Like father like son? Revisiting the role of parental education in estimating returns to education in China

• Published date: 01 February 2019
• DOI: http://doi.org/10.1111/rode.12538
• Date: 01 February 2019

REGULAR ARTICLE

Like father like son? Revisiting the role of parental

education in estimating returns to education in

China

Binlei Gong

Department of Agricultural Economics

and Management, School of Public

Affairs, China Academy for Rural

Development (CARD), Zhejiang

University, China

Correspondence

Binlei Gong, Department of Agricultural

Economics and Management, School of

Public Affairs, China Academy for Rural

Development (CARD), Room 1203

Qizhen Building, 866 Yuhangtang Road,

Zhejiang University, China 310058

Email: gongbinlei@zju.edu.cn

Funding information

The Fundamental Research Funds for the

Central Universities, Grant/Award

Number: 526013*172220182; The

Teaching and Research Development

Program for Faculties in Arts and

Humanities at Zhejiang University, Grant/

Award Number: 126000-541903/026; The

Research Program for Humanities and

Social Science Granted by Chinese

Ministry of Education, Grant/Award

Number: 18YJC790034

Abstract

Parental education has been used as an instrument in the

earnings equation to deal with the endogeneity problem

of education. Recently, however, many have found that

parental education can be a proxy for unobservable net-

working, which directly affects wages. This article revisits

the role of parental education in estimating returns to edu-

cation by introducing the “geographical isolation”theory.

For migrant workers who receive an education and move

around to work, parental education affects their education

but otherwise is unrelated to their wages, which makes

parental education a good instrument in the instrumental

variable approach. For local residents who stay in the

same place during childhood and adulthood, parental edu-

cation can directly affect their wages, and is better intro-

duced as a proxy variable using the control variable

method. This article identifies the heterogeneous effect of

parental education on wages for different Chinese cohorts

and contributes to the debate between the control variable

and instrumental variable methods in returns to education

studies. Moreover, the idea of geographic isolation can

help in the search for good instrumental variables for

migrant workers, which is valuable when studying the

large migrant population in developing countries such

as Mexico, China, India, Vietnam, and many African

countries.

DOI: 10.1111/rode.12538

Rev Dev Econ. 2019;23:275–292. wileyonlinelibrary.com/journal/rode © 2018 John Wiley & Sons Ltd

|

275

1

|

INTRODUCTION

Earnings regressions, at the center of labor economics for decades, are widely used to estimate the

returns to education (RTE), that is, the effect of schooling on personal earnings (Co, Gang, &

Yun, 2005; Wang & Cai, 2008; Dimova, Nordman, & Roubaud 2010). An essential problem that

most studies in the field have to face is the omitted variable bias due to some unobserved wage

determinants. Generally, there are two methods to correct this bias: the instrumental variable (IV)

approach and the control variable approach. For example, most economics students learn in Econo-

metrics 101 how to use parental education as instruments for a person's education to deal with the

endogeneity problem of earnings regression. However, it is still unclear whether parental education

should be used as instrumental or control variables (Li & Luo, 2004). The former approach

assumes parental education affects one's education but is otherwise unrelated to wages, while the

latter approach assumes parental education relates to networking that directly affects wages.

In a relatively fair job market, parental education may only affect one's wages through its effect

on one's own education, and thus the IV approach is preferred. Some scholars (e.g., Lemke & Ris-

chall, 2003) therefore use parental education as instruments to estimate RTE in Western countries.

The same method has also been applied in some RTE studies in China (Mishra & Smyth, 2013).

However, family networking (guanxi) is likely to directly affect a person's occupation and wages

through other channels in China (Bian, 1994; Liou, 2017). For example, succeeding to a parent's

position in state‐owned enterprises (jieban) was not uncommon until recently (Chen & Feng,

2011). The control variable approach that uses parental education as a proxy for this unobservable

family networking may be a better option to demonstrate China's job market and has been used in

some other RTE studies (Aakvik, Salvanes, & Vaage, 2003; Heckman & Li, 2004).

Due to the difficulty of arguing the true effects of parental education on wages in China, recent

studies have avoided the debate by using either institutional features as instruments (Song, 2012;

Xie & Mo, 2014; Campos, Ren, & Petrick, 2016) or twins data to automatically control for hetero-

geneity in parental education (Zhang, Liu, & Yung, 2007; Li, Liu, & Zhang, 2012; Rosenzweig &

Zhang, 2013). However, the former method faces a weak instrument problem and the latter method

requires highly specialized data. Moreover, the effect of parental education on wages has not yet

been clarified.

Inspired by the idea of geographical isolation in Graddy (2006), it might be easier to find good

instruments in RTE studies if there were isolation between where the education is undertaken and

where the work is undertaken. Thanks to China's urbanization, there is a large population of this

kind: migrant workers. Different from normal rural and urban Chinese residents who spend their

childhood and adulthood in the same places, most migrant workers live and study in a rural area

before they go to work in an urban area. Their parental education can do little to help them

directly in the job market. As a result, parental education can be a good instrument, leading to an

unbiased estimation of RTE.

On the other hand, the job market for local residents (urban and rural residents, separately) is

without significant geographical isolation. For these two job markets, the IV approach may not

work well since parental education that affects education in childhood is likely to affect wages

directly through other channels in adulthood such as social networking and connections. As a

result, socioeconomic mobility could be slower and income inequality could be more severe in

these job markets. In other words, more “like father, like son”examples are witnessed. The control

variable method that treats parental education as a proxy variable for networking is preferred in

this circumstance.

276

|

GONG