Heterogeneous Treatment Under Regression Discontinuity Design: Application to Female High School Enrolment
| Author | Yasi˙n Kürşat Önder,Mrittika Shamsuddin |
| Published date | 01 August 2019 |
| DOI | http://doi.org/10.1111/obes.12292 |
| Date | 01 August 2019 |
744
©2019 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd.
OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 81, 4 (2019) 0305–9049
doi: 10.1111/obes.12292
Heterogeneous Treatment Under Regression
Discontinuity Design: Application to Female High
School Enrolment*
Yas˙
inK¨
urs¸at ¨
Onder† and Mrittika Shamsuddin‡
†Cukurambar Mah. Muhsin Yazicioglu Cad. 1474 Sok. Ankara, Turkey (e-mail:
kursatonder@gmail.com)
‡Department of Economics, Dalhousie University, 6214 University Avenue Halifax, NS
B3H0A7, Canada (e-mail: Mrittika.Shamsuddin@dal.ca)
Abstract
This paper undertakes a regression discontinuity (RD) framework with multiple cutoffs
unlike typical RD setting where researchers normalize the score variable and pool all
the observations. This paper explores this heterogeneity in the effect of Islamic mayor on
female secular high schooling in Turkey using the multiple cutoff RD framework developed
in Cattaneo et al. (2016). The presence of many parties in the 1994 municipality election in
Turkeymeans that vote share of the strongest opponent party can vary substantially leading
to different cutoffs. Meyersson (2014) finds that Islamic mayors of 1994 promoted female
high schooling using a normalized and pooled RD framework, which averages the effect
across all the different cutoffs. We extend his work by segregating the effect of Islamic
mayor across different opponent party’s vote shares. Our results suggest that the positive
effect on female secular high school attainment was more pronounced in municipalities
where the strongest opponent party was secular than where the opponent was conservative.
This heterogeneity can be attributed to a policy change in 1999, which restricted religious
high school graduates from entering universities.
I. Introduction
Regression discontinuity (RD) design has become a commonly used tool in social and be-
havioural sciences; many variants of the tool have been proposed and applied (Magruder,
2013; Buser, 2015; Firpo, Ponczek and Sanfelice, 2015; Clark and Del Bono, 2016; Sun
and Zhao, 2016; Khanna and Zimmermann, 2017 and many others). RD design is consid-
ered to allow clean causal inferences in quasi-experimental studies. In a typical RD setting,
the cutoff value in the running variable (also known as the score variable) determining the
assignment of treatment is equal across units. For instance, in two-party races it is implicit
*The authors thank David Card, Beata Javorcik, Sebnem Kalemli-Ozcan, Murat Kirdar, Steven Ongena, Serdar
Ozkan, Jacob Shapiro, AhmetAli Taskin, Insan Tunaliand seminar participants at AALIMS Princeton workshop for
their valuable comments.
HeterogeneousTreatment Under RDD 745
that the winner needs to win at least 50% of the votes. However, this approach has lim-
itations for a wide set of RD applications where multiple cutoffs are present, such as in
electoral races with more than two candidates running for office. By multiple cutoffs, we
mean situations where the treatment is winning an election under plurality rules, which
makes score-determining treatment unknown and variable across units. In such scenarios,
it is possible that each of two candidates receives 33% of the votes and the third candidate
wins with a winning margin of 1%, with 34% of the vote share. To deal with this hetero-
geneity in the cutoff value of voting shares, studies have typically used a winning margin
that enables the researcher to pool all observations for each municipality regardless of the
other parties’ voteshare to draw inferences. Despite the popularity of this normalizing-and-
pooling approach, investigations of existing heterogeneity in the normalized and pooled
RD effects have been limited. Cattaneo et al. (2016) proposes using multiple cutoffs to
investigate observable heterogeneity. This paper uses the methodology proposed by Catta-
neo et al. (2016) to study heterogeneity in the effect of Islamic rule on female schooling in
Turkey.
Contrary to conventional wisdom that Islamic political rule has an adverse effect on
female rights, Meyersson (2014) provides evidence that the election of Islamic mayors in
Turkey in 1994 led to an increase in secular female education attainment for females aged
15–20 years. He uses a standard RD framework to analyse the year 2000 municipal-level
census data, showing that the positive effect on female education had persisted for up
to 17 years; with a reduction in the number of adolescent marriages and an increase in
female political participation. However, in 1994, 14 main parties and many independents
participated in the Turkish election with most receiving much <50% of the votes. Our
study uses multi-cutoff RD to determine whether there is evidence for heterogeneity in
the effects of Islamic mayors. With a pooling approach, there is a strong assumption that
decisions are made by the winning party but approved by the municipal council, whose
seats are shared among parties whose vote count exceeded the 10% threshold limit. Thus,
the political affiliations and ideologies of other members of the municipal council become
crucial, particularly if the winning party has received <50% of the vote. For instance, a
proposal to build religious premises is expected to obtain support from conservative parties
while leftist parties within a council may prioritize other proposals.Thus, investigating such
heterogeneity is crucial for understanding the causal mechanism for the increase in secular
female high schooling.
Our analysis proceeds in four stages. First, using standard RD analysis with optimal
bandwidth as in Meyersson (2014), we replicate the main idea of Meyersson (2014), that
is, election of Islamic mayors led to an increase in female high school completion rates
for the cohort aged 15–20 in the year 2000. Our estimates yield very similar coefficients
but with noisier standard errors.1Second, since 14 parties ran for office in 1994, we rerun
our analysis with a multi-cutoff RD framework as proposed by Cattaneo et al. (2016)
and document that there exists substantial heterogeneity when the strongest opponent’s
1Some of the municipalities had their names changed between 1994 and 2000.Thus, matching municipality names
is a tedious job and we managed to match 65 more observations than did Meyersson (2014). We believe that these
unmatched 65 observations are the source of the difference between the two studies: weuse the data of Meyersson
(2014), and obtain the same estimates.All our results using the data set in Meyersson (2014) can be found in Appendix
S1.
©2019 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd
To continue reading
Request your trial