Rates of Return to Four‐Year University Education: An Application of Regression Discontinuity Design

• Published date: 01 October 2018
• Date: 01 October 2018
• Author: Guochang Zhao,Xin Meng,Elliott Fan,Zhichao Wei
• DOI: http://doi.org/10.1111/sjoe.12251

Scand. J. of Economics 120(4), 1011–1042, 2018

DOI: 10.1111/sjoe.12251

Rates of Return to Four-Year University

Education: An Application of Regression

Discontinuity Design*

Elliott Fan

National TaiwanUniversity, Taipei 100, Taiwan

elliottfan@ntu.edu.tw

Xin Meng

Australian National University, Canberra ACT 0200, Australia

Xin.Meng@anu.edu.au

Zhichao Wei

Guotai Junan Securities Co., Ltd, 601211 Shanghai, China

zhichaowei@gmail.com

Guochang Zhao

Southwestern University of Finance and Economics, 610074 Chengdu, China

guochangzhao@swufe.edu.cn

Abstract

It is invariably difficult to estimate the rate of return to university education, because of the

problem of omitted variable bias. Using a regression discontinuity design, weestimate the effect

of a four-year university education on earnings, and we explorethe pathways through which the

effect operates. Our estimation exploits the centralized,score-based college/university admission

system in China, where the minimum scores required for university admission are externally

determined by the provincial governments.Our findings suggest that being eligible for four-year

university admission implies a sizable increase in earnings. The payoff can be attributed to the

prolonged length of education and improvementsin education quality, although the quality effects

cannot be precisely estimated.

Keywords: China; rate of return to education

JEL classification:I21; I23

*Wegratefully acknowledge comments and suggestions by seminar and conference participants.

Special thanks go to Andrew Leigh and Tsong-Min Wu. Support from the National Natural

Science Foundation of China (Grant No. 71603215) and the Australian Research Council (Grant

No. DP 180102875) is g reatly appreciated. The paper represents the authors’ views and does not

reflect those of the National Natural Science Foundation of China or the Australian Research

Council.

C

The editors of The Scandinavian Journal of Economics 2017.

1012 Rates of return to four-year university education

I. Introduction

In the past 30 years, almost all countries in the world have witnessed the

trend of increasing enrollment in higher education. Between 1980 and 2010,

the proportion of high-school graduates entering post-secondary education

increased in all the 113 countries that kept records. The increase was 41

percentage points for the European Union, 21 for East Asia, 23 for Latin

America, 40 for the United States, 27 for Japan, 22 for China, and 13

for India (World Bank, 2015). For developed and developing countries

alike, the micro- and macro-economic impacts of the expansion of higher

education, particularly the financial rewards for individuals who complete

higher education, have increasingly become an important policy issue.

Potentially, it can be difficult to estimate the economic returns to

higher education because of selection. Earlier studies have attempted

to account for the difficulty mainly using within-twin comparisons or

instrumental variables constructed from natural experiments.1Recently, a

growing number of studies have resorted to the regression discontinuity

(RD) design to address selection, producing a new wave of evidence.

Among these studies, some have focused on the returns to a flagship

university or a set of elite schools (see Hoekstra, 2009; Saavedra, 2009;

Canaan and Mouganie, 2015), or the returns to the field of university

education attended relative to the next-best alternative (Kirkeboen et al.,

2016). Relatively less is known about the rewards for students at the margin

of university enrollment, which is most relevant for shedding light on the

impacts of the global experience of university expansion. To date, only a

handful of studies have addressed the margin at the enrollment of university

(¨

Ockert, 2010; Hastings et al., 2013; Zimmerman, 2014).

Another deficiency in the existing work concerns the shortage of

evidence on the developing world, as the literature has predominantly

focused on high-income countries. To the best of our knowledge, Savvedra

(2009) and Hastings et al. (2013) are the only two exceptions that exploit

data from developing countries.

The main goal of this study is to estimate the rate of return to a four-

year university education in China. We utilize a unique feature of China’s

university admission system that mainly relies on the test scores from the

1For examples of twin studies, see Ashenfelter and Krueger (1994), Berhman et al. (1994),

Miller et al. (1995), Isacsson (1999), and Bingley et al. (2009). Examples of instrumental variable

studies are Angrist and Krueger (1991), Card (1995), Harmon and Walker(1995), Acemoglu and

Angrist (2001), Lochner and Moretti (2004), and Oreopoulos (2006). However,critics have been

skeptical about the returns to education estimated using these techniques. Critics of within-twin

variations include Bound and Solon (1999), Neumark (1999), Leigh and Ryan (2008), and Lee

and Lemieux (2010), and criticisms of natural experiments can be found in Bound et al. (1995)

and Oreopoulos (2006).

C

The editors of The Scandinavian Journal of Economics 2017.

E. Fan et al. 1013

National College Entrance Examination (NCEE), a centralized examination,

as a benchmark for university/college admission. The minimum required

score for university admission, set by the provincial governments, offers a

well-defined cut-off point for eligibility to apply for a seat at any four-year

university.

One estimation strategy that is commonly used in this setting relies

on a fuzzy RD design, using a dummy variable to indicate whether

an individual’s test score exceeds the threshold score as an instrumental

variable. However, the validity of the instrument has recently been

challenged by growing evidence suggesting that various observable and

unobservable factors exhibit abrupt changes at the cut-off point. For

example, Pop-Eleches and Urquiola (2013) applied a fuzzy RD design to

estimate the effects of gaining access to higher achieving secondary schools

in Romania. They found that eligibility in terms of test score cut-off not

only led to a higher probability of attending a better school, but also led

to behavioral responses from teachers, parents, and students.2Because of

all these likely effects on the students’ outcomes, the estimates obtained

from the conventional fuzzy RD design cannot be reliably interpreted as

the causal effect of the quality of high-school education.

Another example is Loyalka et al. (2012), who compared senior

university students whose NCEE scores were located on the two sides of

the cut-off score that allows admission to the first-tier universities in China.

They found that students just above the cut-off score were less likely to

study the most preferred major than those just below the cut-off score.

Although Loyalka et al. (2012) did not estimate the effect of the first-tier

university education on earnings, their finding implies that the estimated

effect would be biased, because the choice of whether to study a prefer red

major might affect a student’s effort, which in turn might affect their future

earnings.

¨

Ockert (2010) used the college admission cut-offs in 1982 in Sweden

to estimate the effects of college admission on earnings. He found that not

only the length of college education, but also the quality of the college

chosen and the timing of the college enrollment were all affected by the

college admission. Therefore, he adopted a reduced-form estimation strategy

to examine different channels through which the college admission would

affect earnings.

2They found the following: students who just passed the cut-off score considered themselves to

be inferior in ability to their peers and felt marginalized; parents of students who just passed the

cut-off scores provided less homework-relatedhelp to the student; and students who just passed

the cut-off score were more likely to be taught by teachers with better qualifications, thereby

reflecting the possibility that teachers sort according to their preference for higher achieving

students.

C

The editors of The Scandinavian Journal of Economics 2017.