Zhang 2021

Journal of Comparative Economics xxx (xxxx) xxx
Contents lists available at ScienceDirect
Journal of Comparative Economics

journal homepage: www.elsevier.com/locate/jce
Gender discrimination in China: Experimental evidence from the

job market for college graduates
Jian Zhang a, Songqing Jin b, Tao Li a, *, Haigang Wang a
a
School of Economics, Central University of Finance and Economics, China
b
Michigan State University, USA
A R T I C L E I N F O A B S T R A C T
JEL Classification: This paper examines employment-related gender discrimination during the initial stages of a
J16 hiring process. It specifically focuses on recent college graduates in China. By examining firms’
J71 responses to fictitious resumes with randomly generated information on gender and other key
J24
attributes of applicants (e.g., school reputation, student’s academic achievement, and leadership
Keywords: experiences), this study is able to separate the effect of gender on a student’s potential for getting
Gender discrimination
an on-site interview from the confounding effects of other factors. The findings reveal that, with
College graduates’ job market
all other factors remaining constant, female applicants, on average, are less likely to be invited by
Correspondence method
China hiring firms to on-site interviews as compared with their male counterparts. Furthermore, gender
discrimination in computer and mathematics, architecture and engineering, and sales occupations
appears to be driving the results of this study. The qualitative evidence based on interviews with
firm recruiters suggests that the findings of this study are generally consistent with the role
congruity theory of prejudice in psychology literature. While the finding that the quality of a job
candidate (academic achievement and leadership experience) does not reduce gender discrimi
nation offers some evidence to support a taste-based discrimination view, we do not have enough
evidence to support a statistical discrimination view.
1. Introduction
Gender discrimination is common in China (Fincher 2013; Steinfeld 2014), and it manifests itself in various forms (e.g.,
discriminatory hiring practices, wage discrepancies, limited promotion opportunities, unfair dismissals, early compulsory retirement
ages, and sexual harassment). Approximately 50% of Chinese women have indicated that they experience gender discrimination at
their workplaces regularly (Wang 2016), and urban Chinese women earn, on average, only 67% of the wages men earn (All-China
Women’s Federation 2011). Although China has enacted a series of laws to address these discriminatory practices (i.e., based on its
constitutional gender equality principle), these laws have generally been ineffective in combating discrimination against women
(Bulger 2000; Han 2001; Yang and Li 2009). Therefore, gender discrimination continues to be pervasive in Chinese society.
The authors thank Changqing Guo, Wei Han, Jing Ma, Rong Pang, Jingchao Wei, Dongfang Yan, Junbin Zhu for excellent research assistance.
The authors also thank Zhen Chen, Fei Di, Hongfeng Lei, Jian Li, Guangtao Yu and Xiaoni Zhang for facilitating a survey and interviews with firm
recruiters. Songqing Jin gratefully acknowledges financial support from Michigan State University AgBioResearch (MICL02608) and Tao Li is
grateful for support from the Beijing Outstanding Young Scientist Program (Grant No.: BJJWZYJH01201910034034).
* Corresponding author.
E-mail address: litao@cufe.edu.cn (T. Li).
https://doi.org/10.1016/j.jce.2021.01.003
Received 2 August 2019; Received in revised form 6 January 2021; Accepted 13 January 2021
0147-5967/© 2021 Association for Comparative Economic Studies. Published by Elsevier Inc. All rights reserved.
Please cite this article as: Jian Zhang, Journal of Comparative Economics, https://doi.org/10.1016/j.jce.2021.01.003
J. Zhang et al. Journal of Comparative Economics xxx (xxxx) xxx
Despite the existence of many studies on wage discrepancies between men and women in China (Meng and Miller 1995; (Meng,
1998); Gustafsson and Li 2000; Bishop et al., 2005; Demurger et al., 2007), only a few delve into the subject of gender discrimination in
the hiring process. Kuhn and Shen (2013) examined explicit gender discrimination in ads posted on an online Chinese job board. Their
results show that (i) only 11% of approximately 1000,000 job ads display a gender preference (while 89% do not) and (ii)
gender-targeted job ads favor women as often as men. This implies that gender discrimination mainly exists in the hiring process in
China in a subtle and implicit way (Steinfeld 2014). This explains why the majority of studies on employment-based gender
discrimination in developed and developing countries have predominately focused on implicit gender discrimination (Kuhn and Shen
2013).
In seeking to understand the reasons behind gender discrimination in the hiring process, existing literature generally offers two
alternative explanations: taste-based discrimination and statistical discrimination. The taste-based discrimination theory, pioneered
by Becker (1957), argues that employers have a distaste for hiring members of certain groups (e.g., females, blacks, or other mi
norities). Therefore, employers may refuse to hire certain groups of people because of this distaste, and if they do hire them, they are
likely to pay them lesser than other equally productive employees. In contrast, the statistical discrimination theory argues that em
ployers tend to have imperfect information on potential employees’ productivity (Phelps 1972; Arrow 1973; Aigner and Cain 1977).
They use group-specific membership to infer potential employees’ productivity. As such, when faced with two applicants having the
same unobservable productivity but belonging to two different groups, an employer will rationally favor the applicant belonging to the
group with a higher expected productivity, thereby resulting in discrimination.
This paper examines implicit gender discrimination (i.e., against females) in the hiring process involving job-seeking college
graduates in China. At this stage, hiring firms solicit job applicants through job fairs and/or online job boards before screening them for
on-site interviews. These two channels are the main avenues through which firms hire employees in China. More specifically, this study
explores the role of gender in screening and inviting applicants for an on-site job interview. To achieve this, it examines the different
responses of firms to male and female college graduate applicants and further discusses the possible mechanisms underlying gender
discrimination against females at the initial stage of hiring. To the best of the authors’ knowledge, this is the first study to examine
implicit gender discrimination in the hiring process in China.
To investigate the existence of implicit gender discrimination in the hiring process in China, the authors conducted a correspon
dence study that examined employers’ responses to randomized, fictitious resumes. The correspondence method was employed
because it overcomes both the omitted variable and self-selection biases plagued with the utilization of observational data (Neumark
et al., 1996; Bertrand and Mullainathan 2004; Petit 2007). The omitted variable bias arises if part of the measured difference in
outcome variables is due to unobserved differences between the two groups (unknown to researchers), while the self-selection bias
arises if there are different self-censorships in job applications between the two groups (e.g., female college graduates may choose to
only apply to prospective employers who are less discriminatory against women). The correspondence method is considered the most
rigorous approach to studying gender or racial discrimination in labor markets. Under this method, gender or racial discrimination is
assessed by comparing hiring firms’ response rates (whether or not a mock applicant is invited for an on-site interview) to fictitious
resumes, wherein applicants’ gender (or race) and other key attributes (such as school background, academic performance, etc.) are
randomly created. As the gender and key characteristics are randomly assigned, they are all independent from each other and are
uncorrelated with the hiring firms’ characteristics. Therefore, the effect of gender on an applicant’s likelihood of receiving an on-site
interview can be separated from the confounding effects of other factors. Despite this obvious advantage of the correspondence
approach in studying employment-based gender discrimination, the number of studies that have used this method, although growing,
is still small.
To conduct a correspondence study, we created identical, fictitious resumes in which gender and several other characteristics (i.e.,
commonly perceived as important in obtaining a job) were randomly assigned. These fictitious resumes were then submitted to
recruiting firms. As such, the omitted variable and selection biases were controlled. The correspondence approach has been employed
in a number of studies measuring discrimination by studying employers’ reactions to identical resumes with randomly assigned
gender, ethnicity, or race (e.g., Levinson 1975; Riach and Rich 1987; Neumark et al., 1996; Bertrand and Mullainathan 2004; Riach
and Rich 2006; Booth and Leigh 2010; Edelman et al., 2017; Chen 2020; Hou et al., 2020).
Our study yields several important findings. First, with all factors remaining constant, female (vs. male) applicants, on average,
receive fewer callbacks from hiring firms. More precisely, the probability of a female applicant obtaining an on-site interview is only
about two-thirds that of a male applicant. This means that a female applicant, on average, will have to submit 55% more applications to
receive the same number of callbacks. Second, a high level of academic achievement of a female applicant (e.g., excellence in academic
performance, student leadership, and strong English skills) does not narrow the gender disparity in hiring. This implies that hiring
firms do not compromise their gender preference even when faced with a high level of achievement of a female applicant. Third,
gender disparity appears to be heterogeneous across firm sizes, ownership types, industries, and occupations. Private firms and
relatively large firms (i.e., with more than 100 employees) tend to prefer male applicants; however, there is no statistically significant
gender disparity for other types of firm ownerships and relatively small firms (i.e., those with less than 100 employees). Gender
disparity is the most striking among industries such as the manufacturing, wholesale and retail trades, information technology, and to a
lesser degree, leasing and commercial services.
This study further explored the possible mechanisms underlying the gender discrimination results. The findings revealed that for
computer and mathematics, architecture and engineering, and sales occupations, females are, on average, less likely to receive call
backs, with all other factors remaining constant. These three occupations account for 43% of the total sample. As such, they appear to
be driving this study’s findings that females are discriminated against in receiving callbacks based on the full sample. Other main
occupations do not show such high levels of discrimination against females. To further reveal the underlying reasons for gender
2
discrimination, we conducted a qualitative analysis by interviewing 20 firm recruiters. The interview results revealed that recruiters
perceive the existence of incongruity between the requirements for these occupations and the stereotyped roles of women. As a result,
female applicants are less likely to be recruited for these occupations. These findings are in fact consistent with the role congruity
theory of prejudice in psychology literature (Eagly and Karau 2002; Eagly and Carli 2003; Koenig and Eagly 2014). While this study
offers some evidence to support a taste-based discrimination view, we do not have enough evidence to support a statistical discrim
ination view.1
Literature on gender discrimination tends to focus on developed countries and finds that gender and racial discriminations are
widespread in developed countries (in the U.S.: Levinson 1975; Neumark et al., 1996; in Australia: Riach and Rich 1987; Booth and
Leigh2010; in England: Riach and Rich 2006). However, even though gender discrimination in developing (vs. developed) countries is
more pronounced, very few studies, especially those that have adopted the correspondence method approach, have examined how
gender discrimination affects job applicants in developing countries. In this respect, this study makes an important contribution to the
gender discrimination literature, as it is the first study to use the correspondence method to ascertain gender discrimination in hiring in
the Chinese job market and also one of the very few such studies focusing on developing countries.2
Kuhn and Shen (2013) studied explicit gender discrimination in the hiring process in China. They find that although it is legal to
express gender preferences in job ads, the share of ads favoring men versus those favoring women is roughly equal. They further find
that as job skill requirements increase, the share of ads stipulating a preferred gender decreases. Kuhn and Shen also show that despite
firms’ idiosyncratic gender preferences for particular jobs, a firm’s gender preferences can be overridden by factors such as a thinner
labor market or a greater incentive to search broadly for the best candidate. However, our study differs from Kuhn and Shen’s (2013)
study in several ways. First, their study specifically explores explicit gender discrimination in the Chinese labor market, while our
study examines implicit gender discrimination during the early hiring stage. Our study finds that even in an environment where it is
legal to express gender preferences in hiring, female applicants are discriminated against for job positions that do not specify gender
preference and do not necessarily require gender-specific skills. Second, our study finds that there is a systematic implicit discrimi
nation against female applicants, while Kuhn and Shen (2013) do not find a systematic explicit discrimination against a particular
gender. Third, our study suggests possible mechanisms underlying the systematic implicit discrimination against female applicants,
while Kuhn and Shen’s study does not seek to explain explicit discrimination against a particular gender. Instead, they seek to un
derstand why the frequency of the explicit gender requirement appearing in job ads (female or male gender) may vary in contrary to
the job’s skill requirement. In summary, although both the studies explore gender discrimination in the labor market, the phenomena
the two studies seek to understand and explain differ.
This paper is organized as follows. Section 2 describes the experimental design. Section 3 presents the experimental results and
focuses on the role of gender in obtaining on-site interviews. Section 4 examines the mechanism driving the gender discrimination
results revealed herein. Section 5 concludes the paper.
2. Experimental design
2.1. Creating randomized fictitious resumes
In our designed experiment, each hiring firm receives one fictitious resume.3 To created fictitious resumes, we used a resume
template commonly used by college students, which is generally a page with information on personal characteristics, education and
academic performance, internship experience, etc. Specifically, the fictitious resumes comprise seven parts: personal information,
education, academic performance, student-leadership experiences, English skills, honors and awards, and internship experience. In
addition, as college student applicants typically include a one-inch small headshot photo on their resumes, we followed this tradition
1
We thank one of the anonymous reviewers for pointing this out.
2
There are few exceptions. Banerjee, Bertrand, Datta, and Mullainathan (2009) study the role of caste and religion in India’s software and
call-center jobs by sending fictitious resumes to hiring firms. Maurer-Fazio (2012) studies the role of ethnicity in China by investigating how Chinese
firms respond to fictitious Internet job board applications from ethnic minority and Han applicants. Hou, Liu, and Crabtree (2020) study anti-Muslim
bias in the Chinese Labor Market by sending fictitious resumes to job advertisements for accounting and administrative positions. Chen (2020)
examines the value of U.S. college education in global markets by sending fictitious online applications to business and computer science jobs in
China. Galarza and Yamada (2014) study the role of ethnicity and attractiveness in receiving callbacks using the correspondence method in the
context of Peru. We are not aware of any other employment-based gender discrimination studies using the correspondence method in the context of
developing countries.
3
An alternative design would be to send two fictitious resumes, one male and one female. In that case, we can estimate the within estimate by
controlling firm fixed effects. Many existing studies on gender, ethnicity, and social class discrimination in job applications or loan applications
adopted this strategy. However, this is not necessary as long as the applicant’s other characteristics and hiring firms’ characteristics are balanced
between those receiving male resumes and female resumes. For practical reasons, sending one fake resume is easier than sending two fake resumes
to the same firm. Our design was partially influenced by Krueger (1999) who examines the impact of class size on student academic performance by
taking advantage of the Tennessee STAR experiment. In the experiment, students are randomly assigned to one of three treatments: small classes,
regular classes with a part-time teacher’s aide, and regular classes with a fulltime teacher’s aide. Our experiment follows the same spirit of the
Tennessee STAR experiment design.
3
and also included a photo of the applicant in the fictitious resumes. To minimize bias due to the appearance of an applicant, we used a
photo of an average-looking male or female in each fictitious resume.4
For the fictitious resume that each hiring firm received, one of the two genders was randomly assigned. Similarly, for the other
components on the resumes (education, English skill, academic achievement, etc.), random assignment was performed in the same
fashion as for the gender. Specifically, we filled in the parts of the resume template, the impacts of which we are interested in seeing on
receiving a callback, with randomized realization. (1) In the personal information section, we randomly assigned a gender and
Communist Party membership status to each fictitious applicant. (2) In the education section, we randomly selected one university
from 15 universities in Beijing (i.e., as the applicant’s graduating university).5 (3) In the academic performance section, each applicant
was randomly positioned as either a “Top 5′′ student (in a class of 40 students) or an average student. (4) In the student-leadership
experience section, each applicant was randomly assigned to have or not have served as a student-body leader. (5) In the English
skill section, we randomly assigned each applicant with either College English Test Level-4 (CET-4) or College English Test Level-6
(CET-6) to measure how English skills would affect college students’ job searches.6 CET-4 is a required English proficiency level for
college students, while CET-6 is an English level that exceeds the requirement for college students and indicates an applicant’s strong
English skills. (6) We filled in the remaining parts of the template with fixed information. When responding to a recruitment ad, we
selected a college major that best matched the job description presented in the ad. To increase the callback rate, we also customized the
applicant’s internship experience to complement the major (i.e., identical for the same major but different across majors) and included
a few honors and awards as in the case of real situations.
To ensure that employers are unable to distinguish our fictitiously created resumes from real resumes, we had a few firm recruiters
assess the fictitious resumes during the resume design stage, and none of them could recognize that our resumes were fictitiously
created. Therefore, the resumes we created are representative of resumes used by college students.
2.2. Sample and the data
There are three channels through which a college graduate looks for a job. The most important and formal channel is a job fair. In
the city of Beijing, where our experiment was conducted, many universities and colleges host job fairs on their campuses for their
graduating students. During these fairs, hiring firms are invited and graduating students are encouraged to attend and submit their
resumes. Additionally, the Beijing College Student Job Placement Guidance Center (hereinafter, the Guidance Center), a government-
funded agency, holds job fairs weekly from September to May, the period during which college students are most actively looking for
jobs. The second channel is an online job board, where college students submit their resumes to introduce themselves to potential
employers. Finally, some college students rely on family and personal connections and associated referrals for obtaining job offers.
As job fairs and online job boards are the two most important channels through which college graduates send their resumes to
potential employers, this study focuses on these two channels. The sample firms selected were those with job openings from October
2013 to May 2014, the period during which most jobs for college graduates are open. For the job fair channel, we visited the career
services websites of major colleges in Beijing and the Guidance Center website weekly starting October 2013 to obtain information
about upcoming job fairs’ locations, time, the list of hiring companies, and job descriptions. For online job boards, we visited the four
major employment websites weekly to collect any recruitment ads listed on these websites specifically designed for college graduates.7
We also obtained information on the size, ownership, and industry of the firms.
2.3. Nature of the experiment
This study adopts a between-subject experimental design, in which the subjects of the experiment are assigned to different con
ditions, with each subject experiencing only one of the experimental conditions. Specifically, as described above, each hiring firm
received one fictitious resume where one of the two genders (male or female) was randomly assigned to the applicant. For the other
4
As we only submitted one resume to each firm, it was practically feasible for us to use the same photo for all the male resumes and the same
photo for all the female resumes.
5
We deliberately selected the 15 universities from three tiers of universities in Beijing, five from each tier. We keep the names of the 15 uni
versities anonymous. The three tiers are specified as follows. A university designated as a Project 985 and Project 211 university by the Ministry of
Education of China is categorized as a Tier 1 university, while a university designated only as a Project 211 university is categorized as a Tier 2
university. A university that is neither a Project 985 university nor a Project 211 one is considered a Tier 3 university. Project 211 is a project of
National Key Universities and Colleges launched in 1995 by the Ministry of Education to promote high-quality research in high-level universities.
Project 985 is a project that was first announced by the former Chinese President Jiang Zemin on May 4, 1998 with the intent to promote Chinese
universities to be world-class ones in the 21st century. Universities that are designated as Project 985 universities or a Project 211 ones have to meet
certain scientific and educational standards imposed by the Ministry of Education and only prestigious universities in China can be designated as
Project 985 or Project 211 universities. Project 985 universities are designated by the Ministry of Education from the list of the Project 211 uni
versities, and it is even harder for a university to be designated as a Project 985 university than as a Project 211 one. Project 985 and Project 211
universities receive large amounts of funding from both the national and local governments.
6
The College English Test (CET) is a national English-as-a-foreign-language test in China organized by the Ministry of Education to examine the
English proficiency of undergraduate students and postgraduate students.
7
The four main employment websites in China are http://www.yingjiesheng.com/, http://www.zhaopin.com, http://www.chinahr.com/, and
http://www.51job.com/.
4
Table 1
Description of randomly generated student characteristics.
Student variable Category Observation Percentage
1. Gender Male 571 50.18

Female 567 49.82
2. English skills CET-6 580 50.97
CET-4 558 49.03
3. Academic score ranking 5/40 572 50.26
20/40 566 49.74
4. Communist Party member Yes 573 50.35
No 565 49.65
5. Student leader Yes 565 49.65
No 573 50.35
6. University
Tier 1 Univ. 1 70 6.15
Univ. 2 79 6.94
Univ. 3 75 6.59
Univ. 4 78 6.85
Univ. 5 66 5.80
Tier 2 Univ. 6 80 7.03
Univ. 7 70 6.15
Univ. 8 69 6.06
Univ. 9 74 6.50
Univ. 10 76 6.68
Tier 3 Univ. 11 84 7.38
Univ. 12 78 6.85
Univ. 13 83 7.29
Univ. 14 85 7.47
Univ. 15 71 6.24
Total 1138 100
components on the resume, random assignment was done in the same fashion as that of gender.
It is important to note that this experimental design allows us to completely control for the omitted variable and self-selection
problems that often plague studies using observational data. The randomization of each component on the resume template was
conducted independently from each other. Therefore, the randomized components are uncorrelated with each other. Because of the
randomization, the group of firms that received resumes with the female gender was identical, on average, to the group of firms that
received resumes with the male gender. Moreover, because of the randomization, the resumes received by the first group of firms were,
on average, at par with those received by the second group of firms with regard to the other randomized components except for gender.
This is also the case for each of the other randomized student characteristics. Therefore, this study was able to overcome the omitted
variable and self-selection problems and obtain unbiased and consistent estimates of the impacts of gender and other student char
acteristics on the probability of receiving a call for an on-site interview.
2.4. Experimental procedure
For the job fair channel, we created a randomized fictitious resume for each of the recruitment ads prior to each of the upcoming job
fairs, following the randomization procedure described above. When a job fair was held, we sent our research assistants to the job fairs
to submit the resumes to the potential employers. In job fairs, potential employers sometime seek to interact with job applicants. To
minimize the possible bias that may have resulted from such interactions, the research assistants were instructed to drop the resumes
with no or minimum interaction with the potential employers. In case they were asked by potential employers for more intensive
interaction, the assistants were instructed to explain that they were submitting their friends’ resumes who were unable to attend the
job fair due to inconveniences.
For online job boards, we created randomized fictitious resumes for each of the recruitment ads listed on the four major
employment websites. Our research assistants then submitted the resumes to potential employers via these websites. As online job
boards, in many cases, require filling in detailed personal information for an identity check, the assistants were only able to send a
limited number of fictitious resumes to a subset of online job openings without such a requirement.
We measured the hiring firms’ responses to the randomized fictitious resumes based on whether the research assistants successfully
received invitations to on-site interviews (e.g., via phone or emails).
3. The effect of gender on invitations for on-site interviews
3.1. Testing for randomness and independence of the experiment
While using experimental data, it is important to check the quality of the randomization design. Based on this study’s data, the
randomization achieved balance, and the realized outcomes of the randomized components are as expected. First, as described above,
5
Table 2
Description of hiring firms’ characteristics.
Category Observation Percentage
Location Beijing 928 81.55

Not Beijing 210 18.45
Size (No. of employees) <=100 304 26.71
100 - 500 451 39.63
500 - 1000 179 15.73
>1000 204 17.93
Ownership State-owned 217 19.07
Private 757 66.52
Foreign 135 11.86
Government 29 2.55
Total 1138 100
this study adopts a between-subject experimental design, wherein the subjects of the experiment are assigned to different conditions,
with each subject experiencing only one experimental condition. Table 1 presents the actual realization of the randomized components
of the resumes. For example, for each fictitious resume, one of the two genders was randomly assigned to the gender component of the
resume. This means that approximately half of the created fictitious resumes should have a female gender, while the other half has a
male gender. Table 1 shows that the random assignment exercise was successful, because the realized outcomes for each randomized
component do not appear to deviate abnormally from the associated theoretical possibilities. In other words, it is expected that a high-
quality randomization would yield a highly balanced distribution of observations across categories for each randomized component.
Second, the randomized student characteristics are independent from the hiring firms’ characteristics (i.e., size, industry, and
ownership). The chi-square independence test shows that the dependence between student characteristics and hiring firms’ charac
teristics is statistically insignificant (Panel A, Appendix Table 1). All the correlation coefficients between student and firm charac
teristics are shown to be close to 0 (Panel B, Appendix Table 1). Third, the randomized student characteristics appear to be
uncorrelated with each other, with the correlation coefficients being close to 0 (Appendix Table 2). This is expected, as the
randomization of the student characteristics was conducted independently. Table 1 and Appendix Tables 1 and 2, taken together, show
that the randomization of the experiment achieved balance and was a success.
3.2. Description of the hiring firms
Table 2 presents a profile of the hiring firms. We successfully sent resumes to a total of 1138 hiring firms. The firms were cate
gorized into four sizes: small (≤100 employees), medium (101–500 employees), large (501–1000 employees), and very large (>1000
employees). Small- and medium-sized firms constituted most of the hiring firms (i.e., 67% overall and 27% and 40% for small- and
medium-sized firms, respectively) (Rows 3–4 and Column 2). Further, 67% of the hiring firms in our sample were private firms (Row 8
and Column 2).
Appendix Table 3 presents the distributions of hiring firms’ industries, preferred majors, and announced job positions. Approxi
mately 26% of the hiring firms are from the manufacturing industry; 20% are from the information-transfer, software, and information
technology services; 12% are from the leasing and commercial services; and 9% are from the wholesale and retail trades industry
(Appendix Table 3, Panel A, Rows 1–4 and Column 2). The four industries account for 67% of the total hiring firms.
Preferences of the hiring firms for different college majors varied (Appendix Table 3, Panel B). Of all the majors, the management
major seems to be the most wanted by hiring firms (43%), followed by the computer science major (24%). Within the management
major, the marketing major appears to be the most sought after major (33%), followed by the accounting major (16%), administrative
management major (13%), human resources management major (11%), and business administration major (8%) (Appendix Table 4).
We then classify the job positions announced by the hiring firms into different occupations as per the Standard Occupational
Classification Code.8 The classification of occupations reveals that computer and mathematical occupations were the most popular
occupations in our sample (20%), followed by business and financial operation occupations (19%) (Appendix Table 3 Panel C). The
sales occupations, architecture and engineering, and office and administrative support were 10%, 13%, and 11% of the sample,
respectively. These five categories put together account for 74% of the total sample.
3.3. Impact of gender on callbacks for on-site interviews
3.3.1. Descriptive results

The results of the differences in the callback rate across different groups reveal that on average, a female applicant is 7.61% less
8
We use an American version of the Standard Occupational Classification Codes. Originally, we attempted to use a Chinese version of the codes
but found out that it is not well structured, and in some cases, it does not clearly distinguish several categories. For example, sales related jobs can be
divided into several categories according to the Chinese version of the code. On the contrary, the American version of the code clearly presents
mutually exclusive categories and is accompanied by examples and explanations. It is practically easy to categorize the jobs according to the
American version of the code. In any event, the classification of job categorization of the American version is similar to the Chinese version.
6
Table 3
Comparison of callback rates (%) across different groups.
Female Male Diff.
Gender 13.93 21.54 − 7.61***

Control variables: Yes No Diff.
CET-6 18.28 17.20 1.07
Score ranked Top 5 17.13 18.37 − 1.24
Communist Party membership 18.50 16.99 1.51
Student Association Leader 17.52 17.98 − 0.45
Project 985 and 211 Universities 18.59 16.21 2.38
Note: ***, **, and * refer to 1%, 5%, and 10% statistical significance level, respectively. Numbers may not foot due to the rounding off. The panel of
control variables compares the callback rates between “Yes” and “No” for each of the control variables. The T-test is performed for the statistical
significance of the differences.
Table 4
Determinants of obtaining an on-site interview.
Variable (1) (2) (3) (4)
Female − 0.0761*** − 0.0772*** − 0.0757*** − 0.0693***

(0.0251) (0.0252) (0.0226) (0.0234)
CET-6 0.0114 0.0117 0.0131
(0.0197) (0.0193) (0.0215)
Score ranked Top 5 − 0.0134 − 0.0191 − 0.0228
(0.0240) (0.0239) (0.0234)
Communist Party membership 0.0238 (0.0206) 0.0320 (0.0212) 0.0340 (0.0226)
Student Association Leader − 0.00520 (0.0174) 0.00300 (0.0183) 0.0114 (0.0195)
University dummies Yes Yes Yes
Firm size dummies Yes Yes
Firm ownership dummies Yes Yes
Firm industry dummies Yes Yes
On-site resume submission (1 = yes) Yes
College major dummies Yes
Firm recruit timing dummies Yes
Job fair dummies Yes
Obs. 1138 1138 1138 1138
Note: Robust standard errors clustered at the job fair level are reported in parentheses. ***, **, and * refer to 1%, 5%, and 10% statistical significance
level, respectively.
likely to receive a callback than a male applicant (Table 3, Row 1). The other student characteristics do not appear to be statistically
significantly correlated with the callback rate, although some characteristics, such as CET-6, Communist Party membership, and
university status, seem to slightly enhance the possibility of an applicant receiving a callback (Table 3).
When examining the callback rates across different firm types, the results reveal that these rates are quite different across different
industries, with the finance sector having the highest callback rate (28%) and the manufacturing sector having the lowest (10%)
(Appendix Table 5, Row 1). The differences in the callback rates across different industries may signify the fact that the demand and job
position requirements by industries vary by nature, with some industries being more selective than others. On the contrary, the
callback rates do not appear to be quite different across different firm sizes and ownerships (Appendix Table 5, Row 2).
3.3.2. Regression results
Main results. To examine the impact of gender on invitations for on-site interviews, we estimate an econometric model that takes the
following regression form:
∑
K ∑
L
(1)
′ ′
Yij = α + βFemalei + Xik γk + Zjl θl + εij
k=1 l=1
In the equation, the dependent variable, Yij, represents whether fictitious applicant i received a callback from firm j for an on-site
interview after the fictitious resume for i was sent to firm j. The gender variable, Femalei , is the gender variable for the fictitious
applicant i, and it is equal to 1 if the resume is associated with the female gender and 0 if otherwise. The coefficient on the female
variable, β, measures the difference between the probability of female and male applicants obtaining an on-site interview, with all
other factors remaining constant. The vectors of resume variables, Xik , and firm variables, Zjl , control for observable applicant and firm
characteristics, respectively. Any unobservable factor that affects the probability of obtaining an on-site interview is captured in εij . As
it is possible that certain recruiting behaviors of the hiring firms are correlated within the same job fairs, the error terms, εij , may be
7
Table 5
Determinants of obtaining an on-site interview by firm size and ownership.
Variable Firm size Firm ownership
<=100 100–500 500–1000 >1000 State-owned Private Foreign
Female − 0.0317 − 0.0798** − 0.115* − 0.0895* − 0.0617 − 0.0728** − 0.0999

(0.0381) (0.0384) (0.0591) (0.0492) (0.0526) (0.0324) (0.0651)
CET-6 0.0379 0.0604 − 0.106 − 0.00655 0.0402 − 0.00528 0.0873
(0.0381) (0.0422) (0.0677) (0.0531) (0.0460) (0.0281) (0.0560)
Score ranked Top 5 0.0521 − 0.0475 0.0340 − 0.0658 − 0.0506 0.00623 − 0.0738
(0.0502) (0.0403) (0.0616) (0.0516) (0.0484) (0.0319) (0.0582)
Communist Party − 0.0395 0.0573 0.139** − 0.0483 0.0601 0.0348 − 0.102*
membership (0.0384) (0.0378) (0.0557) (0.0585) (0.0493) (0.0286) (0.0588)
Student Association − 0.0551 − 0.0263 − 0.0128 0.0770 0.0313 − 0.0173 − 0.0237
Leader (0.0431) (0.0376) (0.0506) (0.0581) (0.0496) (0.0254) (0.0561)
University dummies Yes Yes Yes Yes Yes Yes Yes
Obs. 304 451 179 204 217 757 135
correlated within the same job fairs. Therefore, cluster-corrected standard errors at the job fair level are presented throughout. These
standard errors are robust to arbitrary forms of error correlation within a job fair. As a result of the randomization of our experiment,
the gender variable, Femalei , (as well as the other resume variables) is uncorrelated with the error term, εij . Therefore, a simple or
dinary least squares estimation (OLS) of the linear regression model, as defined in (1), would yield a consistent estimate of the impact
of gender on the probability of obtaining an on-site interview.
Our econometric results show that, with all other factors constant, women’s applications tend to receive far fewer callbacks from
hiring firms (Table 4). First, the regression results show that the coefficient associated with the female variable is negative and sta
tistically significant at the 1% level (Row 1 and Column 1). The OLS result indicates that women’s resumes, on average, are 7.6% less
likely to receive a callback from hiring firms. Given that 21.5% of the men’s resumes received callbacks from the hiring firms, the
probability of a female applicant obtaining an on-site interview was only about 65% of the probability of a male applicant (i.e., (21.5 −
7.6)/21.5 = 65%). This means that female applicants, on average, will have to submit 55% more applications to receive the same
number of callbacks as men. Second, the addition of other student characteristics to the regression barely changes the coefficient of the
female variable (Table 4, Row 1 and Column 2). These results are expected as the female variable is uncorrelated with other student
characteristics. Third, we further controlled for the firm characteristics and a rich set of variables (e.g., how a resume was submitted,
recruiting time, job fair dummies, and in-demand college major dummies) (Columns 3–4). The results show that the coefficient
associated with the female variable is highly robust to the inclusion of those control variables, and both the magnitude and level of
significance hardly change (Row 1).9,10
Heterogeneities across firm sizes, ownership types, and industries. The results revealed that female applicants are significantly less likely
to receive callbacks from hiring firms for on-site interviews. Although these results are important, it is interesting to dig into the
9
Kuhn and Shen (2013) find that the height and looks of a job applicant are important in landing a job in certain industries; however, we believe
that the omitted information on height and looks in our fictitious resumes do not bias our estimation results. In practice, college student applicants
typically do not include height and looks on their resumes except for a one-inch small headshot photo, which is often digitally altered with
image-editing software to look better. To follow this tradition, as we discuss in the experiment design section, we too included a one-inch small
headshot photo of an average looking male or female in each application, depending on the randomly generated gender of the resume. Given that all
the information on the resumes (gender, Communist Party membership status, contact information, education, academic performance, leadership,
English skills, etc.) is not correlated with the omitted information, information on height and looks will not bias our estimation results. Concerns
may arise if certain employers use job fairs as an avenue to privately collect data on applicants’ height and looks. However, this is unlikely to be a
concern in our case, given the reasons explained earlier regarding how resumes are submitted through the job fair channel. Our assistants submitted
the resumes to recruiters without much personal interactions. Nevertheless, in our regression analyses, we included a dummy variable to distinguish
the two resume submission channels to control for any differences between the two channels that may affect the possibility of receiving a callback.
10
We also randomly surveyed approximately 10,000 people on the streets in the main business districts of Beijing to examine the attitude of
regular people toward gender at the workplace. Our results show that the discriminatory attitude toward women appears to be pervasive. First,
approximately 39% of the respondents displayed a tendency of favoring male applicants. This means that the probability of a female applicant
obtaining an interview is only 39% (vs. 61% for an otherwise identical male applicant). Thus, the probability of obtaining interviews for female
applicants is approximately 64% that of male applicants (i.e., 39/61 = 64%). This result is highly consistent with our experimental results, where we
find that the probability of obtaining on-site interviews for female applicants is approximately 65% of the probability for otherwise identical male
applicants. Second, the discriminatory attitude toward women is common, regardless of a person’s gender, marital status, age, or education degree.
Indeed, 30%–45% of the respondents (i.e., male and female, single and married, young and old, and lowly and highly educated) tend to favor male
candidates over female candidates. It is also interesting to note that as many as 33% of the female respondents would choose a male candidate over a
female candidate. Finally, regardless of the geographic locations where people grew up and are from, discriminatory attitude toward women appears
to be common nation-wide.
8
Table 6
Determinants of obtaining an on-site interview by industry.
Variable Manufacturing Wholesale and Information Finance Leasing and Scientific research and
retail trades technology commercial services polytechnic services
Female − 0.0552* − 0.163* − 0.106* − 0.0736 − 0.0891 − 0.0312

(0.0302) (0.0928) (0.0597) (0.118) (0.0598) (0.0924)
CET-6 0.0140 − 0.0138 0.0945 0.0364 0.0551 − 0.0723
(0.0336) (0.0979) (0.0585) (0.124) (0.0714) (0.0799)
Score ranked Top 5 − 0.0372 0.0354 − 0.0529 − 0.0514 0.106 − 0.0646
(0.0324) (0.0724) (0.0490) (0.119) (0.0721) (0.0798)
Communist Party − 0.0155 − 0.0193 (0.0824) 0.00545 (0.0482) − 0.152 0.0746 (0.0743) 0.0611 (0.0776)
membership (0.0320) (0.120)
Student Association 0.0265 (0.0337) 0.0152 (0.100) 0.0127 (0.0522) − 0.149 − 0.0986 (0.0773) − 0.0711 (0.0759)
Leader (0.133)
University dummies Yes Yes Yes Yes Yes Yes
Obs. 299 106 223 72 133 76
Table 7
Determinants of obtaining an on-site interview by job positions.
Variable Computer and Architecture and Sales Business and financial Office and administrative
mathematics engineering operations support
Female − 0.0748* − 0.131** − 0.149* − 0.00278 0.0279

(0.0418) (0.0593) (0.0911) (0.0569) (0.0977)
CET-6 0.0202 0.0456 − 0.0830 0.00353 − 0.101
(0.0422) (0.0534) (0.0888) (0.0466) (0.0913)
Score ranked Top 5 − 0.0449 − 0.0861 − 0.0484 0.00409 0.0678
(0.0380) (0.0564) (0.0785) (0.0563) (0.0977)
Communist Party 0.0128 (0.0456) 0.105 (0.0645) − 0.114 − 0.0270 (0.0508) 0.000348 (0.0788)
membership (0.0838)
Student Association 0.0467 (0.0535) − 0.0557 (0.0455) − 0.134 0.0297 (0.0470) 0.00758 (0.0874)
Leader (0.0945)
University dummies Yes Yes Yes Yes Yes
Obs. 233 152 111 219 128
analysis further by examining whether and the degree to which this effect varies across firm sizes, ownership types, and industries. The
results show that the impact of gender on the possibility of obtaining an on-site interview appears to be heterogeneous across firm
sizes, ownership types, and industries. First, firms with over 100 employees tend to prefer male applicants, while there is no statis
tically significant gender disparity among firms with lesser than 100 employees (Table 5, Row 1 and Columns 1–4). This is probably
due to the fact that it is difficult for small firms to attract qualified candidates in the labor market; thus, they are less likely to screen
applicants based on gender. Second, in the context of private firms, female applicants are much less likely to obtain on-site interviews
(Row 1 and Column 6). Third, gender disparity appears to be the most striking in the manufacturing, wholesale and retail trades, and
information technology industries and to a lesser degree in the leasing and commercial services industry (Table 6, Row 1).
Albeit much less consistent than the female variable, a few key student characteristics also appear to influence applicants’ like
lihood of obtaining on-site interviews. For example, in the context of foreign firms, the coefficient estimate associated with the
Communist Party membership dummy variable is statistically significant and negative (Table 5, Row 4 and Column 7), which is
consistent with our expectations. In terms of the magnitude of the effect, the Communist Party membership would decrease a student’s
likelihood of receiving a callback from a foreign hiring firm by 10.2%. In addition, the findings reveal that the probability of obtaining
an on-site interview varies across different universities. However, a clear pattern across different tiers of universities cannot be
recognized.
Indeed, it would be interesting to ascertain whether any characteristics associated with female applicants (such as excellence in
academic performance, student leadership, and strong English skills) can mitigate the female disadvantage. For example, due to gender
disparities, female applicants may strive for higher academic achievements to enable themselves to compete with male applicants (i.e.,
on an equal footing) in the labor market. We interacted the female dummy variable with the other student-characteristic dummy
variables. The results show that the interaction terms are not statistically significant in all of the regressions (Appendix Table 6). The
coefficient of the female variable is still statistically significant in all the regressions and the magnitude of the coefficient barely
9
changes (vs. those without the interaction terms). These results indicate that an applicant’s high level of academic achievement (e.g.,
excellence in academic performance, student leadership, and strong English skills) does not help narrow the gender disparity in hiring.
This suggests that hiring firms do not compromise on their gender preference even in the context of female applicants’ high academic
achievement. This is significant because it indicates that no matter how strong a female applicant is academically, she is much less
likely to obtain an on-site interview than a male applicant. This finding tends to suggest that the lower callback rates for female
students (vs. male students) is driven by the recruiters’ taste-based discrimination against female.
Heterogeneity across occupations. Past studies have shown that whether female applicants are discriminated against in receiving
callbacks for on-site interviews or not may largely depend on the types of occupations (Riach and Rich 1987; Riach and Rich 2006;
Booth and Leigh 2010; Carlsson 2011). The considerable variation of occupation types in our firm sample allows us to explore the
heterogeneity in gender discrimination across different occupations. We specifically examine whether the gender discrepancy in
receiving callbacks varies across occupations, with a special focus on the five most important occupation categories of the sample. The
results suggest considerable heterogeneity across occupations (Table 7). For occupations such as sales, computer and mathematics, and
architecture and engineering, the coefficients of female applicants are statistically significant and negative, which means that females
are, on average, less likely to receive callbacks, with all other factors remaining constant (Table 7, Columns 1, 2, and 3). On the
contrary, for business and financial operations and office and administrative support occupations, the probability of an average female
applicant receiving a callback is not statistically different from that of an average male applicant. As described above, the three
categories with the statistically significant coefficient of the female dummy (sales, computer and mathematics, and architecture and
engineering occupations) account for 43% of the total sample. As such, they appear to be driving the results that females are
discriminated against in receiving callbacks.
Heterogeneity across channels of job seeking: Job fairs vs. online resume submissions. We further examined the heterogeneities of gender
differences in receiving callbacks for interviews between the two channels of job seeking: job fairs and online job boards. First, based
on the collected data, about 80% of the resumes were sent out through job fairs, while 20% were sent through online job boards.
Second, the results show that firms using these two channels appear to be different (Appendix Table 7). For instance, firms using the job
fair channel are mainly in Beijing, while about 50% of the firms using online job boards are not based in Beijing. In addition, firms
using online job boards appear to be bigger than those using the job fair channel. This is not too surprising given that hiring firms using
job fairs are mainly local firms and those using online job boards include both local and non-local firms. Third, in terms of ownership
and industry, firms display a more or less similar distribution in both the channels. Finally, the results based on the job fair sample
show that gender discrimination is statistically significant, and the magnitude of discrimination is similar to that based on the whole
sample (Appendix Table 8). However, the results based on the online job board sample are statistically insignificant with or without the
control variables (Appendix Table 9). This may be attributed to the small sample size of jobs from the online job board channel.
Comparison of results to those of other studies. Compared to studies on racial discrimination, studies on gender discrimination are fewer,
and discrimination against women at the callback stage is much less apparent in general (Bertrand et al., 2017). Carlsson (2011) found
that in Sweden, women are 3% more likely to be invited for job interviews than men. In addition, there is no statistically significant
difference in response rates between genders in male-dominated occupations, and females are 4% more likely to be invited for in
terviews in female-dominated and mixed-gender occupations (Carlsson 2011). Riach and Rich (2006) found that in the UK, males are
5.2% more likely to receive a favorable response than females in the occupation of engineering, while they are 3.24%, 9.52%, and
9.23% less likely to do so than females in the accountant, secretary, and computer analyst and programmers occupations, respectively.
Petit (2007) found that in France, for high-skilled administrative jobs, females are 20% less likely to receive an interview call than their
male counterparts when the applicants are aged 25, but there is no significant discrimination against female applicants aged 37. Riach
and Rich (1987) found that in Australia, females are 6.8% and 7.9% less likely to be invited for an interview in the occupations of
gardeners and analyst programmers, respectively. Booth and Leigh (2010) found that in Australia, for entry-level jobs, females are 7%
more likely to receive a callback than males in female-dominated occupations, such as data entry jobs (14%) and waitstaff (10%).
A quick review of the limited literature on gender differences in receiving callbacks for on-site interviews reveals the existence of a
considerable variation across different countries and occupations or industries within a country. For example, our overall finding
(7.6%) is comparable to Riach and Rich (1987) using Australian data (6.8%–7.9%) but sharply contrasts the findings of Booth and
Leigh (2010), who also used data from Australia. The female discrimination that we identified (7.6% overall, 7% for computer and
mathematics occupations, and 13% for architecture and engineering occupations) is also much smaller than the 20% identified by the
French study using data for highly-skilled administrative jobs (Petit 2007) but much larger than the 3% in the Sweden study (Carlsson
2011). Although it may be difficult to indicate the exact explanations for the differences in results, these differences may be attributed
to a number of factors, including cultural differences, regulatory environments, labor market conditions (e.g., Kuhn and Shen 2013),
and the composition of specific study samples (i.e., specific types/mix of occupations and industries).
4. The mechanism of gender discrimination
4.1. Interview findings from firm recruiters
In the econometrics analysis, we have consistently shown the existence of significant overall gender discrimination in receiving
10
callbacks for job interviews, and it can be seen that the results are mainly driven by certain occupations (i.e., sale, computer and
mathematics, and architecture and engineering occupations). One may then ask “Why is there gender discrimination against females in
these occupations, while there is not much discrimination in occupations such as business and financial operations and office and
administrative support?” To answer this question, we interviewed 20 firm recruiters from different industries (6 from information
technology, 6 from finance, 3 from education, 2 from manufacturing, 1 from wholesale and retail trades, 1 from construction, and 1
from real estate). Out of the 20 recruiters, 6 were male and 14 were female. The interviews were conducted one-on-one over the phone
in a Q&A manner and lasted approximately 30 minutes. At the start of the interviews, the recruiters were notified that their names and
their firms’ names are unknown to the authors.11 We first asked the recruiters whether and how female and male applicants are
considered differently while recruiting college graduates. We further asked them about the rationales underlying their answers and a
few other relevant questions. We asked each recruiter to answer this question for different occupations. The main interview findings
are summarized below.
For computer, mathematics, architecture and engineering occupations, 14 recruiters (70%) preferred male applicants, while the
remaining 6 recruiters had no particular gender preference. As for the reasons,12 out of the 14 recruiters said that these occupations
are generally quite demanding, and many of these positions are customer oriented, serving other departments of the firms or external
customers. Because of the nature of these jobs, employees constantly work overtime and during off-office hours (sometimes even on
weekends), and they travel a lot. Therefore, recruiters feel that male applicants are better suited to meet these demands than female
applicants, whom they perceive not to possess the toughness and stamina required to effectively perform in these types of jobs. In
addition, two recruiters said that they preferred male applicants because they believe that they are better at math and logical thinking,
which we don’t think should be generalized though.
For the sales occupations, eight recruiters said they have no experience of recruiting for the sales positions and their companies had
no sale positions. Of the remaining 12 recruiters, 5 (42%) indicated that they prefer male applicants for similar reasons as those stated
under computer, mathematics, architecture, and engineering occupations, i.e., these positions are customer oriented and constantly
require employees to work overtime and, in many cases, to travel and socialize with customers. Therefore, male applicants are
considered better suited to meet these demands. When asked why they thought male applicants were more able to meet their job
demands, they gave the same reasons as those mentioned under computer, mathematics, architecture, and engineering occupations:
female applicants are perceived not to possess the toughness and stamina required to meet the needs of these highly demanding jobs.12
The remaining seven recruiters did not have any particular preference toward gender.
For the occupations of business and financial operations, as well as office and administrative support, three recruiters said that they
do not recruit from outside of the firms for these positions. Out of the remaining 17 recruiters, 7 indicated that they had no particular
gender preference, while 3 answered that they preferred female applicants. On the other contrary, the remaining seven recruiters said
that they preferred male applicants because their firms’ female employees outnumber male employees and, therefore, they want to
balance the gender ratio. It is also important to note that in the context of our randomized experiment, such effects are not embodied in
our estimates simply because the nature of our randomized experiment literally means that the gender composition of a firm has been
controlled for.
4.2. Discussion on the interview findings
The responses from the recruiters we interviewed indeed offer a reasonable explanation for the implicit gender discrimination
identified in our study. The findings from the interviews are generally consistent with the role congruity theory of prejudice in psy
chology literature (Eagly and Karau 2002; Eagly and Carli 2003; Koenig and Eagly 2014). According to this theory, a potential for
prejudice exists when social perceivers hold a certain stereotype about a particular social group that is incongruent with the attributes
that are thought to be required for success in certain classes of social roles. This incongruity leads to two forms of prejudices: (a)
perceiving women less favorably than men as potential occupants of certain roles and (b) when fulfilling the role, evaluating women’s
behaviors less favorably than men’s. Consequently, this theory predicts that such prejudice prevents women from accessing certain
roles, such as leadership roles, and occupations, and it also makes it difficult for women to succeed in these roles.
The findings of our study are consistent with the role congruity theory of prejudice. For example, the occupations of computer and
mathematics, architecture and engineering, and sales are perceived by firm recruiters as more challenging and demanding occupations
(long working hours, overtime work, and frequent travels), and females are perceived by these recruiters not to possess the toughness
and stamina required to meet the challenges of these occupations. Therefore, recruiters perceive the existence of incongruity between
the requirements for these occupations and the stereotyped roles of women. According to the role congruity theory, incongruity leads
to prejudice, which in turn prevents women from being recruited for these occupations. The regression results of our study show that
11
In fact, we did not know their names and firms’ names. We obtained the recruiters’ information through universities’ career service centers and
personal referrals, which did not reveal to us the recruiters’ name and firm names. Before we conducted the interview, we obtained the consent of
the recruiters for the interviews.
12
Unfortunately, we do not have the statistical figures to show how demanding the occupations of computer and mathematics, architecture and
engineering, and sales are. The firm recruiters whom we interviewed firmly perceive certain occupations (i.e., computer and mathematics, ar
chitecture and engineering and sales) to be better performed by males rather than females due to the nature of these jobs, which in turn is factored in
their recruitment decisions. As the firm recruiters are the people deciding whom to be invited for a job interview, it is their perceptions that really
matter in hiring decisions.
11
females are, on average, less likely to receive an interview callback for these occupations. On the contrary, for business and financial
operations and office and administrative support occupations, there is no incongruity between the requirements of these occupations
and the stereotyped roles of women. As such, the role congruity theory predicts that females are not at a disadvantage in accessing
these occupations. Consistent with the prediction by the role congruity theory, the regression results of our study show that for these
occupations, the probability for an average female applicant to receive a callback is not statistically different from that of an average
male applicant.
Although the firm recruiters tend to stereotype women as “weak” and “lacking stamina” to perform some jobs, we do not know
whether their claim is from their prejudices toward women or based on their real experience with female workers. If the claim is driven
by their deep seated prejudices toward women, then the discrimination is more aligned with a taste-based discrimination view. If the
claim is supported by their actual experience with female workers and statistical evidence, then we could say that the discrimination is
more aligned with a statistical discrimination view. The recruiters’ main reasons for preferring male applicants are related to
“toughness”, “stamina” and not being “weak”, but they did not offer any concrete statistical evidence to support their claim. The claim
is more based on their perceptions. In addition, our regression results from the previous section that the quality of job candidates (the
student’s academic performance or the student’s leadership experience) do not reduce female discrimination in receiving onsite in
terviews are more consistent with a taste-based discrimination view. Therefore, to sum up, while we have some evidence to support the
taste-based discrimination, we do not have enough evidence to support the statistical discrimination view.13
Our interviews also indicate that pregnancy and childbirth are not the factors driving our results on gender discrimination.
Pregnancy and childbirth by female employees could cause unexpected interruptions and incur great expenses to a firm. Therefore,
firms would prefer males over females to avoid these types of life event disruptions. This could be a potential explanation for our
results. However, when we asked the recruiters whether they would consider pregnancy and childbirth as important factors when
hiring college students, all the 20 recruiters indicated that pregnancy and childbirth are not a concern. Almost all the recruiters
indicated that those who just graduated from colleges are very young (approximately 21–22 years old), and the probability of them
getting married and pregnant right after graduation is low. It has been reported that the average age for the first marriage for females
living in big cities in China is approximately 29 years old.14 Given the high employee turnover rates these days, it is understandable
that recruiters do not consider pregnancy and childbirth as important factors. On the contrary, when asked whether they would
consider these factors for experienced hiring, almost all the recruiters (17 out of the 20) said that they do. We are confident that our
results are not driven by the factors of pregnancy and childbirth.
5. Conclusion
This paper has examined employment-related gender discrimination at the hiring stage with a focus on college graduates in China.
We were able to control for the omitted variable and selection biases associated with job application by examining the responses of
firms to randomized fictitious resumes. This paper finds that with all other factors constant, female applicants, on average, receive far
fewer invitations from hiring firms for on-site interviews. Gender discrimination in the occupations of computer and mathematics,
architecture and engineering, and sales appear to be driving the results of our study. The qualitative evidence based on interviews with
a small number of firm recruiters suggests that the findings of this study are generally consistent with the role congruity theory of
prejudice in psychology literature. While this study offers some empirical evidence to support the taste-based discrimination view, we
do not have enough evidence to support the statistical discrimination view.
Our study complements a large number of studies identifying significant wage gaps between male and female workers in Chinese
labor markets, and finds that significant and pervasive gender discrimination exists at the hiring stage. Despite empirical challenges
(such as the inability to control for the omitted variable and selection biases), a large and significant earning gap between male and
female workers has consistently been identified by a large number of studies using survey data (Meng 1998; Rozelle et al., 2002; Liu
et al., 2000; Maurer-Fazio and Hughes 2002). In addition, the overall gender wage gap is consistently found to have been widening
significantly since the late 1990s (Chi and Li 2008; Zhang et al., 2008; Chi and Li 2014; Lee and Wie 2017). For instance, Lee and Wie
(2017) show that the female-to-male earnings ratio fell from 85% in 1988 to 72% in 2009. This fall is often attributed to gender
discrimination (Chi and Li 2008; Lee and Wie 2017).
The remarkably persistent gender gap in invitations to job interviews implies that gender discrimination in the wage rate may be
biased if the widespread discrimination against women in the hiring process is not taken into account. In light of the significant and
large economic and social costs associated with the widespread gender discrimination, it is important for the government, employers,
and society to consider implementing appropriate measures to curb gender discrimination in the labor market and promote equality.
Our study also extends the existing literature on the determination of labor market outcomes for college graduates in China. Li et al.
(2012a) find that, other things being equal, attending elite colleges gives college graduates a net wage premium of 10.7%. In another
paper, Li et al. (2012b) find that having a cadre parent, on average, can increase the first-job wage of a college graduate by 15%, and
male college graduates earn 19% higher from their first-time jobs following graduation than their female counterparts. Further, college
13
One thing to note is that the root of discrimination can be complex. For example, discrimination can be a result of combination of multiple
factors. Beaman et al. (2009) propose a model where gender discrimination can be caused by a combination of taste-based and statistical
discrimination. Disentangling the causes of discrimination is challenging.
14
Shanghai Women’s Federation and Shanghai Academy of Social Sciences, “A Survey Report on Woman Development for 40 Years since Reform
and Open-up”, 2018.
12
grade point averages (GPAs) and English skills are significant determinants of the first-job wages, while Communist Party membership
and student-leadership experiences do not appear to affect first-job wages. Li and Zhang (2010) find that, other things being equal,
graduates with better college GPAs are more likely to be employed. There is also some weak evidence that the job market may prefer
male graduates over female graduates with similar qualifications. Our study further confirms that while other characteristics generally
matter in affecting hiring firms’ decisions in selecting job applicants for on-site interviews, the applicants’ gender appears to be a more
consistent and significant factor.
Appendix
Appendix Table 1
Test for random assignment of student variables to firms.
Panel A: Chi-square Independence Test Male English skills Academic performance Communist Party member Student leader University
Firm size 3.02 4.34 3.40 2.44 2.06 44.81

Firm industry 15.63 17.83 23.63 24.93 19.27 271.09
Firm ownership 1.65 1.21 4.90 5.03 2.41 38.61
Panel B: Correlation Male English skills Academic performance Communist Party member Student leader University
Firm size 0.027 − 0.051 − 0.009 0.003 − 0.033 − 0.039
Firm industry − 0.001 0.041 0.031 0.027 − 0.006 − 0.011
Firm ownership 0.017 0.031 − 0.033 0.063 − 0.033 − 0.036
Note: In Panel A, the chi-square independence test statistics are reported. Correlation coefficients are reported in Panel B. ***, **, and * refer to 1%,
5%, and 10% statistical significance level, respectively.
Appendix Table 2
Correlation of randomized student variables.
Panel B Male English skills Academic performance Communist party member Student leader University
Male 1.000
English skills − 0.032 1.000
Academic performance 0.032 − 0.044 1.000
Communist Party member − 0.079 0.007 0.007 1.000
Student leader − 0.002 − 0.028 0.011 0.030 1.000
University − 0.018 0.015 0.013 − 0.031 0.020 1.000
Appendix Table 3
Distributions of hiring Firms’ industries, preferred majors, and occupations.
A. Industry Obs. % B. Preferred Major Obs. % C. Occupations Obs. %
Manufacturing 299 26.27 Management 485 42.62 Computer and Mathematics 233 20.47
Information transfer, software 223 19.6 Computer Science 269 23.64 Business and Financial Operations 219 19.24
and IT services
Leasing and commercial services 133 11.69 Mechanical Engineering 81 7.12 Architecture and Engineering 152 13.36
Wholesale and retail trades 106 9.31 Economics 59 5.18 Office and Administrative Support 128 11.25
Scientific research and 76 6.68 Electrical Science 32 2.81 Sales 111 9.75
polytechnic services
Finance 72 6.33 Journalism and 30 2.64 Management 60 5.27
Communication
Education 52 4.57 Civil Engineering 26 2.38 Arts, Design, Entertainment, Sports and 58 5.10
Media Occupations
Real estate 42 3.69 Chemistry 25 2.20 Installation, Maintenance, and Repair 30 2.64
Construction 41 3.60 Arts 21 1.85 Personal Care and Service 29 2.55
Culture, sports, and 29 2.55 English 19 1.67 Educational Instruction and Library 26 2.28
entertainment
Accommodation and catering 19 1.67 Bioscience 17 1.49 Life, Physical, and Social Science 25 2.20
Occupations
Transport, storage, and postal 15 1.32 Law 16 1.41 Legal occupations 17 1.49
services
Others 31 2.72 Medicine 15 1.32 Production occupations 11 0.97
Electronic Information 12 1.05 Others 39 3.43
Others 31 2.72
Total
1138 100 1138 100 1138 100
13
Appendix Table 4
Demand for management majors by hiring firms.
Major Observation Percentage
Marketing 158 32.58

Accounting 80 16.49
Administrative Management 62 12.78
Human Resources Management 52 10.72
Business Administration 37 7.63
Project Management 21 4.33
Public Relations 20 4.12
Financial Management 19 3.92
Others 36 7.42
Total 485 100
Appendix Table 5
Comparison of callback rates (%) of different firm groups.
Panel A Overall Manufacturing Wholesale and Information Finance Leasing and Scientific Research and
Retail Trades Technology Commercial Services Polytechnic Services
Callback 17.75 9.70 23.58 18.38 27.78 22.56 11.84

Obs. 1138 299 106 223 72 133 76
Panel B Firm Size Firm Ownership
<=100 100–500 500–1000 >1000 State- Private Foreign
owned
Callback 15.13 19.07 17.88 18.63 15.21 19.55 14.07
Obs. 304 451 179 204 217 757 135
Appendix Table 6
Regression results on the interaction of the female variable with other student characteristics.
Variable (1) (2) (3) (4) (5) (6)
Female − 0.0494 − 0.0687** − 0.0751** − 0.102*** − 0.0718*** − 0.0705**

(0.0330) (0.0305) (0.0312) (0.0340) (0.0268) (0.0290)
Female * CET-6 − 0.0389
(0.0420)
Female * Score ranked Top 5 − 0.00122
(0.0428)
Female * Communist Party 0.0115 (0.0543)
Membership
Female * Student Association 0.0654 (0.0472)
Leader
Female * (Univ. 1 or Univ. 2) 0.0195 (0.0672)
Female * Tier 1 Univ. 0.00393
(0.0513)
CET-6 0.0327 (0.0309) 0.0132 (0.0216) 0.0130 (0.0215) 0.0143 (0.0215) 0.0133 (0.0214) 0.0132 (0.0215)
Score ranked Top 5 − 0.0225 − 0.0222 − 0.0228 − 0.0229 − 0.0226 − 0.0228
(0.0234) (0.0306) (0.0234) (0.0234) (0.0232) (0.0233)
Communist Party membership 0.0344 (0.0225) 0.0340 (0.0226) 0.0282 (0.0369) 0.0346 (0.0222) 0.0342 (0.0225) 0.0341 (0.0225)
Student Association Leader 0.0107 (0.0196) 0.0114 (0.0195) 0.0114 (0.0194) − 0.0217 0.0113 (0.0195) 0.0113 (0.0195)
(0.0340)
University dummies Yes Yes Yes Yes Yes Yes
Firm size dummies Yes Yes Yes Yes Yes Yes
Firm ownership dummies Yes Yes Yes Yes Yes Yes
Firm industry dummies Yes Yes Yes Yes Yes Yes
On-site resume submission (1 = Yes Yes Yes Yes Yes Yes
yes)
College major dummies Yes Yes Yes Yes Yes Yes
Firm recruit timing dummies Yes Yes Yes Yes Yes Yes
Job fair dummies Yes Yes Yes Yes Yes Yes
Observations 1138 1138 1138 1138 1138 1138
level, respectively. Univ. 1 and Univ. 2 are indisputably the two best universities in China.
14
Appendix Table 7
Description of hiring firms’ characteristics.
Category Job fair resume submission (%) Online resume submission (%)
Location Beijing 89.85 47.30

Not Beijing 10.15 52.70
Size (No. of employees) <=100 28.93 17.57
100 - 500 42.36 28.38
500 - 1000 14.85 19.37
>1000 13.86 34.68
Ownership State-owned 19.32 18.02
Private 67.25 63.51
Foreign 10.70 16.67
Government 2.73 1.80
Industry
Manufacturing 24.24 34.68
Information transfer, software and IT services 20.31 16.67
Leasing and commercial services 12.88 6.76
Wholesale and retail trades 9.17 9.91
Scientific research and polytechnic services 6.33 8.11
Finance 5.68 9.01
Education 5.24 1.80
Real estate 3.28 5.41
Construction 3.82 2.70
Culture, sports, and entertainment 2.84 1.35
Accommodation and catering 1.97 0.45
Transport, storage, and postal services 1.31 1.35
Others 2.93 1.80
Total number 916 222
Appendix Table 8
Determinants of obtaining an on-site interview for job fair resume submission.
Variable (1) (2) (3) (4)
Female − 0.0939*** − 0.101*** − 0.0940*** − 0.0920***

(0.0296) (0.0296) (0.0259) (0.0269)
CET-6 0.0192 0.0210 0.0235
(0.0228) (0.0225) (0.0253)
Score ranked Top 5 0.000114 − 0.00725 − 0.0178
(0.0282) (0.0285) (0.0283)
Communist Party Membership 0.0534** (0.0241) 0.0596** (0.0247) 0.0653** (0.0263)
Student Association Leader − 0.0118 (0.0196) 0.000644 (0.0198) 0.00838 (0.0226)
Obs. 916 916 916 916
15
Appendix Table 9
Determinants of Obtaining an on-site interview for online resume submission.
Variable (1) (2) (3) (4)
Female 0.00439 0.0288 0.0210 0.0188

(0.0377) (0.0400) (0.0446) (0.0471)
CET-6 − 0.00805 − 0.0156 − 0.0370
(0.0364) (0.0362) (0.0398)
Score ranked Top 5 − 0.0487 − 0.0480 − 0.0470
(0.0372) (0.0402) (0.0400)
Communist Party membership − 0.0740** (0.0353) − 0.0472 (0.0376) − 0.0658 (0.0415)
Student Association Leader 0.00743 (0.0409) 0.00998 (0.0398) 0.0109 (0.0439)
Obs. 222 222 222 222
Note: Robust standard errors clustered at the online job board level are reported in parentheses. ***, **, and * refer to 1%, 5%, and 10% statistical
significance level, respectively.
References
Aigner, Dennis, Cain, Glen, 1977. Statistical theories of discrimination in labor markets. Ind. Labor Relat. Rev. 30 (2), 175–187.
All-China Women’s Federation. 2011. “National main data report on the third round survey of Chinese women social Status.” Retrieved from http://www.wsic.ac.cn/
staticdata/84760.htm.
Arrow, Kenneth, 1973. The theory of discrimination”. Discriminat. Labor Market. 3 (10), 3–33.
Banerjee, Abhijit, Bertrand, Marianne, Datta, Saugato, Mullainathan, Sendhil, 2009. Labor market discrimination in Delhi: evidence from a field experiment. J. Comp.
Econ. 37, 14–27.
Becker, Gary, 1957. The Economics of Discrimination. University of Chicago Press.
edited by Bertrand, M., Duflo, E., Banerjee, A., Duflo, E., 2017. Field experiments on discrimination. In: Handbook of Field Experiment, 1. North Holland, Amsterdam,
pp. 309–393. edited by.
Bertrand, Marianne, Mullainathan, Sendhil, 2004. Are emily and greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination.
Am. Econ. Rev. 94, 991–1013.
Bishop, John A., Luo, Feijun, Wang, Fang, 2005. Economic transition, gender bias, and the distribution of earnings in China. Econ. Trans. 13 (2), 239–259.
Booth, Alison, Leigh, Andrew, 2010. Do employers discriminate by gender? A field experiment in female-dominated occupations. Econ. Lett. 107, 236–238.
Bulger, Christine M., 2000. Fighting gender discrimination in the Chinese workplace. 20 BC. Third World L.J. 345.
Carlsson, Magnus, 2011. Does hiring discrimination cause gender segregation in the Swedish labor market? Fem. Econ. 17 (3), 71–102.
Chen, Mingyu, 2020. The Value of U.S. College Education in Global Labor Markets: Experimental Evidence from China. IRS Working Papers, Princeton University.
Chi, Wei, Li, Bo, 2008. Glass ceiling or sticky floor? Examining the gender earnings differential across the earnings distribution in Urban China, 1987–2004. J. Comp.
Econ. 36, 243–263.
Chi, Wei, Li, Bo, 2014. Trends in China’s gender employment and pay Gap: estimating gender pay gaps with employment selection. J. Comp. Econ. 42, 708–725.
Demurger, Sylvie, Fournier, Martin, Chen, Yi, 2007. The evolution of gender earnings gaps and discrimination in Urban China, 1988–1995. Dev. Econ. XLV-1, 97–121.
Edelman, Benjamin, Luca, Michael, Svirsky, Dan, 2017. Racial discrimination in the sharing economy: evidence from a field experiment. Am. Econ. J. 9 (2), 1–22.
Fincher, Leta Hong, 2013. China’s entrenched gender gap. The New York Times (May 20, 2013). Retrieved from http://www.nytimes.com/2013/05/21/opinion/
global/chinas-entrenched-gender-gap.html.
Galarza, Francisco B., Yamada, Gustavo, 2014. Labor market discrimination in lima, Peru: evidence from a field experiment. World Dev. 58, 83–94.
Gustafsson, Bjorn, Li, Shi, 2000. Economic transformation and the gender earnings gap in Urban China. J. Popul. Econ. 13 (2), 305–329.
Hou, Yue, Liu, Chuyu, Crabtree, Charles, 2020. Anti-Muslim bias in the Chinese labor market. J. Comp. Econ. 48 (2), 235–250.
Kuhn, Peter, Shen, Kailing, 2013. Gender discrimination in job ADS: evidence from China. Q. J. Econ. 128 (1), 287–336.
Lee, Jong-Wha, Wie, Dainn, 2017. Wage structure and gender earnings differentials in China and India. World Dev. 97, 313–329.
Levinson, R., 1975. Sex discrimination and employment practices: an experiment with unconventional job inquiries. Soc Probl 22, 533–543.
Li, Hongbin, Meng, Lingsheng, Shi, Xinzheng, Wu, Binzhen, 2012a. Does having a cadre parent pay? Evidence from the first job offers of Chinese college graduates.
J Dev Econ 99, 513–520.
Li, Hongbin, Meng, Lingsheng, Shi, Xinzheng, Wu, Binzhen, 2012b. Does attending elite colleges pay in China? J. Comp. Econ. 40, 78–88.
Li, Tao, Zhang, Juyan, 2010. What determines employment opportunity for college graduates in China after higher education reform? China Econ. Rev. 21, 38–50.
Liu, Pak-Wai, Meng, Xin, Zhang, Junsen, 2000. Sectoral gender wage differentials and discrimination in the transitional Chinese economy. J. Popul. Econ. 13 (2),
331–352.
Maurer-Fazio, Margaret, Hughes, James, 2002. The effects of market liberalization on the relative earnings of Chinese women. J. Comp. Econ. 30, 709–731.
Maurer-Fazio, Margaret, 2012. Ethnic Discrimination in China’s internet job board labor market. IZA J. Migrat. 1–12.
Meng, Xin, Miller, Paul, 1995. Occupational segregation and its impact on gender wage discrimination in China’s rural industrial sector. Oxf. Econ. Pap. 47 (1),
136–155.
Meng, Xin., 1998. Male-female wage determination and gender wage discrimination in China’s rural industrial sector. Labour. Econ. 5, 67–89.
Neumark, David, Bank, Roy J., Van Nort, Kyle D., 1996. Sex Discriminationin restaurant hiring: an audit study. Q. J. Econ. 111, 915–941.
Han, Anna M., 2001. Holding-Up more than half the sky: marketization and the status of women in China. Journal of Contemporary Legal Issues 11 (791).
Petit, Pascale, 2007. The effects of age and family constraints on gender hiring discrimination: a field experiment in the French financial sector. Labour. Econ. 14,
371–391.
Phelps, Edmund, 1972. The statistical theory of racism and Sexism. Am. Econ. Rev. 62 (4), 659–661.
Riach, Peter, Rich, Judith, 1987. Testing for sexual discrimination in the labour market. Aust. Econ. Pap. 26, 165–178.
Riach, Peter, Rich, Judith, 2006. An experimental investigation of sexual discrimination in hiringin the English labor market. B E J. Econom. Anal. Policy 6 (2).
Rozelle, Scott, Dong, Xiao-Yuan, Zhang, Linxiu, Mason, Andrew, 2002. Gender wage gaps in post-reform Rural China. Pacific Econ. Rev. 7 (1), 157–179.
16
Steinfeld, Jemimah., 2014. Women still face great wall of discrimination in China. CNN (March 8, 2014). Retrieved from http://www.cnn.com/2014/03/08/world/
asia/china-gender-discrimination/index.html.
Wang, Huajie, 2016. 50% Chinese women still suffer gender discrimination: survey. Women of China. All-China Women’s Federation (March 29, 2016). Retrieved
from http://www.womenofchina.cn/womenofchina/html1/survey/1603/2239-1.htm.
Yang, Sadie, Li, Ao, 2009. Legal protection against gender discrimination in the workplace in China. Gend. Dev. 17 (2), 295–308.
Zhang, Junsen, Han, Jun, Liu, Pak-Wai, Zhao, Yaohui, 2008. Trends in the gender earnings differential in Urban China, 1988-2004. Ind. Labor Relat. Rev. 61 (2),
224–243.
17

Zhang 2021

Uploaded by

Copyright:

Available Formats

Zhang 2021

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Zhang 2021

Uploaded by

Copyright:

Available Formats

Journal of Comparative Economics xxx (xxxx) xxx

Contents lists available at ScienceDirect

Journal of Comparative Economics

Gender discrimination in China: Experimental evidence from the

2.1. Creating randomized fictitious resumes

2.2. Sample and the data

2.3. Nature of the experiment

1. Gender Male 571 50.18

2.4. Experimental procedure

3. The effect of gender on invitations for on-site interviews

3.1. Testing for randomness and independence of the experiment

Location Beijing 928 81.55

3.2. Description of the hiring firms

3.3. Impact of gender on callbacks for on-site interviews

3.3.1. Descriptive results

Gender 13.93 21.54 − 7.61***

Female − 0.0761*** − 0.0772*** − 0.0757*** − 0.0693***

3.3.2. Regression results

Female − 0.0317 − 0.0798** − 0.115* − 0.0895* − 0.0617 − 0.0728** − 0.0999

Female − 0.0552* − 0.163* − 0.106* − 0.0736 − 0.0891 − 0.0312

Female − 0.0748* − 0.131** − 0.149* − 0.00278 0.0279

4. The mechanism of gender discrimination

4.1. Interview findings from firm recruiters

4.2. Discussion on the interview findings

Firm size 3.02 4.34 3.40 2.44 2.06 44.81

Marketing 158 32.58

Callback 17.75 9.70 23.58 18.38 27.78 22.56 11.84

Female − 0.0494 − 0.0687** − 0.0751** − 0.102*** − 0.0718*** − 0.0705**

Location Beijing 89.85 47.30

Female − 0.0939*** − 0.101*** − 0.0940*** − 0.0920***

Female 0.00439 0.0288 0.0210 0.0188

You might also like

Female − 0.0761* − 0.0772* − 0.0757* − 0.0693*

Female − 0.0494 − 0.0687 − 0.0751 − 0.102* − 0.0718* − 0.0705**

Female − 0.0939* − 0.101* − 0.0940* − 0.0920*