294 H.K. Siphambe Economics of Education Review 19 2000 291–300
Table 2 Characteristics of workers by ownership of organisationfirm
a
Variablesector Age
Educ. head Experience
Hours School
Training Gross income
b
Private sector 31
7.5 8.4
196 7.8
0.33 P736
Public 34
9.4 10.5
169 9.5
0.5 P1115
a
Source: Supplementary Survey data.
b
Gross earnings is average Pula amount per month from the Supplementary Survey data. Table 3
Earnings by level of education, by sector and gender
a
Private employees Public employees
All Male
Female All
Male Female
No school 608
658 210
584 676
348 lowprim
535 609
303 667
972 438
highprim 526
622 402
705 809
455 lowsec
669 774
577 1024
1105 953
highsec 1265
1395 1088
1319 1541
1083 Tertiary
2416 2364
2485 2842
3491 1967
a
Source: Supplementary Survey data.
experience and training than workers in the private sec- tor. However, it is not certain whether the higher earn-
ings in the public sector reflect the higher productivity of workers with more human capital stock, or whether
those human capital stocks are being merely used as screens without necessarily making public workers more
productive than private sector workers. In other words, screening might be prevalent in Botswana’s public sec-
tor.
Table 3 shows average earnings by level of education and by type of employment. Average earnings increase
as the level of education increases. Except for those with no schooling, public sector employees on average earn
more than private sector employees across all education levels. Earnings ratios between those with no education
and those with post secondary education are slightly higher between public employees. Results from Kann et
al. 1988 show that for those with no education and those with primary education the average salary was
higher in government, whereas for those with secondary and tertiary education levels the private sector paid more.
Higher salaries in the public sector for employees with secondary and tertiary education compared with the priv-
ate sector reflect two things. First, the ‘decompression’ exercise of 1990 increased government salaries for
higher level workers quite significantly. Secondly, due to the relative abundance of graduates from secondary
and tertiary levels of education, the private sector no longer needs to pay higher wages than the public sector
for recruitment purposes.
6. Exploring earnings variation
In this section we report results of earnings functions in order to explain the differences in earnings in the sam-
ple by means of differences in the human capital charac- teristics of individuals. The dependent variable is the
natural log of net earnings, where net earnings are defined as gross earnings less tax. Gross earnings are
defined as cash earnings plus wages in kind. Earnings are the money plus in-kind payments reported for the
month preceding the survey period. The independent variables are schooling, training and experience. For the
HIES data, experience is approximated in the usual way as Age-7 for those whose education exceeds 7 years of
schooling and Age-14 for those whose education com- prises 7 years or less. Age-14 is a correction aimed at
avoiding overestimating potential experience for those with fewer years of education.
6
The supplementary sur- vey data include the actual number of years of experi-
ence reported by the respondent. Table 4 summarises the basic earnings function using
HIES data. All the coefficients are significant at the 1 level and have the right signs. The model explains 38
of the variations in earnings. The explanatory power of the model is quite robust and is quite comparable, if not
slightly better than, some of the results that have used
6
See Dougherty and Jimenez 1991 for further discussion on this correction.
295 H.K. Siphambe Economics of Education Review 19 2000 291–300
Table 4 Mincerian earnings function: overall HIES
a
Variable Coefficient
Dependent variable ln monthly earnings Costant
4.19 68.07 Education
0.16 41.05 Experience
0.067 14.04 Experience squared
20.00088 28.7 R
2
adjusted 0.38
Sample size N 2891
a
Values in parentheses are t-statistics. , Significant at the 1 level.
this basic earnings function on developing countries for instance, Kugler Psacharopoulos, 1989; Psacharo-
poulos Steire, 1988; Al-Qudsi, 1989. The education coefficient, which is also the average rate of return to
education is 16. Experience adds positively to earnings until 38 years on the job, beyond which it contributes
negatively to earnings.
7
The results presented in Table 4 are, however, poten- tially subject to a type of selection bias. The results are
based on an equation estimated from data from only those who were working, resulting in a censored sample
of the population. The problem is that the unobserved wage offers of those not working are probably lower than
those for persons in the sample. To correct for this we use Heckman’s technique.
8
This involves a two-step pro- cess. In the first step the probability that an individual
will be gainfully employed and out of school is determ- ined according to a probit regression equation in which
a series of personal characteristics serve as regressors. These are age, education and marital status. The probit
results are reported in Table A1 in Appendix A. From this probit equation a selection variable, the Inverse Mills
Ratio, is created and inserted into the right-hand side of the earnings function. That equation is then re-estimated
for those employed to yield estimates free of censoring bias. To correct the estimates for heteroskedasticity we
use White 1980, 1993 Heteroskedasticity–Consistent covariance matrix estimation.
The results for the corrected estimates are reported in Table 5. The results of this correction are that the aver-
age rate of return to education which is the expected average rate of return is lower by 4. The explanatory
power of the model improves significantly. It is also worth noting that the inverse Mill term is significant even
at the 0.1 level. The selectivity term is negative,
7
The point where experience stops adding positively to earnings is defined by
∂ ln Y
∂ T
= 0, from the earnings function:
ln Y =
a+bS+cT+dT
2
. This is equal to c22d; d,0.
8
See Heckman 1979 for a discussion of this problem. Table 5
Mincerian earnings function: overall — corrected for censoring bias HIES
a
Variable Coefficient
Dependent variable ln monthly earnings Constant
5.7 47.1 Education
0.12 26.5 Experience
0.009 1.5 Experience squared
20.00012 21.1 Inverse Mills Ratio
21.3 214.9 R
2
adjusted 0.41
Sample size N 2891
a
Values in parentheses are t-statistics. , Significant at 1 level.
implying that the observed wage is lower than the wage offers of a randomly selected individual in the popu-
lation. The second implication is that those who are in full time employment do not necessarily have a compara-
tive advantage.
In Table 6 we fit the Simple Mincerian earnings func- tion with continuous education between male and female
workers using HIES data. The results show that the model has a better explanatory power for females with
an R
2
of 51 when fitted on female workers. All coef- ficients are significant at the 1 level of significance and
have the right signs for both sexes. Females have a higher average rate of return of 18 compared with that
of males, which is 12. These paradoxical results are observed in most studies and are attributed to the lower
foregone earnings of females compared with their male counterparts see, for instance, Psacharopoulos Alam,
1991; Gomez-Castellanos Psacharopoulos, 1990; Psa- charopoulos, Velez Patrinos, 1994; Kugler Psachar-
opoulos, 1989. The high private rates of return to females also explain why females have a higher demand
for education than males as evident from their higher average years of education. Table 1 also confirms these
results, as the average earnings of females are less than those of their male counterparts. The Inverse Mills Ratio
is significant at the 0.1 level.
Table 7 explores the effect of adding hours of work and family background variables. Family background is
approximated by the education of the head of the house- hold. For all employees, all coefficients are significant at
the 1 level of significance and the explanatory power of the model improves slightly by one percentage point.
The coefficient of the education of the head of the house- hold is only significant at 1 level of significance for
males; it is not significant for females.
The meaning of a positive significant coefficient to a family background variable is that family background is
important in the determination of earnings. This operates in two ways: i through provision of a better learning
296 H.K. Siphambe Economics of Education Review 19 2000 291–300
Table 6 Mincerian earnings function by gender HIES
a
Variable Male
Female Dependent variable ln monthly earnings
Constant 5.67 42.1
4.8 26.3 Education
0.12 22.4 0.18 22.9
Experience 0.037 5.4
0.018 2.08 Experience squared
20.0006 24.8 20.0003 22
Inverse Mills Ratio 21.29 211.3
20.98 29 R
2
adjusted 0.47
0.51 Sample size N
1587 1304
a
Values in parentheses are t-statistics. Asterisks indicate the level of significance: , 1 level: , 5 level.
Table 7 Mincerian earnings function with log hours and family background variables SS data
a
Variable All
Male Female
Constant 6.4 16.4
6.8 11.2 5.6 12.3
Education 0.08 7.6
0.033 2.2 0.14 10.1
Experience 0.079 10.3
0.049 5.1 0.1 9.4
Experience square 20.0014 24.9
20.00005 21.5 20.0024 25.9
Log hours 20.289 23.9
20.27 22.4 20.23 22.8
Education of head 0.04 3.9
0.079 5.5 0.0079 0.6
R
2
adjusted 0.48
0.49 0.55
N 1026
532 494
a
Values in parentheses are t-statistics. Asterisks indicate level of significance: , 1 level; , 5 level.
environment; ii better contacts, leading to good jobs in the labour market. Given that unemployment is quite
high, especially for primary and lower secondary edu- cation graduates, it is very likely that family contacts are
becoming an important way of learning about and getting a good job. Those with more educated parents are more
likely to get better information about employment and therefore obtain better paying jobs.
As shown in Table 7, hours of work have a paradoxi- cal negative coefficient for both male and female work-
ers. This seems to suggest that those who supply more hours tend to earn less on average or that low paying
jobs are also those that require long hours, e.g. secur- ity guards.
Table 8 presents the results of fitting the Mincerian earnings function to three sectors of Botswana’s econ-
omy: the public, private, and parastatal sectors. We cor- rect for choice of employment among the three sectors.
The model performs better for the public and parastatal sectors, explaining more than 50 or more of the vari-
ation in earnings in those sectors. The explanatory power of the model is, however, relatively weak in the private
sector. For the private sector, the model explains about one-third of the variation in earnings. This could be
explained by the fact that formal education is geared towards producing public sector skills: clerks, teachers,
etc., and not to producing private sector skills. Experi- ence is more important in determining earnings in the
private sector and least important in the public sector. The coefficient of the square of experience is not sig-
nificant at the 5 level for both the public sector and parastatals. Given the lack of information about workers
in the public and parastatal sectors, one would expect experience to be an important variable in determining
earnings. Yet the results here show otherwise. This could be a reflection of wages and salaries in the public and
parastatal sectors being less tied to productivity. In most cases the public sector pays according to educational
qualifications rather than productivity, and also provides automatic promotion. On the other hand, in the private
sector, productivity is an important issue and thus experi- ence shows up as important in determining earnings. In
other words, it may be that unobserved individual influ- ences, such as effort, are more important in the private
sector. The average private rate of return to education is almost equalised between the three sectors. The choice
297 H.K. Siphambe Economics of Education Review 19 2000 291–300
Table 8 Basic Mincerian earnings function by sector of employment — corrected for choice of employment SS Data
a
Variable Public sector
Private sector Parastatals
Constant 6.4 24
4.9 45.08 7.8 14.
Education 0.09 7.8
0.09 7 0.08 5.1
Experience 0.028 2.5
0.07 5.5 0.057 2.5
Experience square 20.00023 20.6
20.0013 22.5 20.00077 21.09
Inverse Mills Ratio 20.6 24.9
0.47 2.8 21.2 24.1
R
2
adjusted 0.57
0.36 0.64
N 276
499 79
a
Values in parentheses are t-statistics. Asterisks indicate level of significance: , 1 level; , 5 level.
variable is significant at the 1 level for all three sectors.
9
7. Private rates of return to education