«Pecuniary and Non-Pecuniary Aspects of Self-Employment Survival Yannis Georgellis*, John Sessions**, and Nikolaos Tsitsianis*** * Department of ...»
Pecuniary and Non-Pecuniary Aspects of Self-Employment Survival
Yannis Georgellis*, John Sessions**, and Nikolaos Tsitsianis***
* Department of Economics and Finance
Middlesex UB8 3PU
**Department of Economics
University of Bath
Bath, BA2 7AY
Department of Economics
University of Hertfordshire
Hertfordshire AL10 9AB
Abstract: We examine the factors that determine self-employment duration in Britain, paying particular attention to self-reported job satisfaction variables and non-pecuniary aspects of self-employment. Based on spell data from the British Household Panel Study, we estimate single-risk and competing-risks hazard models, separately for males and females. Our results show that job satisfaction is indeed a strong predictor of self-employment exit, even after controlling for standard economic and demographic variables. When five domain job satisfaction measures are used, we find that pay, job security and initiative are the three aspects of self-employment most valued by the self-employed themselves. Gender differences regarding the determinants of self-employment survival and exit destination states are also evident.
Keywords: Self-employment, Survival, Job Satisfaction JEL Classification: J23, J28, J60 Word Count: 5844 Electronic copy of this paper is available at: http://ssrn.com/abstract=941154 I. Introduction Compared to the plethora of studies examining self-employment entry, relatively few studies examine the determinants of self-employment survival and exit. Noteworthy recent examples of empirical studies focusing exclusively on self-employment survival include Johansson (2001) and Taylor (1999). Johansson (2001) uses Finnish longitudinal data and finds that the young, the more educated, and those with previous unemployment experience face a higher risk of exiting selfemployment. Based on data from the British Household Panel Survey (BHPS), Taylor (1999) finds that approximately 40 per cent of self-employment ventures in Britain that started in 1991 did not survive their first year in business. Interestingly, a substantial proportion of self-employment spells is not terminated through bankruptcy but rather through moves to alternative employment. Taylor also highlights the importance of previous unemployment experience, whether individuals quit the previous job, and initial capital as important determinants of self-employment survival. Bates (1990) provides some of first evidence on the probability of surviving in business for a sample of male entrepreneurs in the US and finds that those who possess greater business acumen, labour market skills, and capital, are more likely to survive ceteris paribus.1 Failure to address entrepreneurial success and self-employment survival could cast doubts on the efficiency of government programmes designed to facilitate individuals’ entry into selfemployment.2 The design and implementation of such programmes is informed by numerous empirical studies that examine mostly the factors affecting individuals’ decision to become selfemployed, with relatively little information on the factors that determine entrepreneurial success and survival. High self-employment exit rates may be viewed as evidence of poor matches between entrepreneurial availability and skill requirements and can be costly, not only for the self-employed themselves but also for third parties, including banks, customers and governmental and private financial institutions. Understanding the determinants of self-employment survival could allow for a more complete response by policy makers who view entrepreneurship as the key to job creation and prosperity.
Electronic copy of this paper is available at: http://ssrn.com/abstract=941154 In this paper, we provide additional empirical evidence regarding the factors that determine self-employment duration in Britain, paying particular attention to non-pecuniary aspects of selfemployment and self-reported job satisfaction variables. Although recent empirical work on the determinants of self-employment entry acknowledges the importance of job satisfaction for individuals’ choice between salaried work and self-employment, the small number of studies examining self-employment survival focus exclusively on the role of standard economic and demographic variables. Generally, there is very little direct evidence on the correlation between self-reported satisfaction measures and self-employment duration. This is in sharp contrast to the empirical literature on labour turnover and mobility, where a growing number of studies examine the role of job satisfaction and non-pecuniary aspects of a job as strong predictors of quits [see for example, Akerlof et al (1988) and Clark et al (1998)]. More recently, Clark (2001), using employment spell data from the BHPS (omitting self-employment spells), finds that overall job satisfaction is indeed a powerful predictor of separations and quits, even after controlling for wages, demographic and other job variables. Interestingly, Clark (2001) also finds that job security and pay are the two job attributes most valued by workers.
Based on self-employment spell data from the BHPS, our results are generally consistent with the findings of previous studies as far as standard demographic and job variables are concerned. In particular, our results confirm the importance of assets and previous labour market status as important determinants of self-employment duration. Differences in survival rates across gender and different types of self-employment are also evident. However, we also find that job satisfaction and non-pecuniary characteristics of self-employment are strong predictors of selfemployment exit. Furthermore, when overall job satisfaction is broken down to five domain satisfaction variables, we find that satisfaction with pay, satisfaction with job security and satisfaction with work initiative emerge as the most important determinants of self-employment exit. This suggests that these three aspects of self-employment are the most valued by the selfemployed themselves. Similar findings emerge when estimating competing risk survival models, where exit probabilities to alternative destination states are compared. In general, the results of competing risks models lend support to the view that for many workers, self-employment is indeed a stepping stone to paid employment. Thus, the present paper contributes not only to increasing our general understanding of the determinants of self-employment survival, but also to a growing literature on the role of job satisfaction as a powerful predictor of quits and labour turnover, a rather novel aspect of the self-employment survival literature.
The paper is set out as follows: In section II, we describe the data and present some preliminary results based on non-parametric duration models. In section III, we present the results of single risk and competing risks proportional hazard models. In section IV, we conclude.
II. Data and Preliminary Results Our data are from Waves 1 to 8 (1991 to 1998) of the British Household Panel Survey (BHPS), a nationally representative sample of some 5,500 households and 10,000 individuals interviewed annually every fall since 1991. The BHPS is a rich data set providing information not only on a wide range of personal, household, and labour market characteristics, but also on lifetime job histories. Children in the households are interviewed separately once they reach the age of sixteen, and if any member of a household splits to form a new household, then all adult members of both the original and new households are interviewed annually. By such means, the survey remains broadly representative of the British populace over time.
We restrict our samples to include individuals who are between the ages of 18 and 64 for men and 18 and 60 for women. Individuals for whom information on the main variables of interest is not available are excluded from the sample (see Table 1: Note – all Figures and Tables are set out in Appendix B).
By combining information contained in the BHPS spell data files with employment status information collected at each wave, we construct a complete self-employment history for each individual in the sample. Our final sample consists of 1,436 self-employment spells that could be matched to individual interview data after cases with missing data and cases with inconsistencies have been removed.3 In this sample, 170 (11.83%) individuals have multiple spells and 1266 (88.17%) have a single spell for the period 1991-1998. Treating self-employment duration as a continuous variable, the data shows that the mean self-employment duration for males is 67.87 months (5.65 years) whereas the mean duration for females is 50.08 months (4.17 years). The corresponding median figures are 35 months (2.91 years) and 24 months (2 years) respectively.
Non-parametric survival analysis techniques allow us to examine the probability of an individual leaving the self-employment state conditional on having survived in the origin state for the whole interval. Figure 1 (shows the survivor function for men and women using the KaplanMeier estimator (Kaplan and Meier, 1958). The Kaplan-Meier estimates suggest that the selfemployed are more likely to quit during the first 100 months (8.3 years) and that females have lower survival rates than do males. The male hazard rate rises quickly above 400 months (33.3 years), probably due to retirement, while the female rate is somewhat flatter. Looking at the survivor functions, 51 percent (70 percent) of males (females) were eliminated from selfemployment within 200 months (16 years). Gender differences in survival rates may reflect gender differences in the motives for and aspirations from self-employment, differences in labour market opportunities, and even differences in tastes for pecuniary vs. non-pecuniary aspects of entrepreneurship. 4 Turning to the conditional probability of exiting to a new state, the survivor function, shown in Figure 2, illustrates how the exit probabilities vary with both time and the destination state under the competing risks assumption. We focus our attention to three main destinations states [notworking, including unemployment and family care (UNFC); wage employment (WE); and other self-employment (SE)], which account for 78.7 percent of all completed self-employment episodes.
In order to assess whether non-working is distinctly different from unemployment, we also examine the survivor function for transitions from self-employment to unemployment (UN). The probability of moving from self-employment to unemployment/family care (UNFC) is markedly higher than the probability of moving to any other destination states until the 144th month (12th year), where the probabilities of moving to unemployment/family care and wage employment coincide. The most likely destination state is wage employment, followed by unemployment/family care and then by self-employment, supporting the view that self-employment is a stepping stone to paid employment.
A possible explanation for this is that experience of self-employment unveils unobserved characteristics of the individual that increase the probability of finding a better match in the labour market.
Differences in the survivor functions between males and females are also confirmed by the Log-Rank test and Wilcoxon test for the equality of survivor functions (see Table 2).5 Both the LogRank test and the Wilcoxon test, reject the hypothesis of equality of the two survivor functions.
III. Econometric Analysis: Proportional Hazard Models In this section, we examine the determinants of self-employment duration by estimating Cox proportional hazard models [Cox (1972)]. The Cox proportional hazard model is semi-parametric, in the sense that it involves an unspecified baseline hazard instead of requiring further distributional assumptions. This approach, standard in applied work, assumes that censoring does not provide any information about latent failure times beyond that available in the covariates [e.g., Katz and Meyer (1990)]. The Cox proportional hazard model is also convenient for dealing with right censoring.
We model self-employment duration using both single risk and competing risks models. The former refers to the probability of exit from self-employment, irrespective of the state of destination. Competing risks models are used to estimate the probability of exit to different states, namely paid employment, unemployment/family care, unemployment and self-employment.6 Under the assumption of independent hazards, the competing risks model allows us to test whether the explanatory variables have different effects on the propensity to leave self-employment depending on the state of destination. All explanatory variables are measured at the last available interview prior to the termination of the spell.7 Results: Single Risk Proportional Hazard Model The results from estimating the single risk hazard model are shown in Table 3.8 As column 1 shows, males are more likely to survive in self-employment. Gender differences in self-employment survival could be attributed to differences in cultural attitudes towards entrepreneurship, risk-taking, and the availability and/or affordability of family services. Age has a significant and non-linear effect on self-employment survival. As Calvo and Wellisz (1980) suggest, age might be superior to labour market experience as both an indicator of the learning process and as an index of capital accumulation. Furthermore, Evans and Leighton (1989) claim that age may be a proxy for an individual’s attitude towards risk, with mature-aged persons being less willing to bear the stress and risks associated with self-employment. Taylor (1999) includes two age dummies (aged 30 and aged 30-49) and reports that male duration in self-employment is inversely related to age. Similar age effects are apparent for females, but the effect for both males and females disappears when the author desegregates his sample into voluntary and involuntary terminations. Consistent with the findings of Johansson (2001), our results show that there is a clear effect of age on the female risk of exiting self-employment, but such an effect is less pronounced for males.