Gender differences in repeated competition: Evidence from school math contests

https://doi.org/10.1016/j.jebo.2012.12.029Get rights and content

Abstract

The literature shows that males react more favorably than females to competitive incentives. This well-known result, however, is based on experiments in which participants engage in only a one-shot contest. We conduct a series of math contests in elementary schools which are similar to past experiments except for one notable exception: subjects compete in five sequential contests, rather than a one-shot contest typically used. Although males outperform females in the first period contest, we find no evidence of a male advantage in subsequent periods. Females even outperform males in later periods. The data suggests that the relative overperformance of low-ability males and the underperformance of high-ability females are primarily responsible for the first period results. Additionally, even the first period male advantage disappears when we reduce the time pressure or change the task at hand.

Highlights

► We host a series of real-effort math contests to assess gender differences in performance in competitive environments. ► Unlike other gender differences experiments, we repeat treatments multiple times with the same subjects. ► Consistent with past studies we find that males outperform females of similar ability during the first period of competitions. ► The male advantage is not found in any subsequent period, and females even outperform males in later periods.

Introduction

Although women make up 51% of the management and professional workforce in the U.S., they make up a much smaller portion of top management positions. At Fortune 500 firms, only 14% of executive officers and only 4% of chief executives are females.1 The field of law displays a similar trend, with women making up 45% of associates but only 20% of partners at law firms in 2011.2 A sizable literature attempts to explain this gap in achievement between males and females at the highest levels of their professions. Popular explanations include discrimination by employers, and selection differences with males being more likely than females to pursue higher-level positions in the first place. More recently, Gneezy et al. (2003) and Gneezy and Rustichini (2004) present evidence that males respond more favorably than females to competitive incentives. Gneezy et al. (2003) pay college students to solve mazes, either independently or in competition with others, and find that competition increases the relative performance of males compared to females. Gneezy and Rustichini (2004) study footraces between fourth graders and find a similar result: boys run faster when racing against another student, while girls do not. If competitive pressure results in males performing better than females of similar skills, then males will tend to perform better than females in jobs that involve competition (e.g., lawyers competing in a courtroom, entrepreneurs competing for funding, executives engaging in competition against other firms), or when competing for promotions and advancement within their organizations.

Our contribution expands on Gneezy et al. (2003), Gneezy and Rustichini (2004) and related work by conducting a series of experiments in order to better assess the robustness of the observed performance gap between males and females in competitive environments. In particular, we conduct a series of in-classroom contests in which students compete against each other in answering math questions taken from grade-appropriate standardized tests.3 Our methodology is similar to the past analyses with one major exception: where other experiments ended after participants engaged in a single contest, we repeated our experiment multiple times. After one contest ended, we rematched students with a new opponent, and they competed again up to five times.

Although the literature studying gender differences in competition tends to use experiments involving non-repeated competitive interactions, this is not always typical in other areas of experimental economics. Experiments involving other subjects often repeat the experiment procedure with the same subjects over multiple periods, recognizing that first period (or one-shot) results are typically not representative of results after a couple periods of participation (e.g., Crawford, 2002, Ledyard, 1995).4 We find that similar concerns apply to experiments involving gender differences in competition; behavior in the first period does not represent behavior in subsequent periods.

We observe a significant male advantage during the first period contest. This is consistent with results from previous studies in which participants engage in a single contest. This is unsurprising given that the first contest in our experiment is the one contest which would be observed had we followed the literature and not repeated the contests multiple times. More surprising are our results concerning performance in later period contests; our primary contribution to the literature comes from our ability to observe performance across multiple contests.

We find no evidence of a male advantage in any contest after the first period. In all subsequent periods, after the participants have had one period for their nerves or excitement to calm, females tend to perform at least as well as males. In fact, in later rounds of competition we observe females outperforming males. Even when we return to classrooms and rerun the contests two weeks following the original experiment, we find no evidence that the initial male advantage reappears.

A closer look at our data reveals additional insight. The analysis breaks the base sample into three groups: high-, middle- and low-ability participants based on previous-year test scores. In all three ability groups, average male performance was higher than average female performance in the first period contest, with the gap being largest amongst the low- and high-ability groups. In later period contests, the performance gap between males and females fades. Interestingly, high-ability females perform no better on average than their middle-ability peers during the first-period contest; although they do perform better in later periods. The data is consistent with a story in which low- and middle-ability males “over-compete” and high-ability females dramatically “under-compete” in the first period, and that this over- and under-competition is not present in later periods.5

The main results come from the analysis of math contests in which participants compete to answer questions as accurately as possibly, and where ties are broken in favor of the participant with the fastest time. We ran additional treatments that varied the contest design or competitive task slightly, and find no evidence of a male advantage in any period, including the first. For example, we find no evidence of a male performance advantage in any period of competition when we deemphasize the race aspect of the competition, telling the participants that the contest is “not a race” and not breaking ties in favor of the participant with the fastest time. We also find no evidence of a male performance advantage in any period when the contest involves language arts questions rather than math questions. This suggests that the existence of the male advantage, even in the first period, is sensitive to the type of contest and the activity. These results compliment findings in other recent papers. For example, Shurchkov (2012) compares performance differences between males and females under high- and low-time pressure and in math and verbal task environments where the contest lasted one period. Consistent with our first-period results, she finds that the male advantage is greatest in high-time pressure, math settings and that females actually outperform males in low-time pressure, verbal settings.6 Also, Günther et al. (2010) identify a male advantage in maze competitions, but then find no significant difference between male and female performance in competitions involving word games. These results establish that the male advantage in competitive environments does not exist in some settings, but also that it does exist in other settings (e.g., math-based or maze competitions with time pressure, footraces).7 To the extent that workplace and career competition involves the type of competition for which the male advantage does exist, these results imply that the male advantage in competitive settings remains a realistic explanation for the observed gender gap in promotion and achievement.

Our primary contribution comes not from establishing that the male advantage is absent in some competitive settings. Rather, our primary contribution is to illustrate that even when the male advantage does exist (e.g., math contests with a time component), it is not robust to multiple periods of interaction. We find that the male advantage disappears after one period of competition, females tend to perform better than males after five rounds (less than an hour) of competition, and the male advantage does not reappear after even a two week break in competition. These results suggest that even when males initially outcompete females, their initial advantage is short lived, and may not by itself provide a viable explanation for long term achievement differences between males and females generally. More likely, the initial gender differences observed in competitive settings reflects differences in initial excitement about or aversion to competition (e.g., Niederle and Vesterlund, 2007),8 rather than an inherent difference in the ability of males and females to perform well under competitive pressures.

Others consider whether opponent gender affects one's performance. For example, Inzlicht and Ben-Zeev (2000) find evidence that one's performance on difficult verbal and math tests may depend on the gender composition of the group of people sitting in close proximity, even when they do not directly compete. Antonovics et al. (2009) showed that males were more likely to answer trivia questions correctly when a larger fraction of competitors were female. Price (2008) shows that competitive funding can affect time to candidacy in graduate school, with both males and females responding more positively to the competition when more of their peers are female. Although this question is removed from our primary focus, our data are consistent with these previous findings: we observe suggestive evidence that males perform slightly better when competing against females. Similarly, the theoretical literature on contests establishes that in a game theoretic equilibrium, participants should put in more effort the more-evenly matched they are with their opponent. Although our participants know the identity of their opponents, we find no evidence that opponent ability or ability differences have significant effects on one's performance.

Kuhnen and Tymula (in press) conduct an experiment in which they measure how many simple multiplication problems subjects can answer in 90 s, and then repeat the process 20 times. The focus of the paper is very different from ours. In the other paper, there is no direct competition between agents, as all participants receive a fixed payment regardless of performance. Rather, the authors show that those who receive feedback about their relative performance after each round tend to answer more questions correctly compared to those who did not receive feedback. This is consistent with the idea that those who received feedback were more likely to view the situation as a competition and put in more effort compared to those who did not view it as a competition. In our framework, on the other hand, the competitive incentives were the same for all participants, with prizes depending on relative performance and no participant receiving feedback until after the final period of competition. Note that Kuhnen and Tymula (in press) find that males perform better than females even after the first period of competition. This is perfectly consistent with our results; in addition to their setting not being a contest,9 there are also no controls for participant math ability. This means that any difference in average ability between males and females would appear as differences in average performance across all rounds. Other experiments consider contests that last multiple periods, with intermediate feedback coming between periods (e.g., Eriksson et al., 2009). They do not focus on gender differences, only noting that with feedback females may increase effort more than males in later rounds of competition.

Like Gneezy and Rustichini (2004) and Sutter and Rutzler (2010), we use experiments with children in an effort to learn about behavior in general. A number of articles consider whether children react rationally, and whether children behave like adults in economic experiments. For example, Wellman et al. (2001) show that three year olds understand that other people may have false beliefs, Wimmer and Perner (1983) show that before age five most children understand that false beliefs may result in people taking incorrect actions, and Gregan-Paxton and John (1995) show that by age seven children can make decisions accounting for both costs and benefits of their actions. John (1999) provides an overview of the literature and highlights that by age eight “children exhibit more thoughtfulness in their choices…employing a decision strategy that seems to make sense given the task environment” (p. 187). This is consistent with Harbaugh et al. (2007) who show that children as young as age eight play strategies and react similarly to adults in bargaining experiments. Most importantly for our analysis may be the results of Sutter and Rutzler (2010), who find that “the gender gap in competitiveness exists already very early in life, and does not change notably after the age of three years” (p. 2).10 Given these results, we see no reason to believe that our results do not extend to older population groups, although we leave formally testing this for future research.

Section 2 discusses the experimental design and the data. Section 3 analyzes the experimental data. Section 3.1 presents the initial analysis of the math contest data, showing that the male advantage disappears after the first period contest. Section 3.2 provides a more detailed look at the data in an attempt to understand the main result. Section 3.3 analyzes alternative treatments in which we downplay the race component of competition and use language arts questions rather than math questions on the quiz. Section 4 concludes with a discussion of the results.

Section snippets

Experiments and test score data

Working with school officials and teachers, we went into 24 elementary school classrooms to run a series of in-classroom, curriculum-based competitions. We described our purpose to the teachers and parents as examining changes in student performance in competitive settings, while helping the students practice for an up-coming state assessment test. In each period of competition, student participants were randomly paired with another classmate. The students were given a quiz with questions

Math contests with time component

Table 3 presents the results for the standard treatment using state assessment scores as the instrument for ability as described above. Regressions (1) and (2) use data from the first period of competition; (1) controls for the participant's own gender, while (2) also controls for opponent gender. Regressions (3) and (4) do the same for the second period of competition, and regressions (5) and (6) do the same for the combined data from periods two through five. (Table 1, Table 2 report on

Discussion

The main findings of our analysis are as follows. Males perform significantly better than females in the first period of math contests, but the male advantage disappears in subsequent periods of competition. In fact, males perform worse than females of similar ability in later periods. After a two-week break, we find no evidence that the initial male advantage returns.

A closer look at the data provides insight into the initial male advantage. First, we present evidence that in the first period,

References (28)

  • J.C. Cardenas et al.

    Gender differences in competitiveness and risk taking: comparing children in Colombia and Sweden

    Journal of Economic Behavior and Organization

    (2011)
  • U. Gneezy et al.

    Gender differences in competition: evidence from a matrilineal and a patriarchal society

    Econometrica

    (2009)
  • U. Gneezy et al.

    Performance in competitive environments: gender differences

    Quarterly Journal of Economics

    (2003)
  • U. Gneezy et al.

    Gender and competition at a young age

    American Economic Review

    (2004)
  • Cited by (36)

    • Competition, subjective feedback, and gender gaps in performance

      2023, Journal of Behavioral and Experimental Economics
    • The gender of opponents: Explaining gender differences in performance and risk-taking?

      2018, European Economic Review
      Citation Excerpt :

      round, where dollar values of all clues are doubled. Both parameters can relate to the accumulated experience of the contestants within the game and previous research has shown that gender differences may disappear in repeated competition (e.g., see Cotton et al., 2013). If such dynamics could indeed affect women’s performance differently than men’s, then our estimates may be biased if such a variable is not captured.

    • Self-selecting random or cumulative pay? A bargaining experiment

      2018, Journal of Behavioral and Experimental Economics
    • Gender differences and the effect of facing harder competition

      2017, Journal of Economic Behavior and Organization
    • Which explanations for gender differences in competition are consistent with a simple theoretical model?

      2015, Journal of Behavioral and Experimental Economics
      Citation Excerpt :

      We solve cases with two competitors and six competitors separately, showing how the number of competitors affects the predicted performance patterns in certain settings. We compare the theoretical predictions of the model with the empirical evidence from Gneezy et al. (2003) (which involved six competitor experiments) and Cotton et al. (2013) (which involved head to head contests between two competitors). We chose these papers due to data availability, because Gneezy et al. (2003) is the seminal paper on the topic, and because Cotton et al. (2013) involved an experiment involving head-to-head contests that closely resemble the underlying theoretical model with two players.

    View all citing articles on Scopus
    View full text