Estimation Results Directory UMM :Data Elmu:jurnal:J-a:Journal of Economics and Business:Vol52.Issue6.2000:

V. Estimation Results

The results of probit estimation of Equation 1 are presented in Table 4. It can be seen that variables that significantly affect both men and women tend to affect them in the same way. As Tables 2 and 3 suggested, being White raises the likelihood of self-employment, whereas living in an urban location decreases it, for both men and women. This likelihood significantly increases with age, though at a diminishing rate. Table 2. Descriptive Statistics Full-time, Full-year, Nonagricultural Male Workers Wage-and-Salary Workers Self-Employed Workers Mean natural log earnings 10.25 10.30 Percent White 86.6 93.1 Percent living in cities 76.3 75.3 Percent high school graduates 84.9 86.6 Percent college graduates 26.2 34.8 Mean age 40.1 43.9 Percent married 80.4 86.1 Percent with health limitations 5.7 6.1 Percent living in Northeast 21.3 22.6 Percent living in Midwest 24.3 21.7 Percent living in West 20.8 23.5 Percent white-collar 48.9 62.4 Percent service 8.2 3.8 Percent blue-collar 42.8 33.9 Percent fluent 98.0 98.5 Number of observations 33,990 4,025 Table 3. Descriptive Statistics Full-time, Full-year, Nonagricultural Female Workers Wage-and-Salary Workers Self-Employed Workers Mean natural log earnings 9.82 9.43 Percent White 82.9 89.7 Percent living in cities 76.9 70.4 Percent high school graduates 88.0 88.2 Percent college graduates 22.8 22.3 Mean age 39.5 42.5 Percent married 66.0 74.5 Percent living with own minor children 42.3 42.1 Percent with health limitations 4.9 5.7 Percent living in Northeast 20.6 16.9 Percent living in Midwest 23.0 25.5 Percent living in West 19.9 26.2 Percent white-collar 75.5 64.6 Percent service 10.8 29.7 Percent blue-collar 13.6 5.7 Percent fluent 98.2 97.9 Number of observations 25,290 1,377 Gender Differences in Self-Employment 505 The probit results provide evidence that the likelihood of self-employment increases with education for both men and women; having a high-school diploma is critical for women, whereas having a college diploma is critical for men. Living in the west raises the likelihood of self-employment for both men and women; geographic location affects men and women differently and less significantly in the northeast and midwest. Individuals in white-collar occupations are uniformly more likely to be self-employed than individuals in blue-collar occupations. Among men, service workers appear to be the least likely candidates Table 4. Probit Results for Self-Employment Status:Full-time, Full-year, Nonagricultural Workers a Variable Female Equation Coefficient Male Equation Coefficient INTERCEPT 23.713 23.911 .0001 .0001 HS 0.083 20.037 .0694 .1829 COLL 0.040 0.081 .2447 .0002 HLTHLIM 0.068 0.026 .2567 .4953 WHT 0.299 0.326 .0001 .0001 AGE 0.065 0.099 .0001 .0001 AGESQ 20.0006 20.0010 .0012 .0001 CITY 20.156 20.074 .0001 .0006 FLUENT 20.194 20.059 .0522 .4062 NEAST 20.057 0.029 .1474 .2348 MWEST 0.056 20.057 .1185 .0187 WEST 0.193 0.100 .0001 .0001 WHTCLL 0.243 0.175 .0001 .0001 SERV 0.926 20.266 .0001 .0001 DKIDS 0.035 .2610 MARRIED 0.554 0.660 .1898 .0613 MARRIEDAGE 20.031 20.030 .1263 .0874 MARRIEDAGESQ 0.0004 0.0004 .0899 .0689 SPINCOM 5.4810 6 4.3510 7 .0001 .4843 Log likelihood 25016.2 212343.2 Number of observations 26,667 38,015 a p-values are in parentheses. Significant at the 10 level, significant at the 5 level, significant at the 1 level. 506 S. H. Clain for self-employment, whereas workers in this occupation are the most likely candidates for self-employment among women. In and of itself, being married raises the likelihood of self-employment for men. For women, the effect of marriage works most strongly through the spouse’s income: the greater the income, the greater the likelihood of self-employment. For both genders, the effect of marriage on the likelihood of self-employment tends to diminish with age. 18 These estimation results are used to construct selectivity variables for the OLS estimation of the earnings Equations 2 and 3, to avoid sample-selection bias. Tables 5 and 6 present the findings for men and women. Some findings are consistent across gender and type of employment. Earnings tend to increase with education. Health limitations tend to have a negative impact on earnings, whereas living in an urban location tends to raise earnings. Workers in service occupations tend to have the lowest earnings, ceteris paribus. Other findings vary by gender andor type of employment. For example, among wage-and-salary workers, those living in the south tend to have the lowest earnings, ceteris paribus. The same cannot be said of self-employed workers; indeed, for self- employed workers, the impact of regional location varies by gender. In many cases where findings vary across the four types of workers studied, self- employed women stand out as uniquely different. Being White and being married tend to have negative impacts on the earnings of self-employed women, whereas they have positive impacts on the earnings of the three other groups of workers studied. Also, although aging tends to enhance the earnings of the three other types of workers, it erodes the earnings of self-employed women. These patterns may reflect differences in the length of workweek andor workyear worked by self-employed women within the ranges defined as full-time and full-year, by race, marital status and age, and the subsequent impacts on annual earnings. Alternatively, they may reflect the presence of greater constraints andor discriminatory elements faced by self-employed women. 19 One of the most intriguing differences in the findings across the four types of workers is perhaps the most subtle: the signs of the coefficients of the selectivity variables, l se and l ws, are opposite by gender. For men, the findings suggest that the sorting between wage-and-salary employment and self-employment is consistent with positive selectivi- ty. 20 Rees and Shah 1986 interpreted dual positive selectivity as each group having a comparative advantage in its chosen employment. 18 In an effort to check for significant gender differences, men and women were pooled into a single sample for subsequent estimation. A dummy variable for gender was created, as were interaction terms of this dummy variable and all other X variables. Significant interaction terms in a probit estimation using the pooled data pinpointed significant gender differences in the coefficients in the probit equations for men and women. The variables found to be significant at the 10 level were: HS, CITY, NEAST, MWEST, WEST, SERV, and SPINCOM. 19 In the estimation of a full specification of the self-employment earnings equation, for a pooled sample of self-employed men and women, gender differences were found to be significant for the coefficients of l se, WHT, AGE, AGESQ, CITY, WEST, WHTCLL, MARRIED, MARRIEDAGE, MARRIEDAGESQ, and SPINCOM, at the 10 level. In the estimation of a full specification of the wage-and-salary earnings equation, for a pooled sample of men and women employed in the wage-and-salary sector, gender differences were found to be significant for the coefficients of l ws, COLL, HLTHLIM, AGE, AGESQ, CITY, FLUENT, MWEST, WEST, WHTCLL, SERV, MARRIED, MARRIEDAGE, MARRIEDAGESQ, and SPINCOM, at the 5 and 10 levels. 20 The product of the coefficient and the selectivity variable reflects the expected value of the error term in the earnings equation, given the self-selected sample. Since l se .0 and l ws ,0, the products of the observed coefficients and the respective variables are positive. That is, the expected values of the error terms in each equation, conditioned on the sample, are greater than zero, making them greater for the individuals who have selected into the type of employment than for those who have not. Gender Differences in Self-Employment 507 Table 5. OLS Estimation of Earnings Equations: Full-time, Full-year, Nonagricultural Male Workers a Variable Self-Employed Coefficient Wage-and-Salary Coefficient INTERCEPT 27.324 7.889 .0864 .0001 HS 0.216 .0001 COLL 0.707 0.252 .0001 .0001 HLTHLIM 20.074 20.148 .2530 .0001 WHT 1.345 0.079 .0001 .0001 AGE 0.359 0.057 .0001 .0001 AGESQ 20.0036 20.0006 .0001 .0001 CITY 0.073 0.215 .2966 .0001 FLUENT 0.287 .0001 NEAST 0.251 0.113 .0001 .0001 MWEST 20.182 0.077 .0044 .0001 WEST 0.446 0.058 .0001 .0001 WHTCLL 0.893 0.057 .0001 .0001 SERV 21.089 20.112 .0001 .0001 MARRIED 1.766 0.018 .0459 .8336 MARRIEDAGE 20.070 0.004 .0886 .4203 MARRIEDAGESQ 0.0009 20.00003 .0587 .5702 SPINCOM 1.3810 6 .0099 l se 3.932 .0002 l ws 21.089 .0001 F statistic 60.97 868.9 .0001 .0001 R-square 0.186 0.315 Number of observations 4,025 33,990 a p-values are in parentheses. Significant at the 10 level, significant at the 5 level, significant at the 1 level. 508 S. H. Clain Table 6. OLS Estimation of Earnings Equations: Full-time, Full-year, Nonagricultural Female Workers a Variable Self-Employed Coefficient Wage-and-Salary Coefficient INTERCEPT 20.947 7.627 .0001 .0001 HS 0.195 .0001 COLL 0.286 0.352 .0006 .0001 HLTHLIM 20.105 20.071 .3965 .0001 WHT 20.627 0.073 .0407 .0001 AGE 20.203 0.075 .0056 .0001 AGESQ 0.0020 20.0008 .0085 .0001 CITY 0.753 0.168 .0001 .0001 FLUENT 0.351 0.145 .1430 .0001 NEAST 0.251 0.105 .0090 .0001 MWEST 20.196 0.031 .0326 .0001 WEST 20.300 0.150 .1274 .0001 WHTCLL 20.284 0.142 .2954 .0001 SERV 22.582 0.018 .0042 .5724 DKIDS 20.175 20.059 .0147 .0001 MARRIED 24.143 0.423 .0001 .0001 MARRIEDAGE 0.188 20.029 .0004 .0001 MARRIEDAGESQ 20.0022 0.0003 .0006 .0001 SPINCOM 21.0710 5 4.5610 6 .0404 .0001 l se 22.840 .0123 l ws 1.159 .0001 F statistic 24.07 582.45 .0001 .0001 R-square 0.242 0.305 Number of observations 1,377 25,290 a p-values are in parentheses. Significant at the 10 level, significant at the 5 level, significant at the 1 level. Gender Differences in Self-Employment 509 For women, however, the sorting between wage-and-salary employment and self- employment reflects negative selectivity. 21 In finding negative selection into self- employment among Hispanics and Asians, Borjas and Bronars 1989 argued it to be consistent with the existence of consumer discrimination, which reduces the gains from self-employment for the most able members of a minority group. To explore these findings further, the estimated equations are used to predict wage- and-salary earnings, self-employment incomes, and probabilities of self-employment, for the given samples of workers, by gender. For purposes of comparison, the equations estimated for males are also applied to females. The mean predicted values are shown in Table 7. 22 These calculations reiterate the gender differences in earnings, by type of worker, found in Tables 2 and 3: the gender gap in predicted earnings of workers in their chosen work is greater in self-employment 9.426 vs. 10.297 than in wage-and-salary employ- ment 9.823 vs. 10.265. This result is consistent with Moore 1983a. 21 Blau 1985 reasoned that negative selection into farm self-employment in Malaysia tended to support the hypothesis of a noncompetitive labor market in a developing country. 22 To calculate the probabilities, the sample mean value of c is calculated. This value is then used to evaluate 1-Fc, yielding the probability of self-employment for an individual with that sample mean value of c Table 7. Mean Predicted Earnings and Probabilities of Self-Employment Self-Employment Income a Potential Wage-and-Salary Income a Probability of Self-Employment Self-Employed Males 10.297 10.396 .1214 95 CI 10.283, 10.310 10.386, 10.405 99 CI 10.279, 10.314 10.383, 10.408 Females 9.426 9.793 .0725 95 CI 9.398, 9.455 9.775, 9.812 99 CI 9.389, 9.463 9.769, 9.818 Females b 10.113 10.295 .0997 95 CI 10.091, 10.135 10.277, 10.313 99 CI 10.083, 10.142 10.271, 10.319 Wage-and-Salary Income a Potential Self-Employment Income a Probability of Self-Employment Wage-and-Salary Males 10.265 10.152 .0935 95 CI 10.261, 10.269 10.148, 10.157 99 CI 10.260, 10.270 10.146, 10.158 Females 9.823 9.539 .0401 95 CI 9.819, 9.826 9.533, 9.545 99 CI 9.818, 9.828 9.531, 9.547 Females b 10.266 10.165 .0933 95 CI 10.262, 10.270 10.160, 10.170 99 CI 10.162, 10.171 10.159, 10.172 a The natural logarithms of earnings are reported. b These calculations are made using the personal characteristics of the female sample, but the equations estimated from the male sample. 510 S. H. Clain Simulations of outcomes for individuals having the personal characteristics of the female samples, but the equations estimated for men, shed light on the sources of difference. For wage-and-salary workers, the gender gap essentially disappears in the simulated results 10.266 vs. 10.265. That is, differences in the structure of the equations, not in the personal characteristics, account for the gender differences in wage-and-salary earnings. 23 For self-employed workers, the gap is substantially, but not fully, narrowed in the simulated results 10.113 vs. 10.297; differences in the structure of the equations and in the personal characteristics of self-employed men and women contribute to the gender gap in earnings in this sector. Table 7 also reports the mean predicted earnings of workers in the type of work not chosen. Interestingly enough, predicted wage-and-salary earnings are always higher than predicted self-employment earnings. 24 This finding is understandable for workers who have chosen wage-and-salary employment: their choice is consistent with wage maximi- zation. To understand the result for workers who have chosen self-employment, one must assume that there are desirable nonwage characteristics of self-employment that make it a rational choice for these workers in the maximization of expected utility. One could argue that nonwage considerations must influence women to a greater extent; they forfeit a larger percentage of earnings in choosing self-employment than do men. 25 Finally, these calculations reflect the gender differences in selectivity into self- employment. Men who choose self-employment have personal characteristics that are more highly valued in the marketplace than men who choose wage-and-salary employ- ment: the former’s earnings would be higher, on average, no matter which sector they chose 10.396 vs. 10.265, if a wage-and-salary worker; 10.297 vs. 10.152, if self- employed. 26 The opposite is true for women: self-employed women have personal characteristics that are less highly valued in the marketplace than women in wage-and- salary employment 9.793 vs. 9.823, if a wage-and-salary worker; 9.426 vs. 9.539, if self-employed. 27–29 23 The literature exploring gender differences among wage-and-salary workers generally suggests that differences in personal characteristics do play a role in explaining the gender wage-gap in this sector. To measure the contribution of differences in personal characteristics accurately, however, studies rely on more detailed information on occupation, field of study in formal education, and labor market experience. In summarizing these studies, Blau and Ferber 1992 reported that differences in these personal characteristics explained 60 to 66 of the wage gap between men and women with the same education level. 24 A comparison of confidence intervals for these values suggests that these differences are significant 25 Previous researchers have suggested that the ease of underreporting income to avoid taxes is one attraction of self-employment Moore, 1983b; Blau, 1987; Yuengert, 1994. If so, this practice could account for lowered self-employment income, but would not account for the gender difference, unless one were willing to assume women more guilty of it than men. 26 A comparison of confidence intervals for these values suggests that these differences are significant. 27 A comparison of confidence intervals for these values suggests that the differences are significant, though the former only weakly so. 28 Self-employment has been promoted by a number of states and the federal government as a way for individuals to exit the welfare and unemployment insurance programs. See discussion and studies cited in Fairlie and Meyer 1996. Although it is possible that this activity has affected women more than men and has contributed to these findings, it cannot be determined conclusively from the present data set. 29 In the comparisons of mean predicted earnings between the two types of workers, positive negative selectivity of men women into wage-and-salary employment is masked by the systematic differences in personal characteristics of individuals who choose wage-and-salary employment. In and of themselves, these differences in characteristics tend to reduce increase wage-and-salary income of men women in the wage-and-salary sector, making it smaller greater than the self-employment income of self-employed men women. Gender Differences in Self-Employment 511 These results are consistent with the sociological theory that low-wage workers i.e., females are pushed into entrepreneurship, whereas high-wage workers i.e., males are pulled into entrepreneurship by attractive opportunities. However, the result for women contradicts the results of Devine 1994b. Devine used a different data source, with part-time or part-year workers as well as full-time and full-year workers, a different measure of earnings and a methodology not correcting for sample selection bias. She found potential wage-and-salary hourly earnings for the average self-employed female that exceeded the potential wage-and-salary hourly earnings of her wage-and-salary counterpart. This contrast raises the possibility that the mechanisms that determine self-employment status for women differ by length of workweek andor workyear sought. Indeed, the length of workweek andor workyear may be jointly endogenous with the choice of sector. 30 Although this consideration is beyond the scope of the current analysis, it certainly indicates an area in need of further investigation.

VI. Conclusion