have earned US971 less on average had they entered, those who make a transition would have earned an average of US4215 more if they had remained in
a wage-and-salary job. This larger difference could be interpreted as the implicit value of the nonpecuniary benefits in self-employment.
5. Results and discussion
Since this is the first empirical study of self-employment transitions to attempt to control for initial conditions bias, it is useful to first gauge the impact of this
particular part of the estimation. Table 6 presents results from three separate random effects probits. Tax variables are left out of these probits for the purpose
Table 6 Pooled transition probit results: effects of initial conditions bias correction
Ž Ž
Variable Larger sample includes
Smaller sample excludes Smaller sample
initially and previously initially and previously
with correction .
. self-employed individuals
self-employed individuals Ž
. Ž
. Ž
. Age
0.020 0.028 y0.104 0.066
y0.091 0.052 Ž
. Ž
. Ž
. Age squaredr100
y0.018 0.037 0.151 0.094
0.122 0.077 Ž
. Ž
. Ž
. Tenure
y0.003 0.001 y0.005 0.002
y0.006 0.001 Ž
. Ž
. Ž
. Tenure squaredr100
0.054 0.024 0.079 0.059
0.084 0.046 Ž
. Ž
. Ž
. Dropout
0.011 0.073 0.054 0.126
0.031 0.124 Ž
. Ž
. Ž
. Some college
y0.018 0.060 y0.019 0.095
y0.014 0.096 Ž
. Ž
. Ž
. College graduate
0.025 0.069 y0.013 0.107
0.010 0.104 Ž
. Ž
. Ž
. Post-college Ed.
0.054 0.068 0.083 0.116
0.114 0.105 Ž
. Ž
. Ž
. Non-white
y0.142 0.086 y0.315 0.153
y0.283 0.141 Ž
. Ž
. Ž
. North Central
0.0004 0.062 y0.026 0.101
y0.030 0.104 Ž
. Ž
. Ž
. South
y0.041 0.063 y0.101 0.101
y0.106 0.098 Ž
. Ž
. Ž
. West
0.061 0.068 0.082 0.103
0.067 0.095 Ž
. Ž
. Ž
. Married
y0.028 0.064 0.118 0.111
0.100 0.095 Ž
. Ž
. Ž
. Income from capital
0.004 0.003 y0.006 0.015
y0.011 0.012 Ž
. Ž
. Ž
. Part-time
0.094 0.053 0.142 0.093
0.140 0.084 Ž
. Ž
. Ž
. Kids 1 to 2
0.029 0.042 y0.057 0.068
y0.049 0.063 Ž
. Ž
. Ž
. Kids 3 to 5
0.076 0.039 0.225 0.065
0.244 0.059 Ž
. Ž
. Ž
. Kids 6 to 13
0.027 0.025 0.035 0.058
0.050 0.053 Ž
. Ž
. Ž
. Kids 14 to 17
y0.004 0.037 y0.206 0.164
y0.206 0.115 Ž
. Ž
. Ž
. Union
y0.330 0.056 y0.480 0.114
y0.383 0.109 Ž
. Ž
. Ž
. Unemployment rate
y0.020 0.008 y0.047 0.015
y0.047 0.012 Ž
. Ž
. Ž
. MSA
y0.052 0.042 y0.048 0.075
0.017 0.077 Ž
. Mills Ratio
– –
0.958 0.506 N
16,026 5622
5622 Sample transition
0.039 0.033
0.033 probability
Ž .
Entries are random-effects probit coefficients with robust and bootstrapped, for column 3 standard errors in parentheses. Regressions also include indicators for the year of the observation and a constant
term. Statistically significant at the 10 level.
Statistically significant at the 5 level.
of isolating the impact of the initial conditions correction. The first column follows the work in earlier studies, and contains results from a random effects probit on all
person-years of pooled data without any controls for initial conditions bias. My correction procedure requires that I analyze only those individuals who
Ž .
have never previously been self-employed in the panel period , so the second column repeats the process after eliminating all observations from individuals who
were either initially or previously self-employed or who lack information for any of the control variables used in the correction procedure. The third and final
column in Table 6 presents results from the correction procedure, using an inverse
Ž Mills ratio from the first-stage probit of the initial condition Tables 7 and 8
present results from first-stage selection and instrumenting regressions. Also, Table 9 contains results from a baseline probit without random effects. Note that
magnitudes of the coefficients and patterns of significance are essentially un-
. changed in the absence of the random effects .
Comparing the first two columns shows that the different sample required for the initial conditions correction results in only minor differences in patterns of
significance for the control variables. Perhaps more importantly, adding the selection term in column 3 has virtually no effect on signs and significance
patterns for the other control variables. While the coefficient on the Mills ratio is rather large and significant in its own right, the fact that its inclusion has no
Table 7 Initial conditions probit results
Variable No endogeneity control
Ž .
Age y0.078 0.056
Ž .
Age squaredr100 0.065 0.078
Ž .
Dropout y0.050 0.147
Ž .
Some college y0.051 0.124
Ž .
College graduate 0.115 0.146
Ž .
Post-college education 0.216 0.196
Ž .
Union 0.621 0.165
Ž .
Veteran 0.047 0.127
Ž .
Unemployment rate 0.007 0.018
Ž .
MSA 0.323 0.097
Ž .
Nonwhite 0.138 0.184
Ž .
Married y0.035 0.125
Ž .
Income from capital y0.021 0.011
Ž .
Part-time y0.040 0.110
Ž .
Number of children 0.041 0.050
N 1391
Entries are probit coefficients with standard errors in parentheses. This regression also includes indicators for the year of the observation and a constant term. The dependent variable is a dummy
which equals 1 for wage-and-salary employment in the first observed job, and zero for self-employed. Statistically significant at the 10 level.
Statistically significant at the 5 level.
Table 8 Instrumental variables regression results
Variable ATR
MTR Income
Ž .
Ž .
Ž .
Instrumental variable 0.897 0.087
0.519 0.016 0.15 0.19
Ž .
Ž .
Ž .
Age y1.223 0.273
y1.196 0.241 y223.57 262.72
Ž .
Ž .
Ž .
Age squared 0.015 0.004
0.014 0.003 y0.94 3.75
Ž .
Ž .
Ž .
Tenure 0.019 0.006
0.011 0.005 18.11 5.54
Ž .
Ž .
Ž .
Tenure squared y0.00007 0.00003
y0.00007 0.00002 y0.03 0.02
Ž .
Ž .
Ž .
Dropout 4.727 0.718
3.665 0.674 2416.26 810.86
Ž .
Ž .
Ž .
Some college y1.904 0.511
y0.808 0.482 y1536.39 582.76
Ž .
Ž .
Ž .
College graduate y4.647 0.575
y5.815 0.552 y7617.04 654.70
Ž .
Ž .
Ž .
Post-college education y6.180 0.662 y5.041 0.648
y14,653.62 749.76 Ž
. Ž
. Ž
. Non-white
y1.625 0.692 y1.727 0.642
y3167.84 754.57 Ž
. Ž
. Ž
. North Central
y0.102 0.532 0.889 0.494
445.02 580.31 Ž
. Ž
. Ž
. South
y0.769 0.507 0.353 0.467
y1789.82 540.57 Ž
. Ž
. Ž
. West
0.441 0.563 1.193 0.521
y107.02 608.43 Ž
. Ž
. Ž
. Married
0.425 0.426 0.833 0.370
918.17 394.21 Ž
. Ž
. Ž
. Income from capital
0.018 0.051 0.086 0.044
131.16 46.02 Ž
. Ž
. Ž
. Part-Time
y1.356 0.366 y1.296 0.310
y1114.90 323.89 Ž
. Ž
. Ž
. Kids 1 to 2
0.442 0.242 0.123 0.206
634.92 214.00 Ž
. Ž
. Ž
. Kids 3 to 5
0.390 0.248 0.224 0.212
403.88 221.83 Ž
. Ž
. Ž
. Kids 6 to 13
y0.053 0.218 0.153 0.191
y50.43 207.51 Ž
. Ž
. Ž
. Kids 14 to 17
y0.041 0.439 y0.596 0.375
y210.50 393.76 Ž
. Ž
. Ž
. Union
1.147 0.402 0.546 0.353
608.31 381.07 Ž
. Ž
. Ž
. Unemployment rate
y0.114 0.059 y0.051 0.051
y240.23 54.60 Ž
. Ž
. Ž
. MSA
0.399 0.342 0.436 0.304
518.63 336.21 N
5622 5622
5515
2
Ž .
R overall
0.17 0.24
0.29 Entries are GLS random-effects regression coefficients with standard errors in parentheses. Regressions
also include indicators for the year of the observation and a constant term. Statistically significant at the 10 level.
Statistically significant at the 5 level.
discernible effect on the other coefficients should be somewhat reassuring to those researchers who have not performed this correction in previous work. The
statistical significance indicates that initial conditions clearly matter in the self-em- ployment transition process, however, so all remaining probits will include a
selection term.
Before moving to the analysis of tax effects, note that the effects of the other variables in column 3 are generally consistent with findings in earlier studies.
First, age affects the transition probability in a u-shaped manner in the corrected specification. Those at the younger and older extremes of the age distribution are
more likely to enter. Tenure on the wage-and-salary job also affects this probabil- ity in a u-shaped manner. Essentially, those who enter self-employment are likely
to have spent either a very little or a very long time in their pre-transition job.
Minorities and union members are significantly less likely to enter self-employ- Ž
. ment. The minority effect has been previously documented by Meyer 1990
Table 9 Self-employment transition probit without random effects
Variable No endogeneity control
Ž .
Age 0.008 0.027
Ž .
Age squaredr100 y0.003 0.035
Ž .
Tenure y0.003 0.001
Ž .
Tenure squaredr100 0.0006 0.0002
Ž .
Dropout 0.010 0.064
Ž .
Some college y0.041 0.052
Ž .
College graduate 0.014 0.058
Ž .
Post-college education 0.035 0.058
Ž .
Non-white y0.144 0.076
Ž .
North Central y0.000 0.053
Ž .
South y0.046 0.054
Ž .
West 0.068 0.059
Ž .
Married y0.029 0.062
Ž .
Income from capital 0.004 0.002
Ž .
Part-time 0.104 0.052
Ž .
Kids 1 to 2 0.038 0.041
Ž .
Kids 3 to 5 0.076 0.039
Ž .
Kids 6 to 13 0.027 0.024
Ž .
Kids 14 to 17 y0.001 0.038
Ž .
Union y0.319 0.053
Ž .
Unemployment rate y0.021 0.007
Ž .
MSA y0.048 0.039
N 16,026
Entries are probit coefficients with standard errors in parentheses. This regression also includes indicators for the year of the observation and a constant term. For comparison purposes, the sample for
Ž .
this probit is identical to that in column 1 of Table 6 which includes random effects . Statistically significant at the 10 level.
Statistically significant at the 5 level.
Ž .
among others, and is explained by Blanchflower et al. 1998 as the result of racial discrimination in lending markets. The union membership indicator is likely
capturing an important job-lock effect. The effect of children in the household depends on their age distribution. Having more children in the household between
the ages of 3 and 5 increases the transition probability, while having more children who are between the ages of 14 and 17 reduces the probability. This is presumably
a result of the fact that the younger children can be placed in daycare or nursery school facilities, and the older children might be preparing to enter college. A
parent’s attitude toward the inherent risk of becoming self-employed might carry different weights at these times.
While not the primary focus of this study, it is important to address the effect of unemployment on self-employment. Indeed, this relationship has received a great
deal of attention in recent research and no significant degree of consensus has Ž
. emerged. Contrary to Schuetze 1998 and others, I find that higher unemployment
rates have a negative impact on self-employment transitions. This divergence is likely due to different degrees of aggregation in the respective unemployment rate
variables. Schuetze used state-level unemployment rates, while I use county-level unemployment rates. Higher unemployment at the local level might reduce one’s
probability of becoming self-employed by reducing the likelihood that he would be able to regain wage-and-salary employment should his business fail. State
unemployment rates, however, more closely reflect macroeconomic effects such as the degree of downsizing in the economy. In this way, higher unemployment on an
aggregate level indicates that workers might be turned toward self-employment as a result of actually losing their wage-and-salary jobs.
Turning now to Table 10, the effect of controlling for differential tax treatment on the self-employment entry decision is observed. This table presents results from
Table 10 Pooled transition probit results: using marginal tax rate differentials
Variable No endogeneity control
IV for MTR differential Ž
. Ž
. Age
y0.079 0.052 y0.275 0.063
Ž .
Ž .
Age squaredr100 0.109 0.077
0.353 0.093 Ž
. Ž
. Tenure
y0.006 0.001 y0.004 0.002
Ž .
Ž .
Tenure squaredr100 0.100 0.047
y0.011 0.066 Ž
. Ž
. Dropout
y0.014 0.142 0.508 0.157
Ž .
Ž .
Some college y0.011 0.097
y0.051 0.113 Ž
. Ž
. College graduate
0.046 0.106 y0.363 0.116
Ž .
Ž .
Post-college education 0.126 0.106
0.007 0.121 Ž
. Ž
. Non-white
y0.279 0.142 y0.449 0.149
Ž .
Ž .
North Central y0.037 0.103
0.061 0.120 Ž
. Ž
. South
y0.108 0.097 y0.081 0.124
Ž .
Ž .
West 0.048 0.097
0.195 0.123 Ž
. Ž
. Married
0.090 0.095 0.160 0.115
Ž .
Ž .
Income from capital y0.013 0.012
0.008 0.014 Ž
. Ž
. Part-time
0.158 0.085 y0.007 0.107
Ž .
Ž .
Kids 1 to 2 y0.044 0.063
y0.005 0.070 Ž
. Ž
. Kids 3 to 5
0.250 0.061 0.286 0.066
Ž .
Ž .
Kids 6 to 13 0.055 0.055
0.060 0.058 Ž
. Ž
. Kids 14 to 17
y0.240 0.111 y0.263 0.145
Ž .
Ž .
Union y0.407 0.115
y0.309 0.126 Ž
. Ž
. Unemployment rate
y0.045 0.012 y0.055 0.014
Ž .
Ž .
MSA 0.003 0.079
0.060 0.085 Ž
. Ž
. Mills ratio
0.954 0.502 1.173 0.538
Ž .
Ž .
MTR differential 0.017 0.004
y0.123 0.010 N
5622 5622
Sample transition probability 0.033
0.033 Entries are random-effects probit coefficients with bootstrapped robust standard errors in parentheses.
Regressions also include indicators for the year of the observation and a constant term. The first-stage instrumenting equation includes an identical set of variables in addition to the instrument.
Statistically significant at the 10 level. Statistically significant at the 5 level.
Ž random effects transition probits that include the wage-and-salary minus self-em-
. Ž
. ployment difference in marginal tax rates MTR . The first column contains
coefficients and bootstrapped standard errors without instrumenting for the tax rate differential.
12
Column 1 shows that the MTR differential has a very small positive and significant effect on the probability of entry. This indicates that those with a
larger last-dollar tax benefit from entering self-employment are also more likely to be the ones who choose to enter.
The question of endogeneity remains, however, so column 2 presents results from a two-stage instrumental variables estimation process as described above.
While patterns of significance for the non-tax variables remain largely unchanged, the effect of the instrumented tax rate differential is now negative, much larger,
and even more statistically significant. To determine the extent to which endogene- ity is an actual problem in this situation, I performed the test suggested by Rivers
Ž .
and Vuong 1988 . This test involves inserting the potentially endogenous variable along with the estimated residual vector from the proposed first-stage instrument-
ing equation into the transition probit. A significant coefficient on the residual indicates that endogeneity is a serious problem. Indeed, the Rivers–Vuong test for
this random effects probit rejects the null hypothesis of exogeneity — the coefficient on the residual term is statistically significant.
A simulation is helpful in understanding the quantitative significance of the instrumented MTR differential in column 2. Increasing this differential by 5
percentage points causes a reduction in the average self-employment transition probability of about 2.4 percentage points. This yields an elasticity of approxi-
mately y0.60.
Table 11 is similar to Table 10, except that it uses a difference in average tax Ž
. Ž
. rates ATR . As with the MTR differential without instrumenting , the effect of
the ATR differential is small but positive and statistically significant. Column 2 presents results from a similar two-stage instrumental variables estimation process
in order to investigate the potential endogeneity of the ATR differential. Again, patterns of significance for the non-tax variables remain largely unchanged.
However, the effect of the instrumented tax rate differential is now negative, as with the MTR differential, but not statistically significant. In this case, though, the
Rivers–Vuong test fails to reject the null hypothesis of exogeneity.
Using the results in column 1, then, increasing the ATR differential by 5 percentage points would increase the average self-employment transition probabil-
ity by only about 0.4 percentage points. This effect is especially small in
12
Bootstrapped standard errors are generated by repeating the estimation procedure 50 times for each random effects probit. When an instrumental variables process is used, the entire two-stage process is
run 50 times. Experimentation with 150 repetitions required much more computing time but revealed no changes in significance patterns.
Table 11 Pooled transition probit results: using average tax rate differentials
Variable No Endogeneity Control
IV for ATR Differential Ž
. Ž
. Age
y0.082 0.052 y0.099 0.079
Ž .
Ž .
Age squaredr100 0.112 0.077
0.132 0.117 Ž
. Ž
. Tenure
y0.006 0.001 y0.005 0.002
Ž .
Ž .
Tenure squaredr100 0.094 0.045
0.080 0.081 Ž
. Ž
. Dropout
y0.012 0.132 0.061 0.185
Ž .
Ž .
Some college 0.005 0.095
y0.025 0.106 Ž
. Ž
. College graduate
0.056 0.111 y0.017 0.178
Ž .
Ž .
Post-college education 0.166 0.111
0.076 0.239 Ž
. Ž
. Non-white
y0.283 0.142 y0.293 0.182
Ž .
Ž .
North Central y0.030 0.105
y0.031 0.100 Ž
. Ž
. South
y0.105 0.099 y0.111 0.086
Ž .
Ž .
West 0.059 0.097
0.070 0.106 Ž
. Ž
. Married
0.101 0.095 0.103 0.084
Ž .
Ž .
Income from capital y0.012 0.012
y0.011 0.017 Ž
. Ž
. Part-time
0.159 0.085 0.131 0.083
Ž .
Ž .
Kids 1 to 2 y0.051 0.064
y0.046 0.059 Ž
. Ž
. Kids 3 to 5
0.243 0.060 0.246 0.067
Ž .
Ž .
Kids 6 to 13 0.051 0.055
0.050 0.052 Ž
. Ž
. Kids 14 to 17
y0.215 0.115 y0.206 0.174
Ž .
Ž .
Union y0.400 0.109
y0.376 0.117 Ž
. Ž
. Unemployment rate
y0.046 0.012 y0.048 0.016
Ž .
Ž .
MSA 0.013 0.077
0.018 0.076 Ž
. Ž
. Mills ratio
0.994 0.504 0.956 0.417
Ž .
Ž .
ATR differential 0.010 0.003
y0.006 0.031 N
5622 5622
Sample transition probability 0.033
0.033 Entries are random-effects probit coefficients with bootstrapped robust standard errors in parentheses.
Regressions also include indicators for the year of the observation and a constant term. The first-stage instrumenting equation includes an identical set of variables in addition to the instrument.
Statistically significant at the 10 level. Statistically significant at the 5 level.
comparison to the sample transition probability of 3.3, translating into an elasticity of about 0.06.
Table 12 examines whether ATRs and MTRs truly have these opposing effects by including both variables in a single random effects probit. Column 1 contains
results without instrumental variables, while column 2 presents the IV results. The results in Tables 10 and 11 are seen again in this specification and, in fact, are
essentially unchanged. Those contemplating a transition into self-employment are apparently more responsive to changes at the margin than they are to changes in
average tax rates. The Rivers–Vuong test rejects the null hypothesis of joint exogeneity in this case.
Taking the contents of Tables 10–12 together, the overall tax effects become clear. First, the positive and significant effect of the ATR differential indicates that
Table 12 Pooled transition probit results: using average and marginal tax rate differentials
Variable No endogeneity control
IV for ATR and MTR differentials Ž
. Ž
. Age
y0.075 0.052 y0.268 0.075
Ž .
Ž .
Age squaredr100 0.104 0.078
0.344 0.108 Ž
. Ž
. Tenure
y0.006 0.001 y0.004 0.002
Ž .
Ž .
Tenure squaredr100 0.104 0.046
y0.007 0.070 Ž
. Ž
. Dropout
y0.038 0.146 0.481 0.200
Ž .
Ž .
Some college y0.001 0.096
y0.040 0.125 Ž
. Ž
. College graduate
0.070 0.111 y0.337 0.159
Ž .
Ž .
Post-college education 0.157 0.114
0.042 0.206 Ž
. Ž
. Non-white
y0.282 0.143 y0.439 0.152
Ž .
Ž .
North Central y0.036 0.104
0.062 0.120 Ž
. Ž
. South
y0.107 0.098 y0.076 0.124
Ž .
Ž .
West 0.045 0.098
0.192 0.128 Ž
. Ž
. Married
0.093 0.095 0.157 0.118
Ž .
Ž .
Income from capital y0.013 0.012
0.008 0.014 Ž
. Ž
. Part-time
0.167 0.085 0.002 0.115
Ž .
Ž .
Kids 1 to 2 y0.047 0.063
y0.008 0.071 Ž
. Ž
. Kids 3 to 5
0.249 0.061 0.284 0.065
Ž .
Ž .
Kids 6 to 13 0.055 0.056
0.060 0.059 Ž
. Ž
. Kids 14 to 17
y0.242 0.112 y0.263 0.148
Ž .
Ž .
Union y0.415 0.115
y0.315 0.131 Ž
. Ž
. Unemployment rate
y0.045 0.012 y0.054 0.014
Ž .
Ž .
MSA 0.003 0.079
0.059 0.080 Ž
. Ž
. Mills ratio
0.975 0.502 1.175 0.540
Ž .
Ž .
ATR differential 0.006 0.003
0.006 0.026 Ž
. Ž
. MTR differential
0.015 0.005 y0.123 0.010
N 5622
5622 Sample transition probability
0.033 0.033
Entries are random-effects probit coefficients with bootstrapped robust standard errors in parentheses. Regressions also include indicators for the year of the observation and a constant term. The first-stage
instrumenting equations include an identical set of variables in addition to the instruments. Statistically significant at the 10 level.
Statistically significant at the 5 level.
those with the largest tax advantage per dollar of income are slightly more likely to enter self-employment, although the quantitative effect is small. However, the
statistically and quantitatively more significant negative impact of the instru- mented MTR differential reveals an opposite effect.
One explanation for this is that the MTR differential could be indicating a large drop in taxable income as a result of entering self-employment. Those with higher
values of this difference would experience the largest drop in earnings, and should be the least likely to make a transition. Further, those with the lowest, or most
negative, MTR differentials would likely see a dramatic increase in earnings — and marginal tax rates — as a result of becoming self-employed. They might be,
consequently, the most likely to enter.
The extent to which this is an income effect rather than a price effect can be examined by including the difference in after-tax household income as a regressor
in addition to the MTR differential. Table 13 presents results from two probits, Ž
. Ž
. without column 1 and with column 2 similar instrumenting equations for the
MTR and net income differentials. In column 1, the MTR effect is apparently unchanged but the coefficient on the net income differential is strangely positive
Ž .
and significant , indicating that larger expected income reductions from entering self-employment actually increase the probability of entry.
However, the results in column 2 show that instrumenting for both of the potentially endogenous differential variables renders the income effect statistically
insignificant. In fact, the only variables with any significance are the Mills ratio for the initial conditions correction and the MTR differential. The Rivers–Vuong
Table 13 Pooled transition probit results: adding after-tax household income differentials
Variable No endogeneity control
IV for after-tax difference Ž
. Ž
. Age
y0.091 0.056 y0.357 0.654
Ž .
Ž .
Age squaredr100 0.130 0.082
0.352 0.618 Ž
. Ž
. Tenure
y0.006 0.002 0.001 0.020
Ž .
Ž .
Tenure squaredr100 0.102 0.064
y0.108 0.480 Ž
. Ž
. Dropout
y0.046 0.150 1.199 2.567
Ž .
Ž .
Some college 0.026 0.085
y0.477 1.453 Ž
. Ž
. College graduate
0.170 0.104 y2.576 6.933
Ž .
Ž .
Post-college education 0.297 0.115
y4.216 13.786 Ž
. Ž
. Non-white
y0.222 0.163 y1.342 2.744
Ž .
Ž .
North Central y0.070 0.096
0.174 0.936 Ž
. Ž
. South
y0.107 0.075 y0.601 1.376
Ž .
Ž .
West 0.047 0.076
0.179 0.519 Ž
. Ž
. Married
0.084 0.099 0.428 1.157
Ž .
Ž .
Income from capital y0.018 0.015
0.049 0.114 Ž
. Ž
. Part-time
0.213 0.084 y0.327 0.966
Ž .
Ž .
Kids 1 to 2 y0.038 0.054
0.196 0.513 Ž
. Ž
. Kids 3 to 5
0.224 0.062 0.386 0.477
Ž .
Ž .
Kids 6 to 13 0.056 0.062
0.042 0.338 Ž
. Ž
. Kids 14 to 17
y0.208 0.151 y0.304 0.430
Ž .
Ž .
Union y0.449 0.126
y0.166 0.472 Ž
. Ž
. Unemployment rate
y0.039 0.013 y0.121 0.269
Ž .
Ž .
MSA y7.69ey05 0.063
0.229 0.305 Ž
. Ž
. Mills ratio
1.015 0.407 1.068 0.368
Ž .
Ž .
MTR differential 0.014 0.004
y0.124 0.014 Ž
. Ž
. Income differential
1.95ey05 3.48ey06 y2.86ey04 9.75ey04
N 5515
5515 Sample transition probability
0.032 0.032
Entries are random-effects probit coefficients with bootstrapped robust standard errors in parentheses. Regressions also include indicators for the year of the observation and a constant term. The first-stage
instrumenting equations include an identical set of variables in addition to the instruments. Statistically significant at the 5 level.
test rejects the null hypothesis of joint exogeneity, and the effect of the instru- mented MTR differential itself is unchanged.
6. Conclusions