public schools. James, Suryahadi, and King 1996 find that, after controlling for exit- ing test scores, private elementary schools in Indonesia incur lower costs per pupil.
They conclude that private management is more efficient at achieving academic qual- ity. A more recent study goes further and claims that students schooled at private sec-
ular secondary schools enjoy a wage premium of 75 percent over their publicly schooled peers Bedi and Garg 2000.
Unfortunately, neither of the two studies that find that private schooling is superior use a plausibly exogenous source of variation to identify the private school effect.
James, Suryahadi, and King 1996 identify the effect of school type on expenditure using the religious and demographic composition of its subdistrict kecamatan,
meaning they assume that these subdistrict characteristics are orthogonal to schools’ per-pupil spending. Bedi and Garg’s finding that private schooling raises future adult
earnings is based on the identifying assumption that an adult’s province of birth can be excluded from the wage equation. This assumes that province of birth is only cor-
related with adult earnings through its effect on the probability of attending a public or private school. This assumption is violated if province of birth is correlated with
other unobserved factors determining wages, such as personal connections and human-capital accumulation not attributable to school type. In results below, we find
that province of junior high school attendance has a strong effect on school test scores, conditional on numerous covariates including school type, which suggests that
this assumption is flawed.
3
The implication of the Bedi and Garg study, namely that private secular senior sec- ondary schools provide a more valuable education than public senior secondary
schools, is puzzling for two reasons. First, as noted above, public schools in Indonesia look stronger on observed inputs and are widely perceived to be superior to secular
and Muslim private schools. Indeed, studies such as Bedi and Garg find that public and private Christian secondary schools attract observably stronger students. Second,
some public schools in urban areas screen applicants based on the score of their national test following elementary school. Students at these schools benefit from a
higher-scoring peer group. The positive effect of private administration would have to outweigh these peer effects, which appear to be important in other contexts Hoxby
2000; Somers, Mcewan, and Wilms 2003.
III. Model
This section presents a simple model in which two countervailing sources of selection bias may occur. Because public secondary schools appear to be
superior to private secondary schools in Indonesia, past work has focused on the potential for positive selection of the best students into public secondary schools
Bedi and Garg 2000. While the perceived higher quality of public schools is one source of bias, a second source derives from wealthier households’ preference for pri-
vate schools, ceteris paribus. The Journal of Human Resources
532
3. The finding in our study that selection bias is small with respect to the effect of school type on junior sec- ondary school exam scores does not imply that selection bias was not present in these two studies, which
used different dependent variables and a smaller set of control variables.
Consider a two-period model in which a parent selects a school in Period 1. The parent’s utility function is assumed to be concave and increasing in each period’s con-
sumption. The parent’s utility is also assumed to be an increasing function of nonaca- demic features of the school such as its safety or proximity. In the second period,
the child transfers an amount of income determined by the quality of their education, which is consumed by the parent. The parent also earns direct utility in proportion to
the amount of the child’s transfer, according to their personal preference for academic and professional achievement. We assume the parent’s utility from the child’s attend-
ing school j takes the following simple functional form:
4
1
j
. ln
ln U
Y P
Y T
T O
j j
j j
1 2
= -
+ +
+ +
d v
c a
a k
k Y
t
represents the household income in period t, which is exogenous with respect to school type. P
j
is the tuition of school j, and δ is the parent’s discount rate. T
j
repre- sents the student’s transfer to their parents at the end of Period 1, after attending
school j. The amount of the transfer enters the parent’s utility function directly if σ is nonzero, reflecting the household’s valuation of academic achievement, net of income
gains associated with that achievement. Finally, O
j
represents other, nonacademic characteristics of the school valued by parents, such as safety, religious affiliation,
convenience, and discipline. For simplicity, the student’s transfer in the second period school outcome is
assumed to be a positive and linear function of three factors: the quality of the school Q
j
, the parent’s income in Period 1 Y
1
, and the effort and ability of the student of the student E.
2 .
T Q
aY bE
j j
1
= +
+ We assume that parents who have higher income will provide a household envi-
ronment more conducive to learning and professional achievement, meaning that the parameter a is positive. We also assume that student effort and ability is positively
related to the degree to which parents directly value academic performance σ: 3
, .
E d
d where
= +
v ~
Finally, we assume for simplicity that σ, which measures a parent’s taste for edu- cation, is positive and that household income Y
t
is greater than tuition P
j
, for all schools j.
Because public schools are generally considered to be superior on average to pri- vate schools in Indonesia, we assume that Q
pub
Q
pri
. However, the set of schools which a child can attend is typically constrained by the location of the household, and
in some cases, by the child’s score on the national test for elementary school gradu- ates the Ebtanas test score. These constraints, along with other nonacademic char-
acteristics of the school O
j
, may lead parents to choose a private school even if higher-quality public schools exist. The child attends public school if the maximum
utility of the set of public schools they can attend, U
pub
, exceeds the maximum util- ity of the set of private schools they can attend U
pri
.
Newhouse and Beegle
533
4. The exact form of the utility function does not affect the results derived below, as long as utility is con- cave in consumption.
The difference in utilities between the best available public school and the best available private option is:
4 .
ln ln
ln ln
U Y
P Y
P Y
T Y
T T
T O
O
pub pri
pub pri
pub pri
pub pri
1 1
2 2
= -
- -
+ +
- +
+ -
+ -
d d
v c
D a
a a
a a
a k
k k
k k
k Under the assumption that P
pri
P
pub
, it can be shown that: 5
, U
0 and 2
2 v
D 6
. Y
U
1 5
2 2
D This model confirms the intuition that a parent that places higher value on educa-
tion is more likely to send their child to public schools, which are assumed to be of higher average quality. Meanwhile, holding other factors constant, wealthier house-
holds are more likely to send their children to private school. In the United States, where private schools are generally considered to be higher quality, wealthier, and
more motivated students tend to select into private schools. In Indonesia, however, where in general public schools are considered to be of higher quality, the two sources
of selection bias are of opposite sign. Thus, the direction and magnitude of bias in the OLS models is unknown, and depends on the relative strength of unmeasured
wealth and unmeasured motivation in determining students’ choice of school and their test score.
IV. Data
