Competitive Experience and Gender Difference in Risk Preference, Trust Preference and Academic Performance: Evidence from in

April 2018

Abstract

This study examines whether and how competitive experience affects gender differ- ence in the preferences for risk and trust as well as academic performance. By utilizing the provincial differences in college admission rates as an indicator of competitive expe- rience for students, we assess its relationship with gender difference in risk preference, trust preference, and academic performance. We find that females from with lower college admission rates are more risk averse and less trustful, and perform better in more competitive environment, compared with their male counterparts. Our study suggests that observed gender differences may partially reflect the effects of schooling environment rather than inherent gender traits.

Keyword: competition, risk preference, trust, gender difference JEL Classification: D03, J16 1 Introduction

Despite significant educational advancements for women, a substantial gender gap in labor market outcomes, such as women being under-represented in high-paying jobs and high-level occupations, has been commonly observed and extensively recognized (Bertrand, 2011). At the mean time, numerous studies from both behavioral and experimental economics have consistently found that men and women differ in a wide range of economic preferences (Cro- son and Gneezy, 2009). For example, compared with men, women are more risk averse (e.g., Eckel and Grossman, 2008; Dohmen et al., 2011), dislike competitive environments more (Gneezy, Niederle and Rustichini, 2003; Niederle and Vesterlund, 2007; Flory, Leibbrandt, and List, 2014), and are more situationally specific in social preferences (Croson and Gneezy, 2009). In addition, gender differences in economic preferences have been suggested to explain gender differences in labor market outcomes (Buser, Niederle, and Oosterbeek, 2014). Recent studies have investigated the role of socialization in the observed gender differ- ences. Gneezy, Leonard, and List (2009) show that men opt to compete at roughly twice the rate as women in the patriarchal society of Maasai in Tanzania, which is similar to what is observed in standard industrial societies (Niederle and Vesterlund, 2007). By contrast, women often prefer a competitive environment than men in the matrilineal society of Khasi in India. This cultural reversal has been shown to develop through socialization from the young age (Anderson et al., 2013). A similar reversal is also observed for gender difference in spatial ability (Hoffman, Gneezy, and List, 2011). Alesina, Giuliano, and Nunn (2013) examine the historical origins of gender roles, and suggest that societies that traditionally practiced plough agriculture currently exhibit unequal gender norms. Booth and Nolen (2012a, b) observe that gender differences in risk attitude and competitiveness depend on whether the girls have attended a single-gender or mixed-gender school, suggesting the im- portance of schooling environment in shaping gender differences. In this regard, Bertrand (2011) observes,

“is it also the case that women’s attitudes towards, say, risk or other-regarding preferences, have been converging over time towards men’s? This would certainly fit with the view that the gender differences in preferences are not hard-wired but rather a reflection of environmental influences, and warrant more research on the specific changes in the home or schooling environments that might have triggered the convergence in attitudes.”

To further understand the role of socialization, the current study investigates the effects of competitive experience in school in shaping gender differences. In their seminal paper, Nolen- Hoeksema and Girgus (1994) observe that there are no gender differences in depression rates

1 in prepubescent children, but girls and women are about twice as likely to be depressed as boys and men after the age of 15. They propose a hypothesis that compared with males, stress experience leads females to display considerable withdrawal and internalizing behaviors. This hypothesis has been used extensively to explain the observation that females under stress are highly vulnerable to depression while male under stress are more likely to develop alcohol dependence. Following this hypothesis, it is likely that stressful experience leads females to become more risk averse and less trustful than males. To test this hypothesis, we analyze the effects of competitive experience caused by the college entrance examinations in China on gender differences in terms of economic preference and academic performance. The college entrance examination in China is generally regarded as among the most competitive exams in the world. The passing rate is relatively low compared with those of most Western societies. Students compete for slots in China’s nearly 2,000 colleges comprising three different tiers, namely, key universities, regular universities, and technical colleges. The differences among these colleges are based mostly on institutional ranking and duration of programs. Moreover, the national consensus is that studying in a superior university significantly enhances the opportunity to obtain an excellent job in China’s fiercely competitive job market (Li, et al., 2012). Given that the Chinese educational system is typically exam-oriented, college entrance is almost entirely determined by scores in the College Entrance Examination (CEE), commonly known as Gaokao in . Competitiveness and possibly the Chinese culture have led most high schools to dedicate the entire senior year to preparing students for CEE. Students commonly spend several hours studying after returning home from 10 hours in school, as well as foregoing breaks even on weekends. College entrance is definitely the most stressful experience typically shared by high school students in China. We utilize the differences in college admission rates across provinces in China as an em- pirical setup to assess the extent through which competitive experience induced by college entrance affects females’ risk and trust preferences and academic performance, compared with their male counterparts. Essentially, we exploit difference-in-difference (DID) strat- egy, combining gender difference within and cross province difference in terms of college admission rates. Using the Chinese College Students Survey (CCSS), which sur- veyed 8,176 undergraduate students from graduating classes across the different provinces in China, we analyze the effects on risk preference toward gain-oriented and loss-oriented gambles, as well as the trust preference measured by a general trust question. We find that, compared to males, the competitive experience makes females become more risk averse for both gain-oriented and loss-oriented gambles, and less trustful towards people in the society. In addition, we find that competitive experience makes females perform better in those more

2 competitive examines, relative to males. We further conduct several robustness checks to justify the validity of our empirical strategy. Our results are consistent with Nolen-Hoeksema and Girgus (1994) on the effect of stressful life experience on behavior. Namely, competitive and stressful experience in school leads females to become more risk averse and less trustful than males. We further discuss the possibility that competitive experience makes females handle competitive environments better than their male counterparts, thereby leading to better performance of the former. Our study contributes to recent studies on competition and gender gap. In previous experimental studies, Gneezy, Niederle, and Rustichini (2003) show that men were exten- sively more likely to perform better than women in a competitive environment.1 Niederle and Vesterlund (2007) observe that compared to women, men are more likely to opt for a compet- itive environment after controlling for performance, overconfidence, and risk attitude. Flory, Leibbrandt, and List (2015) conduct a natural field experiment on job entry decisions and conclude that women disproportionately shy away from competitive work settings. Buser, Niederle, and Oosterbeek (2014) show that gender differences in competitiveness could par- tially account for gender differences in educational choices. Instead of examining preference toward competition and performance under competition, the current study investigates the effects of competitive experience on gender differences in economic behavior and outcomes. Our study also contributes to the increasing literature regarding the role of culture and socialization on economic preference and outcomes (see i.e., Guiso, Sapienza, and Zingales, 2006; Nunn, 2009 for review). The long-term effects of early childhood environment on later life outcomes have been discussed extensively in human capital literature (Heckman, 2000; Currie, 2001). Moreover, macroeconomic experiences, such as great depression, have been shown to affect preference for risk (Malmendier and Nagel, 2011), preference for corporate financial policies among managers (Malmendier and Nagel, 2011), and preference for redistri- bution (Giuliano and Spilimbergo, 2014). Natural disaster experiences have also been shown to affect both risk preference (Cameron and Shah, 2015) and time preferences (Callen, 2015). Moreover, the willing to trust can be affected by firm-level competition (Francois, Fujiwara, and van Ypersele 2009) and can be traced back to the transatlantic and Indian Ocean slave trades (Nunn and Wantchekon, 2011). In the current study, we show that the competitive experience of adolescents subsequently affects gender differences in economic preference and academic performance. Apart from the aforementioned studies on the role of socialization, recent studies have

1In the real-world settings, Lavy (2013) and Paserman (2010) evaluate gender differences in the perfor- mance of high school teachers and professional tennis players, respectively, and find minimal evidence to support the view that women turn in inferior performance in highly competitive settings.

3 also explored the roles of biological factors in gender differences. In particular, biological differences between men and women, including features related with reproduction, such as hormonal systems, affect the physical, psychological, and behavioral characteristics between genders. For example, the menstrual cycle has been observed to cause absence from work among women, thereby leading to gender wage gap (Ichino and Moretti, 2009), gender difference in bidding (Chen, Katuˇsˇc´ak,and Ozdenoren, 2013; Pearson and Schipper, 2013), and competitiveness (Buser, 2012). The availability of oral contraceptive pills increases human capital investment and labor force participation for women (Goldin and Katz, 2002). Testosterone, which drives aggression particularly among males, has also been linked to economic preferences (for a review, see Apicella, Carr´e,and Dreber, 2014). These studies generally suggest that despite the possible role of biological factors, socialization can trump these influences in shaping gender differences. Our study extends the literature focusing on China. Zhang (forthcoming) finds that ex- ogenously imposed marriage reforms in China increase female competitive inclination but do not eliminate the gender gap. Relatedly, Cameron et al (2013) show that China’s One-Child Policy (OCP) has produced significantly less trusting, less trustworthy, more risk-averse, less competitive, more pessimistic, and less conscientious individuals. Through an experiment conduced in China, Cassar, Wordofa and Zhang (2016) find that gender gap in competitive- ness disappears once incentives are switched from monetary to child-benefiting. Booth et al. (forthcoming) find that exposed to dramatic changes in socio-economic institutions in China after 1949, women are more competitive compared with their male counterparts as well as Taipei women. Cai et al (forthcoming) show that females tend to underperform dur- ing the in China, compared to male counterparts. Our paper extends this line of research by linking experiencing college entrance examination, which is an im- portant experience for many people in China, with gender gap in risk and trust preferences and academic performance. The rest of this paper is divided into the following sections. Section 2 provides background knowledge on college admission in China. Section 3 describes the data used in this study. Section 4 details the empirical strategy we used. Section 5 presents the empirical results. Section 6 concludes this paper.

2 Background: College Admission in China

In order to get into colleges in China, high school graduates need to take a national stan- dardized examination, called the CEE (or Gaokao).2 The total score students get in the CEE

2See Davey et al. (2007) for a detailed description of the CEE system in China.

4 is the most important factor determining college admission.3 Attending colleges especially elite colleges can bring high returns in China (Zhang, et al., 2005; Li, et al., 2012), therefore students begin their preparations for the CEE as early as junior high school or primary school in order to get high scores in the CEE. Because students treat performing well in the CEE as the only goal and they try all their best to obtain high scores, the CEE scores are considered as good measures of student ability and intelligence in China.4 The allocation of college admission quota among provinces is centralized. The Ministry of Education decides how to allocate the admission quota. However, the criteria or the mechanism used by the Ministry of Education is never publicly available. We run a regression to link provincial admission rates with provincial variables (results are shown in Appendix Table A2) and find that the number of high school graduates is significantly correlated with admission rates. Besides, some sporadic information also indicates that the number of high school graduates is an important determinant of admission rates.5 In the Chinese college admission, colleges are categorized into different tiers; those in the high tier have the first priority to admit students. Students rank their college preferences (4 to 6 in each tier) and majors in an application form. Students in different provinces can submit their application form in different timing, prior to the exam, after the exam but prior to knowing the scores, or after knowing the scores. The prevailing admission procedure in most provinces is similar to the Boston mechanism. In the first round, students are only considered by colleges which are listed as their first choice. Colleges only accept students with a total CEE score above the threshold score. Applicants with a score below the threshold are rejected by their first choice and placed in a pool for the college next on their list of options. A college only considers admitting students who do not list it as the first choice if there are still available slots after the first round of admission. Given the limited number of colleges, students usually have few chances to be admitted by their second-choice colleges if the first-choice colleges do not admit them. The selection process is terminated after a student is admitted, which indicates that a student can obtain only one admission offer. As a final step of the college admission, a student decides whether to attend the college admitting him/her or not. Because each student can obtain only one admission offer, the student will not go to any college that year if he/she turns down an offer.6 Although the number of education institutions, as well as new students enrolled,

3Applicants to some special programs, such as art departments (audition), military and police schools (political screening and physical exam) and several sports programs (tryout), are screened using additional criteria. 4Li, et al.(2012) use the CEE score as a proxy for students’ ability. 5e.g., see the news report: http://www.moe.gov.cn/jyb xwfb/s271/201705/t20170510 304227.html, http://news.xinhuanet.com/yuqing/2016-05/15/c 128983170.htm. 6See Chen and Kesten (2015) for detailed analysis for Chinese college admissions.

5 has substantially increased since 1978, a significant number of young people still cannot enroll in higher education institutions. In addition, although the admission rate is increasing every year (from 4.8% in 1977, to 37% in 1995, to approximately 75% in 2012),7 such percentage remains fairly low compared to those in Western countries and cannot meet the demands of students. More importantly, enrolling in leading universities would generally increase the opportunity of success in both the labor and marriage markets. By contrast, less than 10% of candidates can be enrolled in tier-1 universities, and less than 0.2% of Gaokao takers will be accepted in China’s top five universities based on a report by the Economist. The importance and competitiveness of college entrance examination exert enormous pressure on high school students, as well as on their parents and teachers. For example, the China Daily reported that several girls took contraceptives or received injections to prevent the onset of their menstrual cycle during the week of the exam, while a few students studied in a hospital while hooked up to oxygen tanks in hopes of improving their concentration. Therefore, the highly competitive nature of college entrance examinations in China provides an ideal setup to study the effects of competitive experience on gender difference.

3 Data

The data that we use are derived from the second round of the Chinese College Students Survey (CCSS), which was conducted by the China Data Center of in May and June 2011. This survey was conducted from 2010 to 2014. The survey used the method of stratified random sampling (with locations (Beijing, , , North- eastern China, Eastern China, Central China, and Western China) and different tiers of colleges as stratifying variables) to sample colleges.8 In the first round of the survey, 19 out of 2,305 colleges in China were randomly selected as a pretest; the number was expanded to 50 colleges in 2011 and 2012, and 65 colleges in 2013 and 2014. In each college, approximately 300 students from the population of graduating classes were randomly sampled. The questionnaire was designed collaboratively by experts in economics, sociology, and education. In the questionnaire, basic information, such as individual characteristics and family backgrounds, is collected. The questionnaire also collects information on CEE scores, college activities, and student placement after graduation. The survey administrators were trained in Beijing for several days through intensive meetings before the survey was con- ducted. The survey administrators in each collage asked the sampled students to complete

7Statistical Yearbook, 2013 8In the sampling process, three metropolises (i.e., Beijing, Shanghai, and Tianjin) were separated from the rest of China because colleges, particularly the top universities, are concentrated on these cities.

6 the questionnaires. They collected the questionnaire forms and placed the forms inside coded envelopes to guarantee anonymity. The survey team closely monitored the whole survey pro- cess to make sure that the survey was conducted carefully. In this study, we focus on data collected in 2011 because it is the only year when the questions on students’ preferences were included in the questionnaire. The sampled colleges are located in 24 provinces. A total of 8,176 undergraduate students from graduating classes from all 31 provinces in China were selected. Table 1 shows several pre-college characteristics of these students. A total of 45% are female. Their average CEE score is 0.226.9 On average, 25% of their parents have college degrees, 13% are cadres, and 28% work in the public sector. A total of 43% entered magnet classes in high school. Generally, these students performed well in high school. A total of 16% won awards in several city-level or above competitions, and 75% had GPAs in the top 20% level in high school.

[Insert Table 1 Here]

Our main independent variable is competitiveness experience, which is measured by the provincial college admission rate in 2007 when the college graduates in 2011 took the CEE. College admission rate is calculated as the ratio of college admission to the number of students taking the CEE. We obtain the data from three sources. Firstly, we obtain college admission rates of 19 provinces from China Education Examinations Yearbook (2008).10 Then, we obtain college admission rates of 9 provinces from the China Gaokao Information website.11 For the remaining three provinces (, , and ), we search the news report to obtain college admission rates.12 Table A1 lists the detailed information. Table 1 shows that the average value of the competition measurement, that is, one minus college admission rate, is 0.374; its variation across different provinces is large and the standard deviation is 0.121. We investigate two sets of outcome variables in this study. The first set is on economic preferences, including preference toward risk and trust. We measure risk preferences over gains and losses (Kahneman and Tversky, 1979). Risk preference over gain-oriented gamble

9The CEE scores are normalized using mean and standard deviation in the province where the student took the examination. 10The yearbook only includes the information for 19 provinces: Shanghai, , Beijing, , , , , , , , , , , , Tianjin, , , , and Yunan. 11http://gaokao.chsi.com.cn/gkxx/zhiding/200709/20070905/1113417.html. These provinces include , , , , , , Neimenggu, , and . 12We get college admission rates for Hubei, Heilongjiang and Ningxia from http://gaokao.eol.cn/dongtai 5640/20070618/t20070618 238318.shtml, http://zqb.cyol.com/content/2007- 05/15/content 1759511.htm, and http://edu.qq.com/a/20070620/000158.htm,respectively.

7 is measured by the following question: Would you rather (a) gain 1,000 Yuan with certainty, (b) gain 2,000 Yuan with 50%, and nothing with 50%? Option 1 (a), option 2 (b), and option 3 (a) and (b) are indifferent. Risk preference over gain-oriented gamble indexed with “Risk (gain)” is equal to 1 if the answer is option 1; 2 if the answer is option 3; and 3 if the answer is option 2, which indicate risk aversion, risk neutrality, and risk affinity, respectively. Risk preference over loss-oriented gamble is measured by the following question: Would you rather (a) lose 1,000 Yuan with certainty, (b) lose 2,000 Yuan with 50%, and no loss with 50%? Option 1 (a), option 2 (b), option 3 (a) and (b) are indifferent. Correspondingly, the variable “Risk (loss)” is equal to 1 if the answer is option 1; 2 if the answer is option 3; and 3 if the answer is option 2. Similar hypothetical questions are used in Kahneman and Tversky (1979), as they argue “the method of hypothetical choices emerges as the simplest procedure by which a large number of theoretical questions can be investigated”. Recently, Dohmen et al (2011) compare various measurements of risk preference and their economic consequences, and observe that hypothetical lottery questions can predict actual risk taking behavior such as financial investment. Table 1 shows that the average value for “Risk (gain)” is 1.895 and the average value for “Risk (loss)” is 2.227. These results suggest that the students are risk averse over gain-oriented gamble and risk seeking over loss-oriented gamble, thereby demonstrating consistency with experimental observations using real incentives. To measure trust preference, students are asked “Do you agree with the following state- ment: ‘Generally, you can trust people in society.’ Option 1, Strongly Disagree; option 2, Disagree; option 3, Agree; and option 4, Strongly Agree.” This question is known as the general trust question and has been used extensively to measure the willingness to trust, particularly in the General Social Survey and World Values Survey, which has been shown to be predictive for actual economic outcomes such as financial market participation (Guiso,Sapienza, and Zingales, 2008; 2013). Moreover, Francois, Fujiwara, and van Ypersele (2009) show that firm-level competition is shown to have a positive impact on trust at the individual level using the World Values Survey. Table 1 shows that the average value of trust is 2.154, suggesting that students on average tend to disagree that they can trust people in society. The second set of variables is on the academic performance of students in college, includ- ing their GPA; an indicator of whether the student had failed in any course; an indicator of whether the student had won any awards in the provincial level or above; and their scores in the College English Level 4 Test (CET4), a national examination on English proficiency for college students. Table 1 shows the summary statistics of these variables. The average GPA in college is approximately 3 out of 4. A total of 41% of the students had failed in at least one course. Meanwhile, 14% had won awards in the provincial level or above. Their

8 average CET4 score is 467 (out of 710).13

4 Estimation Strategy

To identify the effects of competitive experience (generated by the college entrance) on indi- viduals’ preferences and performance, we compare individuals across provinces with different college admission rates. Two empirical challenges are present; that is, sample selection issue and heterogeneity across provinces. As an illustration, start with a case that there is a random assignment of college admis- sion rates across provinces. The randomization ensures that the college admission rate is not correlated with any other provincial characteristics. However, the different admission rates across provinces could generate a sample selection issue; that is, our sample consists of students admitted to colleges and different admission rates select different quantiles of students even from the same population. For example, provinces with lower admission rates, compared with those with higher admission rates, could have admitted exceptional students. Consequently, the differences in individuals’ preferences and performance across provinces may not be caused by competitive experience but simply reflect the selection caused by the different admission rates. A unique feature of our research setting is that the selection of college admission is primarily based on the CEE score. This one-dimensional selection rule enables us to avoid the sample selection issue by comparing individuals with the same CEE scores across provinces. This strategy resembles the approach used by Lee (2009) in the setting of randomized experiment. As Lee (2009) does not know the exact selection rule in his research setting, he examines the worst and the best selection scenarios to pin down the bounds on treatment effects. In the reality, however, the distribution of provincial college admission rates in 2007, our regressor of interest to determine the different competitive pressures experienced by different individuals, is unlikely to be random. To overcome this identification issue, we focus on the competitive experience effect on gender difference in preferences and performance.14 This DID style strategy enables us to include province-fixed effects in the regressions, which effectively controls for all differences across provinces in the cross-sectional data such as preferences over gender, economic development stage, geographic features, etc.

13A score of 425 is necessary to pass this test. 14Azmat, Calsamiglia, and Iriberri (2016) investigate the gender difference in response to increased in- centives using the similar strategy, comparing gender difference of test scores in tests with different stakes. Cai, et al (forthcoming) study how female and male examination performance are differentially affected by examinations with different competitive pressure by comparing difference in test scores in mock examination (lower pressure) and college entrance examination (higher pressure) between male and female students.

9 Our regression specification is then

yip = βCompetitionp · F emaleip + αF emaleip + CEEip + λp + εip, (1)

where yip are the preferences and performance measures of individual i taking the CEE in province p; F emaleip equals 1 if individual i is a female and 0 otherwise; CEEip is individual i’s CEE score; λp is a set of province fixed effects, capturing all differences across provinces such as preferences over gender, economic development stage, geographic features, etc.; and

εip is the error term. We calculate the standard errors by clustering over the province level to deal with any potential heteroskadasticity problem.

Competitionp is our key regressor, which measures the college admission rate in province p. Specifically, it is measured as 1-college admission rate; hence, the higher value is Competitionp, the more competitive is the college admission. In the baseline analysis, we use the rate in 2007, the year when our concerned individuals took the CEE.15

4.1 Threats to the Identification and Balancing Tests

The identifying assumption underlying our estimation framework (1) is that there is no gender differential response to other provincial characteristics, conditional on province fixed effects and CEE scores. In other words, conditional on province fixed effects and CEE scores, gender differentials are balanced across provinces with different college admission rates before taking CEE. Hence, ex post different gender differentials in preferences and performance across provinces can then be attributed to different degrees of competitive pressures experienced. To check whether this balancing has been achieved in our estimation framework, it is first important to understand what determined the different college admission rates across provinces, and then we can condition out these confounding factors. To this end, we regress

Competitionp on various provincial characteristics in Appendix Table A2, including pop- ulation, GDP per capita, industry output composition, average labor income, government income, trade, educational attainments, ratio of government expenditure on education, num- bers of colleges and college students, and numbers of high schools and high school students. Except for the number of high school graduates, all these potential determinants are highly 16 insignificant. In robustness checks, we include the interaction between F emaleip and the

15To capture the possibility that students may face the competitive pressure of the CEE during the whole senior high school period, we also experiment with the average college admission rate from 2006 to 2007. It is better to use the average college admission rate from 2005 to 2007 since students spend three years in high schools in China; however, provincial college admission rates in 2005 are not available. 16One might be interested to see the insignificant coefficient of GDP per capita which is not consistent

10 number of high school graduates as an additional control to insolate the competitive expe- rience effects. To lend further support to our identifying assumption, we use the micro-level data from the 2005 Chinese population census to check whether there is no gender differential response to other provincial characteristics across provinces with different college admission rates two years before the CEE used in our setting. Specifically, we examine ethnicity, years of school- ings, marriage status, percentage of migrants, household registration status, health condition, and income. Estimation results are reported in Table 2. Through all the specifications, we consistently do not find any statistically and economically significant estimates, implying that gender differentials are balanced across provinces with different college admission rates two years before taking the CEE.17

[Insert Table 2 Here]

Furthermore, we conduct analyses following the idea proposed by Lee and Lemieux (2011).18 First, we check whether the gender difference in pre-determined socioeconomic characteristics and density in college entrance exam scores is similar across provinces. Sec- ond, we examine whether gender difference in individual performance in their middle school periods when there is no CEE is similar across provinces. No significant gender difference in pre-determined characteristics and performance just a few years before the CEE may in- dicate no significant gender different responses to other provincial characteristics, justifying our empirical strategy. Table 3 presents the regression results. In column 1, we consider whether an individual

is a minority, and the coefficient of Competitionp · F emaleip is statistically insignificant and small in magnitude, suggesting that the gender difference in ethnicity are balanced across provinces. In columns 2-4, we examine the balancing of parental characteristics—specifically, whether they have college degrees, whether they are government cadres, and whether they work in public sector. Consistently, we find that our regressor of interest is statistically insignificant and has small magnitudes, suggesting that there is no selection by students’

with the observation that it is easier to get into colleges in cities with higher GDP per capita like Beijing or Shanghai. One potential reason is that these cities might also have more high school graduates, which leads to higher college admission rates in these cities. 17One caveat we need to bear in mind is that the balanced gender difference in demographic and household characteristics across provinces could still not rule out that gender difference in risk attitudes and trust might differ across provinces. Our check can be considered as checking necessary conditions. 18Lee and Lemieux (2011) argue that in the regression discontinuity setting, testing whether the pre- determined variables are balanced at the threshold is a way to check the validity of local randomness.

11 parents such as moving into better provinces.

[Insert Table 3 Here]

In columns 5-7, we examine whether there were any selections due to the admission into middle schools. Specifically, we look at during the middle school period, whether individuals were enrolled in magnet classes, whether they won any competitions at the province level or above, and whether their GPAs were at the top 20% level. We continue to find no statistically and economically significant gender differences across provinces in all these outcomes. These results indicate that there may not be significant selection due to the high school admission. We then check the balancing among individuals’ performance during their senior high school to further examine the selection due to the high school admission and whether there were any changes caused by the competitive experience during the high school entrance exam. Towards this end, we look at during the high school period, whether individuals were en- rolled in magnet classes, whether they won any competitions at the province level or above, and whether their GPAs were at the top 20% level. As shown in columns 8-10, there is no statistically and economically significant gender difference across provinces in all these outcomes. These results suggest that genders were quite balanced during the high school period. Lastly, we check whether there is any discontinuity in the density function of college entrance exam scores across provinces. McCrary (2008) argues that a mixture of discontinu- ities in individual characteristics resulted from selection would further imply a discontinuity in the density distribution. Following this argument, we check and find that there is no gender difference in the number of individuals obtaining same CEE scores across provinces in column 11. In summary, these analyses largely demonstrate that there is no significant selection and genders were balanced across provinces before the CEE, which implies that our research design is generally valid.

5 Empirical Findings

5.1 Main Results

Preferences. We start with presenting the results on economic preferences in Table 4. We use the ordered Logit model to capture the categorical nature of our outcome variables. However, the results from the linear probability model are qualitatively similar and available upon request. [Insert Table 4 Here]

12 In columns 1-2, we consider the effects on risk preference—specifically, two measures so- licited from questions on losing and gaining money, respectively. In a world without any competitive experiences, females are similar as males in their risk seeking towards losing

and gaining money. The coefficients of Competitionp · F emaleip are both negative and significant. These results suggest that the competitive experience enlarges the gender gap in risk preference. Specifically, relative to males, females become more risk averse towards both gain-oriented and loss-oriented gambles after they experience competition in the college entrance. In column 3, we investigate the effects of competitive experience on preference to- wards trust. The coefficient of F emale is positive and statistically significant, implying that in a world with 100 percent college admission (hence free of competitive experience), females are more willing to trust than males. Regarding our central interest, we find the coefficient of Competitionp · F emaleip to be negative and statistically significant. This suggests that the competitive experience makes the gender gap in trust to become considerably narrow. In particular, at the admission rate of 48 percent, females and males become equally trustful.19

Performance. In Table 5, we examine a number of performance indicators by individuals during their four-year college period. Specifically, we look at the GPA (in log), whether an individual had failed any courses, whether an individual had won any awards at the province level or above, and the CET4 score.

[Insert Table 5 Here]

We consistently find that in a world without any competitive experiences, females perform better than males in the colleges—specifically, they have 9.2 percent higher GPAs and 29.1 percent less likely to fail courses. However, we do not find that competitive experience reduces the gender gap in GPAs, likelihood of failing courses and CET4 scores, but the gender gap in winning provincial or national awards is enlarged by the competitive experience. Given that the former two outcomes pertain to competition within a college and CET4 is an English exam, it is much more competitive to win provincial or national awards. These results suggest that the competitive experience improves females’ performance in more competitive environment relative to males.

Summary. We find that competitive experience causes females to become more risk averse, less trustful, and have better performance in more competitive environments, com-

19As trust preference is often linked to risk preference, the observed effect on trust preference could be due to the effect on risk preference. To examine this possibility, we include risk preference in the analysis, and we find that the magnitude of coefficient of Competitionp ·F emaleip remains the same (results available upon request).

13 pared with male counterparts. In the next subsection, we first present several robustness checks to substantiate our findings. Thereafter, we provide an explanation along the psy- chology literature in Section 5.3.

5.2 Robustness Checks

We conduct a battery of robustness checks on our aforementioned results in this subsection. All regression results are contained in the Appendix.

Inclusion of pre-determined characteristics. As the first check, we follow the suggestion by Lee and Lemieux (2010) by including pre-determined characteristics: if our research design is valid, including these controls should have little effect on our estimates. Regression results are reported in Tables A3 and A4 for economic preference and academic performance, respectively. Indeed, we find our estimates barely change in both statistical significance and magnitudes, further implying the validity of our research design.

Inclusion of provincial variables. To condition out the effects from other provincial char- acteristics which determines college admission rates (i.e., the number of high school grad- uates, as shown in Table A2), we further control for the interaction of the number of high school graduates and the female dummy. To capture the impacts of provincial population size, we scale the number of high school graduates by population.20 In addition, Wei and Zhang(2011) show that sex ratio improves marriage competition, which could affect parents’ behavior and therefore their children’s behaviors. To address this concern, we also control for the interaction of sex ratio and the female dummy.21 Estimation results are reported in Tables A5 and A6. Our estimates remain robust to these additional control variables, sug- gesting that they are not biased due to the gender differential impacts from other provincial characteristics.

Inclusion of the interaction of college entrance examination scores with female dummy.One might be concerned that the CEE scores in more competitive provinces are likely to be higher. If women with higher CEE scores are more risk averse, less trusting and perform better aca- demically than men, then our estimates could be biased. To address this concern, we conduce a robustness check by including an interaction of the female dummy and CEE scores in the regression. The results are shown in Tables A7 and A8. The coefficient of the female dummy and competition remains robust, and the coefficient of the interaction of the female dummy

20We would like to thank one referee for pointing out this point. 21Provincial sex ratio is measured in 1997.

14 and CEE scores is mostly not significant. It suggests that our findings are not driven by different effects of CEE scores on women relative to men.

Inclusion of college dummies. Our sample comes from various colleges, which may have different curricula and exam standards. To address this potential incompatibility issue, we further add a set of college dummies in our analysis. However, a compromise is that as many provinces allow students to select colleges after the CEE, competitive experience may lead females and males to select colleges differently; hence, comparing individuals in the same colleges may overlook this competitive experience effect. Regression results are reported in Tables A9 and A10. Similar patterns on gender differences in preferences and performance caused by the competitive experience are also uncovered.

Placebo test. We conduct a placebo test to verify further whether our research design is valid. Specifically, instead of using the real distribution of college admission rates across provinces to measure the competitive experiences across provinces, we randomly assign these false admission rates across provinces, and construct a false variable, Competitionp . We then false replace Competitionp with Competitionp in regression (1). Given this random data gen- false erating process, we expect Competitionp · femaleip to cast zero effects on our outcomes; otherwise, it indicates some mis-specification of our regression. We conduct this exercise 500 times to avoid rare events and increase the power of the test. The mean, standard deviation, and p-value are reported in Table A11. We find that all the mean coefficients false of Competitionp · femaleip are statistically insignificant, lending further support to our research design.

Competitive experience in the high school.To incorporate the possibility that students may face competitive pressures since they were enrolled in high schools, we use the average competition between 2006 and 2007 as the measure of the competitive experience.22 Regres- sion results are reported in Tables A12 and A13. We continue to find the similar patterns of competitive experience effects on gender differences in preference and performance.

Competition measurement. As discussed in data section, we obtain college admission rates for Hubei,Ningxia and Heilongjiang by searching news reports. One might concern about the accuracy of the college admission rates of these three provinces. To check whether it has any impacts on our estimates, we drop these three provinces and re-estimate the impacts. Results are shown in Tables A14 and A15. We can still see the similar impacts of competitive experience on gender differences in preferences and performance.

22It is better to use the average competition between 2005 and 2007 but the competition in 2005 is not available.

15 5.3 Interpretation

We find that competitive and stressful experiences cause females to become less willing to trust and more risk averse, compared with males. To explore potential interpretations, we discuss the literature on gender difference in stress response. In biology and psychology literature, the dominant model of response to stress, originally proposed by Walter Cannon in the 1920s, is characterized by a fight-or-flight response, that is, either confronting a stressor (fight) or fleeing from it (flight). While both men and women show the fight-or-flight pattern of arousal, such as, elevated heart rate and blood pressure, Taylor et al (2000) observe that females are more likely than males to seek support from others in times of stress. They further propose an alternative model of stress response for women, dubbed Tend-and-befriend, in which tending involves nurturing activities designed to protect the self and offspring that promote safety and reduce distress and befriending is the creation and maintenance of social networks that may aid in this process. Men and women differ not only in how they respond to the epoch of stress, but also in how chronic stress experience impacts them in longer term. In particular, stressful life events, many of which are novel and challenging, have been shown to account for numer- ous medical and psychiatric conditions such as anxiety, depression, and alcohol dependence. Nolen-Hoeksema and Girgus (1994) hypothesize that, after experiencing chronic stress, fe- males are more likely to report internalizing symptoms, while males are more likely to report externalizing symptoms. This hypothesis has been proposed to account for the extensively observed gender difference in psychiatric disorders. That is, across different countries and different ethnic groups, the rates of depression are 2 to 3 times higher among women than men, while alcohol dependence is approximately twice as common in males compared to females (Hasin et al., 2007). Maciejewski, Prigerson, and Mazure (2001) find that, among males and females who did not experience a stressful life event, no gender difference is ob- served among those suffering from major depression. By contrast, females who had been exposed to a stressful life event have a threefold increase in risk for major depression rel- ative to males exposed to a stressful life event. Using data from the 2001-2002 National Epidemiologic Survey on Alcohol and Related Conditions, Dawson et al. (2005) find that the association between the number of stressful life events and alcohol consumption is more pronounced among males than females, and that particular stressful life events differentially impacted the level of alcohol consumption among males. These studies support the hy- pothesis of Nolen-Hoeksema and Girgus (1994) on gender difference in response to stressful experiences. We hypothesize that the pressure of college entrance is the most stressful life event that high school students commonly encounter. On the basis of the hypothesis proposed by Nolen-

16 Hoeksema and Girgus (1994), if stress experience leads females to display more withdrawal and internalizing behaviors, it is reasonable to hypothesize that females experiencing more stressful and competitive environment would become less willing to take risk and less willing to trust others. Conversely, if stress experience makes males to exhibit more externalizing behaviors, it is reasonable to hypothesize that males experiencing more stressful and com- petitive environment would become more willing to take risk, and more willing to trust. Thus, our results are consistent with psychological literature on gender difference in stress response. Moreover, our study goes beyond simple correlation, and identifies the casual role of stress experience on gender difference in economic preference. We also find that competitive experience causes females to perform better in more com- petitive environments, compared with male counterparts. One possible explanation is that the gender differential effect on academic preference is partially due to the gender differential effect on economic preference. This argument follows Buser, Niederle and Oosterbeek (2014), in which they find that competitiveness measured in the laboratory setting is positively cor- related with the selection of academic tracks, and the gender difference in competitiveness accounts for a substantial portion of the gender difference in track choice. An alternative explanation is that the competitive experience provides a learning opportunity to handle the subsequent competitive situations and males and females learn differently. For exam- ple, Wozniak, Harbaugh and Mayr (2014) show that feedback about relative performance removes the average gender difference in compensation choices, which suggests that learning may affect males and females differently. Although we could not explore these two possible explanations completely because of data limitations, we attempt to assess the first possi- bility. In particular, we include risk and trust preferences (and their interactions with the female dummy) in the analyses of academic performance presented in Table 5, and examine whether the magnitude of coefficients of Competitionp · F emaleip for academic performance remain the same. We find that the inclusion of preferences does not alter the magnitude for the effect on performance. (Appendix Table A16). This suggests that the differential effect on academic performance is unlikely to be caused by the differential effect on risk and trust preferences.

5.4 Heterogeneous Effects

In this section, we investigate whether and how the effects of competitive experience on gen- der difference in economic preference and academic performance may differ across various groups. We estimate a specification similar with Equation (1) but controlling for parental characteristics, such as indictors for minority, having a college degree, being a government

17 cadre and working in public sector, and pre-college variables measuring students’ perfor- mance in high schools and middle schools. Due to the space limit, Table 6 only presents the estimated coefficients of the interaction of competition and the female dummy.

[Insert Table 6 Here]

First, we compare the heterogeneous effect between students attending magnet high schools and regular high schools, as these two types of schools may have different competitive schooling environments. Columns 1 and 2 are for students attending magnet and regular high schools, respectively. Experiencing the same level of competition, female students attending magnet high schools become more risk averse and less trustful relative to their male counterparts than female students attending regular high schools. However, the effect on gender difference in terms of college performance is relatively the same for students attending magnet and regular high schools. Second, high schools in China offer the track, liberal arts and track, and other more minor tracks. The natural science track is more popular among males while the arts and social science track is more popular among females. Columns 3 and 4 show the results for students taking the natural science track and other tracks, respectively. We can see that with the same amount of increase in competition, female students taking other tracks are less trustful, more risk averse in losing money, and less risk averse in gaining money than female students taking the natural science track relative to their male counterparts. The effect of experiencing competition on gender difference of college performance is approximately the same for students taking different tracks. Third, we examine the differences between those who are single child and those who have siblings, as it has been recently shown that the one child policy has produced significantly less trusting, less trustworthy, more risk-averse, less competitive, more pessimistic, and less conscientious individuals (Cameron et al., 2013). As shown in columns 5 and 6, experiencing the same level of competition, female students having siblings become less trustful and risk averse in gaining money, but perform better in college than those who are single child, relative to their male counterparts.

6 Conclusion

This study investigates the extent to which competitive experience affects gender differences in economic preference and academic performance. To address this issue, we use the em- pirical setup of differences in college admission rates across provinces because the pressure

18 of college admission is probably the most competitive experience commonly shared by high school students in China. Our data reveal that the competitive experience induced by col- lege admission makes females, compared to males, become more risk averse toward both gains and losses, less trustful, and perform better in highly competitive environments. This observed behavioral pattern is consistent with gender difference in stress response as pro- posed by Nolen-Hoeksema and Girgus (1994) in psychology literature. After experiencing stress, females are more likely to display withdrawal and internalizing behavior, such as less willing to take risks and to trust other people. By contrast, males are more likely to display approach and externalizing behavior, such as highly willing to take risks and to trust other people. Our study contributes to the policy implications regarding the effect of college expansion, which is a worldwide phenomenon. From a special report on university education published in the Economist on March 28, 2015, the global tertiary admission rate increased from 14% to 32%, and the number of countries with a ratio of more than half increased from five to 54 in the two decades prior to 2012. Our results suggest that college expansion leading to less competitive environment would have implications on gender differences in economic preference, academic performance, and perhaps labor market outcomes. It would be of interest to further examine the effect of college expansion on other economic behavior and preferences, including willingness to compete (Niederle and Vesterlund, 2007; Flory, Leibbrandt, and List, 2015; Buser, Niederle, and Oosterbeek, 2014). Our study also contributes to understanding the relationship between stress and gender. Based on The Stress in America survey (2012) conducted by American Psychological As- sociation, males and females do differ in the experienced stress and employ different stress management techniques. For example, females are likely to report experiencing extreme stress, and are more likely to engage in social and sedentary activities such as reading and shopping to manage stress. A number of recent studies investigate the role of stress hor- mones in competition in the (Buser, Dreber, and Mollerstrom, 2015; Buckert et al., 2015; Halko and S¨a¨aksvuori,2015, Zhong et al., 2015), while they find that gender difference in stress response could not account for gender difference in willingness to compete. Despite its importance, the effects of stressful experience have not yet been much explored in economics. For example, Evans and Kim (2007) find that the number of years spent living in poverty during childhood is associated with elevated overnight cortisol and a more disregulated car- diovascular response, while concurrent poverty, i.e., during adolescence, does not affect these physiological stress outcomes. More recently, Chemin De Laat and Haushofer (2013) ob- serve that low levels of rain in the preceding year increased the level of cortisol, the stress hormone, among farmers in Kenya. Goh, Pfeffer and Zenios (2015) find that stressors in the

19 workplace contribute substantially to healthcare cost and mortality in the United States. Given the well-established gender difference in stress response in psychology and biology (Nolen-Hoeksema and Girgus 1994; Taylor et al 2000), investigating the effects of different sources of stress arising from early childhood poverty, parenting style, and workplace on gender gap in economic preferences and outcomes would be of significant interest. Finally, we would like to acknowledge some limitations of our study in relation to the measurements of college admission rates, risk and trust preferences, and academic perfor- mance. First, while our results are valuable for those highly competitive environments such as eastern Asian countries/regions including South Korea, Japan, , and where similar examinations exist, caution needs to be taken when one concerns the gener- alization to other less completive settings. In this regard, it would be of interest to study related questions in societies with less competitive examination system. Second, risk and trust preferences are measured using hypothetical choice and academic performance is based on self-report of the students, which may bias the results. It would be highly valuable for the future studies to measure risk and trust preferences with actual incentive and measure academic performance using administrative data.

20 References

[1] Alesina, Alberto, Paola Giuliano, and Nathan Nunn. “On the Origins of Gender Roles: Women and the Plough.” The Quarterly Journal of Economics 128, no. 2 (2013): 469- 530.

[2] Andersen, Steffen, Seda Ertac, Uri Gneezy, John A. List, and Sandra Maximiano. “Gen- der, competitiveness, and socialization at a young age: Evidence from a matrilineal and a patriarchal society.” Review of Economics and Statistics 95, no. 4 (2013): 1438-1443.

[3] Apicella, Coren L., Justin M. Carr´e,and Anna Dreber. “Testosterone and Economic Risk Taking: A Review.” Adaptive Human Behavior and Physiology (2014): 1-28.

[4] Azmat, Ghazala, Caterina Calsamiglia, and Nagore Iriberri. 2016. “Gender Difference in Response to Big Stakes.” Journal of the European Economic Association, 14(6): 1372-1400.

[5] Bertrand, Marianne. 2011. “New Perspectives on Gender.” Handbook of Labor Eco- nomics.

[6] Bertrand, Marianne, Emir Kamenica, and Jessica Pan. “Gender Identity and Relative Income within Households.” The Quarterly Journal of Economics 130, no. 2 (2015): 571-614.

[7] Booth, Alison L., and Patrick Nolen. “Gender differences in risk behaviour: does nurture matter?.” The Economic Journal 122, no. 558 (2012): F56-F78.

[8] Booth, Alison, and Patrick Nolen. “Choosing to compete: How different are girls and boys?.” Journal of Economic Behavior & Organization 81, no. 2 (2012): 542-555.

[9] Booth, Alison, Elliott Fan, Xin Meng, and Dandan Zhang. “Gender differences in will- ingness to compete: The role of culture and institutions” Economic Journal, 2017.

[10] Buser, Thomas. “The impact of the menstrual cycle and hormonal contraceptives on competitiveness.” Journal of Economic Behavior & Organization 83, no. 1 (2012): 1-10.

[11] Buser, Thomas, Muriel Niederle, and Hessel Oosterbeek. “Gender, Competitiveness and Career Choices.” Quarterly Journal of Economics 129(3). (2014). 1409-1447.

[12] Buser, T., Dreber, A., and Mollerstrom, J. ”The impact of stress on tournament entry.” Experimental Economics 20.2 (2017): 506-530.

[13] Buckert, M., C. Schwieren, B.M. Kudielka and C.J. Fiebach, 2015. “How stressful are economic competitions in the lab? An investigation with physiological measures.” Work- ing paper.

[14] Cai, Xiqian, Yi Lu, Jessica Pan, and Songfa Zhong, forthcoming. “Gender Gap under Pressure: Performance and Reaction to Shocks.” Review of Economics and Statistics.

21 [15] Callen, Michael. “Catastrophes and time preference: evidence from the Indian Ocean earthquake.” Journal of Economic Behavior & Organization (2015). [16] Cameron, Lisa, and Manisha Shah. “Risk-taking behavior in the wake of natural disas- ters.” Journal of Human Resources 50, no. 2 (2015): 484-515. [17] Cameron, Lisa, Nisvan Erkal, Lata Gangadharan, and Xin Meng. “Little emperors: behavioral impacts of China’s one-child policy.” Science 339, no. 6122 (2013): 953-957. [18] Cassar, Alessandra, Feven Wordofa, and Y. Jane Zhang, 2016. “Competing for the benefit of offspring eliminates the gender gap in competitiveness.” PANS 113, no. 19, 5201-5205. [19] Chemin, Matthieu, Joost De Laat, and Johannes Haushofer. “Negative rainfall shocks increase levels of the stress hormone cortisol among poor farmers in Kenya.” Available at SSRN 2294171 (2013). [20] Chen, Yan, Peter Katuˇsˇc´ak,and Emre Ozdenoren. “Why can’t a woman bid more like a man?.” Games and Economic Behavior 77, no. 1 (2013): 181-213. [21] Chen, Yan, and Onur Kesten. “Chinese college admissions and school choice reforms: Theory and experiments.” Journal of Political Economy, forthcoming [22] Currie, Janet. “Early childhood education programs.” Journal of Economic perspectives (2001): 213-238. [23] Croson, Rachel, and Uri Gneezy. “Gender differences in preferences.” Journal of Eco- nomic literature (2009): 448-474. [24] Davey, Gareth, Chuan De Lian, and Louise Higgins. “The University Entrance Ex- amination System in China.” Journal of Further and Higher Education (2007), 31(4): 385-396. [25] Dawson, Deborah A., Bridget F. Grant, Frederick S. Stinson, Patricia S. Chou, Boji Huang, and W. Ruan. “Recovery from DSM-IV alcohol dependence: United States, 2001–2002.” Addiction 100, no. 3 (2005): 281-292. [26] Dohmen, Thomas, Armin Falk, David Huffman, Uwe Sunde, J¨urgenSchupp, and Gert G. Wagner. “Individual risk attitudes: Measurement, determinants, and behavioral consequences.” Journal of the European Economic Association 9, no. 3 (2011): 522-550. [27] Eckel, Catherine C., and Philip J. Grossman. “Men, women and risk aversion: Experi- mental evidence.” Handbook of experimental economics results 1 (2008): 1061-1073. [28] Evans, Gary W., and Pilyoung Kim. “Childhood poverty and health cumulative risk exposure and stress dysregulation.” Psychological Science 18, no. 11 (2007): 953-957. [29] Flory, Jeffrey A., Andreas Leibbrandt, and John A. List. “Do Competitive Workplaces Deter Female Workers? A Large-Scale Natural Field Experiment on Job-Entry Deci- sions.” The Review of Economic Studies (2014): rdu030.

22 [30] Francois, Patrick, Thomas Fujiwara, and Tanguy van Ypersele. “Competition builds trust.” Working Paper (2010).

[31] Goldin, Claudia, and Lawrence F. Katz. “The Power of the Pill: Oral Contraceptives and Women’s Career and Marriage Decisions.” Journal of Political Economy 110, no. 4 (2002): 730-770.

[32] Goh, Joel, Jeffrey Pfeffer, and Stefanos A. Zenios. “The Relationship Between Work- place Stressors and Mortality and Health Costs in the United States.” Management Science (2015).

[33] Giuliano, Paola, and Antonio Spilimbergo. “Growing up in a Recession.” The Review of Economic Studies 81, no. 2 (2014): 787-817.

[34] Gneezy, Uri, Muriel Niederle and Aldo Rustichini. 2003. “Performance in Competitive Environments: Gender Differences.” Quarterly Journal of Economics. Vol 118: 1049- 1074.

[35] Gneezy, Uri, Kenneth L. Leonard, And John A. List. “Gender Differences In Competi- tion: Evidence From A Matrilineal And A Patriarchal Society.” Econometrica 77, no. 5 (2009): 1637-1664.

[36] Guiso, Luigi, Paola Sapienza, and Luigi Zingales. “Does culture affect economic out- comes?.” The Journal of Economic Perspectives 20, no. 2 (2006): 23-48.

[37] Guiso, Luigi, Paola Sapienza, and Luigi Zingales. “Trusting the stock market.” the Journal of Finance 63, no. 6 (2008): 2557-2600.

[38] Guiso, Luigi, Paola Sapienza, and Luigi Zingales. “Understand trust.” Economic Journal 123, no. 573 (2013): 1313-1332.

[39] Halko, M. L., & S¨a¨aksvuori,L. (2015). “Competitive Behavior, Stress, and Gender.” Working paper.

[40] Heckman, James J. “Policies to foster human capital.” Research in economics 54, no. 1 (2000): 3-56.

[41] Hoffman, Moshe, Uri Gneezy, and John A. List. “Nurture affects gender differences in spatial abilities.” Proceedings of the National Academy of Sciences 108, no. 36 (2011): 14786-14788.

[42] Ichino, Andrea, and Enrico Moretti. “Biological Gender Differences, Absenteeism, and the Earnings Gap.” American Economic Journal: Applied Economics 1, no. 1 (2009): 183-218.

[43] Kahneman, Daniel, and Amos Tversky. “Prospect theory: An analysis of decision under risk.” Econometrica: Journal of the Econometric Society (1979): 263-291.

23 [44] Lavy, Victor. 2013. “Gender Differences in Market Competitiveness in a Real Workplace: Evidence from Performance-based Pay Tournaments among Teachers.” Economic Jour- nal. Vol 123: 540-573.

[45] Lee, David S., and Thomas Lemieux. “Regression Discontinuity Designs in Economics.” Journal of Economic Literature 48 (2010): 281-355.

[46] Li, Hongbin, Lingsheng Meng, Xinzheng Shi, and Binzhen Wu. “Does Attending Elite Colleges Pay in China?” Journal of Comparative Economics 40 (2012): 78-88.

[47] Maciejewski, Paul K., H. G. Prigerson, and C. M. Mazure. “Sex differences in event- related risk for major depression.” Psychological medicine 31, no. 04 (2001): 593-604.

[48] Malmendier, Ulrike, Geoffrey Tate, and Jon Yan. “Overconfidence and early-life expe- riences: the effect of managerial traits on corporate financial policies.” The Journal of finance 66, no. 5 (2011): 1687-1733.

[49] McCrary, Justin. “Manipulation of the running variable in the regression discontinuity design: A density test.” Journal of Econometrics 142, no. 2 (2008): 698-714.

[50] Nolen-Hoeksema, Susan, and Joan S. Girgus. “The emergence of gender differences in depression during adolescence.” Psychological bulletin 115, no. 3 (1994): 424.

[51] Nunn, Nathan, and Leonard Wantchekon. “The slave trade and the origins of mistrust in Africa.” The American Economic Review 101, no. 7 (2011): 3221-3252.

[52] Nunn, Nathan. “The Importance of for Economic Development.” Annu. Rev. Econ 1 (2009): 65-92.

[53] Niederle, Muriel, and Lise Vesterlund. “Do Women Shy Away From Competition? Do Men Compete Too Much?.” Quarterly Journal of Economics 122, no. 3 (2007).

[54] Paserman, Daniel M. 2010. “Gender Differences in Performance in Competitive Envi- ronments? Evidence from Professional Tennis Players.” Working Paper.

[55] Pearson, Matthew, and Burkhard C. Schipper. “Menstrual cycle and competitive bid- ding.” Games and Economic Behavior 78 (2013): 1-20.

[56] Taylor, Shelley E., Laura Cousino Klein, Brian P. Lewis, Tara L. Gruenewald, Regan AR Gurung, and John A. Updegraff. “Biobehavioral responses to stress in females: tend-and-befriend, not fight-or-flight.” Psychological review 107, no. 3 (2000): 411.

[57] Wei Shang-Jin, and Xiaobo Zhang. “The competitive saving motive: Evidence from rising sex ratios and saving rates in China.” Journal of Political Economy, 119, no. 3 (2011):511-564.

[58] Wozniak, David, William T. Harbaugh, and Ulrich Mayr. “The menstrual cycle and performance feedback alter gender differences in competitive choices.” Journal of Labor Economics 32, no. 1 (2014): 161-198.

24 [59] Zhang, Y. Jane. “Culture, institutions, and the gender gap in competitive inclination: Evidence from the communist experiment in China.” Economic Journal 2017.

[60] Zhang, Junsen, Yaohui Zhao, Albert Park, and Xiaoqing Song. “Economic returns to schooling in urban China, 1988 to 2001.” Journal of Comparative Economics 33, no. 4 (2005): 730-752.

[61] Zhong, Songfa, Idan Shalev, David Koh, Richard P. Ebstein, and Soo Hong Chew. “Competitiveness and Stress.” (2017), International Economic Review, forthcoming

25 Table 1 Summary Statistics Variables Mean S.D. Female 0.446 0.497 Minority 0.074 0.262 Competition 0.374 0.121 CEE scores 0.226 0.623 Parent has a college degree 0.251 0.434 Parent is a cadre 0.133 0.340 Parent works in public sector 0.283 0.450 Enrolled in magnet classes in middle school 0.367 0.482 Win in competition in middle school 0.215 0.411 GPA in top 20% in middle school 0.861 0.346 Enrolled in magnet classes in high school 0.430 0.495 Win in competition in high school 0.162 0.369 GPA in top 20% in high school 0.748 0.434 GPA in college 3.038 0.489 Have Failed in Courses 0.405 0.491 Award in Provincial Level or above 0.142 0.349 CET4 Scores 467.027 56.277 Trust 2.154 0.646 Loss 2.227 0.791 Gain 1.895 0.871 OBS 8176 Note: Competition is measured by one minus college enrollment rate in each province. The values of "Trust" come from answers to the question “Do you agree with the following statement: ‘Generally, you can trust people in the society.' 1. Very Disagree; 2. Disagree; 3. Agree; 4. Very Agree." The values of "Loss" come from the question "You would rather (a) Lose 1000 yuan with certainty; (b) Lose 2000 with 50% and no loss with 50%. 1. (a); 2. (b); 3. (a) and (b) are indifferent." "Loss" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. The values of "Gain" come from the question "You would rather (a) Gain 1000 yuan with certainty; (b) Gain 2000 with 50% and nothing with 50%. 1. (a); 2 (b); 3. (a) and (b) are indifferent." "Gain " is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2.

26

Table 2 Balancing Test Using 2005 Population Mini-Census (1) (2) (3) (4) (5) (6) (7) Having non-agricultural Minority Schooling years Married Migrant Healthy Ln(income+1) Female 0.004 -0.408*** 0.018* -0.002 -0.002 0.008*** -1.268*** (0.010) (0.096) (0.010) (0.005) (0.009) (0.002) (0.257) Female*Competition -0.019 0.096 0.024 0.002 0.004 -0.001 0.868 (0.032) (0.241) (0.024) (0.012) (0.024) (0.006) (0.774) Province fixed effects Yes Yes Yes Yes Yes Yes Yes Constant 0.013*** 10.553*** 0.724*** 0.374*** 0.682*** 0.985*** 4.610*** (0.003) (0.033) (0.003) (0.001) (0.003) (0.001) (0.076) Observations 1,828,689 1,828,689 1,828,689 1,828,689 1,828,689 1,828,689 1,828,689 R-squared 0.245 0.059 0.008 0.102 0.080 0.003 0.035 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%. Competition is measured by one minus college enrollment rate in each province.

27

Table 3 Balancing Test (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Parent Students' Performance in Middle School Students' Performance in High School Ln(# of Students Work in Magnet Win in GPA in Magnet Win in GPA in Having Minority College Cadre Public Class Competition Top 20% Class Competition Top 20% CEE Scores in Sector the Same Decile) Female -0.013 0.025 0.003 -0.001 0.034 -0.049* 0.019 0.012 -0.053 0.053* 0.032 (0.019) (0.055) (0.033) (0.045) (0.034) (0.025) (0.032) (0.030) (0.035) (0.027) (0.223) Female*Competition 0.086 0.094 0.078 0.169 -0.044 0.071 -0.017 0.046 0.098 -0.084 -0.255 (0.067) (0.136) (0.087) (0.120) (0.090) (0.064) (0.088) (0.068) (0.089) (0.064) (0.531) CEE Scores -0.006 0.023** 0.013 0.019* 0.070*** 0.016 0.069*** 0.033** 0.019* 0.047*** 0.396*** (0.009) (0.010) (0.008) (0.010) (0.011) (0.011) (0.014) (0.013) (0.010) (0.009) (0.109) Constant 0.067*** 0.203*** 0.117*** 0.272*** 0.414*** 0.156*** 0.741*** 0.348*** 0.205*** 0.860*** 2.089*** (0.004) (0.007) (0.005) (0.006) (0.005) (0.005) (0.005) (0.006) (0.006) (0.004) (0.051) Observations 6,649 6,212 6,679 6,679 6,640 6,545 6,650 6,636 6,553 6,629 311 R-squared 0.152 0.047 0.028 0.047 0.057 0.033 0.031 0.044 0.037 0.032 0.205 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. Competition is measured by one minus college enrollment rate in each province.

28

Table 4 Gender Difference in the Impacts of Experiencing Competition on Preference: Ordered Logit (1) (2) (3) Loss Gain Trust Female 0.201 -0.136 0.630*** (0.131) (0.136) (0.123) Female*Competition -0.676* -0.615* -1.222*** (0.346) (0.318) (0.315) CEE Scores 0.031 0.019 0.066 (0.032) (0.032) (0.059)

Observations 5,748 5,768 5,866 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. The values of "Trust" come from answers to the question “Do you agree with the following statement: ‘Generally, you can trust people in the society.' 1. Very Disagree; 2. Disagree; 3. Agree; 4. Very Agree." The values of "Loss" come from the question "You would rather (a) Lose 1000 yuan with certainty; (b) Lose 2000 with 50% and no loss with 50%. 1. (a); 2. (b); 3. (a) and (b) are indifferent." "Loss" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. The values of "Gain" come from the question "You would rather (a) Gain 1000 yuan with certainty; (b) Gain 2000 with 50% and nothing with 50%. 1. (a); 2 (b); 3. (a) and (b) are indifferent." "Gain" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. Competition is measured by one minus college enrollment rate in each province.

29

Table 5 Gender Difference in the Impacts of Experiencing Competition on Performance in College: OLS (1) (2) (3) (4) Ln(GPA in Have Failed in CET4 Award in Provincial Level or above College) Courses Scores Female 0.092*** -0.291*** -0.031 14.109 (0.024) (0.064) (0.031) (8.323) Female*Competition 0.016 -0.006 0.153** 23.732 (0.057) (0.146) (0.069) (18.880) CEE Scores 0.007 -0.048*** -0.000 15.799*** (0.005) (0.012) (0.006) (3.546) Constant 1.049*** 0.557*** 0.121*** 453.380*** (0.002) (0.009) (0.004) (1.241) Observations 4,832 5,885 4,944 4,796 R-squared 0.117 0.107 0.015 0.160 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. Ln(GPA in College) is natural logarithm of GPA students reported. "Have Failed in Courses" is a dummy variable, with 1 indicating that students have failed any course in college and 0 vice verse. "Award in Provincial Level or above" is a dummy variable, with 1 indicating that students have won any awards in provincial level or above and 0 vice verse. "CET4 Scores" is scores students got in (Level 4). Competition is measured by one minus college enrollment rate in each province.

30

Table 6 Effects of Competitive Experience by Different Groups (1) (2) (3) (4) (5) (6) Magnet Regular Natural Only High High Science Other Tracks Having siblings Child School School Track Loss -0.832** -0.576 -0.189 -2.582*** -0.215 -0.687

(0.421) (0.755) (0.384) (0.619) (0.640) (0.470) Gain -0.666 -0.148 -0.839** -0.725 -0.803* -0.251

(0.410) (0.513) (0.357) (0.520) (0.471) (0.567) Trust -1.589*** -0.728 -1.134*** -2.242** -0.954** -1.297**

(0.569) (0.671) (0.278) (1.038) (0.381) (0.535) Ln(GPA in College) 0.050 -0.040 0.046 -0.118* 0.087 -0.055

(0.074) (0.060) (0.092) (0.064) (0.074) (0.086) Have Failed in Courses 0.079 -0.218 -0.181 0.223 0.032 -0.019

(0.170) (0.143) (0.169) (0.143) (0.195) (0.121) 0.139* 0.113 0.145* 0.112 0.143 0.192** Award in Provincial Level or above

(0.076) (0.189) (0.085) (0.147) (0.094) (0.087) CET4 Scores 8.598 37.167 17.038 20.335 29.020 13.593 (23.146) (26.432) (21.846) (31.890) (29.451) (28.697) Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%. The same regressions as those in Table A3 and Table A4 are estimated using different groups of samples. The coefficients shown in the table are those on the interaction of the female dummy and the competition measurement.

31

Table A1 Provincial Admission Rates Used in the Paper Hainan 85.10% Shanghai 83.75% Jiangsu 78.61% Tianjin 77.86% Heilongjiang 77.68% Liaoning 76.90% Beijing 73.59% Zhejiang 72.50% Shandong 72.41% Chongqing 72.08% Guangdong 69.41% Jilin 68.70% Tibet 65.10% Qinghai 64.30% Fujian 59.60% Ningxia 58.58% Neimenggu 57.90% Hubei 57.40% Jiangxi 57.31% Guangxi 57.22% Xinjiang 56.60% Hebei 56.19% Sichuan 55.18% Hunan 53.40% Henan 52.34% Anhui 51.51% 49.48% Shaanxi 49.18% Shanxi 47.95% Guizhou 43.00% Gansu 40.98% Source: We get the information from China Education Examinations Yearbook (2008) for 19 provinces, including Shanghai, Liaoning, Beijing, Chongqing, Guangdong, Shandong, Jiangxi, Hunan, Guangxi, Hebei, Henan, Anhui, Shanxi, Gansu, Tianjin, Jiangsu, Sichuan, Shaanxi, and Yunnan . We get the information from China Gaokao Information (http://gaokao.chsi.com.cn/gkxx/zhiding/200709/20070905/1113417.html ) for 9 provinces, including Hainan, Zhejiang, Jilin, Tibet, Qinghai, Fujian, Neimenggu, Xinjiang, and Guizhou. For the remaining three provinces (Hubei, Heilongjiang, and Ningxia), we search the news reports and we get the college admission rates in 2007 from three news reports: http://gaokao.eol.cn/dongtai_5640/20070618/t20070618_238318.shtml, http://zqb.cyol.com/content/2007-05/15/content_1759511.htm , and http://edu.qq.com/a/20070620/000158.htm.

32

Table A2 Determinants of College Admission Rates in Provinces College Admission Rates Population (billion) -4.311 (3.329) GDP per capita (100 thousand) -0.280 (0.608) Percentage of GDP in the first industry 0.518 (1.163) Percentage of GDP in the second industry 0.856 (0.822) Average labor income (100 thousand) -0.026 (0.813) Government income （billion) -0.001 (0.001) Export (10 thousand) 0.000 (0.000) Import (10 thousand) 0.000 (0.000) Ratio of individuals having primary education in 6+ years old population 0.598 (0.754) Ratio of individuals having middle school education in 6+ years old population 0.079 (0.506) Ratio of individuals having high school education in 6+ years old population -1.359 (1.321) Ratio of individuals having college or above education in 6+ years old 1.856 population (1.068) Ratio of education expenditure in government total expenditure -0.683 (0.795) Number of colleges -0.001 (0.003) Number of college students (10 thousand) -0.044 (0.065) Number of college teachers (10 thousand) -0.013 (0.467) Number of high schools 0.000 (0.000) Number of high school students (10 thousand) -0.034 (0.038) Number of high school graduates (10 thousand) 0.256** (0.109) Constant -0.027 (0.957) Observations 31 R-squared 0.889 Robust standard errors in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%.

33

Table A3 Gender Difference in the Impacts of Experiencing Competition on Preference: Ordered Logit (1) (2) (3) Loss Gain Trust Female 0.194 -0.123 0.629*** (0.129) (0.134) (0.124) Female*Competition -0.646* -0.638** -1.198*** (0.346) (0.312) (0.313) CEE Scores 0.019 0.020 0.068 (0.031) (0.033) (0.060) Minority -0.054 0.050 0.042 (0.084) (0.133) (0.166) Parent has a college degree -0.075 0.010 -0.002 (0.072) (0.084) (0.087) Parent is a cadre -0.013 -0.067 -0.077 (0.101) (0.088) (0.094) Parent works in public sector -0.047 -0.058 -0.103 (0.065) (0.056) (0.077) Enrolled in magnet classes in middle school 0.012 -0.008 -0.129** (0.047) (0.052) (0.053) GPA in top 20% in middle school 0.161** -0.061 -0.052 (0.078) (0.086) (0.081) Win in competition in middle school 0.131* 0.124* 0.088 (0.077) (0.074) (0.081) Enrolled in magnet classes in high school 0.018 -0.028 0.097* (0.060) (0.066) (0.053) GPA in top 20% in high school 0.039 0.055 0.048 (0.063) (0.079) (0.072) Win in competition in high school -0.073 -0.020 -0.206* (0.071) (0.073) (0.119) Observations 5,748 5,768 5,866 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. The values of "Trust" come from answers to the question “Do you agree with the following statement: ‘Generally, you can trust people in the society.' 1. Very Disagree; 2. Disagree; 3. Agree; 4. Very Agree." The values of "Loss" come from the question "You would rather (a) Lose 1000 yuan with certainty; (b) Lose 2000 with 50% and no loss with 50%. 1. (a); 2. (b); 3. (a) and (b) are indifferent." "Loss" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. The values of "Gain" come from the question "You would rather (a) Gain 1000 yuan with certainty; (b) Gain 2000 with 50% and nothing with 50%. 1. (a); 2 (b); 3. (a) and (b) are indifferent." "Gain" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. Competition is measured by one minus college enrollment rate in each province.

34

Table A4 Gender Difference in the Impacts of Experiencing Competition on Performance: OLS (1) (2) (3) (4) Award in Ln(GPA in Have Failed in Provincial Level CET4 Scores College) Courses or above Female 0.090*** -0.289*** -0.027 14.849* (0.026) (0.066) (0.030) (8.078) Female*Competition 0.022 -0.015 0.151** 19.424 (0.060) (0.148) (0.066) (18.729) CEE Scores 0.003 -0.038*** -0.003 13.045*** (0.005) (0.012) (0.006) (3.176) Minority -0.009 0.018 0.013 -9.089** (0.009) (0.021) (0.024) (3.459) Parent has a college degree 0.008 -0.009 0.019 10.094*** (0.009) (0.020) (0.017) (1.905) Parent is a cadre -0.005 0.004 -0.005 2.329 (0.008) (0.020) (0.018) (3.316) Parent works in public sector -0.012** 0.046** -0.007 5.208*** (0.006) (0.022) (0.017) (1.772) Enrolled in magnet classes in middle -0.001 -0.008 0.005 2.221 school (0.006) (0.012) (0.009) (1.740) GPA in top 20% in middle school 0.021** -0.066*** 0.022 14.836*** (0.009) (0.021) (0.013) (3.538) Win in competition in middle school -0.012 0.007 0.012 11.739*** (0.008) (0.017) (0.015) (1.949) Enrolled in magnet classes in high school 0.006 -0.049*** -0.007 7.800*** (0.005) (0.012) (0.010) (1.495) GPA in top 20% in high school 0.024*** -0.032* -0.009 5.589*** (0.006) (0.016) (0.013) (1.437) Win in competition in high school 0.044*** -0.068*** 0.149*** 9.163*** (0.009) (0.015) (0.021) (2.773) Constant 1.010*** 0.656*** 0.080*** 425.096*** (0.009) (0.021) (0.017) (3.464) Observations 4,832 5,885 4,944 4,796 R-squared 0.131 0.118 0.043 0.210 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. Ln(GPA in College) is natural logarithm of GPA students reported. "Have Failed in Courses" is a dummy variable, with 1 indicating that students have failed any course in college and 0 vice verse. "Award in Provincial Level or above" is a dummy variable, with 1 indicating that students have won any awards in provincial level or above and 0 vice verse. "CET4 Scores" is scores students got in College English Test (Level 4). Competition is measured by one minus college enrollment rate in each province.

35

Table A5 Gender Difference in the Impacts of Experiencing Competition on Preference: Ordered Logit (1) (2) (3) Loss Gain Trust Female -1.891 1.573 0.566 (1.661) (2.496) (2.182) Female*Competition -0.896** -0.420* -1.461*** (0.425) (0.223) (0.431) Female*(High school graduates/Population) 0.000 -0.001 0.009 (0.003) (0.005) (0.006) Female*Sex Ratio 0.021 -0.017 -0.004 (0.016) (0.025) (0.022) CEE Scores 0.019 0.020 0.069 (0.031) (0.033) (0.059) Minority -0.057 0.052 0.052 (0.085) (0.131) (0.168) Parent has a college degree -0.076 0.012 -0.001 (0.073) (0.084) (0.087) Parent is a cadre -0.013 -0.066 -0.078 (0.101) (0.088) (0.094) Parent works in public sector -0.046 -0.058 -0.102 (0.065) (0.056) (0.077) Enrolled in magnet classes in middle school 0.014 -0.009 -0.129** (0.048) (0.052) (0.052) GPA in top 20% in middle school 0.161** -0.062 -0.052 (0.078) (0.086) (0.081) Win in competition in middle school 0.130* 0.124* 0.088 (0.077) (0.073) (0.081) Enrolled in magnet classes in high school 0.018 -0.028 0.098* (0.060) (0.066) (0.053) GPA in top 20% in high school 0.039 0.055 0.047 (0.063) (0.079) (0.072) Win in competition in high school -0.072 -0.021 -0.207* (0.071) (0.072) (0.118) Observations 5,748 5,768 5,866 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. The values of "Trust" come from answers to the question “Do you agree with the following statement: ‘Generally, you can trust people in the society.' 1. Very Disagree; 2. Disagree; 3. Agree; 4. Very Agree." The values of "Loss" come from the question "You would rather (a) Lose 1000 yuan with certainty; (b) Lose 2000 with 50% and no loss with 50%. 1. (a); 2. (b); 3. (a) and (b) are indifferent." "Loss" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. The values of "Gain" come from the question " You would rather (a) Gain 1000 yuan with certainty; (b) Gain 2000 with 50% and nothing with 50%. 1. (a); 2 (b); 3. (a) and (b) are indifferent." "Gain" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2 . Competition is measured by one minus college enrollment rate in each province.

36

Table A6 Gender Difference in the Impacts of Experiencing Competition on Performance: OLS (1) (2) (3) (4) Have Failed in Award in CET4 Ln(GPA in College) Courses Provincial Level or above Scores Female -0.196 0.754 0.409 89.317 (0.224) (0.668) (0.410) (160.008) Female*Competition -0.033 0.190 0.189** 26.158 (0.061) (0.152) (0.079) (24.506) Female*(High school 0.001** -0.003* 0.000 0.059 graduates/Population) (0.000) (0.001) (0.001) (0.207) Female*Sex Ratio 0.003 -0.009 -0.005 -0.786 (0.002) (0.006) (0.004) (1.549) CEE Scores 0.003 -0.038*** -0.003 13.118*** (0.005) (0.012) (0.006) (3.169) Minority -0.009 0.016 0.014 -8.905** (0.009) (0.021) (0.024) (3.454) Parent has a college degree 0.008 -0.008 0.019 10.149*** (0.009) (0.020) (0.017) (1.900) Parent is a cadre -0.005 0.004 -0.005 2.349 (0.008) (0.020) (0.018) (3.318) Parent works in public sector -0.012** 0.046** -0.007 5.200*** (0.006) (0.022) (0.017) (1.756) Enrolled in magnet classes in -0.001 -0.009 0.005 2.188 middle school (0.006) (0.012) (0.009) (1.751) GPA in top 20% in middle 0.021** -0.067*** 0.022 14.850*** school (0.009) (0.021) (0.013) (3.537) Win in competition in middle -0.012 0.008 0.012 11.763*** school (0.008) (0.018) (0.015) (1.943) Enrolled in magnet classes in 0.006 -0.048*** -0.007 7.800*** high school (0.005) (0.012) (0.010) (1.491) GPA in top 20% in high school 0.024*** -0.032* -0.009 5.575*** (0.006) (0.016) (0.013) (1.442) Win in competition in high 0.045*** -0.068*** 0.148*** 9.100*** school (0.009) (0.016) (0.021) (2.778) Constant 1.010*** 0.656*** 0.080*** 425.045*** (0.009) (0.021) (0.016) (3.468) Observations 4,832 5,885 4,944 4,796 R-squared 0.131 0.119 0.043 0.211 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. Ln(GPA in College) is natural logarithm of GPA

37 students reported. "Have Failed in Courses" is a dummy variable, with 1 indicating that students have failed any course in college and 0 vice verse. "Award in Provincial Level or above" is a dummy variable, with 1 indicating that students have won any awards in provincial level or above and 0 vice verse. "CET4 Scores" is scores students got in College English Test (Level 4). Competition is measured by one minus college enrollment rate in each province.

38

Table A7 Gender Difference in the Impacts of Experiencing Competition on Preference: Ordered Logit (1) (2) (3) Loss Gain Trust Female 0.186 -0.103 0.612*** (0.128) (0.132) (0.123) Female*Competition -0.668* -0.591* -1.227*** (0.350) (0.316) (0.307) Female*CEE Scores 0.071 -0.161 0.116 (0.078) (0.112) (0.074) CEE Scores -0.009 0.078 0.019 (0.040) (0.048) (0.078) Minority -0.054 0.051 0.040 (0.084) (0.131) (0.165) Parent has a college degree -0.075 0.011 -0.003 (0.073) (0.084) (0.087) Parent is a cadre -0.013 -0.065 -0.080 (0.101) (0.087) (0.095) Parent works in public sector -0.046 -0.059 -0.101 (0.066) (0.056) (0.077) Enrolled in magnet classes in middle school 0.012 -0.007 -0.129** (0.047) (0.052) (0.053) GPA in top 20% in middle school 0.157** -0.052 -0.057 (0.079) (0.084) (0.082) Win in competition in middle school 0.130* 0.125* 0.088 (0.077) (0.073) (0.081) Enrolled in magnet classes in high school 0.017 -0.026 0.096* (0.059) (0.067) (0.053) GPA in top 20% in high school 0.038 0.057 0.048 (0.063) (0.079) (0.072) Win in competition in high school -0.070 -0.024 -0.203* (0.071) (0.073) (0.119) Observations 5,748 5,768 5,866 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. The values of "Trust" come from answers to the question “Do you agree with the following statement: ‘Generally, you can trust people in the society.' 1. Very Disagree; 2. Disagree; 3. Agree; 4. Very Agree." The values of "Loss" come from the question "You would rather (a) Lose 1000 yuan with certainty; (b) Lose 2000 with 50% and no loss with 50%. 1. (a); 2. (b); 3. (a) and (b) are indifferent." "Loss" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. The values of "Gain" come from the question "You would rather (a) Gain 1000 yuan with certainty; (b) Gain 2000 with 50% and nothing with 50%. 1. (a); 2 (b); 3. (a) and (b) are indifferent." "Gain" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. Competition is measured by one minus college enrollment rate in each province.

39

Table A8 Gender Difference in the Impacts of Experiencing Competition on Performance: OLS (1) (2) (3) (4) Award in CET4 Ln(GPA in College) Have Failed in Courses Provincial Level or Scores above Female 0.088*** -0.288*** -0.028 13.194 (0.026) (0.066) (0.030) (8.280) Female*Competition 0.020 -0.012 0.147** 15.629 (0.059) (0.148) (0.066) (20.108) Female*CEE Scores 0.012 -0.010 0.013 11.660* (0.010) (0.020) (0.017) (6.117) CEE Scores -0.001 -0.034** -0.008 7.891** (0.006) (0.015) (0.007) (3.546) Minority -0.010 0.018 0.013 -9.109** (0.009) (0.021) (0.024) (3.478) Parent has a college degree 0.008 -0.009 0.019 10.086*** (0.009) (0.020) (0.017) (1.914) Parent is a cadre -0.005 0.004 -0.006 2.164 (0.009) (0.020) (0.018) (3.355) Parent works in public sector -0.012** 0.046** -0.007 5.253*** (0.006) (0.021) (0.017) (1.733) Enrolled in magnet classes in -0.001 -0.008 0.005 2.238 middle school (0.006) (0.012) (0.009) (1.719) GPA in top 20% in middle 0.020** -0.066*** 0.021 14.578*** school (0.009) (0.021) (0.013) (3.611) Win in competition in middle -0.012 0.007 0.012 11.609*** school (0.008) (0.017) (0.015) (1.971) Enrolled in magnet classes in 0.006 -0.048*** -0.007 7.697*** high school (0.005) (0.012) (0.010) (1.472) GPA in top 20% in high 0.024*** -0.032* -0.009 5.522*** school (0.006) (0.016) (0.012) (1.405) Win in competition in high 0.045*** -0.068*** 0.149*** 9.509*** school (0.009) (0.015) (0.021) (2.785) Constant 1.011*** 0.655*** 0.082*** 426.723*** (0.010) (0.021) (0.016) (3.581) Observations 4,832 5,885 4,944 4,796 R-squared 0.131 0.118 0.043 0.213 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. Ln(GPA in College) is natural logarithm of GPA students reported. "Have Failed in Courses" is a dummy variable, with 1 indicating that students have failed any course in

40 college and 0 vice verse. "Award in Provincial Level or above" is a dummy variable, with 1 indicating that students have won any awards in provincial level or above and 0 vice verse. "CET4 Scores" is scores students got in College English Test (Level 4). Competition is measured by one minus college enrollment rate in each province.

41

Table A9 Gender Difference in the Impacts of Experiencing Competition on Preference, Adding College Dummies: Ordered Logit (1) (2) (3) Loss Gain Trust Female 0.278** -0.121 0.584*** (0.136) (0.135) (0.121) Female*Competition -0.822** -0.718** -1.096*** (0.352) (0.308) (0.303) CEE Scores 0.002 0.028 0.068 (0.031) (0.033) (0.059) Minority -0.104 0.020 0.071 (0.086) (0.135) (0.170) Parent has a college degree -0.072 0.019 -0.018 (0.075) (0.084) (0.089) Parent is a cadre -0.016 -0.066 -0.053 (0.098) (0.091) (0.093) Parent works in public sector -0.057 -0.070 -0.113 (0.068) (0.058) (0.074) Enrolled in magnet classes in middle school 0.024 -0.015 -0.131** (0.045) (0.051) (0.054) GPA in top 20% in middle school 0.147* -0.037 -0.069 (0.079) (0.091) (0.081) Win in competition in middle school 0.120 0.140* 0.058 (0.076) (0.073) (0.078) Enrolled in magnet classes in high school 0.001 -0.021 0.109** (0.066) (0.067) (0.051) GPA in top 20% in high school 0.031 0.074 0.047 (0.058) (0.076) (0.074) Win in competition in high school -0.057 0.004 -0.202* (0.070) (0.079) (0.118) Observations 5,748 5,768 5,866 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies and college dummies are controlled in all regressions. The values of "Trust" come from ans wers to the question “Do you agree with the following statement: ‘Generally, you can trust people in the society.' 1. Very Disagree; 2. Disagree; 3. Agree; 4. Very Agree." The values of "Loss" come from the question "You would rather (a) Lose 1000 yuan wit h certainty; (b) Lose 2000 with 50% and no loss with 50%. 1. (a); 2. (b); 3. (a) and (b) are indifferent." "Loss" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. The values of "Gain" come from the question "You would rather (a) Gain 1000 yuan with certainty; (b) Gain 2000 with 50% and nothing with 50%. 1. (a); 2 (b); 3. (a) and (b) are indifferent." "Gain" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. Competition is measured by one minus college enrollment rate in each province.

42

Table A10 Gender Difference in the Impacts of Experiencing Competition on Performance, Adding College Dummies: OLS (1) (2) (3) (4) Award in Ln(GPA in Have Failed in CET4 Provincial Level or College) Courses Scores above Female 0.095*** -0.252*** -0.023 19.966*** (0.024) (0.065) (0.035) (6.353) Female*Competition 0.006 -0.057 0.141* 13.367 (0.059) (0.146) (0.077) (14.302) CEE Scores 0.000 -0.035*** 0.004 7.324*** (0.004) (0.011) (0.007) (2.446) Minority -0.015 0.032 0.003 -12.977*** (0.009) (0.022) (0.024) (3.917) Parent has a college degree -0.000 -0.001 0.028 5.639*** (0.008) (0.018) (0.018) (1.769) Parent is a cadre -0.001 0.003 0.002 -0.523 (0.009) (0.017) (0.019) (2.862) Parent works in public sector -0.014** 0.055** -0.010 2.481 (0.006) (0.021) (0.016) (1.675) Enrolled in magnet classes in middle 0.001 -0.005 0.004 3.500** school (0.005) (0.012) (0.009) (1.654) GPA in top 20% in middle school 0.023** -0.068*** 0.024* 10.582*** (0.010) (0.020) (0.014) (3.795) Win in competition in middle school -0.015* 0.018 0.017 7.049*** (0.008) (0.017) (0.015) (2.005) Enrolled in magnet classes in high school 0.006 -0.052*** -0.001 4.803*** (0.004) (0.013) (0.010) (1.290) GPA in top 20% in high school 0.025*** -0.032** -0.006 1.104 (0.005) (0.014) (0.013) (1.391) Win in competition in high school 0.040*** -0.051*** 0.156*** 3.519 (0.009) (0.016) (0.021) (2.819) Constant 0.987*** 0.711*** 0.283*** 432.081*** (0.016) (0.050) (0.050) (5.740) Observations 4,832 5,885 4,944 4,796 R-squared 0.200 0.158 0.071 0.314 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies and college dummies are controlled in all regressions. Ln(GPA in College) is natural logarithm of GPA students reported. "Have Failed in Courses" is a dummy variable, with 1 indicating that students have failed any course in college and 0 vice verse. "Award in Provincial Level or above" is a dummy variable, with 1 indicating that students have won any awards in provincial level or above and 0 vice verse. "CET4 Scores" is scores students got in College English Test (Level 4). Competition is measured by one minus college enrollment rate in each province.

43

Table A11 Balancing Test, Monte Carlo Experiment Mean S.D. T P-value Loss 0.008 0.377 0.020 0.492 Gain 0.012 0.359 0.033 0.487 Trust -0.024 0.493 -0.049 0.961 Ln(GPA in college) 0.003 0.046 0.076 0.470 Have Failed in Courses 0.002 0.151 0.011 0.496 Award in Provincial Level or above 0.003 0.083 0.032 0.487 CET4 Scores 1.061 22.012 0.048 0.481

44

Table A12 Gender Difference in the Impacts of Experiencing Competition on Preference: Ordered Logit (1) (2) (3) Loss Gain Trust Female 0.197 -0.126 0.628*** (0.128) (0.133) (0.124) Female*Average competition -0.655* -0.632** -1.195*** (0.344) (0.310) (0.313) CEE Scores 0.019 0.020 0.068 (0.031) (0.033) (0.060) Minority -0.054 0.050 0.042 (0.084) (0.133) (0.166) Parent has a college degree -0.075 0.010 -0.002 (0.072) (0.084) (0.087) Parent is a cadre -0.013 -0.067 -0.077 (0.101) (0.088) (0.094) Parent works in public sector -0.047 -0.058 -0.103 (0.065) (0.056) (0.077) Enrolled in magnet classes in middle school 0.012 -0.008 -0.129** (0.047) (0.052) (0.053) GPA in top 20% in middle school 0.161** -0.061 -0.052 (0.078) (0.086) (0.081) Win in competition in middle school 0.131* 0.124* 0.088 (0.077) (0.074) (0.081) Enrolled in magnet classes in high school 0.018 -0.028 0.097* (0.060) (0.066) (0.053) GPA in top 20% in high school 0.039 0.055 0.048 (0.063) (0.079) (0.072) Win in competition in high school -0.073 -0.020 -0.206* (0.071) (0.073) (0.119) Observations 5,748 5,768 5,866 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. The values of "Trust" come from answers to the question “Do you agree with the following statement: ‘Generally, you can trust people in the society.' 1. Very Disagree; 2. Disagree; 3. Agree; 4. Very Agree." The values of "Loss" come from the question "You would rather (a) Lose 1000 yuan with certainty; (b) Lose 2000 with 50% and no loss with 50%. 1. (a); 2. (b); 3. (a) and (b) are indifferent." "Loss" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. The values of "Gain" come from the question "You would rather (a) Gain 1000 yuan with certainty; (b) Gain 2000 with 50% and nothing with 50%. 1. (a); 2 (b); 3. (a) and (b) are indifferent." "Gain" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. Average competition is measured by the average of one minus provincial college admission rates in 2006 and 2007.

45

Table A13 Gender Difference in the Impacts of Experiencing Competition on Performance: OLS (1) (2) (3) (4) Award in Have Failed CET4 Ln(GPA in College) Provincial Level in Courses Scores or above Female 0.090*** -0.290*** -0.026 14.975* (0.026) (0.066) (0.030) (8.101) Female*Average competition 0.023 -0.012 0.149** 19.088 (0.060) (0.148) (0.066) (18.797) CEE Scores 0.003 -0.038*** -0.003 13.045*** (0.005) (0.012) (0.006) (3.176) Minority -0.009 0.018 0.013 -9.089** (0.009) (0.021) (0.024) (3.460) Parent has a college degree 0.008 -0.009 0.019 10.095*** (0.009) (0.020) (0.017) (1.906) Parent is a cadre -0.005 0.004 -0.005 2.329 (0.008) (0.020) (0.018) (3.316) Parent works in public sector -0.012** 0.046** -0.007 5.208*** (0.006) (0.022) (0.017) (1.772) Enrolled in magnet classes in middle school -0.001 -0.008 0.005 2.221 (0.006) (0.012) (0.009) (1.740) GPA in top 20% in middle school 0.021** -0.066*** 0.022 14.835*** (0.009) (0.021) (0.013) (3.538) Win in competition in middle school -0.012 0.007 0.012 11.741*** (0.008) (0.017) (0.015) (1.949) Enrolled in magnet classes in high school 0.006 -0.049*** -0.007 7.799*** (0.005) (0.012) (0.010) (1.495) GPA in top 20% in high school 0.024*** -0.032* -0.009 5.590*** (0.006) (0.016) (0.013) (1.436) Win in competition in high school 0.044*** -0.068*** 0.149*** 9.164*** (0.009) (0.015) (0.021) (2.772) Constant 1.010*** 0.656*** 0.080*** 425.094*** (0.009) (0.021) (0.017) (3.465) Observations 4,832 5,885 4,944 4,796 R-squared 0.131 0.118 0.043 0.210 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. Ln(GPA in College) is natural logarithm of GPA students reported. "Have Failed in Courses" is a dummy variable, with 1 indicating that students have failed any course in college and 0 vice verse. "Award in Provincial Level or above" is a dummy variable, with 1 indicating that students have won any awards in provincial level or above and 0 vice verse. "CET4 Scores" is scores students got in College English Test (Level 4). Average competition is measured by the average of one minus provincial college admission rates in 2006 and 2007.

46

Table A14 Gender Difference in the Impacts of Experiencing Competition on Preference, Dropping Three Provinces: Ordered Logit (1) (2) (3) Loss Gain Trust Female 0.272** -0.080 0.659*** (0.119) (0.128) (0.123) Female*Competition -0.807** -0.698** -1.285*** (0.332) (0.298) (0.303) CEE Scores 0.030 0.018 0.054 (0.031) (0.032) (0.060) Minority -0.020 0.003 0.093 (0.086) (0.136) (0.168) Parent has a college degree -0.067 0.004 -0.063 (0.078) (0.085) (0.067) Parent is a cadre 0.026 -0.066 -0.071 (0.109) (0.091) (0.101) Parent works in public sector -0.040 -0.068 -0.075 (0.069) (0.058) (0.078) Enrolled in magnet classes in middle school 0.002 -0.008 -0.120** (0.046) (0.055) (0.055) GPA in top 20% in middle school 0.139* -0.033 -0.062 (0.083) (0.090) (0.089) Win in competition in middle school 0.118 0.135* 0.051 (0.078) (0.079) (0.083) Enrolled in magnet classes in high school 0.019 -0.023 0.117** (0.064) (0.070) (0.053) GPA in top 20% in high school 0.091 0.055 0.065 (0.058) (0.086) (0.076) Win in competition in high school -0.083 -0.039 -0.222* (0.078) (0.073) (0.128) Observations 5,321 5,340 5,428 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. The values of "Trust" come from answers to the question “Do you agree with the following statement: ‘Generally, you can trust people in the society.' 1. Very Disagree; 2. Disagree; 3. Agree; 4. Very Agree." The values of "Loss" come from the question "You would rather (a) Lose 1000 yuan with certainty; (b) Los e 2000 with 50% and no loss with 50%. 1. (a); 2. (b); 3. (a) and (b) are indifferent." "Loss" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. The values of "Gain" come from the question "You would rather (a) Gain 1000 yuan with certainty; (b) Gain 2000 with 50% and nothing with 50%. 1. (a); 2 (b); 3. (a) and (b) are indifferent." "Gain" is equal to 1 if the answer is 1, 2 if the answer is 3, and 3 if the answer is 2. Competition is measured by one minus college enrollment rate in each province.

47

Table A15 Gender Difference in the Impacts of Experiencing Competition on Performance, Dropping Three Provinces: OLS (1) (2) (3) (4) Award in Ln(GPA in Have Failed CET4 Provincial Level or College) in Courses Scores above Female 0.098*** -0.318*** -0.022 17.187* (0.025) (0.073) (0.033) (8.841) Female*Competition 0.004 0.042 0.143* 14.831 (0.058) (0.164) (0.072) (20.418) CEE Scores 0.002 -0.040*** -0.002 12.774*** (0.005) (0.013) (0.006) (3.238) Minority -0.008 0.016 0.010 -8.712** (0.010) (0.023) (0.024) (3.656) Parent has a college degree 0.009 -0.011 0.027 10.823*** (0.010) (0.021) (0.018) (2.039) Parent is a cadre -0.005 0.005 -0.003 2.087 (0.009) (0.022) (0.019) (3.366) Parent works in public sector -0.011* 0.045* -0.014 5.189*** (0.006) (0.023) (0.018) (1.789) Enrolled in magnet classes in middle -0.000 -0.009 0.009 2.104 school (0.006) (0.013) (0.009) (1.887) GPA in top 20% in middle school 0.018* -0.073*** 0.014 14.716*** (0.010) (0.022) (0.014) (3.534) Win in competition in middle school -0.011 0.004 0.015 12.128*** (0.009) (0.018) (0.015) (2.055) Enrolled in magnet classes in high school 0.007 -0.041*** -0.007 7.268*** (0.005) (0.012) (0.011) (1.594) GPA in top 20% in high school 0.025*** -0.031* -0.009 5.861*** (0.006) (0.017) (0.013) (1.216) Win in competition in high school 0.044*** -0.058*** 0.159*** 8.031** (0.009) (0.015) (0.021) (2.927) Constant 1.007*** 0.665*** 0.086*** 425.086*** (0.010) (0.022) (0.018) (3.574) Observations 4,481 5,445 4,584 4,423 R-squared 0.129 0.122 0.046 0.215 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. Ln(GPA in College) is natural logarithm of GPA students reported. "Have Failed in Courses" is a dummy variable, with 1 indicating that students have failed any course in college and 0 vice verse. "Award in Provincial Level or above" is a dummy variable, with 1 indicating that students have won any awards in provincial level or above and 0 vice verse. "CET4 Scores" is scores students got in College English Test (Level 4). Competition is measured by one minus college enrollment rate in each province.

48

Table A16 Gender Difference in the Impacts of Experiencing Competition on Performance in College: OLS (1) (2) (3) (4) Ln(GPA in Have Failed in CET4 Award in Provincial Level or above College) Courses Scores Female 0.110*** -0.327*** -0.004 10.025 (0.029) (0.085) (0.063) (9.640) Female*Competition 0.011 0.008 0.141* 23.270 (0.059) (0.144) (0.075) (19.066) Loss 0.005 -0.023 -0.003 -1.080 (0.005) (0.016) (0.009) (1.508) Gain 0.003 0.000 0.006 1.769 (0.004) (0.013) (0.008) (1.144) Trust 0.002 0.003 -0.004 -1.825 (0.005) (0.010) (0.009) (1.130) Female*Loss -0.000 0.003 0.010 0.664 (0.006) (0.014) (0.011) (1.354) Female*Gain -0.004 -0.010 0.010 -1.336 (0.006) (0.017) (0.014) (1.969) Female*Trust -0.004 0.018 -0.029 2.069 (0.007) (0.018) (0.019) (2.622) CEE Scores 0.007 -0.047*** -0.001 15.297*** (0.005) (0.012) (0.006) (3.561) Constant 1.027*** 0.601*** 0.123*** 455.684*** (0.015) (0.043) (0.030) (5.083) Observations 4,677 5,663 4,772 4,647 R-squared 0.117 0.108 0.018 0.161 Robust standard errors in parentheses are calculated by clustering over province; * significant at 10%; ** significant at 5%; *** significant at 1%; Province dummies are controlled in all regressions. Ln(GPA in College) is natural logarithm of GPA students reported. "Have Failed in Courses" is a dummy variable, with 1 indicating that students have failed any course in college and 0 vice verse. "Award in Provincial Level or above" is a dummy variable, with 1 indicating that students have won any awards in provincial level or above and 0 vice verse. "CET4 Scores" is scores students got in College English Test (Level 4). Competition is measured by one minus college enrollment rate in each province.

49