Gender, risk preferences and willingness to compete in a random sample of the Swedish population✰

Journal of Behavioral and Experimental Economics 83 (2019) 101467

Contents lists available at ScienceDirect

Journal of Behavioral and Experimental Economics journal homepage: www.elsevier.com/locate/jbee

Gender, risk preferences and willingness to compete in a random sample of the Swedish population✰

T

Anne Boschinia, Anna Dreberb,c, Emma von Essend,e, , Astri Murenf, Eva Ranehillg ⁎

a

SOFI, Stockholm University, Universitetsvägen 10F, SE-106 91 Stockholm, Sweden Department of Economics, Stockholm School of Economics, Sveavägen 65, SE-113 57 Stockholm, Sweden c Department of Economics, University of Innsbruck, Innrain 52, 6020 Innsbruck, Austria d SOFI, Stockholm University, Universitetsvägen 10F, SE-106 91 Stockholm, Sweden e Deaprtment of Economics and Bussiness Economics, Aarhus University, Fuglesangs Allé 4, 8210 Aarhus V, Denmark f Department of Economics, Stockholm University, Universitetsvägen 10A, SE-106 91 Stockholm, Sweden g Department of Economics, Gothenburg University, Vasagatan 1, SE-405 30 Göteborg, Sweden b

ARTICLE INFO

ABSTRACT

Keywords: Gender differences Competitiveness Risk taking Experiment Random representative sample

Experimental results from student and other non-representative convenience samples often suggest that men, on average, are more risk taking and competitive than women. We explore whether these gender preference gaps also exist in incentivized tasks in a simple random sample of the Swedish adult population. Our design comprises four different conditions to systematically explore how the experimental context may impact gender gaps; a baseline condition, a condition where participants are primed with their own gender, and two conditions where the participants know the gender of their counterpart (man or woman). We further look at competitiveness in two domains: a math task and a verbal task. We find no gender gap in risk taking or competitiveness in the verbal task in this random sample. There is some support for men being more competitive than women in the math task in the pooled sample, but the effect size is small. We further find no consistent impact of the respective conditions on (the absence of) the gender gap in preferences.

JEL codes: D91 C83 C91

1. Introduction Gender differences in economic preferences have been proposed as one factor contributing to gender gaps in educational choices and labor market outcomes (e.g. Bertrand, 2011; Croson & Gneezy, 2009). In particular, substantial attention has been given to gender differences in risk preferences and competitiveness, where the experimental literature from both the lab and the field suggests that, if anything, men tend to be more risk taking and competitive than women (e.g. Bertrand, 2011; Croson & Gneezy, 2009; Eckel & Grossman, 2008a, 2008b; Niederle, 2014; Niederle & Vesterlund, 2007). Experimental measures of risk preferences and competitiveness have been shown to relate to important economic choices and outcomes. These preferences seem to play a role in explaining individual outcomes as well as gender differences in outcomes. For example, studies by Bonin, Dohmen, Falk, Huffman and Sunde (2007) and Dohmen et al. (2011) indicate that risk averse individuals are more

likely to work in sectors with little salary variation and less likely to be self-employed. On competitiveness, Zhang (2013) finds that students who are willing to compete in a math task in the lab are more likely to take a competitive high school entrance exam in China than uncompetitive individuals. In a similar vein, Buser, Niederle and Oosterbeek (2014) and Buser, Peter and Wolter (2017) find that competitive individuals choose more math oriented and prestigious high school tracks and that the gender gap in willingness to compete partially explains the gender gap in the choice of educational specialization. Further, Reuben et al. (2017) find that competitive college students have higher expectations for their future salaries. In a large field experiment, Flory, Leibbrandt and List (2015) also find that women are in some, but not most, contexts less likely to apply for jobs with competitive payment schemes than men. The experimental results on gender preference gaps thus largely support the observation that these gaps may have important economic consequences and contribute to gender differences in economic outcomes.

✰ We thank the Austrian Science Fund (FWF, SFB F63), FORTE, the Jan Wallander and Tom Hedelius Foundation (Handelsbankens forskningsstiftelser) and the Knut and Alice Wallenberg Foundation for financial support. ⁎ Corresponding author at: SOFI, Stockholm University, Universitetsvägen 10F, SE-106 91 Stockholm, Sweden. E-mail addresses: [email protected] (A. Boschini), [email protected] (A. Dreber), [email protected] (E. von Essen), [email protected] (A. Muren), [email protected] (E. Ranehill).

https://doi.org/10.1016/j.socec.2019.101467 Received 2 July 2019; Received in revised form 5 September 2019; Accepted 18 September 2019 Available online 20 September 2019 2214-8043/ © 2019 Published by Elsevier Inc.

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

Fig. 1. Number of risky choices by men and women.

In this study, we test whether there are gender differences in risk taking and competitiveness in a random sample of 926 individuals from the Swedish population aged 17–73. While gender preference gaps have been found in many experimental studies in different countries using different types of samples, it is also clear that the existence and strength of gender gaps vary with the context such as the social framing or the gender composition of the reference group, the exact measurement used, and the specific population studied (e.g. Apicella & Dreber, 2015; Booth & Nolen, 2012a,b; Cardénas, Dreber, von Essen & Ranehill, 2012; Croson & Gneezy, 2009; Datta Gupta, Poulsen & Villeval, 2013; Dreber, von Essen & Ranehill, 2011; Filippin & Crosetto, 2016; Gneezy, Niederle & Rustichini, 2003; Gong & Yang, 2012). (Fig. 1). We contribute to this literature by exploring gender gaps in risk taking and the choice to compete in incentivized tasks in a randomly drawn and representative sample. As discussed below, many of the previous larger studies which explore gender gaps in economic preferences in representative samples use hypothetical questions. While these hypothetical questions are generally validated in studies that also use incentivized measures, this does not imply that the hypothetical measures are validated for different dimensions of individual heterogeneity. For example, the gender gaps found in the hypothetical measures are not always replicated in the incentivized tasks (see, e.g., Falk et al., 2018 and Dohmen et al., 2011). We further systematically vary the decision context across four conditions to explore if gender gaps in risk preferences and competitiveness depend on the social context and gender salience. In the first condition, the Baseline condition, participants make decisions anonymously. This condition is the closest to the setting used in most laboratory experiments. In the Priming condition, participants are in a subtle way reminded of their gender before they make any decisions in the experiment. The only other studies exploring the impact of gender priming on risk preferences and competitive choices that we are aware of are Benjamin, Choi and Strickland (2010) and Irriberi and Rey-Biel (2017).

While neither of these studies find that gender priming impacts behavior, our first hypothesis is that, if anything, the prime will make behavior more gender-stereotypical than in the control condition, i.e. that the potential gender gaps in risk taking and competitiveness increase.1 However, our hypothesis is silent on how these increases would occur.2 We further include two conditions where participants are informed about the gender of their counterpart before each decision; these are the Male Counterpart and the Female Counterpart conditions.3 Previous results on whether the gender of the opponent matters for competitiveness, as measured by the change in performance under a tournament scheme versus under a piece-rate scheme, or from the choice whether to get paid according to a tournament scheme or a piece-rate scheme, are mixed.4 For example, Cárdenas et al. (2012), Gneezy et al. (2003), in 1 It is possible that exposure to same- or mixed-sex groups could have priming-related effects and could influence risk preferences and competitiveness. The results from Booth and Nolen (2012a; 2012b) suggest that girls from same-sex schools are more willing to compete, as well as more risk-taking, than girls from mixed-sex schools. Similar results are also reported in a study where first year college students were randomly allocated to all male, all female or coeducational groups (Booth et al. 2014). After eight weeks in a same-sex environment, women are significantly more risk taking than their counterparts in mixed-sex groups. The exact mechanism behind these results remains to be explored, but priming could potentially be involved. 2 Some studies show that both men and women's behavior react to variations of the decision context (e.g. Gneezy et al. 2003; Ellingsen et al. 2012; Ellingsen, 2013; Boschini, Muren & Persson, 2012). However, changes in behavior often occur in the direction predicted by gender stereotypes (Espinosa and Kovářík 2015). 3 It could be argued that these conditions also prime participants with gender, as discussed by Iriberri and Rey-Biel (2017). In the following analysis, however, we divide our sample by condition as was initially intended. 4 There is also a literature on high stakes competitive TV game shows where the gender of the opponent is revealed. This literature suggests that the gender

2

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

Colombia but not in Sweden), de de Sousa and Hollard (2015) find that girls and women compete more against other women, and Gneezy and Rustichini (2004) find that girls in Israel compete less against other girls. Iriberri and Rey-Biel (2017) find that for a mental rotation task men are significantly more competitive than women in terms of improving their performance when informed about the gender of their opponent, but not in a control condition without information. Using randomized assignment of the gender of the competitors, Datta Gupta et al. (2013) find that men compete more against women than against men and Booth and Yamamura (2018) find this increase in competitiveness holds for both genders. Our second hypothesis is, therefore, as in the priming condition, that the gendered behaviors will be more pronounced in these treatment conditions compared to the baseline condition. Similar to the priming condition, our hypothesis is silent on whether men or women or both are impacted by the condition. Finally, we study competitiveness in two different tasks that vary in gender stereotypes: a math task and a verbal task. Some studies show that the competitive task may matter for the gender gap in willingness to compete. While boys or men are often found to be more competitive than girls or women in math-related tasks there is typically no gender gap in verbal tasks (e.g. Dreber, von Essen & Ranehill, 2014; Grosse, Riener & Dertwinkel-Kalt, 2014; Günther, Ekinci, Schwieren & Strobel, 2010; Shurchkov, 2012; though Wozniak, Harbaugh & Mayr, 2014 found that men are more competitive also in a verbal task).5 Our third hypothesis is, therefore, that the typical gender gap, with men being more competitive than women, will be observed in the math task but not in the verbal task in all conditions. The study was conducted using phone interviews. In addition to the economic choice tasks, we also collected basic socio-demographic information, such as age, income, and level of education. To preview our results, we find no support for our first two hypotheses; overall there are no gender differences in risk taking or competitiveness in the verbal task. Pooling the sample, we find some tentative evidence of our third hypothesis. Men are slightly more competitive than women in the math task. However, the effect size is smaller than a Cohen's d of 0.2, which is considered a small effect and considerably lower than what is typically found among students. We further find no support for our first and second hypothesis, finding no evidence of any differences between the baseline and the three other conditions for risk taking or competitiveness in either task. These results hold both in general and for each gender separately. A post hoc power analysis presented in Section 3.4 suggests that the overall null results found are not due to a lack of statistical power. However, to what extent our results for the Swedish population can be generalized to random samples in other countries remains to be explored. Moreover, our risk preference results on this Swedish sample should be interpreted in light of other conflicting results.

In particular, there are other studies exploring gender preference gaps among representative samples of a country population, sometimes with mixed results that could be due to among other things the country, the measure and the sample size. Representativeness in these studies is typically assessed by comparing the general population to the sample at hand, along with a few key variables. The studies differ in sampling methods, using either probabilistic sampling methods, such as simple random samples, where the inclusion probability is known, or nonprobabilistic sampling methods, where the inclusion probability is unknown. For example, Harrison, Lau and Rutström (2007) use the Holt and Laury task to elicit risk preferences in a random sample of the Danish population aged 19 to 75 (253 individuals, 40% response rate). They find no gender difference, and the lack of gender gap is not influenced by the inclusion of some socioeconomic variables. Dohmen, Falk, Huffman and Sunde (2010) and Dohmen et al. (2011) study two different random samples of German adults (1012 participants; risk preferences are elicited among 452 participants) and 22,019 individuals respectively, overall response rates are not reported). Using an incentivized risk measure, Dohmen et al. (2010) find no gender gaps in risk-taking. Dohmen et al. (2011) find a similar result. Women in their sample self-report to be less willing to take risks on average, but this result is not replicated in the incentivized task in the smaller laboratory study. von Gaudecker, van Soest and Wengstrom (2011), using both hypothetical and incentivized measures on a Dutch sample (using the CentER internet panel), find that women on average are less risk-taking than men (1422 participants).6 Andersen, Harrison, Lau and Rutström (2014), using the Holt and Laury task with the procedures of Hey and Orme (2004), study a random sample of the Danish population (413 participants, 21% response rate) and find that women are less risk taking than men.7 Harrison, Lau and Yoo (2019), using the Andersen et al. (2014) plus a different sample, on the other hand, find no statistically significant gender differences in risk preferences. Falk et al. (2018) use a linear combination of a non-incentivized version of the risk task and the self-reported question used on Dohmen et al. (2011) studying random samples of households (with a total of 80,000 people) in 76 countries and find women to be significantly more risk-averse than men at least at the 10% significance level in 82% of the countries, including in Sweden. Sephavand and Shahbazian (2017) study the same self-reported risk question in a stratified (by region) random sample from Burkina Faso, also finding women to be less risk-taking compared to men. Focusing on studies on Sweden, Almenberg and Dreber (2015) use the self-reported question from Dohmen et al. (2011) and find that men are, on average, more risk-taking than women in a random sample of Swedish adults (1300 individuals, 45% response rate). Beauchamp, Cesarini and Johannesson (2017) use a random sample (approximately 11,000 individuals) of the Swedish twin population (the sample of twins is similar to the general population on some selected characteristics). Also using the non-incentivized risk measure mentioned above, they find male twins to be on average are more risktaking than female twins (only looking at same-sex twins). While we know of no previous study using an incentivized task in a representative Swedish sample, these previous positive results on gender differences in risk attitudes indicate that our results should be interpreted with caution. To our knowledge, there are no other studies on competitiveness in a representative sample. Instead, the most related studies are Almås, Cappelen, Salvanes, Sørensen and Tungodden (2015) and

(footnote continued) of the opponent can impact performance or behavior (Antonovics, Arcidiacono, and Walsh 2009; Hogarth, Karelaia, and Trujillo 2012; Säve‐Söderbergh and Lindquist 2017; Jetter and Walker 2018). In particular, Hogarth et al. (2012) find that on a Colombian TV show, women are more likely to exit the game prematurely compared to men, in particular when groups are male-dominated, whereas men's exit decisions are not affected by the gender composition of the group. 5 For example, Günther et al. (2010) and Grosse and Reiner (2010) study gender differences in competitiveness in terms of improved performance comparing a competitive payment scheme and a piece-rate payment scheme and find that the task matters: men are more competitive for a math task but not for a word task. Shurchkov (2012) studies two different tasks and vary whether the tournament is high- or low-pressure. She finds that women perform worse than men in a high-pressure math-based tournament but that the gender gap reverses with women also being more willing to compete in a low-pressure verbal tournament. Other studies find no impact of the task on the gender gap in competitiveness (e.g. Dreber et al. 2011, Wozniak et al. 2014).

6

The recruitment for the CentERpanel is conducted by TNS-NIPO. Households complete an internet based survey every week. When a household leaves the panel it is replaced with another household with similar characteristics. 7 This result is reported in their Appendix. 3

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

Buser et al. (2017), which both elicit willingness to compete in a math task. Almås et al. (2015) study competitiveness in a sample of 523 14–15 year-olds in Bergen, a city which is roughly comparable to the rest of the Norwegian population. They find that family background matters in explaining gender differences in willingness to compete, where the gender gap is higher among individuals with a high socioeconomic background as compared to low. Buser et al. (2017) use a sample of 249 students from a region in Switzerland. Their sample is similar to the regional population concerning the share of women. They find that women are less competitive and that competitiveness can partly explain the gender gap in study choice. The remainder of the paper is organized as follows. In Section 2, we present our experimental design and data. In Section 3, we present the results from each condition separately as well as potential treatment effects. Finally, in Section 4, we discuss our results in comparison with previous literature and then conclude.

polling company. For all decisions, an interviewer read the instructions to the participant.12 Before each decision, participants answered some control questions allowing us to measure participants’ understanding of each decision (no control questions were used for the part measuring risk preferences). Participants received no feedback on outcomes during the experiment. Participants in the phone interview were randomly assigned to one of four conditions; Baseline, priming, female counterpart or male counterpart. In the baseline condition, the interaction was fully anonymous for the participants vis-à-vis each other, and no reference was made to gender.13 In the priming condition, participants were asked to state their gender at the beginning of the interview. Finally, in the female and male counterpart conditions, the gender of the counterpart was revealed before participants made decisions by using gendered pronouns, and this information was repeated and kept constant for each decision involving a counterpart. All decision situations were presented to the participants in standard language. Participants were informed that one of the eight decisions would be randomly chosen for payment by the decision(s) made by the participants involved. If the risk decision was selected for payment, one out of the seven risk decisions was randomly chosen for payment. For a constant budget, we chose to only pay for one decision and have relatively higher stakes for all decisions rather than pay for all decisions and have lower stakes.

2. Experimental design and data We conduct an artefactual field experiment (following the definition of Harrison & List, 2004) on a simple random sample of the Swedish population aged 18–73. The sampling was performed in close collaboration with a professional polling company based in Stockholm, Sweden, with the main sampling and data collection performed in September through November 2011 and additional follow-up collection of income and education data in October 2012. The polling company received a random sample of the Swedish population from Statistics Sweden and collected the data through telephone interviews.8 The polling company then provided us with anonymized data.9

2.2. Experimental measures and demographic questions We elicit risk preferences by using a multiple price list where participants make seven choices between a risky option and a safe option. The risky option was the same across the different decisions, giving SEK 200 or 0 with equal probability, while the safe options varied from SEK 40 to SEK 160 in increments of 20 SEK. We measure risk preferences from the number of times the participant chose the risky option. We measure competitiveness as the binary decision to compete or not in two different tasks: a verbal and a math task. Participants first decided whether to compete in a verbal task and then in a math task. The choices in these two tasks allow us to compare the gender gap in competitiveness in a verbal task with a neutral or potentially female stereotype and a math task with an implicit male stereotype.14 In the verbal task, participants were asked to form as many words as possible of at least three letters from eight given letters for two minutes. In the math task, participants were asked to find as many number combinations as possible that added up to 25 from nine given numbers, also during two minutes. After the task was described to them, but before performing the task, participants chose their preferred payment form – an individual piece-rate payment or a competitive tournament payment. In our individual payment scheme, participants were paid SEK 10 per correct word or number sequence. The tournament payment scheme involved comparison with a randomly selected counterpart (who also chose to compete). The best performer was paid SEK 20 for each correctly solved exercise, and otherwise, they were paid SEK 0. Further, the average interview lasted about 25 min, and we wanted to keep it as short as possible. For these reasons, we chose to not include a stage with individual performance and also not to elicit performance beliefs in the competitive tasks.15 Thus, we cannot perform the analysis

2.1. Setup and conditions Sampled individuals received a letter a few days ahead of the first phone call inviting them to take part in a phone interview study on economic decision-making conducted by researchers at Stockholm University. The letter provided information on the length of the study (approximately 25 min) and earnings (A SEK 100 participation fee plus potentially more depending on the participant's choices).10 In the interview, each participant made decisions in eight independent situations and answered demographic questions. The eight decisions included the following measures and games: the dictator game (in the role of the dictator), the ultimatum game (in the role of the proposer), the trust game (in the role of the trustor), the prisoner's dilemma, the battle of the sexes, risk preferences, and willingness to compete in a math task and willingness to compete in a verbal task.11 In this paper, we focus on risk preferences and willingness to compete (results for the dictator game are reported in Boschini, Dreber, von Essen, Muren, & Ranehill, 2018, and the other results will be reported elsewhere). The order of the decisions was kept constant for logistical reasons for the 8 The polling company (MIND Research) conducted the inverviews according to the standards of Statistics Sweden. Up to 14 attempts to reach each individual in the sample were made, and all interviews and attempts to contact participants were conducted in the afternoon and evening during normal working days. 9 An application to the Stockholm Ethical Review Board (Etikprövningsnämnden i Stockholm: EPN) for the present project was submitted in June 2011. EPN stated that our project did not need to undergo full ethical review since we only handle anonymized information. The data is available for researcher at https://snd.gu.se/en/catalogue/study/snd1021. 10 At the time of the study, SEK 100 was approximately USD 14. University researchers tend to be trusted in Sweden, and no participant voiced any concern. However, although we believe that the payment information was considered credible, we did not ask about this explicitly. 11 We used a postal survey for participants who were recipients in the Dictator Game, proposers in the Ultimatum Game and second players in the Trust Game.

12 In order to minimize the individual differences between the interviews we conducted a pilot were we listened in on a few interviews. 13 Unlike in a standard lab experiment, participants are however not anonymous vis-à-vis the interviewer. 14 See, for example, Nosek et al. (2002) and Steffens et al. (2010) who investigate tasks and implicit gender stereotypes. 15 It is also possible that the length of the study affected decisions although the lenght of our study is considerable shorter than a typical session in the lab.

4

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

done by Niederle and Vesterlund (2007) who study willingness to compete adjusting for individual performance, beliefs and risk preferences. Moreover, in the two conditions where the gender of counterpart was revealed, this information was shared before performing the task. This could affect the choice of whether to compete or not due to many reasons, including performance expectations or taste for specific competitor characteristics. It may also impact performance. Finally, we asked the participants a set of socio-demographic questions. In particular, we asked for age, legal gender, income, and education. For gender, participants were given the binary choice of identifying as a man or woman. Concerning age, we asked for participants’ birth year. Thus, we define this variable as the year the study was conducted deducted by the birth year.16 Income, as well as education, was measured as one of three self-reported categories (Low income: 0–250,000, Middle income: 250,001–750,000, High income: >750,001 and Low education: 0–9 years, Middle education: 10–12 years, High education: >12 years).17 These variables were elicited since some previous work has indicated that they may correlate with gender differences in preferences.18 In research on survey methods characteristics of the interviewer are sometimes found to affect the answers. We, therefore, collected information on the gender of the interviewer.19 Women constituted 40% of the interviewers.

Table 1 Descriptive statistics.

Outcome variables Number of risky choices Share competing in word task Share competing in math task Control variables Female (1=female, 0 otherwise) Age (years at time of interview)* Income <= 250.000 250 001–375 000 >375 000 Education 0–9 years 10–12 years >12 years Gender of the interviewer (1=female, 0 otherwise)

N

Mean/Share

Sd

Min

Max

926 926 926

3.505 0.383 0.288

2.231 0.486 0.453

0 0 0

7 1 1

926 926

0.478 45.66739

0.500 15.6395

0 18

1 74

926 926 926

0.408 0.362 0.230

0.492 0.481 0.421

0 0 0

1 1 1

926 926 926 926

0.148 0.401 0.451 0.397

0.355 0.490 0.498 0.490

0 0 0 0

1 1 1 1

⁎ Since we only have information on birth year, we have defined age as the year the study was conducted deducted by the birth year. We thus assume that all individuals are born on the 1st of January, and the sample will, therefore, include some individuals that are 74 years old.

2.3. Data

observables among the participants in different conditions, p = 0.2265 using a joint chi2 test (see Table A5 in the Appendix). The instructions that were read to the participants can be found in the Supporting information online.

Our data set comprises the 997 individuals that completed the phone interview. The response rate was 52.9%, and there is no evidence of systematic non-response based on gender or age (see Tables A2 and A3 in the Appendix).20 269 individuals were in the baseline condition, 256 in the priming condition, 218 in the male counterpart condition and 254 in the female counterpart condition. About 49% of the participants were female. Table 1 shows an overview of descriptive statistics. Of the 997 participants, 50 did not answer the question regarding income or education, and for an additional 21 observations we did not receive a recorded value for the gender of the interviewer. In the main analyses we do not include the missing observations, and our preferred specification thus comprises 926 participants. Descriptive statistics for the full sample are presented in Table A1 in the Appendix. We have access to population statistics of gender, income, education and age from the same period. Our sample compares well to the Swedish population concerning gender, income and education, but consists, on average, of somewhat older participants than the population. We, therefore, consider our sample representative of the population we wish to investigate (see Table A4 in the Appendix).21 A comparison across conditions also reveals no statistical differences in

3. Results We first explore gender gaps in risk-taking and competitiveness within conditions and the pooled sample. After that, we turn to treatment effects before we present a robustness analysis. Throughout the analysis, we use parametric tests: t-tests and OLS regression analyses for the cardinal risk variable, and test of proportions and Logit regression for the dichotomous competition variable. We employ a significance level of 0.05. To simplify future meta-analyses, we also report effect sizes (Cohen's d). In the regression analyses, we control for the following socio-demographic variables: income, age, age squared, highest obtained educational level, income level and gender of the interviewer. To be able to compare across regressions, we keep the sample constant, restricting the analysis to include only individuals that provided an answer to all the socio-demographic variables. The restricted sample includes 71 individuals less compared to the sample on which the Cohen's d calculations are based. 3.1. Risk preferences

16

We thus assume that all individuals are born on the 1st of January and the sample will, therefore, include some individuals that are 74 years old. 17 In line with regulations from SCB we asked the participants to choose from seven categories of income and four categories of education. Appendix Table A4 shows that the proportions for each of these categories of income and education match the population. However, in our sample some categories contain too few observations. We therefore created three income categories, matching SCBs categorization of low, middle and high income. For education we collapsed the two categories with the least amount of years of schooling. 18 Other socio-demographic measures we elicited, but do not use in the analysis of this paper, were: civil status, number of children below age 18, household income, occupation, occupational sector, and the position within the workplace. Including these variables in the current analysis does not change our result in a qualitative way. 19 The Appendix contains more information about the distributions of the age, income and education variables. 20 These response rates are comparable to standard surveys conducted by Statistics Sweden (www.scb.se). 21 Since our random sample seems to be fairly representative of the population we do not consider population weights necessary.

Despite using a lottery choice task with a safe option, which is one of the measures that most often, but not always, generate a gender gap in risk preferences (Filippin & Crosetto, 2016; Nelson, 2015; Niederle, 2014) we do not find evidence of a gender gap in risk preferences in our sample. In the baseline condition, both men and women choose, on average, 3.6 risky choices out of seven possible (Cohen's d = 0.026, p = 0.834). In the priming condition, men choose on average, 3.2 risky choices and women 3.3 risky choices (Cohen's d=−0.05, p = 0.710). The equivalent numbers for the condition with a female counterpart are 3.8 and 3.6 (Cohen's d = 0.08, p = 0.567) and for the condition with a male counterpart, the numbers are 3.5 and 3.7 (Cohen's d=−0.06, p = 0.649). Pooling across all conditions, we find that both men and women choose an average of 3.5 risky decisions (Cohen's d=−0.005 and p = 0.933). We explore any potential gender gap in risk preferences further in Table 2 using OLS regressions. For each of the four different conditions, and the pooled sample, we run one regression controlling only for 5

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

Table 2 Risk Preferences OLS: Gender differences in the number or risky choices within treatments. Baseline Female

0.026 (0.271)

Age Income: 250 001–375 000 Income: >375 000 Education: 10–12 yrs Education: >12 yrs Age squared Gender of the interviewer Constant 2

Adjusted R Observations

3.581*** (0.177) −0.004 248

Priming 0.075 (0.278) −0.005 (0.061) 0.368 (0.336) 0.837 (0.427) 0.100 (0.418) 0.093 (0.440) −0.000 (0.001) −0.290 (0.279) 3.652** (1.143) 0.001 248

0.200 (0.300)

3.090*** (0.203) −0.002 239

Male count.p 0.544 (0.317) 0.110 (0.064) 0.698 (0.356) 1.336** (0.464) 0.268 (0.545) −0.378 (0.530) −0.001 (0.001) −0.251 (0.293) 0.624 (1.295) 0.072 239

−0.168 (0.318)

3.770*** (0.211) −0.004 203

Female count.p −0.286 (0.321) 0.161* (0.074) 0.168 (0.465) −0.167 (0.468) 0.180 (0.572) −0.190 (0.557) −0.002* (0.001) −0.659* (0.331) 0.847 (1.383) 0.015 203

0.129 (0.290)

3.520*** (0.201) −0.003 236

Pooled 0.303 (0.297) −0.036 (0.059) 1.625*** (0.350) 0.534 (0.441) −0.128 (0.443) −0.798 (0.512) 0.000 (0.001) −0.278 (0.280) 3.810*** (1.132) 0.075 236

0.052 (0.147)

3.480*** (0.099) −0.001 926

0.150 (0.152) 0.045 (0.032) 0.707*** (0.186) 0.661** (0.225) 0.095 (0.248) −0.311 (0.257) −0.001 (0.000) −0.380** (0.146) 2.525*** (0.615) 0.033 926

Robust standard errors. Income: <250 000 and Education 0–9 yrs are used as reference categories.-. ⁎ p < 0.05,. ⁎⁎ p < 0.01,. ⁎⁎⁎ p < 0.001.

condition with a male counterpart becomes non-significant when adding control variables. The gender gap found when we pool the conditions also becomes smaller with larger standard errors when adding socio-demographic controls. We thus find only very limited support for our third hypothesis about a larger gender gap in willingness to compete in the math task compared to the verbal task. The results thus suggest that the small gender gap in competitiveness in our sample in the math task is related to socio-demographic characteristics and risk taking. However, to what extent the results would be different if we could control for individual performance or relative performance beliefs is not clear. It is not the purpose of this study, but little evidence exists on the extent to which behavioral gender gaps are influenced by sociodemographic characteristics, with some exceptions including, for example, Flory et al. (2018) who focus on age. Figure OA1 in the Supporting information online shows that the age distribution is similar for men and women in our sample. Also, in our paper we see few differences in the gender coefficient between the model without control variables compared to the model including control variables.

whether the participant is female or not, and one also controlling for socio-demographic variables. The regression results are similar to the results from the t-tests presented above – we find no gender gaps in risk-taking in any of the conditions, or the pooled sample. Further, including control variables does not change these results. 3.2. Competitiveness Fig. 2 presents the raw shares of men and women choosing to compete in the two tasks. In the verbal task, exploring each condition separately, we find no statistically significant gender differences in the choice of whether to compete or not. In the baseline group, 39% of men and 38% of women choose to compete (Cohen's d = 0.03, p = 0.828). Equivalent numbers for the other conditions are 36% vs 43% (Cohen's d =−0.13, p = 0.297) in the priming condition, 34% vs 43% (Cohen's d=−0.19, p = 0.158) in the male counterpart condition and 35% vs 37% (Cohen's d=−0.05, p = 0.680) in the female counterpart condition. Pooling all four conditions, we find that 36% of men vs 40% of women choose to compete; again this difference is not significant (Cohen's d=−0.08, p = 0.198). The point estimates of the gender gap in the math task are reversed in comparison to the verbal task, and the magnitudes are slightly larger. However, also, this gap is not consistently significant. In the baseline condition, 29% of men and 23% of women choose to compete (Cohen's d = 0.14, p = 0.268). The equivalent numbers in the priming condition and the male counterpart condition are 33% vs 25% (Cohen's d = 0.17, p = 0.182), and 30% vs 23% (Cohen's d = 0.17, p = 0.180) respectively. When the counterpart is a woman, we find that 38% of men vs 25% of women choose to compete. This difference is statistically significant (Cohen's d = 0.28, p = 0.040). Pooling all four conditions also yields a statistically significant gender gap – among all participants 32% of men vs 24% of women choose to compete in the math task (Cohen's d = 0.18, p = 0.004). Tables 3 and 4 display marginal effects from logit regressions for both measures of competitiveness. Results are first presented for each condition separately, and then for the pooled dataset, without and with control variables. The regression analyses confirm previous results with two exceptions. The significant gender difference in the math task in the

3.3. Treatment effects In this section, we further address the first and second hypotheses by testing whether systematically modifying the decision context across the four conditions impact gender gaps. In Table 5, we present the results from nine regressions testing whether the gender gap in risk-taking and competitiveness in the two tasks differ between the baseline condition and the three treatment conditions. All regressions include a variable for condition (reducing the number of conditions to the two in the pairwise comparison), whether the participant is female or not, and the interaction between the two variables. To simplify the interpretation of the interaction coefficient, we use OLS for all regressions.22 22 Interpreting the first derivative (marginal effect) of the multiplicative term between two explanatory variables as the interaction effect is a common mistake (Buis, 2010). For simplicity and to enable comparisons with the results from risk taking we chose to use a LPM. Some researchers argue that for marginal effects it might not matter that much whether a linear or non-linear model is used (Angrist and Pischke, 2008).

6

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

Fig. 2. Competitiveness in the verbal and math task by men and women.

As indicated in the table, we find no impact of the different treatment conditions on the gender gap. Further, we find no impact on the behavior of men and women in general.

3.4. Robustness Before running the study, we did not calculate the optimal sample size or do any type of power calculation. To evaluate whether our

Table 3 Competitiveness in the Verbal Task Logit (marginal effects): Gender differences in the proportion of competitive choices in the verbal task. Baseline Competition Female Number of risky choices

−0.162 (0.263)

−0.387* (0.175)

0.005 (0.284) 0.112 (0.065) −0.192** (0.064) 0.683 (0.356) 1.529*** (0.443) −0.023 (0.449) −0.074 (0.478) 0.002** (0.001) 0.058 (0.285) 2.878* (1.256)

248

248

Age Income: 250 001–375 000 Income: >375 000 Education: 10–12 yrs Education: >12 yrs Age squared Gender of the interviewer Constant 2

Adjusted R Observations

Priming 0.174 (0.265)

Male count.p

−0.502** (0.187)

0.027 (0.300) 0.195** (0.067) −0.030 (0.065) −0.250 (0.364) −0.500 (0.468) 0.141 (0.434) 0.703 (0.432) 0.000 (0.001) −0.109 (0.286) −0.755 (1.364)

239

239

0.370 (0.291)

Female count.p

−0.663** (0.212)

0.522 (0.319) 0.124 (0.071) −0.083 (0.073) 0.626 (0.395) 0.550 (0.455) 0.548 (0.584) 0.854 (0.590) 0.001 (0.001) 0.216 (0.313) −0.561 (1.319)

203

203

Robust standard errors. Income: <250 000 and Education 0–9 yrs are used as reference categories.-. ⁎ p < 0.05,. ⁎⁎ p < 0.01,. ⁎⁎⁎ p < 0.001. 7

0.114 (0.271)

Pooled 0.113 (0.135)

−0.610** (0.188)

0.065 (0.299) 0.224** (0.070) 0.050 (0.063) 0.254 (0.378) 0.265 (0.421) 0.036 (0.454) 0.543 (0.477) −0.000 (0.001) 0.448 (0.294) −3.199* (1.369)

−0.530*** (0.094)

0.148 (0.144) 0.152*** (0.032) −0.056 (0.032) 0.268 (0.179) 0.392 (0.214) 0.134 (0.228) 0.447 (0.233) 0.001 (0.000) 0.157 (0.142) −0.419 (0.633)

236

236

926

926

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

Table 4 Competitiveness in the Math Task Logit (marginal effects): Gender differences in the proportion of competitive choices in the math task. Baseline Competition Female

Priming

−0.356 (0.292)

Number of risky choices

−0.840*** (0.187)

−0.218 (0.336) 0.194** (0.075) −0.129 (0.073) 0.889* (0.414) 2.117*** (0.505) 0.209 (0.546) 0.339 (0.554) 0.001 (0.001) 0.487 (0.311) −0.102 (1.380)

248

248

Age Income: 250 001–375 000 Income: >375 000 Education: 10–12 yrs Education: >12 yrs Age squared Gender of the interviewer Constant Adjusted R2 Observations

−0.392 (0.289)

Male count.p

−0.718*** (0.193)

−0.509 (0.327) 0.208** (0.073) 0.003 (0.071) −0.023 (0.404) −0.238 (0.466) −0.429 (0.482) 0.238 (0.466) 0.000 (0.001) −0.024 (0.306) −1.385 (1.488)

239

239

−0.629* (0.304)

Female count.p

−0.405* (0.205)

−0.539 (0.333) 0.212** (0.077) 0.000 (0.078) 0.108 (0.422) 0.648 (0.499) 1.185 (0.696) 0.945 (0.693) 0.000 (0.001) 0.318 (0.331) −2.859* (1.453)

203

203

−0.356 (0.297)

Pooled −0.420** (0.147)

−0.828*** (0.195)

−0.263 (0.331) 0.096 (0.072) −0.021 (0.066) 0.204 (0.451) 0.847 (0.502) −0.221 (0.478) −0.271 (0.537) 0.000 (0.001) 0.216 (0.302) −0.636 (1.363)

−0.712*** (0.097)

−0.357* (0.160) 0.172*** (0.035) −0.035 (0.036) 0.277 (0.202) 0.778*** (0.234) 0.169 (0.256) 0.350 (0.263) 0.000 (0.000) 0.285 (0.153) −1.264 (0.708)

236

236

926

926

Robust standard errors. Income: <250 000 and Education 0–9 yrs are used as reference categories. ⁎ p < 0.05,. ⁎⁎ p < 0.01,. ⁎⁎⁎ p < 0.001.

the results. While the risk task did not have any control questions, we can identify the share of participants that has multiple switching points between the risky and the safe option. In our sample, a larger share of women (19%) than men (12%) are inconsistent (p = 0.002). Excluding these observations, we still find no gender differences within the baseline or in any of the treatments or pooled samples (see the Supporting information online Table OA5). About 75% and 79% of all participants answered the control question about the competitive part in the verbal task and the math task correctly, and there are no systematic gender differences in the number of correct answers (p = 0.900 for the verbal task and p = 0.594 for the math task, prtest). Excluding the participants who did not answer the control question correctly for each measurement does not impact our results in important ways (see the Supporting information online Tables OA6 and OA7). Multiple testing increases the probability of Type I errors. When designing the study, we did not take Bonferroni corrections into account and do not include the corrections in the text to avoid an increase of Type II errors. In this study, we conduct 24 tests; within each condition and between conditions. A simple Bonferroni correction of the size would imply approximately a p-value threshold of 0.002 (0.05/24) for statistical significance. The only test that survives such a correction is the gender gap in competitiveness in the math task in the pooled sample when no control variables are included. Since we found few gender differences in our study, this correction makes little difference to the overall inference. Revisiting the optimal sample calculations presented above and assuming an alpha of 0.002, we nevertheless have a large enough sample to detect the effect sizes found in previous literature.

Table 5 Comparison of gender differences between baseline and the treatments.

Number of risky choices Competition in the verbal task Competition in the math task Observations

Baseline vs Priming

Baseline vs Male counterpart

Baseline vs Female counterpart

0.174 (0.404) 0.080 (0.089) −0.011 (0.081) 498

−0.194 (0.417) 0.126 (0.092) −0.069 (0.087) 458

0.102 (0.397) 0.065 (0.089) −0.000 (0.081) 492

Robust standard errors *p<0.05,

⁎⁎

p<0.01,

p<0.001.

⁎⁎⁎

sample size would have been large enough to detect an effect as large as those found in previous studies, we have calculated optimal sample sizes after running the study. Regarding risk-taking we assume that the true effect size is the result from Crosetto and Filippin (2016)), who across a large number of studies document an average gender gap corresponding to a Cohen´s d of 0.55 in different versions of the Eckel and Grossman task which is similar to the one we use in that it has a safe option. Assuming a power of 80%, an alpha of 0.05 and using the standardized standard deviation from the pooled regression in column 9 in Table 2, we would have needed a sample size of at least 53 participants per condition to detect an effect size of 0.55. When it comes to willingness to compete in the math task, Niederle and Vesterlund (2007) is the standard reference. Using their results, we find that they have an effect size of 0.80 in terms of Cohen's d (which can be considered a large effect). Assuming a power of 80%, an alpha of 0.05 and using the standardized standard deviation from the pooled regression in column 9 in Table 4, we would have needed a sample size of at least 26 participants per condition to detect an effect size of 0.80. Our choice of sample size, thus, seems to have been large enough to detect the effect sizes known from previous studies at the time of conducting the study. As an additional robustness check, we excluded participants who did not answer the control questions correctly to see if this influenced

4. Discussion To achieve greater gender equality, or increase our knowledge of gender differences in economic preferences, it is important to understand where gender differences come from and in what type of populations they occur. In this experiment on a random and representative 8

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

sample of the Swedish adult population, we find little robust evidence of gender differences in risk preferences or willingness to compete in two different tasks. Our risk result stands in contrast to some previous studies, including others in Sweden with even larger representative samples (Almenberg & Dreber, 2015; Beauchamp et al., 2017; Falk et al., 2018), and might be because we use an incentivized measure of risk preferences. Previous representative samples studies using Swedish samples that we are aware of use the hypothetical question from Dohmen et al. (2010). The risk question or other differences between our study and previous studies, such as the telephone interview component, may contribute to the null result. To our knowledge, there are no previous studies on competitiveness using a random and representative sample. Most studies on students find that women are less competitive than men in math tasks but not necessarily in verbal tasks. While we find an indication of this in our pooled sample, the effect size is small. However, it should be noted that we measure competitiveness differently from most other studies since participants choose payment scheme before performing in our experiment for logistical reasons. One potential explanation of these largely null results could be the specific country studied here: Sweden is one of the most gender-equal countries in the world. The high level of gender equality in Sweden could potentially influence the results; for example, Gneezy and Rustichini (2004) found that boys but not girls reacted to competition by running faster when competing in Israel, whereas Dreber et al. (2011) found no such gender gap in Sweden. However, there are also examples of gender gaps in preferences not always being smaller in more gender-equal countries or cultures (e.g., Cárdenas et al., 2012; Khachatryan, Dreber, von Essen & Ranehill, 2015; Zhang, 2013; and Falk et al., 2018 but see also, e.g., Andersen, Ertac, Gneezy, List & Maximiano, 2013; Gneezy, Leonard & List, 2009). One concern comparing across countries might be the type of sample studied – it is not clear that the selection of who participates in the experiment is similar. More work is thus needed, and the focus on comparable groups across

societies (e.g. random samples of the adult population) may turn out fruitful for understanding why gender gaps in preferences are relatively smaller or larger. Our study differs from the majority of previous studies in the sense that data collection was done through phone interviews. Phone interviews may impact the research context in several ways, such as where the participant takes part in the study, or access to written documentation, and has both advantages and disadvantages. One aspect of employing phone interviews is that it may influence anonymity. Although anonymity is arguably greater in phone interviews than in face to face interviews, the phone interview setting does not allow for the same degree of anonymity as the laboratory. While our intuition is that less anonymity would rather increase gender preference gaps – for example, through a concern to adhere to gender norms prescribing different behavior for men and women – it is, of course, possible that this setting reduces the size of gender preference gaps compared to laboratory experiments. However, while we know of no study that uses phone interviews with incentivized measures of economic preferences, Falk et al. (2018) find gender gaps in risk behavior in many countries in their hypothetical measure using phone interviews. The lack of a priming effect may be due to the fact that the priming occurred at the beginning of our study. In Boschini et al. (2018), where we explore the outcome in the Dictator Game of the same study, we find suggestive evidence of women in the DG giving more than men in the priming condition. It is possible that any effect of the condition had disappeared by the time the participants make decisions in the risk domain and whether to compete. It should also be noted that we did not have any manipulation check and that the priming literature has come under heavy critique during the last few years (see, e.g., Yong, 2012). Further, while it is possible that we are underpowered to detect true, but small-sized, gender differences, our robustness analysis indicates that we have enough observations to detect gender gaps of the size that have previously been found.

Supplementary materials Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.socec.2019.101467. Appendix A: Tables and Figures Table A1–A5

Table A1 Descriptive statistics, full sample.

Outcome variables Number of risky choices Share competing in word task Share competing in math task Control variables Female (1=female, 0 otherwise) Age (years at time of interview)* Income <= 250.000 250 001–375 000 >375 000 Education 0–9 years 10–12 years <12 years Gender of the interviewer (1=female, 0 otherwise)

N

Mean/ Share

Sd

Min

Max

997 997 997

3.520 0.380 0.282

2.240 0.486 0.450

0 0 0

7 1 1

997 994

0.488 45.516

0.500 15.758

0 18

1 74

953 953 953

0.392 0.344 0.220

0.488 0.475 0.414

0 0 0

1 1 1

989 989 989 975

0.150 0.405 0.436 0.401

0.358 0.491 0 0.496 0.490

0 0 0 0

1 1 1 1

⁎ Since we only have information on birth year, we have defined age as the year the study was conducted deducted by the birth year. We thus assume that all individuals are born on the 1st of January, and the sample will, therefore, include some individuals that are 74 years old.

9

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

Table A2 Sample sizes, non-response and under-coverage.

Sample size Response Non-response Under-coverage

Telephone

Postal

2349 996

800 374

Declined No contact

610 271

8 370

Not part of the population No active phone number/address Wrong sampling Late postal responses Older than the population

89 320 56

37 11

6

Table A3 Phone interview: non-response analysis.

Gender Age category*

Non-responders Mean

SD

Responders Mean

SD

p-value ttest

p-value KS**

0.4915 6.5142

0.50 3.0298

0.4885 6.5166

0.50 3.1839

0.896 0.987

1 0.901

⁎ Age categories are; 1 = 18–20, 2 = 21–25, 3 = 26–30, 4 = 31–35, 5 = 36–40, 6 = 41–45, 7 = 46–50, 8 = 51–55, 9 = 56–60, 10=61–65, 11=66–70, and 12=71–73. Since the age for responders is collected as a continuous variable when comparing the figures we forced the responders actual age into the same categories as for non-responders. ⁎⁎ We also tested the equality of distribution between the non-responders and the responders using a Kolmogorov-Smirnov test with the combined p-value for large samples.

Table A4 Comparing population and sample. Population Mean

Sd

Our sample Mean

Sd

Gender (0=man, 1=woman) Age Income

0.502

0.50

0.488

0.50

40.616 Proportion

45.516 Proportion

<= 100.000 100 001–250 000 250 001–375 000 375 001–500 000 500 001–750 000 750 001–1 000 000 >1 000 000 Education < 7 years 8–9 years 10–12 years > 12 years

0.196 0.315 0.299 0.114 0.056 0.012 0.008

23.678 Share of women 0.470 0.392 0.478 0.326 0.261 0.226 0.165

0.133 0.277 0.360 0.113 0.084 0.020 0.013

15.758 Share of women 0.520 0.576 0.493 0.306 0.325 0.263 0.333

0.064 0.119 0.468 0.350

0.462 0.433 0.475 0.550

0.042 0.110 0.409 0.440

0.390 0.422 0.493 0.508

Statistics Sweden provided aggregate estimates based on individual data, from 2011 of the full Swedish population aged 18–73. Table A5 Confirmed random assignment.

Gender Age Income <250 000 251 000–375 000 >375 000 Education 0–9 years 10–12 years >12 years Interaction: gender and age

Control Mean

SD

Priming Mean

SD

Female Counterpart Mean SD

Male counterpart Mean SD

0.4647 45.647

0.500 15.663

0.484 46.439

0.501 15.539

0.523 44.326

0.501 15.843

0.488 45.472

0.501 16.030

0.393 0.429 0.178

0.489 0.496 0.384

0.407 0.327 0.266

0.492 0.470 0.443

0.416 0.349 0.234

0.494 0.477 0.425

0.426 0.331 0.242

0.496 0.471 0.429

0.183 0.425 0.392 21.699

0.387 0.495 0.489 25.587

0.141 0.386 0.472 22.839

0. 348 0.488 0.500 25.920

0.120 0.380 0.50 22.477

0.326 0.486 0.501 24.181

0.155 0.438 0. 406 21.456

0.363 0. 497 0.492 24.547

To test the differences between the means of the conditions we use a multivariate test (mvtest in stata). The chi-square approximation p-value of 0.2265 does not indicate a statistically significant likelihood that the three treatment conditions and the control condition have equal means. 10

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

References

Eckel, C.C., Grossman, P.J., 2008b. Forecasting risk attitudes: An experimental study using actual and forecast gamble choices. Journal of Economic Behavior & Organization 68 (1), 1–17. https://doi.org/10.1016/j.jebo.2008.04.006. Ellingsen, T., Johannesson, M., Mollerstrom, J., Sara, M., 2012. Social framing effects: Preferences or beliefs? Games and Economic Behavior 76 (1), 117–130. https://doi. org/10.1016/j.geb.2012.05.007. Ellingsen, T., 2013. Gender differences in social framing effects. Economics Letters 118 (3), 470–472. https://doi.org/10.1016/j.econlet.2012.12.010. Espinosa, M.P., Kovářík, J., 2015. Prosocial behavior and gender. Frontiers in Behavioral Neuroscience 9 April. https://doi.org/10.3389/fnbeh.2015.00088. Falk, A., Becker, A., Dohmen, T., Enke, B., Huffman, D., Sunde, U., 2018. Global evidence on economic preferences. The Quarterly Journal of Economics 133 (4), 1645–1692. https://doi.org/10.1093/qje/qjy013. Filippin, A., Crosetto, P., 2016. A reconsideration of gender differences in risk attitudes. Management Science February. https://doi.org/10.1287/mnsc.2015.2294. Flory, J.A., Leibbrandt, A., List, J.A., 2015. Do competitive workplaces deter female workers? A large-scale natural field experiment on job entry decisions. The Review of Economic Studies 82 (1), 122–155. https://doi.org/10.1093/restud/rdu030. von Gaudecker, H.-M., van Soes, A., Wengstrom, E., 2011. Heterogeneity in risky choice behavior in a broad population. American Economic Review 101 (2), 664–694. https://doi.org/10.1257/aer.101.2.664. Gneezy, U., Leonard, K.L., List, J.A., 2009. Gender differences in competition: Evidence from a matrilineal and a patriarchal society. Econometrica : journal of the Econometric Society 77 (5), 1637–1664. https://doi.org/10.3982/ECTA6690. Gneezy, U., Niederle, M., Rustichini, A., 2003. Performance in competitive environments: Gender differences. The Quarterly Journal of Economics 118 (3), 1049–1074. https://doi.org/10.1162/00335530360698496. Gneezy, U., Rustichini, A., 2004. Gender and competition at a young age. American Economic Review 94 (2), 377–381. https://doi.org/10.1257/0002828041301821. Gong, B., Yang, C.-L., 2012. Gender differences in risk attitudes: Field experiments on the matrilineal mosuo and the patriarchal Yi. Journal of Economic Behavior & Organization, Gender Differences in Risk Aversion and Competition 83 (1), 59–65. https://doi.org/10.1016/j.jebo.2011.06.010. Grosse, N., Riener, G., Dertwinkel-Kalt, M., 2014. Explaining gender differences in competitiveness: Testing a theory on gender-task stereotypes. Social Science Research Network, Rochester, NY SSRN Scholarly Paper ID 2551206. http://papers.ssrn.com/ abstract=2551206. Günther, C., Ekinci, N.A., Schwieren, C., Strobel, M., 2010. Women can't jump?—An experiment on competitive attitudes and stereotype threat. Journal of Economic Behavior & Organization 75 (3), 395–401. https://doi.org/10.1016/j.jebo.2010.05. 003. Harrison, G.W., Lau, M.I., Rutström, E., 2007. Estimating risk attitudes in denmark: A field experiment*. Scandinavian Journal of Economics 109 (2), 341–368. https://doi. org/10.1111/j.1467-9442.2007.00496.x. Harrison, G.W., Lau, M.I., Yoo, H.I., 2019. Risk attitudes, sample selection and attrition in a longitudinal field experiment. The Review of Economics and Statistics 1–45. June. https://doi.org/10.1162/rest_a_00845. Harrison, G.W., List, J.A., 2004. Field experiments. Journal of Economic Literature 42 (4), 1009–1055. https://doi.org/10.1257/0022051043004577. Hogarth, R.M., Natalia, K., Carlos, A.T., 2012. When should i quit? Gender differences in exiting competitions. Journal of Economic Behavior & Organization, Gender Differences in Risk Aversion and Competition 83 (1), 136–150. https://doi.org/10. 1016/j.jebo.2011.06.021. Iriberri, N., Rey-Biel, P., 2017. Stereotypes are only a threat when beliefs are reinforced: On the sensitivity of gender differences in performance under competition to information provision. Journal of Economic Behavior & Organization 135 (March), 99–111. https://doi.org/10.1016/j.jebo.2017.01.012. Jetter, M., Walker, J.K., 2018. The gender of opponents: Explaining gender differences in performance and risk-taking? European Economic Review, Gender Differences in the Labor Market 109 (October), 238–256. https://doi.org/10.1016/j.euroecorev.2017. 05.006. Khachatryan, K., Dreber, A., von Essen, E., Ranehill, E., 2015. Gender and preferences at a young age: Evidence from Armenia. Journal of Economic Behavior & Organization, Economic Experiments in Developing Countries 118 (October), 318–332. https://doi. org/10.1016/j.jebo.2015.02.021. Nelson, J.A., 2015. Are women really more risk-averse than men? A re-analysis of the literature using expanded methods. Journal of Economic Surveys 29 (3), 566–585. https://doi.org/10.1111/joes.12069. Niederle, M., 2016. Gender. In: Kagle, J., Roth, A.E. (Eds.), Handbook of Experimental Economics, second edition. Princton University Press, pp. 481–553. Niederle, M., Vesterlund, L., 2007. Do women shy away from competition? Do men compete too much? The Quarterly Journal of Economics 122 (3), 1067–1101. https://doi.org/10.1162/qjec.122.3.1067. Nosek, B.A., Banaji, M.R., Greenwald, A.G., 2002. Math = male, me = female, therefore Math Not = Me. Journal of Personality and Social Psychology 83 (1), 44–59. Säve‐Söderbergh, J., Lindquist, G.S., 2017. Children do not behave like adults: Gender gaps in performance and risk taking in a random social context in the high-stakes game shows jeopardy and junior jeopardy. The Economic Journal 127 (603), 1665–1692. https://doi.org/10.1111/ecoj.12355. Sepahvand, M., Shahbazian, R., 2017. “Individual's risk attitudes in Sub-Saharan Africa: Determinants and reliability of self-reported risk in Burkina Faso. Working Paper Series 2017, 11. Uppsala University, Department of Economics. https://econpapers. repec.org/paper/hhsuunewp/2017_5f011.htm. Shurchkov, O., 2012. Under pressure: Gender differences in output quality and quantity under competition and time constraints. Journal of the European Economic Association 10 (5), 1189–1213. https://doi.org/10.1111/j.1542-4774.2012.01084.x.

Almås, I., Cappelen, A.W., Salvanes, K.G., Sørensen, E., Tungodden, B., 2015. Willingness to compete: Family matters. Management Science December. https://doi.org/10. 1287/mnsc.2015.2244. Almenberg, J., Dreber, A., 2015. Gender, stock market participation and financial literacy. Economics Letters 137 (December), 140–142. https://doi.org/10.1016/j. econlet.2015.10.009. Andersen, S., Harrison, G.W., Lau, M.I., Rutström, E., 2014. Discounting behavior: A reconsideration. European Economic Review 71 (October), 15–33. https://doi.org/ 10.1016/j.euroecorev.2014.06.009. Andersen, S., Ertac, S., Gneezy, U., List, J.A., Maximiano, S., 2013. Gender, competitiveness, and socialization at a young age: Evidence from a matrilineal and a patriarchal society. Review of Economics and Statistics 95 (4), 1438–1443. https://doi. org/10.1162/REST_a_00312. Angrist, J.D., Pischke, J.S., 2008. Mostly harmless econometrics: An empiricist's companion. Princeton university press. Antonovics, K., Arcidiacono, P., Walsh, R., 2009. The effects of gender interactions in the Lab and in the field. Review of Economics and Statistics 91 (1), 152–162. https://doi. org/10.1162/rest.91.1.152. Apicella, C.L., Dreber, A., 2015. Sex differences in competitiveness: Hunter-gatherer women and girls compete less in gender-neutral and male-centric tasks. Adaptive Human Behavior and Physiology 1 (3), 247–269. https://doi.org/10.1007/s40750014-0015-z. Beauchamp, J.P., Cesarini, D., Johannesson, M., 2017. The psychometric and empirical properties of measures of risk preferences. Journal of Risk and Uncertainty 54 (3), 203–237. https://doi.org/10.1007/s11166-017-9261-3. Benjamin, D.J., Choi, J.J., Strickland, A.J., 2010. Social identity and preferences. American Economic Review 100 (4), 1913–1928. https://doi.org/10.1257/aer.100. 4.1913. Bertrand, M., 2011. New perspectives on gender. Handbook of labor economics 4. Elsevier, pp. 1543–1590. http://linkinghub.elsevier.com/retrieve/pii/ S0169721811024154. Bonin, H., Dohmen, T., Falk, A., Huffman, D., Sunde, U., 2007. Cross-Sectional earnings risk and occupational sorting: The role of risk attitudes. Labour Economics, Education and Risk Education and Risk S.I. 14 (6), 926–937. https://doi.org/10.1016/j.labeco. 2007.06.007. Booth, A., Cardona-Sosa, L., Nolen, P., 2014. Gender differences in risk aversion: Do single-sex environments affect their development? Journal of Economic Behavior & Organization 99 (March), 126–154. https://doi.org/10.1016/j.jebo.2013.12.017. Booth, A.L., Nolen, P., 2012a. Gender differences in risk behaviour: Does nurture matter? Economic Journal 122 (558). http://ideas.repec.org/a/ecj/econjl/ v122y2012i558pf56-f78.html. Booth, A., Nolen, P., 2012b. Choosing to compete: How different are girls and boys? Journal of Economic Behavior & Organization 81 (2), 542–555. https://doi.org/10. 1016/j.jebo.2011.07.018. Boschini, A., Dreber, A., von Essen, E., Muren, A., Ranehill, E., 2018. Gender and altruism in a random sample. Stockholm University, Department of Economics Research Papers in Economics 2015:7. https://econpapers.repec.org/paper/hhssunrpe/2015_ 5f0007.htm. Boschini, A., Muren, A., Persson, M., 2012. Constructing gender differences in the economics lab. Journal of Economic Behavior & Organization 84 (3), 741–752. https:// doi.org/10.1016/j.jebo.2012.09.024. Buis, M.L., 2010. Stata tip 87: Interpretation of interactions in nonlinear models. The stata journal 10 (2), 305–308. Buser, T., Niederle, M., Oosterbeek, H., 2014. Gender, competitiveness, and career choices. The Quarterly Journal of Economics 129 (3), 1409–1447. https://doi.org/ 10.1093/qje/qju009. Buser, T., Peter, N., Wolter, S.C., 2017. Gender, competitiveness, and study choices in high school: Evidence from Switzerland. American Economic Review 107 (5), 125–130. https://doi.org/10.1257/aer.p20171017. Cardénas, J.-C., Dreber, A., von Essen, E., Ranehill, E., 2012. Gender differences in competitiveness and risk taking: Comparing children in Colombia and Sweden. Journal of Economic Behavior & Organization, Gender Differences in Risk Aversion and Competition 83 (1), 11–23. https://doi.org/10.1016/j.jebo.2011.06.008. Croson, R., Gneezy, U., 2009. Gender differences in preferences. Journal of Economic Literature 47 (2), 448–474. Datta Gupta, N., Poulsen, A., Villeval, M.C., 2013. Gender matching and competitiveness: Experimental evidence. Economic Inquiry 51 (1), 816–835. https://doi.org/10.1111/ j.1465-7295.2011.00378.x. Dohmen, T., Falk, A., Huffman, D., Sunde, U., 2010. Are risk aversion and impatience related to cognitive ability? American Economic Review 100 (3), 1238–1260. https://doi.org/10.1257/aer.100.3.1238. Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J., Wagner, G.G., 2011. Individual risk attitudes: Measurement, determinants, and behavioral consequences. Journal of the European Economic Association 9 (3), 522–550. https://doi.org/10.1111/j.15424774.2011.01015.x. Dreber, A., von Essen, E., Ranehill, E., 2014. Gender and competition in adolescence: Task matters. Experimental Economics 17 (1), 154–172. Dreber, A., von Essen, E., Ranehill, E., 2011. Outrunning the gender gap—boys and girls compete equally. Experimental Economics 14 (4), 567–582. https://doi.org/10. 1007/s10683-011-9282-8. Eckel, C.C., Grossman, P.J., 2008a. Differences in the economic decisions of men and women: Experimental evidence. Social Science Research Network, Rochester, NY SSRN Scholarly Paper ID 1883696. http://papers.ssrn.com/abstract=1883696.

11

Journal of Behavioral and Experimental Economics 83 (2019) 101467

A. Boschini, et al.

Emma von Essen holds a PhD in Economic from Stockholm University. Emma is a researcher at the Institute for Social Research at Stockholm University and the Department of Economic and Business at Aarhus University. Emma's research concerns the understanding of unequal economic outcomes. In her research, she employs data from laboratory experiments, randomized control trails and register data. She is, for example, currently investigating gender differences in online behavior as well as gender differences among the board of directors.

de Sousa, J., & Hollard, G. (2015). “Gender differences: Evidence from field tournaments. ” 1506. CEPREMAP Working Papers (Docweb). CEPREMAP. https://ideas.repec.org/ p/cpm/docweb/1506.html. Steffens, M.C., Jelenec, P., Noack, P., 2010. On the leaky math pipeline: Comparing implicit math-gender stereotypes and math withdrawal in female and male children and adolescents. Journal of Educational Psychology 102 (4), 947–963. https://doi.org/ 10.1037/a0019920. Wozniak, D., Harbaugh, W.T., Mayr, U., 2014. The menstrual cycle and performance feedback alter gender differences in competitive choices. Journal of Labor Economics 32 (1), 161–198. https://doi.org/10.1086/673324. Yong, E., 2012. Nobel laureate challenges psychologists to clean up their act. Nature News Accessed July 27, 2018. https://doi.org/10.1038/nature.2012.11535. Zhang, Y.J., 2013. Can experimental economics explain competitive behavior outside the lab? Social Science Research Network, Rochester, NY SSRN Scholarly Paper ID 2292929. https://papers.ssrn.com/abstract=2292929.

Astri Muren, Professor of Economics, Stockholm University. My research interests are in experimental and behavioral economics, particularly social preferences and gender, also industrial organization and regulation. I received my PhD in Economics from Princeton University in 1989, was Assistant Professor at Temple University, Philadelphia, PA, 1988–1991 and have been at Stockholm University since 1992. I am currently Dean of the Faculty of Social Sciences and Deputy Vice President of the Human Sciences scientific area at Stockholm University.

Anne Boschini is an associate professor and lecturer in economics at the Swedish Institute for Social Research (SOFI), Stockholm University. Her research focuses on gender-specific patterns in behavior, educational choices and labor market outcomes using both experimental and register data. Recently Anne is working on understanding income inequality from a gender perspective as well as the economic impact of male fertility choices.

Eva Ranehill holds a PhD from Stockholm School of Economics and now works as an assistant professor at the University of Gothenburg. She is a behavioral economist whose research employs laboratory and field experiments and empirical studies with archival register data. A large part of her research focuses on behavioral gender gaps—specifically on the robustness of gender preference gaps, what causes such differences, and their economic implications.

Anna Dreber Almenberg is the Johan Björkman professor of economics at the Stockholm School of Economics, and she is also an affiliated professor at the Department of Economics at the University of Innsbruck. She received her PhD in economics from the Stockholm School of Economics in 2009. Her interdisciplinary research focuses on variation in economic preferences between and within individuals and on the reproducibility of scientific results. Her research has been published in top science journals including Nature, Science, Proceedings of the National Academy of Sciences, as well as top journals in economics and psychology such as American Economic Review, Psychological Science and Management Science.

12