Structural analyses of occupational homogamy

Paul Lambert (University of Stirling, UK)Dave Griffiths (University of Stirling, UK)

Mark Tranmer (University of Manchester, UK)

14 November 2011, SOFI, Stockholm

‘Social Networks and Occupational Structure’ www.camsis.stir.ac.uk/sonocs

Three ways of analysing social connections between occupations…

1) Social Network Analysis2) Social Interaction

Distance analysis3) Random Effects

(Multilevel) Modelling

3

Network analysis to look for influential channels of social connections between occs. (camsis.stir.ac.uk/sonocs)

1) Social Network Analysis

Hypothetical network: 469 US OUGs & micro-classes

Green: prof.; Blue: routine non-mnl; Red: manual; Yellow: primary; Green: military

Dental hygienists

Medical professionals

Medical and dental technicians

(Four different isolated components with internal links within microclass but no external links)

(further isolated components)

‘Pseudo-diagonal’ or ‘situs’

<-Hypothetical

Actual composition of occupational networks in USA in 2000: links reflect stratification as much as they do microclasses and psds

Red to violet for low to high CAMSIS (grouped into 7).

Structures similar to CAMSIS scales. Using Kamada-Kawai algorithm and no manual adjustment (expect removing some occs with no ties/relations)

Romania, 2002

Philippines, 2000

Venezuela, 2001

7

2) Social Interaction Distance Analysis (www.camsis.stir.ac.uk : correspondence analysis; RC-II association models)

Husband’s Job Units Occ. Units ↓ → 1 2 .. 407

Derived dimension scores ↓ → 75.0 70.0 .. 10.0 Wife’s 1 72.0 30 15 .. 0

Job 2 72.5 13 170 .. 1

Units .. .. .. .. .. ..

407 11.0 0 2 .. 80

A large cross-tabulation of pairs of occupations is modelled; dimension scores help predict frequency of occurrences in cells; scaled dimension scores arethen presented as CAMSIS scale scores.

• Using CAMSIS approaches, www.camsis.stir.ac.uk

• First dimension of SID scales is usually ‘social stratification’– We’d interpret it as the contour

of social reproduction– Gradational, but ‘lumpy’ for

operational reasons (occ.s)– ‘Specificity’ (many scales!)

• Dimensions: – 1 main one– numerous subsidiary patterns

• Boundaries: – None(?)

8

2040

6080

100

CA

MS

IS

ISEI

MalesFemales

Source: IPUMS-I, N=778k with occ dataData is coded here to ISCO88 3-digit minor groups

Venezuela, 2001

Dimensions=1; Boundaries= none; or maybe 1 in Ro?

Griffiths/Lambert, RC28, April 2011 9

20 40 60 80 100

Venezuela 2001111

131 223311323324 333334

345513515

523600

611

613

614615 711712714 732 831832833834913

914

MC

S

FCS

20 40 60 80 100

Phillipines 2000

12112113

121 215245324 346

348413512 513514 521 522

622629631632641

642 712721 831832913M

CS

FCS

20 40 60 80 100

Romania 2002235

348828MC

SFCS

20 40 60 80 100

USA 2000

72151274 362

384395441 455523534

670730771774 894933

MC

S

FCS

All microdata from IPUMS-I. CAMSIS scales at www.camsis.stir.ac.uk.Histograms show distribution of male scale for all adults in work. Scatterplots show unweighted male-female scores unweighted, ISCO88 3-digit or census SOC for USA

CAMSIS scale distributions

Griffiths/Lambert, RC28, April 2011 10

1101. Jurists1102. Health professionals

1103. Professors and instructors1104. Natural scientists

1105. Statistical and social scientists1106. Architects

1107. Accountants1108. Journalists, authors, and related writers

1109. Engineers1201. Officials, government and non-profit organizations

1202. Managers1203. Commercial Managers

1204. Building managers and proprietors1301. Systems analysts and programmers

1302. Aircraft pilots and navigators1303. Personnel and labor relations workers

1304. Elementary and secondary school teachers1305. Librarians

1306. Creative artists1307. Ship officers

1308. Professional, technical, and related workers, n.e.c.1309. Social and welfare workers

1310. Workers in religion1311. Nonmedical technicians

1312. Health semiprofessionals1313. Hospital attendants

1314. Nursery school teachers and aides3101. Real estate agents

3102. Other agents3103. Insurance agents

3104. Cashiers3105. Sales workers and shop assistants

3201. Telephone operators3202. Bookkeepers and related workers

3203. Office and clerical workers3204. Postal and mail distribution clerks

4101. Craftsmen and kindred workers, n.e.c.4102. Foremen

4103. Electronics service and repair workers4104. Printers and related workers

4105. Locomotive operators4106. Electricians

4107. Tailors and related workers4108. Vehicle mechanics

4109. Blacksmiths and machinists4110. Jewelers, opticians, and precious metal workers

4111. Other mechanics4112. Plumbers and pipe-fitters

4113. Cabinetmakers4114. Bakers

4115. Welders and related metal workers4116. Painters

4117. Butchers4118. Stationary engine operators

Bricklayers, carpenters & related4120. Heavy machine operators

4201. Truck drivers4202. Chemical processors

4203. Miners and related workers4205. Food processors

4206. Textile workers4207. Sawyers and lumber inspectors

4208. Metal processors4209. Operatives and kindred workers, n.e.c.

4210. Forestry workers4301. Protective service workers

4302. Transport conductors4303. Guards and watchmen

4304. Food service workers4305. Mass transportation operators

4306. Service workers, n.e.c.4307. Hairdressers

4308. Newsboys and deliverymen4309. Launderers and dry-cleaners

4310. Housekeeping workers4311. Janitors and cleaners

4312. Gardeners5101. Fishermen

5201. Farmers and farm managers5202. Farm laborers

9990. Members of armed forces

USA

Romania

Phillipines

Venezuela

Male CAMSIS scale scores across four countries using 'microclass' units.

3) Multilevel modelling

• In general, can analyse Individual level data (i) with clustering in higher level units (j) (occupations; person groups)

45

67

89

Labo

ur in

com

e

-40 -20 0 20 40 60Hours

Log monthly income p_1

Overall regression

05

1015

Fitt

ed r

egre

sio

n lin

es

-40 -20 0 20 40 60Hours

Job lines (if n >= 10)4

56

78

9La

bour

inco

me

-40 -20 0 20 40 60Hours of work

qpsu resids Overall

Random intercepts

45

67

89

Labo

ur in

com

e-40 -20 0 20 40 60

Hours of work

qpsu resids Overall

Random coefficients

Source: BHPS, adults in work in Wave A (1991)

Labour income for individuals clustered in occupations

MLM’s on social

network data?

• We could analyse some other process/outcome

• Own job = f(x), i(spouse’s job)

Deviance: 2015365.4Intra-cluster correlation: .42310141Level 1 variance: 124.748Level 2 variance: 91.491045y1 -1.0001604 95.284935 2.2581206 2.4131479 gwage _cons _cons _cons wfcamsis: wfcamsis: lns1_1_1: lnsig_e:e(b)[1,4]

. xtm_var

LR test vs. linear regression: chibar2(01) = 1.4e+05 Prob >= chibar2 = 0.0000 var(Residual) 124.748 .3441898 124.0752 125.4244 var(_cons) 91.49105 11.43816 71.60805 116.8948hocc: Identity Random-effects Parameters Estimate Std. Err. [95% Conf. Interval]

_cons 95.28493 .9529303 99.99 0.000 93.41723 97.15264 gwage -1.00016 .011478 -87.14 0.000 -1.022657 -.977664 wfcamsis Coef. Std. Err. z P>|z| [95% Conf. Interval]

Log likelihood = -1007682.7 Prob > chi2 = 0.0000 Wald chi2(1) = 7592.93

max = 65866 avg = 2022.0 Obs per group: min = 8

Group variable: hocc Number of groups = 130Mixed-effects ML regression Number of obs = 262855

Computing standard errors:

Iteration 1: log likelihood = -1007682.7 (backed up)Iteration 0: log likelihood = -1007682.7

Performing gradient-based optimization:

Performing EM optimization:

. xtmixed wfcamsis gwage if freq > 0 ||hocc:, mle variance

Or analyse the counts of occurrences of links themselves?

Y=# of links; Hocc at level 2, wocc at level 1. (Philippines)

(ICC’s around 1 – 10%)

BHPS own, family & friends’ jobs

020

4060

Spo

use

Par

ent

Chi

ldO

ther

fam

ilyU

nrel

ated

/oth

erFa

ther

rep

Mot

her r

epFr

iend

(wav

e B

)

Frie

nd (w

ave

H)

Frie

nd (w

ave

J)

Frie

nd (w

ave

L)

Frie

nd (w

ave

N)

Frie

nd (w

ave

P)

Frie

nd (w

ave

R)

Alte

r's F

athe

r rep

Alte

r's M

othe

r rep

Alte

r's F

riend

(wav

e B

)

Alte

r's F

riend

(wav

e H

)

Alte

r's F

riend

(wav

e J)

Alte

r's F

riend

(wav

e L)

Alte

r's F

riend

(wav

e N

)

Alte

r's F

riend

(wav

e P

)

Alte

r's F

riend

(wav

e R

)

mean of mcam mean of omcam

Total 481,152 459,456 940,608 Alter's Friend (wave 28,843 26,512 55,355 Alter's Friend (wave 35,912 32,814 68,726 Alter's Friend (wave 30,446 27,585 58,031 Alter's Friend (wave 35,537 32,498 68,035 Alter's Friend (wave 30,902 28,240 59,142 Alter's Friend (wave 24,785 22,599 47,384 Alter's Friend (wave 21,481 19,375 40,856 Alter's Mother rep 28,551 25,826 54,377 Alter's Father rep 45,590 41,846 87,436 Friend (wave R) 3,676 4,238 7,914 Friend (wave P) 6,150 7,219 13,369 Friend (wave N) 7,085 7,934 15,019 Friend (wave L) 9,947 10,541 20,488 Friend (wave J) 10,709 11,619 22,328 Friend (wave H) 8,458 9,031 17,489 Friend (wave B) 9,525 10,335 19,860 Mother rep 12,841 14,066 26,907 Father rep 22,674 22,732 45,406 Unrelated/other 4,079 3,829 7,908 Other family 8,063 6,614 14,677 Child 16,308 19,657 35,965 Parent 21,029 15,972 37,001 Spouse 58,561 58,374 116,935 ego 1. male 2. female Total Alter's relation to sex

Deviance: 569308.55Intra-cluster correlation: .01266878Level 1 variance: .12154432Level 2 variance: .00155958y1 .00001184 .00135396 -1.027e-06 .0961898 -3.2316707 -1.0537382 difscore mcam90 dif_mcam _cons _cons _cons splink: splink: splink: splink: lns1_1_1: lnsig_e:e(b)[1,6]

(xtmixed, j=370 occs, y=spouse link)

Appendices

French occupational codes

Total 152,834 100.00 69. Agricultural workers 3,262 2.13 100.00 66. Skilled driver 25,326 16.57 97.87 61. Skilled industrial artisans 22,529 14.74 81.29 56. Service personnel to indivduals 1,860 1.22 66.55 55. Business employees 2,790 1.83 65.34 54. administratve employees 7,773 5.09 63.51 53. Police and military 2,295 1.50 58.4352. Civil employees, service agents of 4,203 2.75 56.92 48. Supervisors 3,486 2.28 54.17 47. Technicians 4,358 2.85 51.8946. Professional administrative and com 4,540 2.97 49.0445. Professional administration in publ 1,097 0.72 46.07 44. Clergy, religious 20 0.01 45.35 43. Professional social health workers 328 0.21 45.34 42. Teachers and other employees 2,803 1.83 45.13 38. Engineers of technical businesses 948 0.62 43.2937. Executives of administrative and co 1,965 1.29 42.67 35. Professional news, arts and shows 299 0.20 41.3934. Professors, professional scientists 1,140 0.75 41.19 33. Executives of public functions 1,114 0.73 40.44 31. Other professions 964 0.63 39.7123. Business owners with 10 paid worker 678 0.44 39.08 22. Businessmen and employees 14,548 9.52 38.64 21. Artisans 6,219 4.07 29.12 14. Agricultural farming, fishermen 38,289 25.05 25.05 Occupation, unrecoded Freq. Percent Cum.

17

ICC examples analysing occurrences..

Level 2 ICC, by outcome

Model (hocc at level 2, wocc at level 1)

freq (freq >0)

Lnfreq (freq > 0)

freq2 (with zero)

Level 1 camsis score (freq > 0)

1a (Normal null regression) 0.0086 0.1290 0.0136 0.4185

1b (1a + gwage) 0.0086 0.1295 0.0136 0.4231

1f (1a – diag1) 0.0446 0.1487 0.0728 0.2370

Linear outcomes regression

LR test vs. linear regression: chibar2(01) = 7376.34 Prob >= chibar2 = 0.0000 sd(Residual) .3486321 .0002795 .3480846 .3491804 sd(_cons) .0394915 .0017737 .0361637 .0431254soc: Identity Random-effects Parameters Estimate Std. Err. [95% Conf. Interval]

_cons .0961898 .0071229 13.50 0.000 .0822291 .1101505 dif_mcam -1.03e-06 5.86e-08 -17.54 0.000 -1.14e-06 -9.12e-07 mcam90 .001354 .0001364 9.92 0.000 .0010865 .0016214 difscore .0000118 3.17e-06 3.73 0.000 5.62e-06 .0000181 splink Coef. Std. Err. z P>|z| [95% Conf. Interval]

Log likelihood = -284654.27 Prob > chi2 = 0.0000 Wald chi2(3) = 1774.12

max = 42816 avg = 2103.1 Obs per group: min = 10

Group variable: soc Number of groups = 370Mixed-effects ML regression Number of obs = 778159

Logit model

Deviance: 630316.74The intra-cluster correlation: .03463747The level 1 variance: 3.2898681 (by design)The level 2 variance: .11804137y1 .00002195 .01016455 -7.453e-06 -2.172969 -1.0683601 difscore mcam90 dif_mcam _cons _cons eq1: eq1: eq1: eq1: lns1_1_1:e(b)[1,5]

. xtme_var

. est store mlm2

LR test vs. logistic regression: chibar2(01) = 7854.33 Prob>=chibar2 = 0.0000 sd(_cons) .3435715 .015336 .3147907 .3749836soc: Identity Random-effects Parameters Estimate Std. Err. [95% Conf. Interval]

_cons -2.172969 .0617067 -35.21 0.000 -2.293912 -2.052026 dif_mcam -7.45e-06 5.52e-07 -13.51 0.000 -8.53e-06 -6.37e-06 mcam90 .0101646 .0011781 8.63 0.000 .0078555 .0124736 difscore .0000219 .0000304 0.72 0.470 -.0000376 .0000815 splink Coef. Std. Err. z P>|z| [95% Conf. Interval]

Log likelihood = -315158.37 Prob > chi2 = 0.0000Integration points = 7 Wald chi2(3) = 1599.66

max = 42816 avg = 2103.1 Obs per group: min = 10

Group variable: soc Number of groups = 370Mixed-effects logistic regression Number of obs = 778159