Once an entrepreneur, always an entrepreneur? The impacts of skills developed before, during and after college on firm start-ups

Table 5 Trivariate probit model of academic skill diversity, work experience diversity and entrepreneurship

Variable	Academic Diversity (S)	Work Diversity (O)	Entrepreneurship (E)
Male	0.138	0.195	0.159
	(3.03)***	(2.10)**	(3.04)***
Ethnicity	−0.112	−0.214	0.222
	(1.30)	(1.21)	(2.53)*
High school class rank	−0.001	−0.002	−0.005
	(1.27)	(1.57)	(6.30)***
Number of siblings	0.006	0.011	0.013
	(0.45)	(0.41)	(0.89)
Grow up with two parents in household	−0.105	0.110	−0.160
	(1.48)	(0.78)	(2.09)**
Father education	0.014	0.009	0.032
	(0.88)	(0.27)	(1.82)*
Mother education	−0.037	0.036	−0.050
	(2.09)**	(1.01)	(2.60)***
Either of parents started a business	0.045	0.328	0.240
	(0.99)	(3.49)***	(4.62)***
Graduation years
1987-1991	0.069	0.114	−0.140
	(0.95)	(0.76)	(1.78)*
1992-1996	−0.106	−0.308	−0.242
	(1.50)	(2.20)**	(3.19)***
1997-2001	−0.169	−0.623	−0.468
	(2.37)**	(4.45)***	(5.94)***
2002-2006	−0.153	−1.227	−0.692
	(2.17)**	(8.48)***	(8.44)***
Constant	−0.458
	(4.59)***
Parameters for entrepreneurial human capital
\( {\beta}_{\lambda}^O \)	9.565[1.540]***
\( {\beta}_{\lambda}^E \)	1.183[0.257]***
σ ²	0.044[0.014] ***
Number of observations	5295

Note: Dependent variables are binary choices. S is equal to one for lower 25th percentile of AC_SPEC at ISU, and zero for the remaining 75% unspecialized alumni. O is equal to one if the number of occupations held by the end of year 2007 is equal or greater than two, which is the 75th percentile of OCCUPATIONS; E is equal to one if individual has ever started a business and remain as an entrepreneur till the end of year 2007. Probability weights are considered in the model and the standard errors are therefore robust. The number in the bracket is the standard error of the corresponding estimate. *, ** and ***represent statistical significance at 10%, 5% and 1%, respectively. Absolute value of z statistics are in parentheses and standard error in square bracket. \( {\beta}_{\lambda}^S \) is normalized to be one to identify the model.