161 E. Cohn, Y.C. Ng Economics of Education Review 19 2000 159–168
To test whether there are significant differences between the effects of overschooling and underschooling
on earnings between 1986 and 1991, the two samples of observations are pooled. All independent variables are
interacted with the year dummy variable, T91 which has a value of 1 for observations of the 1991 sample, and 0
otherwise, and both the original variables and the inter- action
terms are
included in
the new
earnings regressions. The dependent variable lnINC for both
years is expressed in constant 1990 Hong Kong dollars. Also, a Chow test is performed to test whether the struc-
tural equations for earnings differ between the two years.
3. Data
As indicated earlier, data from the 1986 Hong Kong By-census and the 1991 Hong Kong Census are
employed in this study. Only employees in the age range 15–60 years are included in the analyses. Workers in the
agriculture, fishing, mining and quarrying industries, and those without reported monthly earnings from the main
employment are excluded from the sample. Since infor- mation on the number of years of schooling an individual
attained is not available, years of schooling completed SCHOOL are derived from the highest level of edu-
cational attainment reported, such as Form 5 equivalent to Grade 11 in the US, or DiplomaCertificate courses
in Technical InstitutesPolytechnics. Since actual experi- ence is not available in these two data sets, potential
experience EXP 5 Age 2 SCHOOL-6 and its square EXPSQ are used in the regressions instead.
Earlier research by the Hong Kong Census showed only minor deviations in hours worked. Therefore, the
Hong Kong Census did not collect data on hours of work. Since labor earnings are provided on a monthly
basis, it was not possible for us to convert these earnings into an hourly basis. Therefore, the dependent variable
of the wage equations used here is the natural logarithm of monthly earnings from a worker’s principal employ-
ment lnINC. Other independent variables include industry
dummy variables
CONSTR, WHOLES,
TRANSP, FINANCE, and SERVICE—the reference group is manufacturing, and the marital status of an
individual MAR. The definitions of the variables are presented in Appendix A Table 6 while Appendix B
Table 7 provides basic sample statistics.
4. Results
4.1. Incidence of overschooling and underschooling Employing the objective method following the Kiker
et al. variant, described earlier, the distributions of indi- viduals by adequate schooling ASCH, overschooling
OSCH and underschooling USCH, for the years 1986 and 1991, are shown in Table 1. The data suggest that the
incidence of over- and underschooling was fairly stable between 1986 and 1991. In addition, females were more
likely to be adequately schooled than males and were less likely to be both over- and under-schooled. Finally,
for both sexes and years, the incidence of overschooling is greater than the incidence of underschooling.
4.2. Basic statistics: years of actual schooling, adequate schooling, over- and underschooling, age,
and experience
Table 2 provides means and standard deviations for these variables, cross classified by group ASCH, OSCH
and USCH. Note that the number of years of overschoo- ling is lower for overschooled males than for females,
but that the incidence of adequate, over- and underschoo- ling has been fairly stable over time. Note also that the
mean years of actual schooling for this group is fairly high around 13 years. Concerning underschooled work-
ers, the data indicate that the number of years of under- schooling is greater for females than for males. In con-
trast to overschooled workers, the mean of years of actual schooling is much lower for underschooled work-
ers around 8 years.
We also find that the mean of ADSCH for adequately- schooled persons has increased by one year for females
and by two-thirds of a year for males. This suggests at least a small secular rise in educational requirements for
jobs in Hong Kong, especially for women. In addition, male workers tend to be somewhat older than females,
especially among the underschooled and adequately schooled groups. Also, the 1991 group is slightly older
than the 1986 group. Underschooled workers are older than either overschooled or adequately schooled work-
ers.
Males have more labor-market experience than females, especially among adequately schooled workers,
although differences in work experience by sex appear to be smaller than one might have expected. Finally,
underschooled workers have more labor-market experi- ence than other workers.
Table 1 Percentage Distribution of Individuals by Adequate-, Over- and
Under-schooling 1986
1991 Male
a
Female Male
Female ASCH
35 44
35 44
OSCH 38
32 37
31 USCH
28 24
28 25
Note: See Appendix A for definitions of variables.
a
Percentages do not sum to 100 due to rounding.
162 E. Cohn, Y.C. Ng Economics of Education Review 19 2000 159–168
Table 2 Basic Statistics
Over-schooled Individuals Under-schooled Individuals
Adequately-schooled Individuals 1986
1991 1986
1991 1986
1991 Male
Female Male
Female Male
Female Male
Female Male
Female Male
Female OVERSCH
3.37 3.75
3.29 3.68
– –
– –
– –
– –
1.75 2.36
1.74 2.21
UNDERSCH –
– –
– 3.92
4.16 3.77
4.02 –
– –
– 2.44
2.99 2.42
2.71 ADSCH
9.14 8.98
9.74 10.09
12.03 11.79
11.93 11.65
11.22 11.15
11.85 12.15
2.30 4.22
2.47 2.63
3.54 3.01
2.98 3.05
3.51 3.71
3.41 2.69
SCHOOL 12.50
12.73 13.03
13.77 8.11
7.63 8.18
7.63 11.22
11.15 11.85
12.15 2.66
3.34 2.80
2.78 4.11
4.64 3.87
4.57 3.51
3.71 3.41
2.69 AGE
30.86 30.69
33.79 32.41
39.04 35.37
38.98 36.84
34.25 29.88
35.42 31.29
9.59 9.70
9.85 8.59
11.85 12.09
11.62 11.66
10.36 9.41
10.32 8.89
EXP 12.35
11.96 14.76
12.65 24.93
21.73 24.80
23.21 17.02
12.73 17.57
13.14 9.55
10.66 10.12
9.22 14.06
15.24 13.69
14.76 11.40
11.63 11.58
10.15 N
62795 32456
67103 37050
46126 24861
50560 30704
57728 45377
62226 53023
4.3. Distribution of over- and underschooling by experience
Table 3 provides information on the distribution of over- and underschooling by years of potential experi-
ence. One observation is that a fairly large proportion of overschooled persons have very little experience 0–
5 years. In general, the percentage of overschooled per- sons declines as experience increases. Also, the percent-
age of underschooled workers, both men and women, is greater than the percentage of overschooled workers for
the high-experience groups 21 years or more, and especially so for the highest-experience group 36 years
or more. Further, the percentage of underschooled male
Table 3 Percentage Distribution of Over- and Under-schooling by Years of Potential Experience
EXP in years 1986
1991 Male
Female Male
Female Over
Under Over
Under Over
Under Over
Under 0–5
28.22 7.43
34.68 14.84
20.43 7.88
25.34 12.39
6–10 23.30
10.66 22.91
15.24 19.59
8.81 23.64
11.82 11–15
16.93 12.54
12.65 13.21
18.40 11.38
17.66 11.46
16–20 11.94
13.40 9.11
11.32 14.04
13.12 12.13
11.36 21–25
8.31 10.55
7.47 8.94
10.91 14.06
9.88 11.89
26–30 5.32
8.51 5.24
6.97 7.74
11.36 6.54
10.38 31–35
3.44 8.47
3.61 6.36
5.01 8.40
3.11 7.61
36 1 2.54
28.44 4.33
23.12 3.88
24.99 1.70
23.09 N
62795 46126
32456 24861
67103 50560
37050 30704
Note: Over 5 individuals classified as over-schooled; Under 5 individuals classified as under-schooled.
workers who have relatively low experience 0–15 years is lower than the respective percentage of females.
Finally, although we observe some changes over time for example, in the proportion of both over- and under-
schooled
males and
females with
0–5 year
of experience, overall the 1986 and 1991 distributions are
very similar. 4.4. Effects of overschooling and underschooling on
earnings A number of regressions were run to determine the
effect of adequate schooling, overschooling and under-
163 E. Cohn, Y.C. Ng Economics of Education Review 19 2000 159–168
Table 4 Selected regression coefficients and t-ratios in parentheses for lnINC: Alternative Models, 1986 and 1991
a
Independent 1986
1991 variables
Model 1 Model 2
Model 1 Model 2
Male Female
Male Female
Male Female
Male Female
ADSCH 0.11 254.24
0.09 136.71 0.14 186.71
0.14 121.98 0.13 290.54
0.15 238.17 0.17 202.63
0.19 164.88 OVERSCH
0.04 66.57 0.05 56.04
0.11 89.62 0.10 67.55
0.05 75.76 0.04 47.14
0.11 93.28 0.08 63.02
UNDERSCH 20.04 62.70
20.06 272.42 20.13 293.01
20.10 254.42 20.04 258.54
20.05 268.17 20.12 280.55
20.14 282.00 ADSCHEXP
– –
20.001 238.03 20.003 254.45
– –
20.002 243.02 20.002 235.94
OVERSCHEXP –
– 20.004 264.62
20.002 238.36 –
– 20.004 263.31
20.003 242.27 UNDERSCHEXP
– –
0.003 69.03 0.002 31.39
– –
0.003 58.80 0.003 55.76
R
2
Adjusted 0.43
0.36 0.46
0.38 0.44
0.45 0.46
0.47 N
166649 102694
166649 102694
179889 120777
179889 120777
a
The equations also include controls for potential experience, dummy variables for being married and one digit industry codes.
schooling on earnings, other things equal. Results for two models are shown in Tables 4 and 5.
4.4.1. Returns to years of adequate, over- and underschooling
Results shown in Table 4 for Model 1 are similar in many respects to those obtained earlier for the US
Duncan and Hoffman, 1981; Sicherman, 1991; Cohn and Khan, 1995. We observe high rates of return to
adequate schooling between 9 and 15 percent, lower but positive rates of return to overschooling between 4
Table 5 Selected regression coefficients and t-ratios in parentheses for lnINC: Alternative Models, 1986 and 1991
a
Independent 1986
1991 variables
Model 1 Model 2
Model 1 Model 2
Male Female
Male Female
Male Female
Male Female
SCHOOL 0.09
0.08 0.14
0.13 0.10
0.11 0.17
0.17 219.06
127.30 191.96
117.48 239.32
171.22 211.28
146.73 OSCH
20.19 20.15
20.10 20.08
20.23 20.29
20.17 20.29
266.18 238.71
221.45 213.34
280.22 280.24
233.92 249.03
USCH 0.16
0.03 0.03
0.08 0.19
0.12 0.11
0.05 50.97
6.06 5.39
11.06 55.54
28.24 17.38
7.09 SCHOOLEXP
– –
20.002 20.003
– –
20.003 20.003
274.45 254.13
289.44 257.72
OSCHEXP –
– 20.007
20.007 –
– 20.004
20.003 224.35
219.20 215.89
27.73 USCHEXP
– –
0.005 20.003
– –
0.002 0.003
19.17 210.50
8.80 8.23
R
2
Adjusted 0.38
0.35 0.42
0.37 0.38
0.34 0.42
0.37 N
166649 102694
166649 102694
179889 120777
179889 120777
a
The equations also include controls for potential experience, dummy variables for being married and one digit industry codes.
and 5 percent, and negative rates of return to under- schooling between 2 4 and 2 6 percent. All of these
results are highly statistically significant, as are all of the coefficients in the table. Note that while the 1986 and
1991 results are very similar, the rate of return to adequate schooling for females increased from 9 in
1986 to 15 in 1991.
Augmented by interaction variables for schooling and experience, Model 2 regressions in Table 4 generally do
not alter the conclusions discussed in the preceding para- graph. Among the results that were obtained for Model
164 E. Cohn, Y.C. Ng Economics of Education Review 19 2000 159–168
2 we observe negative coefficients for the interaction of experience with ADSCH and OVERSCH, indicating that
more experienced male and female workers tend to have a lower rate of return to adequate schooling or over-
schooling. These results might represent a vintage effect: more recent graduates received schooling that is more
useful in current production, hence more recent school- ing is rewarded more generously. The results for over-
schooling, moreover, are consistent with the notion that extra
schooling might be
a good substitute
for especially early labor-market experience. What the
present findings suggest, however, is that the wage bene- fits of overschooling decline as one gains more labor-
market experience.
In addition, we find positive coefficients for the inter- action of experience with UNDERSCH. The negative
coefficient for UNDERSCH is reduced as one gains more experience, suggesting that experience is a substi-
tute for schooling in so far as wages are concerned, albeit an imperfect substitute 43.3 [50] years of experience
are required to fully offset the negative coefficient on UNDERSCH for males [females] in 1986; for 1991, the
respective figures are 40 and 46.7.
4.4.2. Returns to being over- and underschooled Results in Table 5 show the returns to actual school-
ing, and for being overschooled and underschooled. As emphasized in Cohn 1992, results from this table
should not be used to compute rates of return to over- and underschooling. Rather, as Sicherman 1991 points
out, one can determine from these regressions whether overschooled or underschooled workers have wages that
are lower or higher than the wages they would have earned in a job for which they are adequately schooled.
Results using Model 1 are generally consistent with those of other studies Verdugo and Verdugo, 1989;
Sicherman, 1991; Cohn and Khan, 1995: overschooled underschooled workers have wages that are lower
higher than the wages they would have earned in a job for which they are adequately schooled. The difference
between earlier results for the US and the present find- ings for Hong Kong is the magnitude of these wage dif-
ferences: in Hong Kong these wage differences are quite substantial.
Concerning Model 2 results, negative interactions are found for all of the coefficients of SCHOOLEXP and
OSCHEXP as well as for the coefficient of USCHEXP for females in 1986. The interpretation of the results for
the first two interaction terms follows the explanation in the preceding section. The negative coefficients of
USCHEXP for females in 1986 is puzzling, especially since the opposite effect has been found for 1991.
Interestingly, calculations based on the 1 Census sam- ple [N 5 10,661] produced a positive and significant
coefficient for USCHEXP for both 1986 and 1991. 4.4.3. Pooled data for 1986 and 1991
To test whether regression results for 1986 differ sig- nificantly from those of 1991, we pooled the two cross
sections and ran the regressions shown in Model 1 of Tables 4 and 5 with the addition of interaction time
dummies each of the relevant variables times T91, where T91 is 1 for 1991 and 0 for 1986. To save space,
the full regressions are not reproduced here they are available from the authors upon request.
To test
the null
hypotheses that,
for each
modelspecification, by sex, the 1986 and 1991 models produce identical coefficients, we conducted a series of
Chow F tests. The results indicate that all of the null hypotheses are rejected at the 5 level, suggesting that
running separate equations for 1986 and 1991 or pooled equations with time interactions was justified.
Corresponding to Model 1 of Table 4, all of the rel- evant interaction coefficients are significant and positive,
except for OVERSCHT91 for females which is nega- tive 2 0.01. The results suggest a slight increase over
the period in rates of return to adequate schooling, over- schooling and underschooling for males 0.02 for
ADSCHT91, 0.006 for OVERSCHT91, and 0.003 for UNDERSCHT91. For females, the results suggest a
fairly large increase 0.07 for adequate schooling and a small increase for underschooling 0.01.
Corresponding to Model 1 of Table 5, all of the inter- action coefficients are statistically significant. Relatively
modest positive coefficients were found for males, regarding actual schooling 0.01 and being under-
schooled 0.02. The respective coefficients for females are also positive, but larger in absolute value 0.03 for
actual schooling and 0.10 for being underschooled. Finally, for being overschooled, both for males and
females,
we find
negative interaction
coefficients 2 0.04 for males and 2 0.14 for females, consistent
with the results shown in Table 5.
5. Discussion