Sheepskin effects in the returns to education in a developing country.
Shabbir, Tayyeb
This paper tests for the sheepskin or diploma effects in the rates
of return to education in a developing country. Pakistan; presumably the
only study for the country that explicitly investigates this important
question. One reason for this paucity of work may have been lack of
appropriate data on an individual's educational status.
The Mincerian log-linear specification of the earnings function is
generalized to allow for the possibility that the returns to education
increase discontinuously for the years when diplomas/degrees are
awarded. This provision is made in three different ways, i.e., by (a)
introducing dummy variables for diploma years, (b) by specifying a
discontinuous spline function, and (c) by specifying a step function.
Empirical evidence based on a nationally representative sample of male
earners shows that substantial and statistically significant sheepskin
effects exist at four important certification levels in Pakistan,
namely, Matric, Intermediate, Bachelor's, and Master's. This
finding is consistent with the screening rather than the convential
human capital view of the role of education. However, it should be noted
that while diplomas seem to matter, it is not true that only diplomas
matter, since even after controlling for diploma years the schooling
coefficient, albeit smaller than before, is still substantial. Again,
regarding the diploma effects, another interesting finding is that such
effects are not significant in case of the Primary and the Middle levels
of schooling.
In terms of the policy implications, it follows that, in the case
of Pakistan, education is an important and significant influence on the
individual earnings. However, to the extent that the diploma effects am
significant, the potential for education as a source of enhancing worker
productivity is lessened, thus reducing the scope of an activist public
policy in this regard. This is particularly true for the Secondary
levels of education. In fact, the findings support a reallocation of the
available public funds away from the tertiary/higher education and
towards the basic education, where the productivity enhancing human
capital effects are relatively more apparent.
1. INTRODUCTION
The observed positive correlation between the labour market
earnings of individuals and their years of completed education is a
widely noted stylized fact. The standard interpretation of this
phenomenon is that of the human capital school, which considers this
correlation (with appropriate controls for labour market experience) as
being consistent with their view that higher earnings reflect the higher
worker productivity caused by increments in education. Amongst the
challengers to this view is the screening theory of education, which
treats it merely (or, at least, mostly) as a signalling device for the
pre-existing abilities that are useful in the word of work. The higher
earnings of the more educated, the proponents of the screening view
argue, really reflect payments for these 'latent' abilities
that are sought by the employers. Both these theories have been
described well in the literature. For instance, see Arrow (1973) and
Spence (1974) for the screening view and Becker (1964) and Mincer (1974)
for the human capital view.
It is important to know which theory is closer to the truth since
their respective views about the role of education have important
implications for public policy. This is particularly important for the
developing countries where any misallocation is extra costly, since the
resources are relatively scarce. However, distinguishing between the
above theories on grounds of empirical evidence is complicated by the
fact that the data on any explicit skills tests which may measure
increments to human capital are rarely available. "In fact, since
the sheepskin hypothesis is considered as one of the testable
predictions of the screening theory, testing this hypothesis is often
considered to be an indirect way of resolving the above debate. Here a
few words about the nature of the sheepskin hypothesis may be in order.
(1) Quoting Riley (1979), the sheepskin prediction states that
"wages will rise faster with extra years of education when the
extra year also confers a certificate".
The objective of the present paper is to test the sheepskin
hypothesis for Pakistan. This exercise is motivated mainly by the
following two factors:
(a) There are no studies that specifically focus on testing the
sheepskin hypothesis for Pakistan in spite of the fact that, since the
1970s, screening theories have figured prominently in the debate about
the effects of education; and
(b) Presently there is a renewed interest in testing this
hypothesis both for the developed as well as the developing countries.
Let me elaborate further on the nature of the above factors.
While apparently there are no studies for Pakistan which explicitly
address the sheepskin hypothesis, there is some incidental and partial
evidence in Hamdani (1977) and Guisinger et al. (1984) which suggests
that there may be a higher (i.e., 'bonus') rate of return for
completed 'Primary' relative to 'Incomplete
Primary'. The above evidence lends support to the presence of
sheepskin effects at the level of Primary School Certificate. However,
the limitations of the data used in these studies of the private rate of
returns to the different levels of schooling do not allow any inferences
regarding the sheepskin question for other certificate/diploma levels.
(2)
In any event, the above evidence is fragmentary at best and the
fact remains that there is a vacuum in terms of studies for Pakistan on
the important topic of the credentialist effects of education. One
reason for this paucity of studies may have been the lack of appropriate
data on an individual's educational status. As discussed further in
the data section of the present paper, most of the available micro level
surveys lack data on dropouts. Also, in such surveys the
individual's education is reported not as years of completed
schooling but rather as a discrete variable whose form does not allow
testing of the sheepskin effect. (3)
Apparently, when it comes to specific studies of the sheepskin
effects of education, the situation for other developing countries is
not very different from that of Pakistan. In general, there have been
few such studies for the developing countries. (4) However, there is
evidence of substantial recent interest in this important area. (5)
Studies by van der Gaag and Vijverberg (1989); King (1988) and Mohan
(1986) are cases in point. The first study, i.e., van der Gaag and
Vijverberg (1989), estimates the returns to schooling for a sample of
male and female wage-earners in Cote d'Ivoire using the 1985 LSMS (Living Standards Measurement Study) data. It finds that when
dichotomous dummy variables that represent completed diplomas are
included in the earnings function, they have positive and significant
coefficients. It turns out that schooling needs completion of diplomas
for most, albeit not all, of its impact. Similar support for the
sheepskin hypothesis is reported by King (1988) which is based on a
1985-86 LSMS sample of 5600 women in Peru. (6) Mohan (1986), on the
other hand, finds that diploma variable is important only for men but
not for women in Colombia.
As against the situation for the developing countries, considerable
work has been done on testing the screening theories for the developed
countries. At this point, in order to provide a broad perspective to the
debate, let me present an overview of the screening-related literature
for the United States.
In fact, starting from the middle 1960s, the fortunes of the
screening theories seemed to have ebbed and flowed. In the early 1970s,
such theories had started to seriously challenge the human capital view.
In fact, Taubman and Wales (1973) was an early important study which
concluded that screening effects were very important in the U. S. labour
market. However, an oft-cited study, Layard and Psacharopoulos (1974),
after critically reviewing several empirical studies of the screening
theory, dismissed it on the grounds that several of its refutable predictions, including the sheepskin hypothesis, were not supported by
the available evidence. With particular reference to the sheepskin
hypothesis they conclude, (7) ".... rates of return to dropouts are
as high as to those who complete a course, which refutes the sheepskin
version of the screening hypothesis".
The response to the above criticism ranged from a complete
acceptance of the views regarding the demise of the screening theory
[Addison and Siebert (1979), p. 139] to a defensive re-statement of the
screening theory which stressed that some versions of the screening
hypothesis do not imply sheepskin effects [Riley (1979)]. However, a
different approach has been taken in a recent article by Hungerford and
Solon (1987). It specifically questions the conclusion reached by Layard
and Psacharopoulos (1974) by arguing that the estimated rates of return
used by them were based on data that failed to disaggregate the earnings
of the dropouts by the number of years of schooling they had actually
completed. In fact, using a U. S. data set, Hungerford and Solon (1987)
present new evidence about "substantial and statistically
significant" sheepskin effects. (8) Thus, they maintain that some
of the earlier studies such as by Layard and Psacharopoulos may have
prematurely "dismissed screening theories of education partly on
the ground that diploma years of education do not confer especially
large earnings gains".
The rest of this paper is organized as follows:
Section 2 presents the proposed methodology while Section 3
describes the data. Section 4 reports the results. Finally, Section 5
contains conclusions as well as certain caveats.
2. METHODOLOGY
Testing the sheepskin hypothesis is tantamount to asking the
question whether the returns to education increase discontinuously in
diploma years. The proposed methodology to do so essentially involves a
re-specification of the traditional human capital earnings function
given by Mincer (1974). In fact, a good starting point for our
discussion would be this 'Mincerian' specification where, as
is well known, log earnings (Y) is posited to be a linear function of
years of completed schooling (S), labour market experience (EXP) and its
square. In the present context, however, it would be convenient to
suppress the experience terms; then, the shortened. (9) Mincerian
earnings function can be written as follows:
Y [alpha] + [[beta].sub.1] S ... ... ... ... (1)
The important point regarding (1) is that the rate of return to
schooling, [differential]Y/[differential]S = [[beta].sub.1], is
constant. In effect, this implies that all years of schooling are
'created equal' in terms of their marginal impact on log
earnings. In particular, there is no 'premium' or
'bonus' rate of return if the marginal year of schooling marks
the completion of a degree/ diploma. (10)
In order to test for the possibility that the returns to education
increase discontinuously in diploma years (i.e., the sheepskin effect
exists), we take the following approaches:
(a) We generalize the human capital log-linear specification (1) to
allow for discontinuities at values of S, which correspond to award of
degrees; and
(b) We specify Y to be a step function of S with a separate step
for each year of completed schooling. Then the 'step size' for
diploma years is compared with that of the years of schooling leading up
to the diploma.
In the discussion that follows, Models I and II correspond to
approach (a) while Model III corresponds to approach (b) above. In order
to further elaborate these approaches, let us discuss them in turn.
First, in order to elaborate approach (a), let us suppose that
there are only two diploma years corresponding, respectively, to ten and
twelve years of completed schooling. Define D10 and D12 as two
dichotomous (0, 1) variables such that D10 = 1 if S [greater than or
equal to] 10 and D12 = 1 if S [greater than or equal to] 12.
Model I: "Dummies for Degrees"
In this case, the relevant discontinuities are allowed for by
simply adding the dummy variables D 10 and D 12 to the traditional human
capital function. The following Equation represents Model I.
Y = [alpha] + [[beta].sub.1]S + [[beta].sub.2]D10 +
[[beta].sub.3]D12 ... (2)
Then, significantly positive regression estimates of D10 and D12
would imply sheepskin effects.
Also note that for every n diploma years, the graph of
[differential]Y/[differential]S gets divided into (n + 1)
'segments' over the domain of S. In the case of Equation (2),
the three relevant regimes defined over the domain of S are given by 0
< S < 10, 10 < S < 12 and 12 < S < [infinity] (See
Figure 1 (a')). The relevant marginal rate of return, r, over the
domain of S are given in the table below:
Years of Completed Schooling (S) Marginal Rate of Return (r)
S < 10 [[beta].sub.1]
S = 10 [[beta].sub.1] + [[beta].sub.2]
10 < S < 12 [[beta].sub.1]
S = 12 [[beta].sub.1] + [[beta].sub.3]
S > 12 [[beta].sub.1]
Model II: Discontinuous Spline Function
Model II posits the relationship between log earnings and schooling
to be a 'discontinuous' spline function with discontinuities
at diploma years. (11) This model is represented by the following
equation.
Y = [alpha] + [[beta].sub.1]S + [[beta].sub.2]D10 +
[[beta]'.sub.2](D10)(S-10) + [[beta].sub.3]D12 +
[[beta]'.sub.3] (D12)(S-12) ... (3)
[FIGURE 1 OMITTED]
Like Model I, here too the dummy variables D10 and D12 allow for
the sheepskin effects which would be implied by positive and significant
regression coefficients for these variables. However, (12) Model II
differs from Model I since the former allows for
[differential]Y/[differential]S to vary across the different regimes
defined over the domain of S, but note that within any given regime,
[differential]Y/[differential]is still presumed to be constant. (Figure
1 Co")).
Again, the marginal rate of return, r, across the domain of S can
be calculated in terms of the parameters of Model II. The appropriate r
is given below.
Years of Completed Schooling (S) Marginal Rate of Return (r)
S < 10 [[beta].sub.1]
S = 10 [[beta].sub.1] + [[beta].sub.2]
10 < S < 12 [[beta].sub.1] + [[beta]'.sub.2]
S = 12 [[beta].sub.1] + [[beta]'.sub.2]
+ [[beta].sub.3]
S > 12 [[beta].sub.1] + [[beta]'.sub.2]
+ [[beta]'.sub.3]
Model III: Step Function
For purposes of exploring the relationship between shooling and
earnings for possible diploma effects, the final specification of
interest is the so-called 'step function'. In this case, no
restrictions are imposed on the earnings-schooling profile - the log of
an individual's earnings is treated as a 'step function'
of years of completed schooling with a separate step for each year. (13)
For K years of completed schooling, such a specification can be
represented by Equation (5), which is given below.
(5) y = [alpha] + [K.summation over (i-1)] [[beta].sub.i] [D.sub.i]
... (5)
Here each [D.sub.i] is a (0, 1) dichotomous variables where
[D.sub.i] = 1 if S = i.
The estimated regression coefficients [[beta].sub.i] can be used to
calculate the implied step size in terms of the 'marginal'
rate of return to an additional year of schooling. Thus, in order to
evaluate the potential sheepskin effects, the step size for the year
conferring a particular diploma can be compared with the step size
corresponding to each of the years leading up to that diploma.
3. DATA DESCRIPTION
The data set used in the present study has been put together by
merging information from the Household Income and Expenditure Survey
(HIES) and the Migration Survey, which were, in fact, two of the four
separate 'modules' (i.e., questionnaires) of the 1979
Population, Labour Force and Migration (PLM) Survey. Conducted as a
joint project of the Pakistan Institute of Development Economics (PIDE)
and ILO-UNFPA, the PLM, a nationally representative survey, was based on
a two-stage-stratified random sample of 11,288 households. (14)
The same households were asked to respond to four sets of
questionnaires, two of which, i.e., HIES and Migration, are relevant
here. These surveys were conducted during the last two quarters of 1979.
Whereas HIES is a survey that is conducted with some regularity, the
Migration Survey was a one-shot thing done only in 1979. However, by the
time this survey was completed, it had spilled over into the first three
or four months of 1980 as well.
HIES has information on some but not all the variables that are
needed for testing the sheepskin hypothesis as outlined in this paper.
More specifically, while HIES has the relevant damon monthly earnings,
age, gender, and employment status of individuals, the information on
their schooling is not available in an appropriate form. In this survey,
the question regarding the individual's schooling is so designed
that the possible responses are 1-digit codes, e.g., 'Primary but
less than Middle' is assigned code 3; 'Middle but less than
Matric' is assigned code 4, and so forth. Thus, it is not possible
to distinguish those who complete a course and stop there from those who
start the next level but drop out. This makes the HIES's schooling
variable inappropriate for the present study since we need information
on the exact number of years of completed schooling for each
individuals. (15) Interestingly, in the Migration Survey, the question
regarding schooling has been designed in a manner that is appropriate
for providing the above information. Since the same households were
targeted for both the surveys, I decided to match individuals across the
two modules and pick up schooling information from the Migration Survey.
Thus, the resultant data set which has been used in this study has
schooling information from the Migration Survey while all other
information is from HIES. This 'merge' enables us to obtain
perhaps the only nationally representative sample for Pakistan where
schooling is measured as a continuous variable measured in terms of the
exact number of years completed.
As a result of restricting the observations to those for male
earners (wage earners or salaried employees) for whom 0 < S < 16
and Y > 0, a sample of size 1568 is obtained. In fact, Table 1
provides the details about the definitions of variables, their sample
means, and standard deviations.
4. RESULTS
Tables 2 through 4 give the regression results for the various
specifications that are relevant for testing the sheepskin effects in
the return to education. The dependent variable in all the cases is the
natural logarithm of an individual's monthly earnings (.Y). Let us
first look at Table 2, where column 1 presents the OLS estimates of the
typical human capital earnings function due to Mincer (1974). This
specification would serve as a 'reference' point. It treats
the log earnings (Y) as a linear function of years of completed
schooling (S), labour market experience (EXP) and its square. The
results show that the coefficients for all these variables are
significant at the 95 percent level of significance. In particular, the
coefficient for S implies that the rate of return to an additional year
of schooling is 9.7 percent.
Columns 2 and 3 of Table 2 give regression estimates that
correspond to Model I, i.e., 'Dummies for Degrees'
specification. (Column 3 differs from column 2 only in terms of having
two additional certification levels). First, note that both these
specifications (columns 2 and 3) turn out to be superior to the human
capital specification (column 1). An F-test of the human capital
specification relative to the alternatives in column 2 or column 3
rejects it at the .01 level of significance. In fact, it is noteworthy
that in the correctly specified model, the schooling coefficient drops
to half of what it was in column 1. Equally importantly, in both cases,
the coefficient estimates for the dummy variables corresponding to the
four important certification levels, (16) namely, Matric (D 10),
Intermediate (D12), Bachelor's (D14) and Master's (D16), are
positive and significant at the 95 percent level. Thus, we find strong
support for the sheepskin hypothesis. In order to further elaborate on
the nature of these diploma effects, (17) let me concentrate on the
results given in column 2.
Note that while the 'estimated effect on the log monthly
earnings of an additional year of schooling' (r) is 0.006 for the
9th year, it more than quadruples to 0.195 (obtained by adding .046 and.
149) for the 10th year (Matric). Again, there are equally spectacular
jumps for the other three diploma years that have been considered here.
For the 12th year (Intermediate) the r is .255 which is up from .046 for
S = 11; again, r is .233 for S = 14 (Bachelor's) relative to .006
for S = 13 and finally, it is .295 for S = 16 (Master's). Thus,
substantial and significant diploma effects in the returns to schooling
are found at these four important certification levels. In fact, the
null hypothesis of no sheepskin effects (i.e., all four dummy variables
D10, D12, D14 and D16 are simultaneously equal to zero) is rejected at
the 0.01 level (using F-statistic).
Again, column 2 of Table 3 presents regression estimates for a
specification that tests for the sheepskin effects in a slightly
different manner. Here, the interaction terms, i.e., D10INTER and
D12INTER allow for "slope" changes in the intra-diploma years.
In fact, this is the specification that we referred to as Model II
(Equation 3) in the section on Methodology. The results show that the
null hypothesis of no sheepskin effect (i.e., the coefficient estimates
for D10, D12, and D14 are all simultaneously zero) is rejected at the
.01 level. However, taken one by one, the estimated coefficients for D10
and D12 are positive and significant while the coefficient estimate for
D14 is positive but not significant. Again, none of the coefficient
estimates of the interaction variables is significant. This implies that
slope changes corresponding to intra-diploma years may not be very
pronounced. This would strengthen the case for the relatively simpler
specification of Model I. However, it is important to note that the
results for Model II are still supportive of the sheepskin hypothesis.
Finally, Table 4 corresponds to Model III, i.e., the step function
specification which provides an opportunity to look more directly at the
data since no restrictions are imposed on an individual's
schooling-earnings profile. Here the log of an individual's
earnings is essentially treated as a step function of the years of
completed schooling with a separate step for each year.
Positive step sizes are reported for all certification levels.
While the step size estimates for S = 5 and S = 14 are not significant,
large "upward" and significant step sizes are noticeable for S
= 8, S = 10 and S = 12.
To conclude this section, the empirical results of this study are
summarized below.
Using a nationally representative sample of male earners in
Pakistan, we find strong evidence for the existence of sheepskin effects
in the returns to education. While the details may differ across the
various specifications that have been considered in this paper, the
above result essentially holds true in all cases. (18) For instance, the
results for Model I show positive and statistically significant diploma
effects on the rates of return to schooling for four important
certification levels, namely, Matric (D 10), Intermediate (D12),
Bachelor's (D14), and Master's (D16). In one of the
specifications, where additional certification levels for the Primary
and Secondary are included, results show that the coefficient estimates
measuring these diploma effects are not statistically significant (Table
2, column 3). Some of the important implications of this observation may
be noted here. First, in terms of the theoretical debate between
credentialism and human capital explanations of the role of education,
it seems that the relative dominance of a given explanation may depend
on the level of schooling being considered. Thus the absence of the
diploma effects at the pre-secondary relative to the secondary and the
post-secondary levels implies that a case for human-capital-type
productivity enhancing the role of schooling can most strongly be made
for the former levels of schooling. At a practical level, this result
implies support for a re-allocation of the available public funds and/or
commitment of new funds to the relatively more basic rather than the
tertiary/higher levels of education.
In any cage, in order to complete the summary of the empirical
results, let us further note that based on the model selection criteria
involving F-statistic, the specifications allowing for sheepskin effects
were found to be superior relative to the typical 'Mincerian'
human capital earnings function. In fact, the true schooling coefficient
is almost 50 percent smaller than the one obtained by using the
misspecified human capital function (compare the coefficient estimate
for the years of schooling (S) as give in column 2 to that in column 1,
Table 2).
5. CONCLUSIONS/CAVEATS
The main conclusions of this study and their policy implications
are as follows. The finding that substantial and statistically
significant sheepskin or diploma effects exist at four important
certification levels in Pakistan, namely, Matric, Intermediate,
Bachelor's, and Master's, is consistent with the screening
rather than the convential human capital view of the role of education.
It is evident that diplomas in their capacity as signals for completed
courses of studies are important determinants of individual earnings and
ignoring them would lead to a serious misspecification of the earnings
function. However, it should be noted that while diplomas seem to
matter, it is not true that only diplomas matter; since even after
controlling for diploma years, the schooling coefficient, albeit smaller
than before, is still substantial. Again, regarding the diploma effects,
another interesting finding is that such effects are not significant in
case of the Primary and the Middle levels of schooling.
In terms of the policy implications of the above conclusions, it is
clear that, in the case of Pakistan, education is an important and
significant influence on the individual earnings. However, to the extent
that the diploma effects are significant, the potential for education as
a source of enhancing worker productivity is lessened, thus reducing the
scope of an activist public policy in this regard. This is particularly
true for the secondary and the post-secondary levels of education. In
fact, the findings support a re-allocation of the available public funds
away from the tertiary/higher education and towards the basic education,
where the productivity-enhancing human capital effects are more
apparent. It may be interesting to note that similar arguments, which
characterize basic education as the relatively more effective social
investment, are being made by others too. (19)
6. CAVEATS
In general, it is possible that our regression estimates showing
the presence of sheepskin effects may be biased due to the omission of
other factors, such as ability or family background, which are
correlated with degree completion. (20) However, presently, there is
little available evidence regarding the above issue, since the data on
ability or family background is not readily available for Pakistan.
However, Olneck (1979), after reviewing the results from a number of
studies for the U. S., reports that the estimated positive sheepskin
effects for college graduation prove to be robust when the variables to
measure ability and family background are included. (21)
Author's Note: I wish to thank Mr Masood Ishfaq for valuable
computer programming and Mr Ejaz Ghani and Mr Ayaz Ahmad for their
excellent research assistance. Discussions with Mr Mubashir Ali
pertaining to the PLM data were extremely beneficial. Of course, I alone
bear responsibility for any shortcomings of the paper.
REFERENCES
Addison, John T., and W. Stanley Siebert (1979) The Market for
Labour : An Analytical Treatment. Santa Monica, Calif.: Goodyear
Publishing Company.
Arrow, K. (1973) Higher Education as a Filter. Journal of Public
Economics 2:3 193 -216.
Becker, Gary S. (1964) Human Capital. New York: National Bureau of
Economic Research.
Behrman, Jere R. (1990) Human Resource-Led Development? Review of
Issues and Evidence. New Delhi: ILO-ARTEP.
Behrman, Jere R., and Nancy Birdsall (1987) Communication on
'Returns to Education: A Further Update and Implications'.
Journal of Human Resources 22:4 603606.
Berry, Albert (1980) Education, Income, Productivity, and Urban
Poverty. In T. King (ed) Education and Income. Washington, D. C.: (July)
153-230. (World Bank Staff Working Paper No. 402)
Chiswick, Barry (1973) Schooling, Screening, and Income. In Lewis
Solomon and Paul Taubman (eds) Does College Matter? New York: Academic
Press.
Guisinger, S. E., J. W. Henderson and G. W. Scully (1984) Earnings,
Rates of Return to Education and the Earnings Distribution in Pakistan.
Economics of Education Review 3:4.
Hungerford, Thomas, and G. Solon (1987) Sheepskin Effects in the
Returns to Education. Review of Economics and Statistics 69:1 175-177.
Hamdani, Khalil A. (1977) Education and the Income Differential: An
Estimation for Rawalpindi City. The Pakistan Development Review 26:2
144-164.
Haque, Nadeem UI (1977) Economic Analysis of Personal Earnings in
Rawalpindi City. The Pakistan Development Review 26:4 353-382.
Irfan, Mohammad (1981) An Introduction to Studies in Population,
Labour Force and Migration. PIDE/ILO-UNFPA Project. Islamabad: Pakistan
Institute of Development Economics. (Research Reports Series, No. 118).
Johnston, J. (1984) Econometric Methods. New York: McGraw-Hill Book
Company. 392-396.
Khan, Shahrukh Raft, and Mohammad Irfan (1985) Rates of Returns to
Education and the Determinants of Earnings in Pakistan. The Pakistan
Development Review 24: 3 & 4 671-680.
King, Elizabeth M. (1988) Does Education Pay in the Labour Market?
Women's Labour Force Participation, Occupation, and Earnings in
Peru. Washington, D. C.: World Bank. (Mimeographed)
Layard, Richard, and George Psacharopoulos (1974) The Screening
Hypothesis and the Returns to Education. Journal of Political Economy
82:5 985-998.
Mincer, Jacob (1974) Schooling, Experience and Earnings. New York:
National Bureau of Economic Research.
Mohan, R. (1986) Work, Wages and Welfare in a Developing
Metropolis. London: Oxford University Press.
Olneck, Michael (1979) The Effects of Education in Christopher
Jencks et al. (eds) Who Gets Ahead? New York: Basic Books.
Psacharopoulos, G. (1980) Returns to Education: An Updated
International Comparison. In T. King (ed) Education and Income.
Washington, D. C.: 73-109. (World Bank Staff Working Paper No. 402)
Riley, John G. (1979) Testing the Educational Screening Hypothesis.
Journal of Political Economy 87 : $227-$252.
Sabot, Richard H. (1989) Human Capital Accumulation in Post-green
Revolution Pakistan: Some Preliminary Results. The Pakistan Development
Review 28 : 4 413-431.
Spence, Michael A. (1974) Market Signalling: Informational Transfer
in Hiring and Related Screening Processes. Cambridge, Mass.: Harvard
University Press.
Taubman, Paul, and T. Wales (1973) Higher Education, Mental Ability
and Screening. Journal of Political Economy 81 : 28-55.
van der Gaag, Jacques, and Wim Vijverberg (1989) Wage Determinants
in Cote d'Ivoire: Experience, Credentials, and Human Capital.
Economic Development and Cultural Change 37 : 2 371-381.
(1) Webster's Dictionary characterizes "sheepskin"
as a colloquial word that refers to a "a diploma, sometimes made of
parchment (the skin of sheep etc., prepared for use as a writing
material)". In this paper, diploma, degree, and certificate will be
used interchangeably to represent formal evidence of completion of a
course of study.
(2) Both these studies focus on calculating rates of return to
different levels of education and are based on the 1975 PIDE-sponsored
'Rawalpindi City Survey'. Hamdani (1977); using the
'direct', i.e., Becket's internal rate of return method,
reports (marginal) rates of return of 7 percent for 'Incomplete
Primary' and 20 percent for 'Completed Primary' while
Guisinger et al. (1984), using the 'indirect' i.e.,
Mincer's earnings function approach, report 3.4 percent and 3.5
percent as the respective rates of return for the same levels of
education. The design of the questionnaire for the Rawalpindi City
Survey led to schooling being measured as a discrete response variable
with responses such as 'Primary and Incomplete Secondary',
'Secondary and Incomplete Bachelor's', etc. Thus, one can
not distinguish between those individuals who finish a certain
certification level, say, Secondary school, and stop there from those
who go on but fail to successfully complete the next certification level
(i.e., Bachelor's). Only for the case of Primary was there a
distinct category for those respondents who did not complete this level.
Hence, in the text, we are able to report the
'sheepskin-related' results only for the Primary level
certificate.
(3) In the past, most of the studies of the rates of return to
education have been based either on the 1975 Rawalpindi City Sample
[Hamdani (1977); Guisinger (1984); Haque (1977)], or the 1979 household
Income and Expenditure Survey [Khan and Irfan (1985)]. In both these
cases, data on education is reported as a discrete variable. However, a
recent study, Sabot (1989), is based on a specialized sample where
education is reported as a continuous variable, i.e., years of schooling
completed.
(4) For the developing countries, while a large number of the
Becket/Mincer type of studies have been done [Psacharopoulos (1980)],it
is only recently that studies challenging the human capital
school's view have appeared [for additional details see Behrman and
Birdsall (1987)].
(5) Amongst the older studies, there is one for the Philippines and
another one for Colombia. Based on the results reported by [Berry
(1980), p. 202], these studies indicate a tendency for higher payoff to
certain completed educational levels relative to their partial
completion.
(6) [King (1988), p. 321 reports that "the estimated rates of
return to post-secondary education without receiving a diploma are
negative for both non-university tertiary and university education ...
-5 and -4 percent a year, respectively. Gaining a diploma greatly
increases the return to post-secondary education; the rates of return to
earning a diploma is 19 percent in the all-Peru estimates". Again,
the [van der Gaag and Vijverberg (1989), p. 378] study shows that, as a
result of introducing controls for diplomas, the rate of return for the
various levels of schooling falls precipitously for all levels except
for university level, where the drop is relatively moderate. More
specifically, the rate declines from 11.9 percent to 2.3 percent for
elementary school, from 20.9 percent to 8.8 percent for junior high
school, from 20.8 percent to -3.2 percent for senior high school, and
from 22.7 percent to 20.8 percent for university level.
(7) [Layard and Psacharopoulos (1974), p. 995].
(8) Their analysis was based on the May 1978 Current Population
Survey data on 16,498 white, male, non-agriculture wage and salary
workers between the ages of 25 and 64. For the individuals in the
sample, earnings increments associated with each year of a course
including the year of its completion are available.
(9) Incidentally, this corresponds to what [Mincer (1974), pp.
9-11] calls 'The Schooling Model', which he uses to set the
stage for his analysis of the effects of experience on individual
earnings.
(10) For the sake of completeness, it should be noted that in some
versions of his empirical earnings function, [Mincer (1974), p. 53] does
allow for non-linear effects of schooling by adding a quadratic S term
to (1). However, since these non-linearities are not specifically linked
to 'completion of diploma' years, the comment in the text
would still be valid.
(11) For a general introduction to the concept of spline functions,
see [Johnston (1984), pp. 392-396]. For a specific application of this
methodology see Hungerford and Solon (1987) which uses U. S. data.
(12) In fact, since setting [[beta]'.sub.2]
[[beta]'.sub.3] = 0 in Model H gives Model I, the latter can be
considered as a special discontinuous spline function. However, Model I
is generally more easily recognizable as simply a case of introducing
dummy variables to the standard human capital specification.
(13) For an application to the U.S. data, see [Hungerford and Solon
(1987), p. 177].
(14) See Irfan (1981) for further details.
(15) Incidentally, the HIES for other years too have the same
design regarding the question of schooling. In fact, extremely few micro
level surveys for Pakistan have measured schooling as a continuous
variable. One exception is the data set used in Sabot (1989). However,
this sample is not national in character since it i s based on a sample
of 800 rural households.
(16) For comparison, the relevant Pakistani diplomas as well as
their U.S. counterparts with their corresponding years of schooling are
given below:
Diplomas
Years of Schooling Pakistan U.S.
5 Primary
8 Middle Elementary
10 Matric
12 Intermediate (F.A./F.Sc.) High School
14 Bachelors (B.A.B.Sc.)
16 Master's (M.A./M.Sc.) College
(17) The estimated coefficient for D5 as well as D8, the two
additional certification levels in column 3, as compared to column 2 are
insignificant while the rest of the results across the two columns are
quite similar.
(18) In fact, as a practical matter as well as on the basis of the
conventional selection criteria such as [R.sup.2], theoretical
consistency, and statistical significance of the coefficient estimates,
Model I may be the most preferred specification.
(19) [Behrman (1990), p. 90].
(20) A related argument is that of Chiswick (1973), who contends
that dropouts are essentially those individuals who overestimated their
ability to benefit from education. However, while Chiswick's
hypothesis is consistent with education as being productivity-enhancing,
it is not in the spirit of the standard Becket/ Mincerian human capital
tradition. So, at the minimum, results showing strong diploma effects
imply a need for reformulating the standard human capital argument.
Again, a direct test of the Chiswick hypothesis would be possible only
if the 'ability to benefit from education' data for all
entrants to a programme are available. In fact, to the extent that the
general measures of ability or family background correlate with the
specific ability Chiswick has in mind, these general measures would
provide a partial control for it.
(21) Table 6.3, pp. 178-9. The education variables are defined on
page 161.
Tayyeb Shabbir is Research Economist at the Pakistan Institute of
Development Economics, Islamabed.
Table 1
Description, Means (X) and Standard Deviations (S.D.) of Some
Important Variables (N =1568; Male Earners)
Variable's X S.D. Variable's Definition
Name
Y 6.42 0.68 Natural logarithm of the person's
monthly earnings which may con-
sist of wages or salary.
S 8.59 3.46 Years of schooling completed.
AGE 33.03 12.10 Age in years.
EXP 18.44 12.46 Total years of labour market experi-
ence; (Age-S-6).
D5 0.87 0.34 Dichotomous, equals 1 if
S [greater than or equal to] 5.
D8 0.65 0.48 Dichotomous, equals 1 if
S [greater than or equal to] 8.
D10 0.49 0.50 Dichotomous, equals 1 if
S [greater than or equal to] 10.
D12 0.21 0.41 Dichotomous, equals 1 if
S [greater than or equal to] 12.
D14 0.11 0.32 Dichotomous, equals 1 if
S [greater than or equal to] 14.
D16 0.03 0.17 Dichotomous, equals 1 if S = 16.
D5INTER 3.83 3.11 (D5) (S-5)
D8INTER 1.76 2.16 (D8) (S-8)
D 10INTER 0.73 1.52 (D 10) (S-10)
D 12INTER 0.29 0.85 (D 12) (S-12)
D 14INTER 0.06 0.35 (D 14) (S-14)
S1 - S16 Dichotomous = 1 if Sn = n
where n=1, ... ,16.
Table 2
Model I: 'Dummies for Degrees' (OLS; Dependent Variable
= Y; Male Earners)
1 2 3
Constant 4.872 * 5.166 * 5.155 *
(94.692) (77.985) (52.980)
S 0.097 * 0.046 * 0.045 **
(24.425) (5.072) (1.790)
EXP 0.1056* 0.056 * 0.056 *
(16.270) (16.649) (16.537)
[(EXP).sup.2] -.0006 * -.0006* -.0006 *
(-8.931) (-9.387) (-9.300)
D5 0.0294
(0.388)
D8 -0.0224
(0.278)
D10 0.149 * 0.159 *
(3.049) (2.627)
D12 0.209 * 0.211
(4.115) (3.052)
D14 0.187 * 0.189 *
(2.959) (2.398)
D16 0.249 * 0.250 *
(2.734) (2.436)
Adjusted [R.sup.2] 0.40 0.42 0.42
N 1568 1568 1568
Note: Results for Model 1 are given in columns 2 and 3.
* Significant at 95 percent level; 2-tailed t-test (t-statistics are
in the parentheses).
** Significant at 90 percent level for 2-tailed t-test yet significant
at 95 percent level for 1-tailed t-test.
Table 3
Model II: Discontinuous Spline Function (OLS; Dependent Variable
= Y; Male Earners)
1 2
Constant 5.135 * 5.165 *
(78.558) (77.888)
S 0.051 * 0.046 *
(5.715) (5.110)
EXP 0.056 * 0.056 *
(16.698) (16.525)
[(EXP).sup.2] -.0006* -0.0006 *
(-.9416) (-9.319)
D10 0.128 * 0.151 *
(2.657) (3.089)
D10INTER -0.102
(-.843)
D12 0.200 * 0.408 *
(3.933) (1.687)
D12INTER 0.131
(0.495)
D14 0.238 * 0.131
(3.936) (0.280)
D14INTER 0.095
(0.339)
Adjusted [R.sup.2] 0.42 0.42
N 1568 1568
Note: Result for Model II are given in column 2.
* Significant at 95 percent level; 2-tailed t-test (t-statistics
are in the parentheses).
** Significant at 90 percent level for 2-tailed t-test yet significant
at 95 percent level for 1-tailed t-test.
Table 4
Model III: Step Function (OLS; Dependent Variable = Y;
Male Earners)
1 Step Size
Constant 6.699 *
(80.288)
EXP 0.055 *
(16.268)
[(EXP).sup.2] -0.0006*
(9.050)
S = 1 -1.245 * -1.245 *
(-5.123) (-5.123)
= 2 -1.559 * -0.314
(14.578) (-1.295)
= 3 -1.401 * 0.158
(-13.485) (1.572)
= 4 -1.314 * 0.087
(13.980) (0.998)
= 5 -1.305 * 0.009
(-15.360) (0.139)
= 6 -1.128 * 0.177 *
(11.788) (2.610)
= 7 -1.273 * -0.145
(12.332) (1.611)
= 8 -1.139 * 0.134 **
(13.242) (1.688)
= 9 -1.209 * -0.07
(-12.132) (-0.944)
= 10 -0.916 * 0.293 *
(-11.333) (4.321)
= 11 -0.973 * -0.057
(-6.894) (0.472)
= 12 -0.621 * 0.352 *
(-0.549) (2.809)
= 13 -0.549 ** 0.072
(-2.249) (0.306)
= 14 -0.341 * 0.208
(-3.826) (0.883)
Adjusted [R.sup.2] 0.42
N 1568
Note: Results for Model III are given in column 1.
* Significant at 95 percent level; 2-tailed 1-test (t-statistics
are in the parentheses).
** Significant at 90 percent level for 2-tailed 1-test yet significant
at 95 percent level for 1-tailed t-test.