the role of small businesses in employing the unemployed and …/file/a1_urwin.pdf · early to mid...
TRANSCRIPT
1
The role of small businesses in employing the unemployed
and inactive
Franz Buscha and Peter Urwin
July 2012
Submission to the Work Pensions and Labour Economics Study Group
Conference
Commissioned by the Federation of Small Businesses
PLEASE DO NOT QUOTE WITHOUT PERMISSION OF THE AUTHORS
Please contact: Prof. Peter Urwin
Centre for Employment Research
University of Westminster
35 Marylebone Road, London NW1 5LS
020 7911 5000 (ext: 3080)
email: [email protected]
This report utilises data from the Labour Force Survey. The LFS is distributed by the Economic and Social Data
Service; Crown Copyright material is reproduced with the permission of the Controller of HMSO and the
Queen's Printer for Scotland.
2
Abstract: This paper presents findings from research that sheds light on the role that small
businesses play in the UK Labour market. We use data from the Labour Force Survey (LFS) to
consider differences in the characteristics of individuals securing employee jobs in small and large
firms. The suggestion from multivariate analysis is that smaller firms are an important bridge to
employment in the UK for those who are either unemployed or inactive, and they have a greater
propensity to hire these groups than larger firms. However, the onset of recession seems to have
reduced this greater propensity for small firms to employ the unemployed and inactive; and there is
some indication that, prior to 2000, larger firms were more likely to hire the unemployed and
inactive.
We also find that employees in micro-businesses and those with 11-49 employees have significantly
higher probabilities of flowing into unemployment and inactivity. Our findings are more consistent
with the suggestion that this small-firm effect arises because we find different types of worker in
firms of different sizes, rather than simply because small firm employment is more risky (whatever
the individual employees characteristics). This is very much a first attempt at investigation in this
area and we flag up a number of areas for possible further investigation.
Key words: Small to Medium Enterprises (SMEs), unemployment, inactivity, labour market
transitions
JEL Classification: J21, J63, J64
3
1. Introduction
This paper considers the importance of the small business sector to the UK economy. We present
findings from new research that sheds light on the role that the sector plays in the UK Labour
market. The study has been commissioned by the Federation of Small Businesses (FSB), which has
concerns over the continuing gaps in understanding of the role that UK small businesses play in
employment.
There is a raft of research that looks into the role that small businesses play in employment
generation (Neumark, Wall and Zhang, 2008; Anyadike-Danes, Bonner, Hart and Mason, 2009;
Haltiwanger, Jarmin and Miranda, 2010), but it provides little insight into the possibility that small
and large firms differ in ways other than their size. In this report we consider the potential for small
businesses to be carrying out a different role in the labour market, to that carried out by larger firms.
We then consider any evidence on the differing roles that small and large businesses play in the
labour market, and ask whether such evidence necessitates a re-think of policy approaches.
Our first addition to the existing literature is to use data from the Labour Force Survey (LFS) to
consider differences in the characteristics of individuals securing employee jobs in small and large
firms. In the initial sections of the report we provide an overview, using tables of transitions in the
UK labour market between unemployment, employment and inactivity. These suggest that smaller
firms are a much more important bridge to employment in the UK for those who are either
unemployed or inactive. We observe many fewer unemployed and inactive individuals transiting to
employment via the route of large firms.
We then move on to confirm that this small-business effect holds even when we control for factors
such as industry sector, occupation and a variety of other controls in a multivariate framework. We
then investigate in much more detail the role of small businesses as a pathway from inactivity and
unemployment, to employment. Having found that the unemployed and inactive as a whole are
more likely to secure employment opportunities in small firms, we use multivariate regression
techniques to identify which of the unemployed and inactive are most likely to transit through
smaller firms. For instance, this allows us to ask whether, amongst the unemployed, are those with
longer unemployment durations more or less likely to be transiting through large or small firms?
We also give some consideration to whether our key findings on small businesses as a pathway to
employment have varied over time. We present findings on the labour market transitions of
individuals from 2005 Quarter 3 to 2007 Quarter 4; and compare these to the behaviours of a
sample of similar size between 1997 Quarter 1 and 1999 Quarter 2. The aim is to analyse a period
prior to the introduction of a weight of Employment Protection Legislation (beginning with the 2002
Employment Relations Act)1 and see if transition rates have altered. In addition, we analyse data
supplied under Special Licence (SL) from the UK Data Archive that allows us to analyse a recessionary
period from 2008 Quarter 4 to 2011 Quarter 1.
1 In the previous recession, unemployment was rising from mid-1990 and peaked around Q3 1993; it then fell
to pre-recession levels in 1997, but continued to fall until 2003. The dates for our earlier period of analysis
have been chosen so that we are considering a period of relative stability, after the recessionary period of the
early to mid 1990s.
4
In this paper we follow usual convention and consider Small to Medium Sized Enterprises (SMEs) as
those that have less than 250 employees. Within this SME category, firms with 1 to 49 employees
are ‘Small’, whilst ‘Micro’ enterprises are those that employ less than 10 employees. Latest figures
suggest that in the UK the SME sector accounts for 59 per cent of all private sector employment and
49 per cent of all private sector turnover2.
The overall number of firms (or number of employee jobs in these firms) at any one point in time is a
snapshot of very dynamic underlying processes. In the case of employment, the ‘net’ number of jobs
at any one point in time is the result of high levels of on-going job ‘creation’ and job ‘destruction’;
the balance between the two determining the overall level of employment within any particular
category of firm size. Put more technically, the net number of jobs measured at a particular point in
time will be a result of significant, on-going gross worker flows into, and out of, employment.
Similarly, the net number of enterprises [particularly in the small firm category] at any one point in
time is underpinned by a large amount of ‘churn’, as firms are constantly set-up and closed down.
Within the literature investigating these more dynamic underlying processes, the question of which
size firms contribute most to job creation has its roots in the work of David Birch (1979, 1981 and
1987), whose studies of the US suggested that between 1969 and 1976, 66% of all net new jobs
could be attributed to firms with 20 or less employees. Amongst the large number of studies that
followed on from Birch’s work, some have provided confirmation and others have challenged the
suggestion that smaller firms predominate in the process of net job creation (for a good review, see
Neumark, Wall and Zhang, 2008).
The debate has tended to focus on evidence that smaller firms create more jobs, but also destroy
more in any given time interval, when compared to larger firms and thus the net job creation picture
hides much more volatile gross worker flows. For instance, in their study of UK manufacturing,
Barnes and Haskel (2002) suggest that during the 1980s, establishments with fewer than 100
employees accounted for approximately 41% of jobs destroyed and somewhere between 59 and
63% of jobs created3. This still details a net contribution and that is true of many of the studies
following on from Birch who continued to identify a disproportionate net contribution to
employment from small firms.
Recent studies by Neumark, Wall and Zhang, (2008) and Haltiwanger, Jarmin and Miranda, (2010)
provide a possible explanation for previous contradictory evidence and underline the importance of
new firms or start-ups in the process of job creation. The suggestion is that the smallest firms (of less
than 20 workers) have disproportionately high rates of job creation (relative to their total
employment share), but also have disproportionately large job destruction rates. Once one accounts
for firm age, any firm-size/job-creation relationship disappears. Furthermore, the importance of
firm-age in explaining job creation rates seems primarily driven by the higher rates of creation and
destruction amongst these youngest (i.e. new) business start-ups.
2 BIS Business Population Estimates for the UK and Regions, 2011.
3 These are midpoints of the actual figures that Barnes and Haskel arrive at using different methods of
calculation, and it should be remembered that, in line with many other studies, the focus of their analysis is
manufacturing establishments.
5
These new studies imply that previous findings on the relationship between firm size and
employment generation are driven by firm age and, more accurately, the important role of new
entrants in net employment generation. However, whilst useful, these findings are relatively
descriptive and provide very limited information on the individuals starting and running these firms;
and similarly limited information on the types of jobs being created and the employees who secure
them.
In all of this discussion we are comparing the employment opportunities created in large (usually
older) firms and small (usually younger) firms as if we were comparing like-with-like. In some ways
this reflects the paucity of firm-level data that allows us to do anything else, but it often reflects a
view that small firms and large firms are carrying out the same (job-creation) role in the economy,
but at different rates and on a different scale. One clear challenge to this view is presented in Urwin
(2011) where the writings of Schumpeter (1934; 1937/1989), Mises (1949) and Knight (1921) are
brought together to show why the creation of new, usually small, businesses is essential to the
entrepreneurial process and why large firms are unlikely to be able to nurture entrepreneurial talent
in the same way. Here we investigate the potential for large and small firms to be carrying out
different job-creation roles in the economy.
2. Data and Methods To analyse the role of small business in employment in the United Kingdom we employ the Quarterly
Labour Force Survey (LFS). This survey is one of the largest repeated household surveys conducted in
the UK and includes information on household/individual characteristics, economic activity,
education, employment, self-employment and firm size & type. The LFS is conducted quarterly and
sample members, once included, are followed for a period of 5 quarters, after which they are
refreshed with new sample members. In every quarter, approximately 1/5th
of sample members thus
exit the survey and are replaced by new individuals. The large size of the LFS makes this dataset
particularly attractive for the analysis of small business’ as every quarter approximately 100,000
people are interviewed, which when weighted to population level, provides a representative picture
of employment changes across the entire UK (60m people).
We select data from 10 quarterly surveys in the late-1990s, mid-2000s and early-2010s and apply
the following selection criteria; we remove anyone under the age of 16 or above the age of 65 (at
any point during the 5 interview waves) and in addition remove all those for whom we don’t have
detailed economic activity information (approximately 2.5% of the remaining sample). To increase
the sample size we pool individuals from ten quarters and create a rolling panel which then traces
various LFS individuals throughout their survey time:
6
Table 1: Exemplar Labour Force Survey data selection for mid 2000s cohortLFS
2005 Q3LFS
2006 Q1LFS
2006 Q2LFS
2006 Q3LFS
2006 Q4LFS
2007 Q1LFS
2007 Q2LFS
2007 Q3LFS
2007 Q4LFS
2008 Q1LFS
2008 Q2LFS
2008 Q3LFS
2008 Q4LFS 2005 Q3 x x x x xLFS 2006 Q1 x x x x xLFS 2006 Q2 x x x x xLFS 2006 Q3 x x x x xLFS 2006 Q4 x x x x xLFS 2007 Q1 x x x x xLFS 2007 Q2 x x x x xLFS 2007 Q3 x x x x xLFS 2007 Q4 x x x x x
This pooling occurs for three different time periods allowing for comparison of employment and self-
employment by firm size across time.4 Descriptive statistics for all three datasets can be found in the
following Table 25:
Table 2: Descriptive statistics of pooled LFS panel data
1990s 2000s 2010s
Number observations 562,425 495,165 450,009
Unique individuals 112,485 99,033 85,743
women 48.53 51.52 51.41
age 38.52 42.27 42.60
nonwhite 0.05 0.08 0.09
degree 0.13 0.28 0.30
child 0.48 0.44 0.44
married 0.57 0.66 0.66
Employment Status
unempld 4.93 3.52 5.26
inactive 21.68 23.31 23.47
empld 1to10 10.34 10.18 10.00
empld 11to49 12.95 13.29 12.61
empld 50+ [combined] 11.58 10.89
empld 250+ 24.18 9.57 8.75
sempl no/e 6.48 7.16 7.66
sempl with/e 2.25 2.08 1.81
public sector 17.19 19.32 19.53
Table 2 suggests that the samples under consideration are evolving slowly over time with more
women responding in the LFS, age rising and the proportion of individuals with a degree increasing
over time. However, examining the distribution of employment activity we see that proportions in
each of the categories has remained relatively stable over time; approximately 10% of individuals are
employed by firms with employee size 1 to 10, a further 13% employed by firms with 11 to 49
employees whilst 20% are employed by firms in the 50+ employee range.
4 A technical disadvantage of these three time periods are that the creation of a rolling panel is complicated by
the changing nature of personal identifier codes in the LFS. Data for the period covering the 2010s had to
come from Special License issues, whilst data from mid 2000s was freely available. Moreover, data for the
1990s had to come from the panel data element of the LFS (the five-quarter longitudinal LFS) which selected
only individuals who responded in all 5 quarters. To allow for cross-comparability across all 3 time periods we
thus created a balanced rolling panel to comply with lowest common denominator – in this case data from the
1990s. 5 LFS weights are used throughout this study
7
The panel data nature of these LFS datasets allows us to produce intra-person correlation/transition
matrixes of employment status. However, in determining whether there are active differences in
transition behaviour by various groups (such as women, the disadvantaged or young) we refrain
from reporting higher order contingency matrixes as it can be difficult to maintain an overview. We
thus resort to regression based modelling which predicts the transition outcome, y, of a specific
individual, i, in time period t, , with respect to various individual characteristics, ix . This type of
analysis allows us to infer whether specific groups are more or less likely to transit into particular
firms. Of particular interest are the group of people who are classified as unemployed or inactive in
time period t. We select this particular sub-group of individuals and model the destination outcome
for all those who transit into some form of employment (or self-employment) 4 quarters later.
However, because outcomes, ity , are in this case measured categorically (i.e. employed firm 1 to 10;
employed firm 11 to 49; employed firm 50 to 249 and employed firm 250+) and we cannot attribute
these categories into a specific rank, we resort to estimating a multinomial logit model:
ijt ijt ijtU β ε′= +x (1)
Where ijx represents a matrix of explanatory variables including age, ethnicity, disability,
unemployment length, children, marital status, full/part-time work, industry and occupation codes.
We assume that an individual will choose the alternatives that maximises their utility and as
McFadden (1973) has shown that if we assume that all ijtε of the j choices are independent and
identically distributed with the Weibull distribution in the form of ( ) exp[ exp( )]f ε ε ε= − − − , then
the resulting model is a multinomial logit model. Then when there are j choices, the probability that
an individual chooses a labour market outcome j is:
(2)
Estimation is carried out iteratively using maximum likelihood. Defining a set of j dummy variables:
��� � 1 if �� � � and 0 otherwise, results in one and only one ��� � 1 for each observation. The log-
likelihood is a generalisation of the binomial logit model and is given by:
(3)
Finally, to ease interpretability, we report exponentiated coefficients giving us an interpretation in
the form of odds-ratio’s.
2
exp( )Pr( | ) , for 1
1 exp( )
ij ji i ij J
ij jj
y j P jβ
β=
′= = = >
′+∑
xx
x
ij
n
i
J
jij PdL loglog
1 1∑∑
= =
=
8
3. Where do the unemployed and inactive get jobs?
Ideally, following the direction of recent research, we would consider both firm size and age, but this
is not possible in the Labour Force Survey– we therefore focus on firm size as our main variable of
interest and a good proxy for age of firm. From Table 3 we begin to get some idea of how much
more important small firms are as a route to employment for groups within the labour market who
find it harder to secure jobs.
Tables 3 and 4 are made up of two sections, both containing the same underlying frequencies
(numbers of individuals) but the first has percentages summing to 100% across the rows and the
second has percentages summing to 100% down each column. This allows us to gain two useful
perspectives on the same data. For instance, the first row of the first section of Table 3 sets out the
proportions moving from unemployment [in the first quarter] to various labour market states in the
final quarter that we observe them. Reading along this first row [where we have 3,477 observations
in total], we can see that 40.65% of those in employment in the first quarter are also in employment
by the final quarter and 20.82% have moved into inactivity. Therefore, just over 60% of the
individuals we observe in unemployment at the start of the year, are either inactive or unemployed
by the end of the year.
However, reading further along this first row, we can see just how important jobs in small businesses
are as a pathway to employment. Taken together, micro-enterprises (7.81%), small businesses
employing 11-49 (9.25%) and medium-sized enterprises with 50 to 249 employees (6.68%), are
employing just under 24% of those who were unemployed in the first period. If we consider this as a
percentage of all those who move from unemployment to employment in private sector enterprises,
these three categories of small firm account for 71% of all movements. If we recognise that self-
employment is essentially the starting-up of a small business, then either starting up a small business
or becoming an employee in a small firm accounts for 88% of all movements from unemployment
into private sector employment.
Reading along the second row, the suggestion is that employment in small firms and self-
employment are just as important as routes to a job for the 18,735 who are observed as inactive in
our first period of analysis. We can get some idea of how hard it is for these individuals to make the
transition to employment, as the first two cells suggest that 89% either remain inactive or move to
unemployment over our period of analysis. However, of the inactive who do move into private
sector employment, 80% do so either by starting up a small business or becoming an employee in a
small firm.
We have identified the importance of working in small firms and self-employment as a pathway to
employment for the unemployed and inactive – the vast majority of both groups who manage to
find a job, do so using these routes. However, it is quite possible that this is simply a result of the
higher rates of job creation that we tend to observe amongst smaller firms. It could be that all firms
employ the same proportion of unemployed or inactive, but that smaller firms create more jobs in
any given time period6 and therefore a higher proportion of the unemployed and inactive find work
through these routes.
6 The issue of correspondingly higher job destruction rates is one that we investigate in detail later in the
report.
Table 3: Rates of transition between labour market
: Rates of transition between labour market states: recessionary period (2008 Quarter 4 to 2011 Quarter 1)
9
ecessionary period (2008 Quarter 4 to 2011 Quarter 1)
10
The second section of Table 3 suggests that this is not that case and that small firms are more likely
to employ the unemployed and inactive. The second section of Table 4 adds further confirmation,
but also raises a particularly interesting question that we return to later in the report.
Reading along the first row of the second section of Table 3, we have column percentages for all but
the principle diagonal7, so the first figure we encounter is 46.66%. As with the first section of the
table, underpinning this figure are the number of individuals transiting from unemployment [in the
first quarter] to inactivity [in the final quarter], but here it is represented as a proportion of all those
who we observe moving to inactivity from another state. Put another way, out of the 2,708
individuals who we observe moving into inactivity from another labour market state, 46.66% come
from the unemployed.
Moving along this first row, we find that 22.23% of the 2,305 individuals who make the transition to
our smallest category of firm [1 to 10 employees] from another labour market state come from the
ranks of unemployed. The figures for our other two categories of small firm are 24.37% [11 to 49
employees] and 21.12% [50 to 249 employees]. When we compare these proportions to the figure
of 19.88% for our largest category of firm, the suggestion is that small firms are more likely to take
on the unemployed, as a proportion of the jobs that they create8. There is an apparent small firm
effect and it does not seem to be simply driven by the greater rates of job creation amongst small
firms. To give some idea of how these figures could have changed our findings, in the first section of
Table 3 we find that only 5.10% of the unemployed transit to the Public sector. However, when we
take some account the number of jobs created, we see that the public sector is particularly likely to
hire the unemployed as 27.75% of individuals flowing to this sector from another labour market
state were previously unemployed.
Given this last finding, our small firm effect may not seem particularly pronounced. However Table 4,
which sets out the same analysis for the period immediately prior to the start of recession, suggests
a much greater small business effect. For the cohorts which make up the analysis between Quarter 3
2005 and Quarter 4 2007, the first section of Table 2 suggests that either starting up a small business
or becoming an employee in a small firm accounts for 88% of all movements from unemployment
into private sector employment. This figure is identical to that for the pre-recessionary period, but as
the second half of Table 4 suggests, in this pre-recessionary period the proportion of all employees
who were previously unemployed is much higher in the smaller firms. In this pre-recessionary period
we find that 24.03% of the 3,103 individuals who make the transition to our smallest category of
firm [1 to 10 employees] from another labour market state come from the ranks of unemployed. The
figures for our other two categories of small firm are 27.74% [11 to 49 employees] and 26.61% [50 to
249 employees]. These figures are much higher than those for our largest category of firm [250+]
where we observe virtually identical proportions in the pre-recession (19.65%) and recessionary
periods (19.88%).
7 The numbers on the principle diagonal are not included in the denominator as these are predominantly
reflective of the stock of jobs within the particular firm-size category. There are individual in the main diagonal
who may be transiting between firms of the same size, but the majority are those who stay within the same
state from the first to the last period that we observe them. 8 Though one should note that the removal of the principle diagonal makes these estimated figures.
Table 4: Rates of transition between labour marke: Rates of transition between labour market states: pre-recessionary period (2005 Quarter 3 to 2007
11
2007 Quarter 4)
12
This is clearly an issue that we need to investigate in more detail, as many factors could be driving
these findings. However, at the very least we can say that evidence suggests that employment in
small firms provides the most important pathway to employment for the unemployed and inactive,
but the onset of recession has reduced the greater propensity for small firms to employ the
unemployed. Interestingly, this changing propensity of small firms to hire those who are outside of
employment does not seem as pronounced for the inactive9.
We now need to become more sophisticated in our analysis and the first issue we deal with in the
next section is to check that the findings in Tables 3and 4 still hold when we control for a variety of
other factors such as industry sector, occupation and region, amongst a number of others. This
allows us to check whether there is an ‘independent’ small firm effect, with our greater propensity
for small firms to hire the unemployed and inactive not simply being driven by a possible greater
concentration of small firms in certain sectors, more likely to employ those from certain
occupations, regions etc.
Before moving on to Section 3 where we consider the transitions to employment of the unemployed
and inactive in more detail, we briefly consider the other elements of Tables 3 and 4 that require
attention.
The cells outlined in green are the flip side of our analysis of transitions to employment in firms of
different sizes, from unemployment and inactivity. For instance in Table 3 [first section], the figure of
2.15% in the first column of cells outlined in green is the proportion of all individuals observed as
micro-business employees in the first quarter who are in unemployment by the last quarter.
Generally the proportion of employees who are subsequently observed in unemployment is lowest
in our largest category of firms [1.36%]. Whilst employment in small businesses is the dominate
pathway to employment for the unemployed, employees in small firms are also more likely to
subsequently experience a spell of unemployment if they are employed in a small firm. This is
consistent with the findings highlighted in the introduction, that small firms have higher rates of job
‘churn’. The second part of the next section specifically models this process to see what lessons
there are for policy.
4. Multivariate analysis
In this section we dig down into the findings outline in red, blue and green from Tables 3 and 4,
using multivariate regression techniques. So that we are considering the most up-to-date
information, the following results are gained from regression equations estimated for the 10 waves
of individuals, the first of whom we observe for five quarters from 2008 Quarter 4 and the last of
whom we observe for 5 quarters from 20011 Quarter 1. Later in this section, we compare these
findings to estimates gained from the same regression equations estimated for a period prior to the
present downturn [10 waves with start dates between 2005 Quarter 3 and 2007 Quarter 4] and
another period prior to the introduction of a weight of Employment Protection Legislation
9 This is not inconsistent with the finding (see for instance, Bell and Smith, 2002) that flows between
employment and inactivity tend to be pro-cyclical, whilst flows between unemployment and employment,
tend to be counter-cyclical.
13
(beginning with the 2002 Employment Relations Act)10
[10 waves with start dates between 1997
Quarter 1 and 1999 Quarter 2].
4.1 Modelling transitions from unemployment/inactivity to employment
Appendix Table 1 sets out the results of a regression equation which tests for the existence of a
small firm effect, having controlled for a variety of additional factors. Our estimated equation tells us
that, having controlled for a variety of factors11
that could be driving our findings in Tables 3 and 4,
we still identify a statistically significant relationship between being unemployed or inactive, and
subsequent employment in smaller firms. For instance, we find that the unemployed who make a
transition to employment are 30% less likely to do so by securing employment in our largest
category of firm [with 250 or more employees], relative to our category of micro-businesses12
. Even
more striking is the fact that, having controlled for a variety of other factors, the inactive who make
a transition to employment are 65% less likely to do so by securing employment in our largest
category of firm, relative to micro-businesses.
It is particularly interesting to note that our regression results uncover a much stronger and more
systematic small firm effect than is suggested in Tables 3 and 4. Looking at Tables 3 and 4, there is a
suggestion that enterprises with less than 250 employees hire a disproportionate proportion of the
unemployed and inactive. However, the effect does not follow a clear pattern; i.e. we do not see the
smallest [micro] firms being the most likely, then the 11-49 category the next most likely to hire
these groups, etc. and ending with the 250+ firms being the least likely. In fact from Tables 3 and 4
there is some suggestion that the group of small firms with 11-49 employees are doing the most to
employ the unemployed and inactive.
What our regression results suggest is that when we control for a variety of additional factors, we
uncover a much clearer and more systematic small firm effect. We have already considered our
finding that the unemployed are 30% less likely to secure employment in firms with 250+
employees, relative to the smallest micro-businesses. Considering firms with between 50 and 249
employees, relative to micro-businesses, we find that they do better than the 250+ category, but not
as well as micro-businesses. More specifically, the unemployed who make a transition to
employment are 22% less likely to do so by securing employment in firms with 50-249 employees,
relative to micro-businesses. Considering that the figure for the inactive is 31.5%, the suggestion is
that the unemployed and inactive are significantly more likely to obtain employment in firms with
50-249 employees [relative to firms with 250+ employees], but still significantly less likely than
through employment in micro-businesses.
10
In the previous recession, unemployment was rising from mid-1990 and peaked around Q3 1993; it then fell
to pre-recession levels in 1997, but continued to fall until 2003. The dates for our earlier period of analysis
have been chosen so that we are considering a period of relative stability, after the recessionary period of the
early to mid 1990s. 11
We estimate a multinomial logistic regression equation (and cluster standard errors) for all individuals who
are observed transiting into one of our categories of firm size, from any category of activity or inactivity. The
different firm size groupings are therefore our dependent categories and we attempt to model transitions into
these different firm-size categories using a range of covariates. We include age, sex, ethnicity (though we are
only able to include a white/non-white distinction), disability status, martial status, education (though only
degree/not degree), dependents, full-time/part-time working, occupation, industry sector, region and [the
variables of interest] whether the individual is employed, unemployed or inactive. 12
All results cited here are significant at the 0.01% level [99.9 per cent confidence], unless otherwise stated
and the reference category for the unemployed and inactive is the employed.
14
Finally, we see an almost insignificant difference between the probability that unemployed
individuals secure a job in firms with 11-49 employees, relative to micro-businesses. There is a
suggestion that the unemployed are 16% less likely to transit through firms with 11-49 employees
than through micro-businesses, but this effect is not as strong as those seen in other parts of our
regression, as it is only significant at the 5% level.
It is important to understand what insight these findings allow. First, we have very strong evidence
that the smaller a firm, the more likely we are to see the unemployed and inactive securing
employment opportunities. Not only does this effect seem not to be driven by other factors, when
we do control for industry, occupation, region etc. we uncover evidence of a more systematic small
firm effect. What this regression equation does not tell us is ‘why’ this small firm effect is happening.
Are our findings driven by the behaviours of the unemployed/inactive, is it a question of the nature
of small firms, or do the two combine in some way to drive our results?13
In the remainder of this
report we attempt to shed some lights on the possible reasons for this.
As a first step we now ask the question of which of the unemployed and inactive are most likely to
move into these smaller firms? To do this, we focus our modelling attention solely on the
unemployed and inactive who transit to employment. In Appendix Tables 2a to 2c we present the
results of a multinomial logistic regression that identifies particular groups within the unemployed
and inactive who are more likely to secure employment in firms of different sizes14
.
Appendix Table 2c presents the section of our multinomial logistic regression results where
comparison is made between the characteristics of unemployed and inactive individuals moving into
our largest category of firms (250+), relative to our smallest category of micro-businesses. Readers
are invited to study the results themselves, as there are a large number of findings and those that
are insignificant are sometimes as interesting as those that are significant. Teasing out findings of
particular interest, we find that,
Comparing the characteristics of the unemployed and inactive who secure employment in firms
with 250+ employees, relative to those who move into work through the route of micro-
businesses:
• The unemployed and inactive without a degree are significantly less likely to secure employment
in a firm of 250+ employees, relative to micro-businesses (a finding which is significant at the 5%
level).
13
More technically, these regression equations are in no way intended to identify causal relationships. Rather,
we are attempting to identify more reliable correlations, to ensure that our results are not being driven by
obvious omitted variables correlated with our explanatory factors. Our results suggest that we are not simply
observing small firms being concentrated in particular industry sectors; employing more individuals from
certain types of occupation and/or regions. If this was the case, then size of firm would simply be acting as a
proxy for these factors, with theoretical and policy implications that would be very different. 14
This multinomial logistic regression equation (with cluster standard errors) is for all unemployed and inactive
individuals who are observed transiting into one of our categories of firm size. The different firm size groupings
are again our dependent categories and we attempt to model transitions into these different firm-size
categories using a similar range of covariates to those described in the previous footnote. The main difference
is that, whilst in the previous specification we include indicators to reflect whether the individual is
unemployed, employed or inactive; we now have only unemployed and inactive individuals in our regression
and so we include indicators for the duration of unemployment and inactivity.
15
• We find that the inactive are particularly unlikely to secure employment in firms with 250+
employees. Relative to those who have been unemployed for one year or less, the inactive are
only 53% as likely to be observed moving into employment in our largest category of firm,
relative to micro-businesses (a finding with a 99.9% confidence level). However, we do not find
that those who have been unemployed for a year or more are any more or less likely to move
into employment in this largest category of firm, relative to those who have been unemployed
for less than a year.
• It is interesting from a policy perceptive that any effect of gender disappears when we introduce
an indicator that controls for the difference between part-time and full-time working. Initially we
observe women being significantly less likely to move into firms with 250+ employees, but this
effect is seemingly driven by the fact that we are only 24.4% as likely to see transitions into part-
time working within our largest category of firm, relative to micro-businesses (99.9%
confidence).
• Considering our indicators of industry sector, we are significantly less likely to see the
unemployed whose occupation is in the skilled trades securing employment opportunities in the
largest firms; relative to micro-businesses, these workers are only 21% as likely to move into
firms with 250+ employees (99% confidence). Similarly, those who consider their profession to
be within the category of Personal Services are only 14% as likely to secure jobs in the largest
firms, relative to micro-businesses (a finding that we might expect, given the ‘boutique’ nature
of employment in this area).
General findings on the characteristics of the unemployed and inactive who secure employment in
firms with either 50 to 249 employees or 11 to 49 employees (Appendix Tables 2b and 2a
respectively), relative to those who move into work through the route of micro-businesses:
• Drawing together the remaining results, we find that the part-time working effect is consistently
significant in all of our regression equations and describes a clear pattern. As already described,
we are only 24.4% as likely to observe the unemployed and inactive securing part-time work in
our largest category of firm, but this figure slowly rises to 56% amongst our firms with 50 to 249
employees and is 68% for our category of firms with 11 to 49 employees. The smaller the firm,
the more likely we are to observe the unemployed and inactive securing part-time employment.
• However, for firms with either 11 to 49 or 50 to 249 employees, we still retain a statistically
significant impact of gender, with unemployed and inactive women only 75% and 73% as likely
to be observed securing employment, respectively. These findings are significant at the 5% level
and suggest that, even controlling for a range of factors including PT/FT working distinctions,
unemployed and inactive women who move to employment are significantly more likely to do so
by securing employment in micro-businesses.
• It is particularly interesting from a policy perspective, to consider the results of our analysis of
unemployed and inactive individuals aged 16-24. As suggested above, we are no more or less
likely to observe this age group moving from unemployment/inactivity to employment in our
largest category of firm, relative to our smallest. However, as we move down the firm size-
bands, it becomes apparent that unemployed and inactive individuals aged 16 to 24 who make
the transition to employment are significantly more likely to secure employment in firms with
either 11 to 49 or 50 to 249 employees. Relative to our category of micro-businesses, we are
16
57% and 68% more likely to see this youngest of age groups working in firms with 50-249 and
11-49 employees respectively.
• We are significantly more likely to see the inactive who move to employment, securing their jobs
in either micro-businesses (as already suggested) or firms with 11-49 employees. Thus we have a
sliding scale that suggests the inactive who move into employment are only 53% as likely to
secure jobs in firms of 250+ employees relative to micro-businesses; this figure rises to 69% for
businesses with 50 to 249 employees and then we find no difference between micro-businesses
and those with 11-49 employees. If we observe an inactive individual moving into employment,
it significantly more likely that this will be through employment in a firm with between 1 and 49
employees.
• One finding that is of particular interest is that, out of all categories of firm size, we are
significantly less likely to observe unemployed and inactive non-white employees securing
employment in firms with 11-49 employees. We are only 56% as likely to observe non-white
unemployed and inactive individuals securing employment in firms of this size, relative to micro-
businesses; whilst for all other size of firm we find no significant difference.
As far as we are aware, this is the first time that any light has been shed on the importance of small
businesses as a route to employment for the unemployed and inactive. In contrast to the debates
that rage over the relative importance of small and large businesses in the process of job creation
and destruction, we have results that better reflect the nuance of this debate. Smaller firms are
carrying out a different role to large firms, as they are much more likely to be acting as a bridge to
employment for those who are on the margins of the labour market.
We may consider that this rather positive finding is perhaps tempered by the suggestion that a
larger proportion of this employment is likely to be part-time. However, there are a number of
additional controls that we add to our base regression equation that shed light on this issue and also
allow us to pursue other avenues of investigation. More specifically,
• Within the Labour Force Survey there are questions that ask individuals whether they are
looking for part-time work, full time work or they have no preference. We find that significantly
larger proportions of those who secure jobs in micro-businesses are looking for part-time work,
when compared to our largest category of firm. Employees who secure jobs in firms of 250+
employees are only 41% as likely to report that they are looking for a part-time, as opposed to a
full-time, job, when compared to those in micro-businesses. There may still be individuals who
take up part-time employment in micro-businesses when they would have preferred a full-time
job, but there are significant numbers for whom a part-time post is their preferred option.
• We also investigate the possible drivers of such a significant impact of inactivity in our regression
equation - are certain categories of inactive individual significantly more likely to be moving into
employment in small firms? Adding these controls to our regression equation begins to push our
‘flows’ data to its limit, and perhaps as a result of our reduced degrees of freedom we find little
that is significant in this approach. However, in Appendix Table 3 we set out the results of a
multivariate regression equation run on the ‘stock’ of employees in the Spring quarter of the
2009 Labour Force Survey15
. This suggests that, relative to those who 12 months ago were in
15
For all 49,593 individuals observed as private sector employees in the Spring 2009 Labour Force survey, we
investigate those factors that are most significantly associated with being an employee in firms of different
17
waged employment, the (i) unemployed, (ii) long-term sick or disabled, those (iii) previously
looking after the family or a home, (iv) retired or (v) a student are significantly more likely to be
employed in smaller firms.
• Whilst we cannot fully distinguish whether the behaviour of the unemployed/inactive as
opposed to small businesses is driving our findings, we can look at indicators such as job search
to see if the unemployed using certain (perhaps less formal) types of job-search are more likely
to be employed in small firms. Our initial findings suggest no significant differences in the job-
search behaviour of unemployed and inactive individuals who we see transiting to employment
in small, as opposed to large, firms.
• The extent to which individuals receive training from firms of different sizes, may be considered
as a reflection of the quality of jobs and will also impact the ability of previously
inactive/unemployed individuals to retain their position in employment. We include an indicator
in our regression of whether an individual (who has made the transition to employment) reports
receiving on-the-job training in the last four weeks. We find no significant difference between
the levels of training taking place in firms of different sizes, in a framework which controls for a
variety of other factors.
• One issue that we are particularly keen to investigate, and which links to our next set of outputs,
is the potential for the unemployed to be securing jobs in smaller firms which are manifestly less
secure, as they are possibly under temporary contracts. This is an issue to which we return in
detail in our policy discussions, but here it is particularly interesting to note the significantly
higher probability that jobs will be offered under temporary contracts in larger private sector
firms. More specifically, we find that the unemployed and inactive moving into employment in
firms with 250+ employees are two-and-a half times more likely to do so under temporary
contract, when compared to those moving into micro-businesses. This is a highly statistically
significant effect and one that is also substantial in its size16
. We find a reduced (but still
statistically significant) difference of just over one-and-a-half (1.7) times between the proportion
of jobs secured by the unemployed and inactive in firms of 50 to 249 employees, relative to
those in micro-businesses. The suggestion is that the unemployed and inactive moving into
employment in firms with less than 50 employees are significantly less likely to be doing so
under temporary contracts.
4.2 Modelling transitions from employment to unemployment/inactivity
In this section of the report we shed light on the reasons why individuals in small businesses seem
more likely to move into unemployment and inactivity. The previous section of our report shows
how important small firms are as a route to employment, but as we might expect, these routes are
potentially more risky for both employer and employee. As suggested in our introduction, small
firms have higher rates of creation and destruction, so the average job is likely to be less secure than
that in a larger firm. In addition the unemployed and inactive, by definition, have less favourable
recent employment histories, references or other evidence of the currency of their skills. Both sides
take on greater risk when entering into these employment relationships.
sizes. This is achieved using an ordered Probit model which has as its dependent variable, firms of different
sizes modelled as an ordered scale (i.e. the largest firms are given a value of [7], the next largest a [6] and so
on until we reach the micro-businesses who are coded as [1]). 16
One can have statistically significant results that are not particularly large.
18
Given that the unemployed and inactive seem much less able to secure jobs in larger firms, it is
important that we consider the evidence on exits from employment in small firms to see how this
risk manifests in the data. Only then can we learn how best to ensure that the jobs in small firms
that the unemployed and inactive are able to access, lead them to periods of employment that can
be sustained (ideally, well beyond the length of any initial contract).
To begin we estimate a logistic regression model to determine whether the small business effect we
identify in Tables 3 and 4 remains when we control for a variety of additional factors. The controls
we include in this regression are almost identical to the specification described in footnote 11, but
here we estimate the equation for all individuals who are employed in private-sector firms of
different sizes in the first period of our analysis; and our dependent categories are the destination
states of (i) unemployment and (ii) inactivity. The results of this analysis suggest that our small firm
effect remains even when we control for sector, occupation, region and a variety of other
firm/individual characteristics.
More specifically, we find that employees working in micro-businesses in the first quarter are over
six times more likely (6.752) than employees in firms with 250+ employees to be observed in
unemployment four quarters later. This is a finding that is highly significant at the 0.1% level (99.9%
confidence) and which is very similar to the figure of 6.554 that we obtain when modeling the
probability of employment to inactivity moves, in micro-businesses relative to firms with 250+
employees.
If we then move on to consider the other categories of small firm in our study, this ratio is smaller,
but still significant for our next largest firms. We find that employees in firms with 11-49 employees
are one-and-a-half times more likely (1.456) to be observed in unemployment by the final quarter
and only just under 30% more likely to be observed in inactivity (an odds ratio of 1.279) relative to
their peers in our largest category of firm. These are statistically significant results at the 1% and 5%
level respectively. We find no significant difference in the probability that employees in firms with 50
to 249 employees will transit to unemployment or inactivity, relative to those in firms with 250+
employees.
If we consider these findings next to those uncovered from our initial analysis of transitions to
employment, there seems also to be some suggestion that, when considering flows from
unemployment and inactivity into employment, our small business effect is particularly focused
amongst micro-businesses and those with 11-49 employees. That is, we find these two categories of
SME have a higher propensity to employ the unemployed and inactive, even after controlling for a
variety of additional characteristics. Similarly, it would seem that employees in micro-businesses and
those with 11-49 employees have significantly higher probabilities of flowing into unemployment
and inactivity.
Our original suggestion that employees in small firms are more likely to transit to unemployment
and inactivity is confirmed. We now consider the types of worker that are more or less likely to
transit from employment to unemployment/inactivity from different sized firms. More specifically,
we attempt to draw some distinction between two competing scenarios:
Does our small firm effect arise because we find different types of worker in firms of different sizes?
For instance, it is generally accepted that those with fewer formal qualifications are more at risk of
19
unemployment at any one time, no matter what type of firm they work in. If small firms have more
employees with lower levels of qualification, then we may expect higher rates of exit into
unemployment amongst small firm employees partly because of the nature of the workforce. In
contrast, we may find from our multivariate analysis that workers of all types (with many formal
qualifications, with few formal qualifications, etc.) have higher probabilities of entering
unemployment and inactivity in small firms, compared to large firms, simply because small firms
have lower survival rates17
. In this second scenario, workers of all types are more at risk in smaller
firms.
Distinguishing the extent to which higher flows into unemployment and inactivity are due to either
(i) the characteristics of employees as opposed to (ii) the size (or perhaps age) of the firm suggests
that the two are separate and distinct. Clearly there is a lot of overlap and it is hard to unequivocally
distinguish the two. However, this is an analysis driven by considerations of policy design. If we are
going to support employers and employees to strengthen one of the few routes to employment that
is heavily utilised by the unemployed and inactive, it is useful to distinguish which of these effects
predominates.
To arrive at some answers to this question, we estimate separate regression equations for each one
of our categories of firm size (Appendix Table 4). The suggestion is that:
• Employees aged 16-24 in firms of all sizes seem more likely (than those aged between 25 and 49)
to experience a move to unemployment or inactivity over the period of our analysis. For this
group some of these moves will possibly be back into education, but for the majority they are
likely to be less desirable destination states. It is interesting that the relative probability of the
16-24 age group moving into unemployment/inactivity, compared to the 25-49 age group, falls
steadily as we move from the largest firm-size category, where it is 1.955, to micro-businesses
where it is 1.470 (and becomes statistically insignificant). The suggestion is that this youngest of
age groups are not [statistically] significantly more likely to move to inactivity/unemployment in
this smallest of business types, compared to their experience in larger firms. This is an issue that
we could perhaps delve down into in more detail, to see whether this finding is driven by moves
into unemployment or particular forms of inactivity.
• Similarly, when we consider employees aged 50 to 65, they seem more likely to transit to
unemployment and inactivity in the larger firms within our sample, relative to those aged 25-49.
In firms with between 50 and 249 employees and those with 250+, they are nearly two-and-a
half times (2.467 and 2.396 respectively) more likely to move into unemployment/inactivity than
those aged 25 to 49; whilst in smaller firms the effect is less [statistically] significant and closer
to one-and-a half times more likely (1.629 and 1.279 in micro-businesses and firms with 11-49
employees respectively). Particularly interesting is the fact that the difference between our
largest and smallest parameter estimate is statistically significant18
and this suggests that the
employees in the 50+ age group are [statistically] significantly more ‘at risk’ in larger firms.
17
In all of this we are unable to distinguish between voluntary and involuntary terminations of the
employment relationship, but shedding some light on the distinction above goes some way to compensate for
this. 18
The ratio of 1.279 for employees aged 50 to 65 in firms with 11-49 employees has an approximate 95%
confidence interval that reaches a value of approximately 1.6 at its upper bound. The ratio of 2.467 for
employees aged 50 to 65 in firms with 50 to 249 employees has an approximate 95% confidence interval that
20
• We find that non-white employees in micro-businesses are significantly more likely to move into
unemployment/inactivity, when compared to their white peers – a finding that is not apparent
in any of our other categories of business. Our use of a categorisation that only distinguishes
ethnicity according to a white/non-white distinction is far from ideal, and primarily driven by the
relatively small numbers of observations we would be left with if we were to distinguish
between, for instance, individuals of Indian and Bangladeshi backgrounds. It is likely that the
significance of this coefficient is driven by differences in the ethnic minority groups that we
observe within micro-businesses, as opposed to larger businesses (see Appendix Table 3); but
again this would require further investigation.
• Those who are employed on a part-time basis in our first period of analysis are significantly more
likely to experience unemployment or inactivity by the fourth quarter that we observe them,
when compared to their full-time colleagues; and this is a finding that is consistent across firms
of all sizes. Again we observe a steady pattern, with this effect being more pronounced the
smaller the size of business we consider. However, whilst the estimated ratio falls steadily from
1.586 for firms with 250+ employees to 2.165 for micro-businesses, there is no suggestion that
these are statistically significant differences.
• In our regression we include an indicator of whether the individual has received some form of
work-based training in the last 4 weeks, and across all but our 50-249 category of firm the
suggestion is that this group have a lower probability of being in unemployment/inactivity 4
quarters later. There is some indication that this impact is particularly pronounced for our
smallest category of firm, but as with other instances above, any difference is not statistically
significant. Similarly, we find that those on temporary contracts in firms of all sizes are much
more likely than their peers on permanent contracts to find themselves in unemployment or
inactivity four quarters later. This is an effect that seems more pronounced for larger firms, but
again we find no significant difference between our parameter estimates in firms of different
sizes.
It is important to understand what these findings are telling us. For instance, when we find that
those aged 50 to 65 are more likely to move into unemployment or inactivity in larger firms, this
does not mean that greater numbers of this age group flow from larger firms into
unemployment/inactivity. Rather it suggests that this group are simply more at risk, relative to their
peers aged between 25 and 49, in larger firms. As suggested in the introduction to this section, we
are particularly interested in the exact nature of the larger flows into unemployment and inactivity
in smaller firms, as it helps us in the first steps of policy design.
Generally we find from the regression equations presented in Appendix Table 4 that there are few
statistically significant differences between the levels of risk faced by different types of employee in
firms of different sizes. This finding is more consistent with the suggestion that we see more small
firm employees moving into unemployment and inactivity because these smaller firms have a
greater number of employees who have characteristics that are traditionally seen as putting them at
‘risk’. However, this avenue of investigation could form an entire research project in itself, as we also
find that there are clear patterns in the estimated coefficients associated with certain characteristics
has a value of approximately 1.8 at its lower bound. We can be more than 95% confident that these two
estimates are significantly different from each other.
21
across firms of different sizes – it is quite possible that the lack of significance is driven by a
combination of small numbers and high variability/heterogeneity.
To be clear, if we were to find that employees with certain characteristics were more at risk of
unemployment and inactivity in smaller firms, it would suggest that the higher rates of small-firm
destruction was one of the main drivers19
. The suggestion here is that, at the very least, small firms
employment of a greater number of individuals with characteristics that put them at risk of
unemployment/inactivity is a significant factor in explaining the larger flows into unemployment and
inactivity from these firms. Small firms take on larger numbers of individuals with more ‘risky’
characteristics and this seems to manifests itself as higher flows into unemployment and inactivity.
4.3 Changes through time
As suggested in the introduction to our study, there is a greater propensity for small firms to hire the
unemployed and inactive, but this falls as we move into recession. In contrast, the [lower]
proportion of unemployed and inactive individuals securing jobs in the largest [250+] firms remains
relatively stable as we move into recession (see the second sections of Tables 3 and 4). However, any
differences are not particularly pronounced and here we present the results of a multivariate
analysis carried out on samples before and after 2008 to determine whether such headline
differences are confirmed when we control for a variety of additional factors. Before doing so, we
consider a much more pronounced difference which becomes apparent when we compare the
transition rates of Tables 3 and 4 with those from a period during the late 1990s.
Table 5 presents comparable figures to those in Tables 3 and 4 for 10 waves of individuals, the first
of whom we follow for five quarters from quarter 1 1997 and the last of whom we follow for five
quarters from 1999 Quarter 2. Unfortunately when we go back this far in the LFS data the question
asked on firm size allows us less differentiation and we lose the ability to distinguish between firms
with 50 to 249 employees and those with 250+ employees. Even with this loss of detail the findings
seem rather striking.
From the first section of Table 5 any difference is not immediately apparent, as the proportion of
unemployed individuals flowing into firms with 50+ employees seems similar to the equivalent
figures, of between 11% and 13%, implied by Tables 1 and 220
. However, when we consider the
second section of Table 3 it becomes apparent just how far the propensity of large firms to hire the
unemployed has fallen in the last 10 to 15 years. During the 1997 and 1999 period that we analyse,
nearly 34% of employees flowing into our largest category of firm (50+) were from the ranks of the
unemployed. This is in contrast to a comparable figure of between 20 and 23 per cent in the periods
we analyse after 2005.
This fall is evident in our mid-sized firm category, but it is nowhere near as pronounced as the fall in
larger firms. Across our three time periods, taken in chronological order, the proportion of
unemployed individuals flowing into firms with 11-49 employees moves very little from 27%, to 28%
and then in the recession it falls to 24%. In contrast, the fall amongst micro-businesses comes closer
19
This is a simplification of the situation, as it is quite possible that we have a number of omitted variables. As
a result we may have individuals with a certain characteristic who are actually very different (on
unobservables) in small and large firms. 20
Summing the proportions accounted for by firms with between 50 and 249 employees, and those with 250+
employees.
22
to the magnitude of fall seen amongst the 50+ businesses, with a chronology of 31% to 24% and
finally to 22% in recession.
Clearly there is a lot that could be driving these findings and in order to shed more light on the
possible reasons for these trends we need to carry out a set of multivariate analyses. However, it is
worth underlining that the choice of our 1997 to 1999 period for analysis is based on the fact that by
this point in time, the economy was well into recovery and unemployment was no longer falling at
the rate it had been for the previous years. We may still reasonably expect a higher proportion of
recruits in firms of all sizes to be taken from the unemployed during this period of recovery, but this
would not seem to explain the relative differences and the subsequent changes.
23
Table 5: Rates of transition between labour market states: pre-recessionary period (1997 Quarter 1 to 1999 Quarter 2)
Unemployed InactiveEmployee [1
to 10]
Employee 11
to 49
Employee
[50+]
self-emplyed [no
employees]
self-emplyed
[with
employees]
Public sector
Unemployed 35.11 21.71 9.46 9.42 13.33 4.23 0.27 6.47 100
Inactive 4.78 81.07 3.51 3.36 3.45 1.51 0.10 2.22 100
Employee [1 to 10] 2.01 5.43 70.39 10.55 6.66 2.01 0.55 2.40 100
Employee 11 to 49 2.25 4.12 7.66 71.50 10.43 1.21 0.25 2.58 100
Employee [50+] 1.75 3.58 2.20 5.79 83.51 0.81 0.10 2.25 100
self-emplyed [no
employees] 1.35 4.05 3.33 2.23 2.50 78.87 6.50 1.18 100
self-emplyed [with
employees] 0.64 1.79 3.33 1.79 0.97 9.23 81.88 0.37 100
Public sector 0.60 3.24 1.16 1.44 2.34 0.45 0.08 90.69 100
Total 3.56 21.24 10.14 12.91 24.04 6.76 2.58 18.77 100
Unemployed InactiveEmployee [1
to 10]
Employee 11
to 49
Employee
[50+]
self-emplyed [no
employees]
self-emplyed
[with
employees]
Public sector
Unemployed 49.43 30.86 27.24 33.59 21.75 3.44 37.03 203.35
Inactive 35.72 11.45 9.72 8.69 7.76 1.27 12.71 87.33
Employee [1 to 10] 15.02 12.36 30.51 16.78 10.33 7.01 13.74 105.76
Employee 11 to 49 16.82 9.38 24.99 26.29 6.22 3.18 14.77 101.65
Employee [50+] 13.08 8.15 7.18 16.74 4.16 1.27 12.88 63.47
self-emplyed [no
employees] 10.09 9.22 10.86 6.45 6.30 82.80 6.75 132.48
self-emplyed [with
employees] 4.78 4.08 10.86 5.18 2.44 47.46 2.12 76.92
Public sector 4.48 7.38 3.78 4.16 5.90 2.31 1.02 29.04
Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00
Status of individual in first
quarter
Status of Individual in final quarter
Status of individual in first
quarter
Status of Individual in final quarter
Weighted Row Percentages
Weighted Column Percentages (minus diagonal)
24
To shed some light on possible impacts that the recession has had, we now estimate a multinomial
regression model for ten waves of individuals, the first of whom we observe for five quarters from
2005 Quarter 3 and the last of whom we observe for 5 quarters from 2007 Quarter 4. The findings
from this analysis of a pre-recession sample are compared to the findings from the analyses
presented to this point, which have been estimated for 10 waves of individuals followed for five
quarters starting between 2008 Quarter 4 and 2011 Quarter 1.
Appendix Tables 5a to 5c set out the results from analysis of the pre-recession sample. The findings
seem to suggest that,
• In the period prior to the recession, women making the transition from unemployment/inactivity
to private sector employment were significantly less likely to do so by securing a job in our
largest category of firm (250+) relative to micro-businesses. The suggestion is that, even having
controlled for a variety of other factors, women were 32% less likely to be seen moving to these
sized firms, relative to micro-businesses, when compared to their male counterparts. This is an
effect that is not evident in the regressions we estimate for 2008-2011 when we include a
variety of controls.
• It is particularly striking that before the recession, those making the transition from inactivity to
employment were not significantly less likely to do so by securing a job in a firm with 250+
employees, compared to a micro-business. This is in stark contrast to the finding that in the
period after the start of recession, the inactive are only 53% as likely to make such a transition in
our largest category of firm21
.
• Similarly, before the recession we were no less likely to see those from the skilled trade
occupations making the transition from unemployment or inactivity to employment through
firms with 250+ employees. However, in the period after the recession, this occupational group
are only 21% as likely to make such a transition into firms with 250+ employees. A similar effect
seems to be apparent for the unemployed and inactive who usually work in the personal service
occupations.
21
The change from statistically insignificant to significant may be due to the larger numbers of inactive making
such transitions after the onset of recession. In this situation the estimated odds ratios would remain roughly
constant, but simply become more significant during recessions. However, here we do seem to observe quite a
substantial change in the estimated coefficient, and it would seem that any changes in estimated standard
errors are not wholly responsible.
25
5. Emerging Conclusions and Policy Discussion
There are numerous specific findings that will hopefully be of interest to policymakers, academics
and others involved in support of the unemployed and inactive. Generally we find that:
Movements from unemployment/inactivity
• Employment in small firms provides the most important pathway to employment for the
unemployed and inactive. When we control for a variety of additional factors in a multivariate
framework, we confirm a statistically significant small firm effect that has an apparently
systematic pattern.
• The suggestion is that this is not being driven by any one particular category of unemployed or
inactive individual. We find that those who were previously unemployed, long-term sick or
disabled, looking after the family/home, retired or a student are significantly more likely to be
employed in smaller firms.
• However, the onset of recession seems to have reduced this greater propensity for small firms
to employ the unemployed and inactive.
Movements to unemployment/inactivity
• We also find that employees in micro-businesses and those with 11-49 employees have
significantly higher probabilities of flowing into unemployment and inactivity; and this is also a
finding that remains significant in a multivariate framework.
• Does this small firm effect arise because we find different types of worker in firms of different
sizes, or simply because small firm employment is more risky (whatever the individual
employees characteristics)? Our findings are more consistent with the former suggestion. We
find some evidence that more small firm employees move into unemployment and inactivity
because these smaller firms have a greater number of employees who have characteristics that
are traditionally seen as putting them at ‘risk’. This is not to say that greater levels of firm
destruction do not account for some of the larger flows into unemployment and inactivity across
smaller businesses; but it would seem that small firms take on larger numbers of individuals with
‘risky’ characteristics and this manifests itself as higher flows into unemployment and inactivity.
• There seems to be a relatively consistent small business effect, but it is particularly pronounced
at the extremes of our firm-size distribution. When considering flows from unemployment and
inactivity into employment, our small business effect is particularly pronounced amongst micro-
businesses and to a lesser extent firms with 11-49 employees. Similarly, it would seem that
employees in micro-businesses have significantly higher probabilities of flowing into
unemployment and inactivity, when compared to the firms with 250+ employees. Firms with
250+ employees are surprisingly inaccessible to the unemployed and inactive; whilst micro-
businesses are surprisingly accessible – firms with sizes in-between these two extremes follow
the same pattern (and differences are often significant), but it is at the extremes of the firm-size
distribution that we see particularly pronounced firm-size effects.
• Finally, when comparing analyses of transitions before and after the onset of recession, there
would seem to be evidence that after the onset of recession, employment in small (particularly
micro-) businesses becomes relatively more important as a route to employment for the
unemployed and inactive. This is consistent with the suggestion that high rates of firm creation
continue into recession (partly due to the fact that the reduction in employee job opportunities
26
pushes people to start their own businesses) and this provides job opportunities for
disadvantaged groups, whilst the largest firms hoard labour and wind down their recruitment.
Irrespective of what drives the greater propensity for small firms to act as a pathway to employment
for the unemployed and inactive, it is reasonable to suggest that this new evidence justifies some re-
thinking of policy approaches. The findings would seem to be of particular interest to those working
in areas of government tasked with helping the unemployed and inactive into employment, and
those concerned with education and training.
For instance, from the evidence presented here, small firms employing the unemployed and inactive
seem no less likely than larger firms to be providing them with opportunities for training and
education. This suggests a fertile ground for those government programmes that attempt to raise
the skill levels of the employed. However, the message here is not simply about more education, it is
about the targeting of support, and recognition that the incentives for those tasked with the
operation of policy are often skewed in the wrong direction. Thus whilst there are economies of
scale from setting up programmes between government departments and large employers, small
employers are much more likely to be employing those who need this support. It would be useful to
identify very small firms in the SME sector who are engaged with programmes such as Skills for Life,
and see what lessons can be learnt from this engagement.
27
References:
Anyadike-Danes, M., K. Bonner, M. Hart and C. Mason (2009), ‘Measuring business growth: high-
growth firms and their contribution to employment in the UK’, Research report, National
Endowment for Science, Technology and the Arts (NESTA), October.
Barnes, M. and J. Haskel (2002), ‘Job creation, job destruction and the contribution of small
businesses: evidence forUK manufacturing’, Working Paper no. 461, Queen Mary Department of
Economics.
Bell, B. and Smith, J. (2002), “On gross worker flows in the United Kingdom: evidence from the
Labour Force Survey”, Bank of England Working Paper Series, ISSN 1368-5562.
Bell and Blanchflower (2009), “What Should Be Done about Rising Unemployment in the UK?”, IZA
Discussion Paper No. 4040
Birch, D. L. (1979), The Job Generation Process, Unpublished report prepared by the MIT Program on
Neighborhood and Regional Change for the Economic Development Administration, US Department
of Commerce, Washington, DC.
Birch, D. L. (1981), ‘Who creates jobs?’, The Public Interest, 65: pp 3–14.
Birch, D. L. (1987), Job Creation in America: How Our Smallest Companies Put the Most People to
Work, New York: Free Press.
Boden, R. (1996), ‘Gender and self-employment selection: an empirical assessment’, Journal of
Socio-Economics, 25(6): 671–82.
Brown, C., J. Hamilton and J. Medoff (1990), Employers Large and Small, Cambridge, MA: Harvard
University Press.
Campbell, M. and M. Daly (1992), ‘Self-employment: into the 1990s’, Employment Gazette.
Clark, K. and S. Drinkwater (2000), ‘Pushed in or pulled out? Self-employment among ethnic
minorities in England and Wales’, Labour Economics, Vol. 7: pp 603–28.
Garicano, L., LeLarge, C. and Van Reenen, J. (2012), "Firm Size Distortions and the Productivity
Distribution: Evidence from France", Centre for Economic Performance discussion paper No. 1128.
Haltiwanger, J., R. Jarmin and J. Miranda (2010), ‘Who creates jobs? Small vs. large vs. young’,
Discussion Paper 101910, Center for Economic Studies (CES), US Census Bureau, August.
Henley, A. (2007), ‘Entrepreneurial aspiration and transition into self-employment: evidence from
British longitudinal data’, Entrepreneurship and Regional Development, 19(3): 253–80.
Knight, F. H. (1921), Risk, Uncertainty and Profit, Boston, MA: Houghton Mifflin Co.
Marshall, A. (1949 [1920]), Principles of Economics, 8th edn, London: Macmillan.
Mises, L. V. (1949), Human Action, New Haven, CT: Yale University Press.
28
Neumark, D., B. Wall and J. Zhang (2008), ‘Do small businesses create more jobs? New evidence for
the United States from the National Establishment Time Series’, IZA Discussion Paper no. 3888,
Institute for the Study of Labor.
Parker, C. (2004), The Economics of Self-employment and Entrepreneurship, Cambridge: Cambridge
University Press.
Schumpeter, J. (1934), The Theory of Economic Development, Cambridge, MA: Harvard University
Press.
Schumpeter, J. (1937), Preface to the Japanese edn of ‘Theorie der Wirtschaftlichen Entwicklung’, in
J. Schumpeter (1989), Essays on Entrepreneurs, Innovations, Business Cycles and the Evolution of
Capitalism, ed. R. Clemence, New Brunswick, NJ: Transaction Publishers, pp. 165–8.
Schumpeter, J. (1989), Essays on Entrepreneurs, Innovations, Business Cycles and the Evolution of
Capitalism, ed. R. Clemence, New Brunswick, NJ: Transaction Publishers.
Urwin, P., V. Karuk, F. Buscha and B. Siara (2008), ‘Small businesses in the UK: new perspectives on
evidence and policy’, Commissioned by the Federation of Small Businesses.
Urwin, P. (2011), Self-employment, Small Firms and Enterprise, Institute of Economic Affairs, 179
pages: ISBN 9780255366106
29
Appendix
Appendix Figure 1
Source: Bell and Blanchflower (2009)
Appendix Table 2a: Multinomial Logistic Regression of Unemployed/Inactive to EmploymentMultinomial Logistic Regression of Unemployed/Inactive to Employment
30
Multinomial Logistic Regression of Unemployed/Inactive to Employment
Appendix Table 2b: Multinomial Logistic Regression of Unemployed/Inactive to EmploymentMultinomial Logistic Regression of Unemployed/Inactive to Employment
31
Multinomial Logistic Regression of Unemployed/Inactive to Employment
Appendix Table 2c: Multinomial Logistic Regression of Unemployed/Inactive to EmploymentMultinomial Logistic Regression of Unemployed/Inactive to Employment
32
Multinomial Logistic Regression of Unemployed/Inactive to Employment
33
34
35
36
Appendix Table 5a: Multinomial Logistic Regression of Unemployed/Inactive to Employment
2005 to 2007
(1) (2) (3) (4) (5)
Base +extra +occu&industr +region +weights
empld_11to49
16 to 24 Agegroup 1.476*** 1.449** 1.417* 1.417* 1.331*
(0.150) (0.197) (0.198) (0.198) (0.191)
25 to 49 Agegroup reference reference reference reference reference
reference reference reference reference reference
50 to 65 Agegroup 0.498*** 0.520*** 0.528*** 0.539*** 0.536***
(0.074) (0.087) (0.089) (0.092) (0.092)
Ethnicity: non-white 0.776 0.769 0.770 0.784 0.811
(0.115) (0.114) (0.116) (0.123) (0.130)
Sex: Female 0.653*** 0.738** 0.728** 0.731** 0.718**
(0.061) (0.072) (0.077) (0.078) (0.078)
Disabled: Yes 1.118 1.118 1.087 1.092 1.075
(0.147) (0.147) (0.143) (0.144) (0.144)
Degree: Yes 1.041 1.005 1.015 0.991 0.971
(0.140) (0.139) (0.150) (0.148) (0.149)
Inactive (4Q ago) 0.819 0.838 0.843 0.874
(0.092) (0.095) (0.096) (0.101)
Unemployed less than 1 year (4Q ago) reference reference reference reference
reference reference reference reference
Unemployed 1+ (4Q ago) 0.697 0.705 0.718 0.702
(0.174) (0.177) (0.180) (0.181)
Married: Yes 0.861 0.864 0.865 0.848
(0.114) (0.117) (0.118) (0.118)
Child: Yes 0.980 0.974 0.982 0.984
(0.108) (0.109) (0.110) (0.112)
Full-Time Job reference reference reference reference
reference reference reference reference
Part-time Job 0.721** 0.643*** 0.645*** 0.631***
(0.080) (0.079) (0.080) (0.081)
1 managers and senior officials reference reference reference
reference reference reference
2 professional occupations 1.501 1.501 1.474
(0.590) (0.597) (0.601)
3 associate professional and technical 1.408 1.434 1.392
(0.432) (0.440) (0.437)
4 administrative and secretarial 0.854 0.853 0.841
(0.230) (0.229) (0.232)
5 skilled trades occupations 0.981 0.980 0.876
(0.302) (0.302) (0.279)
6 personal service occupations 1.258 1.263 1.238
(0.346) (0.348) (0.350)
7 sales and customer service occupation 1.004 0.997 0.970
(0.264) (0.262) (0.263)
8 process 1.787 1.807 1.756
(0.600) (0.612) (0.611)
9 elementary occupations 1.287 1.275 1.233
Industry sector controls+
north east reference reference
reference reference
north west 1.262 1.401
(0.333) (0.375)
merseyside 1.434 1.647
(0.489) (0.575)
yorkshire & humberside 1.257 1.364
(0.329) (0.364)
east midlands 1.057 1.136
(0.276) (0.304)
west midlands 0.996 1.057
(0.251) (0.272)
eastern 1.225 1.308
(0.309) (0.337)
london 1.141 1.197
(0.301) (0.322)
south east 1.066 1.195
(0.258) (0.295)
south west 1.040 1.116
(0.271) (0.297)
wales 0.932 1.083
(0.263) (0.312)
scotland 1.439 1.671
(0.377) (0.447)
northern ireland 2.412 2.793*
(1.110) (1.335)
37
Appendix Table 5b: Multinomial Logistic Regression of Unemployed/Inactive to Employment
2005 to 2007
empld_50_
16 to 24 Agegroup 1.362** 1.544** 1.483* 1.470* 1.412*
(0.155) (0.229) (0.229) (0.228) (0.222)
25 to 49 Agegroup reference reference reference reference reference
reference reference reference reference reference
50 to 65 Agegroup 0.712* 0.764 0.760 0.766 0.780
(0.112) (0.133) (0.135) (0.137) (0.142)
Ethnicity: non-white 0.971 0.958 0.942 1.022 1.026
(0.154) (0.152) (0.153) (0.177) (0.179)
Sex: Female 0.575*** 0.708** 0.696** 0.689** 0.670**
(0.059) (0.077) (0.083) (0.082) (0.082)
Disabled: Yes 0.972 0.988 0.985 0.994 0.975
(0.145) (0.148) (0.150) (0.152) (0.152)
Degree: Yes 1.445** 1.343* 1.254 1.230 1.262
(0.202) (0.191) (0.200) (0.198) (0.207)
Inactive (4Q ago) 0.694** 0.733* 0.746* 0.762*
(0.084) (0.090) (0.092) (0.096)
Unemployed less than 1 year (4Q ago) reference reference reference reference
reference reference reference reference
Unemployed 1+ (4Q ago) 0.542* 0.516* 0.539* 0.514*
(0.149) (0.148) (0.154) (0.154)
Married: Yes 0.970 0.981 0.984 0.938
(0.141) (0.147) (0.149) (0.144)
Child: Yes 1.004 1.001 1.016 1.025
(0.121) (0.123) (0.126) (0.130)
Full-Time Job reference reference reference reference
reference reference reference reference
Part-time Job 0.562*** 0.488*** 0.487*** 0.485***
(0.067) (0.066) (0.066) (0.067)
1 managers and senior officials reference reference reference
reference reference reference
2 professional occupations 1.614 1.573 1.468
(0.680) (0.672) (0.635)
3 associate professional and technical 2.028* 2.090* 2.163*
(0.651) (0.673) (0.709)
4 administrative and secretarial 1.313 1.307 1.323
(0.379) (0.378) (0.392)
5 skilled trades occupations 0.658 0.659 0.601
(0.229) (0.229) (0.213)
6 personal service occupations 0.769 0.764 0.765
(0.251) (0.249) (0.254)
7 sales and customer service occupation 1.756 1.727 1.641
(0.509) (0.501) (0.488)
8 process 2.419* 2.411* 2.518*
(0.852) (0.856) (0.915)
9 elementary occupations 1.413 1.397 1.363
(0.397) (0.391) (0.391)
Industry sector controls+
north east reference reference
reference reference
north west 1.561 1.566
(0.453) (0.463)
merseyside 1.194 1.292
(0.478) (0.525)
yorkshire & humberside 1.182 1.141
(0.349) (0.345)
east midlands 0.916 0.902
(0.273) (0.275)
west midlands 1.028 1.034
(0.291) (0.300)
eastern 1.236 1.248
(0.350) (0.361)
london 0.861 0.873
(0.258) (0.267)
south east 1.056 1.095
(0.289) (0.306)
south west 0.870 0.873
(0.261) (0.267)
wales 0.767 0.806
(0.253) (0.270)
scotland 1.450 1.505
(0.426) (0.453)
northern ireland 2.502 2.711
(1.246) (1.384)
38
Appendix Table 5c: Multinomial Logistic Regression of Unemployed/Inactive to Employment
2005 to 2007
empld_250_
16 to 24 Agegroup 1.215 1.380 1.369 1.365 1.297
(0.168) (0.244) (0.257) (0.258) (0.251)
25 to 49 Agegroup reference reference reference reference reference
reference reference reference reference reference
50 to 65 Agegroup 0.535** 0.631* 0.650 0.651 0.623
(0.114) (0.146) (0.155) (0.157) (0.151)
Ethnicity: non-white 1.282 1.256 1.226 1.209 1.182
(0.232) (0.228) (0.225) (0.239) (0.241)
Sex: Female 0.504*** 0.666** 0.695* 0.705* 0.680**
(0.064) (0.091) (0.102) (0.104) (0.101)
Disabled: Yes 1.064 1.080 1.120 1.154 1.152
(0.195) (0.199) (0.209) (0.217) (0.221)
Degree: Yes 1.954*** 1.772*** 1.406 1.342 1.250
(0.318) (0.288) (0.261) (0.251) (0.236)
Inactive (4Q ago) 0.801 0.852 0.867 0.902
(0.120) (0.130) (0.134) (0.141)
Unemployed less than 1 year (4Q ago) reference reference reference reference
reference reference reference reference
Unemployed 1+ (4Q ago) 0.726 0.692 0.712 0.714
(0.238) (0.230) (0.239) (0.245)
Married: Yes 0.917 0.924 0.929 0.945
(0.163) (0.172) (0.175) (0.182)
Child: Yes 1.100 1.116 1.123 1.075
(0.166) (0.174) (0.176) (0.170)
Full-Time Job reference reference reference reference
reference reference reference reference
Part-time Job 0.401*** 0.378*** 0.376*** 0.383***
(0.059) (0.064) (0.064) (0.066)
1 managers and senior officials reference reference reference
reference reference reference
2 professional occupations 2.605* 2.876* 3.091*
(1.128) (1.261) (1.382)
3 associate professional and technical 1.717 1.771 1.927
(0.648) (0.670) (0.744)
4 administrative and secretarial 0.989 1.005 0.972
(0.339) (0.345) (0.342)
5 skilled trades occupations 0.637 0.639 0.632
(0.258) (0.260) (0.262)
6 personal service occupations 0.507 0.523 0.526
(0.228) (0.236) (0.239)
7 sales and customer service occupation 2.458** 2.472** 2.442*
(0.829) (0.835) (0.848)
8 process 1.643 1.694 1.600
(0.686) (0.711) (0.686)
9 elementary occupations 1.549 1.583 1.602
(0.502) (0.514) (0.535)
Industry sector controls+
north east reference reference
reference reference
north west 1.575 1.628
(0.637) (0.684)
merseyside 1.698 1.725
(0.891) (0.926)
yorkshire & humberside 1.831 1.986
(0.719) (0.807)
east midlands 1.017 1.077
(0.422) (0.460)
west midlands 0.965 1.014
(0.389) (0.423)
eastern 1.117 1.074
(0.450) (0.446)
london 1.400 1.372
(0.565) (0.574)
south east 1.394 1.477
(0.525) (0.578)
south west 1.222 1.273
(0.502) (0.540)
wales 0.650 0.665
(0.322) (0.338)
scotland 2.084 2.282*
(0.824) (0.932)
northern ireland 1.092 1.618
(0.876) (1.332)
N 3113 3113 3113 3113 1653160
Exponentiated coefficients; Standard errors in parentheses: * p<0.05, ** p<0.01, *** p<0.001"
+ Industry sector controls are included in this specification, but not presented in this initial estimation