ESSAYS ON LABOR ECONOMICS

The province administered both the mock examination as well as the Gaokao, therefore, the mock examination was a good indication of thedegree of difficulty and subject material covered b

ESSAYS ON LABOR ECONOMICS

CAI XIQIAN

( B.A XIAMEN UNIVERSITY,

A THESIS SUBMITTED FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

DEPARTMENT OF ECONOMICS NATIONAL UNIVERSITY OF SINGAPORE

2015

I hereby declare that this thesis is my original work and it has been written

by me in its entirety I have duly acknowledged all the sources of information

which have been used in the thesis

This thesis has also not been submitted for any degree in any university

previously

Signed:

First and foremost, I would like to thank to my PhD advisors, Assoc Prof

Hu Guangzhou, Albert and Lu Yi, for supporting me during these past fouryears It would not have been possible for me to complete my thesis withouttheir supervision and collaboration A/P Hu Guangzhou, Albert is a verynice advisor He has given me the freedom to pursue various projects withoutobjection and also provided insightful discussions about the research A/P

Lu Yi is always available and willing to help He shows great patience to

me, and guided me through each step of research Moreover, he is a greatlife mentor, and always gives invaluable suggestions on academic and personaldevelopment matters I would like to express my heartfelt gratitude to both

of them It is my honor to be under their supervision

Secondly, I would like to thank Assoc Prof Davin CHOR, Alberto

SAL-VO, Dr Gong Jie, Jessica PAN, and Zhong Songfa, for their generous ances and valuable comments on my work I would also like to thank Assoc.Prof Luo Xiao, Liu Haoming, Zeng Jingli, Dr Song Changcheng, Yi Junjian,and Zhu Shenghao for their help and suggestions during my study at NUS.Thirdly, I would like to thank all my Ph.D colleagues for making my stay

guid-in NUS that much more pleasurable Thanks Xie Huihua, Zhou Yguid-inke, ZhuLianming, Li Yunong, Mun Lai Yoke, Liu Xuyuan, Sun Yifei, Shen Bo, Shao

Lei, Yang Guangpu, Yang Songtao, Wang Jianguang, Liu Bin, Zeng Ting, LuYunfeng, and others for all the inspirational conversations and unselfish help.Last but not the least, I would especially thank my family, especially mygrandparents and parents for their unparalleled love and constant supportand my wife Lin Yujia, who is always there for me with her unwavering loveand encouragement Their love and support have accompanied me along thejourney and helped me get close to my dream This thesis is dedicated tothem.

1.1 Introduction 1

1.2 Institutional Background 7

1.3 Data and Descriptive Statistics 10

1.4 Gender Gap in Performance on Mock Exam vs Gaokao 13

1.4.1 Empirical Strategy 15

1.4.2 Results 17

1.4.3 Gender Differences in Effort Provision 19

1.4.4 Do Males and Females Respond Differently to Pressure? 25 1.5 Gender Gap in Reaction to Shocks 28

1.5.1 Are the Results Driven by Unobservables? 32

1.5.2 Placebo Tests - Role of External Shocks 37

1.6 Conclusion 40

2 Recover Overnight? Work Interruption and Worker

2.1 Introduction 42

2.2 Context 48

2.2.1 Workplace 48

2.2.2 A Stylized Model 50

2.3 Estimation Strategy 54

2.3.1 Data, Variables, and Specification 54

2.3.2 Identifying Assumption and Validity Checks 57

2.4 Empirical Findings 64

2.4.1 Validity Checks 64

2.4.2 Economic Magnitude 67

2.4.3 Interpretation 68

2.4.4 Additional Results: Heterogeneity and Social Comparison 71 2.5 Conclusion 73

3 The Impact of Temperature on Worker Productivity 75 3.1 Introduction 75

3.2 Context 79

3.3 Data 81

3.3.1 Daily Productivity Data 81

3.3.2 Weather and Air Quality Index Data 91

3.4 Estimation Strategy 93

3.5 Results 95

3.5.1 Main Productivity Results 95

3.5.2 Robustness Checks 97

3.6 Conclusion 100

Appendices 137 Appendix of Chapter One 137 Figures and Tables for Chapter One 94 Figures and Tables for Chapter Two 105 Figures and Tables for Chapter Three 112

Bibliography 124

This thesis consists of three independent chapters.1

The first chapter examines the gender gap in examination performance inresponse to competitive and performance shocks By using a unique datasetfrom the College Entrance Examination in China, We find that females per-form worse than males and are more response to shocks under pressure.The second chapter investigates the effect of interruption on workers’ subse-quent productivity using plausibly exogenous incidents of machine breakdown.Using a dataset from a plastic printing company in China, we find that individ-ual worker productivity declines following a workday with machine breakdown.The last chapter provides the first evidence about how temperature affect-

s worker productivity from a manufacturing firm in a non-climate-controlledenvironment in China Using the exogenous day-to-day variations in temper-ature, we find an inverted U-shaped effect of the daily maximum temperature

on worker productivity, and short-run avoidance behavior to temperaturesthrough absenteeism

1

The first chapter is co-authored with Professor Lu Yi, Jessica Pan and Zhong Songfa, the second chapter is co-authored with Professor Gong Jie , Lu Yi, and Zhong Songfa, while the third chapter is co-authored with Professor Yi Lu

List of Figures

1.1 Distributions of Female and Male Performance on the MockExam and Gaokao 141.2 Relationship between Day 1 Afternoon Exam Score and Morn-ing Exam Score by Gender 301.3 Relationship between Day 1 Afternoon Exam Score and Morn-ing Exam Score by Gender and Distance from Reference Cutoff 33

2.1 Share of Machines that Broke Down (October 2012 March 2014) 562.2 Distribution of Share of Machines that Broke Down 562.3 Distribution of Number of Machines Used by a Worker 582.4 Placebo: Coefficients from Randomly Distributed Interruptions 66

3.1 Standardized Average Productivity For Each Of The Products 833.2 Variation In Productivity By Job Tasks 833.3 Variation In Productivity By Workers 833.4 Variation In Productivity By Day 84

3.5 Daily Temperature, AQI And The Concentration Levels Of

Each Individual Pollutents, By Year 85

3.6 Relationship Between Temperature And Labor Productivity 87 3.7 Relationship Between AQI And Labor Productivity 88

3.8 Robustness Checks- Same Machine and Same Product 89

3.9 Robustness Checks- Individual*Machine*Product FE 90

3.10 Robustness Checks- Without Breakdown 90

3.11 Robustness Checks- First Time Using 90

3.12 Robustness Checks- Absence Yesterday 91

3.13 Robustness Checks- Without Typhoon Events 91

List of Tables

1.1 Summary Statistics 121.2 Gender Gap in Mock Exam and Gaokao Scores 151.3 Regression Estimates of the Gender Gap in Performance 181.4 Regression Estimates of the Gender Gap in Performance byInitial Performance on Mock Exam 211.5 Mock Exam Reference Cut-offs for each University Tier In 2008 231.6 Fraction of Students in Fujian Scoring above each of the Refer-ence Cut-offs in the Mock Exam in 2008 231.7 Fraction of Students in Anxi Scoring above each of the ReferenceCut-offs in the Mock Exam in 2008 231.8 Is the Gender Gap in Performance Larger where it Matters More? 251.9 Do Females Underperform where it Matters More? 281.10 Gender Difference in Afternoon Performance in Response toRelative Performance on Morning Exam on Day 1 31

1.11 Gender Difference in Afternoon Performance in Response toRelative Performance on Morning Exam on Day 1 - Positive vs.Negative Shocks 362.1 Effect of Machine Breakdown on Workers Subsequent Produc-tivity 642.2 Tests on Proficiency and Emotional Reactions 692.3 Heterogeneous Effects and Social Comparison 72

3.1 Summary Statistics For The Main Worker Characteristics 823.2 Summary Statistics For The Weather Elements And Air QualityIndex 843.3 The CorrelationsAnd Partial Correlations Between AQI AndEach Of The Individual Pollutents 853.4 The Distribution Of Temperature And The Percentage Of Pro-ductivity Increment 92

Chapter 1

Gender Gap under Pressure:

Performance and Reaction to

Shocks

A large number of experimental studies suggest that men and women responddifferently to competitive pressures These studies document that women ap-pear to systematically underperform relative to men in competitive settingsand that women may simply prefer to opt-out from competition (for exam-ples, see Bertrand, 2010, Gneezy, Niederle and Rustichini, 2003, Gneezy andRustichini, 2004, Niederle and Vesterlund, 2007, 2011) These studies positthat gender differences in performance and attitudes toward competition mayexplain an important part of the gender gap in educational choices and labormarket outcomes (Buser, Niederle and Oosterbeek, 2014)

A growing line of research has attempted to assess whether such

perfor-mance differences in respond to competition exist in real-world settings terestingly, the results are somewhat mixed Earlier studies by Lavy (2013)and Paserman (2010) examine gender differences in the performance of highschool teachers and professional tennis players, respectively, and find little evi-dence that women perform worse in more competitive settings More recently,

In-a number of studies focusing on reIn-al-world In-acIn-ademic settings show thIn-at menappear to outperform women when competitive pressures are higher, where-

as the reverse holds true in less competitive settings (Azmat, Calsamiglia andIriberri, 2014, Morin, 2013, Ors, Palomino and Peyrache, 2013, Attali, Neemanand Schlosser, 2011, and Jurajda and Munich, 2011)

In this paper, we utilize a unique dataset from China’s National CollegeEntrance Examination (Gaokao) and an arguably cleaner empirical setup toexamine the extent and mechanisms through which competitive pressure af-fects the gender gap in academic performance Having established that womentend to underperform on the high-stakes national Gaokao relative to the low-stakes mock examination held two months earlier, we then turn to a different,but related question In particular, we examine whether males and femalesreact differently to performance shocks on an earlier exam and how these d-ifferences may interact with the degree of competitive pressure faced Thelatter question is also important in the literature on gender gaps in perfor-

mance as there is increasing evidence that women may be less confident andless resistant to negative shocks.1

The Gaokao is widely regarded as one of the most competitive tions in the world - it is practically the only route to admission into universities

examina-of higher education and further success in the test-oriented education system

of China Furthermore, the number of exam takers typically exceed the able places for higher education The admission rate for candidates sitting forthe Gaokao is approximately 75% and students’ performance on the two-dayexamination is typically the sole criteria used to determine their placementinto one of China’s nearly two thousand colleges Each college has separatecut-offs that determine whether students can qualify for various academic pro-grams and entry into a particular college and major is determined almostexclusively by students’ performance on the Gaokao In fact, the examination

avail-is so important that a couple of months prior to the actual examination, eachprovince typically runs a mock examination to ensure that the examiners arefamiliar with the examination protocol and to allow students to gauge theirpreparedness and relative performance on the exam

Drawing on a dataset that comprises the universe of Gaokao takers in Anxi

1

For examples, see Roberts and Nolen-Hoeksema, 1989 and a article in the The Atlantic (http://www.theatlantic.com/features/archive/2014/04/the-confidence-gap/ 359815/).

County in 2008, we are able to directly observe the performance of the sameindividual in the Gaokao and the earlier mock examination By comparing thegender difference in performance in what is essentially the same examination

in a high and low-stakes setting, we are able to cleanly estimate of the effect

of competitive pressures on the gender gap in performance in an importantreal world setting.2 Consistent with previous evidence, we find that malesare significantly more likely to outperform females in the high-stakes Gaokaorelative to the low-stakes mock examination Furthermore, we also exploitadditional variation in stakes, within gender, to examine whether the observedeffects are driven by females underperforming in high-stakes, high pressuresettings, or males improving their performance when the stakes are higher.Next, we extend the previous literature by analyzing whether there is agender difference in the reaction to performance shocks This is also an im-portant question in the literature on gender gaps in performance as there isincreasing evidence that women are less confident (and men tend to be over-confident) and may react more aversely to negative feedback than their malecounterparts (Roberts and Nolen-Hoeksema, 1989 and Goldin, 2013) Thislower degree of confidence and resistance to negative shocks may be anoth-

2

In our setting, relative to the mock examination, the Gaokao is both higher-stakes and more competitive (students compete for a limited number of slots to qualify for their academic program of choice) We do not distinguish the role of exam stakes and the degree of competition and describe the Gaokao as “higher stakes”, “more competitive” and entailing

a greater degree of “competitive pressure” interchangeably.

er reason as to why women are more likely to opt-out of competitive fields(Niederle and Vesterlund, 2007) We exploit the fact that the individual sub-ject components of the examinations are spread out over two days to explorewhether there is a gender difference in the reaction to performance shocks in

an earlier exam To our knowledge, this question remains unexplored in theliterature

Our paper is closely related to a recent literature that examines whethermen and women respond differently to competitive pressures and test stakes

in real-world settings Like our paper, most of these papers focus on academicperformance and exploit differences in the nature of the test or test setting -for example, how competitive the tests are, the level of stakes involved andthe grading scheme used, to examine whether the relative performance of fe-males is affected by the nature of the test For example, Ors, Palomino andPeyrache (2013) show that females outperform males in a less competitive na-tional exam, but for the same cohort of students, males outperform females inthe highly selective competitive entrance exam for admission to a top businessschool in France Morin (2013) exploits a legislative change in Ontario thatexogenously increased competition for university grades and documents thatamong students affected by the change, male performance improved relative

to females Azmat, Calsamiglia and Iriberri (2014) utilizes variation in test

stakes across different exams that Spanish high school students are required

to sit for throughout the year and show that males tend to outperform femaleswhen test stakes are higher

Our setting has a number of features that differentiate it from previousstudies First, the same examination board sets and implements both thehigh and low-stakes exams and the coverage of the test material is identical inboth settings.3 Second, as the Gaokao is the main requirement for admissioninto all colleges in China, there is limited sample selection of individuals intothe actual high-stakes test based on their performance on the low-stakes test.This also ameliorates potential sample selection concerns that individuals with

a greater distaste for competition may choose not to participate in the stakes examination Third, we are able to exploit variation in the perceivedstakes within gender to provide some suggestive evidence of the mechanismsthat lead to female underperformance in more competitive settings Finally,our paper also provides novel evidence documenting gender differences in theresponsiveness to performance shocks on an earlier exam

high-The rest of the paper proceeds as follows high-The next section describes theinstitutional background of the Gaokao in China Section 1.3 outlines the

3

In previous studies (e.g Ors, Palomino and Peyrache, 2013, Azmat, Calsamiglia and Iriberri, 2014), the low-stakes and high-stakes settings considered typically involve different testing strategies, material and timing, which might conflate gender differences in the re- sponse to high vs low-stakes with gender differences in the skills required in the high vs low-stakes settings.

data used and the descriptive statistics The results on the gender gap inperformance on the Gaokao relative to the mock examination are presented

in Section 1.4 Section 1.5 examines gender differences in the reaction toperformance shocks Section 1.6 concludes

The National College Entrance Examination (NCEE), commonly known asGaokao, is an annual two or three day examination that is a pre-requisite forentrance into almost all institutions of higher education at the undergraduatelevel in China.4 There are different tiers of universities in China, namelykey universities (Tier 1), regular universities (sometimes further subdividedinto two different tiers - Tier 2 and Tier 3), and technical colleges, and thedifferences among them are mostly based on ranking of the intuitions and theduration of the programs (Davey, De Lian and Higgins, 2007)

The Gaokao is ultimately under the control of the Ministry of Educationand was once administered uniformly across the country Starting in 2001,some provinces or direct-controlled municipalities arranged separate exam pa-pers while others still adopted the national exam papers The most commonly

4

For example, in 2006, 9.5 million people applied for tertiary education entry in China,

of which 93% were scheduled to take the national entrance exam The remaining applicants were either exempted from the standardized exams (0.3%) or scheduled to take other types

of standardized exams.

adopted examination system across the provinces is the “3+X” system - “3”represents the three compulsory subjects: Chinese, Mathematics and English(each accounting for 150/750 of the total score) and “X” represents the com-bined science subjects comprising physics, chemistry and biology for students

on the science track, or combined arts subjects of history, geography and itics for students on the arts and social sciences track (accounting for 300/750

pol-of the total score).5 All students, regardless of stream, sit for the same Chineseand English exam The coverage of the Math exam is different for students

in the science vs arts stream The “3+X” system is typically held over twodays in June in the following order: Chinese (Day 1, morning), Mathematics(Day 1, afternoon), Combined subjects (Day 2, morning) and English (Day 2,afternoon)

Two months prior to the Gaokao, a formal mock examination, tered by the province, is usually held to allow students to get a sense of theexamination and their relative standing within the province The mock exam-ination results are released about one week after students sit for the exam andthe Department of Education in each province also releases the province-leveldistribution of test scores as well as a set of reference cutoffs for each of thefour university tiers based on the proportion of students who were admitted

adminis-5

Students choose to be in the science stream or arts and social sciences stream in the beginning of the second year of high school.

into each university type in the previous year Students usually sit for theexamination in their last year of senior high school, although there has been

no age restriction since 2001 In different provinces, students either apply foruniversities prior to the Gaokao, after the Gaokao, or after they have learntabout their estimated scores based on the mock Gaokao examination and theirestimated rank in the province

The Gaokao is highly competitive It is commonly described as the “world’stoughest exam” due to the intense pressure and competition that students aresubject to The Gaokao is virtually the only path for Chinese students to

be admitted into universities and the number of exam takers typically exceedthe available places for higher education For example, in 2014, there were9.39 million test takers vying for about 7 million college spots.6 Furthermore,the 2000 or so universities in China are classified into four different tiers,with cutoff points to determine whether students can qualify for each tier ofuniversities Within each tier, each college also has a separate minimum examscore required for admission It is estimated that less than 10% of candidatesenroll into the top tier universities (key universities) and only less than 0.2% ofexam takers will gain admittance into China’s top five universities (Economist,

6

See: gaokao-weekend

http://www.businessweek.com/articles/2014-06-06/china-girds-for-high-stress-2012).7 It is a national consensus that getting into a better university via theGaokao greatly enhances an individual’s chances to obtain a better job inChina’s fiercely competitive job market.

Due to its importance and competitiveness, the Gaokao imposes enormouspressure on test takers, as well as their parents and teachers It is very commonfor students to spend hours studying for the Gaokao after returning home fromten hours of schooling, with little or no break on the weekends Many schoolsdedicate the entire senior year of high school to preparing students for theexam It is common to see astonishing new reports related to the Gaokao inthe local and international media For example, it was reported that somegirls took contraceptives or received injections to prevent the onset of theirmenstrual cycle during the week of the exam.8

Our data consists of test scores and demographic information for the universe

of Gaokao test takers as well as the test scores for all first-time test takers whosat for the mock examination in Anxi, a county of Fujian Province, in 2008.9

7

See: http://www.economist.com/blogs/analects/2012/06/university-entrance-exams 8

http://www.nytimes.com/2009/06/13/world/asia/13exam.html

9

We are only able to utilize data from a single year as prior to 2008, the mock exam data was unavailable Post-2008, the dissemination of exam results was centralized at the province level, thus, we did not have access to the Gaokao data 2008 was the only year that we were able to merge the individual-level mock exam data to the Gaokao data.

Appendix I provides some background information on the social and

econom-ic characteristeconom-ics of Anxi county and Fujian province The 2008 ProvincialMock Examination of Fujian was held in mid-April and the Gaokao was held

in mid-June The province administered both the mock examination as well

as the Gaokao, therefore, the mock examination was a good indication of thedegree of difficulty and subject material covered by the actual examination.Furthermore, the Department of Education in Fujian uses the mock exami-nation to determine the reference cutoff scores for each of the different tiers

of universities based on the proportion of students eligible for each tier in theprevious year.10 Although Gaokao takers in Fujian typically apply to universi-ties after they learn their actual scores and the actual cutoff points, the mockexamination is taken seriously by students as a way to estimate their relativerank in the province and to ascertain the likely tier of university that they willqualify for.11

Our dataset was constructed by merging the mock examination scores tothe Gaokao scores using each test-taker’s name Individuals with the samefirst and last name were dropped as they could not be uniquely identified

10

Appendix Table 1.1 lists the score cutoffs in the 2007 and 2008 Gaokao in Fujian province

as well as the proportion of students that meet the cutoffs for admission into each university tier.

From the Gaokao sample, we further dropped a small number of individualsfor whom we were not able to identify whether they were in the arts of sciencestream Our final merged sample comprises 7,961 individuals - which is 98% ofthe universe of mock examination candidates and 94% of Gaokao candidates.12The summary statistics for the mock examination sample, the Gaokao sampleand the merged sample are reported in Table 1.1 On average, the profile ofstudents in the merged sample is very similar to that in the mock examina-tion.13 Candidates who sat for both examinations (in the merged sample) were

of slightly higher ability than the overall Gaokao sample Nevertheless, thequalitative differences in the actual test scores for the two samples are quitesmall, ranging from 0.4 to 1.3 points out of a 150 or 300 point test There

is also little observed difference in the demographic characteristics across theGaokao and the merged sample Overall, these results suggest that there isvery little selection into the final sample based on an individual’s performance

on the mock examination and that the merged sample is broadly representative

of the universe of first-time Gaokao test-takers in Anxi county

[Insert Table 1.1 here]

Table 1.1: Summary Statistics

12

There were approximately 250 candidates who sat for the Gaokao and not the mock examination Most of these students were either retaking the Gaokao or were private candi- dates (i.e these students were not affiliated with a high school when they took the exam) 13

Appendix A also includes a profile of candidates in Anxi county, Fujian province and China as a whole.

1.4 Gender Gap in Performance on Mock

Ex-am vs Gaokao

Before turning to the formal econometric analysis, we present some suggestivegraphical and descriptive evidence of the gender gap in performance on thehigh-stakes Gaokao vs low-stakes mock examination The top two panels ofFigure 1.1 shows the distribution of total scores separately by gender for theApril mock examination and the June Gaokao As observed in the figure, whilethe female test-score distribution is to the right of the male distribution in themock examination, for the Gaokao, the male distribution appears to converge

to that of the female distribution Since we are presenting non-standardizedscores in this section, it is important to distinguish between students in theScience and Arts stream as one of the exam components, namely, the combinedscience/arts subject, differs across the two groups The middle and bottompanels of Figure 1.1 graph the distributions separately for students in theScience stream and Arts stream Among Science students, males appear tooutperform females at almost all points of the test score distribution, andthe male advantage becomes even more pronounced during the Gaokao Incontrast, among Arts students, we observe a strong female advantage in themock examination at all points of the distribution This advantage appears

to be reduced in the Gaokao, particularly among students in the middle to

upper-tail of the test score distribution.

[Insert Figure 1.1 here]

Figure 1.1: Distributions of Female and Male Performance on the Mock Examand Gaokao

Table 1.2 summarizes the means of the test-score distributions in Figure1.1 Columns (1) to (3) report the mean scores for females, males and thegender gap (female-male), respectively for the mock examination Columns(4) to (6) report similar statistics for the Gaokao Column (7) reports thedifference between the gender gaps reported in Columns (3) and (6) In thefirst three rows, we report the means for the total exam scores for all students

as well as for students in the Science and Arts stream separately On average,female and male students in the Science stream improve their scores on theGaokaorelative to the mock exam Interestingly, the gender gap is narrower inthe mock exam as compared to the Gaokao for science students Male sciencestudents perform about 3 points better on the mock exam relative to females;

in the Gaokao, the male test score advantage increases about three times to

9 points For students in the Arts stream, while females test scores declinedbetween the mock exam and Gaokao, male scores actually increased slightly.The gender gap in performance is also reduced significantly in the high-stakessetting for students in the Arts stream - the gender gap of 23 points in favor offemales on the mock exam nearly halves to about 13 points on the Gaokao In

sum, the raw data shows strong evidence that the gender gap in performance

is larger in high-stakes settings relative to low-stakes settings

[Insert Table 1.2 here]

Table 1.2: Gender Gap in Mock Exam and Gaokao Scores

Next, we turn to a formal empirical framework to more rigorously establish howgender impacts students’ relative performance in high vs low-stakes settings.Assume that an individual’s performance on each subject test on the Gaokao

is represented by the following equation:

Pi,gE,S = αSi + wSg + ZE,S + YiE,S+ xE,Sg + ǫE,Si,g

where E denotes the Gaokao Entrance Exam (M denotes the mock nation in the next equation), S denotes the type of student (science vs artsstream), i denotes individuals, g denotes gender αS

exami-i is the set of specific individual characteristics14 that do not change over the two exams,

stream-wS

g represent gender-specific characteristics that do not change over the twotests, ZE,S represent the Gaokao-specific characteristics (such as the location,temperature) that do not vary across gender, YiE,S is the set of individual

14

This also captures individual selection into the arts/science stream as the selection into streams is individual-specific and does not change across the two exams In our context, the choice of stream is chosen before students sit for either the mock exam or the Gaokao.

characteristics that may affect the two exams differently, xE,S

g captures theGaokao factors that vary by gender and ǫE,Si,g is the error term

Correspondingly, an individual’s performance on the mock examination isgiven by:

Pi,gM,S = αS

i + wS

g + ZM,S + YiM,S + xM,S

g + ǫM,Si,gFor both equations, to remove the effect of the Z’s or the exam-specificcharacteristics that are common to all individuals, we consider standardizedtest scores as the outcomes, that is, ˜Pi,gE,S = P

E,S i,g − ¯ Pi,gE,Sσ

Pi,gE,S

and ˜Pi,gM,S = P

M,S i,g − ¯ Pi,gM,Sσ

Pi,gM,S

.Taking the difference of the two resulting equations, we obtain:

˜

Pi,gE,S − ˜Pi,gM,S = (˜xE,Sg − ˜xM,Sg ) + ( ˜YiE,S − ˜YiM,S) + (˜ǫE,Si,g − ˜ǫM,Si,g )

Notice that this first difference specification allows us to difference out the vidual fixed effects that affect an individual’s performance in both the Gaokaoand mock examination (the αi’s) as well as the gender-specific characteristicsthat do not change across the two tests (the wg’s)

indi-Empirically, we can directly measure ˜Pi,gE,S and ˜Pi,gM,S using our data on testscores We use the female dummy to capture (˜xE,S

g − ˜xM,S

g ), and include schooldummies, zip code dummies and student age to control for ( ˜YiE,S − ˜YiM,S).The regression specification is given by:

˜

Pi,gE,S − ˜Pi,gM,S = β0+ β1F emalei+ Yiγ + ǫi,g (1.1)

As the outcomes have been standardized using the type-specific (arts vs ence) mean and variance, both ˜Pi,gE,S and ˜Pi,gM,S have a mean of 0 and a variance

sci-of 1 For ease sci-of interpretation, we further standardize the difference between( ˜Pi,gE,S− ˜Pi,gM,S) to have a mean of 0 and a variance of 1 so that the coefficient

β1 can be interpreted as the effect of being female on the difference in testscores between the Gaokao and mock examination in standard deviations.15

Table 1.3 reports the female coefficient estimate, β1, from the estimation ofequation (1) for the total score as well as for each of the individual subjects -Chinese, Mathematics, combined subjects and English Column (1) presentsthe raw gender difference in total test scores across the high-stakes Gaokaoand low-stakes mock examination - on average, the difference in score betweenthe Gaokao and mock exam among females is 0.16 standard deviations lowerthan that for males This difference is large and statistically significant and

is virtually unaffected by the addition of covariates (see Column (2)) Asobserved in Columns (3) to (10), most of this effect is driven by a significantlyworse relative performance by females on the combined subject test as well

as the English test The distribution of raw standardized differences across

15

The actual standard deviations for the difference in mock exam and Gaokao test scores ( ˜ Pi,gE,S and ˜ Pi,gM,S) are: Total (0.54), Chinese (0.91), Math (0.66), Combined Arts/Science (0.72), English (0.66).

the two tests by gender are shown in Appendix Figure 1.1 These resultsare consistent with the idea that females underperform relative to males onhigh-stakes vs low-stakes settings.

[Insert Table 1.3 here]

Table 1.3: Regression Estimates of the Gender Gap in Performance

To provide a sense of the magnitude of our estimates, Appendix Table 1.5reports the gender difference in the likelihood of qualifying for a Tier 1 or Tier

2 university based on the reference cutoffs in the mock exam (Column (4)) andthe actual cutoffs in the Gaokao (Column (8)) Based on their performance

on the mock exam, females are 1.1 percentage points less likely to be eligiblefor a Tier 1 university On the Gaokao, the gender difference nearly doubles

to 2.1 percentage points Columns (9) and (10) report the difference in thegender gap in Tier 1 eligibility across the Gaokao and mock exam with andwithout individual-level controls.16 We find that females are significantly (0.8percentage points) less likely than males to be eligible for a Tier 1 universitybased on their Gaokao scores relative to their performance on the mock exam.Given that, on average, 5.5% of Gaokao takers are eligible for Tier 1 universi-ties, this works out to be a relative decline of about 15% (0.8/5.5=0.15) Thesecond and third rows of Appendix Table 1.5 examine the gender gap in the

16

The set of controls are the same as those used in Table 1.3.

likelihood of qualifying for Tier 2 universities and either Tier 1 or Tier 2 versities, respectively The corresponding difference in the gender gap in high

uni-vs low-stakes setting is 7% (1.7 percentage points) for Tier 2 eligibility and8% (2.5 percentage points) for eligibility in either Tier 1 or Tier 2 universities.Next, we evaluate two potential mechanisms that are consistent with theobserved gender gap in performance on high-stakes exams The first possi-bility is that, relative to females, males may take the mock examination lessseriously and only put in their full effort in the Gaokao when the stakes ac-tually matter (see, for example, Attali, Neeman and Scholsser, 2011) Thesecond possibility is that males and females respond differently to competitiveand stressful environments such as the Gaokao We provide several pieces ofevidence, both empirical and qualitative, that appear to support the lattermechanism

If male students take the mock examination less seriously than female dents, this could generate the patterns of relative female underperformance inhigh-stakes settings that we observe in the data, for reasons that are poten-tially unrelated to female performance under pressure While we are unable toentirely rule out the possibility that the observed empirical patterns are driven

stu-by gender differences in effort provision given that we lack direct information

on the effort exerted on the test (e.g time taken to complete the tion, number of questions attempted), we provide some evidence that suggeststhat our results are unlikely to be due to lower effort by males in low-stakessettings.

examina-First, it is worth noting that while the mock exam scores have no ing on the scores used by students to apply for university, the mock exam isthe only time before the Gaokao that students participate in a province-levelexamination Therefore, this is the only time that students can get a realsense of their relative academic standing within the province and the types

bear-of universities that they are likely to qualify for This is important as whatultimately matters for admission into different universities is a student’s per-formance relative to his/her peers at the province-level.17 As such, studentshave an incentive to put in their full effort on the mock exam

To examine this issue empirically, we examine whether the gender gap

in performance varies by students’ ability One might expect that if there areindeed gender differences in effort provision that low ability males might be thegroup that is most likely to reduce their full effort on the mock exam If this istrue, this would suggest that we would observe larger gender performance gapsamong low ability students To examine this possibility, we divide the sample

17

Prior to the provincial mock exam, students have the opportunity to take many practice exams, but these are typically at the school level.

into three groups based on students’ overall performance on the mock exam.Table 1.4 reports similar estimates as Table 1.3 for students in the bottom third(Column (1)), middle third (Column (2)) and upper third (Column 3)) of theoverall test score distribution on the mock exam We find that, compared tomales, females appear to underperform on the Gaokao relative to the mockexam in all three groups, with the relative performance declines more apparentamong higher ability students For example, among students in the bottomthird based on mock exam scores, females performed -0.09 standard deviationsworse on the Gaokao relative to the mock examination compared to males.Among students in the upper third, the female performance decline was about50% larger at -0.13 standard deviations These results indicate that the relativeunderperformance of females in high-stakes settings is observed for students

of all abilities and tends to be more pronounced for higher ability students.This suggests that the overall patterns are unlikely to be driven by lower effortprovision among male students

[Insert Table 1.4 here]

Table 1.4: Regression Estimates of the Gender Gap in Performance by InitialPerformance on Mock Exam

Next, we exploit the fact that the mock examination scores are used tocalculate reference entry cutoffs for different university tiers to generate addi-tional variation in the “stakes” involved in the examinations In particular,

when the Department of Education in Fujian province releases the mock amination scores after the exam, they provide a list of the province-level testscore distribution (in 10 point bins) as well as a set of reference entry cutoffsfor entry into each of the four university tiers that are calculated based on theproportion of students eligible for each tier in the previous year Table 1.5 liststhe reference cutoffs for each stream in 2008 Table 1.6 and 1.7 report the frac-tion of students scoring above each of the reference cutoffs in Fujian provinceand Anxi county, respectively While fewer students in Anxi are projected toqualify for Tier 1 universities relative to province-wide statistics, the fraction

ex-of students in Anxi who score above the reference cutex-offs for Tier 1 and Tier

2 universities is similar to the fraction for the province as a whole AppendixTable 1.2 lists the cutoffs for the Gaokao in 2007 and 2008 and the fraction

of students in Fujian scoring above each of the reported cutoffs Importantly,the fraction of students projected to be eligible for each of the different uni-versity tiers based on the reference cutoffs in the mock exam appears to bevery similar to the fraction of students who were eligible for each tier based

on the actual cutoffs in the 2007 and 2008 Gaokao.18 This suggests that the

18

Note that the slight discrepancy in the fraction of students eligible for each Tier based

on the reference cutoffs in the 2008 mock exam compared to the 2007 Gaokao cutoffs (see Appendix Table 1.4) is likely to be due to the fact that the distribution of mock exam scores are reported in 10 point bins, hence the reference cutoffs are rounded to the nearest ten The reference cutoffs chosen are in fact the mock exam score bins that most closely generate the same fractions of students eligible for each university tier as the 2007 Gaokao cutoffs.

reference cutoffs are indeed informative about the types of universities thatstudents are likely to qualify for.

[Insert Table 1.5 here]

Table 1.5: Mock Exam Reference Cut-offs for each University Tier In 2008

[Insert Table 1.6 here]

Table 1.6: Fraction of Students in Fujian Scoring above each of the ReferenceCut-offs in the Mock Exam in 2008

[Insert Table 1.7 here]

Table 1.7: Fraction of Students in Anxi Scoring above each of the ReferenceCut-offs in the Mock Exam in 2008

We focus on the reference cutoffs for the top two university tiers (Tier

1 and Tier 2) as eligibility for these tiers is more selective - only about 28% of test-takers in Anxi or Fujian are projected to be eligible for entryinto Tier 1 and Tier 2 universities.19 As such, students who are close to thereference entry thresholds for Tier 1 and Tier 2 universities are more likely

22-to face higher pressure on the Gaokao relative 22-to students who are furtheraway from these entry thresholds This arises because students who are close

to the entry thresholds are more likely to experience a larger change in theset of universities (in terms of quality and quantity) that they can apply to

19

From Table 1.6, admission into Tier 3 and the technical universities are a lot less itive, with 40 to 50% of students likely eligible to enter at least a Tier 3 or better university and 80-90% of all test-takers eligible to enter at least a technical university.

compet-resulting from a change in relative performance as compared to students whoare further away from the thresholds This would predict that the gendergap in performance is likely to be larger among students who are close to theentry thresholds relative to students who are further from these thresholds.Moreover, finding differential gender gaps in performance as a function ofstudents’ performance on the mock examination would also alleviate concernsthat certain subgroups of students are simply not taking the mock examinationseriously as this interpretation would imply that the reference entry cutoffsshould have little bearing on the gap in students’ relative performance.Table 1.8 reports the estimates of equation (1) for four different subgroups

of students - those with mock examination scores within 3 points, 5 points, 6

to 10 points and 11 to 20 points of the reference cutoffs required for entranceinto the top two university tiers In Panel (A), we find that consistent withthe idea that female students perform more poorly on the Gaokao relative

to the mock examination when they face greater pressure, the gender gap inrelative performance is largest among students within 3 points of the cutoffsand declines monotonically for students further away from the threshold Morespecifically, compared to their male counterparts, female students perform0.32 (0.25) standard deviations worse on the Gaokao relative to the mockexaminations when they are within 3 (5) points of the cutoff The gender

performance gap is less than half as large at 0.14 and 0.10 for students 6

to 10 points and 11 to 20 points from the reference cutoffs The estimates

in Column (5) provide a formal test of significance of the difference acrossstudents within 3 points and students within 11 to 20 points of the referencecutoffs The difference is -0.22 standard derivations with a standard error of0.14 While the relatively large standard errors of our estimates imply that thedifference is not statistically significant at conventional levels, the magnitude

of the difference is economically large.20 These qualitative patterns are similarfor the individual subjects (Panels (B) to (E)) Overall, these results provideadditional evidence in support of the idea that women underperform relative

to men when subject to more competitive pressures

[Insert Table 1.8 here]

Table 1.8: Is the Gender Gap in Performance Larger where it Matters More?

Pressure?

Next, we turn to the possibility that men and women respond differently tocompetitive and stressful environments such as the Gaokao This explanation

is consistent with a large number of biological studies on gender differences

in stress responses (Taylor et al., 2000, Lee and Harley, 2012) In addition,

20

This is likely to be due to the relatively small sample size of students within 3 or 5 points of the reference cutoffs.

according to the “stress and gender” survey conducted by the American chological Society (2010), given similar self-reported stress levels, women aremore likely than men to report that their stress levels are on the rise, and toreport more physical and emotional symptoms of stress This suggests thatthere could be a substantial gender difference when it comes to test anxietyand stress (Kirschbaum, Wust and Hellhammer, 1992) Moreover, gender dif-ferences in stress response could also potentially reconcile why we only observethe widening of the gender gap in performance in response to competition forDay 2 subjects To cope with challenging situations, the human body con-sumes a lot of energy (Sapolsky, 1994) - therefore, the gender difference inthe response to stress could be more pronounced as students are subject toprolonged stress Consistent with this interpretation, we also find the largestgender performance gaps for the combined subject exam which is half an hourlonger than the other subject tests.

Psy-On the basis of this explanation, there are two possible ways that menand women might differ in performance under pressure First, female perfor-mance may be worse in more competitive settings as they are less able to dealwith the challenges associated with high-stakes settings Alternatively, malesmay be better able to rise to the challenge when faced with higher pressureand competition The latter explanation is consistent with the experimental

literature that has shown that men tend to perform better in more tive situations (Gneezy, Niederle and Rustichini, 2003) To separate these twoeffects, we use the additional variation in stakes generated by distance fromthe reference cutoffs to look within gender to examine whether the observedpatterns are driven by a decline in female performance or an improvement inmale performance in the high-stakes setting.

competi-The first three columns of Panel A in Table 1.9 reports the gender ference of the total test scores for female and male students for subgroups ofstudents classified based on the distance from the reference cutoff For al-

dif-l three subgroups, femadif-les typicadif-ldif-ly do worse on the Gaokao redif-lative to themock examination, with the score gap generally decreasing for female studentsfurther from the thresholds Female students score an average for about 4 to 6points lower on the Gaokao relative to the mock exam when they are within 3points of the cutoffs compared to when they are 11 to 20 points from the cut-off Strikingly, this pattern is reversed for male students - while males within

3 points of the threshold score about 1 point higher on the Gaokao relative

to the mock examination, those who are 11 to 20 points from the cutoff scoreclose close to 4 points lower on the Gaokao Columns (4) and (5) report thedifference of the estimates between Columns (3) and (1) without controls andwith the full set of controls used in the previous tables While the within

gender differences are not significant at conventional levels, the female-maledifference of the test score difference between the Gaokao and mock exam forstudents within 3 points and 11 to 20 points of the cutoffs is marginally sig-nificant at the 10% level without controls Adding in the full set of controlsreduces the coefficient and statistical significance, but the qualitative findingsremain largely similar Panel (B) reports similar estimates using standardizedscores (by type) within each students’ own gender distribution.

[Insert Table 1.9 here]

Table 1.9: Do Females Underperform where it Matters More?

In sum, these patterns suggest that the widening of the gender gap inperformance in high-stakes vs low-stakes settings is driven by a combination

of a decline in females’ performance coupled with an improvement in males’performance in high-stakes settings Female and male students appear to react

to high pressure environments quite differently - while female performanceappears to suffer, male students appear to “up” their game when the stakesare higher

In this section, we exploit the fact that the Gaokao is held over multiplesubjects across a two-day period to explore whether males and females react