See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/24084099

Competition and entrepreneurs' human capital in small business longevity

and growth

Article in Journal of Development Studies · February 2006

DOI: 10.1080/00220380600742050 · Source: RePEc

CITATIONS READS

64 450

1 author:

Taye Mengistae
World Bank
45 PUBLICATIONS 1,083 CITATIONS

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Firms Performance in SSA View project

All content following this page was uploaded by Taye Mengistae on 23 January 2014.

The user has requested enhancement of the downloaded file.

Journal of Development Studies,
Vol. 42, No. 5, 812–836, July 2006

Competition and Entrepreneurs’ Human

Capital in Small Business Longevity and
Growth
TAYE MENGISTAE
The World Bank, Washington DC, USA

Final version received May 2005

ABSTRACT An analysis of data on a sample of small-scale manufacturers shows that a business

is less likely to survive and grows slower the smaller the average price–cost margin in the industry
in which it operates. The probability of survival is also smaller in import competing industries. So
is the mean growth rate among survivors. We interpret this as evidence that small businesses are
less likely to survive and grow slower in industries where the pressure of competition is stronger.
Given competitive pressure and establishment characteristics, the probability of business survival
and the expected growth rate conditional on survival both increase with entrepreneurial human
capital. This is in the sense that the probability of business survival increases with the number of
years of schooling and the number of years of business experience of the entrepreneur as does the
expected growth rate conditional on survival. These results are consistent with another finding
that unobservable influences on business hazard are correlated with those on growth. As a result,
the effect of competition and entrepreneurial human capital on the growth of survivors would be
biased for the effect of the same variables on the expected growth rate of a startup.

I. Introduction
This paper is an empirical investigation of the role of industry competition and
entrepreneurial human capital in small business survival and growth in a low income
economy. Not all of a given cohort of startups would be expected to survive and
prosper in the long run of a competitive industry. Some will inevitably contract or
shutdown altogether in spite of operating in circumstances similar to those of the
more successful members of the cohort in terms of policy environment,
macroeconomic setting, or line of industry. What is it that distinguishes those that
survive and grow from failures? This has been the subject of a small but highly
influential set of theoretical papers and a sizeable body of empirical work on firm
dynamics in advanced economies. One of the better known of the main regularities
the empirical literature has established so far is that smaller businesses are more
likely to fail than larger businesses, but will also normally grow faster when they

Correspondence Address: Taye Mengistae, The World Bank, 1818 H Street, NW, Washington DC, 20433
USA. Email: tmengistae@worldbank.org

ISSN 0022-0388 Print/1743-9140 Online/06/050812-25 ª 2006 Taylor & Francis

DOI: 10.1080/00220380600742050
 Competition, Human Capital and Small Business Longevity 813

survive.1 The age of a business also appears to influence its dynamics in a similar
fashion: younger establishments are less likely to survive, but, among survivors, the
expected growth rate diminishes with age.2 A second common finding in the
literature is that there are significant inter-industry differences in survival
probabilities as well as in the pace of growth among survivors. Thirdly, many
studies show that, for reasons that are not always clear, social and demographic
characteristics of businesses owners such as schooling and ethnicity are strongly
correlated with the longevity or growth performance of the enterprises they run.
Some of these findings have been confirmed in studies based on data from low
income developing economies including some from Africa. In the context of low
income economies, the result that smaller or younger businesses grow faster is
reported in many studies including, for example, McPherson (1996), Ramachandran
and Shah (1999), and Gunning and Mengistae (2001). Almost every paper in this
category detects significant industry effects in growth rates, while some report that
growth is also influenced by the education, gender, or ethnicity of business owners.
Unfortunately, with the exception of McPherson (1996), all firm level growth
equation estimates for sub-Saharan Africa that we have come across so far have been
based on data on surviving businesses only. This makes the interpretation of the
coefficients of the equations rather problematic. Usually our interest is in the
influence of a variable on the expected growth rate of a random draw from the entire
population of businesses, rather than in the mean growth rate of the more successful
segment of it.3 Yet, estimates of the coefficient of the variable in a regression of
growth of survivors would be biased for its effect on the population mean growth
rate if unobservable factors influencing business growth were correlated with
unobservable influences on the probability of business survival. In the presence of
correlation of this kind, coefficient estimates of the growth equation of survivors or
incumbents would be biased for corresponding parameters of the growth equation of
potential startups or fresh entrants.4
Using data from a two-wave repeat sample survey of formal manufacturing
businesses in Ethiopia in the mid-1990s, this paper offers an integrated analysis of
small business growth and survival in the context of a low income economy. The
analysis has three specific objectives. The first of these is to try to contribute to the
understanding of the economic content of industry effects in firm growth and
survival by assessing the role of inter-industry differences in competitiveness in firm
dynamics. Although mean growth rates and survival probabilities are known to vary
significantly between industries in developed and developing economies alike, we are
not aware of attempts at interpreting their economic content in a low income
developing economy. The second objective is to assess the role of entrepreneurial
human capital in business success in as far as it can be measured by the number of
years of formal schooling of business owners and the length of their business
experiences. The third objective is to obtain estimates of the effects of competition,
entrepreneurial human capital, and standard controls such as initial size and initial
age on the expected growth rate of business startups as opposed to the growth rate of
incumbents or survivors. The goal here is to correct the estimates of parameters of
the growth of survivors for possible attrition bias due to business closures. In the
process of attaining this last objective, we hope to provide a richer analysis of the
role of competition and entrepreneurial human capital in business success than could
814 T. Mengistae

be achieved in separate investigations of the determinants of business hazard and of

factors governing the growth of survivors. After all, the failure or closure of a
business often seems to be an extreme outcome of a process of growth (or
contraction).
Briefly, our findings are as follows. First, initial size, lines of activity, and
entrepreneurial human capital are major determinants, not only of business survival,
but also of growth among survivors: (a) smaller business are less likely to survive,
but grow faster when they survive; (b) survival probabilities are higher in some
industries than in others, as are growth rates; and (c) businesses run by entrepreneurs
with greater schooling are more likely to survive and have higher average growth
rates conditional on survival.
With the exception of initial size, all factors that increase with the probability of
survival also increase with the average growth rate given survival. This leads us to
the second result, which is that the effect of a factor on growth that we might
estimate on data on survivors only would be biased for the effect of the same factor
on the expected growth rate of start-ups. In particular, the paper shows that the size
effect on growth among survivors overestimates the size effect for an entire entry
cohort.
Our third result relates to the role of inter-industry differences in competitiveness
in explaining industry effects in business survival and growth. This is the most
distinctive of our findings relative to existing empirical studies of firm dynamics in
developing countries. Using two related but distinct indicators of industry
competitiveness, we find that both the probability of business survival and the
average business growth rate are lower in more competitive industries. This is in as
far as the probability of survival and the expected growth rate of a startup are higher
in industries in which the average price–cost margin is higher and in those not
exposed to competition from imports.
The rest of the paper is organised as follows. Section II discusses the specifications
of the business growth equations and hazard functions that we estimate in the paper.
Section III describes our data in detail. We then report estimation results in Section
IV and conclude in Section V.

II. Empirical Specifications of Business Growth and the Hazard Business Closure
Consider the population of businesses from which our sample was drawn at
the beginning of the observation period, that is, at the time of the first wave of the
survey. Let y1i be the size of business i observed at the time and as measured by the
log of the business’ fixed assets, number of employees on its payroll, or annual sales
revenue for the year ending then. Let y2i be the size of the same business at the time
of the re-visit survey. We assume that the growth rate Dyi ¼ y2i 7 y1i of the business
over the period of observation is linear in a set of parameters d and an unobservable
i.i.d random variable ei so that Dyi ¼ zi d þ ei, where zi is a set of observable business
or industry characteristics orthogonal to ei. The growth rate, Dyi, is observed only for
survivors. Let Survivori be a dichotomous variable equal to unity if i operates
throughout the period of observation, but is zero if the business ceases to operate at
some point during the same period. We assume that the variable Survivori is also
linear in a set of parameters b, a set of observable business or industry characteristics,
 Competition, Human Capital and Small Business Longevity 815

xi, and an unobservable i.i.d random variable, ui, assumed to be orthogonal to xi.
The determination of growth can then be described as
Dyi ¼ zi d þ ei if Survivori ¼ 1; ð1Þ

but is unobserved if Survivori ¼ 0, where

Survivori ¼ xi b þ ui : ð2Þ

The estimator of d we would obtain by applying OLS to Dyi ¼ zi d þ ei on the sub-

sample of survivors only would be biased if e and u are correlated, the magnitude of
the bias being larger the larger is the correlation coefficient.5 However, assessments of
the practical importance of the bias do vary among studies. For example, Evans (1987)
and Hall (1987) find the bias to be negligible while Dunn and Hughes (1994) do not. In
a developing country context, McPherson (1996) finds no evidence of attrition bias in
the estimation of growth equations in informal firms in Southern Africa. The most
common approach to the sample selection problem is to assume that both e and u are
distributed standard normal and estimate the two equations jointly using Heckman’s
two-stage estimator, whereby we obtain the estimate b ^ of b in the selection equation
(2) by maximum likelihood in the first stage, and then apply OLS to
^ þ ni
Dyi ¼ zi d þ seu lðxi bÞ ð3Þ
f½xi b
where seu is the covariance of u and e, and lðxi bÞ ¼ F½x i b
is the inverse Mills ratio
with f(.) and F(.) as the standard normal probability density function and standard
normal distribution function respectively. We will use this approach in this paper
too. We do note, though, that estimating (3) is equivalent to estimating
Dyi ¼ zi d þ suZ l2 ðxi b ^ Þ þ ni , where seZ is the covariance of u and the error term, Z,
2
f½xi b2 
of a probit of business closure, and l2 ðxi bÞ ¼ 1F½x i b2 
is the corresponding business
^
closure hazard rate replacing lðxi bÞ as the selectivity term.6 Now, there does not
seem to be any reason why the hazard function should assume this particular form
other than the need to generate results that can be compared with estimates provided
in other papers. We therefore present in addition results based on the estimation
of more flexible functional forms. Specifically, we have estimated two commonly
used (proportional) hazard models—one parametric and one semi-parametric.
Underlying these models is the assumption that the probability, y, that business i
will have closed before being aged years ti is a product of an age effect, y0(ti), and
the combined effect, y1 ð~ xi Þ, of other factors, x~i , and can be written as
yðti ; x~i Þ ¼ y0 ðti Þy1 ð~
xi Þ. Further assuming that y1 is log linear leads to the semi-
parametric specification
yðti ; x~i Þ ¼ y0 ðti Þexpð~
xi bÞ; ð4Þ
which we estimate by Cox regression. As a parametric alternative to this, we also
estimate by maximum likelihood the Weibull proportional hazard model

yðti ; x~i Þ ¼ ptp1

i expð~
xi bÞ ð5Þ

where p 4 0 and we assume that p 6¼ 1 so that the baseline hazard increases or

decreases as the business gets older depending on whether p 5 1 or p 4 1.7
816 T. Mengistae

III. Data and Variables

Our data come from a repeat small sample survey of manufacturing establishments that
was carried out in the mid 1990s in Ethiopia. The survey was confined to registered
businesses located in the city of Addis Ababa and was conducted in two waves. The
first wave took place during September to October 1993 and covered a random
selection of 220 manufacturing establishments. Of these, 190 were owner-managed (or
entrepreneurial) businesses employing less than 200 people, the rest being medium to
large scale state owned enterprises. The same 220 establishments were then revisited in
October 1995. The survey instrument used in both waves was a written questionnaire
addressed to business managers and accountants and seeking information on a wide
range of financial and production variables including sales, fixed assets, material
purchases, inventories, employment and business organisation and history.
In this paper, we analyse data on the 190 private sector small businesses of the 1993
sample. Not surprisingly, some of these had closed down by the time of the 1995 revisit.
Of the remaining, some had grown quite fast while others had contracted sharply.
What factors could explain this differential in performance? What observable
characteristics distinguish the businesses that expanded from those that contracted,
or from those that closed down? Table 1 provides the definition and descriptive
statistics of the main variables that should matter according to the literature on firm
dynamics, along with those of the variables we seek to explain, namely, growth and
survival status. We can group potentially explanatory variables into three broad groups
of characteristics, namely, those of (a) the business establishment; (b) the industry in
which the establishment operates; and (c) the owner of the establishment (or the
entrepreneur). Apart from the outcome variables of survival and growth, the
establishment-level characteristics of interest are beginning-of-period size, initial age,
and sector of industry. As can be seen from Table 1, the 190 businesses were drawn
predominantly from four light consumer goods industries, namely, food and beverages
(11%), textiles (14%), garments and footwear (24%), and furniture (22%), the balance
coming from other light consumer goods industries.
Ideally, one would measure the size of an establishment by its annual sales revenue
or by its productive capacity, as indicated by the value of its fixed assets.
Unfortunately, neither of these is practical in this case for lack of appropriate price
deflators, and because of well-known problems in the measurement of depreciation
and the valuation of equipment of different vintages. We have consequently used the
number of employees observed at the beginning or the end of the accounting year as
the scale variable. At the beginning of 1993, businesses in our sample ranged in size
from as little as two employees to as high as 167, with an average size of a little over
15 workers. At the time, the average establishment had been in operation for just
under 13 years, observations ranging from a one-year old business to one that had
operated for nearly 50 years.
We measure business growth as the average annual employment growth rate over
the three-year period, defined as the log difference between employment at the
beginning of 1993 and employment at the end of 1995 divided by three. On this
measure, the average business grew at a rate of 2 per cent a year if we considered only
those that survived beyond the 1995 survey. Again, this conceals sharp contrast
between the extremes of a business that has contracted by half and one that has
 Competition, Human Capital and Small Business Longevity 817

Table 1. Descriptive statistics

Variable Definition Obs. Mean SD

End-of-1992 employment Number of employees 190 15.3 20.4

at the end of 1992
End-of-1992 age of Number of years at the 190 12.7 10.4
the business end of 1992 since the
business was set up
Employment growth Average annual rate of growth 158 0.16 0.31
rate (1993–1995) (of employment) during 1993–95
Closed by the end of 1995 Dummy ¼ 1 if the establishment 190 0.17 0.38
was shut down by the end of 1995
Business experience of Number of years since the 190 12.8 10.3
the entrepreneur business owner started her/his
first ever business by 1993
Schooling of the Number of years of schooling 189 8.2 4.83
entrepreneur (years) the business owner
Industry average price-cost Industry (sample) average of the 190 52.7 9.83
margins in 1993 ratio of gross profits to sales
in 1993 (%)
Import competing in 1993 Dummy ¼ 1 if main product of the 190 0.23 0.42
business was Import
competing in 1993
Food and beverages Dummy ¼ 1 if producing mainly 190 0.11 0.31
food or beverages
Textiles Dummy ¼ 1 if producing 190 0.14 0.35
mainly in textiles
Garments and footwear Dummy ¼ 1 if producing mainly 190 0.24 0.43
garments/footwear
Furniture Dummy ¼ 1 if mainly producing 190 0.22 0.42
furniture

expanded by 130 per cent. The average growth rate over the entire sample of 190
would be much lower than 2 per cent when we note that 32 of the 190 establishments
surveyed in 1993 had closed down or ceased all manufacturing operations by 1995.
One of the hypotheses we test is that the average growth rate decreases with initial
size and initial age, while the probability of survival increases with both. Theoretical
explanation for size effects in business survival and growth is provided by well-
known models of market selection. The basic common underlying idea of these
models is that the evolution of a competitive industry is driven by a sorting process
arising from the cost (or productivity) heterogeneity of producers, that is, from the
fact that no two producers can produce the same level of output from a given mix of
traded inputs. Even under pure competition, some firms will always produce more
with a given mix of traded inputs than others, either because they are always better
run as in Lucas (1978), or are better endowed with some other source of quasi rent
such as advantageous location as in Jovanovic (1982), or superior technical
knowledge as in Jovanovic and MacDonald (1994). The resulting inter-firm
differences in unit costs prompts a process of selection whereby only producers
that exceed some productivity threshold can enter an industry (Lucas, 1978;
Lippman and Rumelt, 1982) or survive and grow in it (Jovanovic, 1982; Hopenhayn,
1992; Pakes and Ericson, 1998). Regardless of whether the source of heterogeneity is
818 T. Mengistae

unmeasured entrepreneurial human capital or some other source of quasi rent, one
manifestation of market selection is a positive size effect on the probability of
business survival (Jovanovic, 1982; Hoppenhayn, 1992; Pakes and Ericson, 1998),
and a negative size effect on the expected business growth rate.
Table 2 also shows that there are industry effects in the determination of growth
rates and of survival probabilities, which we need to control for in order to identify
size and age effects.
We see from the same table that a business in our sample was far more likely to fail
between the beginning of 1993 and the end of 1995 in the food and beverages
industries or in garments than if it were in furniture making. The sample failure rate
was one in five over the three-year period in food and beverages, and 11 per cent in
garments. The average growth rate among survivors was also highest in furniture
making (at 11 per cent a year) while the average survivor was actually cutting back
production in food and beverages and in garments (at rates of 3 per cent a year and 2
per cent a year respectively). In theory, these industry effects could be entirely due to
differences in the size or age distribution of businesses across industries. However,
this does not seem to be the case here. First, the average business age does not seem
to differ much across the four industries, although it seems to be higher for
businesses outside of the four. Secondly, the average garments or footwear producer
is larger – with nearly 18 employees – than the average furniture maker (13
employees), which we would not expect to be the case if the difference between
survival probabilities in the two sectors was entirely an industry effect.
Where could the observed industry effects in survival and growth have come from
then? One possible source is inter-industry difference in competitiveness, of which a
common indicator is the industry average price–cost margin. The more competitive
is an industry, we assume, the lower is the average price–cost margin and the higher
the business failure rate in it. One would also expect average business growth rates to

Table 2. Inter industry differences

Sample mean for

Food and Garments and
Variable beverages Textiles footwear Furniture Other

Employment growth 70.03 70.04 70.02 0.11 0.03

rate (1993–1995)
Closed by 1995 0.20 0.04 0.11 0.02 0.06
End-of-1992 employment 14.25 10.11 17.58 13.05 33.94
End-of-1992 age of 13.90 13.81 11.24 12.81 16.65
the business
Business experience of 13.90 14.19 11.33 12.57 16.65
the entrepreneur
Schooling of the 9.00 6.00 7.00 10.00 11.00
entrepreneur (years)
Import competing in 1993 0.10 0.37 0.24 0.12 0.29
Industry average 71.27 35.58 57.40 57.08 48.92
price-cost margins in 1993
Obs. 20.00 27.00 45.00 42.00 56.00
 Competition, Human Capital and Small Business Longevity 819

be smaller in more competitive industries. We use the industry sample mean of the
percentage ratio of gross profits to annual sales for 1993 and 1995 as our proxy for
the industry average price–cost margin.
Although imports have always been a major source of competition for all
manufacturers of consumer goods in Ethiopia, actual competition from this source
seemed to have been exceptionally strong in the garments and footwear industries at
the time of the survey. At that time a sudden surge in imports that a recent
devaluation of the national currency was believed to have induced was being blamed
by industry organisations for widespread closure of many businesses in those
industries. This hints at a second proxy for industry competitiveness, namely,
whether or not the industry in question is import competing. From what we read in
Table 2 it seems that both import competition and average price–cost margins are
reasonably good indicators of industry competitiveness. For example, the fact that
the business failure rate was smaller among furniture makers than in the garments
and footwear industries is consistent with the proportion of import competing
producers being lower in furniture making. The ranking of industries by failure rates
also seems to match ranking by average price–cost margins reasonably well.
As the third category of covariates of business dynamics, two of the
entrepreneurial characteristics on which data were collected in the 1993 survey have
a bearing on subsequent establishment survival and growth. These are the level of
education of the entrepreneur as indicated by years of schooling, and the length of
her business experience as indicated by the number of years since she started her first
ever business. Although there does not appear to be a great deal of inter-industry
difference in the sample in the average length of the entrepreneur’s business
experience (Table 2), average entrepreneurial schooling seems to be lower for the
textiles and garments industries. This could also be part of the reason for business
survival rates being lower for those industries in the sample. Though primarily
intended to explain size effects in firm dynamics, the market selection models of
Jovanovic (1982) and Hopenhyan (1992) also provide an indirect theoretical
rationale for the inclusion of entrepreneurial human capital as covariates of business
survival and growth. If it is indeed the case that differences in unmeasured
entrepreneurial human capital generate size effects in business growth or survival, as
these models imply, then it seems logical to expect that observed differences in
human capital variables such as those registered in the number of years of schooling
should likewise influence the prospects of business success. There is indeed some
evidence that this is the case in some economies. For example, Bates (1990) reports
that the probability of small business survival increases with the level of schooling of
entrepreneurs in the US. McPherson (1996) and Ramachandran and Shah (1999)
similarly find positive association between entrepreneur’s schooling and business
growth rates (conditional on survival) in a number of African countries.

IV. Estimation Results

We now turn to formal tests of the propositions that inspection of the
descriptive statistics of Tables 1 and 2 seems to support. We begin with an
analysis of the influence of the three sets of variables on business growth. We report
in Table 3 results of the estimation of alternative specifications of the first equation
820 T. Mengistae

of (1) by least squares using data on survivors only. The dependent variable in all
eight columns of the table is the average annual employment growth rate between
the beginning of 1993 and the end of 1995. All reported t-values are based on robust
standard error estimates. The estimated coefficients of initial employment are always
negative and statistically significant. We therefore conclude that, controlling for
business age, and line of activity, smaller firms grow faster. At the point estimates
reported in the first column, for example, a business would grow 9.2 percentage
points faster than another that is twice as large. The evidence is also that this size
effect is uniform across industries since none of the coefficients of the interaction of
initial size with industries dummies is statistically significant.
Similar findings in terms of size effects in the growth of survivors are almost
universally reported in the literature. In particular our results confirm those reported
in McPherson (1996), Ramachandran and Shah (1999) and Gunning and Mengistae
(2001) for firms in Sub-Saharan Africa. The way size effects have been reported in
McPherson (1996) and Gunning and Mengistae (2001) does not facilitate comparison
of the magnitude of our point estimates of the effects with theirs. Both these papers
nonetheless report statistically significant negative size effects. So does the paper by
Ramachandran and Shah (1999), in which the estimated growth equation is of the
same functional form as that in Table 3 here. We obtain a stronger (negative) size
effect than the three percentage point fall in the average employment growth that
Ramachandran and Shah estimate to follow from a doubling of initial firm size.
Contrary to what is reported in the three papers on small business growth in
Africa, and indeed many papers in the empirical literature on firm dynamics in
developed economies, there is no evidence in our data that younger establishments
grow faster once we control for initial size. It is possible that this is merely a
manifestation of non-linearity in the age effect, meaning that most businesses in the
sample fall in too low an age range for the inverse relations to set in.8 Alternatively,
it could be that the age effect in the three papers on African firms cited here mix up
calendar time effects with life cycle effects unlike the case in Table 3 here, where
calendar time is fixed at the 1993–1995 interval. Yet another alternative explanation
is that we are using the wrong age variable. The relevant time variable could be the
overall business experience of the business owner or entrepreneur rather than
business age itself. This explanation is suggested by the fact that the coefficient of the
length of business experience of the owner in the employment growth equation is
negative while that of business age is positive. However, in none of the specifications
that we estimate in the table is the coefficient of the business owner’s experience
statistically significant. We therefore tend to infer that there is no evidence of age
effect in business growth conditional on survival (once we control for initial size).
We draw this conclusion while controlling for the other entrepreneurial human
capital variable, namely, schooling. We find the coefficient of this variable to be
positive and statistically significant. McPherson (1996) and Ramachandran and
Shah (1999) report the same result. According to McPherson (1996), for example, a
business run by an entrepreneur who had completed secondary schooling in
Zimbabwe would grow by 3.8 percentage points faster a year than an otherwise
comparable business run by someone with primary education only. At our point
estimates, a business would grow one percentage point faster, with every additional
year of schooling of the entrepreneur. Since secondary schooling normally lasts four
Table 3. OLS estimation of employment growth equation for survivors only Dependent variable: average annual employment growth rate
(1993–1995)

(1) (2) (3) (4) (5) (6) (7) (8)

Log(end-of-1992 employment) 70.092 70.101 70.102 70.095 70.098 70.100 70.095 70.097
(1.91) (3.48) (3.43) (3.35) (3.32) (3.36) (3.32) (3.27)
Log(end-of-1992 employment)
X Food and beverages 70.058
(0.48)
X Textiles 70.064
(0.80)
X Garments and footwear 0.006
(0.09)
Log(end-of-1992 age of establishment) 0.014 0.015 0.016 0.017 0.017 0.016 0.017 0.017
(1.25) (1.33) (1.37) (1.51) (1.48) (1.40) (1.51) (1.50)
Log(business experience of the entrepreneur) 70.007 70.008 70.008 70.009 70.009 70.009 70.009 70.009
(0.62) (0.71) (0.71) (0.81) (0.78) (0.74) (0.81) (0.79)
Number of years of schooling of entrepreneur 0.011 0.017 0.014 0.014 0.014 0.014 0.014 0.014
(1.97) (2.97) (2.68) (2.73) (2.70) (2.55) (2.59) (2.59)
Number of years of schooling of entrepreneur
X Food and beverages 70.003
(0.45)
X Textiles 70.007
(0.80)
X Garments and footwear 70.009
(1.33)
Establishment price-cost margins in 1993 0.003 0.003 0.001
(0.66) (0.53) (0.50)
(continued)
Competition, Human Capital and Small Business Longevity 821
 Table 3. (Continued)
822 T. Mengistae

(1) (2) (3) (4) (5) (6) (7) (8)

Industry average price-cost margins in 1993 70.000 70.002 70.000 70.002

(0.02) (0.34) (0.05) (0.35)
Import competing in 1993 70.118 70.119 70.114 70.117
(2.03) (1.99) (1.95) (1.94)
Food and beverages 0.055
(0.17)
Textiles 0.009
(0.04)
Garments and footwear 70.079
(0.40)
Furniture 0.020
(0.21)
Constant 0.040 0.003 0.019 70.010 0.138 0.002 70.025 0.122
(0.23) (0.04) (0.05) (0.12) (0.33) (0.00) (0.29) (0.29)
Observations 148 148 148 148 148 148 148 148
R-squared 0.15 0.13 0.12 0.14 0.14 0.12 0.14 0.14
 Competition, Human Capital and Small Business Longevity 823

years in the Ethiopian education system, this would mean a four-percentage point
pay premium of secondary school graduates over those who have only completed
primary schooling.9 This is not far from the estimate for Zimbabwe.
The fact that there appear to be substantial differences between industries in
average schooling levels in our case makes this factor one source of inter-industry
differences in average growth rates of surviving businesses. However, there is no
evidence that the ‘rate of return’ to business owners’ schooling varies by industry.
We can see this from the fact that none of the coefficients of the interaction terms
between years of schooling and industry dummies in column two of the table are
statistically significant.
The first column of Table 3 includes industry dummies. Again, however, none of
these has a statistically significant coefficient estimate. Coupled with lack of
statistical significance of estimated coefficients of the firm level and the industry
average price–cost margins in the other columns, this would seem to suggest that
there are no significant industry effects in expected business growth rates conditional
on survival. We note, though, that this bears sharp contrast to the strong industry
effects in survival probabilities to be reported later in the paper. We note also that
the estimated coefficients of one of our indicators of industry competitiveness,
namely, whether or not an industry is import competing is negative and statistically
significant in the growth equation. The average surviving business in an import
competing industry grows slower by about 12 percentage points a year. Which
results of Table 3 carry over to the expected growth rate of new startups? Or, could
some or all of these results suffer from selectivity bias arising from possible
correlation between unobserved determinants of survival with those of the expected
growth rate of new entrants? It is possible that, in spite of the lack of statistically
significant correlation between the industry average price–cost margin and the
expected growth rate of survivors, the unconditional (on survival) growth rate in fact
increases with industry profitability simply because the probability of survival does.
We therefore turn in Table 4 to the question of how the determination of the
expected firm growth rate conditional on survival compares with that of the expected
growth rate of a startup. Each pair of columns in this table is the result of joint
estimation of equations (1) and (2) by maximum likelihood rather than by applying
OLS to equation (3) having first estimated equation (1). Because of the collinearity of
industry dummies with industry characteristics, we estimate both the growth
equation and the selection equation with the two sets as alternatives rather than as
co-determinants. We use industry characteristics in the first four columns, and
industry dummies in the last four. A comparison of the two sets of results shows no
qualitative difference between them in terms of size and age effects in business
growth and survival, and of the effect of entrepreneurial human capital on the same.
We will therefore confine the discussion to the columns in which we have used
industry characteristics as regressors.
The key result here is that the coefficient of the survival rate, lðxi bÞ,^ is positive
and statistically significant in all four specifications in the table. This implies that the
unobservable determinants of survival are positively correlated with unobserved
influences on growth. The estimates of the effects of observables on the growth of
survivors that we report in Table 3 are therefore biased for the corresponding effects
on the unconditional (on survival) mean growth rate. Thus we see from a
Table 4. Maximum likelihood estimation of Heckman selection specification of employment growth equation Dependent variable of employment
growth equation ¼ annual average employment growth rate (1993–1995) Dependent variable of selection equation ¼ not closed by 1995

(1) (2) (3) (4)

824 T. Mengistae

Growth Selection Growth Selection Growth Selection Growth Selection

Log(end-of-1992 employment) 70.055 0.045 70.080 0.351 70.052 0.421 70.066 0.321
(1.74) (0.45) (2.69) (2.30) (1.65) (2.03) (2.12) (1.54)
Log(end-of-1992 age of establishment) 0.015 70.046 0.007 70.001 0.007 70.002 0.012 70.029
(1.21) (1.17) (2.91) (0.05) (2.76) (0.13) (1.12) (0.26)
Log(business experience of the entrepreneur) 70.006 0.047 70.005 0.040
(0.50) (1.22) (0.42) (0.37)
Number of years of schooling of entrepreneur 0.015 0.054 0.012 0.086
(2.57) (1.80) (2.17) (2.58)
Establishment price-cost margins in 1993 0.000 0.001 0.001 0.008
(0.08) (0.30) (1.05) (1.55)
Industry average price-cost margins in 1993 0.014 0.065 0.004 0.096
(2.36) (3.34) (0.62) (3.22)
Import competing in 1993 70.126 70.599 70.038 70.593
(1.94) (2.80) (0.59) (2.19)
Food and beverages 0.113 0.938 0.108 0.929
(1.08) (2.44) (1.07) (2.38)
Textiles 0.061 2.243 0.100 2.521
(0.56) (4.22) (0.93) (4.49)
Garments and footwear 0.095 1.434 0.120 1.470
(0.99) (3.97) (1.29) (3.98)
Furniture 0.252 5.833 0.236 5.208
(2.51) (0.00) (2.47) (0.00)
(continued)
 Table 4. (Continued)

(1) (2) (3) (4)

Growth Selection Growth Selection Growth Selection Growth Selection

Constant 71.224 74.895 70.152 77.386 70.158 71.198 70.258 71.892

(2.77) (2.98) (0.31) (3.33) (1.17) (2.05) (1.82) (2.75)
Lambda 0.322 0.277 0.249 0.255
(14.98) (2.85) (5.53) (6.62)
Log likelihood 766.92 784.55 763.21 758.49
Wald Chi square (no of regressors) 24.45 15.47 21.71 26.13
p-value 0.0019 0.0169 0.0014 0.001
Observations 168 168 168 168

Note: Absolute value of z statistics in parentheses.

Competition, Human Capital and Small Business Longevity 825
826 T. Mengistae

comparison of the first column of Table 4 with column seven of Table 3 that the
effect of initial size is less pronounced in the unconditional distribution of the growth
rate (first column of Table 4) than it is in the distribution of growth conditional on
survival (column five of Table 3).
According to column seven of Table 3, a doubling of initial employment size
among survivors would reduce the average employment growth rate by 9.7
percentage points. However, correcting for selectivity bias in the first column of
Table 4 shows that the same change in employment would only cut down the
expected growth rate of startups by 5.5 percentage points. This sign of the bias is
consistent with the fact that larger firms tend to have higher rates of survival (second
column of Table 4). Even if there were no size effect in the unconditional (on
survival) distribution of the growth rates, the average growth rate of small survivors
would be higher than the average growth rate of larger firms on account of attrition
bias. Still there is clear evidence of an inverse size effect on the unconditional
distribution of growth rates since the coefficient of initial size is still negative and
statistically significant in the first column of Table 4 in spite of the correction for
attrition bias.
Probably the clearest manifestation of selectivity bias in the OLS estimation of
effects on business growth occurs when we move on to the influence of industry
characteristics. We see in the first column of Table 4 that the coefficient of the
industry average price–cost margin is positive and statistically significant. At our
point estimate, a percentage point increase in the margin is associated with an
additional 1.4 per cent employment growth. This is an estimate of effect on the
unconditional mean growth rate, which is on the average higher in industries where
the margin is higher and, in that sense, the pressure of competition is lower. It clearly
contrasts with our OLS estimate of the effect of the same change on the average
growth rate of a survivor not being statistically significant. Again, this suggests that
the industry average price–cost margin is significantly correlated with unobservable
determinants of the probability of business survival. We note that we have controlled
for the establishment level price–cost margin whenever we relate business growth or
business survival with industry profit margins in Tables 3 and 4. The positive
correlation that we infer to exist between industry profitability and business growth
or business survival therefore seems to be a genuine industry effect rather than
reflection of inter establishment differences in efficiency or in market power. We
should note that as an indicator of either of firm level productivity or of market
power, the establishment level price–cost margin does not have statistically
significant influence on the expected business growth rate of a startup. Although
Table 4 suggests that establishment price–cost margins are uncorrelated with the
probability of survival, the opposite result is suggested by our estimates of
alternative specifications of the survival functions.
Turning to the other indicator of industry competitiveness, there seems to be no
important difference between the magnitudes of coefficients of import competition
estimated using OLS and those of the Heckman model. As already noted, all
coefficients of the import competition dummy are negative and statistically
significant in Table 3, suggesting that, on the average, businesses in import
competing industries grow slower when they survive. Other things being equal, the
average surviving business in an import competing industry would grow slower by 12
 Competition, Human Capital and Small Business Longevity 827

percentage points. This estimate holds more or less as the effect of import
competition on the unconditional (on survival) growth rate in the first column of
Table 4, where controlling for the probability of survival does not seem to raise the
growth penalty of import competing firms by much.
Our estimates of the effects of business age and the entrepreneur’s business
experience on the growth rate as read in the first column of Table 4 also remain
largely the same as it is in Table 3, possibly because these variables are uncorrelated
with unobserved determinants of business survival. The same conclusion also applies
to the effect of the number of years of schooling of the entrepreneur. Unlike business
age and the entrepreneur’s business experience, the business owner’s schooling
increases with both the probability of business survival and the expected business
growth rate conditional on survival. However, there is not much difference between
the conditional (on survival) and the unconditional mean growth rates, which
suggests that unobservable determinants of the probability of survival are
uncorrelated with entrepreneurial schooling.
To sum up, a joint reading of Table 3 and Table 4 shows that unobservable
determinants of the probability of business survival are correlated with unobserved
determinants of business growth. As a result, OLS estimates of the parameters of a
business growth equation estimated on a sample of survivors would be biased for
those of the expected growth equation of startups. Correcting for this kind of bias
leads to the conclusion that, other things being equal, the expected growth rate of a
startup would be lower in a more competitive industry. This seems to be more
because the probability of survival decreases with industry competition than because
the latter reduces the growth rate conditional on survival. Given the pressure of
competition, the expected growth rate of a startup would be higher the lower is its
initial size. This is despite the fact that the probability of survival would increase with
initial size, and because the expected growth rate conditional on survival decreases
with initial size. Controlling for industry competition and initial size, the expected
growth rate of a business would be higher the more educated is the business owner.
This is partly because more schooling of the entrepreneur means a higher probability
of business survival. It is partly because greater entrepreneurial schooling raises the
expected growth rate conditional on survival.
The selection equation used in estimating Table 4 is a probit. This obviously
implies a somewhat arbitrary functional form of the underlying hazard of business
closure. In Tables 5 and 6, we investigate if this could possibly have distorted our
findings by estimating more flexible (duration) models of survival for comparison
with the selection equation of Table 4. As a prelude to this we show in Figures 1 and
2 Kaplan-Meier survival function estimates for the full sample (panel (a) of Figure 1)
and by sector and size group. Panel (b) of Figure 1 is consistent with the result in the
selection equation estimates of Table 4 that the survival probability increases with
employment size. Panel (a) of Figure 2 suggests the presence of industry effects in
survival that is not obvious from Table 4. Panel (b) of the same figure confirms the
implication of the estimates of the selection equation of Table 4 that survival
probabilities are lower in import competing industries.
Table 5 reports results of maximum likelihood estimation of alternative log linear
specifications of a Weibull model of proportional hazard. Given at the bottom of the
table but above the rows of test statistics are estimates of the parameter of the
 Table 5. Weibull regression of log relative hazard of business failure

(1) (2) (3) (4) (5) (6) (7) (8)

Log(end-of-1992 employment) 70.726 70.921 70.513 70.562 70.511 70.649 70.676 70.699
828 T. Mengistae

(3.47) (4.14) (2.24) (2.42) (2.20) (2.82) (2.84) (2.75)

Log(business experience of the entrepreneur) 70.618 70.616 70.626 70.673 70.658 70.631
(7.69) (7.55) (7.68) (7.52) (7.49) (7.53)
Number of years of schooling of entrepreneur 70.080 70.076 70.083 70.052 70.049 70.087
(2.06) (1.88) (2.11) (1.28) (1.21) (2.21)
Number of years of schooling of entrepreneur
X Food and beverages 70.004
(0.07)
X Textiles 70.315
(1.27)
X Garments and footwear 70.010
(0.16)
Establishment price-cost margins in 1993 70.170 70.158
(2.83) (2.62)
Industry average price-cost margins in 1993 70.012 70.016 70.017 70.014
(1.97) (2.22) (2.41) (2.22)
Import competing in 1993 0.636 0.842
(1.64) (2.16)
Textiles 72.888
(2.81)
Garments and footwear 71.350
(2.70)
Furniture 72.919
(2.85)
(continued)
 Table 5. (Continued)

(1) (2) (3) (4) (5) (6) (7) (8)

Food and beverages 70.574

(1.04)
Constant 73.825 72.532 78.184 77.946 77.600 4.420 3.905 76.544
(5.74) (3.72) (6.90) (6.64) (6.15) (1.04) (0.93) (5.14)
p 1.36 1.35 8.56 8.51 9.79 9.43 9.18 8.74
Standard error of p 0.19 0.18 1.13 1.14 1.32 1.27 1.24 1.18
Log likelihood 798.6 782.7 733.4 731.73 727.7 725.4 724.1 728.53
LR Chi square (no of regressors) 13.58 45.38 126.8 130.2 138.2 143 145.5 136.55
Prob 4 chi square (no of regressors) 0.0 0 0 0 0 0 0 0

Note: Absolute value of z statistics in parentheses.

Competition, Human Capital and Small Business Longevity 829
 Table 6. Cox-regression of the hazard of business failure

(1) (2) (3) (4) (5) (6) (7) (8)

830 T. Mengistae

Log(end-of-1992 employment) 70.723 70.908 70.573 70.676 70.551 70.530 70.616 70.561
(3.40) (3.99) (2.15) (2.40) (2.07) (1.92) (2.19) (1.99)
Log(business experience of the entrepreneur) 70.832 70.840 70.868 70.934 70.836 70.831
(5.61) (5.50) (5.75) (2.67) (5.48) (5.47)
Number of years of schooling of entrepreneur 70.097 70.081 70.108 70.105 70.113 70.103
(2.24) (1.74) (2.43) (2.03) (2.45) (2.22)
Number of years of schooling of entrepreneur
X Food and beverages 0.001
(0.02)
X Textiles 70.373
(1.45)
X Garments and footwear 70.059
(0.82)
Establishment profit margins in 1993 70.015 70.015 70.014 70.015
(2.03) (1.66) (1.86) (2.04)
Industry average price-cost margin in 1993 70.165 70.001
(2.74) (0.04)
Import competing in 1993 0.758 0.716
(1.77) (1.66)
Textiles 72.903
(2.81)
Garments and footwear 71.391
(2.74)
(continued)
 Table 6. (Continued)

(1) (2) (3) (4) (5) (6) (7) (8)

Furniture 72.924
(2.85)
Food and beverages 70.584
(1.04)
Observations 190 190 187 187 187 187 187 187
Log likelihood 7129 7114 770.59 768.12 768.9 763.71 766.68 767.54
LR Chi square (no of regressors) 13.11 44.35 110.2 115.18 113.6 123.99 118.05 116.34
p value 3E-04 0 0 0 0 0 0 0

Note: Absolute value of z statistics in parentheses.

Competition, Human Capital and Small Business Longevity 831
832 T. Mengistae

Figure 1. Kaplan–Meier survival function estimates: (a) full sample and (b) by size

baseline hazard, p, and its standard error. The estimates suggest values of p in excess
of unity. This in turn means the that baseline hazard rate is higher for an older
cohort of survivors, which is an interesting result in as far as it is contrary to findings
in other studies. The result in Table 4 that age is not a statistically significant
influence may well thus be the outcome of misspecification of functional form. In
trying to interpret the increase of business hazard with establishment age, it is
important to note that the hazard rate actually decreases with the entrepreneur’s
business experience. This is quite in line with what one would expect, and suggests
that the right business age variable in the context of the growth of entrepreneurial
establishments could be the length of the business experience of the entrepreneur
rather than the age of the establishment.
From the first row of the table we see that the estimated effect of initial size on the
hazard rate is negative and statistically significant, which is consistent with the
positive and statistically significant estimate for the coefficient of the same variable in
the survival probit in Table 4. Thus, we see in column seven that a business’
probability of failure is 68 percentage points lower than that of a business that is half
 Competition, Human Capital and Small Business Longevity 833

Figure 2. Kaplan–Meier survival function estimates: (a) by industry and (b) source of
competition

as large. The effect of the industry average price–cost margin on survival comes out
strongly also in this table. So do our estimates of the effect of exposure to
competition from imports and the effect of years of schooling of the entrepreneur.
For example, an additional year of schooling of the business owner would reduce the
hazard rate by up to nine percentage points in column eight.
Despite the differences between Tables 4 and 5 regarding the influence of business
age on the hazard of business closure, Table 5 supports the essentials of the
conclusions that we draw from the estimation of survival probits in Table 4. These
include that (a) smaller businesses are more likely to fail than larger businesses; (b)
greater entrepreneurial schooling or experience lowers the risk of business failure;
and (c) businesses are less likely to fail in industries where price–cost margins are
higher or in those not exposed to competition from imports.
834 T. Mengistae

These results come out more or less as strongly in the Cox regression reported in
Table 6. They cannot therefore be attributed to the parametric assumptions involved
in Table 4 or in Table 5. The only significant difference between Tables 5 and 6 is
that, contrary what we seen in Table 5, industry average price–cost margins and
competition from imports do not exhibit joint statistical significance in the same
equation in Table 6.

V. Summary and Conclusion

This paper has analyzed data from a sample survey of small, owner-managed
manufacturing establishments in Ethiopia’s main industrial city. The survey involved
a 1995 revisit of a sample that was first surveyed in 1993. Between the beginning of
1993 and the end of 1995, a sizeable fraction of the original sample closed down. Of
the remaining, some grew considerably while others contracted sharply. In the
analysis, we have sought to identify the factors that distinguished those that survived
from those that did not. Based on this and the growth patterns of survivors, we have
also estimated the influence of the same factors on the expected growth of potential
entrants or startups.
The most distinctive finding of the paper in the context of the literature on small
business dynamics in low income economies is that an enterprise is less likely to
survive and grows slower the more competitive is the industry in which it operates.
This is implied by the fact that the probability of survival and the mean growth
rate among survivors are both smaller in import competing industries, while
increasing with the industry average price–cost margin. We also find that
entrepreneurial human capital is an important determinant of small business
growth and survival. Our evidence for this is that the probability of business
survival increases with the number of years of the entrepreneur’s schooling, as does
the average growth rate. Both the probability of survival and the expected growth
rate conditional on survival also increase with the number of years of business
experience of the business owner.
The fact that factors affecting the probability of survival also affect the growth rate
of survivors is consistent with our finding that unobservable influences on business
hazard are negatively correlated with unobservable influences on growth. As a result,
the effect of competition and entrepreneurial human capital on the expected growth
rate of survivors would be biased for the effect of the same variables on the expected
growth rate of startups. In correcting for this bias, we have highlighted the direct link
that exists between the analysis of business hazard and the estimation of the expected
growth rate of startups.
We obtain these results while controlling for initial business size and initial
business age. Consistent with current empirical literature on firm dynamics in
developed and developing economies alike, we find that smaller businesses are less
likely to survive than larger ones. However, they also grow faster when they survive.
On the other hand, contrary to reports typically reported for developed economies
and for some developing countries, there is no evidence that the conditional (on
survival) mean growth rate of a business depends on its age once we control for
initial size. The probability of survival does diminish with business age, though,
controlling for initial size.
 Competition, Human Capital and Small Business Longevity 835

An immediate implication of our finding on the relationship between firm

dynamics and competition is that policies that directly enhance the competitiveness
of domestic industry as defined here should be expected to increase the hazard rate of
incumbents. In particular, trade liberalisation measures that expose hitherto
protected industries to import competition would lower business survival rates. A
second policy implication of the paper is that private and social returns to schooling
are by no means confined to realisation of greater labour market earnings. They
could also manifest themselves in greater entrepreneurial success as measured by
business longevity and growth.
Three limitations of the study should quality our results. We hope that future
research will address all three as more and better data become available on small
business dynamics in developing economies. One is the smallness of the sample size
of the data analyzed here by the standards of similar studies carried out in developed
economies, where the typical sample size runs into thousands of observations. A
second limitation is that our observations come from a handful of light consumer
goods manufacturing industries in which business life cycle patterns may not be the
same as those found in the services, agriculture or heavy industry. Thirdly, because
we have observed each establishment only at two points in time, we might not have
adequately controlled for unobserved heterogeneity of producers that could well bias
our estimates even for industries covered by our data.

Notes
1. Numerous studies report this result, particularly those following Evans (1987a, b) and Hall (1987) on
US datasets. Early among these are Dunne, Roberts and Samuelson (1989) on US data and Dunne and
Hughes (1994) on British data.
2. Moreover, there is substantial evidence that age and size effects in firm dynamics account for a
significant share of fluctuations in aggregate employment besides that commonly attributed to industry
or business cycle effects (for example, Leonard, 1987; Davis and Haltiwanger, 1992).
3. This should certainly be the case if our ultimate preoccupation is with the determination of aggregate or
industry level growth.
4. In an analysis of a large sample of informal business establishments in Southern Africa, McPherson
(1996) estimates this bias to be not statistically significant. However, we cannot assume that this result
generalises to other low income economies, or to samples with a larger proportion of formal sector
firms.
5. This is a standard sample selection (or censoring) problem as set out in Maddala (1983), and discussed
in Hall (1987), Evans (1987a, b), and Dunn, Roberts and Samuelson (1989) in the specific context of
firm growth.
6. The equations to be estimated in this case would be
Dyi ¼ zi d þ ei if closedi ¼ 0
but would be unobserved if closedi ¼ 1
where closedi ¼ xi b2 þ Zi, and Zi is an iid random error term orthogonal to xi.
7. See, for example, Lancaster (1990: 162–70) for the derivation of the log likelihood function we
maximise in estimating equation (4) and (5) and the corresponding covariance matrices. Audretsch and
Mahmood (1995) and Honjo (2000) are examples of the use of the semi-parametric and parametric
maximum likelihood employed here in the context of business survival analysis in developed economies.
McPherson (1995) analyzes the hazard of closure in a large sample of predominantly informal firms in
Southern Africa using what is, basically, the same methodology as pursued here.
8. The addition of second order terms in the log of age in our estimation did not alter this result.
9. This assumes that rates of return do not vary across grade levels.
836 T. Mengistae

References
Audretsch, D. B. and Mahmood, T. (1995) New firm survival: new results using a hazard function, Review
of Economics and Statistics, 77(1), pp. 97–103.
Bates, T. (1990) Entrepreneur human capital inputs and small business longevity, Review of Economics and
Statistics, 72(4), pp. 551–9.
Davis, S. and Haltiwanger, J. (1992) Gross job creation, gross job destruction and employment
reallocation, Quarterly Journal of Economics, 107(3), pp. 819–64.
Dunne, P. and Hughes, A. (1994) Age, size, growth and survival: UK companies in the 1980s, Journal of
Industrial Economics, 42(2), pp. 115–40.
Dunne, T., Roberts, M. and Samuelson, L. (1989) The growth and failure of US manufacturing plants,
Quarterly Journal of Economics, 104(4), pp. 671–98.
Evans, D. (1987a) Tests of alternative theories of firm growth, Journal of Political Economy, 95(4),
pp. 657–74.
Evans, D. (1987b) The relationship between firm growth, size and age: estimates for 100 manufacturing
industries, Journal of Industrial Economics, 35(4), pp. 567–81.
Gunning, J. and Mengistae, T. (2001) Determinants of African manufacturing investment: The
microeconomic evidence, Journal of African Economies, 10(1), pp. 48–80.
Hall, B. (1987) The relationship between firm size and firm growth in the US manufacturing sector,
Journal of Industrial Economics, 35(4), pp. 583–606.
Honjo, Y. (2000) Business failure of new firms: an empirical analysis using a multiplicative hazards model,
International Journal of Industrial Organization, 18(4), pp. 557–74.
Hopenhayn, H. (1992) Entry, exit, and firm dynamics in long run equilibrium, Econometrica, 60(5),
pp. 1127–50.
Jovanovic, B. (1982) Selection and the evolution of industry, Econometrica, 50(3), pp. 649–70.
Jovanovic, B. and MacDonald, G. (1994) Competitive diffusion, Journal of Political Economy, 102(1),
pp. 24–52.
Lancaster, T. (1990) The Econometric Analysis of Transition Data (Cambridge: Cambridge University
Press).
Leonard, J. (1987) In the wrong place at the wrong time: the extent of frictional and structural
unemployment, in K. Lang and J. Leonard (eds) Unemployment and the Structure of Labour Markets,
pp. 128–53 (New York: Blackwell).
Lippman, S. and Rumelt, R. (1982) Uncertain imitability: analysis of inter-firm differences in efficiency
under competition, Bell Journal of Economics, 13(2), pp. 418–38.
Lucas, R. (1978) On the size distribution of business firms, Bell Journal of Economics, 9(2), pp. 508–23.
Maddala, G. S. (1983) Limited Dependent and Qualitative Variables in Econometrics (Cambridge:
Cambridge University Press).
McPherson, M. (1995) The hazards of small firms in southern Africa, Journal of Development Studies,
32(1), pp. 31–54.
McPherson, M. (1996) Growth of micro and small enterprises in southern Africa, Journal of Development
Economics, 48(2), pp. 253–77.
Pakes, A. and Ericson, R. (1998) Empirical implications of alternative models of firm dynamics, Journal of
Economic Theory, 79(1), pp. 1–45.
Ramachandran, V. and Shah, M. (1999) Minority entrepreneurs and firm performance in Sub-Saharan
Africa, Journal of Development Studies, 36(1), pp. 71–87.
View publication stats