223 S.P. Vahey Economics of Education Review 19 2000 219–227
Table 2 Continued
Name Definition
Males Females
Education OVER vector
OGRADE Overed. req. GRADE 5 1; otherwise 5 0
0.07 0.25
0.08 0.27
OSOME Overed. req. SOME 5 1; otherwise 5 0
0.07 0.26
0.05 0.22
OHIGH Overed. req. HIGH 5 1; otherwise 5 0
0.11 0.31
0.16 0.37
OCOLL Overed. req. COLL 5 1; otherwise 5 0
0.02 0.13
0.02 0.14
OBACH Overed. req. BACH 5 1; otherwise 5 0
0.03 0.18
0.01 0.12
UNDER vector USOME
Undered. req. SOME 5 1; otherwise 5 0 0.03
0.16 0.01
0.12 UHIGH
Undered. req. HIGH 5 1; otherwise 5 0 0.10
0.30 0.06
0.24 UCOLL
Undered. req. COLL 5 1; otherwise 5 0 0.05
0.21 0.04
0.18 UBACH
Undered. req. BACH 5 1; otherwise 5 0 0.05
0.21 0.05
0.22 UPOST
Undered. req. POST 5 1; otherwise 5 0 0.02
0.15 0.01
0.08
dard deviations are given in Table 2. The dependent vari- able is generated as follows. Respondents were asked to
estimate their personal income in 1981 and the number of hours worked per week. Removing those who did not
work year round, I calculate the earnings per hour usu- ally worked.
If Berg’s proposition is correct, the coefficients on the overeducation dummies should be negative.
4. Results
The results from the OLS regressions are presented in Table 3. The first column includes only the personal
characteristics of the respondents and the control vari- ables. The next column includes these and the education
variables from Eq. 1. The third column includes the same variables but the sample is restricted to males; and
the next column includes only females. The final column refers to single, urban-based females.
From the first column, it is apparent that the personal characteristic variables are generally significant at the
10 level t-statistics in parentheses and have the expected signs.
6
The occupational and the industry dummies are all significant. Employment in the retail and
other services RET has a particularly strong negative impact on earnings; employment in the extraction and
construction industries EXTR has an impact of similar magnitude, though the sign is reversed. The location
6
The Breusch–Pagan–Godfrey test Breusch and Pagan, 1979; Godfrey, 1978 indicates that the null hypothesis of
homoskedasticty can be rejected at the 5 level. The reported standard errors are heteroskedasticity-consistent, calculated
using White’s 1980 method.
dummies are insignificant, apart from those for living in British Columbia BC or living in a community with a
population 100 000 CITY, which both have posi- tive coefficients.
The inclusion of the education variables reveals that the returns to over and undereducation vary with the
required level of education.
7
Many of the educational mismatch dummies are insignificant at the 10 level.
However, the overeducation dummy at the BACH level of required education and the undereducation dummy at
the HIGH level are significant. At the 15 significance level, the undereducation dummy at the SOME level is
also significant. In these cases, overeducation is associa- ted with higher earnings and undereducation with
lower earnings.
8
Restricting the sample to males and females in turn reveals some startling differences between the sexes.
9
First, the coefficients for the required schooling variables at the BACH and POST levels are much larger for
females. Second, although the impacts of the educational mismatch variables for males are similar to those for the
7
As with the previous equation, the Breusch–Pagan–Godfrey test indicates that the errors are heteroskedastic and the reported
standard errors are heteroskedasticity-consistent.
8
The over and undereducation dummies are jointly signifi- cant at the 5 significance level. The hypothesis that two
dummy variables
one for
overeducation, one
for undereducation capture the effects of educational mismatch,
regardless of the level of required education, cannot be rejected at the 10 level.
9
An F-test of the hypothesis that there is no difference between the male and female coefficients indicates that the null
can be rejected at the 5 significance level.
224 S.P. Vahey Economics of Education Review 19 2000 219–227
Table 3 Regression equations; dependent variable ln Y
Variable All
All Males
Females Females
EXP 0.016
0.016 0.019
0.014 0.018
3.404 3.401
3.140 1.704
1.124 EXP2
—0.033 2
0.024 2
0.028 2
0.023 2
0.040 2 3.423
2 2.644 2 2.491
2 1.257 2 1.080
UNION 0.041
0.058 0.055
0.059 0.125
1.177 1.690
1.354 0.950
1.018 SEX
0.312 0.293
9.071 8.855
TEN 0.021
0.021 0.017
0.022 0.006
3.659 3.731
2.825 2.147
0.334 TEN2
2 0.051
2 0.059
2 0.051
2 0.049
2 0.015
2 2.415 2 2.895
2 2.345 2 1.254
2 0.256 BIL
0.083 0.050
0.127 2
0.071 2
0.063 1.892
1.192 2.666
2 0.918 2 0.431
EXTR 0.198
0.184 0.155
0.262 0.171
3.764 3.689
2.817 1.503
0.364 DIST
0.091 0.042
2 0.004
0.147 0.120
1.809 0.869
2 0.084 1.236
0.556 PUB
0.079 2
0.011 2
0.009 2
0.021 2
0.021 1.759
2 0.251 2 0.198
2 0.216 2 0.128
INFO 0.108
0.018 0.076
2 0.025
0.194 1.836
0.322 0.992
2 0.227 1.039
RET 2
0.197 2
0.218 2
0.155 2
0.267 0.167
2 3.782 2 4.287
2 2.442 2 2.732
0.873 PROF
0.490 0.194
0.215 0.148
0.049 9.412
2.507 2.447
1.337 0.228
SEMI 0.452
0.263 0.229
0.282 0.180
9.265 4.576
3.325 3.020
0.950 SUPER
0.229 0.123
0.158 2
0.002 0.361
4.087 2.232
2.629 2 0.014
1.093 SKILL
0.195 0.114
0.130 0.070
0.088 4.870
2.815 2.901
0.996 0.645
ATL 2
0.075 2
0.049 2
0.083 0.007
2 0.049
2 1.179 2 0.791
2 1.271 0.075
2 0.150 QUE
2 0.027
0.002 2
0.076 0.110
0.183 2 0.647
0.053 2 1.776
1.568 1.429
PRA 0.009
0.009 2
0.025 0.045
0.179 0.187
0.195 2 0.413
0.624 1.220
BC 0.145
0.144 0.210
0.026 2
0.190 2.862
2.876 3.175
0.283 2 1.302
CITY 0.068
0.047 0.012
0.114 2.091
1.526 0.363
2.075 RGRADE
2 0.358
2 0.406
2 0.305
2 0.391
2 4.737 2 4.868
2 2.037 2 1.082
RSOME 2
0.136 2
0.148 2
0.106 2
0.254 2 1.702
2 1.587 2 0.844
2 0.968 RCOLL
0.021 2
0.030 0.058
0.142 0.384
2 0.431 0.574
0.733 RBACH
0.193 0.139
0.262 0.494
2.181 1.188
1.989 2.261
RPOST 0.359
0.226 0.801
1.275 3.665
1.939 3.424
3.511 OGRADE
2 0.037
0.051 2
0.152 0.212
2 0.436 0.492
2 1.007 0.621
Continued.
225 S.P. Vahey Economics of Education Review 19 2000 219–227
Table 3 Continued
Variable All
All Males
Females Females
OSOME 2
0.001 2
0.059 0.117
0.899 2 0.008
2 0.618 0.777
2.757 OHIGH
0.017 2
0.012 0.042
2 0.056
0.285 2 0.168
0.448 2 0.359
OCOLL 2
0.028 2
0.203 0.132
0.333 2 0.271
2 1.347 0.730
1.133 OBACH
0.145 0.129
0.099 0.009
2.139 1.634
0.446 0.032
USOME 2
0.181 2
0.259 2
0.048 0.317
2 1.572 2 2.369
2 0.204 0.628
UHIGH 2
0.202 2
0.243 2
0.122 0.593
2 3.170 2 3.365
2 0.950 1.978
UCOLL 2
0.047 2
0.106 0.093
2 0.699 2 1.302
0.607 UBACH
0.025 0.100
2 0.057
2 0.099
0.277 0.888
2 0.400 2 0.423
UPOST 0.040
0.122 2
0.263 0.355
0.965 2 0.721
Constant 1.534
1.700 2.046
1.642 1.608
25.080 25.570
23.670 11.550
7.135 N
993 993
569 424
98 R
¯
2
0.363 0.418
0.385 0.304
0.264 Note: t-statistics in parentheses.
full sample,
10
for females all these terms are insignifi- cant.
11
One interpretation of the difference in the returns to educational mismatch by gender is that Frank’s 1978
hypothesis holds.
12
Female job search is geographically constrained and, as a result, competition within the lab-
our market is insufficient to generate the usual returns to educational mismatch. In order to test this theory, I
10
For males, the over and undereducation dummies are jointly significant at the 5 level; but the hypothesis that two
dummy variables capture educational mismatch, regardless of required education, cannot be rejected at the 10 level. For
females, the educational mismatch dummies are jointly insig- nificant at the 10 level.
11
Recall from Table 1 that for high educational requirements, the numbers of mismatched females are smaller than those of
mismatched males. This contributes to the imprecision of the estimates.
12
It is often argued that unions reduce the variance in their members’ earnings. Hence, differences in union coverage could
account for the differences in the returns to educational mis- match by gender. However, the hypothesis that there is no dif-
ference between union and non-union workers cannot be rejected for either males or females at the 10 level. In both
cases, the results for both union and non-union workers are similar to those obtained using all workers.
restrict the sample to females that are unmarried and urban-based. The returns to skill mismatch, shown in the
final column, are larger than for all female workers. However, only two mismatch variables are significantly
different from zero at the 10 level, OSOME and UHIGH; and the coefficient on the latter implies that
undereducation raises wages relative to otherwise ident- ical workers. These results are difficult to reconcile with
Frank’s theory.
Using Oaxaca’s 1973 decomposition, Table 4 shows the contributions of the explanatory variables to the
male–female earnings gap. I decompose the gap into the difference in the sample means multiplied by the esti-
mated male coefficient first column and the difference
Table 4 Male–female earnings decomposition
Characteristics Returns
PC 0.092
0.397 REQ
0.011 2
0.069 OVER
0.002 2
0.008 UNDER
2 0.015
2 0.006
Total 0.090
0.314
226 S.P. Vahey Economics of Education Review 19 2000 219–227
in the coefficients multiplied by the female mean second column. The first component shows the difference due
to male–female characteristics; the second, the return to these characteristics. The latter is sometimes attributed
to discrimination. The largest proportion of the gap is explained by the returns to personal characteristics. The
overeducation and undereducation variables have rela- tively small impacts; and their combined effect is nega-
tive. The majority of the earnings gap accounted for by the educational mismatch variables is due to the charac-
teristics themselves, rather than the returns to the charac- teristics.
5. Conclusions
