESR). The findings and views reported in this paper, however, are those of the author and should not be attributed to either FaHCSIA or the MIAESR. This research project was funded by FaHCSIA under the Deed of Agreement for the provision of Social Policy Research Services. Peter Butterworth is supported by NHMRC Career Development Award No. 525410. Bryan Rodgers is supported by NHMRC Research Fellowship Award no. 471429.

AbstractThe Household, Income and Labour Dynamics in Australia (HILDA) survey includes twelve items that assess different psychosocial characteristics of work. However, these items are not drawn from an established scale and, therefore, we do not know the best way to combine the items, or indeed the validity of doing so. The current study uses several different statistical methods to develop measures of the psychosocial characteristics of jobs using these items. Consistent with previous research and theory, the results show that the twelve HILDA survey items reflect three key components of psychosocial job adversity: job demands and complexity, job control and job security. This factor structure was consistent across the seven waves of the survey data available for analysis. Based on the current findings, we plan to use the psychosocial job quality items to investigate the relationship between job adversity and physical and mental health over time.

Keywords: job quality, measurement, factor analysis, longitudinal, HILDA survey


68

1 IntroductionThe psychosocial characteristics of jobs such as workloads, employee control and job security may be important determinants of physical and mental health. A study conducted by Broom et al. (2006) showed that combinations of adverse psychosocial job conditions were associated with poor physical health, greater depression and more frequent visits to a general practitioner. Similarly, Griffin et al. (2007) found that higher levels of adverse psychosocial job conditions were associated with greater anxiety and depression. Much evidence from the Whitehall II study of British civil servants and other large epidemiological surveys has found that physical health problems, such as coronary heart disease (Chandola, Brunner & Marmot 2006; Chandola et al. 2008), are linked to stressful psychosocial conditions at work. A meta-analysis by Stansfeld and Candy (2006) concluded that psychosocial work stressors are predictive of an increase in psychological distress. Recently, researchers have used the term ‘job quality’ to classify jobs along an employment continuum, characterised by ‘sets of work features which foster the wellbeing of the worker’ (Green 2006, p. 9; Grzywacz & Dooley 2003).

Given the theoretical and empirical association between psychosocial job characteristics and health outcomes, as well as the association with other factors such as social and economic disadvantage (Siegrist 2000), psychosocial job quality may be an important construct to assess in large-scale population surveys. However, the selection of appropriate constructs, items and scales is problematic. There are various different models of psychosocial job quality or stress. The demand/control/support (DCS) model posits that a stressful work environment entails:

�� high psychological demands within the job (involving work pace, volume of work, the strain involved and the complexity and mental demands of the job)

�� low levels of control over the work process

�� a lack of social support at work (Griffin et al. 2007; Karasek 1979).

Others have identified job insecurity as an important element of work quality or stress (De Witte & Naswall 2003). Job security has been characterised in two ways: objective insecurity, having temporary or casual employment contracts; and subjective or perceived job insecurity, based on workers’ worry about the continuation of their employment (De Witte & Naswall 2003). These two measures of insecurity are correlated, but perceived job insecurity has been shown to be more strongly associated with psychological distress (Ferrie 2001; Sverke, Hellgren & Naswall 2002). The role played by job insecurity has gained greater attention in recent years, paralleling growth in precarious employment situations such as casual, part-time and contract positions (Noyelle 1990; Quinlan, Mayhew & Bohle 2001).

Strazdins et al. (2004) proposed that increased technology and the growth of service and knowledge industries, combined with more just-in-time, contract and contingent jobs, characterise contemporary labour markets. Thus, measurement of work stress should include a combination of job demands and complexity, job control and job security. Strazdins et al. (2004) used data from the PATH Through Life survey, which included four items assessing job demands and 15 items assessing job control. These 19 items were taken from the Whitehall

derIvINg AN evIdeNce-BASed MeASure of JoB quAlIty froM the hIldA Survey

69

study (Bosma et al. 1997), which originally adapted the items from the Job Content Questionnaire (Karasek 1979). Strazdins and colleagues also investigated the utility of a single item assessing perceived job security (Sverke, Hellgren & Naswall 2002). Using these items, the researchers developed a combined five-category measure of demands, control and security, which they labelled ‘job pressure’ and that was shown to have synergistic associations with physical and mental health. Measures of psychosocial job quality that are based on theory and are supported by psychometric analysis, such as those from the Whitehall study and the combined scale developed by Strazdins et al. (2004), have greater validity than those without this evidence.

2 Current projectThe Household, Income and Labour Dynamics in Australia (HILDA) survey is a widely used, national household panel survey that examines labour force status, family relationships, financial and economic circumstances, health, and a variety of other constructs and measures. It is a particularly important resource for Australian researchers and policy makers. The context and focus of the HILDA survey makes it an ideal resource to examine the psychosocial characteristics of work and their consequences. Indeed, a number of papers have used some of the items included in the HILDA survey to investigate research questions about job quality, satisfaction and security (see Adam & Flatau 2006; Dockery 2003; Long 2005). However, as the survey does not include the widely used, psychometrically-based scales of psychosocial job quality used in the epidemiological literature, this previous research has considered individual items. Since the initial wave, the survey has included in the self-completion questionnaire a module of 12 items that assess job conditions. These 12 items were drawn from various scales and surveys and include four items taken from the International Social Science Surveys—Australia (IsssA) (Kelly & Evans 1999). The cohesion and validity of using this set of items to assess psychosocial job quality has not to our knowledge been previously investigated.

This paper reports analyses that seek to develop valid measures of psychosocial job quality based on the 12 items that are available in all seven waves of the HILDA survey. The objective is to use the measures in future research to assess the longitudinal relationship between psychosocial job quality and health outcomes. As well as providing a basis for our own research, we consider that this work may have broader benefits for the Australian research community and policy makers. At the end of 2008 there were approximately 1,050 researchers registered to use the HILDA survey data (HILDA 2008 Annual Report) and some may be interested in assessing psychosocial job quality. In addition, having valid and reliable measures of psychosocial job quality available in the HILDA survey would provide a unique dataset with which to investigate these constructs longitudinally and add to the research literature on the psychosocial aspects of work.

One simple approach to derive a measure of job quality from the 12 items included in the HILDA survey would be to sum the scores for each item (after reversing the scores for items that are negatively worded). This could provide a total scale score, where higher values represented better job quality. However, without further information about the theoretical basis of this combination of items, or examination of the consistency of item performance longitudinally, a scale such as this has little theoretical or empirical support.


70

Therefore, the primary aim of the present analysis is to identify the factor structure, or the underlying latent constructs that underpin the 12 job quality items in the HILDA survey. Factor analysis is ‘a statistical technique that seeks to summarise, or group observed variables into a smaller number of variables’ (Bray 2001, p. 19). Once a factor structure is identified, its consistency across seven waves of data will be tested to evaluate its validity for longitudinal analysis. We will also test the validity of the factors through comparison with another measure of job quality (an additional nine items assessing job demands and control) that was introduced in Wave 5 of the survey. As our theoretical basis, we take the model of job quality described by Strazdins et al. (2004), and hypothesise that three factors representing job demands and complexity, job security and job control will be observed in the HILDA survey data. Prior to conducting any analyses, we expect that the items will load on the factors in the manner shown in Figure 1. Most of the item loadings shown in Figure 1 are intuitively reasonable. However, item 3 ‘I get paid fairly for the things I do in my job’ is somewhat problematic in that it does not fit neatly in any of the three factors. We included the item in the job security factors because pay and security are both external aspects of the job and distinct from intrinsic features of work such as control and job demands. However, we recognise the limitations of this decision and thoroughly investigate the appropriate treatment of this item in subsequent analyses.

Figure 1: Hypothesised factor structure

1. My job is more stressful than I had ever imagined

2. I fear that the amount of stress in my job will make me ill

7. My job is complex and difficult

8. My job often requires me to learn new skills

9. I use my skills in current job

10. I have freedom to decide how I do my own work

11. I have a lot of say about what happens on my job

12. I have freedom to decide when I do my work

3. I get paid fairly for the things I do in my job

4. I have a secure future in my job

5. Company I work for will still be in business in 5 years

6. I worry about the future of my job(a)

Job demands& complexity

Job control

Job security

(a) Item 6 has been reverse scored.


71

3 MethodologyData for the analyses were drawn from Waves 1–7 of the HILDA survey. The HILDA survey is a national household based panel study that commenced in 2001, designed and managed by The Melbourne Institute for Applied Economic and Social Research (The University of Melbourne), and funded by the Australian Government through the Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA). Data are collected annually from all adult members in each sampled household. For further information on the study design and sample of HILDA survey see Wooden and Watson (2002).

For the analyses reported in this manuscript, participants were in scope if they responded to the self-completion questionnaire in Wave 1, and also participated in at least two subsequent waves of the survey (total at least three waves; n=12,092). However, at various times job quality data were not available for some of these participants who were either unemployed, were not in the labour force or did not provide valid responses to the items. Therefore, the final number of respondents included in the cross-sectional (Wave 1) analysis of job quality items was 6,685. The final number of participants included in the longitudinal analyses across Waves 1–7 of the survey was 9,282. Data from the fifth wave of the survey (n=6,257) were used to examine the convergent and divergent validity of the derived measures of job quality against another set of job quality items used in previous research (Strazdins et al. 2004).

Job quality measures

In all waves of the HILDA survey, the self-completion questionnaire included 12 items that assessed respondents’ reported psychosocial job conditions (Box 1). Each item originally used a seven point scale ranging from 1 ‘strongly disagree’ to 7 ‘strongly agree’, which was recoded to 0 ‘strongly disagree’ to 6 ‘strongly agree’, to ensure comparability with other scales in HILDA (which are used in a subsequent paper).

Box 1: The 12 items asked in all waves of the HILDA survey

1) My job is more stressful than I had ever imagined2) I fear that the amount of stress in my job will make me physically ill3) I get paid fairly for the things I do in my job4) I have a secure future in my job5) The company I work for will still be in business in 5 years from now6) I worry about the future of my job7) My job is complex and difficult8) My job often requires me to learn new skills9) I use many of my skills and abilities in my current job10) I have a lot of freedom to decide how I do my own work11) I have a lot of say about what happens on my job12) I have a lot of freedom to decide when I do my work

Note: Responses to each item ranged from 0 ‘strongly disagree’ to 6 ‘strongly agree’. Item 6 is reverse scored in all subsequent analyses as it is worded differently to the other items loading on the ‘job security’ factor.


72

Wave 5 of HILDA introduced nine additional job quality items that have a stronger psychometric evidence base (Box 2). Three of these items measured job demands and six measured job control. The new items were a subset of the items analysed by Strazdins et al. (2004) in the PATH study, and were taken from the British Whitehall study (Bosma et al. 1997). Although these nine items are available for use from Wave 5 onwards of the HILDA survey, they are not present in earlier waves.

Box 2: The nine additional items introduced at Wave 5

1) I have a lot of choice in deciding what I do at work (Control)2) My working times can be flexible (Control)3) I can decide when to take a break (Control)4) My job requires me to do the same things over and over again (Control)5) My job provides me with a variety of interesting things to do (Control)6) My job requires me to take initiative (Demands)7) I have to work fast in my job (Demands)8) I have to work very intensely in my job (Demands)9) I don’t have enough time to do everything in my job (Demands)

Note: Responses to each item ranged from 0 ‘strongly disagree’ to 6 ‘strongly agree’.

Statistical analyses

The analyses were conducted in four steps. First, a cross-sectional Exploratory Factor Analysis (EFA) was conducted on the 12 job quality items in Wave 1 to identify potential factor solutions. Second, two cross-sectional Confirmatory Factor Analyses (CFAs) were conducted, again using Wave 1 data. The first CFA tested a model with three factors based on Strazdins and colleagues’ (2004) theory that job quality comprises three components: job control, job demands and job security. The specific items hypothesised to load onto each factor are shown in Figure 1. The second CFA tested a model with four factors, based on EFA results suggesting a possible fourth factor. All models allowed for correlations between factors (oblique rotation). The EFAs and CFAs were conducted using the Statistical Package for the Social Sciences (SPSS) and Mplus (Muthen & Muthen 2006).

After demonstrating that a three factor solution was appropriate, Longitudinal Factor Analysis (LFA) was then conducted to determine whether the three factor model fits the data consistently across the seven waves of the HILDA survey. This set of analyses compared a constrained model in which all loadings, intercepts and residual variances were constrained to be equal across all time points against an unconstrained model where these estimates were free to vary across time points. In the case where an unconstrained model fits substantially better than a constrained model, it can be concluded that there are considerable differences across time in item parameters. Items and factors were correlated across waves to account for lack of independence across time. Model fit was evaluated using the Root Mean Square Error of Approximation (RMSEA), the Comparative Fit Index (CFI) and the Tucker Lewis Index (TLI) (Muthen & Muthen 2006). The 2 statistic was not used as an indicator of model fit as it is sensitive to, and therefore


73

inappropriate with, a large number of observations (Muthen & Muthen 2006). The longitudinal factor analyses were conducted using the statistical software Mplus. In both the single wave cross-sectional analyses and the longitudinal analyses, participants with missing data on all items were excluded. Those cases with partial missing data remained in the analyses, and were included in model estimations by adopting the EM algorithm in maximum likelihood estimation (Muthen & Muthen 2006).

Finally, in addition to examining factor scores, subscales for job demands and complexity, job control and job security (and a total scale score representing overall job quality) were constructed by summing the relevant items. The validity of these subscales was assessed in a number of ways. First, each subscale was correlated with the factor scores. Second, we assessed the impact of removing two potentially problematic items. Third, the factor scores and subscales were correlated with the scale scores derived from the new items included in Wave 5.

4 Results

descriptive statistics

Descriptive statistics for the 12 job quality items are provided in Tables 3 and 4. Table 1 shows means for each item at Wave 1. Low scores indicate strong disagreement with the statement, whereas high scores indicate strong agreement. The potential scores for each item ranged from 0 to 6, with a mid-point of 3. Overall, the means were close to the mid-point, with slightly more extreme scores shown for item 2 ‘I fear that the amount of stress in my job will make me physically ill’ (m=1.59), item 5 ‘The company I work for will still be in business in 5 years from now’ (m=4.62), and item 9 ‘I use many of my skills and abilities in my current job’ (m=4.42). Table 2 shows the correlations between items. The correlations between items hypothesised to load onto the same factor (see Figure 1) were all significant (p<0.001), though a few were relatively low. There were also some significant correlations evident between items across factors, suggesting some overlap in the constructs of job demands and complexity, job control and job security.


74

Table 1: Descriptive statistics for each of the items at Wave 1 (n=6,685)

Job quality items Mean (SD)

1. My job is more stressful than I had ever imagined 2.38 (1.67)

2. I fear that the amount of stress in my job will make me physically ill 1.59 (1.68)

3. I get paid fairly for the things I do in my job 3.61 (1.80)

4. I have a secure future in my job 3.77 (1.86)

5. The company I work for will still be in business in 5 years from now 4.62 (2.71)

6. I worry about the future of my job(a) 3.92 (1.93)

7. My job is complex and difficult 2.92 (1.93)

8. My job often requires me to learn new skills 3.70 (1.87)

9. I use many of my skills and abilities in my current job 4.42 (1.63)

10. I have a lot of freedom to decide how I do my own work 3.89 (1.83)

11. I have a lot of say about what happens on my job 3.48 (1.92)

12. I have a lot of freedom to decide when I do my work 2.63 (2.10)

(a) Item 6 has been reverse scored.Note: Reported data is unweighted.


75

Tabl

e 2:

Co

rrel

atio

ns b

etw

een

the

12 it

ems

at W

ave

1

Q1

(D)

Q2

(D)

Q3

(S)

Q4

(S)

Q5

(S)

Q6

(S)

Q7

(D)

Q8

(D)

Q9

(D)

Q10

(C)

Q11

(C)

Q12

(C)

Q1

(D)

–

Q2

(D)

0.64

**–

Q3

(S)

–0.2

0**

–0.1

8**

–

Q4

(S)

–0.0

3*–0

.09*

*0.

24**

–

Q5

(S)

–0.0

2*–0

.07*

*0.

19**

0.50

**–

Q6

(S)(a

)–0

.20*

*–0

.27*

*0.

06**

0.39

**0.

23**

–

Q7

(D)

0.44

**0.

35**

–0.0

8**

0.17

**0.

08**

–0.1

2**

–

Q8

(D)

0.23

**0.

17**

0.00

0.16

**0.

15**

–0.0

7**

0.51

**–

Q9

(D)

0.14

**0.

06**

0.10

**0.

26**

0.18

**0.

020.

37**

0.48

**–

Q10

(C)

–0.0

6**

–0.0

7**

0.17

**0.

23**

0.10

**0.

08**

0.16

**0.

14**

0.36

**–

Q11

(C)

0.01

–0.0

3*0.

16**

0.28

**0.

08**

0.09

**0.

22**

0.17

**0.

36**

0.69

**–

Q12

(C)

–0.0

6**

–0.0

6**

0.14

**0.

12**

–0.0

10.

05**

0.07

**0.

03**

0.15

**0.

57**

0.56

**–

(a)

Item

6 h

as b

een

reve

rse

scor

ed.

Not

e:

*p<

0.05

, **p

<0.

001,

‘–’ n

ot a

pplic

able

. The

fact

or e

ach

item

was

sus

pect

ed to

load

ont

o is

in b

rack

ets

(D—

job

dem

ands

, C—

job

cont

rol,

S—jo

b se

curi

ty).

Gro

upin

gs fo

r eac

h fa

ctor

are

als

o sh

aded

.


76

exploratory factor Analyses (efA) (using Wave 1 data)

The eigenvalues associated with the EFA are presented in Figure 2. The plot provides support for either a three or four factor solution. The largest difference between factors (or the greatest slope shown) was between factors 2 and 3, indicating some support for a three factor solution. However, using the criteria of eigenvalue closest to 1, the four factor solution was the best model. Despite some ambiguity, it was clear that a model with fewer than three, or with more than four, factors was not appropriate.

Figure 2: Scree plot of eigenvalues and factors (based on Wave 1 data)

1 2 3 4 5 6 7 8 9

Factors

Eige

nval

ue

3.0

2.5

2.0

1.5

1.0

0.5

0.0

Additional fit statistics from the EFA suggested that a four factor solution yielded the best fit to the data. Table 3 shows that the eigenvalue, chi-square statistic, RMSEA and RMSR were all substantially lower for a four factor than three factor model. The RMSEA and RMSR for the four factor solution indicated adequate model fit (RMSEA<0.06; RMSR<0.08), however this was not the case for the three factor solution.

Table 3: Model fit indices for the Exploratory Factor Analysis (based on Wave 1 data)

3 factor model 4 factor model

Model fit indicesEigenvalue 1.59 1.01Chi-square 2601.836, p<0.001 375.843, p<0.001RMSEA 0.108 0.047RMSR 0.0496 0.0180


77

confirmatory factor Analyses (cfA) (using Wave 1 data)

Two CFAs were conducted using Mplus to investigate both the three and four factor solutions further. The factor loadings for each of the solutions are shown in Table 4. The four factor solution included a factor comprising items 1 and 2, focused on stress and strain. In the three factor model these items loaded onto the factor representing job demands and complexity along with items 7, 8 and 9.

Table 4: Factor scores for the Confirmatory Factor Analyses (based on Wave 1 data)

3 factor CFA solution 4 factor CFA solution

Factor scores

Factor scores

Job demands and complexity Demands

Q1 My job is more stressful than I had ever imagined 0.543 –

Q2 I fear the amount of stress in my job will make me physically ill

0.454 –

Q7 My job is complex and difficult 0.797 Q7 0.729

Q8 My job often requires me to learn new skills 0.628 Q8 0.690

Q9 I use many of my skills and abilities in my current job

0.493 Q9 0.602

Job security Security

Q3 I get paid fairly for the things I do in my job 0.247 Q3 0.277

Q4 I have a secure future in my job 0.964 Q4 0.901

Q5 The company I work for will still be in business in 5 years from now

0.517 Q5 0.554

Q6 I worry about the future of my job(a) 0.399 Q6 0.416

Job control Control

Q10 I have a lot of freedom to decide how to do my own work

0.827 Q10 0.832

Q11 I have a lot of say about what happens on my job 0.844 Q11 0.840

Q12 I have a lot of freedom to decide when I do my work

0.667 Q12 0.664

Stress/strain

Q1 0.881

Q2 0.726

(a) Item 6 has been reverse scored.


78

The results of the four factor CFA model were a better fit to the data (CFI=0.875, TLI=0.828 and RMSEA=0.095), than the three factor model (CFI=0.747, TLI=0.673 and RMSEA=0.132). However, the modification indices from the three factor solution signalled that the model would be improved if the error terms associated with items 1 and 2 were correlated: these were the same items found to load onto the potential fourth factor in the four factor solution. Allowing the error terms for these two items to be correlated greatly improved the model fit statistics for the three factor model (CFI=0.859, TLI=0.814 and RMSEA=0.100), such that there were negligible differences between the three and four factor solutions. Given the theoretical support (for example, Strazdins et al. 2004) for the three factor solution, the comparability of the models when the error terms between items 1 and 2 were correlated, and concerns about the inclusion of item 2 (see later discussion), it was decided to retain this model in subsequent analyses. This decision was also based on previous research suggesting that job complexity is one aspect of job demands (De Jong, Mulder & Nijhuis 1999).

The fit indices obtained in both the three and four factor CFA models were marginally poorer than those typically used to demonstrate adequate model fit (CFI and TLI >0.95, and RMSEA<0.06). Therefore, the CFA analyses for both models were repeated in SPSS to obtain additional information about the appropriateness of the data for factor analysis. The results from these analyses suggested the data were appropriate for factor analysis: Bartlett’s test of sphericity was significant (X2=23581.525 (66), p<0.001), indicating significant correlations between the variables, and the Kaiser-Meyer-Olkin Measure of Sampling Adequacy was above the recommended value of 0.6 (0.732).

longitudinal factor Analyses (lfA) (using Waves 1–7)

The results of the LFA showed that the three factor model fitted the data equally across all time points. Table 5 shows the fit statistics for both the unconstrained and the constrained models. While values for the TLI and CFI fell just short of the standard criterion of 0.95, the RMSEA indices demonstrated adequate model fit (Bentler & Bonett 1980). Most importantly, there was little variation in the fit statistics between the unconstrained and constrained models (see Table 5). Table 6 shows the factor loadings for the unconstrained model and Figures 3, 4 and 5 plot the loadings across the seven waves of HILDA. As can be seen by Figures 3, 4 and 5, when left unconstrained the factor loadings for each wave were similar, suggesting there was little difference in the factor structure across time points.

Table 5: Fit statistics for each model with the waves unconstrained and constrained (n=9,298)

CFI TLI RMSEA

Unconstrained 0.915 0.899 0.029

Constrained 0.906 0.895 0.029

Note: Adequate model fit indices: CFI≥0.95, TLI≥0.95, RMSEA≤0.06.


79

Table 6: Factor loadings for each item at each wave, when the model was unconstrained

Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 Wave 6 Wave 7

Job demands and complexity

Q1 0.469 0.457 0.495 0.474 0.508 0.470 0.468

Q2 0.351 0.402 0.413 0.384 0.397 0.381 0.380

Q7 0.752 0.741 0.781 0.771 0.786 0.794 0.787

Q8 0.697 0.686 0.675 0.665 0.685 0.678 0.661

Q9 0.579 0.568 0.544 0.531 0.568 0.549 0.533

Job security

Q3 0.261 0.277 0.268 0.298 0.295 0.291 0.304

Q4 0.938 0.933 0.913 0.927 0.949 0.933 0.934

Q5 0.538 0.495 0.482 0.494 0.484 0.464 0.442

Q6 0.397 0.414 0.469 0.437 0.452 0.469 0.414

Job control

Q10 0.824 0.829 0.838 0.835 0.839 0.848 0.847

Q11 0.848 0.840 0.864 0.864 0.863 0.867 0.862

Q12 0.656 0.630 0.639 0.656 0.642 0.637 0.640

Figure 3: Unconstrained factor loadings for job control items across each wave

Item number

Fact

or lo

adin

gs

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

Q10 Q11 Q12

Wave 1 Wave 2 Wave 3 Wave 4

Wave 5 Wave 6 Wave 7


80

Figure 4: Unconstrained factor loadings for job demand (complexity) items across each wave

Q1 Q2 Q7 Q8 Q9

Fact

or lo

adin

gs

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

Item number



Figure 5: Unconstrained factor loadings for job security items across each wave

Fact

or lo

adin

gs

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

Item number

Q3 Q4 Q5 Q6




81

validity assessment of subscales and total scale scoreUsing Wave 1 data, the items representing job demands and complexity, job control and job insecurity (with item 6 reversed) were summed to create subscales. A total score of all 12 items representing overall job quality was also calculated, with the items loading on the job demands scale reversed so that higher scores on each scale represented better job quality. Cronbach’s alpha for each of these subscales and the total scale was adequate and reported in Table 7, along with descriptive statistics. The correlations between the factor scores (based on the three factor solution) and the simple summary scores for each subscale were also calculated. The Pearson’s correlation coefficient between the subscale and the factor score for job demands and complexity was 0.98, for job control was 0.99, and for job security was 0.96. These findings suggest that the simple summary scores adequately represent each of the factors.

Table 7: Descriptive statistics for the total scale, each subscale, and options for item exclusion (at Wave 1)

Cronbach’s alpha

Mean Range Standard deviation

Skewness and Kurtosis

Total job quality scale with all items(a) 0.60 40.70 0–72 9.58 –0.13, 0.19

(excluding item 2) 0.55 36.30 0–66 8.72 –0.01, 0.21

(excluding item 3) 0.57 37.09 0–66 8.92 –0.14, 0.23

(excluding items 2 & 3) 0.51 32.72 0–60 8.01 –0.09, 0.21

Demands subscale 0.72 14.96 0–30 6.14 –0.12, –0.35

(excluding item 2) 0.70 13.37 0–24 5.24 –0.33, –0.34

Control subscale 0.82 9.99 0–18 5.02 –0.21, –0.83

Security subscale 0.59 15.78 0–24 4.97 –0.47, –0.21

(excluding item 3) 0.64 12.20 0–28 4.24 –0.53, –0.37

(a) Items in the demands subscale (1, 2, 7, 8, 9) have been reverse scored before inclusion in the total job quality scale, to ensure that a higher score on the total scale represents better job quality (that is, lower demands, more control and more security).

Table 7 also shows the results of additional analyses that examine the influence of two unusual and potentially problematic items (item 2 and item 3). The mean for item 2 ‘I fear that the amount of stress in my job will make me physically ill’ was considerably lower than for the other job demands items (see Table 1), and the correlations between this item and the other job demand items were also lower (see Table 2). In addition, the wording of the item suggests that it not only assesses the experience of job stress, but also measures the potential health outcomes of highly


82

stressful jobs. Therefore, researchers may wish to exclude this item from scale construction when seeking to examine the association between job adversity and health. While item 3 ‘I get paid fairly for the things I do in my job’ was found to load most strongly on the factor labelled job security, the item clearly measures a somewhat different construct such as effort–reward imbalance. Evidence that the loading of this item was much lower than the other job security items is consistent with this conclusion (see Table 4). Further, the correlations between this item and the other job security items were also low (see Table 2). Therefore researchers seeking to develop a scale that more purely assesses job security may choose to omit this item. The results presented in Table 7 show that there is little difference in the characteristics of these scales if items 2 and 3 are excluded, though alpha does decrease as items are removed.

The job demands and complexity, and job control subscales were also validated against the additional nine job quality items assessing job demands and job control that were introduced to the HILDA survey in Wave 5. Unfortunately, no additional items were introduced assessing job security and therefore, the validity of this factor could not be further investigated. Table 8 shows the correlation between the job demands and complexity, and the job control subscales and the similar subscales derived from the items introduced at Wave 5. The correlations shown in bold are of main interest. Each is statistically significant and their strength (moderate to strong) suggests the items available at all waves provide a valid measure of job demands and control. The cross-domain correlations between the control and demand subscales are much lower and provide evidence of divergent validity, suggesting a clear distinction between the demand and control factors. Correlations between the job demands and control subscales derived from the Wave 5 items, and the factor scores created from items available at all waves, were also examined. The correlations were virtually identical, although slightly weaker, to those displayed in Table 8 for the subscales.

Table 8: Correlations between subscales for items available at all waves, and those introduced at Wave 5

Items available at all waves

Demands and complexity subscale

Demands and complexity subscale

(without item 2)

Control subscale

Items introduced at Wave 5

Demands subscale 0.561** 0.547** 0.053**

Control subscale 0.172** 0.226** 0.775**

Note: **p<0.001.

5 DiscussionGood measurement is fundamental to good research. Our results suggest that the 12 job quality items included in the HILDA survey can be combined to form a valid and psychometrically-sound measure of psychosocial job adversity. Factor analytic techniques were used to assess the


83

factor structure of the items across seven waves of data. The findings showed that the three factor theory of job quality proposed by Strazdins et al. (2004) provided an adequate solution to summarise the 12 items, both cross-sectionally and across time. Consistent with theory, the three factors were labelled: a) job demands and complexity, b) job control, and c) job security.

The CFAs conducted also showed some support for a four factor model, involving the separation of items 1 and 2, and items 7, 8, and 9, within the job demands and complexity factor. However, model fit improved greatly when the error terms for items 1 and 2 were allowed to correlate and, for reasons of parsimony, it was decided to retain the three factor solution. De Jong, Mulder and Nijhuis (1999) define the subcomponents of job demands as including ‘a wide range of qualitative and quantitative aspects of the job, such as working under time pressure, working hard, strenuous work and job complexity’ (p. 1151). In our factor analysis, items 1 and 2 seem to be measuring work stress/strain, whereas items 7, 8, and 9 are assessing job complexity and heavy demands. This conceptualisation of job demands also lends support for the three factor solution in which stress and complexity are aspects of the same construct. To reflect the emphasis on these two aspects, this paper has used the label job demands and complexity for this factor.

using the hIldA survey job quality items

There are a number of ways in which users of the HILDA survey data could apply the current findings to their research and construct valid indicators of psychosocial job quality. First, users could use factor analysis to create factor scores for each of the three factors proposed—job demands and complexity, job control and job insecurity. Factor analysis packages such as Mplus and AMOS allow for the calculation of factors that have correlated error terms, such as was evident for items 1 and 2, allowing for factor scores that may be more representative of the underlying construct. The factor scores generated could then be used separately or combined in subsequent research (see Broom et al. (2006) and Strazdins et al. (2004) for different methods of combining measures of job demands, control and security into a single measure).

Second, users could simply add the items shown to load on each factor to create summed scores representing job demands and complexity, job control and job insecurity (after reversing item 6, which is worded in the opposite direction to the other items in the insecurity scale). This method is cruder than creating factor scores as it does not apply different loadings or relative importance to the individual items in each scale. However, the analyses reported in the current study suggest that these scales adequately represent the more sophisticated factors. This approach is also less statistically demanding and more transparent than generating factor scores, and it avoids the potential difficulty of having different factor loadings in different waves of the study.

limitations

There are some caveats on the interpretation of our analyses that should be recognised. First, it is important to note that there is some ambiguity about how best to treat item 2 ‘I fear that the amount of stress in my job will make me physically ill’ and item 3 ‘I get paid fairly for the


84

things I do in my job’ in the construction of scales/factors. The focus of item 2 is on the health consequence of job stress and, therefore, to avoid potential circularity in the operationalisation of predictors and outcomes, we suggest it would be best to exclude this item from factor and summary scale scores when the outcome of interest is health-related. Similarly, item 3 assesses effort–reward imbalance rather than job security and, therefore, to maintain the integrity of the scale/factor we would suggest excluding this item. However, item 3 could be included as a distinct measure of perceived effort–reward imbalance and contribute to multidimensional measures of job quality.

Second, it is important to acknowledge that the fit indices obtained in both the three and four factor cross-sectional CFA models were somewhat poorer than those typically used to evaluate adequate model fit (CFI and TLI>0.95, and RMSEA<0.06). However, we consider that the fit indices obtained are adequate: they guide our efforts to summarise the items already collected in the HILDA survey in a theoretically and empirically meaningful way for use in subsequent analyses.

A further limitation is that this study focused on the statistical properties of the existing job quality items rather than the conceptual comprehensiveness of the factors identified. While it was shown that the three factors matched the components of job quality hypothesised by Strazdins et al. (2004), and that they demonstrated convergent and divergent validity, our measure of job demands largely reflected job complexity and stress rather than other aspects such as time pressure or workload. Finally, we recognise that other potentially important psychosocial job characteristics, such as social support at work (Griffin et al. 2007; Karasek 1979), have not been measured in the HILDA survey.

6 ConclusionIn future research we plan to use the items available in all waves of the HILDA survey to examine the relationship between psychosocial job quality and health over time (across seven waves). We will also examine the association between these psychosocial assessments of job quality and more objective measures such as income, employment tenure and occupational status. The results of the current study provide confidence that our future analyses based on the HILDA survey job quality items will provide meaningful results that are valid across time.

ReferencesAdam, M & Flatau, PR 2006, ‘Job insecurity and mental health outcomes: an analysis using Wave 1 and 2 of HILDA’, Industrial and Labour Relations Review, vol. 17, no. 1, pp. 143–70.

Bentler, PM & Bonett, G 1980, ‘Significance tests and goodness of fit in the analysis of covariance structures’, Psychological Bulletin, vol. 88, pp. 588–606.


85

Bray, JR 2001, Hardship in Australia: an analysis of financial stress indicators in the 1998–99 Australian Bureau of Statistics Household Expenditure Survey, Occasional Paper no. 4, Department of Family and Community Services, Canberra.

Bosma, H, Marmot, MG, Hemingway, H, Nicholson, AC, Brunner, E & Stansfeld, SA 1997, ‘Low job control and risk of coronary heart disease in Whitehall II (prospective cohort) study’, British Medical Journal, vol. 314, no. 7080, pp. 558–65.

Broom, DH, D'Souza, RM, Strazdins, L, Butterworth, P, Parslow, R & Rodgers, B 2006, ‘The lesser evil: bad jobs or unemployment? A survey of mid-aged Australians’, Social Science and Medicine, vol. 63, no. 3, pp. 575–86.

Chandola, T, Britton, A, Brunner, E, Hemingway, H, Malik, M, Kumari, M, Badrick, E, Kivimaki, M & Marmot, M 2008, ‘Work stress and coronary heart disease: what are the mechanisms?’, European Heart Journal, vol. 29, no. 5, pp. 640–48.

Chandola, T, Brunner, E & Marmot, M 2006, ‘Chronic stress at work and the metabolic syndrome: prospective study’, British Medical Journal, vol. 332, no. 7540, pp. 521–25.

De Jong, J, Mulder, M & Nijhuis, F 1999, ‘The incorporation of different demand concepts in the job demand–control model: effects on health care professionals’, Social Science and Medicine, vol. 48, pp. 1149–60.

De Witt, H & Naswall, K 2003, ‘“Objective” vs “subjective” job insecurity: consequences of temporary work for job satisfaction and organizational commitment in four European countries’, Economic and Industrial Democracy, vol. 24, pp. 149–88.

Dockery, AM 2003, Happiness, life satisfaction and the role of work: evidence from two Australian surveys, School of Economics and Finance Working Paper No 3.10, November, Curtin Business School, Western Australia.

Ferrie, JE 2001, ‘Is job insecurity harmful to health?’, Journal of the Royal Society of Medicine, vol. 94, pp. 71–76.

Green, F 2006, Demanding work: the paradox of job quality in the affluent economy, Princeton University Press, Princeton, New Jersey.

Griffin, JM, Greiner, BA, Stansfeld, SA & Marmot, M 2007, ‘The effect of self-reported and observed job conditions on depression and anxiety symptoms: a comparison of theoretical models’, Journal of Occupational Health Psychology, vol. 12, no. 4, pp. 334–49.

Grzywacz, JG & Dooley, D 2003, ‘"Good jobs" to "bad jobs": replicated evidence of an employment continuum from two large surveys’, Social Science and Medicine, vol. 56,pp. 1749–60.

Karasek, RA 1979, ‘Job demands, job decision latitude, and mental strain: implications for job redesign’, Administrative Science Quarterly, vol. 24, no. 2, pp. 285–308.


86

Kelly, K & Evans, MDR 1999, ‘Australian and international survey data for multivariate analysis: the IsssA’, Australian Economic Review, vol. 32, pp. 298–302.

Long, A 2005, ‘Happily ever after? A study of job satisfaction in Australia’, The Economic Record, vol. 81, no. 255, pp. 303–21.

Noyelle, T 1990, ‘Toward a new labor market segmentation’, in T Noyelle (ed.). Skills, wages and productivity in the service section, Westview Press, Boulder, Colorado, pp. 212–24.

Melbourne Institute of Applied Economic and Social Research 2009, Household, Income and Labour Dynamics in Australia (HILDA) survey, Annual Report 2008, Uni Print, Melbourne.

Muthen, LK & Muthen, BO 2006, Mplus: Statistical analysis with latent variables. User's guide version 4.1, Muthen & Muthen, Los Angeles, California.

Quinlan, A, Mayhew, C & Bohle, P 2001, ‘The global expansion of precarious employment, work disorganisation, and consequences for occupational health: placing the debate in comparative historical context’, International Journal of Health Services, vol. 31, pp. 507–36.

Siegrist, J 2000, ‘Place, social exchange and health: proposed sociological framework’, Social Science and Medicine, vol. 51, no. 9, pp. 1283–93.

Stansfeld, S & Candy, B 2006, ‘Psychosocial work environment and mental health—a meta-analytic review’, Scandinavian Journal of Work, Environment and Health, vol. 32, no. 6,pp. 443–62.

Strazdins, L, D'Souza, RM, Lim, LL, Broom, DH & Rodgers, B 2004, ‘Job strain, job insecurity, and health: rethinking the relationship’, Journal of Occupational Health Psychology, vol. 9, no. 4,pp. 296–305.

Sverke, M, Hellgren, J & Naswall, K 2002, ‘No security: a meta-analysis and review of job insecurity and its consequences’, Journal of Occupational Health Psychology, vol. 1, pp. 9–26.

Wooden, M & Watson, N 2002, The Household, Income and Labour Dynamics in Australia (HILDA) survey: an introduction, University of Melbourne, Melbourne.

87

Sequence matters: understanding the relationship between parental income support receipt and child mortalityPeng Yu


Acknowledgements

The author thanks anonymous referees, colleagues at the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA), and also participants of the Economics Program seminar at the Australian National University, FaHCSIA’s STAR seminar, the National Centre for Social and Economic Modelling (NATSEM) seminar at the University of Canberra, and the 2008 Australian Population Association 14th Biennial Conference, for helpful comments and suggestions. The data used for this research come from the Youth in Focus Project, which was jointly funded by the Australian Government Department of Education, Employment and Workplace Relations, FaHCSIA, Centrelink and the Australian Research Council (Linkage-Project LP0347164), and carried out by the Australian National University. However, the opinions, comments and/or analysis expressed in this paper are those of the author and do not necessarily represent the views of the Australian Research Council or the Minister for Families, Housing, Community Services and Indigenous Affairs.

AbstractPrevious research indicates there is a complex relationship between parental income support receipt and child mortality. This research improves understanding of the relationship using a unique administrative dataset, the Second Transgenerational Data Set (TDS2), which contains information on 127,826 Australian children, almost a whole birth cohort, and their parents. Generally, parents of children who died under age 15 years were more disadvantaged and were on income support for longer periods than were other parents. A robust finding of the research is that the association between child mortality and parental income support receipt varied significantly with the time of the receipt—before, at or after child death. In particular, the incidence of parental income support receipt reduced significantly following the death of a child, probably due to a temporary loss of income support eligibility. The research suggests


88

that income support receipt has more complicated implications than simply as an indicator of economic disadvantage in such a case, and recommends enhanced social and economic support to bereaved parents and families.

Keywords: child mortality, income support, economic disadvantage, Australia

1 IntroductionChild death is a rare event in Australia,1 but when it happens it significantly impacts on families. Mainly due to lack of appropriate data, child mortality is not well explored in Australia outside of the health discipline, especially at an individual level.

Yu (2008) investigated the underlying influencing factors of child mortality at an individual level using a unique administrative dataset, the Second Transgenerational Data Set (TDS2). Consistent with other Australian studies (for example, ABS 2007b; AIHW 2005; AIHW 2006; Draper, Turrell & Oldenburg 2004; Turrell & Mathers 2001), Yu (2008) found an association between increased risk of child mortality and parental disadvantage, as indicated by factors such as low income, Indigenous status, teenage motherhood and living in a disadvantaged neighbourhood. Longer income support duration, often used as an indicator of long-term economic disadvantage, was also found to be associated with higher child mortality. However, using different measures of income support receipt2 leads to different results. In particular, contrary to the estimation of using total income support duration, income support receipt at a point in time—another commonly used indicator of current economic disadvantage—is associated with a significantly lower risk of child mortality. This seemingly contradictory result calls for a closer look at the relationship between parental income support receipt and child mortality. Specifically, it raises several related issues: what does income support mean with respect to child mortality? Does income support serve as an indicator of economic disadvantage, as expected? What are the implications of the relationship between child mortality and income support for social policy?

Income support in this paper, as in Yu (2008), has a narrow and specific meaning. It refers to a range of Australian Government payments managed by Centrelink and targeted at low-income individuals, including Age Pension, Disability Support Pension, Carer Payment, Wife Pension, Parenting Payment Single/Partnered, Newstart Allowance (for the unemployed), Youth Allowance and Sickness Allowance.3 These income support payments have no time limit. They are a flat rate, paid from general government revenue and means tested (including income and asset tests). Some payments, such as Newstart Allowance and Youth Allowance, are also subject to an activity test—a requirement to participate in job searching, training or other approved activities. Australian income support payments are fairly generous. For instance, the basic rate for single recipients of the Disability Support Pension or Parenting Payment Single is $615.80 per fortnight, which is equivalent to earning the minimum wage ($14.31 per hour) for 43 working hours.

In addition, the Australian Government provides various non–income support family payments, such as Family Tax Benefit, Child Care Benefit and Baby Bonus, to help with the costs of caring for children. Except for a small proportion of wealthy families, the majority of Australian families

SequeNce MAtterS: uNderStANdINg the relAtIoNShIP BetWeeN PAreNtAl INcoMe SuPPort receIPt ANd

chIld MortAlIty

89

with children receive some family payments from the Australian Government (although not all of them are eligible for income support payments). Parents on income support are generally eligible for maximum rates of family payments. In the sample drawn from the TDS2, all the families had received some non–income support family payments (with low to middle family income), while over half also received income support at some stage in addition to their family payments (with low income).

The presence of a dependent child is essential for the eligibility of payment in many cases; for instance, Parenting Payment (Single/Partnered) and Carer Payment (if caring for a child with disability) are conditional on having the responsibility of caring for a child. In addition, the number of dependent children is an important factor in meeting the income test and deciding the rate of payment for most income support and family payments. Centrelink requires its customers to report any change of relevant circumstances, including number of children and marital status, within a required timeframe (usually within 14 days).

In the context of this income support system, the correlation between parental income support receipt and child mortality may reflect the compound effects of several factors:

�� As income support is means tested, receipt often indicates low family income—low socioeconomic status is commonly found to be associated with increased child mortality (for example, AIHW 2006; Draper, Turrell & Oldenburg 2004; Turrell et al. 2006).

�� Income support provides extra income to families on low income; alternatively, given the same level of family disposable income, parents receiving income support have more free time to look after their children than those whose income is all from working. Both higher income and longer time of care tend to lower child mortality risk.4

�� Several Australian studies—for example, Butterworth, Crosier & Rodgers (2004) and Kalb (2000)—link welfare receipt with mental health problems and stigma effects, which may indirectly contribute to lower quality of care and thus a higher child mortality risk.5

However, another study (Lee & Oguzoglu 2007) found that the stigma effect associated with income support receipt was not significant for young Australians. As such, on balance, it is inconclusive whether income support receipt should necessarily be associated with a higher or lower risk of child mortality.

One more complexity is that child mortality may also affect parental income support receipt. Without doubt, child death is one of the most stressful life events, and the significant impact it has on parents and family is well documented, especially in psychological literature.6 There may be substantial economic costs associated with the death of a child, and levels of depression, anxiety and stress tend to increase and negatively affect the health conditions, productivity and relationship stability of bereaved parents and family members.7 Together, these factors worsen a family’s financial situation and are likely to increase the incidence of income support receipt.

Another possibility to be noted is that parents may lose their income support payment following the loss of a child if their income support was conditional on caring for the child (for example, Parenting Payment Single recipients).


90

The relationship between parental income support receipt and child mortality is not clear cut, and which effects dominate is an empirical issue. Understanding the nature of the relationship has both methodological and policy implications; for instance, whether a snapshot measure of income support receipt merely indicates economic disadvantage and how disadvantaged families with children can be better helped. This paper uses a Centrelink administrative dataset on a cohort of Australian children, which has full and accurate records of family income support history, to explore the relationship in depth.

In short, the estimation results show that the correlation between two events—parents receiving income support and child mortality—varies significantly with the sequence of these events and also by child gender. At any point in time, previous income support receipt is associated with a higher mortality risk for boys (insignificant) and with a lower mortality risk for girls. Probably due to a temporary loss of child-related income support eligibility, parental income support receipt significantly reduces following the death of a child, whether a boy or a girl; however, the receipt increases gradually but substantially in the long term. As such, income support receipt has more complex implications than simply as a proxy for economic disadvantage—at least in certain circumstances—and enhanced social and economic support to bereaved parents and families is recommended.

The remainder of this paper is structured as follows. Section 2 introduces the dataset, the sample and the methodology and Section 3 reports on and discusses the estimation results. This is followed by the conclusion.

2 Data and methodologyThe data source used for this research is the TDS28 of the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA). The dataset was created in April 2005 from Centrelink administrative records to examine outcomes for children whose parents receive government income support. To facilitate this, the TDS2 links the administrative records of a cohort of almost 130,000 Australian children born between 1 October 1987 and 31 March 1988 to the administrative records of their parents.

The TDS2 provides good coverage of the Australian birth cohort for the period, noting that the data excludes a small number of children from high-income families whose parents never claimed income support or non–income support family payments, at least up until their child’s 17th birthday. Since the TDS2 was extracted from Centrelink administrative records, the information on income support is expected to be accurately recorded, and information on children, as a critical eligibility criterion for benefits, is also likely to be more reliable than in most survey data.

However, infant deaths within the first couple of months after birth are likely to be underrecorded in the TDS2 and this may not be random, as discussed in detail in Yu (2008). As such, children who died within the first two months of birth are excluded from the sample.9 In addition, children born overseas or those whose parent identifiers (unique individual identifications—used to link parents’ records with children’s records) are missing are also excluded. Children in the birth


chIld MortAlIty

91

cohort are referred to as the primary children in the TDS2 and parents who provided the longest care are referred to as the primary parents.10

Most children only appear in the TDS2 as a dependent of their parents, and thus do not have records in their own right. As a result, except for sex, date of birth and date of death, no other child-level variables are available for the entire sample. Therefore, the study is mainly based on the information of primary parents. In most cases in the sample, each primary child is associated with one primary parent, while a few parents are the primary parent of more than one primary child.

The research focuses on the deaths of primary children that happened before their 15th birthday and the TDS2 contains records up to April 2005, beyond the time they would have had their 17th birthday. This allows for at least two years of observation of parental income support receipt for the last recorded death before age 15 years (longer for deaths at younger ages) and makes it possible to investigate the correlation between child mortality and parental income support receipt for a relatively long period of time after the death.

In total, there are 119,414 primary children in the working sample, of whom 61,283 (51.3 per cent) are boys. There are 516 deaths recorded as having happened before age 15 years, of which 306 (59.3 per cent) were boys.

The vast majority of primary parents are female (approximately 97 per cent), non-Indigenous (96.7 per cent), and born in Australia (80 per cent). Among the primary parents of children who died, the percentages of female, non-Indigenous and Australian-born people are 98.6, 94.2 and 84.7 per cent respectively. Therefore, there are relatively larger proportions of female, Indigenous and Australian-born people among parents whose child had died than in the entire sample.

Table 1 provides a statistical summary of the primary parents by gender and death of a child. As shown in the table, in comparison with the parents of children who survived to their 15th birthday, the parents of children who died are significantly more likely to: have income support records, have longer family income support duration, have lower family income, be a non-birth parent,11 have more children, be younger at the first date of care, have more changes in marital status, have more changes in home address, and have more family and individual income support spells as recorded in the TDS2. Overall, these factors indicate that the parents of children who died are relatively more disadvantaged than others.


92


chIld MortAlIty

93

Tabl

e 1:

St

atis

tical

sum

mar

y of

prim

ary

pare

nts

Pare

nts

of a

ll bo

ysPa

rent

s of

boy

s w

ho d

ied

Pare

nts

of a

ll gi

rls

Pare

nts

of g

irls

who

die

d

(n=

60,9

77)

(n=

306)

(n=

57,9

21)

(n=

210)

Fem

ales

(per

cen

t)96

.77

98.6

9*97

.29

98.5

7

Indi

geno

us (p

er c

ent)

3.27

5.56

3.35

6.19

Coun

try

of b

irth

Aus

tral

ian

(per

cen

t)79

.77

83.9

980

.05

85.7

1*

Mai

n En

glis

h-sp

eaki

ng c

ount

ries

(pe

r cen

t)7.

694.

90*

7.64

6.67

Oth

er c

ount

ries

(pe

r cen

t)12

.54

11.1

112

.31

7.62

*

Inco

me

supp

ort r

ecei

pt e

ver i

n th

e TD

S20.

57 (0

.50)

0.74

(0.4

4)*

0.57

(0.5

0)0.

69 (0

.46)

*

Tota

l fam

ily in

com

e su

ppor

t dur

atio

n (d

ays)

1,25

5 (1

,548

)1,

739

(1,6

50)*

1,25

7 (1

,549

)1,

575

(1,6

06)*

Tota

l ind

ivid

ual i

ncom

e su

ppor

t dur

atio

n of

pri

mar

y pa

rent

(day

s)83

4 (1

,250

)1,

202

(1,5

41)*

831

(1,2

28)

932

(1,1

51)

Rela

tive

fam

ily in

com

e (a

s pe

r cen

t of s

ampl

e m

ean)

104.

43 (6

2.72

)93

.95

(69.

09)*

104.

76 (7

1.75

)96

.31

(47.

81)*

Bir

th p

aren

ts0.

75 (0

.43)

0.51

(0.5

0)*

0.76

(0.4

3)0.

47 (0

.50)

*

Tota

l no.

of c

hild

ren

for b

enefi

t pur

pose

3.12

(1.6

9)4.

06 (1

.84)

*3.

12 (1

.72)

4.02

(1.9

9)*

No.

of c

hild

ren

befo

re p

rim

ary

child

1.63

(0.9

7)1.

80 (0

.99)

*1.

63 (0

.98)

1.52

(0.8

1)

Dis

abili

ty0.

03 (0

.18)

0.07

(0.2

6)*

0.03

(0.1

8)0.

04 (0

.19)

Age

at fi

rst d

ate

of c

are

28.6

8 (6

.21)

27.3

4 (5

.69)

*28

.66

(6.1

9)27

.43

(6.3

1)*

Coun

t of m

arit

al e

vent

s in

the

TDS2

3.12

(2.1

3)3.

75 (2

.84)

*3.

12 (2

.11)

3.59

(2.5

7)*

Coun

t of c

hang

es in

hom

e ad

dres

s4.

51 (3

.17)

5.25

(3.9

5)*

4.51

(3.2

1)4.

86 (3

.46)

Coun

t of f

amily

inco

me

supp

ort s

pells

1.25

(1.5

4)1.

58 (1

.54)

*1.

25 (1

.55)

1.53

(1.5

3)*

Coun

t of i

ndiv

idua

l inc

ome

supp

ort s

pells

2.12

(2.7

3)3.

28 (3

.38)

*2.

12 (2

.74)

2.69

(2.8

7)*

Not

e:

Stan

dard

dev

iati

ons

are

in b

rack

ets.

*

Sign

ifica

ntly

diff

eren

t fro

m p

aren

ts o

f chi

ldre

n w

ith

the

sam

e se

x at

the

5 pe

r cen

t lev

el.


92


chIld MortAlIty

93

Figures 1 to 3 respectively compare income support receipt as at the last date of each year from 1985 to 2004 between parents of boys who died and parents of boys who are alive; between parents of girls who died and parents of girls who are alive; and between parents of boys who are alive and parents of girls who are alive. Among the four categories of parents, parents of boys who died stand out with the largest proportion of income support recipients at any point in time, while the other three categories of parents—parents of boys who are alive, parents of girls who died and parents of girls who are alive—are not remarkably different.

Figure 1: Income support recipients—parents of boys who died versus parents of boys who are alive

Per c

ent

Point in time (31 December)

1985 1987 1989 1991 1993 1995 1997 1999 2001 2003

Boys alive Boys who died Differences

0

5

10

15

20

25

30

35

Source: Constructed from the TDS2.


94

Figure 2: Income support recipients—parents of girls who died versus parents of girls who are alive

Per c

ent


1985 1987 1989 1991 1993 1995 1997 1999 2001 2003

Girls alive Girls who died Differences

–5

0

5

10

15

20

25

30


Figure 3: Income support recipients—parents of boys who are alive versus parents of girls who are alive

Per c

ent


1985 1987 1989 1991 1993 1995 1997 1999 2001 2003

Girls Boys

0

5

10

15

20

25



chIld MortAlIty

95

Another point to note is the clear increasing trend in income support incidence over time for both children who died and those who are alive, especially between 1985 and 1997.12 This is due to changes in macroeconomic conditions and the welfare policies of the Australian Government (for example, an economic recession and broadening of the eligibility criteria for parenting payments). This trend illustrates the importance of controlling for time and supports the use of a duration model in the econometric analysis.

A standard Cox Proportional Hazards model was applied for the analysis, to compare the conditional probability of death by parental income support status at the same age of a child. The control variables included parental income support receipt, Indigenous status, country of birth, number of older siblings, teenage motherhood, birth parent, disability, marital status, rural residence, Socio-Economic Indexes for Areas (SEIFA) disadvantage index and remoteness of living areas (see Table A1 in the Appendix for definitions of the variables).13 All variables are from the records of the primary parents. The income support receipt variable, marital status, rural residence, SEIFA index and remoteness may change over time. Since significant gender differences were found in a preliminary analysis using a pooled sample of boys and girls, the final models were estimated separately for them.14

Eight different measures of income support receipt around any point in time (noted as t) are used respectively in the survival analysis:

(1) income support receipt seven to 12 months ago (time t–2)

(2) income support receipt in the last six months (time t–1)

(3) current income support receipt (time t)

(4) income support receipt in the following six months (time t+1)

(5) income support receipt seven to 12 months later (time t+2)

(6) income support receipt so far (time T0 – time t)

(7) income support receipt later (time t – April 2005)

(8) income support receipt ever in the TDS2 (time T0 – April 2005).15

Time T0 refers to the first recorded time in the TDS2, and the last TDS2 records were in April 2005.

By comparing the estimation results using these eight measures, it may be inferred which potential effects of income support receipt discussed in the introduction dominate—that is, indicating economic disadvantage, providing extra resources (time or money) for care, lowering quality of care, or as a result of entrenched and deepened disadvantages or loss of income support eligibility following child death.

Most Centrelink payments are made fortnightly, whereas the death of a child is recorded to a specific date, so the estimate of current income support receipt at time t may show mixed effects of both affecting factors of child death (such as pre-existing disadvantages and extra resources for care) as well as affected factors (such as negative impacts of child death on parental health


96

and income, and loss of income support eligibility). The estimate of income support receipt at time t–1 is relatively less likely to be affected by child death (anticipated or not), and the estimate of income support receipt at time t–2 is even less likely to be affected.

In contrast, the estimate of income support receipt at time t+1 is more likely to show the impact of child death than any measure of income support receipt at previous time periods, and the delayed impact of child death on the parent is likely to be more apparent in the estimate of income support receipt at time t+2.

In addition, it may be expected that the estimate of income support receipt:

�� so far is consistent with the estimates of income support receipt at time t–1 and time t–2

�� afterwards is consistent with the estimates of income support receipt at time t+1 and time t+2

�� ever in the TDS2 show a compound impact of all factors, and should be consistent with the estimates of total duration of income support receipt in the TDS2 as reported in Yu (2008).

3 Estimation resultsTable 2 compares the estimation results of eight duration models for boys and girls where eight different measures of income support receipt around any time t are used respectively.16 Hazard ratios are reported in Table 2; a value greater than 1 implies a positive correlation and a value less than 1 implies a negative correlation.

Table 2: Estimated child mortality risk (2 months–15th birthday) and parental income support receipt around the time

Parental income support receipt around time tEstimated child mortality risk at time t

(hazard ratio)(a)

Boys Girls

(1) 7–12 months ago (time t–2) 1.32 (0.22)* 0.60 (0.15)**

(2) in the last 6 months (time t–1) 1.34 (0.22)* 0.53 (0.14)**

(3) currently on income support (at time t) 0.69 (0.12)** 0.31 (0.09)***

(4) in the following 6 months (time t+1) 0.70 (0.11)** 0.41 (0.09)***

(5) 7–12 months later (time t+2) 0.67 (0.11)** 0.43 (0.10)***

(6) so far (time T0 – t) 1.07 (0.19) 0.55 (0.13)***

(7) later (time t – April 2005) 1.42 (0.21)** 1.09 (0.18)

(8) ever in the TDS2 (time T0 – April 2005) 1.33 (0.20)* 1.12 (0.20)

(a) In comparison to parents not on income support at a given time or period of time.Notes: Standard errors in brackets. * significant at 10 per cent; ** significant at 5 per cent; *** significant at 1 per cent. Other control variables include Indigenous status, country of birth, number of siblings, teenage

motherhood, birth parent, disability, marital status, SEIFA index, remoteness and rural residence. Time T

0 refers to the first recorded time in the TDS2.


chIld MortAlIty

97

On the one hand, noticeable gender differences exist in the pattern of the relationship between mortality risk and parental income support receipt. As shown in the first column of Table 2, boys whose primary parent was receiving income support at time t–1 or at time t–2 are consistently subject to higher mortality risks at time t but the coefficients are only marginally significant at the 10 per cent level (generally insignificant considering the large sample size). In contrast, for girls, as shown in the second column of Table 2, parental income support receipts at time t–1 and time t–2 are consistently correlated with significantly lower child mortality risks.

On the other hand, boys and girls share some common patterns with respect to the relationship. First, for both genders, parental income support receipt at time t and at the following two time periods (time t+1 and time t+2) are all associated with significantly lower mortality risk of children (at least at the 5 per cent level). Second, the correlations between mortality risk and parental income support receipt after time t are positive for both genders and the coefficient is statistically significant for boys. Jointly these two findings indicate that parental income support receipt increases gradually but significantly in the long term after child death. In addition, the results reveal that for both boys and girls the correlations between child mortality and parental income support receipt change dramatically when different measures of parental income support receipt around child death are used.

robust tests

Before discussing the implications of the results, it is worth considering how robust these main findings are. Several exercises were undertaken to determine this.

First, as discussed in Section 2 and also Yu (2008), deaths within the first few months after birth are likely to be underrecorded in the TDS2. If the underrecording is non-random, the estimation may be biased. To tackle this, a method was applied to exclude all deaths recorded in the first two months after birth. However, if deaths after the first two months are also non-randomly underrecorded, the results are still problematic. Therefore, one test was to exclude those deaths within the first six months (the test was repeated for three and four months) after birth; the findings were generally robust.

Second, as children get older, mortality is influenced by more factors outside the family, and the relative importance of parental and family factors may fall accordingly. The reported results in Table 2 are based on analysis of deaths up to the child’s 15th birthday. One robust test, which excluded those who died after age 12 years (randomly chosen), showed similar results. This, along with the first set of tests, suggests that the main results are not sensitive to arbitrary cut-off points of child age.

Third, the working sample includes some middle to high-income families for whom income support receipt might be less relevant or irrelevant, as they only received family payments but were never on income support. Another robust test showed that excluding those who had never received income support does not change the conclusions.


98

Fourth, a period of six months—immediately before and after time t—is used for the estimations reported in Table 2. This period may be too long. Therefore, one test was undertaken by using one month or one fortnight instead of six months immediately before or after time t. The patterns of the correlation between child mortality and parental income support receipt remained the same.

In addition, variations of model specifications were tested. For instance, all models were estimated: first only with a single explanatory variable of parental income support receipt, then including a few demographic variables (parental country of birth, Indigenous status and teenage motherhood), and then by adding all other variables. Some model specifications also use continuous variables of parent age at the first date of care (also age2, age3 and age4) and SEIFA index (also SEIFA2) instead of category variables. Overall, the key findings are robust to these tests.

discussion

Several robust findings emerged from the research, most of which have rarely been reported in the existing literature and thus deserve detailed discussion.

first, previous parental income support receipt is not associated with a significantly higher risk of child mortality.

As discussed in the introduction, in general there are three potential mechanisms through which parental income support receipt may affect child mortality:

�� serving as an indicator of low income

�� working as a protective factor—by providing extra family income or allowing more time for a parent to provide care (with a given family income)

�� receiving income support may be associated with poorer quality of care if the receipt has a stigma effect and leads to low self-esteem, increases levels of stress and depression of parents,17 or is associated with unhealthy lifestyle behaviours, such as drinking, smoking, poor nutrition and physical inactivity.18

The estimation results suggest that the effect of parental income support receipt as a protective factor either balances or outweighs its effects through the other two mechanisms.

Second, child mortality is followed by a reduced incidence of parental income support receipt.

As discussed in the introduction, the death of a child is likely to have a significant impact on parents and correspondingly other members of the immediate and extended family. Grief may negatively affect parental health and relationship stability, and may result in behaviour change and a poorer family financial situation.19 As such, a higher incidence of parental income support receipt after the death of a child would be expected.


chIld MortAlIty

99

However, some parents may lose income support eligibility as a result of the death of a child if the payment—for instance, Carer Payment for a child with disability and Parenting Payment Single—is conditional on caring responsibility for the child (some parents may transfer to other income support payments that do not depend on the presence of an eligible child but are usually less favourable, for example, Newstart Allowance).

As illustrated in Figures 4 and 5, significant differences exist in the threshold of eligible income between Parenting Payment Single/Carer Payment and Newstart Allowance.20 As such, if the income of a single Parenting Payment Single/Carer Payment recipient falls between $777.67 and $1,416.35 per fortnight, or the joint income of a partnered recipient falls between $1,421.68 and $2,328.50 per fortnight, they are likely to lose their income support eligibility after the death of their child because they will not be able to satisfy the income test for Newstart Allowance. In addition, recipients of Parenting Payment Single/Carer Payment and Newstart Allowance are subject to different activity tests, with activity testing in the latter case generally requiring higher levels of participation.

Figure 4: Child death and income support eligibility for single parents

NSA, single, no child

$499.70

$410.60

$777.67 $1,416.35

PPS/CAR, one child

Paym

ent

Income

Note: PPS=Parenting Payment Single; CAR=Carer Payment; NSA=Newstart Allowance.Source: Constructed by the author based on the 2006 income test requirements (Centrelink 2006).


100

Figure 5: Child death and income support eligibility for partnered parents

$417.20

$375.50

NSA, partnered, no child

$1,421.68 $2,328.50

CAR, partnered, one child

Paym

ent

Income

Note: CAR=Carer Payment; NSA=Newstart Allowance.Source: Constructed by the author based on the 2006 income test requirements (Centrelink 2006).

As such, the estimated negative correlation between child mortality and parental income support receipt after the death of a child may be driven by the parent losing income support eligibility. In the TDS2, reasons for ceasing receipt of income support payments are not always recorded, but where they are, the most frequent reasons for primary parents no longer receiving their current income support payment after the death of a child include internal benefit transfer,21 excess income, death of child, no dependent children, base rate of Parenting Allowance (PGA),22 and bereavement period ended.23 These reasons, except for internal benefit transfer, generally support the conjecture of losing income support eligibility as a result of the death of a child.

third, in contrast to the immediate decrease in parental income support receipt after the death of a child, in the long term the receipt is likely to increase substantially.

There are several possible explanations for this finding:

�� The death of a child negatively impacts on parental physical and mental health and relationships,24 and these effects take time to become apparent.

�� Parents may decide to have another baby to ‘replace’ the one who died—replacement effect (Cain & Cain 1964; Olsen 1980)—which may increase their income support incidence.

�� Parents may change their behaviour after losing a child and withdraw from the labour force to spend more time with their remaining children.


chIld MortAlIty

101

Furthermore, income support receipt may indicate a lower ability to find and retain employment. In this case, losing a child may temporarily lead to loss of income support eligibility, but this does not necessarily improve a parent’s employment perspective and in fact may make it worse (Corden 2002; Stebbins & Batrouney 2007). Therefore, some time after the shock of losing a child, parents return to income support by way of different paths.

fourth, the correlations between parental income support receipt and child mortality show gendered patterns.

As shown in Table 2, parental income support receipt in the last six months or previous seven to 12 months is associated with approximately 30 per cent higher death hazard for boys (marginally significant at 10 per cent level), but associated with approximately 40 per cent lower death hazard for girls (significant at 5 per cent level). In addition, parental income support receipt so far (including time t) is associated with a 7 per cent higher death hazard for boys (insignificant) but a 45 per cent lower death hazard for girls (significant at 1 per cent level). These results indicate that parental income support receipt is more likely to be a protective factor for girls than for boys.

How to explain this pattern? Table 3 lists the leading causes of death for two age groups of Australian children in 199125—less than 1 year old, and 1 to 14 years old—and shows that the causes of death are also gendered. Boys have relatively larger probabilities of dying for sudden external reasons (for example, accidents), while girls are more likely to die of long-term illness or disease.26

Notwithstanding the generally decreasing child mortality (ABS 2007a) and also child mortality due to external causes27 (ABS 2004), the gendered pattern in causes of death has not changed in the last decade. According to an ABS report (ABS 2004), the mortality rate for children aged 0 to 14 years is 55.1 per 100,000 persons for boys, 27.8 per cent higher than that for girls (43.1 per 100,000 persons). However, in the same age group, mortality caused by external causes is 9.3 per 100,000 persons for boys, more than 63 per cent higher than figures reported for girls (5.7 per 100,000 persons).The United Nations Children’s Fund (UNICEF) (2001, cited in ABS 2006) reports that it is generally true—regardless of the child’s age and across all Organisation for Economic Co-operation and Development (OECD) countries—that boys are not only more likely than girls to experience an injury, they are also more likely to die as a result of an injury.


102

Table 3: Leading causes of death before age 15 years, by age and sex, 1991

Cause of deathMales Females

No. % No. %Deaths at ages under 1 year old 1,049 100.00 787 100.00

Certain conditions originating in the perinatal period

444 42.33 350 44.47

Congenital anomaly 265 25.26 210 26.68

Sudden death, cause unknown 228 21.73 130 16.52

Deaths at ages 1–14 years 462 100.00 327 100.00

Accidents, poisonings and violence 233 50.43 129 39.45

Motor vehicle traffic accidents 93 20.13 41 12.54

Accidents caused by submersion, suffocation and foreign bodies

66 14.29 33 10.09

Malignant neoplasms 58 12.55 55 16.82

Diseases of the nervous system and sense organs

47 10.17 34 10.40

Congenital anomalies 40 8.66 36 11.01

Source: ABS (1994), Tables 6–8, p. 27.Note: For simplicity, only leading causes are listed. As such, the subcategories may not add up to the totals.

In comparison to deaths caused by diseases, deaths as a result of sudden external reasons may be more likely to happen in a disadvantaged family living in a poor neighbourhood where children are exposed to higher risks of accidents and injury (for example, traffic accidents), have poorer access to health services, and also may experience lower quality parenting (Blakemore 2007; Shore 1997). Since boys are more likely to engage in high-risk behaviour than girls (Blakemore 2007), the injury and death risks of boys may be notably higher if they live in a disadvantaged family in a disadvantaged neighbourhood. Longer caring time but lower quality care associated with parental income support receipt is probably less effective in offsetting the increased risks for boys. It should be noted that these interpretations of the gendered pattern are far from conclusive and the issue deserves further exploration.


chIld MortAlIty

103

4 ConclusionsThis paper examines in greater detail the complex relationship between parental income support receipt and child mortality observed in Yu (2008)—one measure of income support receipt (total income support duration) is associated with higher child mortality risk, while another (income support receipt at a point in time) is associated with lower child mortality risk, although these two measures are positively correlated. The investigation is made possible by an administrative dataset, the TDS2, which contains a full, detailed and accurate family income support history of nearly a whole birth cohort of Australian children up to April 2005, when the children turned 17 years old. The hazards of death at any time before age 15 years are compared by parental income support status using a duration model.

The most striking finding of the research is that parental income support receipt decreases significantly following the death of a child rather than increases as would be expected. A plausible interpretation of this finding is the loss of the primary parent’s income support eligibility after losing a child. If this is true, then the parent suffers not only psychologically, but also economically. Considering the low incidence of child mortality and the significance of its impact on families, the benefits of enhanced social and economic support to bereaved families would be much higher than its costs. Further investigation on this point is warranted.

A related conclusion from this finding is that sequence matters with respect to the correlation between child mortality and parental income support receipt. Using different measures of income support before and after the death of a child leads to dramatically different results. This finding may be also valid for other significant life events and other measures of disadvantage. In many datasets, records of some significant events may not be available, and even when they are the dates of the events may not be accurately recorded. Whether the measure was taken before or after a significant event (such as the death of a child) may have very different implications on the research findings.

In addition, while on average the death of a child leads to a lower incidence of parental income support receipt immediately afterwards, in the long term parental income support receipt is likely to increase significantly. Several plausible explanations for this were provided in the last section. Whatever the reasons, the results suggest that the death of a child puts parents and family under greater risks of disadvantage, and thus reaffirm that early and ongoing social and economic support is badly needed to help bereaved parents and families recover.

Furthermore, the results reveal a gendered pattern in the correlation between child mortality and parental income support receipt, which also requires further investigation. Generally, the death of a boy is more likely to be caused by sudden external causes (for example, accidents) rather than long-term illness or disease (as it is for girls). Longer caring time, but poorer quality care associated with parental income support receipt, may also be less effective in reducing deaths caused by accidents and injuries, indicating that parental income support receipt may be less protective for boys than for girls. As such, the results highlight the importance of taking the increased vulnerability of boys to accidental death into account when targeting policy intervention.


104

AppendixTable A1: Definition of key variables in the survival analysis

Variable Definition

Parental income support receipt

Income support receipt ever in TDS2

1, if a primary parent had received an income support payment as recorded in the TDS2; 0, otherwise

Income support receipt so far 1, if up to a point in time, primary parent had income support records in the TDS2; 0, otherwise

Income support receipt later 1, if after a point in time, primary parent had income support records in the TDS2; 0, otherwise

Currently on income support (at time t)

1, if at a point in time t, primary parent was recorded as being on an income support payment; 0, otherwise

Income support receipt in the last 6 months (time t–1)

1, if within the last 6 months before a point in time, primary parent had been on an income support payment; 0, otherwise

Income support receipt 7–12 months ago (time t–2)

1, if within the period of 6 to 12 months before a point in time, primary parent had been on an income support payment; 0, otherwise

Income support receipt in the following 6 months (time t+1)

1, if within the following 6 months after a point in time, primary parent had been on an income support payment; 0, otherwise

Income support receipt 7–12 months later (time t+2)

1, if within the period of 6 to 12 months after a point in time, primary parent had been on an income support payment; 0, otherwise

Other variables

Indigenous 1, if a primary parent self-identified as being Indigenous; 0, otherwise

Country of birth of primary parent

1, Australia; 2, main English-speaking countries (Canada, Ireland, New Zealand, South Africa, the United Kingdom, the United States); 3, other countries

Number of older siblings Count of children a primary parent ever cared for before a primary child

Teenage motherhood 1, if a primary parent is female and younger than 20 years old when the primary child who cared for was born; 0, otherwise


chIld MortAlIty

105

Table A1: Definition of key variables in the survival analysis (continued)

Variable Definition

Birth parent 1, if a primary parent started to take care of a primary child from the date of birth; 0, otherwise

Disability 1, if a primary parent had a medical condition that attracted an impairment rating of 20 points or more; 0, otherwise

Marital status of primary parent 1, single; 2, partnered

Rural residence, identified with postcode

1, if most population with the postcode live in rural area (based on Census data); 0, otherwise

SEIFA disadvantage index Based on the Index of Relative Socio-economic Disadvantage, one of the Socio-Economic Indexes for Areas developed by the ABS from Census data. Matched with postcode.

Remoteness Based on the Australian Standard Geographical Classification remoteness classification. Matched with postcode.0, major city; 1, inner regional area; 2, outer regional area; 3, remote area; 4, very remote area.

Endnotes1 According to the ABS (2007a), in 2006 the age-specific death rates for males of

0 to 12 months of age, 1 to 4 years old, and 5 to 14 years old were respectively 5.3, 0.2, and 0.1 deaths per 1,000 population; the rates for females of these three age groups were respectively 4.1, 0.2 and 0.1 deaths per 1,000 population.

2 For instance, income support receipt at a point in time, income support duration while caring for a child, income support duration before the last date of care, income support duration since the first date of care, and income support duration in a fixed window (1987–2005)—as used in Yu (2008).

3 See Whiteford (2000) for an overview of the Australian income support system. Refer to <www.centrelink.gov.au> for details of the payments delivered by Centrelink.

4 For instance, using aggregate data from 16 European countries over the 1969 through 1994 period, Ruhm (2000) finds that rights to parental leave are associated with substantial decreases in deaths of infants and young children, and suggests parental time is an important input into the wellbeing of children. Baum (2003) finds that maternal employment in the first year of life has detrimental effects on child development, which are partially offset by the positive effects of increased family income from employment.


106

5 Welfare receipt is often associated with unemployment, which is usually found to be a risk factor for health and mortality in the literature—for instance, Dooley, Fielding and Levi (1996), Gerdtham and Johannesson (2003), Johansson and Sundquist (1997), and Mathers and Schofield (1998).

6 See, for instance, Corden (2002), Dijkstra and Stroebe (1998), Dyregrov and Dyregrov (1999), Goodenough et al. (2004), Li et al. (2003) and Najman et al. (1993).

7 A recent study by Stebbins and Batrouney (2007) conducted in-depth research on the topic of the social impact and economic costs of the death of a child on the family during the first three years following the death. After interviewing 103 bereaved families, they estimated that the economic costs of child death include about $3,160 out-of-pocket medical and health-related expenses, $3,800 one-off costs associated with funeral expenses, and $59,500 lost income from employment. The social impact for parents is also significant, and includes disharmony and arguments, less support and closeness between partners, and a dramatic decrease in both the frequency and importance of social activities.

8 For detailed background information on the data, refer to Breunig et al. (2007); for a detailed discussion of the advantages and limitations of the data, refer to Yu (2008).

9 A few robust tests have been undertaken for this issue, such as including all primary children, or excluding children who died within the first three or four or six months. The findings are generally consistent.

10 One point to note is that a ‘parent’ recorded in the TDS2 for benefit purposes is not necessarily the biological parent of a child (although in most cases this is true). It can be their grandparent, older sibling, relative or any other person who acted as the primary guardian and also claimed benefits for the child.

11 In this research, birth parent is defined as a primary parent who started taking care of a primary child from the birth of the child.

12 This increasing pattern is consistent with the overall picture of income support receipt in Australia (see FaCSIA 2006), although the sample in the TDS2 is not directly comparable with the income support recipients reported in FaCSIA (2006)—the former refers to the same cohorts, while the latter includes new entrants.

13 For details of the SEIFA index and remoteness classification, refer to the ABS website at <www.abs.gov.au>.

14 With the pooled sample the differences in the associations between child mortality and parental income support receipt around time t are more significant, and boys consistently have significantly higher mortality risks than girls.

15 Time t is equivalent to the age of a child, time t–1 refers to the period of time immediately before time t (for example, the last six months), and time t–2 refers to the period of time before the last one (time t–1).


chIld MortAlIty

107

16 Full estimation results of the paper can be provided on request. In the estimation, time t is equivalent to the age of a child (all the primary children are roughly at the same age), time t–1 refers to the period of time immediately before time t (for example, the last six months), and time t–2 refers to the period of time before the last one.

17 See Lee and Oguzoglu (2007) and Kalb (2000) for stigma effects of welfare receipt in Australia, and Bingley and Walker (1997), Edin and Lein (1997), Hoynes (1996), Moffitt (1983), and Rodgers-Dillon (1995) for other countries.

18 An analysis by the author on a survey of young people aged 18 years, a stratified random sample of the TDS2 children, showed that intensive family income support receipt was significantly associated with young people’s smoking, illicit drug use and less frequent participation in moderate or intensive physical activity. McGinnis and Foege (1993) estimated that three lifestyle variables—tobacco, diet and activity, and alcohol consumption—explained around 38 per cent of premature mortality in the United States in 1990.

19 See references listed in endnotes 6 and 7.

20 The demonstration is based on the 2006 income test requirements (see Centrelink 2006).

21 Transferring from one income support payment to another income support payment.

22 Base rate PGA is a non–income support payment superseded by Family Tax Benefit Part B, on 1 July 2000.

23 Other reasons include ‘fail to reply to correspondence’, ‘did not lodge form’, ‘departure overseas permanently’ and ‘full-time employment’.

24 For instance, prevalence of depression and stress and increased mortality of bereaved parents are reported in Goodenough et al. (2004) and Li et al. (2003). Najman et al. (1993) also show that after the death of an infant, marital relationships deteriorate and marital break-up rates increase.

25 Data for earlier years are not available, and data for later years show a generally similar pattern.

26 Recent annual report of the New South Wales Child Death Review Team (2007) also shows that in New South Wales girls were more likely than boys to die of diseases and morbid conditions (80.0 per cent of females compared with 64.8 per cent of males), while boys were much more likely to die of external causes or ill-defined and unknown causes.

27 However, over 1998–2002, age-specific death rates of external causes of death increased in the age group of 15 to 19 years (ABS 2004, p. 21).


108

ReferencesAustralian Bureau of Statistics (ABS) 1994, Causes of death, Australia, 1993, cat. no. 3303.0, ABS, Canberra.

——2004, Deaths from external causes, 1998 to 2002, cat. no. 3320.0, ABS, Canberra.

——2006, Yearbook Australia, cat. no. 1301.0, ABS, Canberra.

——2007a, Deaths, Australia, 2006, cat. no. 3302.0, ABS, Canberra.

——2007b, Health of children in Australia: a snapshot, 2004–05, cat. no. 4829.0.55.001, ABS, Canberra.

Australian Institute of Health and Welfare (AIHW) 2005, The health and welfare of Australia’s Aboriginal and Torres Strait Islander peoples, ABS cat. no. 4704.0; AIHW cat. no. IHW14, ABS & AIHW, Canberra.

——2006, Australia’s health 2006, cat. no. AUS 73, AIHW, Canberra.

Baum, CL 2003, ‘Does early maternal employment harm child development? An analysis of the potential benefits of leave taking’, Journal of Labor Economics, vol. 21, no. 2, pp. 409–48.

Bingley, P & Walker, I 1997, ‘The labour supply, unemployment and participation of lone mothers in in-work transfer programs’, Economic Journal, vol. 107, pp. 1375–90.

Blakemore, T 2007, ‘Examining potential risk factors, pathways and processes associated with childhood injury in the Longitudinal Study of Australian Children’, Australian Social Policy 2006, pp. 27–52.

Breunig, R, Cobb-Clark, D, Gorgens, T & Sartbayeva, A 2007, User’s guide to the Youth in Focus data version 1.0, Youth in Focus Project Discussion Paper Series No. 1, Australian National University, Canberra, <http://youthinfocus.anu.edu.au/publications.htm>.

Butterworth, P, Crosier, T & Rodgers, B 2004, ‘Mental health problems, disability and income support receipt: a replication and extension using the HILDA survey’, Australian Journal of Labour Economics, vol. 7, no. 2, pp. 151–74.

Cain, AC & Cain, BS 1964, ‘On replacing a child’, Journal of the American Academy of Child Psychiatry, vol. 8, pp. 443–56.

Centrelink 2006, A guide to Australian Government payments, 1 July–19 September 2006, Centrelink, Canberra.

Corden, A 2002, ‘Financial effects for families after the death of a disabled or chronically ill child’, Child Care, Health and Development, vol. 28, no. 3, pp. 199–204.

Dijkstra, IC & Stroebe, MS 1998, ‘The impact of a child’s death on parents: a myth (not yet) disproved?’, Journal of Family Studies, vol. 4, no. 2, pp. 159–85.


chIld MortAlIty

109

Dooley, D, Fielding, J & Levi, L 1996, ‘Health and unemployment’, Annual Review of Public Health, vol. 17, pp. 449–65.

Draper, G, Turrell, G & Oldenburg, B 2004, Health inequalities in Australia: mortality, Health Inequalities Monitoring Series no. 1, cat. no. PHE 55, Queensland University of Technology and the AIHW, Canberra.

Dyregrov, A & Dyregrov, K 1999, ‘Long-term impact of sudden infant death: a 12- to 15- year follow-up’, Death Studies, vol. 23, no. 7, pp. 635–61.

Edin, K & Lein, L 1997, Making ends meet: how single mothers survive welfare and low-wage work, Russell Sage Foundation, New York.

FaCSIA (Australian Government Department of Families, Community Services and Indigenous Affairs) 2006, Income support customers: a statistical overview 2002, Statistical Paper no. 1, FaCSIA, Canberra.

Gerdtham, U & Johannesson, M 2003, ‘A note on the effect of unemployment on mortality’, Journal of Health Economics, vol. 22, pp. 505–18.

Goodenough, B, Drew, D, Higgins, S & Trethewie, S 2004, ‘Bereavement outcomes for parents who lose a child to cancer: are place of death and sex of parent associated with differences in psychological functioning?’, Psycho-Oncology, vol. 13, pp. 779–91.

Hoynes, HW 1996, ‘Welfare transfers in two-parent families: labor supply and welfare participation under AFDC-UP’, Econometrica, vol. 64, no. 2, pp. 295–332.

Johansson, SE & Sundquist, J 1997, ‘Unemployment is an important risk factor for suicide in contemporary Sweden: an 11-year follow-up study of a cross-sectional sample of 37789 people’, Public Health, vol. 111, pp. 41–45.

Kalb, G 2000, Labour supply and welfare participation in Australian two-adult households, Department of Econometrics and Business Statistics Working Paper no. 35, Monash University, Melbourne.

Lee, W & Oguzoglu, U 2007, ‘Income support and stigma effects for young Australians’, Australian Economic Review, vol. 40, no. 4, pp. 369–84.

Li, J, Precht, DH, Mortensen, PB & Olsen, J 2003, ‘Mortality in parents after death of a child in Denmark: a national follow-up study’, Lancet, vol. 361, pp. 363–67.

Mathers, CD & Schofield, DJ 1998, ‘The health consequence of unemployment: the evidence’, Medical Journal of Australia, vol. 168, pp. 178–82.

McGinnis, J & Foege, WH 1993, ‘Actual causes of death in the United States’, Journal of the American Medical Association, vol. 279, pp. 2207–12.

Moffitt, R 1983, ‘An economic model of welfare stigma’, American Economic Review, vol. 73,pp. 1023–35.


110

Najman, JM, Vance, JC, Boyle, F, Embleton, G, Foster, B & Thearle, J 1993, ‘The impact of a child death on marital adjustment’, Social Science & Medicine, vol. 37, no. 8, pp. 1005–10.

NSW Child Death Review Team 2007, Annual Report 2006, prepared by Virginia Winter & James Gosley, NSW Commission for Children and Young People, Sydney.

Olsen, R 1980, ‘Estimating the effect of child mortality on the number of births’, Demography, vol. 17, no. 14, pp. 429–43.

Rodgers-Dillon, R 1995, ‘The dynamics of welfare stigma’, Qualitative Sociology, vol. 18,pp. 439–56.

Ruhm, CJ 2000, ‘Parental leave and child health’, Journal of Health Economics, vol. 19, pp. 931–60.

Shore, R 1997, Rethinking the brain: new insights into early development, Families and Work Institute, New York.

Stebbins, J & Batrouney, T 2007, Beyond the death of a child, research summary report on the social impacts and economic costs of the death of a child, The Compassionate Friends Victoria Inc., Canterbury.

Turrell, G & Mathers, C 2001, ‘Socioeconomic inequalities in all-cause and specific-cause mortality in Australia: 1985–1987 and 1995–1997’, International Journal of Epidemiology, vol. 30, pp. 231–39.

Turrell, G, Stanley, L, de Looper, M & Oldenburg, B 2006, Health inequalities in Australia: morbidity, health behaviours, risk factors and health service use, Health Inequalities Monitoring Series no. 2, cat. no. PHE 72, Queensland University of Technology and the AIHW, Canberra.

UNICEF 2001, ‘A league table of child deaths by injury in rich nations’, Innocenti Report Card, no. 2, UNICEF Innocenti Research Centre, Florence.

Whiteford, P 2000, The Australian system of social protection—an overview, Policy Research Paper No. 1, Australian Government Department of Family and Community Services, Canberra.

Yu, P 2008, ‘Mortality of children and parental disadvantage’, Australian Social Policy, no. 7, pp. 1–60.

111

Regional living and community participation: are people with disability at a disadvantage?Samara McPhedran


Acknowledgements

This paper uses unit record data from the Household, Income and Labour Dynamics in Australia (HILDA) survey. The HILDA project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (Melbourne Institute). The findings and views reported in this paper, however, are those of the author and should not be attributed to either FaHCSIA or the Melbourne Institute. The opinions, comments, and/or analyses expressed in this paper do not necessarily represent the views of the Minister for Families, Housing, Community Services and Indigenous Affairs, and cannot be taken in any way as expressions of Government policy.

The author thanks Paula Mance and the Research Projects and Publications Section (Research and Analysis Branch), Dr Andrea Lanyon (Research and Analysis Branch), and Marcia Kingston and Ruth Ganley (Disability and Carers Payments Branch) for their valuable comments on earlier drafts of this work.

AbstractThere is considerable evidence that people with disability are at a heightened likelihood of experiencing disadvantage in many facets of life. A greater likelihood of disadvantage may in turn increase the risk of exclusion from a range of opportunities, including social participation. However, factors that may lead to ‘double disadvantage’ among people with disability—such as living outside major cities—have not been well assessed in Australia in relation to social connectedness. The current study compared various socioeconomic, life satisfaction, community participation and social support measures among prime working age regional people with and with no disability. People with disability experienced greater relative disadvantage and reported lower levels of perceived social support compared with people with no disability. Irrespective of disability status, men in regional Australia reported lower levels of social support than women. However, engagement in community activities such as volunteering did not differ as a function of disability status. This in turn suggests potential avenues for consideration in terms of


112

strengthening social connectedness among regional people with disability, and addressing the risk of social exclusion for this group.

Keywords: disability; social participation; regional; community; social support

1 IntroductionThere is growing awareness of associations between disability and social exclusion. While much of the literature to date has focused on outcomes for people with intellectual or developmental disability, disability is a multidimensional concept that encompasses impairments with physical function or structure (for example, blindness), activity limitations (such as needing help with mobility), and/or restrictions in participation (for example, attending school or work, having opportunities for social interaction). Although disability is not a necessary or sufficient condition for social exclusion, there is considerable evidence that people with disability are at a heightened likelihood of experiencing disadvantage in many facets of life. Two well-studied areas where people with disability can be excluded from full participation are education and employment. These two areas have been covered by a substantial body of Australian and international literature, the diversity and findings of which are beyond the scope of the current study to review.1

In contrast, the social dimensions of disability, such as the relationship between disability and social connectedness (or social isolation—the lack of connectedness), have not received a great deal of attention. While there are a number of documented benefits of social connectedness for individual and community wellbeing, relatively little Australian research has examined social connectedness in the context of co-occurring indicators of potential disadvantage. This gap in knowledge may reflect, in part, the conceptual and analytical difficulties of determining the relative contribution to a given outcome of each of a number of (often interrelated) indicators of disadvantage. Poor education, for instance, may simultaneously serve as an indicator of disadvantage, an outcome of disadvantage and a contributor to disadvantage.

In the context of disability research, the challenge of disentangling pathways between different indicators of disadvantage is compounded by the dearth of information about the direction of the relationship between disability and disadvantage. While disability may increase the risk of experiencing disadvantage, it is also possible that disadvantage may increase the risk of disability. Understanding the direction of the relationship, as well as its interaction with the severity of disability, has significant application within policy development and program design, and future work is planned to provide information in this area. The current study, however, focused on whether two broad factors that have been associated with disadvantage have a cumulative impact on social connectedness. This work goes part of the way to addressing how different indicators of disadvantage each contribute to a specific set of social participation outcomes.

Although existing work suggests that social connectedness can be affected by a nuanced interplay of socioeconomic and demographic factors, situational factors that may affect social connectedness for people with disability—such as location—have not been well considered in an Australian setting. Location can affect access to services, infrastructure, training and

regIoNAl lIvINg ANd coMMuNIty PArtIcIPAtIoN: Are PeoPle WIth dISABIlIty At A dISAdvANtAge?

113

employment opportunities, and could also affect social connectedness. This may occur directly, by increasing the distance that must be travelled to access social opportunities, or indirectly, for instance, by affecting economic participation opportunities, in turn limiting the resources available to facilitate social participation.

The proposition that location can detrimentally influence social connectedness draws upon the ‘double disadvantage’ theory. The double disadvantage theory suggests that health-related disadvantage (for example, limits on the type or amount of paid work that can be undertaken) is exacerbated by locational disadvantage (such as a lack of services in a particular area). It has been proposed that people with disability who live outside major cities may, due to the combination of disability and location, face a heightened risk of disadvantage (Gething 1997). The term double disadvantage is reminiscent of the often-encountered phrase ‘multiple disadvantage’. However, multiple disadvantage is typically used in a descriptive sense to indicate the presence of more than one source of potential disadvantage, whereas double disadvantage is used herein to denote a testable hypothesis. Specifically, it expresses the hypothesis that when disability and location co-occur, those factors have a cumulative influence on social connectedness (see McPhedran (in press) for additional discussion of this issue).

The double disadvantage theory rests on three testable assumptions. The first assumption is that disability and disadvantage are related, which has been amply demonstrated through existing research (for example, Australian Institute of Health and Welfare 2008; Jenkins & Rigg 2004; Mavromas et al. 2007; Winn & Hay 2009). The next assumption is that location and disadvantage are related, for which there is also broad support. For example, on measures such as employment and income, rural and regional2 Australians are generally less advantaged than their counterparts in major cities. In recent decades, Australians living in regional and rural areas (particularly areas traditionally connected with agricultural production) have faced increasing levels of unemployment, poor health, and financial hardship—factors that reflect in part the impacts of drought (Alston & Kent 2004; Hall & Scheltens 2005), demographic trends (an ‘ageing community’ and negative population growth), primary production policy changes, increasing urbanisation, and a shift away from rural industries (Archer 2000; Talbot & Walker 2007).

The final premise of the double disadvantage theory is that disability and locational disadvantage have cumulative negative impacts; that is, people with disability who live outside major cities will fare less well than either people with disability in major cities or people with no disability who live outside major cities. However, even if the double disadvantage theory applies to socioeconomic circumstances, it is not known whether this extends to social connectedness—and if so, in what ways. Consequently, it is important to determine whether the double disadvantage theory applies to social participation, and, if so, how. This knowledge can in turn inform ways to support the ability of regional people with disability to participate more fully in community life.

Previous work has addressed parts of the double disadvantage theory of disability and regional living. McPhedran (in press) found that relative to people with disability in major cities, people with disability in regional Australia were less likely to be employed, were more likely to receive income support, and had a lower likelihood of attaining Year 12 or tertiary education. Additionally, a number of bivariate differences were found between regional and city respondents in terms of life satisfaction and social participation.


114

For example, regional respondents reported lower satisfaction with their employment opportunities and financial situation, less frequent non–face to face contact with family and friends and less often making time to keep in touch with friends. However, these differences were mitigated by the inclusion of socioeconomic factors (particularly labour force status). This suggests that for people with disability, location may have an indirect influence on certain types of social connectedness and specific aspects of life satisfaction, via locational differences in socioeconomic circumstances. Similarly, locational differences in feelings of needing help but not being able to get it (which was experienced more frequently by regional respondents with disability) were accounted for by the inclusion of education level (respondents with a Year 11 or lower education were less likely to feel that they could access help).

While socioeconomic differences accounted for many of the bivariate relationships between location and certain types of social participation, regional living continued to be associated with higher satisfaction with feeling part of the local community, stronger perceived neighbourhood relationships (such as a close-knit neighbourhood where people are willing to help one another and can be trusted), and more frequent community involvement in the form of volunteering and attending community events. These findings suggest that while locational socioeconomic disadvantage may indirectly affect various forms of social connectedness for people with disability, such as having contact with friends, location can also be positively associated with other types of social connectedness, such as community engagement. This accords with the social cohesion theory of regional living, which proposes that regional locations provide a stronger sense of ‘community’ for residents, relative to major cities. A body of literature supports this theory (Boyd et al. 2008; Gething 1997; Obst, Smith & Zinkiewicz 2002), but the possible co-existence of social cohesion and double disadvantage has not been considered.

Although McPhedran (in press) demonstrated locational differences in various socioeconomic and social indicators when disability status was held constant (that is, when all respondents had a disability, but location varied), that study did not examine disability-related differences in socioeconomic and social connectedness indices when location is held constant. This information is crucial for assessing the third premise of the double disadvantage hypothesis—that disability and location have cumulative negative impacts. To demonstrate this, it is necessary to identify items on which regional people with disability fare less well than people with disability in major cities and people with no disability in regional areas. Similarly, to examine the social cohesion theory it must be established whether levels of community engagement through activities such as volunteering are comparable between regional people with and with no disability.

Therefore, the current study further tested the double disadvantage and social cohesion theories, and their potential application to measures of social connectedness for regional people with disability. It assessed the frequency with which different types of community involvement occurred, levels of perceived social support, neighbourhood relationships and self-reported life satisfaction among working age regional people with disability, with emphasis on comparing their experiences with those who did not report having a disability. This provided information about differences in social connectedness as a function of disability status, to assist in disentangling associations between disability and social connectedness from relationships between location and social connectedness.


115

The research had three specific aims. First, it compared life satisfaction, community participation and social support among prime working age regional people, by disability status. Second, it assessed the extent to which differences on those measures could be accounted for by the inclusion of demographic and socioeconomic variables. Finally, it tested the double disadvantage hypothesis by examining whether disability status was associated with differences on the specific socioeconomic, life satisfaction and social connectedness indices where locational variation had previously been observed for people with disability. The specific indices identified were: education level, labour force status, income support receipt, equivalised income, satisfaction with employment opportunities, satisfaction with financial situation, frequency of non–face to face contact with family/friends, frequency of making time to keep in touch with friends, and the frequency of needing help but not being able to get it.

2 Methods

data source and sample

Sample selection procedures and methods have been described in detail elsewhere (McPhedran in press). Briefly, data were drawn from Wave 6 (2006) of the Household, Income and Labour Dynamics in Australia (HILDA) survey, for respondents aged between 18 and 45 years (that is, prime working age) who lived in regional Australia. Location was defined using the Accessibility/Remoteness Index of Australia (ARIA) (Australian Bureau of Statistics (ABS) 2001), with the categories of ‘inner’ and ‘outer’ regional collapsed into one group to ensure sufficient sample size.

Respondents were classified as having a disability if they reported a long-term health condition, impairment or disability that restricts them in their everyday activities and has lasted, or is likely to last, for six months or more. HILDA adopts a broad definition of disability, drawn from ABS surveys. In this sense, chronic illness is recognised as a form of disabling condition. It should be noted that this broad approach does not reflect the method used to assess disability for the purposes of determining Disability Support Pension (DSP) eligibility, and will result in elevated estimates of disability prevalence relative to studies that focus on profound disability or core activity limitations (that is, self-care, mobility and communication). However, this is not considered to be a significant constraint to the analyses herein, given it was not the purpose of the current study to estimate the prevalence of disability, or to compare prevalence estimates derived using different definitions and data sources.

demographic and socioeconomic information

Age, sex and marital status were considered. Marital status was coded into one of five groups: legally married, living with someone in a relationship (de facto), divorced/separated, not living in a relationship, or widowed. The highest education level attained by respondents was recoded into one of four categories: Year 11 or below, Year 12, certificate or diploma,3 and tertiary.


116

Respondents were classified into one of three broad labour force categories: employed, unemployed and not in the labour force (NILF). For the purpose of descriptive analysis these categories were further disaggregated into employed—full time, employed—part time, NILF—marginally attached to the labour force, and NILF—not marginally attached to the labour force. In the interests of maximising cell sizes, subsequent analyses used only the three broad categories.

Three financial variables were assessed: equivalised household income (based on gross financial year financial income), financial perception and receipt of income support. Income was treated as a continuous variable, and equivalised using procedures described elsewhere (ABS 2003). Financial perception (how well individuals believe they and their family are faring financially, given their current needs and responsibilities) was coded dichotomously, with 0 representing responses from ‘reasonably comfortable’ through to ‘prosperous’ (classed as ‘doing well’), and 1 indicating a perception of ‘just getting along’ through to ‘very poor’ (classed as ‘doing poorly’). Income support was dichotomised, with respondents coded ‘yes’ (=1) if they were in receipt of any income support payment, allowance or other pension.

Three separate area characteristic measures were included: the Socio-Economic Index for Areas (SEIFA) (ABS 2001) relative advantage/disadvantage scale, housing tenure and area attachment. The SEIFA advantage/disadvantage measure considers area characteristics such as the proportion of families with a high income, and proportion of people employed in a skilled occupation. The relative advantage/disadvantage index is a continuum where lower scores indicate greater relative disadvantage and higher scores indicate greater relative advantage. Housing tenure was coded dichotomously: own/paying off (=0) or rent/board/other (=1). While housing tenure can also be considered an ‘individual’ characteristic, in the current instance it was used as an area indicator due to its potential connection with area attachment (see endnote 4). Area attachment was assessed on a five-point Likert scale with 1 representing a ‘strong preference to stay’ and 5 indicating a ‘strong preference to leave’ the area of current residence. Lower scores indicate higher area attachment.

Social connectedness and life satisfaction variables

Wave 6 of the HILDA survey contains 12 questions about the frequency of community involvement, assessed on a six-point Likert scale ranging from ‘never’ through to ‘very often’. The community involvement scale incorporates contact with friends/family and neighbours, engagement with civic activities such as volunteering and giving to charity, and involvement with political activism (for example, contacting local politicians, union activity). Each question was analysed separately, to establish whether disability status was associated with differences in specific types of community involvement.

Respondents also provided information about whether they were an active member of a sporting, hobby or community-based club or association (a dichotomous variable, no/yes) and, if so, the number of groups of which they were currently a member.


117

Perceived social support was assessed through a series of 10 statements such as ‘I often feel very lonely’ (reverse coded item) and ‘I seem to have a lot of friends.’ Respondents indicated the extent to which they agreed with each of the statements on a seven-point Likert scale ranging from ‘strongly disagree’ to ‘strongly agree’. Again, each question was assessed individually. Respondents were also asked how frequently they had social get-togethers with friends or family not living with them.

Neighbourhood relationships were evaluated using five items, with a seven-point scale of agreement for statements such as ‘This is a close-knit neighbourhood’ and ‘People in this neighbourhood can be trusted’.

The life satisfaction scale in HILDA includes eight items sampling self-reported satisfaction in areas such as an individual’s financial situation, health and free time. There is also one stand-alone measure of ‘global’ satisfaction, sampled by the item ‘all things considered, how satisfied are you with your life?’. Each life satisfaction item is scored on a scale of 0 (the lowest level of satisfaction) to 10 (the highest level of satisfaction).

Analyses

Bivariate comparisons of demographic and socioeconomic variables and measures of social connectedness were undertaken using independent samples t-tests or logistic regression as appropriate. Items where differences were identified between disability status groups were subject to multivariate analysis. Multivariate analysis was undertaken using a series of multiple linear regressions (missing data excluded pairwise). The full model contained six steps:

1 Base model (disability status only) (reference category=no disability)

2 Base model, demographics (marital status reference category=legally married; sex reference category=male)

3 Base model, demographics, education (reference category=certificate/diploma)

4 Base model, demographics, education, labour force status (reference category=employed)

5 Base model, demographics, education, labour force status, economic circumstances (income support reference category=not receiving support; financial perception reference category=doing well)

6 Base model, demographics, education, labour force status, economic circumstances, area characteristics (housing reference category=own/paying off home).


118

3 Results

demographics

2,031 regional respondents were aged between 18 and 45 years. Of these, 243 were excluded due to non-completion of the necessary self-report measures. This resulted in a final sample of 1,475 respondents without disability (652 men, 823 women) and 313 respondents with disability (151 men, 162 women; 18.8 per cent and 16.4 per cent of the sample by sex, respectively). The mean age of respondents with disability was 34.16 years (SD=8.22), and for respondents without disability it was 31.72 years (SD=8.08). In light of this small difference, age was entered in subsequent multivariate analyses. Additional descriptive data are shown in Table 1.

Table 1: Demographic and socioeconomic characteristics by disability status

Disability statusNo disability

(n=1,475)Disability (n=313)

% % OR (CI) p value

Marital status(a)

Married 45.4 37.4 0.71 (0.55–0.92) 0.007

Separated/divorced 8.9 14.7 1.75 (1.22–2.51) 0.002

Living in a relationship (de facto) 17.8 17.3 0.96 (0.70–1.33) 0.808

Not married, not living in a relationship

27.0 30.4 1.18 (0.90–1.54) 0.226

Housing tenure

Own/paying off mortgage 62.5 55.31.36 (1.06–1.74) 0.014

Rent/board/other tenure(b) 37.5 44.7

Employment status

Employed

Full time 56.2 34.20.26 (0.20–0.34) <0.001

Part time 27.0 22.4

Unemployed

Looking for work 3.1 10.5 3.66 (2.30–5.83) <0.001

NILF

Marginally attached 6.0 13.43.09 (2.34–4.08) <0.001

Not marginally attached 7.7 19.5


119

Table 1: Demographic and socioeconomic characteristics by disability status (continued)

Disability status

No disability (n=1,475)

Disability (n=313)

% % OR (CI) p value

Highest education level

Tertiary 16.2 12.5 0.74 (0.51–1.06) 0.098

Certificate/diploma 33.4 37.4 1.19 (0.93–1.54) 0.173

Year 12 21.4 13.4 0.57 (0.40–0.81) 0.002

Year 11 or lower 29.1 36.7 1.42 (1.10–1.83) 0.008

Income support receipt 17.4 45.2 3.93 (3.03–5.09) <0.001

Equivalised income—FY ($) 42,032 32,944 t=6.06 <0.001

(a) There were 8 widowed respondents in the no disability group, comprising 0.5 per cent of that group, and 1 widowed respondent in the disability group.

(b) Of this category, the majority were rentals.Note: Percentages may not sum to 100 due to rounding error. OR=odds ratio; CI=confidence interval.

Relative to people with no disability, people with disability were less likely to have a Year 12 education or to own their own home. People with disability had a lower mean equivalised income, were more likely to be unemployed or not in the labour force, and in receipt of income support, and separated or divorced.

In keeping with the assumption of disability-related disadvantage, 41.9 per cent of people with disability lived in areas in the lowest two deciles of relative advantage/disadvantage, compared to 29.6 per cent of respondents without disability (Table 2).

Table 2: SEIFA relative advantage/disadvantage decile by disability status

Disability statusNo disability Disability

Decile Number (%) Number (%)1 (most disadvantaged) 199 (13.5) 70 (22.4)2 237 (16.1) 61 (19.5)3 224 (15.2) 53 (16.9)4 197 (13.4) 38 (12.1)5 211 (14.3) 29 (9.3)6 143 (9.7) 18 (5.8)7 109 (7.4) 17 (5.4)8 92 (6.2) 13 (4.2)9 51 (3.5) 14 (4.5)10 (most advantaged) 12 (0.8) 0 (0.0)


120

Preferences to stay in their area of current residence did not differ between groups. Both groups reported, on average, a moderate preference to continue living in that area (Disability: Mean=2.08, SD=1.17; No disability: Mean=2.06, SD=1.19; p=0.435), with responses positively skewed around a strong preference to stay.

There were small but significant differences between respondents with and without disability on social participation items related to contact with family and friends. People with disability consistently reported lower mean levels of contact (Table 3). However, involvement with the wider community, in terms of activities such as volunteering and attending community events, was comparable between groups. A slightly lower proportion of respondents with disability (29.3 per cent) reported that they were active members of a club or association, relative to people with no disability (34.0 per cent). However, this difference was not statistically significant (OR=0.80, CI=0.61–1.05, p=0.106). The number of clubs people were active members of did not differ between groups.

Table 3: Social participation by disability status

Disability statusNo disability DisabilityMean (SD) Mean (SD) p value

Have telephone, email or mail contact with friends or relatives not living with you

4.70 (1.20) 4.34 (1.31) <0.001

Chat with your neighbours 3.38 (1.36) 3.54 (1.44) 0.061Attend events that bring people together such as fetes, shows, festivals or other community events

3.26 (1.19) 3.18 (1.21) 0.268

Get involved in activities for a union, political party, or group that is for or against something

1.59 (0.94) 1.61 (0.92) 0.744

Make time to attend services at a place of worship

1.80 (1.34) 1.91 (1.44) 0.161

Encourage others to get involved with a group that’s trying to make a difference in the community

1.95 (1.15) 2.08 (1.28) 0.078

Talk about current affairs with friends, family or neighbours

3.55 (1.33) 3.46 (1.39) 0.318

Make time to keep in touch with friends 4.31 (1.11) 3.92 (1.20) <0.001Volunteer your spare time to work on boards or organising committees of clubs, community groups or other non-profit organisations

2.19 (1.44) 2.26 (1.52) 0.498

See members of your extended family (or relatives not living with you) in person

3.72 (1.30) 3.44 (1.38) 0.001

Get in touch with a local politician or councillor about issues that concern you

1.34 (0.72) 1.50 (0.86) 0.001

Give money to charity if asked 3.33 (1.27) 3.17 (1.28) 0.041

Note: SD=standard deviation.


121

In terms of social get-togethers, the most commonly reported frequency of contact within each group was ‘about once a week’. Respondents with disability were more likely to report that they had social contact with family or friends not living with them less than once every three months (OR=2.96, CI=1.87–4.70, p<0.001), but there were no differences within any other frequency categories.

With the sole exception of feeling better after talking to people about something that is on their mind, people with disability reported lower perceived social support than people with no disability. Although the differences were small, they were nonetheless statistically significant (Table 4).

Table 4: Perceived social support by disability status

Disability statusNo disability DisabilityMean (SD) Mean (SD) p value

People don’t come to visit me as often as I would like(a)

3.55 (1.67) 3.88 (1.78) 0.002

I often need help from other people but can’t get it(a)

2.34 (1.43) 3.06 (1.75) <0.001

I seem to have a lot of friends 4.59 (1.54) 4.11 (1.78) <0.001I don’t have anyone that I can confide in(a) 2.27 (1.58) 2.76 (1.88) <0.001I have no one to lean on in times of trouble(a) 2.14 (1.52) 2.58 (1.76) <0.001There is someone who can always cheer me up when I am down

5.44 (1.56) 5.12 (1.75) 0.002

I often feel very lonely(a) 2.61 (1.70) 3.49 (1.96) <0.001I enjoy the time I spend with the people who are important to me

6.34 (0.95) 6.19 (1.42) 0.022

When something is on my mind, just talking with the people I know can make me feel better

5.62 (1.36) 5.49 (1.44) 0.131

When I need someone to help me out, I can usually find someone

5.68 (1.33) 5.13 (1.61) <0.001

(a) Higher score indicates less support.Note: SD=standard deviation.

With the exceptions of satisfaction with their home and free time, life satisfaction differed by disability status for all items. Again, the differences were small but statistically significant. In most cases satisfaction—while lower for people with disability—was relatively high. However, there was considerable within-group variation in scores, indicating a wide range of responses.


122

Table 5: Life satisfaction by disability status

Satisfaction with:

Disability status

No disability Disability

Mean (SD) Mean (SD) p value

The home in which you live 7.71 (1.96) 7.77 (1.92) 0.611

Your employment opportunities 7.37 (1.97) 5.99 (2.76) <0.001

Your financial situation 6.27 (2.07) 5.00 (2.62) <0.001

How safe you feel 8.29 (1.44) 7.79 (2.00) <0.001

Feeling part of your local community 6.87 (2.05) 6.44 (2.33) 0.001

Your health 7.80 (1.54) 5.81 (2.32) <0.001

The neighbourhood in which you live 7.85 (1.81) 7.37 (2.14) <0.001

The amount of free time you have 6.08 (2.48) 6.12 (2.86) 0.397

Your life 7.93 (1.31) 7.24 (1.86) <0.001

Note: SD=standard deviation.

There was evidence for variation by disability status in specific types of perceived neighbourhood relationships. Respondents with disability were slightly more likely to believe that people in their neighbourhood generally did not get along, or share the same values (Table 6).

Table 6: Neighbourhood relationships by disability status

Disability status

No disability Disability

Mean (SD) Mean (SD) p value

This is a close-knit neighbourhood 3.93 (1.46) 3.96 (1.58) 0.737

People around here are willing to help their neighbours

4.39 (1.48) 4.42 (1.57) 0.703

People in this neighbourhood can be trusted

4.56 (1.43) 4.50 (1.45) 0.484

People in this neighbourhood generally do not get along with each other(a)

2.66 (1.37) 2.89 (1.52) 0.010

People in this neighbourhood generally do not share the same values(a)

3.13 (1.39) 3.43 (1.48) 0.001

(a) Higher score indicates less cohesion.Note: SD=standard deviation.


123

Multivariate analysis

When demographic, social and economic factors were controlled for, disability continued to be associated—albeit less strongly—with 14 out of the 23 items that bivariate analysis identified as significantly different between regional people with and with no disability. These outcomes are summarised in Table 7.4

Table 7: Multivariate analysis

Base model

Demographic Education Labour force status

Economic Area

B B B B B B R2

Social participation

X1 –0.36** –0.29** –0.27** –0.22** –0.18* –0.17* 0.11

X2 –0.39** –0.33** –0.33** –0.29** –0.24** –0.24** 0.10

X3 –0.28** –0.20* –0.20* –0.15 –0.12 –0.12 0.09

X4 0.15** 0.12* 0.12* 0.10 0.09 0.08 0.04

X5 –0.16* –0.19* –0.17* –0.13 –0.09 –0.09 0.09

Social support

X1 –0.56** –0.50** –0.49** –0.41** –0.32** –0.33** 0.11

X2(a) 0.33** 0.28* 0.28* 0.27* 0.18 0.19 0.06

X3(a) 0.73** 0.67** 0.65** 0.52** 0.40** 0.40** 0.13

X4 –0.48** –0.39** –0.39** –0.30** –0.22* –0.22* 0.07

X5(a) 0.49** 0.41** 0.39** 0.29** 0.23* 0.23* 0.11

X6(a) 0.45** 0.34** 0.32** 0.25* 0.18 0.18 0.11

X7 –0.31** –0.19 –0.18 –0.13 –0.08 –0.08 0.07

X8(a) 0.88** 0.76** 0.74** 0.67** 0.61** 0.62** 0.15

X9 –0.14* –0.09 –0.07 –0.05 –0.03 –0.04 0.05

Life satisfaction

X1 –1.36** –1.31** –1.25** –0.79** –0.59** –0.58** 0.24

X2 –1.27** –1.18** –1.14** –0.90** –0.55** –0.55** 0.31

X3 –0.50** –0.46** –0.46** –0.43** –0.31** –0.32** 0.10

X4 –0.42** –0.45** –0.43** –0.38** –0.26 –0.27* 0.17

X5 –1.99** –1.93** –1.93** –1.83** –1.73** –1.73** 0.22

X6 –0.48** –0.48** –0.47** –0.42** –0.32* –0.32** 0.26

X7 –0.70** –0.61** –0.61** –0.60** –0.50** –0.52** 0.14

Neighbourhood relationships

X1(a) 0.23* 0.23* 0.19* 0.16 0.10 0.09 0.09

X2(a) 0.30** 0.31** 0.29** 0.21* 0.16 0.15 0.08

(a) Denotes items where positive values indicate lower outcomes for people with disability.Notes: * Significant at p≤0.05; ** Significant at p≤0.01. (Notes continue on p. 124.)


124

Notes: (continued) Social participation—X1: Have telephone, email or mail contact with friends or relatives not living

with you; X2: Make time to keep in touch with friends; X3: See members of your extended family (or relatives not living with you) in person; X4: Get in touch with a local politician or councillor about issues that concern you; X5: Give money to charity if asked.

Social support—X1: When I need someone to help me out, I can usually find someone; X2: People don’t come to visit me as often as I would like(a); X3: I often need help from other people but can’t get it(a); X4: I seem to have a lot of friends; X5: I don’t have anyone that I can confide in(a); X6: I have no one to lean on in times of trouble(a); X7: There is someone who can always cheer me up when I am down; X8: I often feel very lonely(a); X9: I enjoy the time I spend with the people who are important to me.

Life satisfaction—X1: Your employment opportunities; X2: Your financial situation; X3: How safe you feel; X4: Feeling part of your local community; X5: Your health; X6: The neighbourhood in which you live; X7: Your life.

Neighbourhood character—X1: People in this neighbourhood generally do not get along with each other(a); X2: People in this neighbourhood generally do not share the same values(a).

Social participation

Although controlling for demographic and socioeconomic variables reduced the strength of association between disability status and various indices of social participation, disability status continued to be associated with less frequent participation for two items: respondents’ having telephone, mail or email contact with friends or family not living with them, and making time to keep in touch with friends. In contrast, the inclusion of labour force status reduced to statistical non-significance the association between disability status and the frequency with which respondents saw members of their extended family, got in touch with a local politician or councillor, and gave money to charity. In the full model (not shown in Table 7), sex was independently associated with the frequency of non–face to face contact and making time to keep in touch with friends, with women reporting more frequent contact (B=0.570) and more often making time to keep in touch (B=0.463).

Social support

Disability status continued to be associated with lower perceived social support on a number of indices, after socioeconomic factors were taken into account. People with disability were less likely to agree that when they need someone to help them out they could usually find someone and less likely to perceive that they had a lot of friends. In addition, people with disability were more likely to feel that they often needed help from others but cannot get it, do not have anyone to confide in, and often feel lonely. Other predictors that were common across the items included sex (women reported higher perceived support for all items, except feeling lonely), marital status (respondents who were divorced/separated were less likely to have someone to confide in and more often felt lonely), and financial perception (respondents who believed that they were doing poorly financially reported lower perceived support on all items).


125

life satisfaction

The inclusion of socioeconomic variables reduced the relationship strength between disability and levels of satisfaction with employment opportunities, financial situation, feelings of safety, feeling part of the local community, the neighbourhood of residence and life overall. However, despite the mitigating influence of socioeconomic factors, disability continued to be significantly associated with lower satisfaction on each of these items. The perception of doing poorly financially was associated with lower satisfaction on each of the items. Marital status was an independent predictor of respondents’ satisfaction with their financial situation and life overall (individuals who were divorced/separated or not married and not living with someone were less satisfied), while education level and current employment status were related to satisfaction with employment opportunities (tertiary education was associated with higher satisfaction, while unemployment or not being in the labour force was related to lower satisfaction).

Neighbourhood relationships

The inclusion of labour force status negated the relationship between disability status and the belief that people in the neighbourhood generally did not get along. Similarly, controlling for economic variables reduced to non-significance the association between disability status and the belief that people in the neighbourhood generally did not share the same values.

double disadvantage indices

Five of the items selected for multivariate analysis had been pre-identified as possible indices of double disadvantage. These were: respondents’ frequency of non–face to face contact with family/friends, making time to keep in touch with friends, feelings of needing help but not being able to get it, satisfaction with employment opportunities, and satisfaction with financial situation. In the current study, disability-related differences persisted for each of the five indicators once demographic and socioeconomic variables were controlled for.

The extent to which variance was accounted for by the full model differed between items, ranging from 4 per cent (getting in touch with a politician) to 31 per cent (satisfaction with financial situation).

4 DiscussionThe current study explored the double disadvantage theory of regional living and disability, with specific reference to social connectedness. It assessed levels of self-reported life satisfaction, community participation, and perceived social support among prime working age regional people, as a function of disability status. Next, it examined the extent to which differences on measures of social connectedness and life satisfaction could be accounted for by the inclusion of demographic and socioeconomic variables. In conjunction with earlier data, this enabled


126

identification of indices on which regional people with disability fared more poorly than both people with disability in major cities and regional people with no disability. Also, the study looked at the social cohesion theory of regional living, by examining whether the previously observed higher levels of community engagement among regional people with disability relative to people with disability in major cities were consistent with the levels of community engagement reported by regional people with no disability.

demographic and socioeconomic indicators

In terms of demographics, regional people with disability were more likely to be divorced or separated than their counterparts with no disability. Marital status offered predictive utility in relation to a number of indices of satisfaction and social connectedness (particularly financial satisfaction and perceived social support), with divorce/separation or having no partner generally associated with lower satisfaction and social connectedness. Therefore, the greater likelihood of people with disability to have experienced divorce or separation relative to people with no disability may increase the likelihood that those individuals will also experience social exclusion.

Furthermore, people with disability were less likely to own their own home than their counterparts without disability.5 While home ownership proportions did not differ for people with disability across locations (McPhedran in press), the current study indicates proportionally higher levels of renting among people with disability relative to those with no disability. Existing work demonstrates the vulnerability of renters (particularly renters with limited resources) to experience financial stress in response to shifts in housing affordability (Yates 2007). Findings about the greater likelihood of people with disability both renting and being unemployed or not in the labour force (see below), as well as the documented relationships between housing stability and overall wellbeing (AHURI 2002; Burke & Hulse 2002; Robinson & Adams 2008), raise issues for further consideration about access to appropriate housing for people with disability in regional areas (see AHURI 2003).

The findings confirm the expectation that disability is associated with greater relative socioeconomic disadvantage in regional Australia. People with disability in regional Australia were less likely to be employed, less likely to have a Year 12 education, had a lower income, and were more likely to receive income support than people in regional Australia who did not report disability. These socioeconomic characteristics were associated with lower levels of satisfaction and social connectedness on a range of measures. Given that that a greater proportion of people with disability report these socioeconomic characteristics, this places those individuals at an increased likelihood of social exclusion. In conjunction with earlier observations of locational differences in socioeconomic characteristics for people with disability, the current findings for education, employment, income and income support receipt are consistent with the theory of double disadvantage and indicate a cumulative negative influence of location plus disability.

The current study does not demonstrate specific pathways from demographic characteristics or socioeconomic factors through to social isolation, although this is an issue that merits consideration using longitudinal data. However, the results highlight a range of potential


127

demographic and socioeconomic risk factors for lower social connectedness that are more commonly found among people with disability, many of which have been studied in other settings with different samples. For example, there are known relationships between marital breakdown and subsequent financial changes and housing transitions for both men and women (for example, AHURI 2008). The observed relationships between marital status, socioeconomic indicators and various measures of satisfaction and social connectedness used in this study suggest a need to consider access to social and other supports (for example, affordable accommodation near family or friends who provide informal support, assistance to establish and maintain social connections) for regional people with disability who live alone or are experiencing relationship dissolution.

An additional socioeconomic indicator of note was the SEIFA measure of relative advantage/disadvantage, which showed that a greater proportion of people with disability lived in the lowest two SEIFA deciles (indicating the greatest levels of relative disadvantage), compared with people who did not report disability. This confirms the premise that disability and disadvantage are related, an observation predicted to relate to social connectedness. Indeed, it has been documented elsewhere that areas with fewer facilities and resources report lower levels of social cohesion (Cattell 2001). However, this was not the case in the current study, where multivariate analyses (discussed below) demonstrated that SEIFA was not consistently associated with social connectedness. The reasons for this result are unclear, although it is possible that the ‘area attachment’ measure—which was associated with most social connectedness items used in this study—captures the variance that would otherwise be associated with a range of more specific indices such as community facilities and resources, housing tenure, and so forth. The usefulness of the ‘area attachment’ measure in social connectedness research, and the variety of other attributes and constructs it may capture, merits further study.

life satisfaction

Previous examination of locational differences in life satisfaction revealed few differences in life satisfaction between people with disability in major cities and regional Australia. The differences that did emerge were either positive (for example, regional people being more satisfied with feeling part of the community) or in cases where regional people were less satisfied (financial situation and employment opportunities) negated by the inclusion of demographic and socioeconomic variables (such as education level and marital status). In contrast, there was a persistent relationship between disability and lower satisfaction in many different life areas, even when socioeconomic and demographic indicators that may be related to (or influenced by) disability status were controlled for. While this does not provide broad evidence for double disadvantage in relation to life satisfaction, it does suggest an independent influence of disability on life satisfaction, over and above the influence of socioeconomic factors.

life satisfaction and double disadvantage

There were two items that provided evidence for double disadvantage: respondents’ satisfaction with their employment opportunities and financial situation. These two indices,


128

where disability-related differences persisted in the current study, had previously been found to be negatively influenced by regional location (through an influence of socioeconomics). This indicates that the occurrence of double disadvantage on the objective socioeconomic indicators of education, employment and income support extends to subjective evaluations on related satisfaction measures. These evaluations are likely to affect financial perception, which was in turn associated with a number of indices of social connectedness. It would be beneficial to further investigate potential relationships between income and employment, subjective evaluation of financial wellbeing and employment prospects, and social connectedness, to determine whether objective measures of labour force status and financial circumstances, or the subjective evaluation of those circumstances, are most strongly associated with social connectedness.

Social connectedness

Relative to previous comparisons between people with disability who lived in major cities and regional Australia, where few differences emerged on social connectedness indicators, the current study found that regional people with disability fared less well than their regional counterparts with no disability on a number of indicators of social connectedness. This applied particularly to perceived social support, where disability-related differences persisted for items relating to friendships, confidants and loneliness. These findings suggest that disability status, rather than location, negatively affects social connectedness (although socioeconomic factors may reduce the extent to which disability is associated with lessened social support).

A finding that merits further consideration was the association between sex and social support, with men (irrespective of disability status) reporting lower support. This is consistent with previous research that has documented the negative impacts associated with male social isolation (Cloutier-Fisher & Kobayashi 2009; Eng et al. 2002; Grant, Hamer & Steptoe 2009; Meisinger, Kandler & Ladwig 2009), and conversely, has shown benefits in encouraging men who are not in paid work to participate in activities outside their home and family (for example, ‘men’s shed’ programs, social and community activities with their peers) in order to increase their social networks, as well as encourage purposeful activity and a sense of self-worth (Golding 2009). It is reasonable to propose that disability would exacerbate the risk of low social connectedness among men, leading to a heightened risk of social exclusion. While considerable effort has been devoted to improving women’s access to services and support in rural and regional Australia, these results point to the importance of taking into account the different needs and preferences of men and women with disability.

Social connectedness and double disadvantage

The current study provided evidence of double disadvantage on three specific social connectedness indicators: non–face to face contact with friends or relatives, making time to keep in touch with friends, and often needing help from other people but not being able to get it. Previous work showed that socioeconomic variation accounted for differences between city and regional people with disability on these measures (McPhedran in press). It was thus speculated


129

that the greater relative disadvantage in regional areas explained those differences in contact and support, for example, by reducing the ability to afford private transport or access to formal support.

If those locational differences were simply related to socioeconomics, however, it would also be expected that any differences between people with and with no disability in regional locations should be mitigated by controlling for socioeconomics. This did not occur. Therefore, it appears that disability status and location both affect specific types of social contact and support, through a direct influence of disability, and an indirect influence of location (via the greater relative socioeconomic disadvantage in regional locations). This suggests that the theory of double disadvantage can be applied in the context of social connectedness, but only to specific indicators rather than across the board.

Interestingly, however, the actual frequency of reported social get-togethers with friends and family was broadly similar between regional people with and with no disability, regardless of the differences in self-reported frequency of making time to keep in touch with friends. One explanation for this apparent discrepancy is that disability status is associated with variation in perceptions of contact frequency. For example, people with disability may place greater importance on contact with friends and family, or desire more frequent contact with friends and family than people with no disability. A person with no disability who sees family and friends two or three times a month may be more inclined to consider that ‘often’ than a person who places greater subjective importance on the frequency of contact and may therefore view contact of that frequency as ‘sometimes’ or ‘occasionally’.

Also, the assessment of how frequently respondents had social get-togethers included family and friends, a more general question than the assessment of keeping in touch with friends (rather than family and friends). Therefore, an additional explanation for the difference in objective and subjective assessment of contact frequency is that disability status affects the formation and maintenance of friendships to a greater degree than it affects contact with family. This accords with the finding that people with disability were less likely to see themselves as having a lot of friends, and suggests that while family can represent an important source of social contact, people with disability may nonetheless experience lower social connectedness in regard to having friends outside their family.

Potentially, differences in labour force status could also influence the frequency of contact with friends. The workplace in itself is a source of social contact and friendships (Morris & Abello 2005; Morrison 2009), hence not being in paid work can be associated not just with socioeconomic circumstances but also with lower social contact. Consequently, people who are not in paid work may lack the social networks available to people who are in paid work. This may lead to a desire for more frequent contact with others, leading to dissonance between preferred or perceived levels of contact and actual levels of contact. This proposal is partially supported by the finding that controlling for labour force status reduced to statistical non-significance the differences between disability status groups on the item ‘people don’t come to see me as often as I like’; contact with people at work may reduce the extent to which other social contact is desired.


130

The above three possible explanations for the difference in objective and subjective assessment of contact frequency suggest the value of further considering both opportunities for regional people with disability to have regular social contact outside the family, and the ability of people who are not able to undertake paid work to engage in ‘substitute’ activities. A useful direction for future investigation will be to determine whether perceived social connectedness is associated with the number of hours of social contact within a week (for example, the hours associated with full-time paid work), specific types of social contact (for example, contact with family, compared with contact with friends), or both. If relationships between the amount and type of social contact and perceived social connectedness can be better understood, then effective means to facilitate social contact may be identified for people who are not able to undertake paid work. It is important to gather this information, given that almost 45 per cent of regional respondents with disability were either unemployed or not in the labour force.

Social cohesion

In previous work, the only locational differences that persisted when demographic and socioeconomic factors were controlled for were areas where regional people with disability fared better than their counterparts in major cities—chatting with neighbours, attending community events, volunteering, satisfaction with their home, feeling part of the local community and the perceived strength of neighbourhood relationships.

Disability status was associated with differences on only one of these measures in the current study—satisfaction with feeling part of the local community—and that difference was accounted for by the inclusion of economic variables. Financial perception and housing independently associated with this item, suggesting that the perception of not being well off affects people’s judgements of being part of their community, and confirming earlier literature that shows an association between home ownership and community cohesion (for example, Stone & Hulse 2007). However, it must be recalled that regional people with disability nonetheless felt more a part of their local community than their counterparts in major cities (McPhedran in press). This indicates a positive association between regional location and feelings of inclusion, potentially offsetting the negative influence of disability.

The frequency of attending community events and volunteering was consistent between regional people with and with no disability. The frequency of participation was low overall, however, suggesting that a number of respondents either do not engage in community activities or that they take part only on occasion. It had previously been speculated that relatively low levels of participation, while higher in regional areas relative to cities, may be related to disability status as well as barriers to participation (McPhedran in press). However, the absence of variation by disability status in the current study suggests that if barriers to participation exist, they also apply to people with no disability. Barriers to participation may include limited opportunities or minimal diversity of options, lack of awareness about available ways in which to engage with the community, or, alternatively, other uses of time may take precedence (for example, spending time with family or engaging in work). However, people with disability may nonetheless experience different barriers to participation than people without disability.


131

It has been argued that social connectedness, indexed by measures such as levels of social support and community participation, relates positively to an individual’s mental health and overall wellbeing (for example, Eime, Payne & Harvey 2008; Moore, Townsend & Oldroyd 2007). The small amount of Australian study into relationships between social connectedness and wellbeing is consistent with this proposal. For example, Ziersch and others (2009) found that trust of others and social involvement were positively related to mental health, while earlier work showed a positive association between perceptions of neighbourhood relationships and mental health (Ziersch et al. 2005). Similarly, Moore, Townsend and Oldroyd (2007) found higher life satisfaction among members of community conservation groups, relative to matched controls.

In light of the potential relationships between community engagement and general wellbeing, it would be beneficial to further examine whether community involvement can offer protection against social exclusion for people with otherwise low levels of social connectedness. For example, community participation through activities such as volunteering may benefit individuals who have limited contact with family or friends. Furthermore, this may highlight the need to support and promote participation opportunities in regional Australia (while acknowledging that people with disability may have different preferences and capabilities to people with no disability).

Although there were differences in the perceived extent to which neighbours got along with one another and the homogeneity of neighbourhood values (both of which were accounted for by socioeconomic differences), the strength of neighbourhood relationships—in terms of neighbours being close-knit, trusting and helping one another—was consistent between people with and with no disability. This suggests that the perceived homogeneity of a neighbourhood does not necessarily relate to the perceived quality of neighbourhood relationships, and in conjunction with the findings for community engagement, supports the social cohesion theory of regional living.

limitations and future directions

The current study did not address whether disability leads to disadvantage, or whether disadvantage precedes disability. Disentangling the direction of the relationship between disability and disadvantage through the use of longitudinal data will provide useful insights into possible interventions and forms of support that can assist regional people who are disadvantaged. Similarly, this work did not examine pathways between disadvantage and social connectedness, or the role of social connectedness as a protective factor against negative outcomes on measures such as mental health. Work currently underway will address these questions.

An important area for future study is how, for people with disability, the extent of limitation affects participation in society. For example, for people with disability in regional Australia, the influence that the extent of limitation has on the types or amount of paid work that can be undertaken may be compounded by limitations in the types of paid work that are available (such as a greater availability of work that requires strenuous physical activity). Investigation into


132

whether people in regional Australia with similar levels of limitation to their city counterparts nonetheless experience greater disadvantage is necessary to better develop ways to support regional people with disability. The extent of limitation may also account for the higher amount of variation in social connectedness among people with disability, relative to people with no disability.

5 ConclusionsThe current study demonstrates that regional people with disability experience disadvantage on many socioeconomic and social connectedness indicators, and, in some instances, that disability and location have cumulative negative influences. Irrespective of these results, however, the findings highlight that regional living can be positively associated with community engagement and community-oriented activities, irrespective of disability status. These results also indicate that the characteristics of regional Australia may play a distinct, yet multifaceted, role in the experiences of people with disability. This in turn suggests potential avenues for consideration in terms of strengthening social connectedness among regional people with disability, and addressing the risk of social exclusion among regional people with disability.

Endnotes1 For an overview, the reader is referred to the Australian Institute of Health and Welfare (2008).

2 The term most common in existing literature is ‘rural’. However, this term is often used in a broad conceptual sense, rather than representing any formal or agreed method of geographic classification. There is also debate over how to define ‘rural’ (Fuguitt, Brown & Beale 1989). For this reason, when the phrase ‘rural and regional’ is used herein it is intended to capture the informal use of ‘rural’ and the more formal classification designation of ‘regional’ location, which overlaps with the concept of ‘rural’, and encompasses the spatial and population elements of ‘rural’ location (Wilkinson 1991).

3 This category excluded certificates or diplomas obtained through graduate study, which fell into the ‘tertiary’ category.

4 Full results are available from the author.

5 While housing tenure in itself was not generally associated with social connectedness, area attachment was associated with almost every measure of satisfaction and connectedness. It is likely that the ‘area attachment’ measure used in this study is a global construct that subsumes the influence of more specific measures such as housing, and captures a broader attitudinal orientation to which variables like social contact, employment, housing and economic perceptions all contribute.


133

ReferencesAlston, M & Kent, J 2004, Social impacts of drought: report to NSW Agriculture, Centre for Rural Social Research, Wagga Wagga, NSW.

Archer, J 2000, The politics of metrocentrism, paper presented at the Australasian Political Studies Association, Canberra, 4–6 October.

Australian Bureau of Statistics (ABS) 2001, ABS view on remoteness, cat. no. 1244.0, ABS, Canberra.

——2003, Household income and income distribution, cat. no. 6523.0, ABS, Canberra.

Australian Housing and Urban Research Institute (AHURI) 2002, Housing, housing assistance and wellbeing, AHURI Research and Policy Bulletin no. 10, AHURI, Melbourne.

——2003, The key role of housing in regional development, AHURI Research and Policy Bulletin no. 20, AHURI, Melbourne.

——2008, How does the loss of a partner affect housing outcomes?, AHURI Research and Policy Bulletin no. 100, AHURI, Melbourne.

Australian Institute of Health and Welfare (AIHW) 2008, Disability in Australia: trends in prevalence, education, employment and community living (Bulletin 61), AIHW, Canberra.

Boyd, CP, Hayes, L, Wilson, RL & Bearsley-Smith, C 2008, ‘Harnessing the social capital of rural communities for youth mental health: an asset-based community development framework’, Australian Journal of Rural Health, vol. 16, pp. 189–93.

Burke, T & Hulse, K 2002, Sole parents, social wellbeing and housing assistance, AHURI Final Report no. 15, Melbourne.

Cattell, V 2001, ‘Poor people, poor places, and poor health: the mediating role of social networks and social capital’, Social Science and Medicine, vol. 52, pp. 1501–16.

Cloutier-Fisher, D & Kobayashi, KM 2009, ‘Examining social isolation by gender and geography: conceptual and operational challenges using population health data in Canada’, Gender, Place and Culture, vol. 16, no. 2, pp. 181–99.

Eime, RM, Payne, WR & Harvey, JT 2008, ‘Making sporting clubs health and welcoming environments: a strategy to increase participation’, Journal of Science and Medicine in Sport, vol. 11, pp. 146–54.

Eng, PM, Rimm, EB, Fitzmaurice, G & Kawachi, I 2002, ‘Social ties and change in social ties in relation to subsequent total and cause-specific mortality and coronary heart disease incidence in men’, American Journal of Epidemiology, vol. 155, no. 8, pp. 700–709.

Fuguitt, G, Brown, D & Beale, C 1989, Rural and small town America, Russell Sage, New York.

Gething, L 1997, ‘Sources of double disadvantage for people with disabilities living in remote and rural areas of New South Wales, Australia’, Disability and Society, vol. 12, pp. 513–31.


134

Golding, B 2009, Older men’s learning through social inclusion, seminar, Brotherhood of St Laurence, Fitzroy, Australia, 20 August.

Grant, N, Hamer, M & Steptoe, A 2009, ‘Social isolation and stress-related cardiovascular, lipid, and cortisol responses’, Annals of Behavioural Medicine, vol. 37, pp. 29–37.

Hall, G & Scheltens, M 2005, ‘Beyond the drought: towards a broader understanding of rural disadvantage’, Rural Society Journal, vol. 15, no. 3, pp. 347–58.

Jenkins, SP & Rigg, JA 2004, ‘Disability and disadvantage: selection, onset, and duration effects’, Journal of Social Policy, vol. 33, pp. 479–501.

Mavromaras, K, Oguzoglu, U, Black, D & Wilkins, R 2007, Disability and employment in the Australian labour market, Melbourne Institute of Applied Economic and Social Research, Melbourne.

McPhedran, S in press, ‘Disability and community life: does regional living enhance social participation?’, Journal of Disability Policy Studies.

Meisinger, C, Kandler, U & Ladwig, K-H 2009, ‘Living alone is associated with an increased risk of Type 2 diabetes mellitus in men but not women from the general population: the MONICA/KORA Augsburg Cohort Study’, Psychosomatic Medicine, vol. 71, pp. 784–88.

Moore, M, Townsend, M & Oldroyd, J 2007, ‘Linking human and ecosystem health: the benefits of community involvement in conservation groups’, EcoHealth, vol. 3, pp. 255–61.

Morris, A & Abello, D 2005, Disability support pension new customer focus groups, Social Policy Research Centre, Sydney.

Morrison, RL 2009, ‘Are women tending and befriending in the workplace? Gender differences in the relationship between workplace friendships and organizational outcomes’, Sex Roles, vol. 60, pp. 1–13.

Obst, P, Smith, SG & Zinkiewicz, L 2002, ‘An exploration of sense of community, part 3: dimensions and predictors of psychological sense of community in geographical communities’, Journal of Community Psychology, vol. 30, pp. 119–33.

Robinson, E & Adams, R 2008, Housing stress and the mental health and wellbeing of families, Australian Family Relationships Clearinghouse Briefing no. 12, Australian Institute of Family Studies, Melbourne.

Stone, W & Hulse, K 2007, Housing and social cohesion: an empirical exploration, AHURI Final Report no. 100, Melbourne.

Talbot, L & Walker, R 2007, ‘Community perspectives on the impact of policy change on linking social capital in a rural community’, Health and Place, vol. 13, pp. 482–92.

Wilkinson, K 1991, The community in rural America, Greenwood Press, Westport, CT.

Winn, S & Hay, I 2009, ‘Transition from school for youths with a disability: issues and challenges’, Disability and Society, vol. 24, pp. 103–15.


135

Yates, J 2007, Housing affordability and financial stress, Australian Housing and Urban Research Institute Sydney Research Centre Research Paper no. 6, Sydney.

Ziersch, AM, Baum, FE, Darmawan, ING, Kavanagh, A & Bentley, R 2009, ‘Social capital and health in rural and urban communities in South Australia’, Australian and New Zealand Journal of Public Health, vol. 3, no. 1, pp. 7–16.

Ziersch, AM, Baum, FE, MacDougall, C & Putland, C 2005, ‘Neighbourhood life and social capital: the implication for health’, Social Science and Medicine, vol. 60, pp. 71–86.


136

137

Social policy noteAustralian Social Policy Journal No. 9


138

139

Work–life tension and its impact on the workforce participation of Australian mothers1

Ibolya Losoncz and Benjamin Graham


Acknowledgements

The authors are grateful for the helpful comments and advice from colleagues and the valuable programming assistance provided by Jason Brandrup at FaHCSIA.

AbstractThis paper expands on earlier work by Losoncz and Bortolotto (2009), which identified six distinctive groups of working mothers using six waves of the HILDA survey. The focus of this paper is on the labour market behaviour of working mothers in each cluster, and whether reducing working hours or leaving the workforce has benefits for the health and wellbeing of mothers in each cluster, particularly those experiencing high conflict between work and life. Mothers from clusters with high work–life conflict did not show a higher tendency to exit from paid work than mothers from other clusters. The most likely to exit paid work were mothers who had lower regard for the working mother role. As such, role preference seems to have a greater influence on work decisions. Leaving work or reducing hours did not lead to improved satisfaction with family life or parenthood in any of the clusters, while in some clusters, leaving work or reducing hours improved mental and physical health. Policies that promote greater work–life balance may have a different level of influence on mothers’ workforce participation based on their level of work role identification and particular events in their life course.

Keywords: mothers; working mothers; work–life balance; spillover; cluster analysis

1 IntroductionAustralian working mothers balance the important aspirations of participating in paid work and caring for their families and children. Mothers who can successfully combine these two goals benefit from enhanced personal satisfaction and a stronger financial position, which also benefits their families. It also promotes other societal goals, such as social inclusion, economic


140

participation and gender equity. However, for some mothers it can mean continual management of a variety of intersecting tensions and demands.

To simultaneously consider the different aspects of working mothers’ lives Losoncz and Bortolotto (2009) explored the use of cluster analysis. The research identified six major homogenous groups, each with a distinctive profile in terms of work–life experience as well as main sociodemographic, work and family characteristics. Two of the six clusters were characterised by strong work–family strain. Mothers in these two clusters tended to have long working hours, high work overload and a lack of support from others, as well as lower outcomes on health measures and low satisfaction with family life and parenthood.

This social policy note explores the labour market behaviour of working mothers across different clusters. The results of longitudinal analysis are presented to answer two questions of interest, whether:

�� mothers with strong work–family strain were more likely to reduce their working hours or exit paid work than mothers in other clusters

�� reduction in working hours or exit from paid work led to improved physical and mental health and satisfaction with family life and parenthood.

2 Related researchThe participation of Australian mothers in the labour market has increased steadily over the last quarter of a century (ABS 2006). At the same time, most mothers continue to have primary responsibility for family care and domestic matters. Not surprisingly, the work–life experience of Australian women—particularly those with dependent children—has gained increased attention in both public and policy discourse. Australian research on work–life balance has also increased considerably in recent years (for example, Alexander & Baxter 2005; Baxter et al. 2007; Baxter 2009; Bowman 2009; Pocock, Skinner & Williams 2007; Reynolds & Aletraris 2007; Strazdins et al. 2008). What is clear from these studies is that the work–life experiences of Australian mothers are far from uniform and are shaped by a wide range of factors. To explore the influence of these factors, Losoncz and Bortolotto (2009) used cluster analysis to identify major groups of working mothers, each with distinctive profiles in terms of their work–life experiences.

the six cluster description of Australian working mothers

Losoncz and Bortolotto (2009) identified six major clusters. Figure 1 is a conceptual representation of the six clusters on the two major continuums of role preference2 (y axis) and work–family strain (x axis). The two dimensions are based on natural groupings of the answers given by respondents on work–life balance questions. Role preference reflects the degree to which mothers aspire to and value the ‘working mother’ role, with high scores reflecting a belief that being a working parent is good. Work–family strain reflects mothers being forced to compromise on family or work activities—for instance, missing a family event due to

Work–lIfe teNSIoN ANd ItS IMPAct oN the Workforce PArtIcIPAtIoN of AuStrAlIAN MotherS

141

work. Descriptive analysis found systematic differences between the six clusters in terms of sociodemographic background, individual and family characteristics, and factors related to work. Below is a brief description of each cluster.

Figure 1: Dimensions and distributions of the six work–life balance clusters of Australian mothers in paid work

Aspiring &struggling

(15%)

Valuing theworking mother

role

Highly functioning& fulfilled

(19%)

Guilty coper(17%)

Treading water(21%)

Indifferent yetsuccessful

(16%)

Indifferent &struggling

(13%)

High

LowLow High

Managing work–family nexus

Note: Percentages do not sum to 100 due to rounding.Source: Losoncz & Bortolotto 2009.

highly functioning and fulfilled cluster (19 per cent of mothers)

The mothers in this cluster highly value their working mother role (such as the extent it makes them a more rounded person or a better parent) and are successful at managing the practical impact of the work–life nexus (such as finding the time for and enjoying family activities, or enjoying time at work and engaging in work activities).

Descriptive analysis found that mothers in this cluster reported the lowest level of stress at work. Their hours at work were just below the average (27 hours per week) and the time they spent on domestic tasks was well below the average. Their partners tended to spend average hours at work and on domestic tasks. Women in this cluster reported the highest levels of satisfaction with family relationships, division of household tasks and support received from others. They also had the highest scores for physical and mental health.


142

Indifferent yet successful cluster (16 per cent of mothers)

Mothers in this cluster place a relatively low value on their working mother role, but they manage the day-to-day impact of the role just as well as mothers in the ‘Highly functioning and fulfilled’ cluster. So, while these mothers tend to be indifferent to the working mother ideal, they are successful at it.

In terms of their characteristics, mothers in this cluster were the most likely to be married, to be self-employed or working for a family business, and to be a casual worker. Compared to the other clusters, they worked the fewest hours (21 hours per week), with the majority happy with their hours. While they spent the highest number of hours on domestic tasks, their combined paid and unpaid working hours were still the lowest. Mothers in this cluster appear to have a gender-based domestic arrangement with their partners, who spent above average time at work but below average hours on domestic tasks. Mothers in this cluster also reported parenthood as a positive experience.

Aspiring and struggling cluster (15 per cent of mothers)

Mothers in this cluster place a high value on being a working mother. However, when it comes to the day-to-day aspects of their life, they report a very strong tension between work and family.

This cluster had the highest proportion of mothers working more than 45 hours per week. Mothers in this group spent the longest hours at work (33 hours per week), and this did not affect the time they spent on domestic tasks, leading to the highest combined (paid and unpaid) work hours of any cluster. Even though mothers in this cluster had high occupational status and a high level of job control, they also had the highest level of work overload and stress at work. Their partners had combined paid and unpaid working hours just above the average. However, because of the long hours the mothers spent in paid work, families in this cluster had the second highest average working hours per adult in the family (39 hours per week). Mothers in this cluster reported the lowest physical and mental health scores, and low satisfaction with family relationships, parenthood and support from others.

Indifferent and struggling cluster (13 per cent of mothers)

Mothers in this cluster place a relatively low value on the working mother role, and when it comes to the practical, day-to-day aspects of managing, they report the strongest tension between work and family.

Mothers in this cluster were the most likely to be single or widowed. They worked the second longest hours (31 hours per week) and over half of them wanted to work fewer hours. They reported the second highest level of overload and work stress, low levels of job control and flexibility, and the lowest level of work satisfaction. Mothers in this cluster worked the second longest hours, and if partnered, their partners also reported above average working hours. At the same time, the average equivalised household disposable income in this cluster was the lowest. Mothers in this cluster reported the lowest satisfaction with family relationships, parenthood and support from others, and low physical and mental health.


143

treading water cluster (21 per cent of mothers)

Mothers in this cluster are just below the average both in terms of the value they place on their working mother role and the extent to which they are successful at managing the practical impact of this role. So, while they experience considerable tension between work and family life, they are coping with it.

In terms of their characteristics, they reported average values on sociodemographic, work and family indicators. Mothers in this cluster spent average hours both at work (29 hours per week) and on domestic tasks. Their partners also tended to work average hours. Mothers in this cluster reported an average level of satisfaction with family relationships, division of household tasks and level of support from others, as well as average physical and mental health scores.

guilty copers cluster (17 per cent of mothers)

Mothers in this cluster place a relatively high value on the working mother role. In terms of managing the practical aspects of the work–life nexus, they reported scores well above the average. However, they often worry about their children while at work, and feel that working leaves them with little energy to be the kind of parent they want to be.

Mothers in this cluster were found to be similar in nearly all their characteristics to the overall sample. The only notable difference was their high level of conscientiousness reported on the Personality Trait scale (Losoncz 2009), which may explain their tendency to worry about their children while at work.

Work–family strain and workforce participation

It is apparent from the above cluster descriptions that there is a connection between experience of work–family strain and the number of hours spent at work. While research has established the bidirectional nature of this relationship, most of the attention has been given to the impact of long working hours on work–life balance (Keith & Schafer 1980; Gray et al. 2004; Pocock, Skinner & Williams 2007; Strazdins et al. 2008). For example, Pocock and her colleagues found a strong and consistent association between long work hours (over 45 hours per week) and work–family strain.

While long working hours may have a negative impact on family and personal wellbeing, other aspects of work can be equally influential. For example, workers who have a poor fit between their actual and preferred hours are likely to have poorer work–life outcomes (Fagan & Burchell 2002; Messenger 2004; Pocock, Skinner & Williams 2007). The quality, complexity and skill level of the job, flexibility and pace of work, and the level of job security and amount of control over work schedule have all been identified as important contributing factors (Allan, Loudoun & Peetz 2007; Galinsky 2005; Strazdins et al. 2008).

Research on the impact of work–family strain on labour supply is more limited. One potential outcome of unresolved, long-term conflict between work and family life is temporary or permanent withdrawal from work. One analysis of work–life conflict and its influence on changing


144

industry or job found a small, but statistically significant relationship between work–life tension and withdrawal from public accounting (Greenhaus et al. 1997). In terms of its influence on permanent withdrawal from paid work, Raymo and Sweeny (2006) found that work–life conflict was positively connected to retirement preferences.

Analysis specific to work–life conflict and the intention to temporarily or permanently withdraw from work found that negative spillover from work to family was strongly associated with the intention to change jobs or to withdraw from work (Forma 2008). After controlling for sociodemographic variables, family characteristics and work characteristics, the association was weaker but still present, suggesting that the work–family relationship is a relevant determinant of the supply of labour.

The present research furthers Forma’s findings by exploring the relationship between work–life conflict and actual withdrawal from the workforce, as well as lowered workforce participation. It also examines the impact of reduction in working hours, or exit from paid work, on self-reported work–life balance, physical and mental health, and satisfaction with family life and parenthood.

3 Data and methodsData for this analysis were drawn from the first six waves (2001–2006) of the Household, Income and Labour Dynamics in Australia (HILDA) survey,3 a nationally representative household panel survey focusing on employment, family and income issues. The sample for this study, around 1,300 mothers in each wave, was limited to working mothers (partnered and unpartnered) with parenting responsibilities for any children aged 17 years or less who undertook the 13-item questionnaire on work–life balance in the self-completion part of the survey. The items were included in the survey to measure three themes relating to the impact of combining work and family responsibilities on self, work and family.

The focal method in this research is cluster analysis. Cluster analysis groups individuals according to their similarity on selected features—in this instance their responses to the 13 work–life balance questions. (For more information on cluster analysis refer to Box 1.) A two-stage analysis was applied. In the first—partitioning—stage, a hierarchical procedure was used to investigate if the six clusters identified by Losoncz and Bortolotto (2009) using Wave 5 data would fit the data in each of the six waves. The original six cluster solution presented a good differentiation between groups and provided a consistent solution with only minimal variation in cluster scores across the six waves.

In the second—fine tuning—stage, respondents were reassigned around the common seed points (across the six waves) of each cluster. The purpose of this second step was to create more homogenous, or alike, groups and to increase comparability between waves. Other statistical methods employed in this paper included descriptive statistics and analysis of variance (ANOVA).


145

Box 1: Cluster analysis method

Cluster analysis is an exploratory statistical technique that groups respondents based on their characteristics, rather than variables, to detect natural groupings in the data. Clustering simplifies complex data and can reveal patterns in multifaceted phenomena by identifying how types of people are similar and dissimilar (Adlaf & Zdanowich 1999).

One of the benefits of this approach is that it enriches understanding of how a set of concepts finds expression in different kinds of people. In doing so, it suggest possible causal relationships, which may assist in developing hypotheses for future research (Adlaf & Zdanowich 1999; Berry 2008). To enhance the research utility of clusters, initial cluster analysis is usually followed by a number of other techniques, such as descriptive analysis, multivariate analysis of variance or regression analysis.

Classifying working mothers into largely homogenous groups based on their work–life experience can advance understanding of the connection between work–life balance, economic productivity and the wellbeing of families and individuals, to enable policy development.

limitations of the study

A main limitation of the study is that it did not include an unquantified proportion of mothers who did not join the workforce because of the anticipated or actual tension between work and family responsibilities. As such it may underreport the proportion of mothers who see their engagement in the workforce as an important aspect of their life, but consider work and family responsibilities too difficult to manage together. Furthermore, some mothers may already work reduced hours, and presumably have lowered aspirations in order to manage their work–life balance.

4 Results

exiting paid work by cluster types

Two different types of exit from paid work were considered: exit to care for a newborn and exit for other reasons. Figure 2 shows the proportion of all mothers in each cluster compared to the proportion of mothers leaving work in each cluster. Mothers from clusters with a low value on their working mother role (that is, ‘Indifferent yet successful’ and ‘Indifferent and struggling’) were the most likely to exit from paid work to care for a newborn, while mothers from the ‘Aspiring and struggling’ cluster were the least likely (Figure 2).

Exiting from paid work for other reasons showed a somewhat different pattern. Mothers from the ‘Indifferent yet successful’ cluster were still the most likely to exit paid work, while mothers from the ‘Highly functioning and fulfilled’ cluster were the least likely to exit from paid work (Figure 2).


146

Figure 2: Proportion of mothers exiting paid work from each cluster compared to average size of that cluster

0

5

10

15

20

25

Average cluster distribution in the six wavesDistribution of mothers who left paid work to care for a newbornDistribution of mothers who left paid work for other reason

Highlyfunctioningand fulfilled

Per c

ent

Indifferentyet

successful

Aspiringand

struggling

Indifferentand

struggling

Treadingwater

Guiltycopers

Source: HILDA, Waves 1–6.

The main reason mothers gave for exiting paid work differed considerably4 between clusters. Mothers in the ‘Highly functioning and fulfilled’ cluster most commonly left work because they were laid off, retrenched or made redundant. Mothers in the ‘Indifferent yet successful’ cluster were more likely to leave paid work due to the temporary or seasonal nature of their job than mothers in other clusters. This result is consistent with the finding that this cluster had the highest proportion of self-employed or casual mothers. Mothers from the ‘Aspiring and struggling’ and ‘Treading water’ clusters were more likely to leave paid work due to own sickness, disability or injury. The most common reason for leaving paid work among mothers in the ‘Indifferent and struggling’ cluster was to look after children, the house or someone else. Mothers in the ‘Guilty copers’ cluster were the most likely to leave paid work to obtain a better job, start a new job or to study.

lowered workforce participation by cluster types

Those mothers who reduced their working hours by at least 10 per cent between waves reported a relatively substantial reduction; an average reduction of 10 hours per week, or 31.7 per cent. Variations between clusters in the distribution of instances of reducing working hours were negligible. However, when they chose to reduce hours, mothers from the ‘Indifferent and


147

struggling’ cluster did so to a larger degree (11.4 hours per week) than the other clusters. In contrast, mothers from the clusters successfully managing work and family responsibilities (‘Indifferent yet successful’ and ‘Highly functioning and fulfilled’) reported the smallest reduction (8.3 and 8.8 hours per week respectively). The distribution of clusters by decline in paid working hour quartiles found similar results.

the impact of exiting paid work

Earlier descriptive analysis of clusters indicated several common characteristics among mothers in the ‘Aspiring and struggling’ and ‘Indifferent and struggling’ clusters: long working hours, high work overload, lower scores on health measures, and low satisfaction with family life and parenthood. This raises the question as to whether exiting from paid work, or reducing paid working hours, is likely to lead to improved outcomes for mothers in these clusters.

Figures 3 and 4 present the self-reported physical and mental health scores of mothers who exited the workforce5 (excluding mothers who left the workforce to care for a newborn, or due to own sickness, disability or injury) for each cluster. Scores are reported for the year the mother left the workforce as well as for the preceding and subsequent years. In addition, the mean score of the total sample6 for each cluster is included as baseline information. Compared to this total sample, mothers who left paid work generally reported considerably lower physical and mental health scores in the year prior to leaving paid work. Two clusters were exceptions: the mean physical health scores of mothers in the ‘Indifferent and struggling’ and ‘Guilty copers’ clusters were both higher than the total sample mean.

Additional analysis of mothers who left paid work found a lower level of work participation in the previous year, with these mothers spending 30 per cent less time in paid work than mothers who remained in paid work. This lower level of participation prior to leaving paid work may be associated with their generally lower self-reported physical and mental health.

Mothers’ self-reported physical health scores, on average, remained the same after leaving paid work in most clusters (Figure 3). Some variation between clusters is observable (notably the consistent improvement among mothers in the ‘Aspiring and struggling’ cluster); however, this difference in change between clusters did not reach statistical significance.

Mental health after leaving paid work scores displayed a different pattern. Self-reported mental health score in the ‘Highly functioning and fulfilled’, ‘Indifferent and struggling’ and ‘Guilty copers’ clusters showed an observable improvement, which increased even further in the subsequent wave.7 In contrast, self-reported mental health of mothers in the ‘Treading water’ cluster showed a notable decline (Figure 4). However, this difference in change between clusters did not reach statistical significance.

Similar analysis of contrasting mean cluster scores for satisfaction with family life and parenthood before and after exiting paid work found no observable differences for any of the clusters.


148

Figure 3: Self-reported physical health before and after exiting paid work(a) by work–life balance clusters

60 65 70 75 80 85 90

Highly functioningand fulfilled

Indifferentyet successful

Aspiringand struggling

Indifferentand struggling

Treading water

Guilty copers

Self-reported physical health (SF-36) score

Left paid work sample, Wave X+1 Left paid work sample, Wave XLeft paid work sample, Wave X–1 Total sample, Wave 5

(a) Excluding mothers who left the workforce due to own sickness, disability or injury, or to care for a newborn.


Figure 4: Self-reported mental health before and after exiting paid work(a) by work–life balance clusters

60 65 70 75 80 85 90





Treading water

Guilty copers

Self-reported mental health (SF-36) score


(a) Excluding mothers who left the workforce due to own sickness, disability or injury, or to care for a newborn.



149

the impact of reduced working hours

Figures 5 and 6 present the self-reported physical and mental health scores of mothers who reduced their working hours8 (by more than 10 per cent) between waves. Mothers who reduced their paid working hours reported similar or somewhat higher physical and mental health scores in the year before reducing their paid working hours as the total sample in all clusters, with the exception of the ‘Indifferent and struggling’ cluster.

Analysis of mean cluster scores reported before and after reducing paid working hours found only marginal differences in physical and mental health (Figures 5 and 6). An exception was the ‘Indifferent and struggling’ cluster where mothers reported a notably higher mental health score in waves subsequent to reducing their working hours compared with other clusters. However, this difference in change between clusters did not reach statistical significance.

Similar analysis of satisfaction with family life and parenthood before and after reducing paid working hours found no observable differences for any of the clusters.

Figure 5: Self-reported physical health before and after reducing paid working hours by work–life balance clusters

60 65 70 75 80 85 90





Treading water

Guilty copers

Self-reported physical health (SF-36) score




150

Figure 6: Self-reported mental health before and after reducing paid working hours by work–life balance clusters

60 65 70 75 80 85 90





Treading water

Guilty copers

Self-reported mental health (SF-36) score



the impact of exiting paid work or reducing working hours on household income

Although the range of responses that mothers gave as their main reason for exiting paid work did not include household income, it is reasonable to assume that it was an influencing factor in their decision. Figures 7 and 8 present the equivalised household disposable income of mothers (excluding mothers who left the workforce to care for a newborn, or due to own sickness, disability or injury) for each cluster, in the year they left the workforce or reduced their paid working hours as well as scores for the previous and subsequent years. In addition, the Wave 5 mean score for the total sample of each cluster is included as baseline information.

Compared to this total sample, mothers who left paid work reported a substantially lower household disposable income the year before they left paid work than the total sample (Figure 7). This may partly be due to the considerably lower level of work participation by mothers prior to leaving work (as noted earlier). In the years after leaving paid work, their average household disposable income remained the same. There was some variation between clusters, such as an observable increase after exiting paid work in the ‘Highly functioning and fulfilled’ cluster and a decline in the ‘Treading water’ cluster. However, these variations did not reach statistical significance.

Additional analysis, not shown here, found that while mothers’ gross wages and salary for the financial year declined by more than half in the year after leaving paid work, at the household level the ratio of work or investment income to pensions and benefits remained the same for


151

most clusters. An exception was the ‘Aspiring and struggling’ cluster where a notable increase in pensions and benefits was observable. The relatively constant disposable income of families despite mothers’ exit from paid work may partly be explained by the relatively low hours in paid work of these mothers (prior to leaving paid work) and a subsequent increase, at a similar rate, in pensions and benefits and in the wages and salary of their partner9 and/or income from business and investment. This topic merits further investigation.

Among mothers who reduced their working hours, a slight increase in their disposable household income was evident in all clusters (Figure 8). While some between-cluster variation was observable, it did not reach statistical significance.

Additional analysis, not shown here, found that mothers’ gross wages and salary for the financial year increased by 8 per cent on average despite their reduced hours in paid work. In addition, the ratio of work or investment income to pensions and benefits at the household level showed a small decline across all clusters. Increasing disposable income of families despite reduced paid work hours by mothers could be explained by the increase in their own as well their partner’s9 wage and salary and/or income from business and investment.

Figure 7: Equivalised household disposable income before and after exiting paid work(a) by work–life balance clusters

$0 $10,000 $20,000 $30,000 $40,000





Treading water

Guilty copers

Total sample

Equivalised household financial year disposable income in 2007 dollars


(a) Excluding mothers who left the workforce due to own sickness, disability or injury, or to care for a newborn.Source: HILDA, Waves 1–6.


152

Figure 8: Equivalised household disposable income before and after reducing paid working hours by work–life balance clusters

Equivalised household financial year disposable income in 2007 dollars


$0 $10,000 $20,000 $30,000 $40,000





Treading water

Guilty copers

Total sample


5 DiscussionLosoncz and Bortolotto (2009) used a single wave of HILDA to identify six homogenous groups of Australian working mothers, each with a distinctive profile of work–life experience. Independent analysis of the first six waves of HILDA found a consistent emergence of the same six clusters in each wave. Descriptive analysis found that these clusters also have distinctive profiles in terms of sociodemographic background, family characteristics and factors related to work. Knowledge of these differences can inform targeted policy development and assist program delivery.

The distinctive demographic profile of each of the six clusters means that the unique effects of particular factors—for instance, work conditions or amount of domestic work—are not the focus. Rather than examining the contribution that a particular factor makes across a whole population, cluster analysis reveals groups in the population who experience particular arrangements of these factors. The focus is on mothers experiencing the sum of multiple negative characteristics, rather than on the characteristics themselves.

Of the six clusters, Losoncz and Bortolotto (2009) found that mothers from the ‘Indifferent and struggling’ and ‘Aspiring and struggling’ clusters experience the strongest tension between work and family responsibilities. Mothers in these two clusters were characterised by long working hours, high work overload and a lack of support from others. As their long hours in paid work were not compensated by reduced time spent on household activities, they spent the most time in combined (paid and unpaid) work. For unpartnered mothers, this is attributable to the absence


153

of another adult in the household. Yet even in coupled families, the time spent in paid work by a partner was often unaffected by the long working hours of the mother, with both parents working long hours.

This finding indicates a number of possible approaches to improve work–life balance for these mothers. Encouraging support from others within and outside of the home could help reduce the level of overload they face. More manageable combined hours of paid and unpaid work would be desirable for those mothers in the high tension clusters. Assisting fathers to adjust their working hours would achieve more balanced working hours for the family unit as a whole and allow fathers play a greater role in family and household activities, which would also reduce the pressure on mothers.

The two clusters with strong work–life balance tension also shared lower outcomes on health measures and low satisfaction with family life and parenthood. This raised the question of whether mothers in high work–life tension clusters were more likely to exit paid work or reduce their working hours than mothers in other clusters and, if so, whether this would lead to improved physical and mental health, and satisfaction with family life and parenthood.

These questions were examined in the present paper. Mothers from clusters with high work–life conflict did not show a higher tendency to exit from paid work than mothers from other clusters. The clusters with the highest prevalence of exiting paid work, either to care for a newborn or for other reasons, were the ‘Indifferent yet successful’ cluster followed by the ‘Indifferent and struggling’ cluster. This suggests that mothers’ role preference may be a stronger predictor of withdrawing from paid work than their actual experience of work–family strain, noting that mothers’ role preference is likely to be influenced by their future plans around workforce participation. This finding indicates that policies that promote greater work–life balance may have a different level of influence on mothers’ workforce participation based on their level of work role identification and particular events in their life course. In other words, there is unlikely to be a ‘one size fits all’ path to improving the lives of mothers experiencing work–life tension as mothers may be choosing to remain in high tension situations in order to follow their life aspirations. Support for these mothers will necessarily take a different form.

The proportion of mothers who reduced their work hours was generally the same across the clusters. However, the magnitude of reduction was greater among mothers in high work–life conflict clusters. For mothers who reduced their workforce participation, those from the ‘Highly functioning and fulfilled’, ‘Indifferent and struggling’, and ‘Guilty copers’ cluster reported a notable improvement in their mental, but not physical, health in waves subsequent to leaving paid work or reducing their paid working hours. Mothers from the ‘Aspiring and struggling’ cluster did report some improvement in their physical health after leaving paid work, but the impact of reduced working hours was negligible. Contrary to expectations, reducing working hours or exiting paid work did not lead to improved satisfaction with family life and parenthood in any of the clusters.

Interestingly, mothers’ exit from paid work or reduction in paid working hours, on average, did not lead to a decline in household disposable income or an increase in household pension or benefits. This topic requires further investigation; however, initial analysis suggests that for mothers exiting paid work this may partly be explained by their already low working hours during


154

the preceding year, and a continuance of some employment benefits, such as paid maternity leave, after leaving paid work. Other feasible explanations for non-decline, on average, in household disposable income in families where mothers reduced their workforce participation may include an increase in their partner’s wage and salary, and/or income from business and investment.

Australian working mothers are faced with the challenge and responsibility of reconciling family and work commitments. While during recent decades mothers have increased their participation in the labour market greatly, fathers have not increased their participation in unpaid household work to a matching degree. Without equal sharing of the dual role of earner and carer between mothers and fathers, mothers will inevitably feel the work–family tension more keenly. It is notable that the positive impact of reduced labour force engagement on self-reported mental health was limited to the ‘Indifferent and struggling’ cluster. This tends to indicate that manageable paid working hours may improve the work–life experience of mothers with low work role identification. However, for mothers with high work role identification other strategies should be considered, such as support from others outside and within the home, and taking shared responsibility for the balance of caring and work responsibilities within couples.

While the focus of this policy note was on mothers experiencing high work–family strain, it should be noted that over one-third of Australia working mothers (35 per cent) reported low work–family strain while an additional 38 per cent felt that they more or less manage the day-to-day demands of their work and family life. The size and the characteristics of these groups indicates that Australian mothers can successfully combine work and family without compromising their health or aspirations—as long as they can share the load.

Endnotes1 This social policy note is based on a paper presented at the Household, Income and Labour

Dynamics in Australia (HILDA) Survey Research Conference 2009, Melbourne, 16–17 July. It uses unit record data from the HILDA survey (release 6.0). The HILDA project was initiated and is funded by the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA) and is managed by the Melbourne Institute of Applied Economic and Social Research (MIAESR).

2 Role and lifestyle preferences may contribute to the degree of work–life strain—that is, people whose reality does not match their preference tend to experience higher level of strain (Brunton 2006).

3 For more information see Watson and Wooden (2002).

4 Due to the small cell sample size a test to establish statistical significance was not run.

5 The 421 instances of exiting paid work in the sample included some multiple episodes by the same mothers. As subsequent analysis on selected outcome indicators did not show a significant difference between the multiple and single episode groups, only first episodes (n=391) were selected for further analysis.


155

6 The authors included scores from Wave 5 to provide continuity with earlier analysis results. Additional analysis, not presented in here, found no statistically significant differences between the first six waves on these measures.

7 Over one-quarter of mothers re-entered the workforce in the subsequent wave, but the proportion of mothers re-entering the workforce was evenly distributed between the clusters.

8 The 1,140 instances of reducing paid working hours included some multiple episodes by the same mothers. Distribution analysis of single versus multiple episodes by cluster types found multiple episodes to be more frequent (proportionally) in the ‘Treading water’ and ‘Guilty copers’ clusters. However, this difference did not reach statistical significance. As subsequent analysis on selected outcome indicators did not show a significant difference between the multiple and single episode groups only first episodes (n=883) were selected for further analysis.

9 Due to the high frequency of missing values for partner’s wages and salary a direct analysis of this measure was not included in this paper.

ReferencesAdlaf, EM & Zdanowich, YM 1999, ‘A cluster-analytic study of substance problems and mental health among street youth’, American Journal of Drug and Alcohol Abuse, vol. 25, pp. 639–60.

Alexander, M & Baxter, J 2005, ‘Impacts of work on family life among partnered parents of young children’, Family Matters, vol. 72, pp. 18–25.

Allan, C, Loudoun, R & Peetz, D 2007, ‘Influences on work/non-work conflict’, Journal of Sociology, vol. 43, no. 3, pp. 219–39.

Australian Bureau of Statistics (ABS) 2006, Australian Social Trends, cat. no. 4102.0, ABS, Canberra.

Baxter, J 2009, ‘Mothers’ timing of return to work by leave use and pre-birth characteristics’, Journal of Family Studies, vol. 15, pp. 153–66.

Baxter, J, Gray, M, Alexander, M, Strazdins, L & Bittman, M 2007, Mothers and fathers with young children: paid employment, caring and wellbeing, Social Policy Research Paper no. 30, Australian Government Department of Families, Community Services and Indigenous Affairs, Canberra.

Berry, HL 2008, Twelve types of Australians and their socioeconomic, psychosocial and health profiles, paper prepared for the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs, Canberra.

Bowman, DD 2009, ‘The deal: wives, entrepreneurial business and family life’, Journal of Family Studies, vol. 15, pp. 167–76.

Brunton, C 2006, ‘Work, family and parenting study: research findings’, Centre for Social Research and Evaluation, Ministry of Social Development, New Zealand.


156

Fagan, C & Burchell, B 2002, Gender, jobs and working conditions in the European Union, European Foundation for Improvement of Living and Working Conditions, Dublin.

Forma, P 2008, ‘Work, family and intentions to withdraw from the workplace’, International Journal of Social Welfare, vol. 18, pp. 183–92.

Galinsky, E 2005, ‘Children’s perspectives of employed mothers and fathers: closing the gap between public debates and research findings’, in DF Halpern & SE Murphy (eds), From work–family balance to work–family interaction: changing the metaphor, Lawrence Erlbaum Associates, London, pp. 219–37.

Gray, M, Qu, L, Stanton, D & Weston, R 2004, ‘Long work hours and the wellbeing of fathers and their families’, Australian Journal of Labour Economics, vol. 7, no. 2, pp. 255–73.

Grenhaus, JH, Collins, KM, Singh, R & Parasuraman, S 1997, ‘Work and family influences on departure from public accounting’, Journal of Vocational Behaviour, vol. 50, pp. 249–70.

Keith, PM & Schafer, RB 1980, ‘Role strain and depression in two-job families’, Family Relations, vol. 29, no. 4, pp. 483–88.

Losoncz, I 2009, ‘Personality traits in HILDA’, Australian Social Policy, vol. 8, pp. 169–98.

Losoncz, I & Bortolotto, N 2009, ‘Work–life balance: the experience of Australian working mothers’, Journal of Family Studies, vol. 15, no. 2, pp. 122–38.

Messenger, JC 2004, Working time and workers’ preferences in industrialized countries: finding the balance, Routledge, London.

Pocock, B, Skinner, N & Williams, P 2007, ‘Work, life and time’, The Australian Work and Life Index, 2007, Centre for Work & Life, Hawke Research Institute, University of South Australia.

Raymo, JM & Sweeney, MM 2006, ‘Work–family conflict and retirement preferences’, Journal of Gerontology & Social Sciences, vol. 61, pp. 161–69.

Reynolds, J & Aletraris, L 2007, ‘Work–family conflict, children, and hour mismatches in Australia’, Journal of Family Issues, vol. 28, pp. 749–72.

Strazdins, L, Lucas, N, Mathews, B, Berry, H, Rodgers, B & Davies, A 2008, ‘Parent and child wellbeing and the influence of work and family arrangements: a three cohort study’, report to the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs.

Watson, N & Wooden, M 2002, The Household, Income and Labour Dynamics in Australia (HILDA) survey: Wave 1 survey methodology, HILDA Project Technical Paper Series (no. 1/02), University of Melbourne.

157

Book reviewAustralian Social Policy Journal No. 9


158

159

Utopias and revolutionsGornick, J & Meyers, M (eds) 2009, Gender equality: transforming family divisions of labor, Verso, London and Esping-Anderson, G 2009, The incomplete revolution: adapting welfare states to women’s new roles, Polity Press, London

Reviewer: Julie Connolly

School of Political Science and International Studies, University of Queensland

Introduction

Policy attempts to facilitate work–life balance, while simultaneously advancing social welfare, are the subject of the two texts under review here. Gender equality: transforming family divisions of labor is the sixth publication in Verso’s Real Utopia Project. Gornick and Meyers have assembled a number of international experts from both Europe and the United States (US) to critically evaluate their ‘real utopia’ consisting of a suite of policies to facilitate women’s long-term attachment to the workforce and minimise gender specialisation in household tasks while supporting child welfare. Esping-Anderson has assembled recent data pertaining to the plight of the welfare state’s attempts to ameliorate the distributive consequences of women’s changing roles in order to advance social equality. Read together these two texts provide an invaluable guide to current research on social policy in the related areas of work and welfare, contextualised by what Esping-Anderson refers to as the ‘incomplete revolution’ in women’s roles, household labour and caring responsibilities. Each text is argumentative and engaged, providing an opportunity for experts and non-experts alike to reflect on international policy debates about family structure, employment opportunities and caring responsibilities.

real utopias

Gornick and Meyers argue that achieving symmetry between men and women, presumably in heterosexual couples, is a real possibility, but one which will nonetheless take some ‘mildly coercive’ social policy to achieve. Their concerns about gender inequity, the time crunch experienced by working parents, and child welfare, are warranted and familiar to those already engaged with social policy debates. Their antidote to these problems involves extending family leave arrangements, having greater working time regulation and investing in early childhood service provision. For the most part their interlocutors agree that these mechanisms would help accelerate a resolution to the conflicts that have emerged between work and care. But there are some noticeable exceptions.


160

Book structure

Following the introductory essay in which Gornick and Meyers set out their real utopia for a dual earner/dual care giver society, the collection of essays is divided into three parts—debates over the:

�� principles entailed in this ideal world

�� details of the proposal

�� practicalities of implementing the proposal.

While intuitively reasonable, this structure does not neatly compartmentalise the debate because the authors included in each section stray into areas of significance for the others. One factor that becomes increasingly apparent on reading the contributions is that the proposals under analysis are affected by existing institutional and policy infrastructure. This means that the politics of change will differ between countries. And, moreover, that one size does not fit all. The inclusion of commentators from the United States demonstrates this point throughout the text. Even where commentators agree that policy can provide solutions to the problems under analysis, there is considerable concern that importing policies perhaps more suited to Scandinavian social democracy will fail in the US. The book could have benefited from an alternative configuration of contributors, perhaps according to country of origin, to better structure the debate. This would work to reveal tensions in Gornick’s and Meyer’s real utopia and the limits of its applicability.

Arguments offered by contributors

Barbara R Bergmann, arguing from the perspective of a highly skilled professional in the US labour market, insists that family leave simply reinforces a gendered division of household responsibilities. Moreover, she believes that work–time regulation will be almost impossible to achieve in highly competitive industries where job redesign might facilitate women’s caring responsibilities but inhibit career aspirations. For Bergmann the only solution is to outsource care. But as other commentators note, this solution presupposes the availability of unskilled and likely low-paid workers to assume responsibility for care-related and other domestic tasks.

This argument points to the chief shortcoming of Gornick and Meyers’ real utopia: the type of equality envisaged seems to be between working couples, not between social classes. The absence of a class perspective in their real utopia is further highlighted by Ruth Milkman, who argues that class status is predictive of family formation and women’s aspirations. It is interesting that this sensitivity to the class dynamics that sustain social inequalities, including those between men and women, is only articulated by the American contributors, with further evidence introduced by Shireen Hassim whose concern, however, is the appropriateness of such a real utopia for third world countries.

Other areas of dissension between the contributors include the following debates:

�� Should the objective of further policy intervention be substantive equality between men and women as envisaged by Gornick and Meyers, or more simply to diminish constraints on individual decision-making when it comes to managing work, care and self-realisation?

utoPIAS ANd revolutIoNS

161

�� What are the precise attitudinal and institutional barriers to the implementation of the proposal? Do women’s preferences matter? Is greater working time regulation even possible?

�� Do Gornick and Meyers presume the continued existence of a welfare state that has been largely eclipsed by a combination of neo-liberalism and globalisation?

Explorations of these questions and various answers to them are contained in the subsequent essays, a complete analysis of which is beyond the scope of this review. Nonetheless the inclusion of different perspectives based on different country experiences is what makes this book worth reading.

the incomplete revolution

Esping-Anderson’s text is less self-consciously utopian. Nonetheless, his principled commitment to a re-articulated welfare state as an engine for social equality is clear. Indeed there is a hidden utopianism that animates his arguments. Namely, that despite the evidence of increasing polarisation among income groups, facilitated by assortative marriage between highly educated and thus high-earning couples, greater social equity is also within reach. Obtaining this end requires significant social investment in family policy, child welfare and support for retirement incomes. The function of the welfare state in social protection from cradle to grave is thus reconstructed with an eye to intergenerational equity and the long-term productivity of political economies dependent on a shrinking labour force.

Part 1

In the first part of the book, Esping-Anderson describes the social transformations that have ushered in the incomplete revolution. He describes growing social inequalities and their implications for families who aspire to more children than it seems they can afford, children born into disadvantage, and future retirees who appear to lack secure incomes. While the arguments and data presented here are reasonably convincing, the analysis is not entirely compelling because assortative marriage can hardly be thought to exhaust the reasons for social inequality however much it may be implicated in the reproduction of certain class advantages. This text is largely silent on the changing structure of economic development in advanced economies, which likely carries explanatory significance for structures of inequality.

Nonetheless, Esping-Anderson’s rejoinders to welfare state critics throughout the book are noteworthy. He criticises social analysts who too quickly presume that social structures have validity, particularly normative validity, beyond their immediate functionality. Esping-Anderson does not engage in debates about what form the family should take. Instead, he is interested in achieving a ‘social equilibrium’ between family structures, conditions of employment and the capacity to provide care. His interest is in maintaining social cohesion rather than a particular set of family arrangements.

Part 2

The second part of the book is devoted to an analysis of those policy areas involved in managing the incomplete revolution: family, childhood and ageing. Esping-Anderson draws on a wealth of


162

data from European countries to make a convincing argument that long-term investment in all three arenas is consistent with the sustainability of the welfare state and the health of national economies that increasingly rely on improvements in human capital to generate additional wealth and productivity.

Esping-Anderson’s pragmatic touchstone throughout this text is that certain kinds of costs, such as caring for children and the elderly, will rise regardless of government in/action. Therefore the important question becomes how will these costs be met and who will be most advantaged. Acknowledging economic principles, Esping-Anderson aims for the most ‘efficient’ solution to social policy problems; whether or not he has succeeded is likely to be the subject of future scholarly and public debates.

conclusion

In their concluding essay, Gornick and Meyers are confident that their proposals for establishing a real utopia to improve both women’s labour force participation and child welfare have withstood the barrage of criticisms contained in the collection of essays in the book and that with some small modifications may be amenable to implementation in a wide variety of circumstances. I was not so convinced, but applaud the audacity of the attempt. Reading these essays in conjunction with Esping-Anderson’s book is particularly instructive. One book poses questions that the other answers. Both books are written from a commitment to social equity so that what could have been dry policy analysis is enlivened by a real interest in social transformation. Both books will be of interest to expert and non-expert audiences interested in the future of the welfare state and its capacity to support women’s new roles.

163

Guidelines for contributorsAustralian Social Policy Journal is currently published annually and features current research and analysis on a broad range of issues topical to Australia’s social policy and its administration. Regular features include major articles, social policy notes and book reviews.

Australian Social Policy Journal is a fully refereed academic journal; all submissions of major articles and social policy notes to the journal are subject to an external blind peer review. The journal is recognised by the Australian Research Council’s (ARC) Excellence in Research for Australia (ERA) Ranked Journal List of refereed journals.

Submissions are accepted from academic researchers, government employees and other relevant practitioners. Submissions that contribute to current social policy research issues and debates are particularly encouraged.

Submissions can be forwarded by email to <[email protected]>, or posted to:

Publications UnitResearch and Analysis BranchAustralian Government Department of Families, Housing, Community Services and Indigenous AffairsPO Box 7576Canberra Business Centre ACT 2610

For more information:

Phone: (02) 6244 5458Fax: (02) 6133 8387Email: [email protected]

Content is compiled by the Research and Analysis Branch of the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs (FaHCSIA).

Copies are available to the public in both printed and electronic format, and can be downloaded from the FaHCSIA internet site at <www.fahcsia.gov.au/research>.

164

Writing content

1. Major articles should be between 4,000 and 8,000 words, social policy notes between 2,500 and 4,000 words and book reviews 1,200 to 1,500 words. Authors should discuss any possible variations to these lengths with the Publications Unit.

2. All submissions should include an abstract and up to six keywords.

3. It is understood that submissions have not been published previously, nor are they being considered for publication elsewhere, and that content is original. If author(s) wish to have dual publication considered, they must seek permission from the Publications Unit to reprint. Full acknowledgement of prior publication in Australian Social Policy Journal must be given where applicable.

4. Authors must make every effort to ensure that, to the best of their knowledge:

�� all information presented in the paper is accurate and appropriately referenced

�� conclusions drawn are appropriately supported by the analysis and/or the evidence presented in the paper

�� correct spelling and grammar is employed consistently throughout the paper.

5. Content should be written in plain, non-technical English, and must not use:

�� discipline-specific jargon unless necessary, in which case this should be clearly defined within the article

�� acronyms unless preceded by the full title in the first instance

�� language which discriminates on the basis of gender, ethnicity, age and/or disability.

6. Professional practices regarding the reproduction of copyright material must be followed.

7. Authors should:

�� use the Harvard system (author–date) for referencing

�� use endnotes rather than footnotes

�� clearly source figures, tables and other data where necessary.

More information on Commonwealth Style guidelines is available from Snooks & Co. 2002, Style Manual 6th Edition, John Wiley & Sons.

165

Submitting content1. The name, title, affiliation(s) and contact details of the author(s) must be included with the

submission.

2. Submissions must include an electronic copy of the submission emailed to <[email protected]>. Text should be sent as an attached Microsoft Word Document (.doc) or Rich Text Format (.rtf) file. All figures (that is, charts and diagrams) presented in the submission should be sent as Microsoft Excel (.xls) files (or eps files) and must include the data from which the figures were generated. Authors who wish to use alternative word processing, spreadsheet or graphics packages must contact the Publications Unit.

3. Content, particularly figures and tables, should be presented in greyscale (black and white). Colour is not employed in the Australian Social Policy Journal.

4. Anyone whose personal details (name, contact details, etc.) are to appear in a publication must provide written approval these details to be printed and/or promoted by electronic media including the internet.

review process

1. Submissions will be reviewed by the Research Publications Editor. At this point, the author may be contacted to provide further information or revise content. Authors are requested to respond to these comments in a timely manner.

2. A decision whether or not to publish a submission will be based on the following criteria:

�� relevance to the Australian Government Department of Families, Housing, Community Services and Indigenous Affairs portfolio

�� contribution to current social policy research issues and debates

�� academic quality and readability of the paper

�� blind peer review process

3. All submissions of major articles or social policy notes to the Australian Social Policy Journal are subject to an external blind peer review. Two referees are engaged to review and comment upon submissions. The identity of the author(s) and referee(s) will not be revealed to the other party.

Authors will be required to respond to these comments, and revise submissions where necessary, within one month of receiving the referee report.

166

Publishing process

1. Submissions will be proofread and style-edited by the Publications Unit and/or a professional editor. Authors will be required to review style changes and respond within two weeks. Content changes are not permitted.

2. Once content has been cleared and submitted for typesetting, author revisions are not permitted.

3. The Publications Unit will nominate a release date for the publication. Once the publication is released it will be available to the public in both printed and electronic format, and can be downloaded from the FaHCSIA internet site at <http://www.fahcsia.gov.au/about/publicationsarticles/research/austsocialpolicy/Pages/default.aspx>

4. Author(s) will receive two printed copies of the published work.

Subscription Form (Research Publications)

PLEASE NOTE: Subscription, publications and delivery are free-of-charge.

Compulsory fields, marked ‘*’ must be completed for the subscription to be processed.

Please write clearly.

* Name:

E-notification

I would like to be notified via email when new FaHCSIA research publications are released.

* Email:

Printed copies

I would like to automatically receive printed copies of the following publications:

Australian Social Policy Journal (annual)

Social Policy Research Papers

Occasional Papers

Statistical Papers

Position: Department/Section:

* Organisation / Business Name:

* Postal Address:

* Suburb * State: * Postcode: * Country

�� Print copies of these publications are available (free-of-charge) from our distributor, telephone 1800 050 009 (freecall).

�� Electronic copies are available from the FaHCSIA website http://www.fahcsia.gov.au/research.

Return by mail: Research Publications Unit

Research & Analysis Branch

Department of Families, Housing, Community Services and Indigenous Affairs PO Box 7576 Canberra Business Centre ACT 2610

Return by fax: (02) 6133 8387

Return by email: [email protected] (subscription form can be downloaded from www.fahcsia.gov.au/research)

PO Box 7576Canberra Business Centre ACT 2610Tel 1300 653 227TTY 1800 260 402Fax (02) 6244 6589Email [email protected] www.fahcsia.gov.au

australian social policy journal no. 9 2010

Documents