Earnings and education among the self employed in Colombia

THE WORLD BANK EDT7o Discussion Paper EDUCATION AND TRAINING SERIES Report No. EDT70 Earnings and Education Among the Self Employed in Colombia George Psacharopoulos Ana-Maria Arriagada Eduardo Ve1ez April 1987 Education and Training Department Operations Policy Staff The views presented here are those of the author(s), and they should not be interpreted as reflecting those of the World Bank.  Discussion Paper Education and Training Series Report No. EDT70 EARNINGS An EDWCMTION AMDDG THE SELF-EMPLOYED IN COLOMBIA George Psacharopoulos Ana-Maria Arriagada Eduardo Velez Research Division Education and Training Department April 1987 The World Bank does not accept responsibility for the views expressed herein, which are those of the author(s) and should not be attributed to the World Bank or its affiliated organizations. The findings, interpretations, and conclusions are the results of research or analysis supported by the Bank; they do not necessarily represent official policy of the Bank. Copyright 0 1987 The International Bank for Reconstruction and Development/ The World Bank  Abstract A major debate in the economics of education is the extent to which observed earnings differentials between more and less educated workers reflect differences in productivity. This study focuses on the earnings of the self-employed and those in the private sector of the economy. It is argued that in a competitive economy wage differences within such sectors is a good proxy for productivity differences by level of education. Data from a 1984 sample of about 21,000 workers in Colombia support the hypothesis that the returns to education among the self-employed are highest relative to any other grouping of workers. The policy implications of such findings are discussed.  EARNINGS AND EDUCATION AMONG THE SELF-EMPLOYED IN COLOMBIA George Psacharopoulos*, Ana Maria Arriagada*, Eduardo Velez** I. Introduction There are at least two reasons for studying the labor market performance of those in self-employment or in the informal sector of the economy. The first relates to a perennial debate in the economics of education, and the second to the formulation of educational policy, especially in developing countries. Economics of Education. The theoretical cornerstone of the economics of education is productivity differences between workers with more and less education. However, given the difficulty of directly measuring worker productivity, empirical estimates have typically relied on earnings differences by level of education. Although such approximation may be valid in competitive labor market environments, it has understandably been criticized in the literature as having limited applicability in non-competitive economic environments. By restricting the observations to those in self-employment activities, one better approximates the elusive productivity differential. The self-employed simply receive as pay the true worth of what they produce, (net of the returns to other factors they may use, like capital and raw materials). To put it in other words, focusing on the labor * The World Bank, Washington, DC, and (**) Instituto SER de Investigacion, Bogota, Colombia. The views expressed here are those of the authors and should not be attributed to the above institutions. We wish to thank Avi Dor, Bernardo Kugler and Morton Stelcner, for their valuable comments on an earlier draft of this paper.  - 2 - By restricting the observations to those in self-employment activities, one better approximates the elusive productivity differential. The self-employed simply receive as pay the true worth of what they produce, (net of the returns to other factors they may use, like capital and raw materials). To put it in other words, focusing on the labor earnings of the self-employed corresponds to a natural, non-econometric pricing of education as a factor of production. Educational Policy. An increasing number of graduates of the school system in developing countries are absorbed in non-formal wage, non-modern sector activities. The issue hence arises whether the education such persons received has a social economic payoff. Unfortunately, typical labor market surveys focus on those employed in large establishments or the civil service. But because of institutional factors and the non-profit maximizing behavior of the public sector, such data cannot be used to answer the question: Does it pay to expand this or that level of education? In this paper we use data from a sample of 21,300 workers in Colombia to answer the above questions. In the following section we present the sample and key mean characteristics of workers in self-employment, the private sector and the public sector 1/. Section III reports the results of earnings functions within economic sectors and compares them to prior estimates in the literature. Section IV addresses the issue of selectivity bias and who enters into the different sectors in the first place. Section V focuses on different types of self-employed workers. The final section discusses the policy implications of the results. 1, Estimates for public sector employees are also presented as a point of reference.  -3- II. The Sample The data used in this paper come from the "National Household Survey' carried out by DANE (Departamento Administrativo Nacional de Estatistica) in the ten largest cities of Colombia in June 1984 2/. The sample includes about 25,000 households or 42,000 individuals 12 years old and over who declared any work activity, regardless of earnings, during the week before the survey. The questionnaire raised information usually not provided by household surveys, e.g., whether the household operates a business, who are the household members working in that business, and how many workers does the business employ. Given the focus of our study and the content of the data set, we restrict the analysis to 21,300 male wage earners and self-employed between 15 and 65 years of age with non-zero earnings who reported having only one main occupation 3/. The exclusion of those workers who reported more than one job truncated our sample. However, since those performing more than one job are similarly distributed among wage and self-employed individuals, we trust that our estimates have not been biased one way or 2/ The sample is representative of Colombia's urban population and the socio-economic composition of each city. The cities in the sample are: Bogota, Medellin, Cali, Barranquilla, Bucaramanga, Cucuta, Manizales, Pasto, Pereira, and Villavicencio. For a more detailed description of the sample see, DANE, (1984) 3/ This limitation was dictated because hours of work in the main occupation was the only proxy for labor supplied in the data set, while earnings from labor were lumped together in one question for all jobs performed by each individual worker. Therefore, for workers with more than one job there is no information on the overall number of hours they work to account for their reported earnings.  - 4 - another. Females were also excluded in order not to confound issues of labor productivity with sex discrimination 41. Domestic workers were also excluded given that their occupation-specific compensation arrangements include payment in kind -- an item not properly recorded in the survey. The earnings variable for wage workers is an aggregate of all payments received from their wage employment5/. The equivalent variable for the self-employed explicitly refers to the net labor income received from their business. Even though the survey did not gather information on business expenditures, capital stock, or inventories, the design of the questionnaire is such that the self-employed income mostly derives from their labor 61. As is well known, most self-employed workers in Colombia do not keep accounting of any sort which may bias their reported labor income. However, this will not affect the estimated coefficients but only the earnings differences between salaried and self-employed workers. In this respect, caution is advisable when comparing earnings figures for the two groups of workers. 4/ Also, there is no information in this data set on actual female work experience, number of children and the like which are necessary to analyze their sex-specific labor supply decisions. 5/ According to some observers of the Colombian labor market, salaried earnings in household surveys may be downwardly biased by as much as 25Z because they do not ask specifically for "other payments" received by salaried workers (such as meals, clothing and transportation). 6/ The first question in the survey was: "How much was your net income from this business or profession last month?"; followed by: "Did you receive income from any other source other than labor last month?".  - 5 - Table 1 provides key statistics of the sample. There are substantial differences between the age and educational characteristics of wage earners in the public and private sectors, and the self-employed, as well as the mean monthly earnings. Table 1 Mean Sample Characteristics by Type of Employment Characteristic Self- Private Sector Public Sector Employed Employees Employees Earnings (pesos per month) 24,599 19,367 29,592 (36,192) (20,701) (22,531) Years of Schooling 6.4 7.0 9.3 (4.0) (3.7) (4.5) Age 38.8 31.1 36.8 (12.2) (11.0) (10.3) Hours of Work per week 51.4 50.0 48.0 (15.4) (11.7) (11.0) Social Security 0.104 0.584 1.00 Employer 0.194 - - E'irm Size: - 1 worker 0.615 0.030 - - 2-5 workers 0.310 0.252 - - 6-10 workers 0.040 0.129 - - 11+ workers 0.033 0.588 1.00 Household Business 0.188 - Number of workers 6,727 12,219 2,355 Note: Numbers in parenthesis are standard deviations.  On average, the self-employed are 7 years older than private sector wage workers, they have almost one year less education, and they work more hours per week. The raw mean earnings of the self-employed appear substantially higher than those of the private sector salaried workers, and lower than those of public sector employees. As expected, the self-employed distributions show greater dispersion than those of the salaried workers, particularly in regard to earnings. Table 2 shows the mean earnings and educational level of the three kinds of workers. Table 2 Educational Atta4nment and Earnings by Type of Employment Self-employed Private Sector Public Sector Employees Employees Educational Educat. Mean Educat. Mean Educat. Mean Level Attain. Monthly Attain. Monthly Attain. Monthly Earnings Earnings Earnings (Z) (pesos) (2) (pesos) (%) (pesos) None 5.3 11,646 2.6 13,435 1.6 17,610 Primary incomplete 25.0 14,942 19.5 13,669 10.8 19,116 Primary complete 24.3 18,814 21.9 15,839 16.4 20,658 Secondary incomplete 26.0 24,148 34.1 16,980 25.3 22,965 Secondary complete 10.3 34,924 12.9 22,982 19.9 27,886 Higher incomplete 2.6 40,277 3.5 30,074 6.8 29,243 Higher complete 6.5 72,827 5.5 55,888 19.2 54,749 Overall 100.0 24,599 100.0 19,367 100.0 29,592 In general, the self-employed appear with higher earnings than private sector employees for all educational levels, with the exception of those without any schooling. As expected, the public sector workers are more educated than the rest, and the less educated workers in that sector receive higher pay than workers in other sectors with the same schooling.  -7- I:I. Basic Earnings Functions We start by following the standard human capital model (Mincer 1974) to estimate earnings functions for the different types of workers in the sample: ln(Y) = a + bS + cEX + dEX2 + eZ + u where, Y = monthly labor earnings, S = years of schooling, EX = years of working experience computed as Age - S - 6, Z = a vector of adjustment variables which includes: - hours of work per week - access to social security - firm size u = error term As it is well known, the use of total earnings as the dependent variable can produce bias in the estimates of the parameters depending on the labor supply responses to the wage rate 71. Therefore, in a second stage we added to the right hand side of the above equation the log of hours of work per week, and a dummy variable, "access to social security", as a control variable. Access to social security in Colombia, like in many other developing countries, is a proxy for a host of non-easily measurable conditions of employment. Jobs with social security are "better jobs" since they fulfill institutional regulations regarding employment stability, wages, health care, leave, and so on. If this is so, we expect that access to social security will have a positive and significant influence on earnings. We also included a set of firm size 7/ Blinder (1973) suggests the use of wage rates as the dependent variable in the equation instead of total earnings. Since in our case the coefficient of the variable hours included in the right hand side of the equation turned out to be significantly less than one, we decided not to use wage rates but to include the log of hours worked in the standard human capital model.  - 8 - dusuies in order to further account for the presence of the so called "protected sector" effect. The importance of the employment size of the firm accounting for wage differentials has been empirically documented in LDC labor markets (Mazumdar 1983, House 1984). This is because workers of larger firms may earn more than workers with similar endowments in smaller firms by virtue of minimum wage policies, the presence of labor unions, and the like. If this is the case in Colombia, we expect to find a premium in earnings associated with larger enterprises. Taking the private sector employees and self-employed workers first, pooled together and then, separately, we used a Chow test for the equality of the overall structure of earnings for both types of employment. The results indicated that we can reject the null hypothesis of a similar earnings structure for wage and self-employed workers. Therefore, a separate treatment for each subsample is in order. Table 3 presents the results of the basic earnings functions. The sign of the coefficients conform with human capital theory and the explanatory power of the model is consistent with previous research in Colombian urban labor markets (Fields and Schultz 1977, Bourguignon 1980, Psacharopoulos 1983). Also, as expected, the model performs better for wage workers. As seen in Table 3, there appear to be no earnings differentials between self-employed and wage workers. This finding questions the validity of a "segmented labor market" in Colombia, stating that "informal sector" jobs (i.e. self-employment) are characterized by lower wages than those of the "formal sector" (i.e. salaried employment).* * However, given the limitations of the earnings figures reported in household surveys described in section II, this finding needs further research.  -9- In terms of the returns to education, our estimates deviate from the findings of other studies (Chiswick 1976, Henderson 1982) since ours show similar returns to schooling for the self-employed compared to those for salaried workers. In fact, our estimates suggest higher private returns to education for the self-employed (12.9%) versus (10.6%) for the wage earners. However, when testing the equality of the two regression coefficients (via a t-test), we found that the partial effect of schooling on the earnings of salaried workers and self-employed workers is similar. In terms of the returns to experience, the model confirms the expectation that earnings increase with age. Wage and self-employment labor markets may be expected to differ regarding the way experience is valued in them, and the impact of experience in wage employment would be more important than it is in self-employment (Mazumdar 1973). The coefficients obtained in our regressions, show a very similar effect of experience on the earnings of wage and self-employed workers 5.8% and 5.7%, respectively which suggests similar experience earnings profiles. Summarizing the results thus far, we can conclude that education is an important and strong source of variation in earnings, and investment in education exhibits decent rates of return across sectors.  - 10 - Table 3 Basic Earnings Functions by Type of Employment Independent Self- Private Self-employed Public variable employed Sector Private Sector Public Sector employees employees employees (1) (2) (1+2) (3) Years of schooling 0.129* 0.106* 0.114* 0.095* (50.6) (81.8) (91.2) (40.7) Experience 0.057* 0.058* 0.057* 0.043* (19.3) (43.6) (44.8) (15.9) Experience squared -0.0007* -0.0007* -0.0007* -0.0005* (16.1) (29.6) (32.6) (10.5) Constant 7.982 8.226 8.150 8.635 R2 (adjusted) 0.286 0.382 0.324 0.429 Sample size 6,676 12,166 18,843 2,344 Mean dep. variable (log earnings/month) 9.666 9.648 9.654 10.118 Notes: t-values in parenthesis. * Statistically significant at 1 percent level or better. Comparison with Other Studies Colombia is one of the very few developing countries for which estimates of the returns to education exist over time. However, most of the earlier studies used data from surveys in Bogota's labor market, which limits the comparability of over time trends (see Appendix Table A-1). Nevertheless, the comparison of our basic earnings functions with prior estimates seem to confirm the hypothesis that human capital investments, like any other investment, exhibit diminishing returns over time  - 11 - (Table 4). Following the expansion of Colombian education 8/ we witness a general pattern toward the decline of the impact of education on earnings. Table 4 The Returns to Education over Time (percent) 1965 1971 1974 1979 1984 All workers 17.3a/ 16.7a/ 12.8a/ - 11.0 Self-employed - - 13.2a/ 14.8 12.9 Private sector workers - - 14.6 12.6 10.6 Public sector workers - - 13.4 12.9 9.5 Source: Based on Appendix Table A-1. Notes: All estimates refer to males. - not available. a/ Bogota only. Bourgouignon (1980) and Psacharopoulos (1983) already detected this trend when analyzing 1974-1975 data for Bogota and all urban Colombia respectively. Horn's (1986) results using 1979 urban data, confirm the same pattern. Our findings indicate that a further decline in the returns to education occurred between 1974 and 1984. From a high of seventeen percent return to an additional year of schooling estimated by Schultz in 1965, the returns to education decreased to fifteen percent in the mid-seventies (Psacharopoulos, 1983), and to about eleven percent in 1984 (according to our estimate for all private sector workers). Earlier estimates (Bourgouignon, 1980) show returns to education for the self-employed (13.2Z) similar to those for wage earners (12.8%), where 8/ The mean years of schooling of the male labor force increased from 2.8 years in 1964 to 5.0 years in 1978. See Psacharopoulos and Arriagada (1986).  - 12 - the test on their difference is not statistically significant. Our results, using a sample comparable to Bourgouignon's 9/, are: 12.4% and 9.4% for self-employed and wage earners, respectively. These figures suggest that the decline in the returns to education has taken place largely among the wage earners. In fact, the self-employed returns to an additional year of schooling remained practically the same during the 1974 - 1984 period. Comparing our estimate for public sector workers (9.5%) with prior results --13.4% by Psacharopoulos (1975) -- we can see that public sector workers have experienced the largest decline in the returns to education during this period. Extension of the Standard Model We reestimated the basic model including the amount of labor supplied (log of hours worked per week), access to social security (dummy variable where 1 indicates belonging to any social security scheme, and 0 otherwise), and employment size of the firm (three dummy variables: 2 to 5 workers, 6 to 10 workers, and more than 10 workers, the omitted category being one-worker firms). Table 5 presents the result of the extended model. 9/ That is, excluding "professionals" with university education.  - 13 - Table 5 Extended Earnings Functions Self- Private Self-employed Public Independent employed Sector & Private Sector variable Employees Employees Employees (1) (2) (1+2) (3) Years of schooling 0.113* 0.095* 0.109* 0.096* (43.6) (70.4) (86.2) (40.7) Experience 0.047* 0.050* 0.054* 0.043* (16.3) (37.9) (42.9) (15.9) Experience squared -0.0006* -0.0007* -0.0007* -0.0005* (12.8) (26.2) (30.7) (10.3) Log hours worked 0.294* 0.206* 0.264* 0.170* (10.9) (12.2) (17.6) (4.3) Social Security 0.306* 0.118* 0.108* - (9.7) (11.3) (9.5) Firm Size 1/: 2-5 workers 0.180* -0.048 0.100* - (8.9) (1.8) (8.2) 6-10 workers 0.497* 0.002 0.154* - (10.5) (0.1) (9.1) 11+ workers 0.690* 0.114* 0.199* - (13.2) (4.4) (14.4) Constant 6.963 7.466 7.010 7.912 R2 (adjusted) 0.349 0.419 0.360 0.433 Sample size 6,676 12,166 18,843 2,344 Mean dep.variable (log earnings/month) 9.666 9.647 9.654 10.118 Notes: t-values in parenthesis. 1/ the omitted group is one-worker firms. - omitted since all public sector employees belong to social security. * Statistically significant at the 1 percent level or better.  - 14 - Besides the fact that all variables and the explained variance exhibit the expected behavior, the returns to schooling and experience for both subsamples remain quite stable when labor supplied, social security and size of the firm are added to the equations. The coefficients of hours of work per week, (0.294) and (0.206) for the self-employed and wage workers respectively, are statistically significant for both groups. Thus, for the self-employed an increase in the hours of work per week by one percent raises their monthly earnings by 0.29 percent, while for the wage earners the same rise in hours of work increases their monthly earnings by 0.20 percent. The coefficients of access to social security (0.306) for the self-employed and (0.118) for the wage earners show a positive and significant effect on earnings. For the wage earners the interpretation of this coefficient is straightforward: belonging to a social security scheme is likely to be determined exogeneously since it comes attached to the job and is mandatory. And jobs providing social security are those offering better employment conditions, which includes a higher pay. For the self-employed this positive coefficient does not necessarily have the same interpretation. Since for them enrollment in a social security scheme is voluntary, it is quite possible that those who earn higher incomes are likely to enrol. Hence, there may be simultaneity bias for the self-employed, in that social security access is a function of earnings, and earnings are a function of social security. The coefficients of firm size show a stronger impact regarding self-employment earnings than they do on salaried earnings. An increase in the size of the firm is associated with a greater positive and significant impact on the earnings of the self-employed. For wage  - 15 - employment this effect is significant only for firms with more than 10 workers. These results for wage workers might be due to the fact that in this data set all firms with 10 or more workers were lumped together in one category. Previous research in Colombia (Fields 1978) shows that only larger firms (50 workers and more) have a significant impact on earnings. IV. Selectivity Bias As mentioned earlier, most of the literature on the returns to education is restricted to wage workers. Only a few studies include the self-employed (Kuznetz 1966, Blaug 1974, Chiswick 1977, Fields and Schultz 1982). However, such studies suffer from selectivity bias (Heckman 1979) since they do not consider how the individuals in the sample have sorted themselves out into wage labor or self-employment 10/. Because an individual's decision to participate in wage or self-employment presumably affects his productivity differently, we would expect some systematic censoring in the sample, which would bias the estimated coefficients of the basic earnings functions. The bias arises from a specification error (or "omitted variables") where the OLS regressions confound the behavioral parameters of interest with the parameters that determines the probability of one being in the sample. Thus, OLS estimates are biased because the error terms of the earnings equation and the labor supply equation are correlated, and the conditional mean of the error term is not included among the explanatory variables. Our purpose in this section is to 10/ It should be noted that previous studies in Colombia (Kugler 1975, Fields 1976) show that "family background" does not play a significant and direct role in earnings determination. Accordingly, we do not expect selectivity bias to operate because of omitted background variables.  - 16 - correct the unadjusted rates of return for selectivity bias, that is, to account for the individual's choice between wage and self-employment. In doing so, we followed Heckman's (1979) method. This method consists in a two stage estimation that arrives at consistent estimates of the "earnings-offer" function from the "observed-earnings" sample 11/. The omitted variable is represented by the ratio of the density of a standard normal f(Ni, A), where Ni represents labor supply determinants and A the parameters to be estimated, to the distribution function f(Ni, A), referred to as the inverse of the Mill's ratio (denoted by lambda 2). Lambda is estimated for each individual from the probit function where the probability of being in the sample is Pi = i(Ni, A). The adjusted OLS equation then becomes Ln(Y) = Xi P + 42 + ui where ui, is the error term, Ai is the computed value from the probit,4'j is the covariance between the error. term of the earnings function (Gal) and the error term of the labor supply equation (dii) respectively, and Cis the standard error of 6,2 According to economic theory, the utility maximizing individuals in the sample will allocate their non-leisure time between wage and self-employment activities 12/. An individual will choose self-employment if his expected marginal returns from self-employment are higher than his expected returns from wage employment, and vice versa. 11/ The first stage consists in the estimation of a probit function predicting the probability of being in one type of employment. 121 In fact, mutually exclusive alternatives between wage and self-employment are empirically the most common choices. Therefore, to exclude the option "wage and self-employment" does not represent a mis-specification of the structure of individual choice.  - 17 - Given the large size of our sample, we drew a random subsample of ten percent private sector employees and self-employed individuals. The analysis in the rest of this section refers to this particular subset of workers. We first estimated a probit function determining the choice between wage and self-employment based on the following variables: a dummy variable indicating 1 for urban and 0 for rural place of birth, years of residence in the actual place of residence, two dummies indicating whether the current place of residence is a mid-size city, a large city other than Bogota (Medellin, Cali, Barranquilla), age, and years of schooling. Thus, we assume that the choice of sector of employment depends on the knowledge of and exposure to urban markets and opportunities; the size of the current place of residence which accounts for differential labor market conditions and structure of employment in the various urban centres; the level of education possessed by the individual; and his position in the household. This last variable can be interpreted as a proxy for stability, where household heads are expected to have greater preferences for wage employment because of family obligations. Next, we reestimated our earnings functions conditioned on the actual choice of sector of employment. The probability of being self-employed for the entire sample was estimated in 0.361. Table 6 presents the results of the maximum likelihood estimates of the probit model, which represent the parameters of the underlying structure of the choice between wage and self-employment.  - 18 - Table 6 The odds of being Self-Employed: Probit Estimates Independent variable Coefficient Age 0.077* (4.40) Age squared -0.0007* (3.03) Years of schooling 0.008 (1.03) Urban born 0.187* (1.98) Years resident in city 0.001 (0.57) City size: - Mid-size city 0.388* (4.23) - Large city 1/ 0.309* (3.58) Head of household 0.246* (2.78) Constant 2.751 (8.50) Chi-squared 224.11 Degrees of freedom 8 Sample size 1,886 Notes: Asymptotic t-values in parenthesis. * Statistically significant at the 1 percent level or better. 1/ Omitted category is Bogota. The results have several interpretations. First, as expected from the existence of upper age limits for entering most jobs in the wage sector, the probability of holding a salaried job is clearly higher for younger workers. Second, and contrary to the commonly held view, it  - 19 - appears that urban birth increases the likelihood of being self-employed. Third, residence in a mid-size or large city (other than Bogota) also increases the probability of self-employment. Fourth, also against what is usually proposed by labor market dualists, the coefficient of years of schooling indicates that education by itself is not an important factor determining the choice of sector of employment. Finally, and contrary to our expectations, being a household head not only does not decrease the probability of self-employment, but it increases it. The next step was to reestimate our OLS earnings functions including the adjustment for the probability of being in the sample. The results for the basic earnings functions are found in table 7. Table 7 Basic Earnings Functions by Type of Employment Adjusted for Selectivity Bias Independent Self- Private Sector variable employed employee Years of schooling 0.117* 0.093* (112.5) (18.9) Experience 0.044* 0.049* (2.9) (7.9) Experience squared -0.0006* -0.0007* (2.9) (8.3) Selectivity factor -0.357 0.317* (1.50) (2.09) Constant 8.357 8.269 R2 (adjusted) 0.270 0.354 Sample size 682 1,204 Mean dep. variable 9.758 9.635 (log earnings/month) Notes: t-values in parenthesis. * Statistically significant at the 1 percent level or better.  - 20 - The t-statistic on the coefficient for lambda for the self-employed shows that the selectivity problem is non-significant for this group of workers. This means that being self-employed does not have an impact on their earnings. On the other hand, the lambda coefficient for wage workers indicates otherwise. That is, there is a positive selectivity bias in the earnings of employees. The inclusion of the adjustment term changes the rest of the OLS estimated coefficients, indicating that our previous estimates of the education and experience variables were slightly upwardly biased. However, the returns to an additional year of schooling decrease by approximately only one percentage point for both, self-employed and wage workers. The returns to experience show a similar change. In any case, these adjusted estimates reinforce our previous result showing that human capital raises the individual's productivity regardless sector of employment. Table 8 presents the results of the extended earnings functions including the adjustment for self-selection.  - 21 - Table 8 Extended Earnings Functions Adjusted for Selectivity Bias Independent Self- Private Sector variable employed employees Years of schooling 0.094* 0.079* (10.2) (15.5) Experience 0.032* 0.038* (2.2) (6.0) Experience squared -0.0004* -0.0006* (2.3) (6.8) Log Hours of work 0.259* 0.151* (3.1) (2.8) Social security 0.337* 0.154* (3.3) (4.4) Firm Size 1l: 2-5 workers 0.186* -0.052 (2.9) (0.6) 6-10 workers 0.613* -0.022 (4.1) (0.2) 11+ workers 0.893* 0.082 (7.2) (1.0) Selectivity Factor.; -0.330 0.408* (1.4) (2.75) Constant 7.817 7.756 R2 (adjusted) 0.364 0.389 Sample size 682 1,204 Mean dep.variable (log earnings/month) 9.758 9.635 Notes: t-values in parenthesis. * Statistically significant at the 1 percent level or better.  - 22 - Results in Table 8 show again the existence of an upward bias in our original education and experience estimates, as well as in the impact of hours of work in the earnings of employees and self-employed workers. On the other hand, we verify that the coefficients of social security affiliation were downwardly biased in the original regressions. Additionally, the already detected effect of the firm size for the earnings of the self-employed appears even stronger while totally losing significance for wage workers. Again the coefficient of lambda among employees exhibits positive and significant impact on their earnings stressing the presence of a positive selectivity bias. This indicates that low wage employment probability individuals (a large lambda) who are employees, actually earn more, controlling for the other characteristics that determine earnings. Summarizing the results of this section, the inclusion of self-selection to a particular sector of employment does not invalidate the substance of our previous findings in terms of decent and similar returns to education in the wage and self-employment.  - 23 - V. Alternative Types of Self-employment Thus far we considered the self-employed as one group. However,> examination of the workers engaged in non-wage employment shows that, 19 percent are "employers" (with one or more hired workers), and 81 percent "own account" individuals (working alone). Thus we investigate whether the already detected effect of education on the earnings of the self-employed is similar among the two sub-groups. To do this, we first separated employers (EMP) from one-worker "own account" enterprises (OA), referred to by their acronyms from now on. Table 9 presents the basic characteristics of EMP and OA. Table 9 Mean Characteristics by Type of Self-employment Type of Age Years of Hours of Monthly Sample Self-employment Schooling Work per Earnings Size Week (Pesos) Employer (EMP) 41.1 7.8 54.2 45,131 1,305 (11.4) (4.3) (14.5) (60,664) Own Account (OA) 38.1 6.1 50.7 19,658 5,422 (12.3) (3.9) (15.5) (24,688) Note: Numbers in parenthesis are standard deviations. EMP are on average older, more educated and work more hours than OA. As expected, EMP earn substantially more -- twice as much -- than OA workers. In order to examine whether these differences in earnings imply a differential effect of education on earnings, we estimated the separate earnings functions presented in table 10 for the two types of self-employment.  - 24 - Table 10 Basic Earnings Functions by Type of Self-employment Independent Employers Own Account variable Workers Years of schooling 0.118* 0.118* (21.6) (41.8) Experience 0.039* 0.054* (5.5) (17.2) Experience squared -0.0006* -0.0007* (3.6) (13.9) Constant 8.679 8.015 R2 (adjusted) 0.269 0.258 Sample size 1,295 5,380 Mean dep.variable (log.earnings/month) 10.275 9.518 Note: t-values in parenthesis. * Statistically significant at the 1 percent level of better. Using again the Chow test, we found that the structure of the earnings equations for EMP and OA is not the same (F = 142.5). As expected, the estimates show statistically significant differences in earnings between employers and self-employed who work alone. Interestingly, the returns to schooling in the two forms of self-employment are identical. It may thus be concluded that the economic effect of education is strongly positive not only for wage earners and self-employed, but also for the two groups of self-employed. A one-person enterprise is very likely to have its owner's human capital as its main asset. Enterprises with hired labor are likely to have somewhat larger operations which may imply a higher amount of non-human capital.  - 25 - In a second set of regressions we incorporated the log of hours worked and access to social security obtaining estimates consistent with those shown in section III above. Returns to schooling remained stable, around 11 percent for both forms of self-employment. As we mentioned earlier, we do not have information in this data set to control for the share of the declared income of the self-employed attributable to returns to non-human capital (eg, machines) and it is reminded that EMP should have reported only their income deriving from labor. Attempting to capture somehow the potential bias in our estimates for the self-employed due to the lack of direct control on the owner's non-labor inputs in the business, we added a set of "place of work" dummies (works in the street, in a household, with a vehicle, or in specific working premises) to our regressors as a proxy for non-human assets. We assume that those self-employed working in the street have the lowest amount of non-human capital and that those working in specific premises --such as a factory, office, or store, have the highest amount of physical assets. Table 11 shows the mean characteristics of EMP and OA self-employed according to the place in which they carry out their business.  - 26 - Table 11 Mean Characteristics of Employers and Own Account Workers by Place of Work Employment Age Years of Hours of Monthly Sample Location Schooling Work per Earnings Size Week (Pesos) Street: - Employer 40.7 6.0 52.4 30,722 109 (12.1) (4.0) (15.0) (51,937) - Own Account 36.4 5.0 49.4 14,708 1,705 (12.7) (3.4) (14.7) (16,037) - Both 36.7 5.1 49.6 15,670 1,814 (12.7) (3.5) (14.8) (20,418) Household: - Employer 40.0 6.0 55.2 27,356 399 (11.2) (3.4) (14.6) (39,219) - Own Account 39.3 5.8 50.6 16,243 1,725 (12.6) (3.6) (17.0) (23,156) - Both 39.4 5.8 51.5 18,331 2,124 (12.4) (3.5) (16.7) (27,252) Vehicle: - Employer 39.3 5.9 56.5 27,090 37 (11.8) (3.7) (16.5) (17,608) - Own Account 38.5 6.4 53.4 23,536 656 (11.4) (3.4) (15.3) (18,598) - Both 38.6 6.3 53.6 23,726 693 (11.4) (3.4) (15.3) (18,552) Special Premises: - Employer 41.8 9.0 53.9 57,408 755 (11.4) (4.5) (14.1) (68,956) - Own Account 38.8 7.9 51.1 28,480 1,336 (11.9) (4.5) (14.6) (33,955) - Both 39.9 8.3 52.1 38,969 2,091 (11.8) (4.5) (14.5) (51,487)  - 27 - Table 11 reinforces our previous remarks in terms of the great diversity among the self-employed. Regardless the place of work, EMP earnings are substantially higher than those of the OA self-employed. With the exception of those working in vehicles, EMPs have more education. EMPs and OAs working in special premises exhibit the highest amounts of schooling among the self-employed. Interestingly, and against our expectations, EMPs working in the street appear with much higher mean monthly earnings'than those EMP working in a household or vehicle. Table 12 presents the extended earnings functions for the two groups of self-employed including in the regressors the set of place of work dummies (the excluded category is "works in special premises").  - 28 - Table 12 Extended Earnings Functions for Employers and Own Account Workers Independent variable Employers Own Account Workers (EMP) (OA) Years of schooling 0.089* 0.106* (15.2) (36.1) Experience 0.032* 0.046* (4.7) (14.9) Experience squared -0.0003* -0.0006* (3.2) (12.0) Log Hours of work -0.027 0.299* (0.4) (10.7) Social security 0.372* 0.193* (7.0) (5.0) Works in household -0.321* -0.315* (6.5) (11.8) Works in street -0.313* -0.281* (4.1) (10.2) Works with vehicle -0.230 0.076 (1.9) (2.2) Constant 9.198 7.222 R2 (adjusted) 0.331 0.312 Sample size 1,295 5,380 Mean dep.variable (log earnings/month) 10.275 9.518 Notes: t-values in parenthesis. * Statistically significant at the 1 percent level or better. The above results exhibit several interesting features. First, the returns to education behave quite differently for EMP and OA when controlling for place of work. The coefficient of the education variable  - 29 - (years of schooling) in the EMP regression drops two percentage points compared to our previous results, while the same coefficient for the OA renains almost unchanged. In our view, this effect possibly reflects the relevance of non-human capital assets for employers -- an item that should have been excluded when responding to the survey. Second, the previously found impact of the amount of labor supplied (hours of work) by the EMP disappears, while it maintains both level and significance for the OA self-employed. Third, there is a strong and negative effect on earnings from operating a street or a household business. Regardless whether the individual is an employer or works on his own account, operating a street or household business is associated with lower earnings. Surprisingly, this negative effect is stronger for those operating a household business than it appears for those working in the street. Chiswick (1980) suggests that negative earnings differentials for household enterprises could be associated with less efficient firms, with a greater use of unpaid family labor, or with a relatively high proportion of in-kind income. Regrettably, this data set does not permit to test any of these hypotheses. Nevertheless, we explored the individual characteristics of those operating street and household businesses. We looked at the years operating the enterprise, the position of the worker in the household (main or secondary worker), the total years of working experience and the migratory status. We did not find any individual characteristic which differentiates household and street self-employed from the rest. The only differences we found lie in the sector of employment. Nearly 80 percent of the self-employed operating street businesses and 45 percent of the self-employed operating in households work in the retail commerce and the  - 30 - services sectors. Twenty three percent of those in household businesses work in the construction sector. If earnings in these activities are lower than in the rest, this could account for the detected negative earnings differentials of the self-employed in household and street enterprises. However, results from regressions including economic sector dummies did not support this hypothesis. In fact, the economic sector was positive and significant only for EMPs and OAs working in agricultural activities. VI. Concludint Remarks There are several conclusions stemming from the above analysis of earnings and education among different types of workers in Colombia and over time comparisons of the returns to education. (a) The non-competitiveness of LDC labor markets regarding the earnings-productivity debate might have been exaggerated. In fact, restricting the analysis to the non-screened competitive-market self-employed yields an average rate of return to investment in education of the order of 13 percent, as compared to a rate of return of 9.5 percent based on the earnings of the most non-competitive segment of the market -- the public sector employees. (b) The issue of selectivity bias in standard human capital earnings functions might have been exaggerated as well. When correcting for such bias in this data set, the returns to education fell by about only one percentage point. (c) The issue of overinvestment in education and the declining returns over time may have been exaggerated as well. In the 10 year 1974-1984, the educational attainment of the Colombian  - 31 - urban labor force has doubled. Yet the returns to education have fallen by only 1 to 2 percentage points. The present level of returns the order of 10 percent (according to the most conservative estimates), justifies further expansion of education. In particular, expansion of secondary education seems the appropriate target, given the fact that the urban labor force has already attained near universal primary education. (d) The growing informal sector in LDC economies does not only have a residual employment, safety valve function -- those employed in it make full use of their human capital stock, acquired by education and, according to our estimates, realize returns similar or even higher, to those engaged in more conventional types of employment.  - 32 - APPENDIX Table A-1 Previous Earnings Functions Results for Colombia Schultz 1Z Bourgoulgeon 2/ Psacharopoulos 4 t Horn 1979 5/ Independent Bogota Bogota BQgta 1974 Colombla 1975 Colombia 1979 variable 1965 1971 Wage Self-m- Pubilc S. Private S. Piblic Private Self-em- L.Force L.Force Workers ployed 3/ Workers Workers Wage Wage ployed Years of 0.173* 0.167' 0.127* 0.132* 0.134* 0.146* 0.129* 0.126* 0.148* schooIrg (13.4) (38.9) (31.1) (12.0) (44.2) (73.8) (30.46) (52.25) (29 .95) Experience 0.121* 0.078* 0.067r 0.057* 0.041' 0.070' 0.055* 0.068* 0.082* (8.8) (17.6) (22.5) (8.1) (11.2) (32.3) (10.36) (27.46) (12.65) Exper.squared -0.001* -0.001' -0.000* -0.000* -0.000* -0.001* -0.001* -0.001' -0.001' (7.3) (12.6) (15.8) (7.0) (7.3) (23.3) (6.99) (19.79) (10.70) Constant 4.30 5.08 - - 6.38 5.88 6.995 6.869 6.637 R2 0.881 0.629 0.385 0.237 0.606 0.515 0.543 0.452 0.367 Sample size 722 1,016 2,160 762 1,376 5,507 816 3718 1666 Note: All estimates refer to males. t-values in parenthesis. * Statistically significant at 1 the percent level or better. - omitted In orlignal source. Source: 1/ Schultz (1968). Sample refers to the Bogota male labor force. Surveys carried out by CEDE, Universidad de los Andes, Bogota, between 1963 and 1966. 2/ Bourgouignon (1980). Sample refers to the Bogota male labor force. Data from employment surveys carried out by DANE in 1971 and 1974. 3/ This sub-sample excludes individuals with university education. 4/ Psacharopoulos (1983). Sample refers to urban male workers. Private sector workers include wage earners and self-employed. Data from urban labor markets survey carried out by DANE in 1975. 5/ Horn (1986). Sample refers to urban male labor force in Bogota, Medellin, Cali, Barranquilla, Bucaramanga, Cucuta, Manizales, Pesto, Pereira, and Villavicencio. Data from employment survey carried out by DANE in 1979.  - 33 - REFERENCES Blau, D.M., "Self-employment, Earnings, and Mobility in Peninsular Malaysia," World Development, 14, 7, 1986: 839-853. Blaug M., "An Economic Analysis of Personal Earnings in Thailand", Economic Development and Cultural Change, October 1974, pp. 1-32. Blinder A., "Wage Discrimination: Reduced Form and Structural Estimates", Journal of Human Resources, 8, No.4, 1973, pp.436-455. Bourguignon F., "The Role of Education in the Urban Labor Market During the Process of Development: The Case of Colombia", Paper Presented at the VI World Congress of the International Economic Association, Mexico, August 1980. Cain G., "The Challenge of Segmented Labour Market Theories to Orthodox Theory",Journal of Economic Literature, 14, No.4, 1976, pp. 1215-1257. Chiswick C., "On Estimating Earnings Functions for LDCs", Journal of Develop-ment Economics, 3, 1977, pp.67-78. ------- "Analysis of Earnings from Households Enterprises: Methodology and Application to Thailand", Review of Economics and Statistics, Vol. 65, No. 4, 1983, pp. 658-662. DANE, "Metodologia del Diseno Muestral de las Encuestas de Hogares", Bogota, 1984. Doeringer P., Piore M., Internal Labor Markets and Manpower Analysis, Lexington Massachussets-Heath Lexington Books, 1971. Fields G., "Education and Economic Mobility in a Less Developed Country", Economic Growth Center, Yale University, Center Discussion Paper No. 237, revised version, June 1976. Fields G.,"Analyzing Colombian Wage Structure", The World Bank, Studies in Employment and Rural Development No. 46, May 1978. Fields G., Schultz P., "Sources of Income Variations in Colombia: Personal and Regional Effects", Paper No. 272, Economic Growth Center, Yale University, 1977. ------ "Income-Generating Functions in a Low Income Country: Colombia", The Review of Income and Wealth, March 1982, pp. 71-87.  - 34 - Gusinger S., Henderson J., Scully G., "Earnings, Rates of Return to Education and the Earnings Distribution in Pakistan", Economics of Education Review, 3,4, 1984: 257-268. Heckman J., "The Common Structure of Statistical Models of Truncation, Sample Selection and Limited Dependent Variables and a Simple Estimator For Such Models", Annals of Economic and Social Measurement, 5/4, 1976,m pp. 475-492. ----- "Sample Selection Bias as a Specification Error", Econometrica, vol.47, January 1979, pp. 153-161. Henderson, J., "Earnings Functions for the Selfemployed. Comment", Journal of Development Economics, 13, 1982, pp.97-102. House, W., "Labor Market Segmentation: Evidence From Cyprus", World Development, vol.12, No.4, 1984. pp. 4003-417. Horn R., "The Earnings Impact of Education Experience and Formal Job Training (SENA) Among the Self-employed in Colombia", Education and Training Department, The World Bank, 1986. Krause W., "An Empirical Study of Labour Market Segmentation: A Comment", Industrial and Labour Relations Review, 30, No.2, 1977, pp. 219-224. Kugler B., " Influencia de la Educacion en los Ingresos del Trabajo: El Caso Colombiano", Revista de Planeacion y Desarollo, Enero 1975. Kuznetz, S., Modern Economic Growth, Rate, Structure and Spread, New Haven, Yale University Press, 1966. Mazumdar, D., "Labor Supply in Early Industrialization: The Case of the Bombay Textile Industry", Economic History Review, August, 1973. ------ The Urban Informal Sector", World Development, vol. 4, No.8, August 1976, pp.655-679. ------ "Paradigms in the Study of Urban Labour Markets in LDCs: A Reassessment in the Light of an Empirical Survey in Bombay City", World Bank Staff WorkinR Paper, No.366, 1979. ------ "Segmented Labor Markets in LDCs", American Economic Review, vol.73, No.2, May 1983, pp. 254-259. Mincer J., Schooling Experience and Earnings, Columbia University Press, New York, 1974. Mohan R., "The Determinants of Labor Earnings in Developing Metropolis: Estimates for Bogota and Cali, Colombia", Urban and Regional Economics Division, DRD, The World Bank, November 1980.  - 35 - Osterman P., "An Empirical Study of Labor Market Segmentation", Industrial and Labour Relations Review, 28, No.4, 1975, pp. 508-523. PREALC, Sector Informal. Funcionamiento y Politicas, Oficina Internacional del Trabajo, Chile, 1978. Psacharopoulos, G., "Education and Private Versus Public Sector Pay", Labor and Society, vol. 8, No. 2, April-June 1983. Psacharopoulos G., Arriagada A.M., "The Educational Attainment of the Labor Force: An International Comparison", International Labor Review, Vol. 125, No. 5, December 1986, pp: 561-574. Schultz P., "Returns to Education in Bogota, Colombia", Rand Corporation, RM 5645 RC/AID, 1968. SER, La Magnitud y las Caracteristicas Socioeconomicas del Sector Informal Urbano: Personas, Hogares y Negocios, IFT-104, Instituto SER de Investigacion, Bogota, March 1986.