You are on page 1of 20

The Effect of Part-Time Employment on Academic Achievement:

Evidence from Dhaka University

Akib Khan (005)*, Shoaib Anwary (027)*,†, Sania Aiman (062)*, Md. Nazmus Sakib (087)*,
Afsana Abedin (099)*, Shams-E-Tabriz (050)*, Nazifa Tabassum (121)*
*
Department of Economics, University of Dhaka, Dhaka, Bangladesh

Representative (Group 9)

Abstract
This study estimates the impact of hours worked at a part time job on GPA among 6 th semester students from
different departments of Dhaka University. We have used two stage least squares (2SLS) with unearned income
categories and class-level employment rate, which have considerable explanatory power for work intensity, as
instruments. Results show that GPA monotonically declines with increments in (part-time) labor supply, though the
decline is fairly moderate. Moreover, 2SLS estimates are considerably higher (in absolute value) than that from
ordinary least squares.

Key words: Part-time employment, GPA, 2SLS estimation, Unearned income, Class-level employment rate.

Table of Content
s
1. Introduction........................................................................................................................................3
2. Literature Review................................................................................................................................4
3. Data.....................................................................................................................................................5
3.1. Sample Specification....................................................................................................................5
3.2. Variable Specification..................................................................................................................6
3.3. Students’ Participation in the Part-time Job Market: An Overview..............................................6
4. Findings...............................................................................................................................................7
4.1. Econometric Specification............................................................................................................7
4.2. Summary Statistics.......................................................................................................................9
4.3. Regression Results.....................................................................................................................10
5. Concluding Remarks..........................................................................................................................12
Bibliography..............................................................................................................................................13

List of Tables and Figures

Table 1: Part-time Employment Scenario in Subsamples..........................................................................15


Table 2: Hours Worked and Mean (observed) - Unearned Income, Employment Rate & GPA.................16
Table 3: Summary Statistics for Endogenous and Exogenous Variables....................................................16
Table 4: Impact of Hours Worked on GPA (OLS and 2SLS Estimates)........................................................17
Table 5: Coefficients of Instruments in 1st Stage Hours Equation.............................................................17
Table 6: Self-reported Impact of Part-time Employment on Academic and Social life (by Hours Worked)
.................................................................................................................................................................. 18

Figure 1: Relative Share of Different Types of Part-time Job.....................................................................18


Figure 2a: Hours Worked and Predicted GPA (instruments: Unearned Income and Employment Rate)...19
Figure 2b: Hours Worked and Predicted GPA (instruments: Unearned Income) …………………………………...19
1. Introduction

Estimating the causal impact of part-time employment on the academic accomplishments of university
students has been of particular significance to both researchers and policymakers. This is because, the
trade-offs associated with a student’s labor-supply decision while studying (as an undergraduate) are
considerably different and counter-acting. Expected positive contribution of enhanced skill and practical
job experience (through participation in a part-time job) to ensuring a better post-graduation job-market
entry may be nullified by its potential detrimental impact on the academic outcome.

This paper concentrates on investigating the latter issue and in particular, has aimed to estimate the
ceteris paribus impact of part-time employment on undergraduates’ academic performance. We have
chosen GPA to proxy for the academic outcome since it affects both the placement in graduate schools
(for those who will continue to invest in education) and, the wage and occupational choice (for those
who will join the full-time labor force immediately upon graduating from the university). Work intensity,
on the other hand, has been defined as the average working hours per week (over the period of
concern).

Skill and experience that a paid job offers can substitute for pure academic knowledge. In that case,
part-time employment may be conducive to decreasing the amount of time available to improve
academic achievements, thus having a negative impact on GPA. Higher work intensity may also work
towards decreasing academic productivity by causing fatigue and/or distraction. Job earnings often
increase the possibility of being exposed to addictive consumption goods. Thus short-run gains, financial
or non-financial, can be prioritized over investments in education. On the other hand, engagement in
part-time job(s) is also capable of complementing academic dexterity via developing confidence,
diligence, efficiency, time-management capability and responsibility (D’Amico 1984). In addition, time
spent working, at least, substitute for the leisure time during which they had a higher probability of
being engaged in illegal, unethical or counter-productive activities.

Recent developments in this area of research, both theoretical and empirical, require the estimation of
the causal effect of employment to overcome the problem of endogeneity inherent in the labor-supply
decision which, in turn, can give rise to bias and inconsistency in the estimation process. Put formally,
variations in working hours may be attributable to certain unobservable characteristics (of an individual
student) which also have an impact on the academic outcome. Resultant will be the presence of
spurious correlation – thus biasing the treatment effect. As an example, a positive relationship may
emerge if those attaining better GPA are characterized by attributes that also promote productivity in
both schoolwork and paid employment – thus creating an upward bias in the OLS estimator. The worst
case scenario would be where reverse causation takes place, i.e., when performance determines the
level of labor (to be supplied).

This paper is the first of its kind in Bangladesh in the sense that no other attempts to our knowledge has
yet been made in analyzing this issue. This, despite providing us with some privilege, also poses several
serious problems. While researchers abroad have exploited the results from prior studies (of diverse
background) in choosing valid instruments and in making necessary case-specific methodological
adjustments, the dearth of such research has left us with limited opportunities but to follow their
empirical approach. Despite facing such and other limitations, in order to set an example, we have opted
to estimate the causal effect of part-time employment on undergrads’ academic score. To obtain
exogenous variations in hours worked (per week), we have used a set of indicators representing
different categories of unearned income as well as the class-level employment rate(s) as instruments.
From both theoretical and empirical point-of-view, these (as will be described in details in the following
sections) satisfy the requirements for being valid instruments. Our estimates (using a random sample of
DU students) also appear to be consistent with the recent-most empirical findings (discussed in Section
2 below).

Hereafter, the paper will proceed as follows. Section 2 reviews, in brief, the stock of prior literatures,
whereas Section 3 provides sample and variable specifications as well as an overview of the part-time
employment scenario for the DU students. Empirical strategy and findings (both descriptive and
inferential) are presented in Section 4. Finally, the paper concludes through Section 5 by making some
final comments, discussing the limitations and highlighting the potentials of additional investigative
studies.

2. Literature Review

Academic interest in estimating the impact of part-time employment on academic accomplishment


blossomed in the early 1980s. Steinberg et al. (1982) found that first-time workers spent less time on
homework and had lower GPAs (from a sample of California 10th and 11th graders). On the contrary,
on the basis of National Longitudinal Survey of Youth (NLSY) data (covering the period 1979-1982),
D’Amico’s (1984) estimations suggest that working up to 20 hours a week was correlated with
improvements in class rank and mitigated dropout rate. In addition, working beyond 20 hours per week
was found to have no association with class percentile, but was responsible for a rise in dropout rates –
though for only a very few subgroups. Similar estimates were obtained by Schill et al. (1985), who
estimated that those working less than 20 hours per week had higher GPAs as opposed to those who did
not work or worked more than 20 hours.

Empirical research in this field, as time progressed, continued to produce such mixed evidence. Studies
which specified employment and hours worked as exogenously determined succeeded to find positive
association between labor supply and performance, at least for modest level of (part-time) labor.
Examining the early-1980s High School and Beyond (HSB) data, Marsh (1991) concluded that hours
worked in 10th grade had positive association with the subsequent rate of dropout. Besides, his
estimations suggest that higher total working hours during the final three years of high school, on
average, reduced senior year grades and standardized test scores. Some other single-equation studies
used National Education Longitudinal Survey (NELS-88) data (on students who were 10th graders in 1990
and 12th graders in 1992). For the 10th graders, Schoenhals et al. (1998) found that while absenteeism
escalated slightly with increased degree of participation, adverse employment effects on grades and
homework time could be entirely attributed to pre-existent divergences across the students working at
varying intensities. Singh (1998), however, estimated that for employed 10th graders, employment had
an inverse relationship with both standardized achievement scores and grades. According to Quirk et al.
(2001), cumulative GPA among 12th graders was highest for those working about six hours per week,
but was lower for those who worked more than 12 hours per week than for those who self-selected
away from the labor market.

Unfortunately, seeking to control for the endogeneity of the labor supply decision in the academic
achievement equation has also failed to provide definitive evidence. Investigating data on 11th and 12th
graders in the 1987 National Assessment of Economic Education Survey, Lillydahl (1990), by means of
instrumental variable regression, found that GPA increased with work intensity up to about 14 hours per
week, while working beyond that caused a decline in performance; but, working also enhanced scores
on the Test of Economic Literacy. Warren et al. (2000), too, estimated a 2SLS model, using NELS-88 data
and identifying school-year work with zip code level employment rates as instruments. Their results
were in agreement with those obtained by Schoenhals et al. (1998); they found that high school
employment did not have any impact on the grades (with pre-existing heterogeneity between more and
less intensively employed students explaining the negative association between work hours and grades).
Oettinger (1999) opted to estimate the treatment effect with individual-specific fixed effects for
continuously enrolled NLSY respondents with comprehensive GPA and weekly employment history
information during 11th and 12th grades. His results revealed that, employment and grades were
unrelated for Whites but had a substantial negative relationship for Blacks and Hispanics. Tyler (2003)
and DeSimone (2006) offer the most recently published evidence on the subject. Both of them
estimated a 2SLS model, but with different instruments; Tyler used state child labor laws, whereas
DeSimone instrumented work intensity using indicators of unearned income. Tyler found that 10
additional weekly work hours decreased math achievement scores by 0.2 standard deviations. Albeit
being appealing intuitively, the analysis had two noticeable limitations: the data were very old (almost a
decade), and the instrument did not demonstrate convincing explanatory power. DeSimone, however,
has managed to overcome some of these limitations quite successfully. His estimations also indicate the
presence of concavity in GPA-employment relationship: GPA rises with work intensity until 15 hours per
week and then declines. Moreover, 2SLS estimates, there, are notably larger than those yielded by OLS.

3. Data
3.1. Sample Specification

In recognition of the significant potential implications for both academic and non-academic arena, we
have exclusively concerned ourselves with (the students from) the University of Dhaka (henceforth, will
be referred to as DU), the largest public university in Bangladesh. Furthermore, being subjected to
various financial, institutional and physical constraints, we decided to restrict our population (and
hence, the resultant sample) to the students (from the 7 th Semester) corresponding to the faculties of
Arts, Social Sciences and Business Studies. Six departments 1 were randomly chosen, followed by the
collection of a 30% random sample from each of them so as to ensure cost-effectiveness. The sample,
thus drawn, has provided us with required information about 138 students.

3.2. Variable Specification


1
Banking, International Business and Tourism & Hotel Management from the Faculty of Business Studies;
Economics, Psychology and Mass Communication & Journalism from the Faculties of Arts and Social Sciences.
The sampling units (i.e., the students) were surveyed by means of a self-administered questionnaire
during normal class periods.2 Information on the labor supply were extracted from responses to the
question concerning the average hours worked (per week) during the last, i.e., the 6 th semester.
Academic performance – since we are interested in estimating the causal effect of part-time
employment - has been inferred from their self-reported figure of GPA obtained in that semester. 3
Apart from these which are key to our analysis, in order to account for differences across individuals
along several other dimensions which are probable candidates to create spurious correlations between
job-engagement and academic outcome - either exogenously determined (such as age, gender, paternal
& maternal education) or alterable (such as relationship status) or both (such as religious affiliation,
mode of accommodation, expected future investment in education, health status), the respondents
were required to provide some additional information. 4 Moreover, to be utilized later as natural
instruments (elaborated on in Section 4), data on unearned income (i.e., monthly allowance received
from the family) and employment rate for each class were collected. 5

As was stated in the preamble as one of the objectives, to construct a summary image of the would-be
graduates’ current employment status, we asked the sampled (employed) students to enlighten us
about some other aspects of (their) part-time employment. They were requested to specify the job-
type, to report the degree of flexibility permitted (by the employer), perceived rate of employment
among their classmates & friends and average monthly wage, and lastly, to comment on the
consequences of such engagement for (their) academic and social life. 6

3.3. Students’ Participation in the Part-time Job Market: An Overview

Table 1 provides with the share(s) of employed students in different subsamples (differentiated on the
basis of different indicators, namely gender, current mode of residence, religious orientation, paternal
and maternal schooling, relationship status and sibling’s association with DU). Estimates reveal that,
more than 41% of the students (surveyed) worked on a part-time basis in the 6 th semester.
Approximately three-fourth of them are male, thereby indicating the presence of a strong gender bias.
Focusing on the intra-gender participation rate, half of the male respondents reported themselves as
employed, whereas the figure is considerably low – only 27% - for the female ones. Despite being the
minority (about 40% of the sample), students dwelling at Hall or Mess constitute a large portion (46%) of
the working students, thanks to (their) higher rate of participation (47%) as compared to that of the
other group, i.e., those who are living with their parents and/or relatives (37%). In our sample, 8% of the
Hindu students (around 8.5% of the sample) chose to work part time, while the figure is 43.8% among

2
Please find the questionnaire attached herewith at the end of the document.
3
Since findings of Cassady (2001) suggest that self-reported GPA values have been “remarkably similar to official
records” and hence can be considered “highly reliable” and “sufficiently adequate for research use”, we are not
expecting the use of self-reported rather than actual GPA to introduce bias to our estimates.
4
For the complete list of questions, please see the questionnaire attached herewith.
5
Rate of employment for the respondent’s class was calculated by conducting an instant survey (requiring the
students to raise their hands if the reply is affirmative) preceding the distribution of questionnaire.
6
Note that, such comments, since they are contingent upon the experiences from the last semester, are ex-post,
and hence, more representative of the reality.
the Muslims (88% of the respondents). 7 Almost 43% of the students with varsity-educated father self-
selected themselves into the market for part-time jobs, whereas for the other group (characterized by
paternal schooling of less than 16 years), the participation rate is 38%. Indeed, the trend is reversed
when we sub-sample on the basis of maternal education: 63% of those with graduate mother(s) decided
to supply zero hours of part-time labor. But note that, the proportion of students whose father(s)
received (at least) 16 years of formal education - in the sample - is typically higher than that marked by
the same criterion but for maternal educational investment (66% vs. 32%). One-fourth of the (sampled)
students are married or engaged in a relationship, and approx. 31% of them were involved in a part-time
job. The rate of employment is higher, namely 44.7%, among the single ones. Finally, those who have
sibling(s) in DU (prior to their admission) have demonstrated a participation rate of 39% which is smaller
than that associated with the other group (43.4%).

In response to the question “what type(s) of job DU students are taking up as part-time commitment?”,
Figure 1 represents the job-wise share of part-time employment. The salient feature, indeed, is the
dominance of ‘Tuition’ over the other categories. Private and coaching-center based tuition, together,
share 84% of the entire pie. It has been followed by ‘Journalism’ – though a long way down the
spectrum – which covers only 11% of the working students. And according to the data at hand, as
expected, almost all of them (5 out of 6) belong to the Dept. of Mass communication of Journalism.
‘Internship’ and ‘Research Assistance’ are the remaining two categories (which are allocated 2% and 3%
of the employment pie, respectively). Hence, the part-time jobs, apparently, are concentrated in the
area of ‘Tuition’ – private or otherwise. This, however, isn’t so auspicious since, in general, without
being related to the discipline of education (currently inhabited by a student), part-time jobs have little
chance to make a significant potential contribution to a student’s future employment prospect.

4. Findings
4.1. Econometric Specification

This paper chooses to estimate the causal impact of hours worked (per week) on GPA. A sizeable
proportion of the hitherto-cited studies have found the employment-achievement relationship to be
characterized by non-linearity. In accommodating this non-linearity (to be precise, concavity), almost all
of them fitted a quadratic model. Whether such specification is appropriate in our case is purely an
empirical question. Table 2, which reports mean GPA across different hour-groups, however, fails to
enable us to discern any general pattern – linear or non-linear, which, in turn, has led us to opt for a
linear model at first and then perform certain statistical procedures to test for functional
misspecification.

As was articulated in the beginning, labor supply becomes endogenous in an academic outcome
equation if unobserved factors (e.g., innate ability) are correlated with both employment and
achievement and/or if part-time job participation is influenced by, instead of only influencing,
achievement. In either case, the consequence will be the introduction of both bias and inconsistency in
all the parameters (to be estimated) via a correlation (positive or negative) between the scalar errors
and e and v in the system of equations:
7
The figures are 100% and 2% respectively, for those who have been raised as Christians.
H = α 0 + Z α 1 + X α2 + v (1st stage)

G = β0 + Hβ 1 + X β2 + e (2nd stage)

Here, H denotes hours worked (per week), G stands for GPA, X represents the vector of exogenous
covariates (see Table 3 for the complete list) and the vector Z comprises instrumental variables that
have an impact on labor-supply decision but not on GPA. α and β are the parameters to be estimated; β1
is of particular importance to us since it is to capture the ceteris paribus impact of employment on
achievement. As explained above, the correlation of e with H through its (linear) association with v will
result in the OLS estimator of β1 being biased and inconsistent. In order to obtain a consistent estimator
for β1, we have used the traditional tool of 2SLS (Two-stage Least Squares) instrumental variable
regression along with heteroskedasticity-robust standard errors (for inferential purpose). Under this
statistical treatment, we are effectively adding the fitted values, obtained from running the first-stage
(OLS) regression, to the second-stage equation along with other exogenous controls (i.e., X), thus
allowing β1to be construed as the causal effect of working.

The instruments, as is well-known to an econometrician, must satisfy two conditions: they must be
highly jointly significant (as will be reflected in the joint F statistic) in the first-stage equation (or in other
words, they must have a high joint partial correlation with working hours), and they must be
uncorrelated with GPA (in the population) after controlling for other exogenous covariates (i.e., Corr(e,Z)
= 0). In verifying the latter, an overindentification test will be performed (under the null hypothesis that
all the instruments are exogenous).

Drawing from DeSimone (2006), Warren, et. al (2000), Ruhm (1997) and Oettinger (1999), we have used
instruments at both individual (as captured by indicators of unearned income) and aggregate levels (as
represented by employment rate of the respondent’s class). In what follows, we will elaborate on the
reasons behind such selection of instruments.

Three distinct categories of unearned income (with the ‘no unearned income’ category as the
benchmark) as well as the over-all employment rate (in a student’s class) are utilized as instruments to
identify the hours-worked equation. The primacy of this strategy (over others) stems from the fact that
these categories are likely to provide discrete pieces of information. Theoretical developments in Labor
Economics stipulate unearned income as a major causal factor in decisions regarding labor-supply (both
full and part-time). The assumption of diminishing marginal utility of income implies that an enhanced
level of unearned income reduces the marginal return to an additional hour of paid work, thus is
indicative of a strong negative (partial) correlation with labor supply. Also, DeSimone (2006) found
considerably large first-stage partial correlations between unearned income and hours worked. On the
other hand, a high participation rate among classmates may induce a student to join the part-time labor
market via the working of the peer effect. This is parallel to the arguments given in favor of the use of
zip-code or state employment rates as instruments (Warren, et al. 2000, Ruhm 1997 and Oettinger
1999). The presence of such theoretical linkage and empirical evidence leads us to expect the
instruments to have strong explanatory power for the observed variations in the labor supply (of DU
students). The estimates of mean unearned income and employment rate - corresponding to different
hour-groups (see Table 2) - albeit being crude, provide further support in favor of such an expectation. A
secular decline in the unearned income with increases in the working hours is evident from the data.
However, observed association of overall employment rate with working hours provides a rather weak
hint of a positive relationship (between them).

The remaining question to be answered, therefore, is validating the exclusion restrictions. Heckman and
Carneiro (2003) have suggested that child educational investment decisions are determined not by
current parental income but rather by family-fixed effects (such as parental education levels) which are
determinants of permanent income. Their conclusions are corroborated by empirical works from
Cameron and Heckman (1998) and Chevalier and Lanot (2002). As our model incorporates both paternal
and maternal educational background, maternal employment status, which can proxy for the long-run
economic solvency of a family, we can plausibly expect the unearned income to be left uncorrelated
with academic performance after controlling for such indicators of socio-economic background.
Furthermore, empirical evidence is provided by Dustmann et al. (1996), who pursued a similar
identification scheme to identify work hours in an exam-outcome equation. Presumably, parental
income is more closely associated with the permanent income of a family (and hence, GPA) than the
allowance allocated to the student (i.e., student’s unearned income) – which should be a very small
component of total permanent income and is one that should be more closely related to the decision
regarding part-time job-engagement. But again, the decision to provide a student with pecuniary
support (by a family) may be dependent upon his/her academic progress. On the exclusion of
employment rate (among classmates) from the structural model, we have sufficient empirical evidence
in favor of us (e.g., Hanushek 1972, 1992). Despite being possible candidates of containing unobserved
cluster effects, employment rate for the class, being an aggregate measure, can plausibly be expected to
have no explanatory power for the academic performance, ceteris paribus.

4.2. Summary Statistics

Since we have already discussed the summary statistics for some of the variables (included in the model)
in Section 3.3, here, we’ll focus on the remaining ones: 8

The sample average (standard deviation) of GPA is estimated to be 3.412 (0.305). Average time spent
working (in a paid job) has been observed to be around 4.4 hours (per week) with a sizable standard
deviation of 7.97; self-reported maximum (minimum) number of working hours is 50 (0). Furthermore,
about 27.5% of the students worked less than or equal to 10 hours a week (see Table 2), whereas the
share of students who reported a higher work intensity is only 14%.

Rate of employment in the class, one of the (natural) instruments, on average, is 41% - the highest
(lowest) being 85.7% (22.2%) corresponding to Department of Mass Communication and Journalism
(Banking). Only 8.6% of the (sampled) students did not receive any allowance from their family, whereas
each of the remaining three categories contains around 30% of the students. The percentage of

8
In contrast to Table 1, estimates from Table 3, however, are exclusively based on the subsample used for OLS
and 2SLS estimation. But the difference in the sample size is not large enough to warrant a re-discussion of the
summary statistics.
students, whose mother have been in an occupation after their birth, is merely 7%. One-fifth of our
surveyed students have sibling(s) who got admitted into DU (prior to their enrollment). Half of them
evaluated their health status as above-average, i.e., good or excellent. Average age has been observed
to be approximately 22.4 (years). About 78% of the respondents said that they would continue study
upon the completion of baccalaureate.

4.3. Regression Results

Table 4 presents the key findings from our multivariate analysis (of academic performance). Estimates
from both OLS and 2SLS models (with varied combination of instruments) are reported here. Column 1
shows the relevant OLS estimates. Column 2 reports the relevant (1 st and 2nd stage) estimates from the
model with both unearned income (categorized) and class employment rate used as instruments,
whereas column 3 (4) corresponds to the 2SLS IV model which instruments hours worked with unearned
income (class-level employment rate) only. Hours worked (per week), across all the models, enters in
level form.9 Furthermore, the discussion of Table 4 will, often, be supplemented by referring to Table 5
which reports the 1st stage estimates in detail. 10

The OLS estimate of the ceteris paribus impact of hours worked is negative but pretty small (only 0.003
in absolute value) and more importantly, statistically insignificant. Moving to the 2SLS model(s), the
instruments are jointly significant (in the 1 st stage hours equation) at the 5% level in the first two
specifications (see Tables 4 and 5). Class-level employment rate, however, has little explanatory power
for time spent working (the associated t statistic, in model 4, is significant at the 10% level only; also
note that it loses individual statistical significance when coupled with the unearned income categories in
model 2). The associated F statistic is highest when only the indicators of unearned income are used as
instruments as opposed to that utilizing both unearned income and employment rate (though the
Partial R2 is higher in the latter case). The sizable Partial R2s (despite the long list of explanatory
variables) in these two cases (namely, 0.221 and 0.253) are promising in the sense that we can expect,
as a result, lower standard errors in the 2 nd stage estimation.

Table 5 shows that all the categories of unearned income (with zero income as the baseline) are very
significant, both statistically and practically – the magnitude of effect being more/less equal across the
groups (see columns 2 and 3). 1 st stage estimations following 2SLS model 2 (3) suggest that on average,
the difference between any category with respect to the ‘no unearned income’ group ranges from 12 to
13 (13.5 to 14.5) hours, with negative signs to them. Put in plain words, after controlling for the other
determinants of GPA (which also are likely to influence labor-market productivity), each of the three
groups (included) is predicted to work about 12 (or 13 in the case of model 3) hours less (on average)
than that containing students with no financial backup from their family.

9
Ramsey’s RESET test has not provided any evidence whatsoever of functional form misspecification. In addition,
quadratic term (i.e., square of working hours) was added to all the specifications, but without any benefit: it
remained statistically insignificant, both individually and jointly (with the level term).
10
For the sake of lucidity and simplicity, we have refrained from reporting the coefficient estimates associated with
other covariates both in Tables 4 and 5; they, indeed, are readily available from the author(s).
Henceforth, we will be focusing on the parameter of interest, i.e., that associated with hours worked in
the 2nd stage regression. The 2SLS estimates, across all specifications, are substantially different from
that obtained by single-equation OLS estimation (as also confirmed - though rather weakly - by the
exogeneity test statistics in 2 out of 3 cases). Model 4 produces an estimate (namely, -0.020) that is, in
terms of magnitude, considerably differs from those yielded by the first two 2SLS specifications (-0.010
and -0.009, respectively), but is of limited importance due to having no statistical significance which,
perhaps, can be attributed to the lack of explanatory power in the instrument used (class-level rate of
employment) – as was referred to in the discussion above. The estimates from (2SLS) specifications 2
and 3 are quite similar in size and also statistically significant (at the 5% and 10% levels of significance,
respectively).11 Identifying students’ labor supply with both indicators of unearned income and class-
level employment rate results in an estimate which can be interpreted as follows: On average, an
additional hour of paid work, ceteris paribus, leads to a decline in GPA by 0.010 points. Practically
speaking, the estimated effect is fairly moderate considering the implication that working for another 5
(10) hours in a part-time paid job is predicted to cause an average fall in GPA equal to 0.05 (0.10) points
only. The corresponding coefficient (accompanying hours worked) in the third specification (that
excludes employment rate from the set of instruments) estimates this decline to be 0.045 (0.09) points.
Figures 2a & 2b explicitly demonstrate the moderate diminishing impact of hours worked on academic
performance (as proxied for by GPA). The insignificant overidentification statistics (for the first two 2SLS
specifications; see Table 4), as expected, also provide support in favor of excluding the instruments from
the GPA equation.

To further investigate the impact of part-time employment on academic triumphs, we directly asked (as
also was stated before) the employed students to comment on the (experienced) influence of such
engagement on their academic and social life. Table 6 reports the percentage(s) of students, pertaining
to different work-intensity groups, who felt that part-time employment had caused considerable or
great impairment to their educational or social status. Here, contrary to the results obtained via
regression analysis, higher working hours are not clearly associated with a higher degree of
dissatisfaction, when it comes to considering the interaction with academic life. But in the case of part-
time job’s interference with social life, we can observe an exponential rise in the share of discontent
students up to 25 hours of weekly work. Those who work beyond that (only 3.6% of respondents) are,
surprisingly, have not showed any discontentment – may be because they are fond of their job which
has enabled them to be willingly working so extensively.

5. Concluding Remarks

This paper sought to estimate the partial effect of part-time employment on academic outcome. We
found no evidence of non-linearity characterizing the association between these two variables.

11
Against one-sided alternatives (hypothesizing that the impact is negative), both of them are significant at the 5%
level of significance.
Identifying students’ labor supply with unearned income categories and class-level rate of employment
(which are jointly significant in the 1 st stage equation and have convincingly passed the over-
identification test), we found a negative association between work intensity and subsequent academic
performance – albeit the impact is modest enough. While class-level employment rate, alone, has
proved to be a rather weak determinant of part-time employment (thus, resulting in an insignificant and
divergent estimate), unearned income indicators are found to be capable of explaining a sizable
proportion of the observed variation in the weekly hours worked, ceteris paribus.

Our study, however, has certain notable limitations. IV regression models are typically estimated using
fairly large samples, so as to obtain a better degree of precision in the estimators (via smaller standard
errors). The sample we have worked with, in this regard, is considerably small – thus restricting us,
statistically speaking, to make strong and persuasive conclusions. Our population, also, is rather less
representative in the sense that it has only considered departments from only three faculties. Thus, the
results obtained, despite being significant, are not generalizable (to all DU students – which would
certainly be more desirable). We have not - mainly due to time constraint - checked the robustness of
2SLS estimates to exclusion restriction variations which has become a standard practice among the
applied econometricians in validating the choice of instruments. Another criticism follows from the
postulation that there may be heterogeneity in the impact of both the instruments on work-intensity
and work-intensity on academic achievement. In that case, our estimates are properly construed as the
local average treatment effect among students who responded to the instruments (Imbens and Angrist
1994).

Despite these drawbacks, this paper, being the first in Bangladesh in attempting to capture the causal
effect of part-time working on academic outcome, can serve as a useful point of departure rather than a
full-blown empirical analysis. Indeed, if provided with adequate financial and institutional support, we
will again take up the issue and further our investigation.

Bibliography
Arnaud, Chevalier, and Gauthier Lanot. "The Relative Effect of Family Characteristics and Financial
Situation on Educational Achievement." Education Economics.
Cameron, Stephen V, and James J. Heckman. "Life Cycle Schooling and Dynamic Selection Bias: Models
and Evidence for Five Cohorts of American Males." Journal of Political Economy.

Cassady, Peter. "Self-Reported GPA and SAT Scores: A Methodological Note." Practical Assessment,
Research & Evaluation.

D’Amico, Ronald. "Does Employment During High School Impair Academic Progress?" Sociology of
Education, 1984.

DeSimone, Jeff. "Academic Performance and Part-time Employment among High School Seniors." Topics
in Economic Analysis and Policy, 2006.

Dustmann, Christian, Najma Rajah, and Arthur van Soest. "Part-time Work,School Success and School
Leaving." Working Paper, Department of Economics, University College London.

Eckstein, Zvi, and Kenneth I. Wolpin. "Why Youths Drop Out of High School: The Impact of Preferences,
Opportunities, and Abilities." Econometrica.

Hanushek, E A. "Conceptual and empirical issues in the estimation of educational production function."
Journal of Human Resources, 1979.

Hanushek, E A. "Education and race: an analysis of the educational production process." 1992.

Heckman, James, and Pedro Carneiro. "Human Capital Policy." National Bureau of Economic Research
Working Paper 9495.

Imbens, Guido W, and Joshua D. Angrist. "Identification and Estimation of Local Average Treatment
Effects." Econometrica, 1994.

Jr., McNeal, and Ralph B. "Extracurricular Activities and High School Dropouts,” ." Sociology of Education,
68(1).

Lillydahl, Jane H. "Academic Achievement and Part-Time Employment of High School Students." Journal
of Economic Education.

Marsh, Herbert W.Subversion of Academic Goals?”. "Employment During High School: Character
Building or a ." Sociology of Education.

Oettinger, Gerald S. "Does High School Employment Affect High School Academic Performance?"
Industrial and Labor Relations Review, 53(1).

Quirk, Kimberly J., Timothy Z. Keith, and Jeffrey T. Quirk. "Employment During High School and Student
Achievement: Longitudinal Analysis of National Data." Journal of Educational Research,
September/October 2001.

Singh, Kusum. "Part-time Employment in High School and Its Effect on Academic Achievement." Journal
of Educational Research.
Staiger, Douglas, and James H. Stock. "Instrumental Variables Regression with Weak Instruments."
Econometrica, 65(3).

Steinberg, Laurence D., Ellen Greenberger, Laurie Garduque, and Sheila McAuliffe. "High School
Students in the Labor Force: Some Costs and Benefits to Schooling and Learning." Education Evaluation
and Policy Analysis.

Tyler, John H. "Using State Child Labor Laws to Identify the Effect of School- Year Work on High School
Achievement." Journal of Labor Economics.

Warren, John Robert, Paul C. LePore, and Robert D. Mare. "Employment During High School:
Consequences for Students’ Grades in Academic Courses." American Educational Research Journal.
Table 1: Part-time Employment Scenario in Subsamples

Engaged in a Part-time Job


Yes No Total
By - 57 81 138
(41.30) (58.70) (100.00)
14 38 52
Female (26.92) (73.08) (100.00)
[24.56] [46.91] [37.68]
Gender
43 43 86
Male (50.00) (50.00) (100.00)
[75.44] [53.09] [62.32]
26 29 55
Hall/Mess (47.27) (52.73) (100.00)
Mode of [45.61] [35.8] [39.86]
Accommodation 31 52 83
Private or Relative’s
(37.35) (62.65) (100.00)
residence
[54.39] [64.20] [60.14]
3 0 3
Christian (100.00) (0.00) (100.00)
[5.26] [0.00] [2.19]
1 12 13
Religious Affiliation Hindu (7.69) (92.31) (100.00)
[1.75] [15.00] [9.49]
53 69 122
Muslim (43.80) (56.20) (100.00)
[92.98] [85.00] [88.32]
18 29 47
No (38.30) (61.70) (100.00)
Father graduated [31.58] [35.80] [34.06]
from University 39 52 91
Yes (42.86) (57.14) (100.00)
[68.42] [64.20] [65.94]
41 53 94
No (43.62) (56.38) (100.00)
Mother graduated [71.93] [65.43] [68.12]
from University 16 28 44
Yes (36.36) (63.64) (100.00)
[28.07] [34.57] [31.88]
46 57 103
Single (44.66) (55.34) (100.00)
Relationship [80.70] [70.37] [74.64]
Status 11 24 35
Married or
(31.43) (68.57) (100.00)
in a relationship
[19.30] [29.63] [25.36]
46 60 106
No (43.40) (56.60) (100.00)
Sibling(s) studied [80.70] [77.92] [79.10]
in DU 11 21 32
Yes (39.29) (60.71) (100.00)
[19.30] [22.08] [20.90]
Note: Within parentheses (brackets), reported is relative frequency within its row (column) of each
cell – exclusive to the corresponding scheme of sub-sampling.
Table 2: Hours Worked and Mean (observed) - Unearned Income, Employment Rate & GPA

Mean
Hours Unearned
Fraction
Worked Freq. Income Rate of
of Sample GPA
(per week) (monthly, Employment
in 000’s)
0 81 58.70 3.72 0.38 3.44
1-5 16 12.32 2.72 0.46 3.45
6-10 21 15.22 3.39 0.39 3.31
11-15 4 2.90 2.75 0.53 3.33
16-20 6 4.35 2.67 0.50 3.55
21-25 4 2.90 2.00 0.51 3.25
>26 3 3.62 0.33 0.64 3.40
Note: N=135

Table 3: Summary Statistics for Endogenous and Exogenous Variables

Mean SD Min Max


Dependent Variable
GPA (2nd Stage) 3.412 0.305 2.500 3.950
Hours worked per week (1st Stage) 4.405 7.971 0.000 50.000
Exogenous Variables
Instruments
Rate of Employment (in the Class) 0.410 0.144 0.222 0.857
Unearned Income (weekly) Category:
Tk. 0 (omitted) 0.086 0.281 0.000 1.000
Tk. 240-625 0.320 0.468 0.000 1.000
Tk. 626-1000 0.305 0.462 0.000 1.000
Tk. 1000+ 0.289 0.455 0.000 1.000
Other Controls
Gender (Male=1, Female=0) 0.609 0.490 0.000 1.000
Accommodation (Private or Relative’s Residence=1,Hall/Mess=0) 0.602 0.492 0.000 1.000
Graduate Father (Yes=1, No=0) 0.672 0.471 0.000 1.000
Graduate Mother (Yes=1, No=0) 0.328 0.471 0.000 1.000
Working Mother (Yes=1, No=0) 0.070 0.257 0.000 1.000
Sibling(s) studied in DU (Yes=1, No=0) 0.203 0.404 0.000 1.000
In a Relationship/Married (Yes=1, No=0) 0.266 0.443 0.000 1.000
Health Status (Good/Excellent=1, Average/Poor=0) 0.500 0.502 0.000 1.000
Age (in years) 22.414 1.031 20.000 25.000
Will continue study after Graduation (Yes=1, No=0) 0.777 0.152 0.000 1.000

Note: N=128. Summary statistics are calculated from the subsample yielding the regression estimates in Table 4.
Table 4: Impact of Hours Worked on GPA (OLS and 2SLS Estimates)

Dependent variable = Hours worked


(1st stage)
(1) (2) (3) (4)
Model
OLS 2SLS 2SLS 2SLS
Instruments
Unearned income (categorized) - Yes Yes No
Rate of employment - Yes No Yes
Joint Instrument Statistics
Partial R2 - 0.253 0.221 0.087
Robust F-statistic (instruments) - 2.809** 3.399** 3.499*
Dependent variable = GPA
(2nd stage)
Hours worked (per week) -0.003 -0.010** -0.009* -0.020
(0.003) (0.005) (0.005) (0.014)
Overidentification test χ2 statistic - 3.045 2.972 -
Exogeneity test F statistic - 2.871* 1.360 2.677*
Note: N = 128 across all models. Robust standard errors are reported in parentheses. ** p<0.05, * p<0.1.
All the models include covariates representing gender, age, self-evaluated health status, paternal and
maternal education, maternal occupational status, current mode of accommodation, sibling’s association
with DU, relationship status and expected future investment in education.

Table 5: Coefficients of Instruments in 1st Stage Hours Equation

Dependent variable = Hours worked


(1st stage)
Model (2) (3) (4)
Instruments
-12.899*** -14.528*** -
Tk. 240-625
(3.996) (4.566) -
-12.029*** -13.481*** -
Tk. 626-1000
(4.138) (4.686) -
-13.042*** -14.061*** -
Tk. 1000+
(4.290) (4.743) -
11.248 - 17.605*
Rate of Employment (in the Class)
(7.522) - (9.412)
Joint Instrument Statistics
Partial R2 0.253 0.221 0.087
Robust F-statistic (instruments) 2.809** 3.399** 3.499*
Note: N = 128 across all models. Robust standard errors are reported in parentheses.
*** p<0.01, ** p<0.05, * p<0.1.
Table 6: Self-reported Impact of Part-time Employment on Academic and Social life (by Hours Worked)

Hours % of the employed students who feel part-time job as


Worked Frequency considerably detrimental to their -
(per week) Studies Social Life
1-5 17 47.06 17.65
6-10 21 52.38 19.05
11-15 4 25.00 25.00
16-20 6 33.33 33.33
21-25 4 50.00 75.00
>26 5 0.00 0.00
Mean - 42.11 22.81
Note: N=57

5% 2% 11 Tuition (Coaching Center)


%
4% Internship

Journalism

Reseach Assistance

79% Private Tuition

Figure 1: Relative Share of Different Types of Part-time Job


Predicted Values for GPA -- gpa

Predicted Values and 95% CI 3.5

2.5

2
0 20 40 60
Hours Worked (per week)

Figure 2a: Hours Worked and Predicted GPA


Notes: In all cases, plotted are predicted GPA
(instruments:
Predicted Values for GPA --Unearned
gpa Income and Employment Rate)
along with 95% confidence intervals.

4
Predicted Values and 95% CI

3.5

2.5

2
0 20 40 60
Hours Worked (per week)

Figure 2b: Hours Worked and Predicted GPA (instruments: Unearned Income)
Calculations are based on the estimates in Table 4. All other covariates, in each case, are taken at their sample
means.

You might also like