Gender differences in competition : are women taking requirements more serious?

(1)

GENDER DIFFERENCES IN

COMPETITION: ARE WOMEN

TAKING REQUIREMENTS MORE

SERIOUS?

Abstract

There is still a difference between men and women on the labor market. This difference might be present because men and women make different decisions about when to apply for a job. This thesis studies the relationship between posing requirements on a competitive environment and competition entry by women. This is done by an experiment at a Dutch high school, where participants in the control group are not confronted with requirements before making an entry decision, and participants in the treatment are confronted with requirements before making the decision to enter a competition. Main findings are that the requirements on itself have no effect on competition entry. Women are less likely to enter a competition. But women that are confronted with requirements are more likely to enter the competition. An explanation for this finding is sought by looking at confidence but the requirements does not have effect on the confidence of women in the treatment.

Lydia van der Vegt

10183329

Master Thesis Business Economics - Track Managerial Economics and Strategy 15 ECT

(2)

1

Statement of Originality

This document is written by Student Lydia van der Vegt who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

2

1. Introduction

To this day, women have a different position on the labor market than men. Women generally earn less than men, and are underrepresented in top positions (Bertrand & Hallock, 2001). Though the difference between men and women is partly due to discrimination on the labor market, men and women also tend to behave different in competitive environments. Different behavior might also affect the difference between men and women on the labor market. There is a popular statement saying men already apply for a job when they meet 60% of the requirements, when women only apply when they meet 100% of the requirements. Meaning that men apply when they are not completely qualified, while women only apply when they know they are qualified for a job. But to this day there is no scientific research to back up this statement. So do men and women react different to the job requirements in vacancies? Mohr (2014) claims that women are more likely to see requirements in a vacancy as guidelines or rules, causing women to take them more strict. Mohr also argues that women are socialized to follow rules more strict than men, and this might cause the difference in behavior of men and women on the labor market. The combination of women looking at requirements as guidelines or rules, and their socialization to follow rules can result in women not applying for the job they are qualified for, while underqualified men do apply and are in a position to get that job.

Applying for a job often involves competition, more people apply for the same job and the company picks the person that is best suited for the job. Therefore, economic experiments model the labor market as a competition. In these experiments, a common finding is that men are more likely to self-select into a tournament payment scheme, showing that men are prefer competition more than women do. A common explanation for the difference between men and women is confidence, where men seem to be overconfident about their ability to win the tournament and women are found to be under confident (Niederle & Vesterlund, 2007). This thesis further expands on this literature by looking at the effect of adding requirements to enter a competition. These requirements in the experiment are used to simulate the requirements in a vacancy. Since women might be socialized to “follow the requirements” more than men, it is interesting to see if there is a different response to adding requirements to a competition. Therefore, the research question is: “Do requirements affect the decision of women to enter a competition?”.

This question is answered through an economic experiment at a Dutch high school. In the experiment, participants make a decision to enter a tournament or not. In the main treatment, participants get a set of requirements before they enter the competition, while in the control

(5)

4

group these requirements are absent. The findings replicate the results found in other literature that find that women in general are less likely to choose a competitive payment scheme. Although when women are confronted with requirements they need to fulfill before they can enter the competition, women become more likely to enter a competition. Thereby the addition of requirements can be used as a way to close the gender gap found in competition entry. This could be explained because the requirements give upfront information about how likely someone is to do well in the competition. When looking at the effect of requirements on confidence, no effect is found.

The thesis is structured as follows. Section 2 provides an overview of related literature. Section 3 explains the methodology used in the experiment. Section 4 shows the results of the experiment and section 5 concludes.

(6)

5

2. Literature Review

Gender and competition is a widely studied subject in the past years. The focus of this literature review is on the most important papers, and will start with an analysis of literature on differences in preferences between men and women in section 2.1. Section 2.2 elaborates on explanations for the gender gap that is often found. Section 2.3 explains research on closing the gender gap. Section 2.4 explains research on how a gender gap is present for children at a young age. And section 2.5 elaborates on experimental research that is closely related to this paper.

2.1 Gender differences in preferences

A common hypothesis for gender differences is that men and women have different preferences when they are making economic decisions. Croson and Gneezy (2009) reviewed literature that studied gender differences in the domains of risk preferences, social preferences and competitive environments. They find a robust result for risk preferences, where men are structurally more likely to make risky decisions than women. This result is attributed to confidence, since men tend to be more often overconfident and women are more often under confident. For social preferences, they find that different literature finds different results regarding gender. This difference is mainly because women make decisions that are more context dependent than men which makes women more likely to react different to small differences in experimental design (2009, p. 464). Lastly they find that most literature on gender differences in competitive environments finds that men have a preference for competitive environments while women do not.

Common reasons for the differences in preferences for competition are: “backlash”, nature or nurture, and confidence. Backlash is described in Bowles, Babcock and Lai (2007). They study gender differences in outcomes in negotiations. They find that it is often rational for women not to initiate a negotiation and enter a competitive environment, because they end up worse off while men often end up better off after a negotiation. They find that women anticipate that negative outcome and that women act rational by avoiding this competitive environment.

Other common explanations for differences found in preferences in competition are part of a nature versus nurture debate which will be further elaborated in section 2.2.1. Another reason for finding a difference in behavior between men and women that is not part of Croson and Gneezy’s (2009) analysis, is confidence. Since it might be the case that men overestimate

(7)

6

their opportunities of winning a competition, while women underestimate their chances of winning. Confidence is further explained in section 2.2.2.

2.2 Explanations for the “Gender Gap”

This section will give an overview on research trying to explain the gender gap. The sub-sections will provide possible explanations for the differences found in competitive environments. Section 2.2.1 will describe research focused on explaining the differences according a nature or nurture argumentation. Section 2.2.2 will describe literature on confidence and how that might affect the decision made in competitive environments.

2.2.1 Nature versus Nurture

This subsection digs deeper into literature trying to explain possible explanations for the gender gap found in competitive environments. On the one hand, the literature explores the biological factors of the difference, while other literature suggest that socialization plays a role.

Biological differences can be measured through hormone levels, where the main idea is that hormonal balances have an effect on the choices made by women. This is studied by Wozniak, Harbaugh and Mayr (2014), they use the natural fluctuations in hormone levels caused by the menstrual cycle of women to distinguish whether the women that is participating is currently in a high hormonal phase, or a low hormonal phase. The authors find that women are more likely to select into a tournament payment scheme during a high hormonal phase. This is contrary to the findings of Buser (2012), who finds that women are more likely to select into a tournament during a low hormonal phase. These different findings could be attributed to a very different experimental design in both papers. Though both authors find that the biological hormonal levels play a role for women when selecting a tournament payout scheme or a piece-rate, the findings are not robust when different experimental designs are used. The main difference in design is that in the experiment of Wozniak et al. (2014) the participants play against participants that are in a different payment scheme. The participants make a choice about the type of payment scheme they want and get paid accordingly to their choices, but the opponents in a tournament can be in a different payment scheme option. This makes their problem not an individual decision problem, because the participants have to take the beliefs about the tournament scheme choice of their opponents into account. This is not the case in the experiment of Buser (2012). And could drive the different findings of Wozniak et al (2014).

Contrary to the argument that those differences can be attributed to biological factors, there is also research that suggests socialization plays a role in explaining the difference in

(8)

7

preferences. Gneezy, Leonard, & List (2009) examined the difference between matrilineal and patriarchal societies. They run an experiment in both the Khasi and Maasai tribe, where the Khasi is a matrilineal society, and the Maasai tribe a patriarchal society. They find that Maasai men are more likely to select into a tournament, than Maasai women. However in the matrilineal society they find that Khasi women are more likely to select into a tournament payment scheme than Khasi men. Anderson et al. (2013) also researched differences between patrilineal and matrilineal societies, but their research focuses on children. The matrilineal society is again the Khasi tribe, but the patrilineal society is studied through the Kharbi tribe. In both tribes the authors do not find a difference in competitiveness when children are younger than 12. In the matrilineal society, they find no difference in competitiveness between boys and girls older than 12, but in the patrilineal society they find that girls become less competitive when they get older than 12.

Differences between a matrilineal and patrilineal society is not the only way to study societal differences, as stereotypes are often different across societies and do not need to have a biological cause. Günther et al. (2010) study the effect of stereotypes on performance in competition. In western societies men are more often associated to be better in mathematical tasks, where women are more often said to be better in verbal tasks. So the authors look at the effects of stereotypical tasks on performance, where the manly task involves mathemetical problems, the neutral task involves a verbal task, and the female task involves memory performance and pattern matching. The findings of Günther et al. suggest that when participants are confronted with a manly task, only men increase their performance when faced with competition. In a neutral task both men and women increase their performance when faced with competition, and when faced with a typical female task only the women increase their performance.

The above findings suggest that both biological factors and societal factors can play a role in explaining the cause of different preferences between men and women. The difference found between patrilineal and matrilineal societies shows that societal differences do play a significant role. Anderson et al. (2013) find that the difference emerges when children get older than 12, the age puberty starts and hormonal levels change. Therefore, it can also be linked to biological factors explained by Buser (2012) and Wozniak et al. (2014). This experiment is conducted with children aged 14-18, so any difference found can be attributed to both societal and biological differences according to the literature cited above and no claims could be made about what causes any difference found regarding nature or nurture.

(9)

8

2.2.2 Confidence

Regardless of where the difference in competitiveness is stemming from: whether from biological differences between men and women, or societal differences affecting the behavior of people, confidence seems to be the common explanation. Confidence can be split up into two different types of confidence, absolute confidence and relative confidence. Absolute confidence is about the assessment whether one’s answer is correct, where relative confidence is about ranking oneself into a group of peers. The definition of relative confidence is relevant in the case of research regarding competition, since people need to take other people into account when deciding to enter a competition, and they need to be confident enough that they can win the competition before deciding to enter.

Not everybody is able to assess their skill level relative to peers correctly. Ola Svenson (1981) found that half of the drivers in Sweden think they are in the top 30% of high skilled drivers. In the US half of the drivers beliefs they are in the top 20%. This shows that at least some people are overconfident about their driving skills. Though this study did not differentiate for gender, so it is not sure whether overconfidence is mainly present in male’s behavior or in females’ behavior. Niederle and Vesterlund (2007) studied the behavior of men and women in a competitive environment and give overconfidence as one of their main explanations for the gender gap they have found. They find that men tend to be overconfident and women are more likely to be under confident about their relative performance.

Kamas and Preston (2012) extend the research of Niederle and Vesterlund (2007) regarding confidence by differentiating between 3 different types of confidence. They find that placement confidence, earlier referred to as relative confidence, is the most important type of confidence related to the gender difference in competition. Women expect to end at a lower place in the competition than men expect themselves to end up. They also tested for absolute confidence, and find no significant difference between men and women.

These articles suggest that confidence and especially overconfidence and under confidence plays a role in the difference between men and women in competition. This thesis builds on their findings, by looking at the role requirements have on confidence. Since the requirements that are attached to the competition give a signal to the participant whether they are more or less likely to do well in the competition.

(10)

9

2.3 Research on “closing” the gender gap

Besides research on showing there are differences between gender, other researchers are looking for ways to bring the behavior of men and women closer to each other and reduce the gender gap. This section elaborates on three ways research has sought solutions to close the gender gap. This is relevant for this thesis because adding requirements to a competition might also have an effect on widening or closing the gender gap found in competition.

Section 2.3.1 goes deeper into how team competition can help close the gender gap. Section 2.3.2 explains the role of performance feedback. Section 2.3.3 gives an overview on research done on gender quota.

2.3.1 Team competition

Team competition can have an effect on the participation in tournament because the decision to compete is not made by only one person. Dargnies (2012) studies the effect of team competition in a lab experiment. Participants can enter the competition as a team or alone. They find a large and significant difference in entry between men and women when people enter the competition alone. But when people enter the competition as team, they find that females enter the competition at the same rate as when they entered alone, but men enter significantly less than when they are in the competition alone. The main reason the authors give is that men become less sure about the performance of others and therefore don’t want to enter the competition. So performing in a team competition has the effect of closing the gender gap by influencing the choices of men and making them less likely to participate in the competition.

2.3.2 Performance feedback

In research looking at performance feedback, the participants get feedback on their absolute or relative performance. This has an effect on entry decisions because participants get more information on how good they are relative to other participants making their entry decision not dependent on their self-assessed confidence alone. Wozniak, Harbaugh and Mayr (2014) study not only the effect of the menstrual cycle on competition entry, but also look at performance feedback. In their experiment, the authors give the participants information on their relative performance compared to others. They find no significant difference in choices made by people who know their relative performance. They also look at the cost of making a concerning suboptimal choice and they find that the cost of women making a suboptimal choice is significantly higher than the cost of men making a suboptimal choice.

Ertac and Szentes (2011) also study the role of performance feedback on the entry decision. They find that the gender gap between men and women declines significantly when

(11)

10

participants are given feedback on their performance. They argue that this result is driven by ambiguity aversion, but how this mechanism works remains unclear. But feedback might have an effect on the decision to enter a competition made by women, because it gives them more information about their performance in the competition. The requirements added to the competition in this thesis might also give more information before entering a competition, since the requirements are based on how well someone can perform, therefore it can have an effect on the entry decision made by women.

2.3.3 Affirmative action

Another way to close the gender gap that is often studied is affirmative action. Gender quota is an affirmative action that is widely studied. In these experiments a certain percentage of the winners has to be female. Niederle, Segal and Vesterlund (2013) study affirmative action, specifically the effect of a gender quota, experimentally and find that adding a gender quota to the competition changes the decisions to enter significantly. Women tend to enter a competition more often, while men enter the competition less. This result is replicated by (Czibor & Dominguez-Martinez (2017). Both authors show that a policy change in the end of the competition can affect the entry in the beginning, while this thesis tries to find whether a change at the beginning of the competition has an effect on the entry in a competition. Though both the gender quota as well as the requirements before entering a competition give extra information about the chances to win the competition before deciding to enter. The quota gives women certainty that they have a higher chance to win the competition, while the requirements before the competition give information about how likely it is to do well in the competition.

So a difference in the gender gap found in competition entry decisions often have to do with more uncertainty, in the case of team competition, or more information, in the case of personal feedback and the gender quota. More uncertainty affects the choices of men, because men are less likely to choose competition when they enter in a team, while more information affects the choices of women, and makes it more likely for women to enter a competition. Posing requirements upon a competition that need to be fulfilled before someone can enter, gives information about which type of person would do well in the competition, and thus is likely to affect the choices of women in their decision to enter.

2.4 Competition at a young age

Since the research conducted in this paper is done with high school students, this section explains literature on how findings might differ when using children aged 14-18 instead of

(12)

11

university students as subject pool. Willingness to compete might differ with age, when it is mainly social factors that cause a difference between men and women. Then it is likely that a difference between boys and girls happen later in life, since the children have to learn what the difference is in how they should behave. When a difference in preference for competition is already present at a young age, it is more likely that the difference is due to biological factors. Either way it is necessary to know if other researchers found a difference already at age 14-18 since this might shape the results in this thesis.

Sutter and Glätzle-Rützler (2015) studied children aged three to 18, and found a difference between boys and girls as early as children aged three. They find that boys from three to 18 are more likely to choose a competition, and girls are less likely to shy away competition. They also studied some subjects 2 years after their first experiment and found that this difference persists over time. Anderson et al. (2013) also studied differences in age, but also between a patrilineal and matrilineal society. They find no significant differences in both societies until the children are around 13-14 years old. But when they hit puberty around that age, they find a difference in the patrilineal society, where girls shy away the competition while boys enter at the same rate as when they were younger. They find no difference between boys and girls in the matrilineal society.

Gneezy and Rustichini (2004) also studied competition at a young age in a field study. Their subject group was around 9-10 years old, and had to perform a running task. The first round there was no difference between the performance of boys and girls, though when the competition was added, the boys ran significantly faster than girls. So the competition causes an increase in performance for boys. Because the authors find this difference in performance before puberty, they argue that the difference does not come from socialization and that it is more likely that biological factors play a role. Because the research in this paper is done with subjects that already hit puberty, it is likely that a difference will be found, though the source can be both biological or because of socialization.

2.5 Experimental evidence

Literature on gender differences in competition are mainly based on lab experiments, though also some field experiments have been conducted. This part will give a review of both types of experimental evidence.

It is often hypothesized that gender differences on the labor market are due to different reactions of men and women on competitive environments. Gneezy, Niederle, & Rustichini (2003) study this in an experiment. In their experiment they find that men and women, who are

(13)

12

able to perform equally in a non-competitive environment, perform differently in a competitive environment. The participants are asked to solve mazes under different payment systems, varying from non-competitive to competitive. They find that women do not significantly increase their performance when they competing against a mixed gender group, while men do increase their performance significantly in that situation. It is remarkable that when women are in a same sex tournament, their performance is increased significantly. This implies that for women not only the competition itself is important, but also who they are competing against.

But performance is not the only difference found between men and women in a competition. In the experiment of Geezy, Niederle & Rustichini (2003) people are put in a certain compensation scheme, therefore their results do not reflect the preferences of men and women. On the labor market the differences might be due to different preferences, because people have the ability to “self-select” into certain jobs that might be more or less competitive by applying to jobs with different competitive specifications. Niederle & Vesterlund (2007) conducted an experiment where participants are able to choose their preferred payoff scheme. They selected a task where men and women perform equally. When they let participants select their preferred payment scheme, men are significantly more likely to select the payment scheme that involves competition, while women are more likely to select a piece rate. They find that confidence is the main factor explaining this difference. Men are more likely to be overconfident about their performance while women seem to be under confident.

Flory, Leibbrandt and List (2015) also find a difference in preferences for competition between men and women in their natural field experiment. They randomized almost 9000 job seekers into vacancies with different compensation regimes. The compensation regimes varied in competitiveness, and the authors find that when the compensation regime in the vacancy is more dependent on competition the applicant pool becomes more male dominated. They find that the gender gap is not necessarily present because men like competitive workplaces better than women, but that women have a stronger aversion to competition compared to men (Flory, Leibbrandt, & List, 2015, p. 135). This result extends the analysis of Niederle and Versterlund (2007) because Flory et al. provide a clear link to the labor market. Though the vacancies used in their experiment only vary in compensation scheme, they are not able to distinguish what the roles of “requirements” or criteria in the vacancies are in determining the choices made by their participants.

Another experiment that also links competition to the labor market is from Buser, Niederle and Oosterbeek (2014). Their experiment is conducted with Dutch high school students and they find that boys are more likely to choose a science based tracks, which are

(14)

13

more mathematically challenging whereas girls are more likely to choose a track with easier math. Those track choices are a good indication of future career choices, since math and sciences are highly correlated with college attendance and completion. In their experiment they test competitiveness, and the track chosen by the participants, and the authors find a high correlation. This indicates that competitiveness is correlated with career choices and this effect is already causing self-selection in high schools.

The research mentioned in this subparagraph shows that multiple studies found differences in gender regarding preferences for competition. Especially the research of Niederle and Vesterlund (2007) is very similar to the research in this paper. This paper uses a similar experimental design, but requirements are added before subjects make the decision to enter the competition.

(15)

14

3. Model

Section 3.1 explains the hypotheses tested in this thesis. Section 3.2 elaborates on the experimental design and section 3.3 shows the characteristics of the participants.

3.1 Hypotheses

This thesis is built upon the following hypotheses.

Hypothesis 1: The addition of requirements to the competition will cause women to enter less often than men enter the competition.

In the literature described in the previous section, authors find a very robust result that women are less likely to enter a competition. Combined with the statement that women only apply for a job when they meet 100% of the requirements while men already apply when they meet 60% of the requirements and the statement of Mohr (2014) that women are socialized to follow requirements more often, leads to the hypothesis that adding requirements to a competition causes women to enter less often than men do. It is unlikely that women think they meet all the requirements, and therefore they decide not to enter the competition.

Hypothesis 2: The requirements work as a signaling device, decreasing confidence for women and increasing confidence for men.

Since numerous authors in the previous section name confidence as one of the main reasons for women to enter a competitive environment less often, it is likely that the addition of requirements have an effect on confidence. Because requirements give more information about how likely it is someone will do well in the competition. Since it is expected that women enter the competition less often when there are requirements attached, it is likely that the requirements decrease their confidence about their ability in the competition.

3.2 Experimental design

The aim of this paper is to test wether requirements have an effect on the behavior of women in entering a competitive environment by means of an experiment. The experimental design is based on Niederle and Vesterlund (2007) with the main differences that their experiment used a within subject design and this experiment will be a between subject design. This means that in this experiment subjects are randomly matched to be either in the control group or in the treatment group. This experiment adds requirements before making the decision to enter the competition as the main treatment and when using a within suject design the purpose of the research might be too clear for the participants, which might influence results. Following the

(16)

15

design of Niederle and Vesterlund (2007) gives the opportunity to compare results with earlier research.

The experiment consists of four rounds. In the first round participants have 5 minutes to solve as many math problems as possible and will be paid according to a piece rate of € 0.50 per correct answer. The problems are adding up five random two digit numbers, since ability in this task is not necessarily dependent on gender. The problems look as follows:

Problem:

1

2

3

4

5 Answer

1.

84

90

81

21

58

Table 1: Example of exercises in Experiment

The participants had to fill in their answer in the last column.

In the second round, participants play a tournament. The participants get 5 minutes to complete the same type of questions as in round one. Participants are randomly matched into groups of four. They are not able to know the other participants in their group, this means they do not know who they are competing against. The winner of the tournament is the one that gets most answers correct of all players. In case of tie, the winner is determined randomly. The winner gets paid €1.50 per correct answer, the other players receive nothing. The first two rounds are the same for the treatment and the control group.

The people in the experiment do not know who the four people are that they are competing against, because this might influence their decision to participate. Datta Gupta, Poulsen, & Villeval (2013) show that the decision to play a competition or not depends on the other players in the competition. In this particular experiment, all participants are classmates instead of being put together for the experiment with people they have never seen before. This means that they already know each other really well on a personal level. If participants in this experiment know who they are competing against their decision might be influenced by the knowledge they have about the personal characteristics of their classmates. Therefore in the entire experiment no one is aware which people they are competing against. This is done by handing participants a number that they have to keep secret during the entire experiment, the number contained information about who is in which group, but the participants do not know the numbers of the other participants.

In the third round, participants are able to choose their preferred payout scheme. They can choose whether they want to get paid according to the piece rate, or if they want to compete in the tournament. If a participant wants to get a piece rate, they get paid the same as in round one, €0.50 per correct answer. If a participant chooses the tournament they play against the

(17)

16

same people as in round two, and the results of those players of round two. This way, the competitors of the participant are also in the tournament payment scheme and the decision to compete is not dependent on the beliefs about other participants competing in round three, making this an individual decision problem.

The participants in the control group only make the decision to enter the competition or not, and then make the exercises. The treatment group gets a set of requirements they need to fulfill before they can enter the competition. The requirements are used to simulate the requirements that are often stated in a vacancy. In a vacancy a company sketches their ideal candidate through a mixture of “soft” personality based skills and “hard” skills. To simulate this, three of the six requirements in the vacancy are based on personality and the other three are based on mathematical ability. The six requirements used are:

Personality based requirements - Eager to win attitude - High level of perseverance

- The ability to work concentrated on single task Skill based requirements

- Good mathematics grades

- Preferably a high science level high school course profile (NG/NT profile at Dutch high school level)

- Good numerical understanding

The skill based requirements are formulated quite vaguely, “good mathematics grades” instead of “an average math grade of 7” to let the personal assessment of these requirements play a more important role, which is also often the case with vacancies. Before making the decision to enter the tournament, participants are asked to mark every requirement they think applies to them. This makes it able to see if different people who think they fulfill the same amount of requirements make the same decision about tournament entry, and more importantly if this decision differs across genders.

In the fourth round, participants again need to make a decision which payment scheme they would like to apply to their performance. This payment scheme will be applied to their performance in the first round, so they do not have to exert any effort in this round.

The experiment is conducted with pen and paper. Every round the participants get a new set of instructions necessary for that round. They also get a sheet containing the excercises for that round. The participants could in total make a total of 30 problems every round, this is to make sure they do not run out of problems during the 5 minutes and it makes them able to skip

(18)

17

problems they find hard to solve. The experiment is based on a real effort task because this gives the ability to show how a different payment scheme affects real effort and the performance of the participants.

3.3 Participant Characteristics

The experiment is conducted with high school students. All participants are in their 4th_{year of}

high school and follow either Senior General Secondary Education (HAVO) or pre-university education (VWO). All students follow the course Economics at either Havo or VWO level, though the students chose different track in high school. The most common track in the sample is Economics and Society (EM) with 106 students, second largest is Nature and Technology (NT) with 15 students, and the smallest track is Nature and Health (NG) with 4 students. Because both the Nature and Technology and the Nature and Health track contain science classes like chemistry, physics, biology and higher level mathematics the people in these tracks are pooled together under Science track in the further analysis.

Table 2 shows a summary of characteristics, both in the control group, the treatment group and the total is given. In total 125 high school students participated, these students where either in the control group or in the treatment group because of the between subject design of the experiment. Of these students 22 followed pre-university education and 103 followed Senior

Variable Treatment Control Total

Sample Average Age 16.02 15.92 15.97 #Female 33 28 61 #Male 29 35 64 VWO 0 22 22 HAVO 62 41 103 Economics track 49 57 106 Science track 13 6 19 Math 13 7 20 Sport 52 50 102 SportCompetitive 40 41 81 SportCooperative 36 39 75 Hobby 26 24 50 HobbyCompetitive 8 15 23 HobbyCooperative 8 10 18 Total observation 62 63 125

Table 2: Total sample descriptives

Notes: This table shows total observations in both the treatment group, the control group and the total sample

(19)

18

General Secondary Education. 20 students have a higher level mathematics course. 81 participants played Sports, of which 81 participants labelled their sports as competitive and 75 labelled their sport as cooperative. 50 participants indicated that they have a hobby, of which 23 labelled their hobby as competitive and 18 labelled their hobby as cooperative.

(20)

19

4. Results

This section will explain the results of the experiment. Section 4.1 will explain the randomization in the experiment. Section 4.2 gives results on performance in the experiment. Section 4.3 elaborates on the results on the first hypothesis, the preferences for competition. Section 4.4 eexplains the results of hypothesis two, that the requirements affect confidence of participants. Variable Regression type Observations R-squared P-value F-Test Constant OLS Regression 0.212 125 0.557 (-0.9613) <0.001 Notes: This table shows the results of a linear regression on the treatment variable, Significance levels shown with an *, * significant at the 10% level, ** significant at the 5% level and ***significant at the 1% level. The p-value of the F-test gives the significance of the F-test for joint significance of all independent variables Hobby Competitive -0.044 (0.1507) Hobby Cooperative -0.052 (0.1476) Sport Cooperative 0.034 (0.1390) Hobby 0.161 (0.1114) Higher Level Math

Sport 0.085

(0.1391)

Sport Competitive -0.096

(0.1542) Table 3: Regression to check

randomization Treatment Age -0.004 (0.0589) Female 0.009 VWO Science Track 0.260 (0.1868) (0.1188) 0.052 (0.1889) (0.0922) -0.680***

(21)

20

4.1 Randomization in the experiment

The experiment is done during the regular classes of the students. This could lead to randomization problems the between subject design causes a class to be either in the control group or the treatment group. As can be seen in table 2, there are some discrepancies between the treatment and control group regarding high school level, high school track and higher level mathematics. To test whether these differences are significant, all variables are regressed on treatment. If randomization between the control and treatment group is executed correctly, no coefficient should be statistically different from 0. The results of this regression are shown in table 3.

Table 3 shows that the coefficient for VWO (pre-university education) is significant at the 1% level. The coefficients for science track and higher level mathematics are not statistically significant. This result can be further elaborated by checking whether all coefficients are jointly significant. This is done by an F-test which is also shown in table 3, the p-value for the F-test is smaller than 0.001, indicating that all coefficients are jointly significantly different from zero. Meaning that randomization in the experiment failed and it failed at the high school level.

Because randomization failed at high school level, the observations of the participants who are in the VWO class are removed from the analysis. This causes the number of observations in the control group to fall from 63 to 41, leading to larger standard errors in the control group. But dropping the VWO observations from the data makes the control and the treatment group more comparable and will lead to less biased estimators. Since the VWO students are only in the control group dropping the VWO observations does not matter for any analysis done only on the treatment group.

4.2 Performance

The experiment consists of four rounds, during the first three rounds the participants had to make exercises in which they added 5 random 2 digit numbers. Their performance is measured by the total answers correct in each round. The maximum a participant could get was 30 answers correct. Table 4 shows the average correct answers in each round. In the first round the participants were in a piece rate payment scheme, in the second round participants were in a competitive payment scheme, and in the third round participants could choose their preferred payment scheme.

(22)

21

Besides the average answers correct table 4 also shows p-values of a two sided t-test on differences in correct answers between gender within the control or treatment group and differences between the rounds within the control and treatment group.

Within the control group, women have a significantly lower performance than men in the second round, this is statistically significant at the 10% level. So in the control group, women that are confronted with a mandatory competition scheme are performing worse than

Average Correct Female

Average Correct Male

P-value difference female and male

First round 6.70 (1.0434) 7.78 (0.6394) 0.361

Second round 6.30 (0.6011) 8.27 (1.0187) 0.081

Third round 7.87 (0.6692) 8.56 (1.0640) 0.573

p-value difference first and

second round 0.52 0.39

p-value difference second

and third round 0.01 0.63

third round 0.06 0.17 Average Correct Female Average Correct Male p-value difference female and male

First round 6.72 (0.4672) 7.62 (0.6187) 0.247

Second round 8.38 (0.7164) 7.03 (0.5303) 0.13

Third round 8.24 (0.7254) 8.09 (0.5407) 0.87

second round 0.56 0.16

third round <0.01 0.32

Control group Treatment _{Control and Treatment}p-value difference

First round 7.17 (0.5801) 7.14 (0.3826) 0.97

Second round 7.17 (0.5737) 7.66 (0.4429) 0.50

Third round 8.17 (0.5935) 8.16 (0.4413) 0.99

second round 1 0.16

third round 0.02 0.01

Notes: This table shows the average answers correct in each round, seperated by control and treatment. Within these categories averages for male and female are given, as well as the overall averages within the treatment and control group. P-values correspond to a two sided t-test. The numbers between parenthesis are the corresponding standard errors.

Table 4: Average of correct answers in treatment and control group

Treatment Control

(23)

22

man that are confronted with the competitive payment scheme. These differences are not found in the Treatment group. This might be due to the slightly higher number of females in the treatment group that are in a science track and have higher level mathematics in high school. When looking at the overall result of men and women combined there is no statistically significant difference in performance between the Treatment and the Control group.

For men there is no significant difference in scores between the different rounds of the experiment in both the control and the treatment group. So male performance is not influenced by the different payment schemes. Women on the other hand perform significantly better in the third round than in the first and second round, and no significant difference is found between the first and the second round, this result is found in both the treatment and the control group. The better performance in the third round could be explained by the fact that women are able to choose the payment scheme under which they can perform best. So women who are better in a piece rate payment scheme can choose a piece rate, and women who are better in a competitive payment scheme can choose a competitive payment scheme and therefore maximize their output.

4.3 Preferences for competition

This section explains findings for the first and main hypothesis in this paper. The hypothesis is that the addition of requirements causes women to enter the competition less often. First, the overall differences between treatment and control will be explained. Second, the differences between gender in the control and treatment group will be explained.

Table 5 shows the proportions in both the treatment and control group that choose a competitive payment scheme in both round three and round four in the experiment. Also, the p-values for the two-sided t-test for equal proportions in the treatment and control group are given. Overall there is an increase in the proportion choosing competition when requirements are shown before the competition in both round three and four. Though the p-value for this increase is 0.19 in both rounds, therefore overall there is no significant difference in choice for

Round 3 Control Round 3 Treatment p-value two sided t-test round 3 p-value t-test C<T round 3 Round 4 Control Round 4 Treatment p-value two sided t-test round 4 p-value t-test C<T round 4 Female 0.09 (0.059) 0.36 (0.084) 0.02 0.01 0.13 (0.070) 0.33 (0.082) 0.08 0.04 Male 0.56 (0.117) 0.48 (0.093) 0.63 0.67 0.33 (0.111) 0.34 (0.088) 0.94 0.48 Overall 0.29 (0.071) 0.42 (0.063) 0.19 0.10 0.22 (0.065) 0.34 (0.060) 0.19 0.10

Table 5: Differences in proportion in choice for competition

Notes: Numbers shown in the table are the proportion that chooses for competition in the control and treatment group. The p-value corresponds to a two-sided t-test for equal proportions. Numbers within parenthesis are

(24)

23

competition in both treatments. But when testing if the proportion in the control group is smaller than the proportion in the treatment group, the p-value shows that this is the case at the 10% level in both the third and fourth round in the experiment. So overall the proportion that chooses competition is smaller in the control group than in the treatment group. For men there is no statistical difference found in choice for competition.

Counter to what was hypothesized, the proportion of women choosing competition when there are requirements added to the competition is larger. The proportion of women that chooses competition in round three in the control group is 0.09, while when the requirements are added the competition, the proportion choosing for competition is 0.36. When looking at the p-values, both the two sided and the one sided t-tests are significant at the 5% level, indicating that women enter a competitive environment more often when there are requirements attached to the competition.

The robustness of this result is tested through a probit regression that includes several control variables. Appendix 1 shows the correlations between all independent variables. This table shows high correlations between the interaction variable Fem*Requirements and Female, and between Fem*Treatment and Requirements. This can be explained by the fact that Fem*Treatment is an interaction variable that consists of the variables Female and Requirements. More important is the high correlation between Higher Level Math and Science Track, this correlation is 0.79, which indicates that these variables are subject to multicollinearity. This is because higher level mathematics often is part of the Science Track courses in high school. Appendix 2 shows regression results of the probit regressions used to decide whether to drop the variable Science Track, or Higher Level Math. Because the log-likelihood is larger when Higher Level Math is used, the variable Science Track is dropped as an independent variable and Higher Level Math is used for estimating the choice for competition. All other correlations show no sign of multicollinearity.

Table 6 shows results of a probit regression analysis for choice of competition. Since choice for competition is a binary variable, a probit regression is used. The first regression shows only the treatment effect of adding requirements to a competition. The positive number indicates that adding requirements to the competition makes it more likely for people to choose a competitive environment, though this is not significant. This result stays positive when the variables that indicate being female and an interaction variable showing the effect of being a female in the treatment group are added to the regression. Though when other control variables are added the result for the treatment effect becomes negative this indicates that adding requirements to the competition makes it less likely for people to choose a competitive

(25)

24

environment. Though neither coefficient is statistically significant, so overall the treatment itself does not have an effect on the choice for competition.

Being a woman makes it less likely for a participant to choose a competitive environment in the second and the third regression. In both regressions the parameter is negative and significant at the 1% significance level. This negative effect is in line with earlier research by Niederle & Vesterlund (2007), Gneezy, Niederle, & Rustichini (2003), Flory, Leibbrandt and List (2015) and Buser, Niederle and Oosterbeek (2014).

The effect of being a woman that is confronted with requirements before deciding to enter the competition is captured by an interaction variable between treatment and female. This coefficient is positive in both regressions, and significant at the 5% level. This means that a woman in the treatment is more likely to choose a competitive environment than a woman in the control group, which remains stable when control variables are added. This is counter to the

Variable

Regression type Observations LogLikelyhood Higher Level Math

-0.546*** Competitive Sports

Constant Age Fem*Rules

Table 6: Regression results for choice for competition Choice for competition Choice for competition Female Choice for competition Requirements 0.342 (0.2629) 0.183 -0.328 (0.3788) (0.3906) -1.500*** -1.524*** (0.4774) (0.4999) 1.194** 1.294** (0.5769) (0.5902) (0.4204) 0.688 0.435** (0.2040) 0.296 (0.2964) Competitive Hobby -0.102 (0.4051) -0.140 -7.058** (0.1681) (0.2979) (3.2900) -66.951 -60.875 -56.6221

Notes: Significance levels shown with an *, * significant at the 10% level, ** significant at the 5% level and ***significant at the 1% level. The numbers in parenthesis are robust standard errors

Probit Probit

103 103 103

(26)

25

hypothesis that the addition of requirements to the competition will cause women to enter less often.

An explanation for this result can be that the women in the treatment group get more information about how likely it is that they will do well in the competition. On average, women in the treatment group indicated that 3.97 out of 6 requirements apply to them. When a woman indicates that more than half of the requirements apply to her, she might feel more confident about her chances of winning the competition. This results can thus be driven by a similar force that drives the results of Ertac and Szentes (2011), where women get feedback about their performance, or what happens when women are confronted with a gender quota (Czibor & Dominguez-Martinez, 2017) (Niederle, Segal, & Vesterlund, 2013). In both the performance feedback case and the gender quota case, women get more information about their chances to win the competition, which influences their decision to enter. The requirements could have worked as a way to give the women more information about how likely they are to succeed in a competition, causing them to choose the competition more often. Further explanations for the exact effect requirements have on confidence are given in section 4.4.

The effect of age is also positive and significant, indicating that older participants are more likely to enter a competition than younger participants. In line with what was expected based on Buser, Niederle and Oosterbeek (2014), higher level mathematics students are more likely to choose competition, because this coefficient is positive though it is insignificant.

Competitive sports and competitive hobbies both have no effect on the choice of competition. A sport is marked as competitive when the sport has a competitive element in it, for example tennis, where a big part of the sport is playing matches against other people. This could indicate Hawthorne effects, because people who are competitive in their day to day life, by playing a competitive sport or have a competitive hobby, are not more likely to choose a competitive payment scheme in the experiment. Indicating that people might make a different decision because they are in an experiment. But playing a sport or having a hobby might contain very different tasks than the task in this experiment, so someone who for example plays tennis might not be good at doing mathematical problems and is therefore not choosing a competition. Besides, when someone is playing a sport, people also train to get better, and may therefore be more likely to be competitive in that case, while in this experiment people don’t have the opportunity to train the task in this experiment.

Because there are some problems with randomization in the experiment regarding high school level, the regressions in table 6 show the results without the VWO observations in the treatment group. To check for robustness of the results with these observations, the regressions

(27)

26

are also done with the observations of the people that are in VWO, this table is shown in appendix 3. Adding the VWO observations to the regressions does not change any of the results found above. The treatment still has no significant effect on the choice for competition. Being female makes it less likely to choose competition, this effect is significant at the 5% level. Being a female confronted with the requirements in the treatment makes it more likely to choose competition, which is significant at the 10% level. Age makes it more likely to choose competition. The coefficient for higher level mathematics is still positive, tough becomes significant at the 10% level. Competitive sports and hobbies still have no significant effect on the choice for competition. So with the observations of people who follow the VWO level of high school the effect of being a female and being a female in the competition does not change from the original analysis. Also, all effects found are still in line with the effects found in other literature.

The next question that arises is does it hurt a woman’s payout not to choose a competition payout scheme. In other words, could women be better off by making another decision? Someone is optimizing their behavior if they chose the competitive payment scheme in round three and also won the competition. In Figure 1, the percentages of the winners that chose for competition and the winners that chose a piece rate in the third round are shown. Only 28% of the female winners chose the competitive payment scheme, and thus, optimized their behavior. This means that the 72% of the female winners of the competition could have gotten a higher payout if they made a different choice. The figure also shows that male winner chose the competitive payout scheme more often than women, 73% of male winners optimized their behavior by choosing the competitive payment scheme.

(28)

27

Does the addition of requirements to the competition help women to optimize the decision and choose for competition? The effect of the treatment on women is given by the interaction variable in table 7. The treatment itself has an insignificant effect on the choice for competition by the winners. Being female winner makes it less likely that the competitive payment scheme is chosen, this effect is significant at the 5% level. Being a female winner in the treatment has no significant effect on the choosing competition. Thus adding requirements to the competition does not help female winners to optimize their decision for competition.

Table 7 also looks at the losers of the competition. Because for a loser, it would be optimal to choose the piece rate payment scheme, then they get €0.50 per correct answer, where as they chose a competitive payment scheme they got nothing. In this regression you can see that the requirements do not have an effect on choosing competition. Being a female loser makes it less likely to choose the competitive payment scheme, and therefore they are more likely to choose the optimal payment scheme. But being a female in the treatment does not have a significant effect on the choice for competition by the losers, and therefore the treatment has no effect on women optimizing their payout.

Variable Regression type Observations LogLikelyhood 32 -17.653 1.564 (1.0245) 1.068 (0.5861) Probit regression -39.688 Notes: Significance levels shown with an *, * significant at the 10% level, ** significant at the 5% level and

***significant at the 1% level. The numbers in parenthesis are robust standard errors

Female -1.185* (0.6256) Probit regression 71 Constant -0.349 (0.3864) Female*Treatment 1.008 (0.7361) -2.135** (0.8289)

Table 7: Choice for competition by the winners and losers in round 3 Competition Requirements 0.095 (0.4792) Losers Winner Competition -0.637 (0.7282)

(29)

28

4.4 The role of confidence

This section discusses the results for the second hypothesis that the requirements work as a signaling device, decreasing confidence for women and increasing confidence for men. First the results on absolute confidence are given, than the results on relative confidence are shown. Confidence is measured in two ways. The first level of confidence, called absolute confidence, and is measured by asking participants how much questions they have correct. The second level of confidence is relative confidence, which is measured by asking the participants their expectation about the place they ended in the competition. This is measured only for people who are in the competitive states in the experiment, so in the third round, people only filled in this question if they entered the competition or not. No counterfactual beliefs are measured, so participants who chose a piece rate did not fill in which place they expected to end if they had chosen competition.

Because the experimental design for the control and the treatment is completely similar in the first two rounds, no difference in confidence is expected in these rounds. The requirements are only added in the third round of the competition, so if the requirements affect confidence there should be a difference in confidence between the control and treatment group in round three. Table 9 shows the results of a linear regression on confidence in absolute confidence. Absolute confidence is measured by:

𝐴𝑏𝑠𝑜𝑙𝑢𝑡𝑒 𝐶𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 = # 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑐𝑜𝑟𝑟𝑒𝑐𝑡 𝑎𝑛𝑠𝑤𝑒𝑟𝑠 − # 𝑎𝑐𝑡𝑢𝑎𝑙 𝑐𝑜𝑟𝑟𝑒𝑐𝑡 𝑎𝑛𝑠𝑤𝑒𝑟𝑠.

Someone is overconfident when his estimation about the number of correct answers exceeds the correct answers he had, leading to a positive number for confidence. Someone is under

First round Second Round Third round

Confidence Confidence Confidence

Requirements 0.297 0.931* 0.979 (0.9898) (0.5422) (0.6168) Female -1.814* 0.523 -1.273* (1.0713) (0.6503) (0.6846) Female*Treatment 0.997 -0.907 -0.640 (1.2195) (0.9294) (0.8917) Constant 0.944* 0.00 1.05** (0.9012) (0.3583) (0.4906)

Regression type OLS OLS OLS

Number of observations 103 103 103

R-squared 0.078 0.016 0.144

Table 9: Differences in absolute overconfidence

(30)

29

confident when the estimated number of correct answers is lower than the correct answers someone actually had, leading to a negative number for confidence.

The results in Table 9 show that for absolute confidence the addition of requirements to the competition has no significant effect. The effect of being female is negative and significant at the 10% level in both the first round and the third round, indicating that women are under confident in these rounds. On average women in the first round estimate they have 1.18 fewer answers correct than they actually did, making them under confident about their absolute performance. In the third round they estimate their performance is 1.27 answers fewer than their actual performance. There is no significant effect in the second round. So when competition is mandatory, there is no difference in absolute confidence between men and women. When competition is absent, in the first round, or optional, in the third round, females tend to be under confident in their performance.

Table 10 shows the results on relative confidence. This confidence is measured by the difference between the place someone ended in the competition and the place this person estimated he ended. So the equation looks as follows:

𝑅𝑒𝑙𝑎𝑡𝑖𝑣𝑒 𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 = 𝑒𝑠𝑡𝑖𝑚𝑎𝑡𝑒𝑑 𝑝𝑙𝑎𝑐𝑒 𝑖𝑛 𝑡ℎ𝑒 𝑐𝑜𝑚𝑝𝑒𝑡𝑖𝑡𝑖𝑜𝑛 − 𝑎𝑐𝑡𝑢𝑎𝑙 𝑝𝑙𝑎𝑐𝑒 𝑖𝑛 𝑡ℎ𝑒 𝑐𝑜𝑚𝑝𝑒𝑡𝑖𝑡𝑖𝑜𝑛

When relative confidence is positive this indicates the person is being overconfident about their place in the competition, if relative confidence is negative, the person is being under confident about their place in the competition. The coefficient for requirements is not significantly different from zero in both rounds, meaning that the addition of requirements has on average no effect on relative confidence. In the second round being a woman decreases relative confidence because on average women guessed they ended 1.457 places lower than they actually ended in the competition in the second round. Women in the treatment are significantly overconfident and guess they ended 1.444 places higher than they actually ended in the competition in the second round. Diminishing the effect of being a woman and making them on average as confident about their place as men are. Though this effect is already present in the second round, where there are no differences in experimental design between the control group and the treatment group.

The negative effect of being a women is still significant in the third round making women on average under confident about the place they ended in the competition. The effect is smaller than in the second round, because in the third round the question about the beliefs on the place someone ended up in the competition is only asked to people who actually chose for competition, meaning the effect could become smaller because of a self-selection bias. People that are confident are self-selecting into the competitive payment scheme and the people that

(31)

30

are on average less confident self-select into the piece rate. This indicates that confidence in itself plays a role when people make the decision for selecting a competitive payment scheme or a piece rate. But that the requirements in itself has is not affecting confidence.

Second Round Third round

Confidence Confidence Requirements -0.397 0.126 (0.4014) (0.3203) Female -1.457*** -1.031* (0.4550) (0.5243) Female*Treatment 1.444** 0.959 (0.5531) (0.6889) Constant 0.5 0.231 (0.3328) (0.2012)

Regression type OLS OLS

Number of observations 103 51

R-squared 0.128 0.132

Table 10: Differences in relative overconfidence

(32)

31

5. Discussion

This last paragraph provides a discussion of the main findings. Section 5.1 gives a discussion of the main findings in this thesis. Section 5.2 shows results for an additional analysis about how strict women take the rules.

5.1 Discussion of the main results

This thesis tried to answer the question “Do requirements affect the decision of women to enter a competitive?”. Because one of the reasons there still is a gender gap in the labor market could be that women do not apply for jobs they could be qualified for because they take the requirements in a job vacancy more serious than men do. This is studied by means of an experiment, where the subjects are assigned to either the control or treatment group. In the control group, subjects are not confronted with requirements before making the decision to enter a competition or not, while in the control group subjects get a set of requirements they need to fulfill before making the decision to enter the competition. The main finding is that the requirements in itself have no significant effect on entering the competition. Being a female has a significantly negative effect on entering the competition, this effect is as predicted by the literature. Being a female in the treatment group, makes it significantly more likely that someone enters the competition. Thus having requirements have the effect that is more likely for a female to enter the competition.

This result can be explained that the requirements before the competition give a signal about how well someone would do in the competition. Because the average number of requirements filled in by female participants is 4 on average, which is more than half, and could give women a signal that they are likely to do well in the competition and therefore they become more likely to choose competition. Further research could look at what would happen when the requirements are harder to fulfill and the average of requirements filled in would be lower.

Comparing the result that women in the treatment are more likely to choose competition with the results of relative confidence in table 10, where the women in the treatment are already more confident in round two of the experiment when they are not yet confronted with the requirements, could also indicate that the women in the treatment group are already more confident than the women in the control group. This problem could have been addressed by using a within subject design, where there is the ability to control for within subject differences between the control and the treatment. So further research can look if the effect found in this

Gender differences in competition : are women taking requirements more serious?