Introduction Entrepreneur

1 Introduction
The entrepreneur is at the same time one of the most intriguing and one of the
most elusive characters . . . in economic analysis. He has long been recognised as
the apex of the hierarchy that determines the behaviour of the firm and thereby
bears a heavy responsibility for the vitality of the free enterprise society. (Baumol,
1968, p. 64)
Self-employment is unquestionably the oldest way by which individuals offer and

sell their labour in a market economy. At an earlier time, it was also the primary
way. Despite this history, its principal features and the characteristics that dif-
ferentiate self-employment from wage and salary employment have attracted the
attention of only a handful of students of the labour market. (Aronson, 1991, p. ix)
Entrepreneurship is increasingly in the news. Governments all over the

world extol its benefits and implement policies designed to promote
it. There are several reasons for this interest in, and enthusiasm for,
entrepreneurship. Owner-managers of small enterprises run the majority
of businesses in most countries. These enterprises are credited with
providing specialised goods and services that are ignored by the largest
firms. They also intensify competition, thereby increasing economic effi-
ciency. Some entrepreneurs pioneer new markets for innovative products,
creating new jobs and enhancing economic growth. In a few cases, today’s
small owner-managed enterprises grow to become tomorrow’s industrial
giants. Even those that do not may create positive externalities, like the
development of supply chains that help attract inward investment, or
greater social inclusion. It is sometimes also claimed that the decentrali-
sation of economic production into a large number of small firms is good
for society and democracy, as is the fostering of a self-reliant and hardy
‘entrepreneurial spirit’.
Entrepreneurship has only recently come to be regarded as a subject.
A complete view of it recognises its multidisciplinary academic under-
pinnings, drawing as it does from Economics, Finance, Business and
Management, Sociology, Psychology, Economic Geography, Economic
History, Law, Politics and Anthropology. This heterogeneous provenance
reflects the multidimensional nature of entrepreneurship, which partly
Downloaded from https://www.cambridge.org/core. Indian Statistical Institute (Delhi), on 08 Apr 2021 at 18:08:13, subject to the Cambridge
Cambridge Books Online © Cambridge
Core terms of use, available at https://www.cambridge.org/core/terms. University Press, 2009
https://doi.org/10.1017/CBO9780511493430.003
2 The Economics of Self-Employment and Entrepreneurship
contributes to the elusiveness of the entrepreneur alluded to by William

Baumol.
1.1 Aims, motivation and scope of the book

There is general agreement that entrepreneurship is sadly neglected in
most economics textbooks (see, e.g. Rosen, 1997; Kent and Rushing,
1999). Some commentators have attributed this neglect to an inherent
unsuitability of economics for studying entrepreneurship. These com-
mentators allege that economics is concerned only with equilibrium out-
comes in competitive markets with perfectly informed agents, whereas
entrepreneurship embodies imperfect information and unexpected inno-
vations that disrupt equilibria (Barreto, 1989; Kirchhoff, 1991; Harper,
1996; Rosen, 1997). While it is true that economists usually take the
state of technology as given, this point should not be pushed too far.
For a start, there is much more to entrepreneurship than just innova-
tion. This book makes the case that the tools of modern economics are
invaluable for understanding the determinants of, and constraints on,
entrepreneurship, the behaviour of entrepreneurs, the contribution of
entrepreneurs to the broader economy and the impact of government
policies on entrepreneurship. It is hoped that the book will go some way
towards correcting mistaken and prejudiced impressions about the con-
tribution of economics to this field, and will convince the reader that it is
not an oxymoron to talk about the ‘economics of entrepreneurship’.
One aspect of entrepreneurship that economists have researched quite
thoroughly is occupational choice. Despite the limitations of employ-
ing such a narrow definition – an issue we shall discuss further below –
many applied labour economists have equated self-employment with
entrepreneurship, and have analysed behavioural choice between paid-
and self-employment. Unfortunately, here too the economics textbooks
have been guilty of neglect: the above quotation from Richard Aronson
about self-employment remains as true today as it did over a decade ago.
To appreciate the economic importance of self-employment, consider the
following facts:
1. Around 10 per cent of the workforces in most OECD economies
are self-employed. The figure climbs to about 20 per cent when in-
dividuals who work for the self-employed are also included (Haber,
Lamas and Lichtenstein, 1987). Two-thirds of people in the US
labour force have some linkage to self-employment, by having expe-
rienced self-employment, by coming from a background in which the
household head was self-employed, or by having a close friend who
is self-employed (Steinmetz and Wright, 1989). And by the end of
https://doi.org/10.1017/CBO9780511493430.003
Introduction 3
their working lives, about two-fifths of the American workforce will

have had at least one spell of self-employment (Reynolds and White,
1997).
2. Between 80 and 90 per cent of businesses are operated by self-
employed individuals (Acs, Audretsch and Evans, 1994; Selden,
1999).
3. Many employees in industrialised countries claim that they would
like to be self-employed. For example, according to Blanchflower
(2000), 63 per cent of Americans, 48 per cent of Britons and
49 per cent of Germans stated a preference for self-employment over
paid-employment.1
Why have the economics textbooks so conspicuously neglected self-
employment and entrepreneurship? The answer is unclear, but might
involve a lingering mistrust of entrepreneurship among economists, and
an inability to pigeon-hole this multifaceted subject. There is certainly no
lack of research on self-employment and entrepreneurship by economists.
Indeed, we hope that one of the contributions of this book is to organ-
ise and assess the current state of this branching, acquisitive and rapidly
growing literature. The book is intended to serve as a comprehensive
overview and guide to researchers and students of entrepreneurship in a
variety of disciplines, not just in Economics. The book can also be used
to support teaching in modules such as Small Business Economics and
Entrepreneurship. Readers with a working knowledge of basic undergrad-
uate calculus, statistics and economics should cope easily with the book’s
modest technical demands. Readers without this technical background
will still be able to read the majority of the book without difficulty, and
understand the gist of the remainder by ‘reading between the Greek’.
For brevity and focus, some topics will be excluded. These in-
clude entrepreneurship in education; ‘social entrepreneurship’ and
‘intrapreneurship’;2 organisational, strategic and managerial decision
making by entrepreneurs; network and organisational ecology approaches
to entrepreneurship; ‘evolutionary economics’; and practical advice
(including ‘how to’ information) to entrepreneurs. Nor will we provide
descriptive case studies either of individual entrepreneurs, or of small
firms or the industries in which they operate. These topics are ably cov-
ered in numerous texts in the Business/Management literature, and will
not be repeated here.
1.2 Structure of the book

The remainder of this chapter is organised as follows. Section 1.3 dis-
cusses issues in the definition and measurement of entrepreneurship and
https://doi.org/10.1017/CBO9780511493430.003
self-employment. The distinction between these two concepts explains

their joint presence in the title of this book. Sections 1.4 and 1.5 intro-
duce the reader to some stylised facts about the self-employed, including
their number, incomes and income inequality. Section 1.6 describes some
useful econometric models of occupational choice referred to extensively
in the book.
Part I deals with theories of entrepreneurship and the character-
istics of entrepreneurs. Chapter 2 outlines economic theories of the
entrepreneur and entrepreneurship, both ‘early’ (Section 2.1) and
‘modern’ (Section 2.2). The former tend to paint a broad-brush picture
of entrepreneurship, whereas the latter apply the tools of microeco-
nomics to the problem of entrepreneurship as an occupational choice.
Chapter 3 fills out some detail by considering the roles of pecuniary
factors, individual characteristics, family background and environmen-
tal variables for explaining the decision to become an entrepreneur.
Chapter 4 considers separately issues of race, gender and immigration
as they relate to entrepreneurship.
Part II treats the important problem of raising finance for en-
trepreneurial ventures. Chapter 5 analyses the economic issues arising
from bank finance of new start-ups, bank loans being the largest and
most frequently used source of outside funds by entrepreneurs. Although
we will touch on such ‘practical’ issues as collateral and bank–borrower
relationships, the focus of this chapter will be on the market failures that
can arise as a consequence of asymmetric information, including credit
rationing, under-investment and over-investment. These issues carry im-
portant policy implications about whether governments should encour-
age or discourage entrepreneurship. Chapter 6 considers other sources
of funds, including equity finance, trade credit, group lending schemes
and borrowing from family and friends. Chapter 7 summarises evidence
on whether, and to what extent, entrepreneurs face credit rationing.
Part III investigates what happens to new ventures after they are
launched. Chapter 8 discusses theory and evidence about job creation
by entrepreneurs, and their labour supply and retirement behaviour.
Chapter 9 analyses the growth, innovation and exit behaviour of new
ventures and small firms. With all of this apparatus in place, it is possi-
ble to explore systematically the scope for governments to intervene in
the market to promote entrepreneurship. This is the subject of part IV
(chapter 10). Government policies can take several forms, including
credit market interventions, taxation and direct assistance and regula-
tion. Chapter 11 concludes, and draws together some suggestions for
future research where our understanding of particular issues is especially
incomplete.
https://doi.org/10.1017/CBO9780511493430.003
Introduction 5
1.3 Definition and measurement issues

The problem of defining the word ‘entrepreneur’ and establishing the boundaries
of the field of entrepreneurship still has not been solved. (Bruyat and Julien, 2001,
p. 166)
Our first and most pressing task is to define entrepreneurs and
entrepreneurship. Unfortunately, this happens to be one of the most
difficult and intractable tasks faced by researchers working in the field.
There is a proliferation of theories, definitions and taxonomies of en-
trepreneurship which often conflict and overlap, resulting in confusion
and disagreement among researchers and practitioners about precisely
what entrepreneurship is (see Parker, 2002a). For example, consider the
following illustrative and abbreviated set of viewpoints. In applied work,
labour economists often equate entrepreneurs with the self-employed, on
the grounds that the self-employed fulfil the entrepreneurial function of
being risk-bearing residual claimants. However, others think this defini-
tion is too broad, claiming that only business owners who co-ordinate
factors of production (in particular, those who employ workers) are re-
ally entrepreneurs. Still others think that the economist’s definition is too
narrow, because it excludes entrepreneurship in the corporate and social
spheres. Then there are those steeped in the Schumpeterian tradition who
argue that entrepreneurship is identified primarily with the introduction
of new paradigm-shifting innovations. Others again have emphasised psy-
chological traits and attitudes supposedly peculiar to entrepreneurship.
And so the list goes on.
It is easier to define the terms ‘self-employed’ and ‘self-employment’,
though even here there are measurement problems and disagreements,
which we discuss below. Given the widespread availability of data on the
self-employed in government and private surveys world wide, it is also
an easier entity to operationalise in empirical research (Katz, 1990). To
cut through a paralysing and ultimately fruitless debate, and to achieve
consistency, we will adopt the following convention in this book. At the
conceptual level, the terms ‘entrepreneur’ and ‘entrepreneurship’ will be
used; at the practical level, where issues of measurement, estimation and
policy are involved, we will use the closest approximation to the manifes-
tation of entrepreneurship that appears to be suitable. That will usually
be ‘self-employment’, though occasionally the term ‘small firm’ will be
more relevant. Note that the problems of defining ‘small’ firms are also
non-trivial. Firm-size definitions are arbitrary and industry-specific, and
are not obviously congruent with entrepreneurship. Not all entrepreneurs
run small firms, and not every small firm is run by an entrepreneur (Brock
and Evans, 1986; Holtz-Eakin, 2000).
https://doi.org/10.1017/CBO9780511493430.003
The self-employed are often taken to be individuals who earn no wage

or salary but who derive their income by exercising their profession or
business on their own account and at their own risk. Likewise, partners
of an unincorporated business are usually classified as self-employed. It
is sometimes helpful to partition the self-employed into employers and
own-account workers (the latter of which work alone), or into owners of
incorporated or unincorporated businesses. In most countries, incorpo-
ration of a business renders it susceptible to company law, which requires
the owner to publicise stipulated data about the business. In the UK and
USA, for example, these include audited accounts if business turnover
exceeds a specified level. Despite the costs and inconvenience involved,
there are several advantages to incorporation. They include: protec-
tion from creditors via limited liability; favourable pension contribution
rules; greater credibility with customers (Storey, 1994b); and payment
of corporation tax on company profits, which at high owner incomes
may be substantially lower than personal income tax rates (Fuest, Huber
and Nielsen, 2002). Incorporated business owners can either receive an
employee salary as a director of their company, or they can pay themselves
dividends – so escaping payroll taxes in some jurisdictions, including
the UK. Most self-employed people in most countries own unincorpo-
rated businesses,3 which renders their incomes liable to personal income
tax.
The first definitional problem is that in many countries (including the
UK and USA), owners of incorporated businesses are defined as em-
ployees rather than self-employed. This is despite them resembling in all
other respects (e.g. residual claimant status) the ‘self-employed’. This is
sometimes an important distinction in applied self-employment research.
Second, for some individuals, the legal and tax-based definitions of self-
employment are at variance with each other. In law, the issue comes down
to whether there is a contract of service or a contract for services. The first
indicates paid-employment, the second self-employment.4 In contrast,
different criteria are often used for determining who is self-employed for
the purposes of income taxation and social security eligibility. There is no
shortage of examples where the legal and tax definitions fail to coincide
(Harvey, 1995; Dennis, 1996).
Third, in many government surveys used in empirical research, self-
employment status is self-assessed by the survey respondents. This can
lead to further differences in the classification of workers, compared with
legal and tax-based definitions (Casey and Creigh, 1988; Boden and
Nucci, 1997). Partly for this reason, some surveys (e.g. the UK Labour
Force Survey (LFS) and the US Characteristics of Business Owners
(CBO)) use tax-based definitions of self-employment.
https://doi.org/10.1017/CBO9780511493430.003
Introduction 7
Fourth, there appears to be a ‘grey area’ between paid-employment and

self-employment. Some workers classified as self-employed with apparent
autonomy over their work hours are effectively employees, being ‘periph-
eral’ workers subordinated to the demands of one client firm (Pollert,
1988; Harvey, 1995). For example, Harvey (1995) contends that, in the
construction industry, most self-employed workers are to all intents and
purposes direct employees, working exclusively for one contractor, and
providing independently only their labour.5 Employers provide the mate-
rials, capital and plant and set the terms of work and pay. This is a poten-
tial matter of concern to policy makers to the extent that self-employed
workers are engaged on worse terms than employees, lacking job secu-
rity, entitlement to holidays, sick pay, employment protection, or trades
union rights. It is sometimes argued that employers actively seek to or-
ganise their workforce in self-employment contracts, to cut costs and to
avoid their social obligations.6
This debate is related to one on contracting-out of labour by large firms,
a phenomenon that was thought to have been particularly pronounced
in the 1980s, a decade when self-employment rates in several countries
increased dramatically. Hakim (1988) pointed to some evidence that UK
firms made some limited moves during the 1980s to outsource work to
self-employed contractors. However, according to Blackburn (1992) sub-
contracting has always existed and the increase in the UK in the 1980s
was relatively slight. Also, case study evidence from Leighton (1982)
suggested that many employers prefer hiring employees to self-employed
contractors even in industries where self-employment is relatively com-
mon. Benefits of hiring employees directly include greater control, disci-
pline and stability. If this argument is valid, it suggests that the amount of
‘grey’ self-employment in the form of outsourced labour has not grown
dramatically in the recent past, and is presumably likely to remain stable
in the near future.
Other examples of workers in the ‘grey area’ between employment and
self-employment include commission salespersons, freelancers, home-
workers and tele-workers,7 workers contracted through temporary em-
ployment agencies and franchise holders. Regarding the latter, it is
often difficult to determine whether a franchise is an independent small
business or part of a large firm. Felstead (1992) argued that many ‘self-
employed’ franchisees are effectively directed by their franchisor, who
holds most of the ownership rights and has a senior claim on profits.
With relatively little discretion about the format of their business, one
could certainly claim that some franchisees resemble branch managers
more than independent entrepreneurs. On the other hand, one could
argue that a self-employed retailer who is pressurised into stocking the
https://doi.org/10.1017/CBO9780511493430.003
goods of mainly one goods manufacturer also has limited discretion about
the nature of his or her business. And like other self-employed workers,
franchisees face uncertainty. In their case, not only is their income un-
certain, but there is also a possibility that the franchisor will either go out
of business, or refuse to renew the franchise agreement at the expiration
of its term.8
Two other ‘grey’ categories include unpaid family workers who work
in a business run by a self-employed person; and members of worker
co-operatives, who are not obviously either employees or self-employed
workers in the conventional sense of the term. Both groups tend to be
more numerous in developing than in developed countries, and in rural
than in urban areas. According to Bregger (1996, p. 5), ‘Unpaid family
workers are persons who work on a family farm or in a family business
for at least 15 hours a week and who receive no earnings or share of
the profits of the enterprise’. Blanchflower (2000) detected substantial
variation within developed countries in the proportion of self-employed
workers who are unpaid family workers, being as high as 33 per cent in
Japan, compared with 14 per cent in Italy and just 1.7 per cent in the USA.
As Blanchflower points out, it may not make sense just to discard unpaid
family workers from the self-employment count, since they often share
indirectly (e.g. via consuming household goods) the proceeds generated
by the business. Worker co-operatives are also relatively uncommon in the
UK, and tend to be larger and better established in European countries,
such as France, Italy and Spain (Spear, 1992).
Section 1.4 documents some international evidence on levels of, and
trends in, aggregate self-employment rates. At the aggregate level, it is
likely that alternative measures of self-employment are highly correlated
with each other, allowing trends within a given country to be identified
fairly reliably (Blau, 1987). However, to the extent that different countries
use different definitions of self-employment, cross-country comparisons
of levels have to be treated with caution.
1.4 International evidence on self-employment rates

and trends
There is great diversity in the level and time-series pattern of self-
employment rates across countries. This is evident from tables 1.1, 1.2 and
1.3, which summarise data for a selection of OECD countries, Eastern
European transition economies and developing countries, respectively.9
Two additional features of these tables stand out. One is that self-
employment rates are higher on average in developing than developed
countries. A second is that the treatment of agricultural workers makes
https://doi.org/10.1017/CBO9780511493430.003
Introduction 9
Table 1.1 Aggregate self-employment rates in some selected OECD

countries, 1960–2000a (per cent)
1960 1970 1980 1990 2000
A All workers
USA 13.83 8.94 8.70 8.50 7.33
Canadab 18.81 13.20 9.74 9.52 10.66
Japan 22.68 19.18 17.18 14.05 11.34
Mexico 34.25 31.29 21.67 25.64 28.53
Australia 15.86c 14.09 16.16 15.05 13.49
Franceb 30.51 22.17 16.79 13.26 10.56
Italy 25.93 23.59 23.26 24.53 24.48
Netherlandsb 21.87 16.65 12.23 9.64 10.46d
Norway 21.79 17.90 10.03 9.24 7.03
Spainb 38.97 35.59 30.47 26.27 20.49
UK 7.28 7.36 8.05 13.32 11.34
B Non-agricultural workers
USA 10.45 6.94 7.26 7.51 6.55
Canadab 10.17 8.33 7.05 7.40 9.46
Japan 17.38 14.44 13.75 11.50 9.35
Mexico 23.01 25.20 14.33 19.89 25.48
Australia 11.01c 10.00 12.73 12.34 11.72
Franceb 16.90 12.71 10.71 9.32 8.06
Italy 20.60 18.97 19.20 22.24 23.21
Netherlandsb 15.08 12.02 9.06 7.84 9.25d
Norway 10.14 8.61 6.53 6.12 4.83
Spainb 23.60 21.55 20.63 20.69 17.69
UK 5.89 6.27 7.11 12.41 10.83
Notes: a Self-employment rates defined as employers plus persons working on their own
account, as a proportion of the total workforce.
b Includes unpaid family workers c 1964 not 1960 d 1999 not 2000.
Source: OECD Labour Force Statistics, issues 1980–2000, 1970–81 and 1960–71.
a substantial difference to measured self-employment rates in most (but

not all) countries.
1.4.1 The OECD countries

Consider the USA first. According to table 1.1,10 if agricultural workers
are included, the US self-employment rate has been in continual decline
since at least 1960. According to Steinmetz and Wright (1989), this de-
cline can actually be traced back to the 1870s, when the self-employment
rate stood at just over 40 per cent of the labour force, and it underwent
an especially steep decline between 1950 and 1970 (see also table 1.1).11
https://doi.org/10.1017/CBO9780511493430.003
Table 1.2 Aggregate self-employment rates in some selected transition

economies, 1980–1998/1999a (per cent)
1980 1990 1992 1994 1998/99b
Poland (all workers) 25.44 27.17 22.44

(non-agricultural) 3.37 9.16 11.70
Russian Federation 0.76 5.29
Czech Rep. (all workers) 10.18 14.59
(non-agricultural) 10.25 14.50
Hungary (all workers) 16.93 14.56
Slovak Rep. (all workers) 6.21 7.80
account, as a proportion of the total workforce.
b ‘1998/99’ is 1998 for Poland and 1999 for the Russian Federation.
Source: UN Yearbook of Labour Statistics, various issues.
Table 1.3 Aggregate self-employment rates in developing countries,

1960s–1990sa (per cent)
1960sb 1970sb 1980sb 1990sb
Africa
Mauritius 13.03 10.30 n.a. 16.72
Egypt 29.19 26.14 28.20 27.19
Americas
Bolivia n.a. 48.86 40.27 34.81
Costa Rica 20.78 17.10 21.80 24.70
Dominican Rep. 44.79 29.44 36.46 37.11
Ecuador 42.97 37.81 37.27 37.03
Asia
Bangladesh 7.33 45.56 38.83 29.59
Korean Rep. 44.04 33.92 33.07 28.02
Pakistan 21.94 46.90 55.95 48.18
Sri Lanka 26.94 22.90 24.74 26.68
Thailand 29.83 29.65 29.75 28.45
account, as a proportion of the total workforce. Includes agricultural workers.
b ‘1960s’ is either 1960, 1961, 1962 or 1963 for all countries; ‘1970s’ is some year between
1970 and 1976; ‘1980s’ is 1980 or 1981 except Ecuador (1982), Costa Rica (1984) and
Bolivia (1989); ‘1990s’ is some year between 1990 and 1996.
Source: UN Yearbook of Labour Statistics, various issues.
https://doi.org/10.1017/CBO9780511493430.003
Introduction 11
However, if agricultural workers are excluded, table 1.1 shows that a re-
vival in US self-employment occurred during the 1970s and 1980s, a
finding that has also been observed by some other authors.12 But unlike
these other authors, table 1.1 reveals that this revival in non-agricultural
US self-employment has apparently come to an end: by 2000 the non-
agricultural self-employment rate had fallen back to below its 1970 level.
The US experience is mirrored by France, which has also seen its over-
all self-employment rate decline steadily since the start of the twentieth
century (Steinmetz and Wright, 1989).13 However, this trend is not ob-
served in every OECD country. For example, table 1.1 shows that both
Canadian self-employment rates exhibited U-shaped patterns, increas-
ing particularly strongly in the 1990s.14 In contrast, both measures of the
UK self-employment rate increased dramatically in the 1980s, a finding
that attracted substantial research interest when it was first discovered
(Hakim, 1988; Campbell and Daly, 1992). It declined in the 1990s, es-
pecially among males (Moralee, 1998). According to Storey (1994a), the
UK historical trend in self-employment was one of steady decline be-
tween 1910 and 1960, followed by increase from 1960 to 1990, with the
rate in 1990 being similar to that in 1910.
Most researchers tend to exclude agricultural workers from their def-
initions of self-employment, on the grounds that farm businesses have
very different characteristics to non-farm businesses. It has been known
at least since Kuznets (1966) that the agricultural sector tends to decline
as an economy develops – and that this may distort self-employment
trends (Blanchflower, 2000). Consequently, to analyse trends we focus
henceforth on panel B of table 1.1. These data indicate a striking variety
of patterns over 1960–2000. Four countries (Japan, France, Norway and
Spain) had steadily declining self-employment rates. Six witnessed a re-
vival in self-employment at some point within the period (USA, Canada,
Mexico, Italy, the UK and the Netherlands); and one (Australia) had a
relatively stable self-employment rate.
Finally, we will say a word about the relative importance of small firms
in the economy. The overwhelming majority of US businesses employ
fewer than five individuals (see, e.g. White, 1984; Brock and Evans, 1986,
chapter 2, for details). There are high rates of business formation and dis-
solution among small firms, especially in industries like retailing, where
low capital requirements make entry easy and keep profits modest. The
aggregate number of small businesses grew in the USA in the post-war
period, but their relative economic importance (measured in terms of
their employment share or share of gross domestic product (GDP)) de-
clined somewhat over that period. The most recent evidence suggests that
the share of private non-farm GDP accounted for by small businesses in
https://doi.org/10.1017/CBO9780511493430.003
the USA has now stabilised, at around 50 per cent over the last two
decades (SBA, 2002a). That the earlier decline was not greater is mainly
attributable to the growth of the service sector, in which small firms are
disproportionately concentrated. We will return to the issue of changing
industrial structure later in the book.
1.4.2 The transition economies of Eastern Europe

Of special interest are the so-called ‘transition economies’ of Eastern
Europe, which underwent a switch from communist central planning to
a more market-based system at the end of the 1980s. In the words of
Earle and Sakova (2000, p. 583): ‘it is difficult to imagine a regime more
hostile towards self-employment and entrepreneurship than the centrally
planned economies of Eastern Europe.’ These regimes fixed prices and
wages, placed restrictions on hiring workers and acquiring capital and
levied confiscatory taxes on entrepreneurs. This is reflected in the low
non-agricultural self-employment rates observed in Poland in 1980, and
in Russia in 1992 (see table 1.2).15
Part of the interest in studying the transition economies is that they
serve as a test-bed for the strength of dormant entrepreneurial vigour
that could be released after market liberalisation.16 Real opportunities
emerged for entrepreneurs to exploit market gaps left by the previous
communist regimes, especially in the provision of services and the pro-
duction of consumer goods. The incentive to become self-employed was
no doubt enhanced by declining opportunities in wage employment and
growing unemployment as the sprawling former state-run companies be-
gan to contract in the 1990s. But although legal barriers to private enter-
prise and self-employment came down after 1989, bureaucracy and the
limited rule of law has continued to stifle productive entrepreneurship in
many of these countries (Baumol, 1990, 1993; Dutz, Ordover and Willig,
2000). For example, Baumol contended that, although the supply of en-
trepreneurs varies from country to country, probably a greater source of
variation in entrepreneurs’ social productivity is the scope to engage in
privately profitable but socially unproductive rent-seeking and organised
crime. If payoffs to such activities are sufficiently high, entrepreneurs will
rationally divert effort from productive innovation to exploit them. It re-
mains to be seen what will happen to productive entrepreneurship in the
transition economies as we move further into the twenty-first century.17
1.4.3 Developing countries

In the early post-war period, researchers attached great importance
to fostering entrepreneurship in developing countries. In the words of
https://doi.org/10.1017/CBO9780511493430.003
Introduction 13
W. Arthur Lewis (1955, p. 182): ‘Economic growth is bound to slow

unless there is an adequate supply of entrepreneurs looking out for new
ideas, and willing to take the risk of introducing them.’ According to Leff
(1979), interest in the issue had dwindled by the 1970s. Leff asserted that
this was because of a perception that the entrepreneurial problem had
been ‘solved’, with high rates of real output growth serving as evidence of
entrepreneurial vigour. Subsequently, slower growth, high rates of pop-
ulation growth, widespread failures of state-owned enterprises (SOEs),
constraints on public sector employment and the spread of free-market
beliefs reactivated interest in promoting entrepreneurship in developing
economies.
Table 1.3 shows that, on average, developing countries have markedly
higher self-employment rates than developed countries. Nevertheless,
table 1.3 also reveals considerable variation in these rates. It has been not-
ed elsewhere that Asian countries often have very high self-employment
rates, sometimes exceeding half the workforce.18 However, in contrast to
claims by some previous researchers that the trend in developing countries
is away from self-employment (Blau, 1987; Schultz, 1990), the evidence
in table 1.3 reveals that no such trend can be generally established.
Why are self-employment rates so high in developing countries? We will
return to the broad issue of economic development and entrepreneurship
later in the book; here we consider just two specific factors. First, the data
in table 1.3 (and those used in most other studies of these countries) in-
clude agriculture. Agriculture plays a prominent role in the economies of
many developing countries, in which high self-employment rates are tra-
ditionally found. Second, high self-employment rates may reflect limited
development of formal economic and financial markets. For example,
Leibenstein (1968) argued that entrepreneurship in developing coun-
tries often simply involves overcoming constraints caused by poor eco-
nomic and financial infrastructure, and is quite basic in nature. This
viewpoint is related to the long-standing dual labour market model of
development, comprising a formal urban sector in which employees earn
premium wages, and an informal rural sector in which entrepreneurs re-
ceive below-average incomes (Lewis, 1955; Harris and Todaro, 1970).
The model predicts that as poor economies develop, labour will move
from the informal to the formal sector, with a decline in self-employment.
However, evidence from the field refutes both the prediction of higher in-
come in paid-employment in developing countries, and the prediction of
workers shifting from self- to paid-employment as they age.19 Arguably,
more recent economic theories offer greater scope for explaining high
self-employment rates in developing countries. We return to this topic in
chapter 3, section 3.3.
https://doi.org/10.1017/CBO9780511493430.003
1.5 Self-employment incomes and income inequality

This section attempts three tasks. First, we discuss issues relating to the
definition and measurement of self-employment incomes, and review ev-
idence about the levels of and trends in average self-employment incomes
relative to average paid-employment incomes. Second, we analyse the in-
equality of self-employment incomes. Third, we review evidence from
earnings functions on the determinants of self-employment incomes.
1.5.1 Incomes and relative incomes

Measurement issues
Self-employment income can be measured in several different ways. Con-
sider the following identity:
Net profit ≡ Revenue − Costs ≡ Draw + Retained earnings.
Net profit from running an enterprise is a widely used measure of self-
employment income. An alternative is Draw, the amount of money drawn
from the business on a regular basis by the owner. This represents the
consumption-generating value of the business and as such may be less
prone to income under-reporting. A less frequently used third measure
is Draw augmented by the growth in business equity (Hamilton, 2000).
It should be stressed at the outset that any analysis of self-employment
income data should be performed with the utmost caution. There are
several reasons for this:
1. Income under-reporting by the self-employed. This is possibly the most se-
rious problem with using self-employment data. It is partly attributable
to self-employed respondents who over-claim business tax deductions,
or under-report gross incomes to the tax authorities, mistrusting inter-
viewers’ claims that they are truly independent of the tax inspectorate.
Ways of estimating self-employment income under-reporting rates are
discussed in chapter 10, subsection 10.4.1.
2. Different ways of treating owners of incorporated businesses. Incorporated
self-employed individuals are usually treated as employees of their
company. Because they are richer on average, their exclusion from
the self-employed sample may bias downwards the average income of
the self-employed.20 On the other hand, including the incorporated
self-employed is not without its problems, since it is not clear how to
interpret the salary that an incorporated self-employed business owner
chooses to pay her/himself.
3. Relatively high non-response rates to survey income questions by the self-
employed. This problem can be quite pervasive and can substantially
bias estimates of absolute and relative returns to self-employment
https://doi.org/10.1017/CBO9780511493430.003
Introduction 15
(Devine, 1995).21 There are several possible reasons for survey non-
response. One is mistrust of survey interviewers by self-employed re-
spondents, for the reasons given above. A second is that richer people
(of whom a disproportionate number are self-employed – see below)
have a higher marginal valuation of time, so participate less in sur-
veys. Third, many self-employed do not accurately know their incomes,
which according to Meager, Court and Moralee (1994) accounted for
two-thirds of missing British income cases in the 1991 British House-
hold Panel Survey (BHPS) – rather than refusal to co-operate with the
survey.
4. A failure to deal properly with negative incomes and ‘top-coding’ can in-
troduce biases (Devine, 1995). Many researchers either drop negative
income observations or round them up to a small positive number
before applying logarithmic transformations; both practices impart an
upward bias to average self-employment income. Top-coding on the
other hand, which is a procedure of truncating very high earnings val-
ues to a maximum level, imparts a downward bias.
5. Ignoring employee fringe benefits that are unavailable in self-employment bi-
ases upwards any relative income advantage to self-employment. Some
of these benefits can be substantial, especially employer contributions
to health care and occupational pension schemes (Holtz-Eakin, Penrod
and Rosen 1996; Wellington, 2001).
6. Self-employment incomes include returns to capital as well as returns to
labour. National Accounts’ experts have long argued about how best
to disentangle these returns, which might explain why few researchers
choose to separate them in practice.22 Headen (1990) proposed an
especially straightforward approach, which works as follows. Let h j ,
y j and w j denote an individual’s work hours, total ‘labour’ income
and wage rate respectively in sector j = {E, S }, where E is paid-
employment and S is self-employment. Also, let yk denote returns
to capital in self-employment. While h j and y j values are observed
in most data sets, w j s are not and must be calculated (in E) or esti-
mated (in S ). We have yE = w E h E and yS = w Sh S + yk . To estimate
yk , first calculate w E = yE / h E and estimate an ‘earnings function’ like
ln w E = β X + u, where X is a vector of personal characteristics, β is
a vector of coefficients and u is a random disturbance (see chapter 1,
subsection 1.5.3). This yields parameter estimates β̂. Second, predict
ŵ S for the self-employed from ln w S = β̂ X. This can be taken as the
return to self-employed labour assuming (i) that employee incomes
are purely returns to labour, and (ii) that the self-employed have the
same rates of return to personal characteristics as employees do (see
subsection 1.5.3 for a critical assessment of this assumption). Finally,
https://doi.org/10.1017/CBO9780511493430.003
denoting sample mean values by overbars, calculate

yk / h S = yS/ h S − ŵ S .
Using this approach Headen (1990) estimated the returns to labour
and capital as, respectively, 84 and 16 per cent for a sample of US
self-employed physicians in 1984.
In practice, few researchers have addressed any of the above six prob-
lems in much depth. Only the first two have attracted any real attention,
but even here the coverage has been uneven. This caveat should be borne
in mind when interpreting the following evidence.
International evidence on relative average self-employment

incomes and trends
Standard economic theory predicts that workers move between occupa-
tions until incomes in each occupation are equalised. However, that pre-
diction must be qualified when one takes into account the heterogeneity
of individual abilities and job characteristics, including non-pecuniary
compensating differentials, risk and over-optimism among individuals
choosing risky occupations. As will be seen below, few studies have
found equality between average incomes in self-employment and paid-
employment.
In the USA, the evidence on the relative income position of the self-
employed is mixed. This is no doubt partly attributable to the different
income definitions and sampling frames used in different data-sets. How-
ever, there is a tentative emerging consensus that the self-employed earn
less on average than employees do.23 Hamilton (2000) conducted an
especially thorough study of relative self-employment incomes. Because
of the pronounced income inequality of self-employment incomes (see
below), Hamilton analysed median rather than mean incomes. Hamilton
controlled for experience when comparing incomes in self-employment
and paid-employment, and utilised all three different measures of self-
employment income described at the start of this section. Using 1984
Survey of Incomes and Program Participation (SIPP) data on personal
characteristics, Hamilton estimated that on average all individuals except
those in the upper quartile of the self-employment income distribution
would have earned more, and enjoyed higher future income growth rates,
if they had quit self-employment and become employees. For example,
for individuals in business for ten years the median earnings differential
was 35 per cent in favour of paid-employment. This finding, which was
found to be robust to different definitions of self-employment income,
may under-state the true differential since it does not take account of
employee fringe benefits such as employer-subsidised health insurance.
https://doi.org/10.1017/CBO9780511493430.003
Introduction 17
Conversely, however, Hamilton may have over-stated the income differ-

ential by ignoring the possibility of income under-reporting and business
tax deduction opportunities for the self-employed.
There is also evidence that US median incomes in self-employment
have lagged behind median employee incomes for several decades
(Carrington, McCue and Pierce, 1996: Current Population Survey
(CPS) data, 1967–92).24 Aronson (1991: US Social Security data, 1951–
88) showed that a 48 per cent income advantage to the self-employed in
1951–4 had dwindled to a 23 per cent advantage by 1975–9. This be-
came a 10 per cent disadvantage by 1980–4, widening to a 20 per cent
disadvantage by 1985–8. It appears that a similar story holds irrespective
of occupation and education (SBA, 1986); and adjusting for the longer
average work hours of the self-employed reduced further their relative
income position – by around 70 per cent according to Aronson (1991).
The UK has also witnessed a downward trend in relative average
self-employment incomes since the 1970s (Robson, 1997; Clark and
Drinkwater, 1998).25 In contrast to the USA, the UK evidence points to a
relative income advantage to self-employment. Disagreement centres on
how large this advantage is. General Household Survey (GHS) micro-
data point suggests a small premium to self-employment of 7 per cent
over 1983–95, according to Clark and Drinkwater (1998) (see also
Meager, Court and Moralee, 1996). In contrast, aggregate UK National
Accounts data suggest a greater difference, of 35 per cent in terms of
pre-tax gross income in 1993 according to Robson (1997). It may be rel-
evant that the latter, unlike the former estimate, includes an adjustment
for income under-reporting by the self-employed.
Evidence from eleven OECD countries supports the notion that in
many countries the self-employed are not well remunerated relative to em-
ployees. According to OECD (1986), only in West Germany did the ratio
of median self-employment to paid-employment incomes exceed unity.
In Finland, Sweden and Japan the ratios were below that of the USA.
Similar evidence was also found independently by Covick (1983) and
Kidd (1993) for Australia; and Covick noted the same downward trend
in relative self-employment incomes as observed in the USA and the UK.
The special circumstances prevailing in the transition economies of
Eastern Europe may help explain why opposite findings have been found
there. Earle and Sakova (2000) studied self-employment choices and
incomes in six Eastern European countries between 1988 and 1993:
Poland, Russia, Slovakia, Bulgaria, Hungary and the Czech Republic.
They found that, in all countries apart from Poland, the mean income of
employees was less than that of own-account self-employed individuals,
which in turn was less than the mean income of self-employed employers.
https://doi.org/10.1017/CBO9780511493430.003
We conclude this section with three puzzles. One is why, if the self-
employed in the USA earn less on average than employees do, and if they
are only moderately older on average than employees are, they neverthe-
less possess substantially greater savings and asset holdings (Quadrini,
1999; Gentry and Hubbard, 2001). A second puzzle is why individuals
remain in self-employment despite apparently earning less and work-
ing longer hours (see chapter 8, section 8.2) than employees do. Third,
why do entrepreneurs invest in undiversified and hence risky private busi-
nesses, when they could obtain similar rates of return from less risky pub-
licly traded equity (Moskowitz and Vissing-Jørgensen, 2002)? A possible
answer to the first puzzle is income (but not asset) under-reporting by the
self-employed, reflecting the greater taxation of income than wealth. A
tentative answer to the second puzzle is proposed in section 8.2. Possible
answers to the third include non-pecuniary benefits to entrepreneurship,
a preference for skewed returns and systematic over-estimation by en-
trepreneurs of the probability of survival and success in entrepreneurship.
1.5.2 Income inequality

It is now well established that in most countries the incomes of the self-
employed are more unequal than employees’ are. This fact usually be-
comes immediately obvious when histograms of incomes are graphed sep-
arately for the two groups.26 Relatively large numbers of the self-employed
are concentrated in the lower and upper tails of their income distribution,
compared with employees. Consequently, when data from the two occu-
pations are combined, the self-employed are invariably observed to be
disproportionately concentrated in both the upper and lower tails of the
overall income distribution. In their British analysis based on 1991 BHPS
data, Meager, Court and Moralee (1996) showed that this result is not an
artefact of sampling error, and remained after controlling for observable
characteristics such as gender, work status, work hours, age, education,
industry and occupation.27
The same story of pronounced self-employment income inequality is
observed when sample data are mapped into scalar inequality measures,
such as the Gini coefficient or the mean log deviation.28 Typical results
were obtained by Parker (1999b), who computed a range of inequality
measures from UK Family Expenditure Survey (FES) data over 1979–
1994/5. Parker reported that self-employment income inequality indices
were between two and five times as great as those for employees, depend-
ing on the year and the particular inequality index chosen.29 Arguably the
UK context is particularly interesting because self-employment income
inequality grew especially rapidly in the 1980s, the same decade when
https://doi.org/10.1017/CBO9780511493430.003
Introduction 19
self-employment itself expanded substantially (see subsection 1.4.1). The

increased self-employment income inequality in the 1980s was large
enough for Jenkins (1995) to cite it as the primary reason for the growth of
overall UK income inequality in the 1980s, exacerbated by a greater self-
employment population share. Subsequently, Parker (1999b) reported
that by the mid-1990s UK self-employment income inequality had fallen
back from its 1991 peak.
Why does self-employment income inequality tend to be so high? More
generally, what are the underlying factors generating self-employment in-
come inequality? There is little theoretical guidance at present to help
us answer these questions. While Rosen’s (1981) theory of ‘superstars’
explains why the most talented individuals command salaries that are
disproportionately higher than those of their nearest rivals, this theory
cannot explain income dispersion among the bulk of the self-employed,
who earn much more modest incomes. Likewise, while it follows that
if idiosyncratic ability augments the returns of entrepreneurs but not of
employees then the upper tail of the income distribution will be dispro-
portionately full of entrepreneurs (Lazear, 2002), this assumption about
ability is demanding and may not be warranted. And Parker’s (1997b)
parametric model of the self-employment income distribution, which
combines several stylised facts about income dynamics and firm size
growth processes, does not really reveal the underlying causes of income
dispersion.30
Unfortunately, empirical work has not improved our understanding
of the causes of self-employment income inequality much either. Using
income inequality decomposition techniques, Parker (1999b) found that
none of age, gender, marital status, region, occupation, work status, or
educational qualification explained more than a small fraction of UK
self-employment income inequality levels or trends. In contrast, these
variables helped explain a sizeable share of employee income inequality.
Parker concluded that these results reflect the marked heterogeneity of
the self-employed, heterogeneity that increased in Britain in the 1980s
along some unmeasured dimensions. It will be seen below how attempts
to explain average self-employment incomes themselves (rather than their
inequality) have also met with only limited success.
Another issue is the extent to which self-employment facilitates earn-
ings and social mobility. Holtz-Eakin, Rosen and Weathers (2000)
reported that becoming self-employed increases upward earnings mo-
bility for low-income Americans, relative to remaining as employees.
But the opposite was found for high-income Americans, for whom
self-employment entailed downward mobility. Regarding social mobility,
measured in terms of social class, the sociological literature reports mixed
https://doi.org/10.1017/CBO9780511493430.003
results (e.g. compare Mayer, 1975 with Bland, Elliott and Bechhofer,
1978).
Finally, we mention for completeness that even less is known about
the wealth distribution of entrepreneurs. Although there are models of
entrepreneuria wealth transfers and accumulation (Shorrocks, 1988;
Banerjee and Newman, 1993; Parker, 2000), none explains why wealth
distribution takes its observed shape. According to estimates compiled by
Parker (2003b), older self-employed Britons enjoy above-average wealth
holdings, yet only moderate wealth inequality.
1.5.3 Earnings functions

Methods
In this subsection we ask whether it is possible to explain individuals’
self-employment incomes in terms of a few personal and economic vari-
ables. The most popular method for attempting this is estimation of a so-
called ‘earnings function’. Earnings functions were originally developed
by human capital theorists to explain the determinants of employment
earnings. An earnings function typically regresses log earnings, ln y, on
a set of explanatory variables that includes age or experience, a, years of
education, s ch and a vector of other personal and family characteristics,
X. Let u be a stochastic disturbance term, and index individuals by i in
a sample of size n. Employee earnings functions typically take the form
ln yi = β0 + β1 ai + β2 ai2 + β3 s chi + γ Xi + ui i = 1, . . . , n ,
(1.1)
where the βs and γ are coefficients. The β3 coefficient measures the rate
of return to an extra year of education, and for this reason is of particular
interest to human capital theorists.
In principle, it is a straightforward matter to estimate (1.1) using a
sample of self-employed individuals. However, there are several reasons
why one would expect the coefficients of (1.1), and their interpretation,
to differ from those obtained using employee samples.
First, the self-employed rate of return to schooling may differ from
employees’ rate of return. On one hand, entrepreneurial success is likely
to depend on numerous factors other than formal education, implying
that the self-employed β3 will be low relative to its value for employees
(Brown and Sessions, 1998, 1999). Indeed, formal education might
even inculcate attitudes that are antithetical to entrepreneurship (Casson,
2003). On the other hand, if employers demand education from their
workers primarily as an otherwise unproductive screening device (the
‘screening hypothesis’), then the self-employed who do not face this
https://doi.org/10.1017/CBO9780511493430.003
Introduction 21
requirement can be expected to quit education before its rate of return

falls as low as that of employees’. This implies that, if screening occurs,
the self-employed β3 will be relatively high (Riley, 1979).31
Second, the self-employed earnings–age profile may differ from that
of employees. In terms of (1.1), a steeper profile implies a larger β1
and/or a smaller β2 coefficient. There are at least three reasons why
the self-employed earnings–age profile may be steeper than that of
employees. First, the self-employed do not share the returns of their hu-
man (or physical) capital investments with employers, who might smooth
out their costs and returns over employees’ lifetimes. Second, if the self-
employed learn about their abilities over time with the ablest surviving
( Jovanovic, 1982), then one might expect to see any self-employed co-
hort’s average returns increase over time. Third, investment in physical
capital reduces earnings of the young self-employed, while the returns of
that investment accrue to the older self-employed – again implying a steep
earnings-age profile. On the other hand, if employees can shirk on the job,
then employers may respond by steepening employees’ earnings profiles
in order to elicit appropriate worker effort (Lazear and Moore, 1984).
Naturally, no such agency problem arises in self-employment where the
principal is the agent.32
These considerations suggest that earnings functions should be esti-
mated separately for employees and the self-employed. It is not advisable
to estimate an earnings function that pools data on the two occupations,
such as
ln yi = β Xi + δzi + ui (1.2)
(as in, e.g. Amit, Muller and Cockburn, 1995), where zi is an indicator
(dummy) variable for self-employment/paid-employment status. While
(1.2) looks attractive by providing a direct estimate of relative occupa-
tional earnings advantage, it imposes the strong restrictions of identical
rates of return to all of the explanatory variables for both occupations,
which, for the reasons given above, are unlikely to hold.
When estimating earnings functions, it is necessary to avoid selection
bias. Incomes are observed only in the occupation that individuals choose
to participate in; and those who participate in self-employment might
not be a random sample from the population. Rather, they might pos-
sess characteristics that make them particularly favourably disposed to
self-employment. Without correcting for this, the estimated coefficients
of (1.1) will be susceptible to bias. By correcting for this bias, it becomes
possible to ask whether self-employed people (and employees) could im-
prove their lot by switching into the other occupation. A popular practical
way of removing selection bias is Heckman’s (1979) method.
https://doi.org/10.1017/CBO9780511493430.003
Heckman’s method comprises two steps. The first step estimates the
participation equation
zi = ω Wi + vi , (1.3)
where zi is an indicator variable equalling 1 if individual i is self-employed
and 0 otherwise; Wi is a set of explanatory variables, ω is a vector of
coefficients and vi is a disturbance term, with unit variance. Second,
having computed fitted values ẑi from this regression, the ‘Inverse Mills
Ratio’ λi = −φ(ẑi )/(ẑi ) is added to the right-hand side of (1.1), where
φ(·) and (·) are the density and cumulative distribution functions of
the standard normal distribution. Thus one estimates the augmented
earnings function
ln yi = β0 + β1 ai + β2 ai2 + β3 s chi + γ Xi + αλi + ui , (1.4)
where α > 0 implies positive self-selection into self-employment.
Results
Various estimates of earnings functions of type (1.4) have been per-
formed, using a variety of data sources and explanatory variables.33
Even more studies have estimated (1.1), i.e. without correcting for
self-selection. Despite the heterogeneity of the studies to date, several
empirical regularities are detectable. First, rates of return to schooling
tend to be lower for the self-employed than for employees, and not
consistently positive and significant. For example, in their overview of the
literature, Van der Sluis, Van Praag and Vijverberg (2003) documented
an average rate of return of 6.1 per cent for self-employed Americans,
compared with 7–9 per cent for American employees. In other countries,
self-employed rates of return are usually lower still. These findings are
consistent with the idea that entrepreneurial skills are non-academic in
nature. They also do not support the screening hypothesis.34 Indeed, the
screening hypothesis is thrown into further doubt by additional findings
that the self-employed acquire as much – and sometimes even more –
formal education and vocational training as employees (Wolpin, 1977;
Fredland and Little, 1981; Parker, 1999b).
Second, the evidence consistently points to flatter earnings–age profiles
for the self-employed than for employees. While this finding is in line with
the agency cost model of Lazear and Moore (1984) mentioned above,
several caveats to this interpretation can be mentioned. They include
mismeasurement of self-employment experience, and a failure to control
for parental managerial experience, which is known to be very important
for the self-employed (Lentz and Laband, 1990).
Third, few explanatory variables possess much explanatory power in
self-employed earnings functions, resulting in poor goodness-of-fit: it is
https://doi.org/10.1017/CBO9780511493430.003
Introduction 23
common to find R 2 values of 10 per cent and less. This is consistent

with the poor performance of univariate decompositions of self-employed
income inequality measures described earlier. In contrast, employee
earnings functions tend to provide much better fits.35 Subsequently, re-
searchers have tried including a range of non-human capital variables in
an effort to improve the explanatory power of self-employment earnings
functions. These include dummies for industries, disabilities, ill-health
and immigration status. But few have improved matters noticeably in
this respect.
Fourth, on the sample selection issue, there is general disagreement
among the studies that have estimated (1.4) about the direction and sig-
nificance of sample selection effects. There is therefore no clear-cut ev-
idence that the self-employed select into self-employment because they
enjoy a comparative earnings advantage there. In contrast, many stud-
ies find positive selection effects for wage employment, implying that the
self-employed would earn more if they became employees.36 This may
reflect greater human capital (skills and experience) possessed by the self-
employed, which commands a higher return in paid-employment than in
self-employment.
Extensions
Williams (2001) extended the analysis of earnings comparisons to ask
whether self-employed workers would earn more by becoming a fran-
chisee (F) or an independent business owner (I). Williams estimated
selectivity-corrected profit functions in both F and I using data on 14,550
firms taken from the 1987 CBO data set. Selectivity effects were found
to be important, with individuals in F earning less than those in I, after
controlling for personal characteristics. Furthermore, franchisees were
estimated to be substantially worse off if they became independent own-
ers. These findings suggest that franchisees have relatively low average
ability, whereas independent business owners have relatively high aver-
age ability. A policy implication is that legal restrictions on franchising
activities may therefore have deleterious effects on the material wellbeing
of this group of self-employed people.
It is also possible that individuals mix their work hours between self-
employment, S, and paid-employment, E. Work mixing appears to be
more widespread in developing than in developed countries (Sumner,
1981; Vijverberg, 1986).37 If work mixing occurs, then (1.3) and (1.4)
are inappropriate, and one should estimate a composite earnings function
of combined individual wages w i :
ln w i = (1 − ξi )β E XEi + ξi β S XSi + ui , (1.5)
https://doi.org/10.1017/CBO9780511493430.003
where ξi is the proportion of work hours i devotes to S; the βs are vec-

tors of coefficients; and the Xs are matrices of observations on personal
and occupation-relevant characteristics. Because ξi may be endogenously
chosen, (1.5) can be estimated only after an equation for ξi is specified
on a set of exogenous variables. This permits direct comparison of the
two occupations’ β̂ coefficients.
1.6 Some useful econometric models

This section describes several econometric models that have become
widely used for explaining individuals’ decisions to participate in, to enter
and to exit from, self-employment. The non-technical reader may skip
this section, though at the risk of missing some of the subtleties of the
empirical results discussed later in the book.
1.6.1 Occupational choice and probit/logit models

Consider a cross-section of data on n individuals, indexed by i : i =
1, . . . , n. There are two occupations denoted by j : self-employment,
S, and paid-employment, E. Each individual has a vector of observed
characteristics Wi and derives utility Ui j = U(Wi ; j ) + ui j if they work in
occupation j , where U(·; ·) is observable utility and ui j is idiosyncratic
unobserved utility. Define the ‘latent’ variable (i.e. the relative advantage
to S ) as
zi∗ = U(Wi ; S) − U(Wi ; E) − ui E + ui S . (1.6)
If we assume that U(·; ·) is linear, taking the form U(Wi ; j ) = β j Wi , where
β j are vectors of coefficients, then we can write
zi∗ = α + β Wi + vi , (1.7)
where β := β S − β E is a vector of coefficients, α := E[ui S − ui E ] is an
intercept, and where vi := ui S − ui E − α ∼ I I D(0, σ 2 ) is a disturbance
term. Henceforth we shall incorporate the intercept term in Wi as a set
of ones, so β will be treated as the complete set of coefficients.
Individual i chooses self-employment over paid-employment if zi∗ ≥ 0.
Hence define the observable occupational indicator variable

1 if individual i is observed in S, i.e. if zi∗ ≥ 0
zi :=
0 if individual i is observed in E, i.e. if zi∗ < 0
Therefore the probability that an individual with characteristic vector Wi
is drawn from the population and appears in the sample is
Pr(zi = 1) = Pr(zi∗ ≥ 0) . (1.8)
https://doi.org/10.1017/CBO9780511493430.003
Introduction 25
The probit model assumes that the distribution of the disturbance

term vi is normal. Hence Pr(zi = 1) = (β Wi /σ ) and Pr(zi = 0) =
1 − (β Wi /σ ), where (·) is the (cumulative) distribution function of
the normal distribution. The likelihood function is
n
L= (β Wi /σ )zi [1 − (β Wi /σ )]1−zi . (1.9)
i =1
Non-linear methods are needed to maximise (1.9) to estimate the β pa-

rameters (up to a scalar transformation since σ is unknown). It is standard
to normalise σ 2 to unity without loss of generality.
The logit model arises if the distribution function of vi is assumed to
be that of the logistic distribution, in which case (1.8) becomes
exp{β Wi }
Pr(zi = 1) = . (1.10)
1 + exp{β Wi }
The likelihood function can be formed in a similar way as above and β
estimated in a like manner.
In practice, estimates of β tend to be insensitive to whether the probit
or logit assumption is made. Both estimators are widely implemented
on computer software packages. Both dominate ordinary least squares
(OLS) estimation of zi = β Wi + vi (called the linear probability model),
since OLS is an inefficient and heteroscedastic estimator in this context,
and problematically can predict probabilities outside the unit interval
(see, e.g. Maddala, 1983).
It is often of interest to calculate the effects of changes in the kth ex-
planatory variable of Wi , i.e. Wi k , on the probability of self-employment.
These effects are given by

∂ Pr(zi = 1) βk for the linear probability model
:= βk Pr(zi = 1)[1 − Pr(zi = 1)] for the logit model
∂ Wi k βk φ(β Wi ) for the probit model,
(1.11)
where φ(·) is the density function of the standard normal distribution. In
practice, these calculations are usually evaluated at sample means of the
variables.
There are at least three distinct applications of the probit/logit model to
occupational choice. One (I) focuses on the probability that individuals
are self-employed rather than employees. A second (II) asks the different
question of what factors affect the decision to become self-employed, as
opposed to remaining in paid-employment. A third (III) investigates the
decision to leave self-employment, as opposed to continuing in it. All of
these applications can be handled by the probit/logit model: they merely
https://doi.org/10.1017/CBO9780511493430.003
require different definitions of the dependent variable, zi . However, it is

possible to dispute the relative merits of each. For example, Evans and
Leighton (1989b) argued that II is preferable to I, on the grounds that the
latter confounds entry and survival effects. This is because the probability
of being self-employed at time t depends on the probability of switching
into self-employment at some previous time and then surviving until t.
On the other hand, as Wellington (2001) has pointed out, application
II excludes people who are already successfully self-employed, which is
evidently a group of some interest. Low annual switching rates from paid-
employment to self-employment (see chapter 2, section 2.2) also means
that II sometimes suffers from having small numbers of self-employed ob-
servations; and the characteristics of switchers may well be different from
those of non-switchers. There is less disagreement about the importance
of application III, which we discuss separately in chapter 9. We would
simply assert that all three applications generate useful information and
play an important role in applied research.
1.6.2 The structural probit model

Microeconomic theory teaches us that relative prices often affect individ-
ual choices. If this precept is true for occupational choice, then one of the
explanatory variables in the matrix W above should be relative income, or
its logarithm: (ln yi S − ln yi E ). However, we know from subsection 1.5.3
that occupational incomes are endogenous, and prone to selection ef-
fects, so this information needs to be incorporated into the probit model
to obtain efficient and unbiased estimates of the parameters of interest.
The structural probit model is a popular method that accomplishes this.
The first stage of the structural probit model is to estimate selectivity-
corrected earnings functions separately for the self-employed and
employees. Extending the discussion in subsection 1.5.3, and letting M
denote the vector of explanatory variables used in the earnings functions,
one estimates the equations
zi = β Wi + vi i ∈ {S, E} (1.12)
[ln yi S|zi = 1] = γ S Mi + ϑ Sλi S + ui S i∈S (1.13)
[ln yi E |zi = 0] = γ E Mi + ϑ E λi E + ui E i ∈ E, (1.14)
where λi S = −φ(ẑi )/(ẑi ) and λi E = φ(ẑi )/[1 − (ẑi )] are the Inverse
Mills Ratios to correct for selectivity into each occupation. Equa-
tion (1.12) is called the ‘reduced form probit’ and is not of direct interest:
its principal role is to correct for selection bias in the earnings functions
(1.13) and (1.14).
https://doi.org/10.1017/CBO9780511493430.003
Introduction 27
The second stage of the structural probit model generates the pre-
dicted log incomes from both occupations derived from (1.13) and (1.14),
namely ln yi S and ln
yi E . The third stage estimates the ‘structural probit’
model
zi = α[ln
yi S − ln
yi E ] + ω Xi + ṽi , (1.15)
where X = M is a further vector of explanatory variables such that W =

X ∪ M, and where ṽi , ui S , ui E and vi are all assumed to be normally
distributed disturbance terms. If occupational choice depends on relative
financial returns in each occupation, then α > 0. Standard t-statistics
can be used to test the hypothesis that an individual is more likely to be
self-employed the greater is their relative income in self-employment.
Estimation at each stage is easily accomplished using standard econo-
metric software, taking note of the following three points. First, because
the relative income variable in (1.15) has been generated from previ-
ous regressions (i.e. it is a ‘generated regressor’), Newey–West corrected
standard errors should be used. Second, income under-reporting by the
self-employed may bias yi S and hence the coefficients in (1.15). This
problem should be overcome by applying income under-reporting cor-
rections (if available) to the data at the outset. Third, one should ensure
that the disturbances of the earnings equations are normally distributed
(Bernhardt, 1994). This can be achieved by suitable transformations of
the earnings variables in (1.13) and (1.14) (Parker, 2003a). Unfortu-
nately, many previous applications of the structural probit model have
ignored these points, leading to biases of unknown magnitude.38
1.6.3 Extensions to cross-section models of occupational choice

While the models described above are widely used and appropriate for
most applications, it is sometimes necessary to extend them. We describe
three extensions below.
First, although individuals may wish to become self-employed, they
may not have the opportunity to do so. This motivates the use of the bi-
variate probit (BVP) model, which separates individuals’ opportunities to
∗
become self-employed from their willingness to do so. Let z1i be a latent
∗
variable representing the former effect and let z2i represent the latter ef-
fect. Potentially different factors impinge on willingness and opportunity,
suggesting the specifications
∗
z1i = β1 W1i + v1i (1.16)
∗
z2i = β2 W2i + v2i , (1.17)
https://doi.org/10.1017/CBO9780511493430.003
where (v1i , v2i ) are distributed as bivariate normal variates with correla-
tion coefficient ρ: the joint distribution function is (·, ·; ρ). As before,
define zi as an indicator variable, equal to 1 if individual i becomes self-
employed, and 0 otherwise. Evidently
∗ ∗
Pr(zi = 1) = Pr[min(z1i , z2i ) ≥ 0] = (β1 W1i , β2 W2i ; ρ) ,
and analogous to (1.9), the likelihood function is

n
L= (β1 W1i , β2 W2i ; ρ)zi [1 − (β1 W1i , β2 W2i ; ρ)]1−zi .
i =1
To identify this model, it is necessary that W1 = W2 . The researcher must

impose identifying restrictions on the basis of a priori reasoning.
A second extension recognises that there may be more than two occupa-
tions to choose from. For example, Earle and Sakova (2000) studied the
problem of choosing between employer self-employment, own-account
self-employment, paid-employment, or unemployment. Multiple occu-
pational choices can be handled by using multinomial choice models.
Perhaps the most popular model within this class is the multinomial logit
(MNL) model, which can be regarded as an extension to the simple logit
model described above. In this model, individual i must choose between
j = 1, . . . , J alternatives. Define zi j as equal to 1 if i chooses j , and 0
otherwise. Then the MNL model proposes that the probability that i
chooses j is
exp{β j Wi + γ Xj }
Pr(zi j = 1|Wi , Xj ) =
. (1.18)
j exp{β j Wi + γ Xj }
Here Wi is a vector of variables whose values vary across individuals,

whereas Xj is a vector of variables whose values vary across occupations.
The β j coefficients must vary across occupations or else they cannot be
identified. Analogous to the genesis of the simple probit/logit model dis-
cussed in subsection 1.6.1, (1.18) is the probability that results from a
choice problem in which individuals maximise utility across each alter-
native, where the utilities are given by Ui j = β j Wi + γ Xj + ui j . Also as
there, the β j and γ coefficients can be estimated by maximum likelihood,
a procedure that has now been incorporated in many standard economet-
ric software packages.
All of the models discussed so far have assumed that individuals choose
to spend all of their time either in self-employment or in paid-employment.
As noted in subsection 1.5.3, some individuals mix their work time be-
tween multiple occupations. A third extension is needed to handle work
mixing. Let h E denote the proportion of total available work hours that
https://doi.org/10.1017/CBO9780511493430.003
Introduction 29
an individual is observed to spend in paid-employment, and let h ∗E be

a latent variable relating the desired proportion of work hours in paid-
employment to the regressors X. Consider the model

1 if hi∗E ≥ 1
hi E = hi E if 0 < hi∗E < 1
∗
(1.19)

0 if hi∗E ≤ 0
hi∗E = β Xi + ui .
This model can be estimated by double-limit tobit maximum likelihood
(see Vijverberg, 1986, for details). However, in the light of the discussion
in subsection 1.6.2, a limitation of this model is its omission of relative oc-
cupational returns from the set of explanatory variables, X. Incorporating
this extension into (1.19) can be expected to complicate the estimation
substantially – which may account for its absence from the applied en-
trepreneurship literature to date.
1.6.4 Issues arising from the use of time-series and panel data
Time-series applications take a set of T time-series observations on the
self-employment rate, s t , and regress them against a set of explanatory
variables Xt , as follows:
s t = γ Xt + vt t = 1, . . . , T , (1.20)
where γ is a vector of regression coefficients and vt is a disturbance
term.
One rationale for estimating a time-series model is that, unlike
cross-section studies, it becomes possible to analyse trends in self-
employment. Also, the time-series approach can identify determinants
of self-employment rates that are uniform for all or most members of a
cross-section at a given point in time, e.g. income tax variables, interest
rates and other macroeconomic variables.
Prior to the 1990s, the preferred technique for estimating (1.20) was
ordinary or generalised least squares (OLS or GLS: see, e.g., Blau, 1987,
Steinmetz and Wright, 1989). Since then, however, it has become known
that least squares estimators are inappropriate when any of the regressors
in (1.20) are non-stationary.39 The application of least squares estimators
to non-stationary data is known to generate spurious regressions and ren-
ders classical inference invalid (Phillips, 1986). The R 2 goodness-of-fit
measure is no longer informative and t and F statistics can no longer be
used for hypothesis testing. The importance of this point is underlined
by the fact that self-employment rates in the USA, UK, and most other
https://doi.org/10.1017/CBO9780511493430.003
OECD countries appear to be non-stationary in practice (Parker, 1996;

Parker and Robson, 2000). Hence this point appears to be of consider-
able practical importance. When time-series variables are non-stationary,
it is necessary to check whether they cointegrate, i.e. whether there exists
at least one linear combination of the variables (called a cointegrating vec-
tor) that is stationary. If so, there is said to be a long-run (non-spurious)
relationship between the variables. It is then possible to obtain consistent
estimates of the coefficients of that relationship, and to perform appro-
priate hypothesis tests on the coefficients. It is also possible to examine a
dynamic (‘short-run’) error-correction model that describes how agents
behave out of equilibrium.40
Following Parker (1996), we set out below a ‘working guide’ for es-
timating and performing inference on the parameters of (1.20) using
cointegration methods:
1. Check that each variable in (1.20) is non-stationary using unit root
tests.
2. If at least two variables are non-stationary, use a multivariate cointe-
gration estimator to identify the number of cointegration vectors.
3. If there is a unique cointegration vector, test if s t is weakly exogenous.
If so, perform significance tests on each element of γ .
4. Estimate an error-correction model using the cointegrating residuals
v̂t in order to determine the short-run determinants of aggregate self-
employment.
If time-series data are available for the same set of individuals (or firms
or countries) then a ‘panel’ of data is available. Panel data combines
the case-specific variation of cross-section data with the temporal varia-
tion of time-series data, and enables the researcher to control for cohort
and person-specific effects that are absent from a pool of repeated cross-
sections.
There is now a large literature on panel data estimation; below, we
describe only the most widely used ones in applied self-employment re-
search. Suppose that the panel comprises N individuals i observed over T
time periods t. A simple pooled regression model corresponding to (1.20)
is
s i t = γ Xi t + α + vi t , (1.21)
where α is an intercept common to all individuals. A more general spec-

ification allows the intercept to vary across individuals, giving rise to the
fixed-effects model:
s i t = γ Xi t + αi + vi t . (1.22)
https://doi.org/10.1017/CBO9780511493430.003
Introduction 31
The random-effects model is similar to (1.22), except it assumes that the

intercepts are drawn from a common distribution with mean α and vari-
ance σα2 .
Some applications of pooled and fixed- /random-effects models of self-
employment have taken the cross-section units to be countries rather than
individuals. For example, Robson and Wren (1999) and OECD (2000a)
both estimated the following dynamic specification:
ln s i t = αi + β Xi t − γ ln s i t−1 + ω Xi t−1 + ϑt + vi t , (1.23)
where X is a matrix of explanatory variables; αi is a country-specific
intercept term; and ϑt is a set of time dummies. The β coefficients
capture short-run effects of variables on self-employment whereas the
ω coefficients pick up long-run effects. However, as in the time-series
case, least squares estimates of panel models will be biased if any of
the variables in the model is non-stationary. Cointegrated panel tech-
niques are appropriate in this case (see Parker and Robson, 2000, for
details).
N OT E S
1. Of course, these types of surveys can be severely criticised for asking hy-
pothetical questions, without forcing individuals to bear the constraints of
self-employment as they would if they acted upon their declared preferences.
Indeed, much lower rates of serious entrepreneurial intention emerge from
longitudinal analysis of wage and salary workers (Katz, 1990). For this reason,
we will often adopt in this book the standard economic practice of ignoring
studies that report interviewees’ declared preferences, focusing instead on re-
vealed preferences.
2. Intrapreneurship – the practice of entrepreneurship within corporations – is
an emerging field in economics. For an important recent contribution, see
Gromb and Scharfstein (2002).
3. According to CPS data compiled by Bregger (1996), 38 per cent of self-
employed Americans run incorporated businesses. These tend to be businesses
that employ others, which might explain why incorporation rates among new
(and typically small) entrants to self-employment are about half this rate (Evans
and Jovanovic, 1989).
4. Harvey (1995) cites the UK legal case of Young and Woods v. West, whereby
the criteria for a worker being under a contract of service includes the worker
not determining their own hours, not supplying their own materials and equip-
ment, not allocating or designating their own work, not being able to nominate
a substitute to work in their place and not setting their rate of pay (see also
Leighton, 1983).
5. See also Marsh, Heady and Matheson (1981), Casey and Creigh (1988) and
Hakim (1988).
https://doi.org/10.1017/CBO9780511493430.003
6. Firms certainly appear to exercise some discretion about the mode of em-
ployment contract they offer. For example, in her exploration of new laws
penalising companies that misclassify employees as self-employed to avoid
tax payments, Moralee (1998) found that in response to the new laws the
number of ‘employees’ in the construction industry increased sharply while
the number of ‘self-employed’ workers decreased sharply.
7. According to Moralee (1998), 13 per cent of the UK self-employed in 1997
were home-workers, with little change in self-employed home-working taking
place over the 1990s. Moralee also observed that 61 per cent of teleworkers
were self-employed.
8. Williams (2001) argued that franchisees take less risk than independent self-
employed business owners, because of profit sharing arrangements with fran-
chisors and lower demand uncertainty resulting from selling a known prod-
uct. Williams also observed a lower variance of self-employment incomes
among franchisees than non-franchisees in his 1987 CBO sample of full-time
self-employed workers. However, he appears to over-state the case, not least
because exit rates are higher among franchisees than independent business
owners (Bates, 1994).
9. Data limitations can be quite severe, especially for developing and transition
economies. They largely determined the countries selected, and account for
the exclusion of Germany in particular.
10. The primary source for the US entries in table 1.1 is the CPS Monthly House-
hold Labour Force Survey.
11. In his historical study, Phillips (1962) characterised US self-employment as
a ‘shrinking world within a growing economy’. Phillips predicted that self-
employment would eventually serve as a refuge only for older, handicapped
or unproductive workers as a safeguard against unemployment.
12. See Blau (1987), Steinmetz and Wright (1989), Aronson (1991), Bregger
(1996) and Williams (2000).
13. See also Kuznets (1966, table 4.2) who documented declining self-
employment shares between the mid-nineteenth and mid-twentieth century
in Germany, Switzerland, Canada and the UK, as well as in the USA and
France.
14. See also Lin, Picot and Compton (2000), Manser and Picot (2000), Kuhn
and Schuetze (2001) and Moore and Mueller (2002). Self-employment ac-
counted for most overall job growth in Canada in the 1990s, which was con-
centrated among own-account workers.
15. Agriculture was never fully collectivised in Poland, which accounts for its high
rate of self-employment for all workers inclusive of agriculture.
16. According to Blanchflower, Oswald and Stutzer (2001), Poles topped the
list of respondents to a survey of 25,000 people in twenty-three countries
asking whether they would prefer to be self-employed to being a wage worker:
80 per cent responded in the affirmative. Blanchflower, Oswald and Stutzer
concluded that there is no shortage of potential entrepreneurs in the transition
economies.
17. For further discussion about the state of entrepreneurship in Eastern Europe,
see OECD (1998, ch. XIII) and Smallbone and Welter (2001). Tyson, Petrin
and Rogers (1994) and Luthans, Stajkovic and Ibrayeva (2000) describe the
https://doi.org/10.1017/CBO9780511493430.003
Introduction 33
environmental and psychological challenges facing entrepreneurial develop-

ment in transition economies.
18. This includes the Philippines and Indonesia (Le, 1999) and Nepal (Acs,
Audretsch and Evans, 1994). According to Acs, Audretsch and Evans, the
self-employment rate in Nepal in the 1980s reached over 85 per cent, com-
pared with only 3.1 per cent in Botswana.
19. See, e.g., Blau (1985, 1986) and Teilhet-Waldorf and Waldorf (1983). Other
studies reporting higher self-employment than paid-employment incomes
in developing countries include Chiswick (1976) and Bertrand and Squire
(1980) for Thailand; Mazumdar (1981) for Kuala Lumpar; and House,
Ikiara, and McCormick (1993) for Kenya.
20. For example, SBA (1986) estimated that incorporated business owners
earned over twice as much on average as the unincorporated self-employed.
21. Devine (1995) calculated self-employed non-response rates of about
30 per cent (CPS data, 1976–91), compared with 17–19 per cent for wage
and salary workers. In related work, Devine (1994a) estimated non-response
rates of 25 per cent for the incorporated and 40 per cent for the unincorpo-
rated self-employed.
22. Early efforts to separate factor returns from self-employment incomes include
Kravis (1959), Denison (1967), Christiansen (1971) and Chiswick (1976).
See Carrington, McCue and Pierce (1996) for a discussion of the issue with
regard to two US data sets: the CPS and the (Michigan) Panel Study of
Income Dynamics (PSID).
23. Studies finding higher average incomes in paid-employment include Fain
(1980), Becker (1984), SBA (1986), Haber, Lamas and Lichtenstein (1987),
Carrington, McCue and Pierce (1996) and Hamilton (2000). In contrast,
Form (1985), Borjas (1986), Evans and Jovanovic (1989), Ferber and
Waldfogel (1998) and Quadrini (1999) reported an income advantage to self-
employment on average, while Borjas and Bronars (1989, table 1) reported
similar average incomes. For early work see Johnson (1954) and Lebergott
(1964).
24. Carrington, McCue and Pierce (1996) also reported that median hourly
wages and annual incomes in self-employment were substantially and sig-
nificantly more volatile and pro-cyclical than those of employees; and that
while economic downturns were associated with lower wages for all workers,
the decline for the self-employed was some three or four times greater than
that for employees.
25. Robson (1997) attempted to explain the trend in terms of macroeconomic and
fiscal variables and the aggregate self-employment rate. However, causality
can run both ways between average self-employment incomes and the aggre-
gate self-employment rate.
26. To our knowledge, most applications have been based on UK data. See,
e.g., Curran, Burrows and Evandrou (1987), Hakim (1989a), Rubery, Earn-
shaw and Burchell (1993), Goodman and Webb (1994), Meager, Court and
Moralee (1994, 1996) and Storey (1994a).
27. However, part-timers and females are especially likely to be found in the
poorest self-employment groups in Britain (Hakim, 1989a). Of course, it is
possible that the preponderance of self-employed in the lower tail might be
https://doi.org/10.1017/CBO9780511493430.003
partly explained by individuals choosing to reinvest profits directly in the

business rather than consuming them as income.
28. For British evidence, see Rees and Shah (1986), Pissarides and Weber (1989),
Dolton and Makepeace (1990), Jenkins (1995), Meager, Court and Moralee
(1994, 1996) and Parker (1997b, 1999b). For a Dutch example, see Nisjen
(1988).
29. Parker corrected for self-employment income under-reporting and deployed
inequality measures that are robust to mean-preserving measurement error.
However, he did not attempt to separate returns to labour from returns to
capital.
30. Parker’s model ultimately gave rise to the Pearson Type VI distribution as
a parametric form for the density function of the self-employment income
distribution. This is a unimodal and positively skewed distribution, which
seems to fit self-employment income data quite well. It also has the advantage
of being able to handle zero and negative incomes, an important property
given that over 6 per cent of UK self-employees had zero or negative incomes
in 1991.
31. There are caveats to this hypothesis, however. The self-employed may invest in
education as a hedge, or in order to work for others before commencing a spell
of self-employment. Customers, suppliers of credit and government agencies
may also screen self-employed workers. See Fredland and Little (1981) for
further discussion of these points.
32. Flat self-employment earnings–age profiles can also emerge if the self-
employed optimally do little investment in human capital on the job. This and
other possible explanations of flat profiles are explored by Kawaguchi (2003).
33. UK studies include Rees and Shah (1986), Dolton and Makepeace (1990),
Taylor (1996), Burke, Fitz-Roy and Nolan (2000), Clark and Drinkwater
(2000) and Parker (2003a). US studies include Brock and Evans (1986), Gill
(1988), Borjas and Bronars (1989), Evans and Jovanovic (1989), Fujii and
Hawley (1991), Fairlie and Meyer (1996) and Hamilton (2000). Kidd (1993)
is an Australian study; Maxim (1992) and Bernhardt (1994) are Canadian
studies; and de Wit (1993) and de Wit and van Winden (1989, 1990, 1991)
provide evidence for the Netherlands. Earle and Sakova (2000) used Eastern
European data; and Blau (1985) used Malaysian data. For a detailed meta-
analysis, see van der Sluis, Praag and Vijverberg (2003). The latter highlighted
several limitations in the estimation methods used in previous studies, includ-
ing failure to control for ability, endogeneity of schooling, and measurement
error.
34. Fredland and Little (1981) showed that estimates of self-employed rates of
return can be sensitive to the inclusion or exclusion of professionals in the
sample – an important practical point that should be borne in mind when
estimating (1.1) or (1.4) using data on the self-employed.
35. Fredland and Little (1981) suggested that relatively low self-employed earn-
ings function R 2 s can be taken to support the screening hypothesis. However,
low self-employed R 2 s could be caused by a variety of factors, including un-
observed heterogeneity among the self-employed.
https://doi.org/10.1017/CBO9780511493430.003
Introduction 35
36. These findings are consistent with those of Evans and Leighton (1989b),
Headen (1990), Maxim (1992) and Hamilton (2000). Of these studies,
Hamilton provides the most detailed evidence of the gains to the self-
employed from switching to paid-employment.
37. For example, Vijverberg (1986) reported that some 20 per cent of respondents
in a 1976 Malaysian survey data set performed work mixing. The author’s
calculations using 1994/5 FES data revealed that only 1.4 per cent of UK
workers mixed paid-employment and self-employment. The proportion of
self-employed people doing work mixing (3.3 per cent) exceeded the propor-
tion of employees doing so (1.1 per cent).
38. An alternative to the three-step approach is structural estimation of the param-
eters of an underlying utility maximisation model (Brock and Evans, 1986).
However, this approach is complicated, not obviously superior and has not
been widely used.
39. In simple terms, a non-stationary process is one in which there is no mecha-
nism forcing values of the series to revert to the mean.
40. See, e.g., Harris and Sollis (2003) for an introduction to cointegration
analysis.
https://doi.org/10.1017/CBO9780511493430.003
https://doi.org/10.1017/CBO9780511493430.003

Introduction Entrepreneur

Uploaded by

Copyright:

Available Formats

Introduction Entrepreneur

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Introduction Entrepreneur

Uploaded by

Copyright:

Available Formats

1 Introduction

Self-employment is unquestionably the oldest way by which individuals offer and

Entrepreneurship is increasingly in the news. Governments all over the

contributes to the elusiveness of the entrepreneur alluded to by William

1.1 Aims, motivation and scope of the book

their working lives, about two-fifths of the American workforce will

1.2 Structure of the book

self-employment. The distinction between these two concepts explains

1.3 Definition and measurement issues

The self-employed are often taken to be individuals who earn no wage

Fourth, there appears to be a ‘grey area’ between paid-employment and

1.4 International evidence on self-employment rates

Table 1.1 Aggregate self-employment rates in some selected OECD

1960 1970 1980 1990 2000

a substantial difference to measured self-employment rates in most (but

1.4.1 The OECD countries

Table 1.2 Aggregate self-employment rates in some selected transition

1980 1990 1992 1994 1998/99b

Poland (all workers) 25.44 27.17 22.44

Source: UN Yearbook of Labour Statistics, various issues.

Table 1.3 Aggregate self-employment rates in developing countries,

1960sb 1970sb 1980sb 1990sb

1.4.2 The transition economies of Eastern Europe

1.4.3 Developing countries

W. Arthur Lewis (1955, p. 182): ‘Economic growth is bound to slow

1.5 Self-employment incomes and income inequality

1.5.1 Incomes and relative incomes

denoting sample mean values by overbars, calculate

International evidence on relative average self-employment

Conversely, however, Hamilton may have over-stated the income differ-

1.5.2 Income inequality

self-employment itself expanded substantially (see subsection 1.4.1). The

1.5.3 Earnings functions

requirement can be expected to quit education before its rate of return

common to find R 2 values of 10 per cent and less. This is consistent

ln w i = (1 − ξi )β E XEi + ξi β S XSi + ui , (1.5)

where ξi is the proportion of work hours i devotes to S; the βs are vec-

1.6 Some useful econometric models

1.6.1 Occupational choice and probit/logit models

The probit model assumes that the distribution of the disturbance

Non-linear methods are needed to maximise (1.9) to estimate the β pa-

require different definitions of the dependent variable, zi . However, it is

1.6.2 The structural probit model

where X = M is a further vector of explanatory variables such that W =

1.6.3 Extensions to cross-section models of occupational choice

To identify this model, it is necessary that W1 = W2 . The researcher must

Here Wi is a vector of variables whose values vary across individuals,

an individual is observed to spend in paid-employment, and let h ∗E be

OECD countries appear to be non-stationary in practice (Parker, 1996;

where α is an intercept common to all individuals. A more general spec-

The random-effects model is similar to (1.22), except it assumes that the

environmental and psychological challenges facing entrepreneurial develop-

partly explained by individuals choosing to reinvest profits directly in the

You might also like

where X = M is a further vector of explanatory variables such that W =

To identify this model, it is necessary that W1 = W2 . The researcher must