1
Explaining the Beauty Premium:
The Role of Competition
Master Thesis
Author: Suzanne Stevens
Student number: 10005781
Thesis supervisor: Jeroen van de Ven
Date: May, 2014
Faculty of Economics and Business
MSc Business Economics
Specialization: Organisation Economics
ECTS: 15
2
Abstract
This thesis is about the beauty premium. The literature shows that there is a lot of evidence for the existence of the beauty premium, but there is relatively little agreement on the reasons for this premium. The gender gap can partly be explained by preference for competition and this thesis tests whether the beauty premium can also partly be explained by this. The results from an experiment, performed among 112 high school students, show no significant relation between attractiveness and preference for competition. The standard errors are high, so there is also no disproof of the relation. Controlling for performance, confidence, risk and sports does still not give significant results for attractiveness. However, confidence and risk do significantly influence the choice for competitiveness. Further research should be performed to investigate this relation, using a better data set.
3
Table of Contents
Introduction ________________________________________________________________________4 Related Literature ___________________________________________________________________5 The Beauty Premium _______________________________________________________________5 The Beauty Premium: Evidence _____________________________________________________5 The Beauty Premium: Possible Explanations ________________________________________ 11 The Gender Gap and Preference for Competition ______________________________________ 14 The Beauty Premium and Preference for Competition __________________________________ 17 Methodology _____________________________________________________________________ 18 Experimental design ______________________________________________________________ 18 Subject pool __________________________________________________________________ 20 Variables _____________________________________________________________________ 21 Hypotheses ___________________________________________________________________ 25 Results __________________________________________________________________________ 26 Summary Statistics _______________________________________________________________ 26 Attractiveness and Preference for Competition ________________________________________ 27 Entry Decision___________________________________________________________________ 39 Gender and Preference for Competition _____________________________________________ 40 Possible Limitations ______________________________________________________________ 41 Discussion and conclusion ___________________________________________________________ 43 References _______________________________________________________________________ 45 Appendix A: Instructions Experiment __________________________________________________ 47 Appendix B: Non-Parametric tests ____________________________________________________ 57 B1. Attractiveness and Competition Choice ___________________________________________ 57 B2. Attractiveness and Beliefs ______________________________________________________ 57 B3. Attractiveness and Lottery Choice _______________________________________________ 58
4
Introduction
In 1994, Hamermesh and Biddle officially showed the existence of what is now called the beauty premium. Using real data, they show that attractive individuals earn around 5 percent more than individuals with average looks and that less attractive individuals earn 5 to 10 percent less than average looking individuals (Hamermesh and Biddle, 1994). Several studies have since then tried to find the reasons for the existence of the beauty premium. These studies show that the beauty premium may exist of partly discrimination from the side of the employer and partly through confidence channels from the employee’s view (Hamermesh and Biddle, 1994; Mobius and Rosenblat, 2006). It is important to research the reasons for the existence of the beauty premium. If discrimination plays a large role, the government may need to take steps to try to diminish this discrimination. Every possible explanation for the existence of the beauty premium should be investigated.
Besides a wage differential because of attractiveness, a lot of research has given attention to explaining the gender gap; the wage differential between men and women. Several studies have shown that the preference for competition may explain the wage gap between man and women (Niederle and Vesterlund, 2011). Since the wage gap may be explained by a different preference for competition between men and women, it may be possible that this preference for competition can also explain the beauty premium. Therefore, this study focuses on answering the following research question:
What is the relation between attractiveness and preference for competition?
This study will look at whether above-average attractive people have a higher preference for competition and whether below-average attractive people have a lower preference for competition than average attractive people. An experiment was performed in four high school classes. The results show no significant evidence of a relation between attractiveness and preference for competition. Including the possible influence of performance, confidence, risk and sport, still no relation is found. However, confidence, risk and performance in the tournament do significantly influence the choice for tournament.
The next section will show an overview of the literature relevant for this topic, the third section will explain the methodology used to answer the research question, the fourth section will give an overview of the results and the last section will conclude and discuss.
5
Related Literature
The Beauty Premium
In 1994, Hamermesh and Biddle were the first to critically study the phenomenon that is now known as the beauty premium. These authors use three broad household surveys (two from the United States and one from Canada) to study the marginal effect of looks, “after accounting for all the other causes of variations in earnings that are usually measured” (Hamermesh and Biddle, 1994, p. 1180). In all three surveys, the interviewer had to evaluate the attractiveness of the respondent on a five-point scale (Hamermesh and Biddle, 1994, p. 1179). The findings show that plain people receive around 5-10 percent lower wages than average-looking people. This is called the plainness penalty. The good-looking people earn more than the average-looking people, this is the beauty premium. Although the study only uses North American data, it clearly shows the need for more research into this phenomenon. After Hamermesh and Biddle, multiple other studies have evaluated the beauty premium. This section will discuss the most influencing studies in this field.
The Beauty Premium: Evidence Field studies
This section will discuss six important field studies which have shown the existence of the beauty premium. These studies all use different samples to evaluate the influence of beauty on success.
First of all, French (2002) uses a worksite-based dataset and a self-assessed measure of physical appearance to see whether the earlier findings on the beauty premium continue to hold. His results show that for both attractive and unattractive employees, relative to average-looking employees, the wage differentials are present. However, only the differentials for female workers are significant. Thus, he shows that appearance is important in the two worksites which he uses as a sample, although the result in his study is gender-specific (French, 2002, p. 572). His measure of attractiveness, based on self-report, may be a measure of self-esteem rather than physical appearance. However, his result does show a new perspective relative to previous research (French, 2002, p. 572).
Second, Berggren et al. (2007) study the effect of beauty in politics. They investigate the effect of the appearance of almost 2,000 Finnish political candidates on the number of
6
votes this politician gets. Their results show that an increase in beauty by one standard deviation is associated with an increase of 17-20 percent in the number of votes that an average non-incumbent candidate gets (Berggren et al., 2007). The study only evaluates the effect in Finland and may therefore not be generalizable. However, it does show that beauty does not only matter in businesses, but also other situations such as political elections.
Third, Fletcher (2009) studies the wage returns to attractiveness for young high school graduates. He examines whether the beauty premium represents unmeasured ability and whether beauty is complementary with ability in determining wages. He shows that a one standard deviation increase in ability raises wage by 3 to 6 percent. Attractive or very attractive individuals earn 5 to 10 percent more than average-looking individuals. Fletcher also finds the plainness penalty. Unattractive people earn 3 to 5 percent less than average-looking individuals (Fletcher, 2009, p. 321). Moreover, for very attractive individuals, increases in ability are associated with increases in wages, while for below-average levels of attractiveness the returns to ability may be negative. Fletcher includes many variables, to control for a variety of usually unmeasured variables (Fletcher, 2009, p. 323). Examples are multiple health measures, household income and a personality rating. The health of a person during childhood may for example influence both attractiveness, ability and wage later in life. Therefore, it is good that Fletcher controls for these factors. However, the sample of Fletcher only includes young adults who only finished high school education. This may decrease the external validity to people with higher educational levels.
Fourth, Robins et al. (2011) look into the effects of beauty, personality and grooming on wages. They use data from the National Longitudinal Study of Adolescent Health to find a significant beauty premium when beauty is the only personal trait included in the model. They do not find any evidence of the plainness penalty (Robins et al., 2011, p. 244). However, when both personality attractiveness and grooming are added to the model, the beauty premium is no longer significant for women (Robins et al., 2011, p. 236). For women, personality attractiveness matters more than physical appearance. For men, physical appearance generates a higher premium than having an attractive personality or being well groomed (Robins et al., 2011, p. 236). These authors conclude that including other personal traits in the model reduces the beauty premium, and part of the beauty premium found in earlier studies may be reflecting better grooming for men and women, and an attractive
7
personality for women (Robins et al., 2011, pp. 244-245). This study teaches us that other non-cognitive personal traits are also important determinants of labor market success (Robins et al., 2011, p. 245).
Fifth, looking at the choices made in a television game show, Belot et al. (2012) show that although there is no difference in performance between attractive and unattractive players, unattractive players are substantially more likely to be eliminated by other players. This process leads to costs of about €440 on average, so eliminating unattractive players is a costly process (Belot et al., 2012, p. 853). A questionnaire among subjects shows that subjects predict less attractive players to be eliminated sooner. This shows that they are implicitly aware of some discrimination or bias towards attractive players (Belot et al., 2012, p. 869). Thus, we can learn from this study that although there is no explicit awareness of discrimination, it does happen in a public setting and the society should look into it.
Last, López Bóo et al. (2013) perform a randomized field experiment to analyze the link between beauty and the labor market. They created fictitious resumes by digitally creating fictitious photos of people, ranging from unattractive to attractive, and submitted these to real job openings. They then analyzed the responses to find that attractive people received 36 percent more call-backs than unattractive people (López Bóo et al., 2013, p. 170). Although this research gives evidence for the beauty premium, the study does have some limitations. First, the study does not look at the actual hiring of the applicants, but only at the interviewing process. Moreover, only six occupational categories are taken into account. Furthermore, the pictures are created by a computer system. This system is new and not proven to work perfectly (López Bóo et al., 2013, p. 172).
Experiments
Next to field studies, several experiments have been performed, to research the influence of beauty on the behavior of people. Hereafter, four influential experiments will be discussed.
First, Solnick and Schweitzer (1999) study the influence of physical attractiveness and gender on ultimatum game decisions. They show that there are no significant differences in the decisions made by attractive and unattractive men and women. However, there are differences found in how attractive people are treated by the other participants of the game. Attractive people were offered more by the other players. Moreover, less was demanded
8
from attractive people (Solnick and Schweitzer, 1999, p. 201). The expected earnings of attractive people were 8-12 percent higher than for unattractive people (Solnick and Schweitzer, 1999, p. 210). Thus, attractive people may earn more because they are offered more. However, this game is simple and non-repeated and past relationships, future expectations and possible communications in real life do not influence this game. Thus, both contextual and individual factors might influence the results of this experiment.
Second, Wilson and Eckel (2006) research the effect of beauty in trust games. Attractive people earn more in the first stage of the game, since they are perceived as being more trustworthy. However, they are also expected to give more in the first stage of the game. Therefore, in the second stage, attractive players receive a beauty penalty because they do not live up to the expectations of the other players. The punishment is harder for attractive players then for unattractive players (Wilson and Eckel, 2006). This experiment teaches us that people trust strangers, trust pays and attractive people are trusted more. However, these players are strangers who are unlikely to interact in the future. If earlier interactions and possible future interactions are present, the results might be influenced.
Third, to explore beauty and gender, Andreoni and Petrie (2008) perform a laboratory experiment in the form of a linear public goods game, where there are benefits to group cooperation. They do not perform a direct test of the beauty premium or the wage gap, but they use the public goods game to explore the wage differences in an employment setting (Andreoni and Petrie, 2008, p. 74). There are two treatments in this study, the Information treatment and the No Information treatment. The results of the No Information treatment show that attractive people make 7 percent more than middle attractive people and 12 percent more than unattractive people, where the differences between the three categories of attractiveness are significant (Andreoni and Petrie, 2008, p. 79). This is evidence of the beauty premium. However, in the Information treatment, where the participants are informed on the contributions of the other participants, the beauty premium turns into a beauty penalty. In this treatment, unattractive people earn 8 percent more than attractive people (Andreoni and Petrie, 2008, p. 80). Thus, this study teaches us that when there is no information, beauty is rewarded and when there is information on performance, there is no beauty premium. However, a limitation of this study is that only an experiment is used and no real labor market evidence is given.
9
Fourth, Rosenblat (2008) uses a dictator game to study the influence of gender and physical attractiveness on behavior. Her results show that female allocators give more to recipients that are physically attractive. Men’s allocations are unaffected by physical attractiveness. This result holds when the allocators both see and hear the recipients (Rosenblat, 2008, pp. 467-468). This study teaches us that appearance matters in negotiation situations. However, the study does not show us the exact process. Moreover, in more complex situations, where both parties can influence the outcome and where there is richer communications, the results are less likely to hold (Rosenblat, 2008, p. 478). Therefore, the external validity should be tested in the real-world.
Body shape
Several studies investigated the influence of beauty in the form of body shape on earnings. In 2004, Persico et al. use data from two samples to research the effect of height on wages. They focus their attention on white, non-Hispanic men, to avoid the influence of race and gender (Persico et al., 2004, p. 1025). Using regression analysis, these authors show that the average wage of shorter men is about 10 to 11 percent lower than the wage of taller men (Persico et al., 2004, p. 1029). Although the data is limited in focusing only on white, non-Hispanic men, this study does show us that height matters.
Moreover, in 2005, Heineck evaluated whether taller workers earn more than their shorter counterparts. He finds that there are no wage differentials based on height for female workers and male East German workers. However, he does find a wage differential for the height range up to 195 cm for male workers from West Germany of about 4 percent. The data set is quite limited and the author does not take into account productivity differentials and self-selection of taller workers into specific jobs. The study does show some evidence that height matters.
Furthermore, Case and Paxson (2008) show that in developed countries, height is still highly related to success in the labor market. In both the US and the UK, taller workers mostly work in highly-skilled jobs. This is true for both men and women (Case and Paxson, 2008, p. 499). In numbers, they show that a one-inch increase in height leads to an increase in weekly earnings of about 1.4 to 2.9 percent and an increase of about 1.0 to 2.3 percent in average hourly earnings. Again, height matters.
10
Lastly, both Cawley (2004) and Averett and Korenman (1996) show that white women receive a wage penalty when being overweight. Cawley (2004) shows that for white females, a weight difference of two standard deviations is associated with a 9 percent difference in wages. Averett and Korenman (1996) show that white women who are obese at the ages 16 to 24 and 23 to 31 earn around 20 percent less than white women who are not obese (Averett and Korenman, 1998, p. 24). Both studies show that weight matters, but only for white women. They do not have clear results on why weight only matters for white women and not for men and other races.
Hair color
Next to looking only at total attractiveness of workers, several studies also looked specifically at the influence of hair color on wages. Price (2008) studies the effect of hair color and attractiveness on door-to-door fundraising results. He finds that the returns to physical appearance are 71.6 to 76 percent greater for blonde women than for brunette (minority) women. These results depend on the origin of the households that live in the homes where the women go fundraising. Blondes receive significantly higher donations from Caucasian households, but significantly lower donations from non-Caucasian households (Price, 2008, p. 351). This study shows that hair color does impact productivity in a door-to-door fund-raising experiment, but the effect depends on the characteristics of the donor.
Furthermore, Johnston (2010) uses U.S. panel data to study the effect of hair color on women’s own wages and on their spouse’s wage. He finds that blonde women receive a large wage premium. This premium is similar in size to the return from an extra year of schooling. The spouses of blonde women also receive higher wages (Johnston, 2012, p. 10). Probably, blond women have higher fortunes in the marriage market, since they are perceived as being more attractive and they have higher wages (Johnston, 2012, p. 11). This study again shows that hair color impacts earnings.
Moreover, in 2012, Guéguen studies the effect of employees’ hair color on wages. He uses an experiment to avoid that the results are influenced by other factors, like the age of the women, the length of their hair, race etc. (Guéguen, 2012, p. 370). By using an experimental method, the study becomes more controlled. Guéguen looks at the effect of hair color on the wage of waitresses. His results show that waitresses with blond hair receive
11
more tips from male customers. Moreover, they receive larger tips (Guéguen, 2012, p. 371). The hair color had no effect on female customers. Having blond hair could increase the wages of the waitresses by about 6.1 percent (Guéguen, 2012, p. 371). However, the behavior of the waitress could change with hair color and she could therefore get more tips. Therefore, it is not clear where the effect comes from.
The Beauty Premium: Possible Explanations
A lot of studies have shown the existence of the beauty premium; attractive people do earn higher wages than unattractive people in many situations. Moreover, also in other situations (politics, bargaining and negotiation situations, television game shows) attractive people have advantages over unattractive people. Multiple studies have also shown the existence of a plainness penalty; below-average looking people earn less than average looking people. However, earlier studies are not clear on the exact reasons for the existence of the beauty premium and/or plainness penalty. This section will elaborate on the evidence given towards possible explanations of the premium. The most noteworthy reasons for the existence of the beauty premium are discrimination, both by the customers and the employer, stereotyping and confidence of the employee.
The initiators of the existence of the beauty premium, Hamermesh and Biddle (1994) gave three possible reasons for the beauty premium. The first is pure employer discrimination. Hereby, employers simply discriminate against the unattractive (Hamermesh and Biddle, 1994, p. 1178). The second is customer discrimination/productivity. It is possible that customers prefer to deal with better-looking individuals. It is also possible that in some occupations above-average attractive individuals can better interact with coworkers. Both factors make attractive people more productive, and therefore they can earn higher wages (Hamermesh and Biddle, 1994, p. 1177). The third and last possible reason is occupational crowding. Hereby, restricting unattractive workers in some occupations brings down the wages of all workers in these occupations (Hamermesh and Biddle, 1994, pp. 1178-1179). These authors only find limited evidence for occupational sorting. The strongest evidence for possible reasons for the beauty premium is for employer discrimination, stemming from employer/employee tastes (Hamermesh and Biddle, 1994, p. 1193). Later data from Robins et al. (2011) also shows that both employer discrimination and a higher productivity (possibly through customer discrimination) play an important role in raising the beauty premium.
12
In 2006, Mobius and Rosenblat performed a study to research the reasons of the beauty premium. They found a sizeable beauty premium and identified three channels through which this premium could arise. All three channels can influence the estimation of the worker’s ability by the employer. The first channel is the confidence channel. According to this channel, physically attractive people are more confident and this higher confidence can influence the level of wage when there is oral interaction between the employer and employee. The second channel is the visual stereotype channel. Here, the employer has the expectation that attractive employees perform better than unattractive employees. The third channel is the oral stereotype channel. Under this channel, the attractive worker can have better oral skills, leading their employer to expect more from this worker and therefore raising his or her wage (Mobius and Rosenblat, 2006, pp. 222-223). Mobius and Rosenblat vary the degree of visual and oral interaction between the worker and employer to decompose the beauty premium in these three effects. They also measure employee confidence (Mobius and Rosenblat, 2006, p. 222). Their results show that all three channels are important. About 15-20 percent of the beauty premium is transmitted through the confidence channel. The visual stereotype channel and the oral stereotype channel each account for about 40 percent of the beauty premium (Mobius and Rosenblat, 2006, p. 234). Thus, both confidence and stereotyping play an important role in the existence of the beauty premium. Due to the fact that an experimental data set is used, instead of a real-world data set, the results could be different in a real-life setting. Moreover, Mobius and Rosenblat could have improved their study by lengthening the interaction period.
Andreoni and Petrie (2008) use a public goods experiment to research the influence of cooperation on the beauty premium. In this game, all players have to decide on how much to contribute to a public good, so how much to cooperate with the other players. The attractive players do not contribute more or less than unattractive players. They do earn more in the No Information treatment, where all players do not receive information on the contributions of the other players. However, in the Information treatment (where all players receive information on the contributions of the other players) there is a matter of a beauty penalty. Both these effects are due to stereotyping. In the No Information treatment, players expect more cooperation from attractive people. As stated by Andreoni and Petrie: “attractive people are consistently judged and treated more positively” (Andreoni and
13
Petrie, 2008, p. 82). Therefore, the other players are more cooperative with attractive players. However, since these attractive players do not contribute more than average, in the Information treatment, they are penalized for not living up to the expectations. Attractive people are thus stereotyped as being better cooperators, and this may explain part of the higher wage and advantages offered to them when there is no information of real performance yet.
Contrary to what Andreoni and Petrie (2008) found, Belot et al. (2012) find in their study that subjects predict that attractive people will be less cooperative. It seems that the stereotype of attractive people is not constant over different samples. These authors study evidence from a television game show in order to show the existence of the beauty premium. They also study the possibility of a performance differential to be the reason of the premium. However, they show that there is no evidence that attractive people perform better or worse than unattractive people (Belot et al., 2012, p. 859). Moreover, they also show that confidence is not a reason for the beauty premium. Their data show that attractive players do not behave more confidently than unattractive players (Belot et al., 2012, p. 860). Their main reason for the beauty premium is taste-based discrimination. They show that discrimination favors attractive people because of consumption value considerations (Belot et al., 2012, p. 867). Either discriminators want to be accompanied by attractive people and want to pay for this, or they enjoy being watched on television surrounded by attractive people. Thus, Belot et al. (2012) show that taste-based discrimination is the primary reason for the existence of the beauty premium in their study.
Solnick and Schweitzer (1999), Wilson and Eckel (2006) and Rosenblat (2008) all use behavioral game theory to study other possible channels through which the beauty premium can work. Solnick and Schweitzer (1999) show in a bargaining ultimatum game that attractive players do not demand more than unattractive players. However, they are treated differently and offered more. It is not clear whether this stems from discrimination or stereotyping. Wilson and Eckel (2006) studied behavior in a trust game to research the beauty premium. Their results are similar to the results of Andreoni and Petrie (2008). Attractive players earn more in the first stage because they are trusted more. Probably there is some stereotyping involved which makes attractive people seem more trustworthy. However, in stage 2, the attractive players do not live up to the expectation, and are being
14
penalized. So when performance of attractive players is unknown, they are trusted more and can therefore receive higher payoffs/wages. Rosenblat (2008) investigates a new channel through which the beauty premium can be explained. She calls it the negotiation channel. It states that attractive workers negotiate more effectively and will therefore receive higher wages (Rosenblat, 2008, p. 466). Using a dictator game, she shows that female allocators give more to attractive people when they can both hear and see them. This evidence shows that attractiveness matters, but the recipients also need to convince the allocator to give more to them (Rosenblat, 2008, pp. 467-468).
In short, Hamermesh and Biddle suggested three possible reasons for the existence of the beauty premium, namely employer discrimination, customer discrimination and occupational crowding. Further research has found evidence for employer discrimination (Belot et al., 2012; Robins et al., 2011) and customer discrimination (Robins et al., 2011). Mobius and Rosenblat (2006) find evidence for both the confidence channel and the stereotyping channel. Belot et al. (2012) do reject both channels and only find evidence for taste-based discrimination. However, Andreoni and Petrie (2008) and Wilson and Eckel (2006) support the evidence of the stereotyping channel. Rosenblat (2008) adds to the discussion by suggesting the influence of the fact that attractive people may be better negotiators. All in all, the evidence on the exact reasons for the beauty premium is still mixed; both pure discrimination and discrimination because of different expectations may influence the wage differential.
The Gender Gap and Preference for Competition
Next to seeing a wage differential based on attractiveness, many discussions are focused on the wage differential between men and women. For the last 50 years, women started to participate more actively in the labor-force. However, this change is not matched by a narrowing of the wage gender gap (O’Neill, 2003, p. 309). The female-to-male ratio of median annual earnings has stabilized at around 80 percent in 1993. As O’Neill (2003) finds, much of the gender gap can be explained by the difference in work experience. The percentage female in occupation also explains part of the gender gap (O’Neill, 2003, p. 313). Thus, the gender gap can largely be explained by nondiscriminatory factors, but part of the gap still remains unexplained (O’Neill, 2003, p. 314). Other studies identified discrimination
15
and the fact that men and women differ in both their abilities and their preferences over jobs as possible explanations of the gender gap (Niederle and Vesterlund, 2011, p. 602).
Over the past decade, many researchers have studied another possible explanation for the persisting gender gap. It is possible that gender differences in preference for competition may explain part of the gap. As Niederle and Vesterlund (2011) state: “If women are more reluctant to compete, then they may be less likely to seek promotions or to enter male-dominated and competitive fields” (p. 602). This may explain why women work in functions where wages are lower. Niederle and Vesterlund (2011) review the growing literature on gender differences in competition. They evaluate both experiments and field evidence, and they investigate what measures can be taken to alter the gender gap in willingness to compete. This section will discuss two articles on this subject in detail, and will further refer to Niederle and Vesterlund (2011) for important conclusions.
In 2007, Niederle and Vesterlund were the first to study the relation between gender and preference for competition in an economical experiment. They look at the type of compensation scheme chosen by both men and women, when holding other job characteristics constant (Niederle and Vesterlund, 2007, p. 1067). As men will more often choose to compete, they will probably also win more often. So they will also work in more lucrative jobs, since they have won the competitions for promotions more often. Niederle and Vesterlund use a controlled laboratory experiment where groups of two women and two men have to perform a real task, namely adding sets of numbers for five minutes (2007, p. 1068). They first perform the task under a piece-rate payment scheme. Second, they perform the task under a tournament payment scheme. For the third task, the participants have to choose which compensation scheme will be applied. The results show that twice as many men as women select the tournament. Low-ability men seem to enter the tournament too often, while high-ability women do not enter the tournament enough (Niederle and Vesterlund, 2007, p. 1069). These authors investigate multiple possible explanations for their finding that men more often choose tournament than women. They conclude that the difference can partly be explained by overconfidence, but risk aversion and feedback aversion do not seem to play a role. Furthermore, controlling for overconfidence, risk, and feedback aversion, the gender difference remains significant and large. The authors conclude that “a sizeable part of the gender difference in tournament entry is explained by
16
men and women having different preferences for performing in a competitive environment” (Niederle and Vesterlund, 2007, p. 1070). So men simply seem to like competition more than women, and therefore enter a tournament of competitive environment more often than women. This causes them to get promoted more often and earn higher wages. So this preference for competition can explain part of the gender gap.
Buser et al. (forthcoming) look at the influence of the difference in preference for competition on actual decision making by people. They research the educational choice of high school students and the influence of competitiveness on this choice. Male students choose a math heavy study track more often than female students, even though some are equally gifted (Buser et al., forthcoming, p. 2). Competitiveness may explain this difference. Buser et al. (forthcoming) use the same measure of competitiveness as Niederle and Vesterlund (2007). The high school students perform a real task of adding numbers, and have to decide in task 3 whether to enter the tournament or not. The results show that male students choose prestigious academic tracks more often than female students, even though the performance of females (including math grades) is are least as good as that of males. Moreover, males are twice as likely to choose the tournament payout scheme, even though performance is similar (Buser et al., forthcoming, p. 3). The preference for competition accounts for about 20 percent of the gender gap between choosing the lowest and highest ranked study track. Confidence and risk attitude do not affect this effect. When controlling for these two factors, the preference for competition still accounts for 16 percent of the gender gap. These results also validate the measure of Niederle and Vesterlund (2007) (Buser et al., forthcoming, p. 4).
Niederle and Vesterlund review multiple studies and conclude that these results on the gender gap in tournament entry are robust (2011, p. 605). They also show that differences in beliefs only explain part of the gender gap, risk attitudes only play a limited role and controlling for performance does not remove the gender gap (Niederle and Vesterlund, 2011, p. 611). Field studies show that the findings in the laboratory continue to hold in multiple real-life situations. Moreover, the performance gap between men and women seems to increase when there is more competitive pressure (Niederle and Vesterlund, 2011, p. 618). Thus, preference for competition has a significant influence on the persistence of the gender wage gap.
17
The Beauty Premium and Preference for Competition
The previous sections have shown that there is a lot of evidence suggesting the existence of the beauty premium. The evidence on reasons for this premium remains mixed. Furthermore, the gender gap in wages persists to exist. Recent research has shown the influence of preference for competition on the gender gap. Males prefer to compete more often than women and therefore enter in competition for promotions more often. They will win these competitions more often, since they enter more often, and will therefore work in higher jobs and earn more than women. This can explain part of the gender gap.
Previous research has not yet combined these two streams of research. The beauty premium may also be explained by preference for competition. Below-average attractive people may be reluctant to compete for promotions and higher jobs, and may therefore earn less than above-average attractive people, who do compete. This study will research whether preference for competition can also explain part of the beauty premium, just like it explains part of the gender gap in wages.
If the data show that above-average attractive people choose tournament more often, this can indicate that in real-life they will also compete more often. Therefore, it is possible that they enter the competition for promotions more often and win these competitions more often. If this is the case, then attractive people will work in higher positions and earn higher wages than unattractive people. The opposite will hold for below-average attractive people.
18
Methodology
Experimental design
To answer the research question, an experiment was performed. In this first section, I will describe the content of the experiment and the next sections will describe the subject pool, the variables and the hypotheses of this study.
An experimental design is chosen to exclude the influence of factors such as discrimination or family preferences. This experiment is based on the experiments of Niederle and Vesterlund (2007) and Buser et al. (forthcoming). In this experiment, the participants have to solve a simple addition task three times. Task 1 works with a piece-rate compensation schedule, task 2 with a competitive tournament compensation schedule and in the third task, the participants can choose which compensation schedule they prefer. Moreover, the participants have to complete a questionnaire with background information on themselves, their confidence, risk attitude and attractiveness. This provides me with information on both the preference for competition and attractiveness, to determine whether attractive and unattractive people of equal performance choose the same compensation scheme.
Task
The task of the experiment is to add four two-digit numbers. The participants are given 25 problems on a sheet, and they have three minutes to solve as many problems as possible. They are not allowed to talk during the experiment, or to use a calculator. However, they are allowed to use the empty space on the sheet as scrap paper.
There are two versions of each task, to ensure that two participants sitting next to each other cannot copy answers from each other. The experimenter pays attention to make sure that there is no cheating. The numbers are randomly drawn by excel, using a distribution from 10 to 99. This is done to ensure that the difficulty level of the problem sets across the three tasks and the versions are not significantly different. A problem looks as follows:
19
The participants have to fill in the number 259 in the blank cell. The score of a participant is determined by the number of correct answers given within three minutes. Incorrect answers are not subtracted from the score. Following Niederle and Vesterlund, this simple addition task is chosen because it requires both skill and effort and there are no gender differences in performance (2007, p. 1074). It will have to be checked whether there are also no differences between the performance of attractive participants and the performance of unattractive participants.
Compensation
For each experiment, so in each of the four classes, the computer randomly selects four participants, one participant for each part of the experiment, to receive a compensation. The participants do know that only one randomly determined participant will be paid out in each part, however, they do not know which participant is chosen. After receiving the instructions with specific information on the payout method of task 1, the experiment will start with this first task. In task 1, the participants can earn €0.25 for each correct answer given within the three minutes. With the total of 25 problems per sheet, €6.25 can be earned in task 1.
Task 2 is the tournament payout task. The participants again have to solve as many problems as possible within three minutes. However, now they will only receive a payment if they perform better than three other people in the participant’s group. Before the experiment, the computer randomly matched each participant to three other participants within the class. The participants know that they are in a group with three other participants, however, they do not know which participants. If randomly chosen, the participant will only be paid if his or her performance is better than the performance of the three other people in his or her group. If he or she performs best, the payout will be €1.00 for each correct answer. Otherwise, the compensation will be €0.00.
In task 3, before completing the problem set, the participants have to choose whether they will be paid a piece-rate or be paid like the tournament compensation schedule. After being given instructions, they have to make a decision on how to be paid. After this choice, they receive the third problem set with 25 problems. If randomly selected by the computer, the participant will be paid according to his or her choice. If the participant chooses the piece-rate compensation scheme, payment will be €0.25 for each correct
20
answer. If the participant chooses the tournament compensation scheme, he or she will receive €1.00 for each correct answer, provided that his or her performance exceeds the performance of the three other group members in task 2. Otherwise, compensation is €0.00. After task 3, the participants have to complete a questionnaire with background information, and questions to infer the level of confidence, risk attitude and attractiveness. They can earn money with two questions. In question 2, they can earn €1.00 if they guess their correct rank. In question 4, they have to decide between 5 different lotteries. They can earn up to €6.00, depending on their decision and on chance. The lotteries can be found in the questionnaire in appendix A.
Subject pool
A high school in Heemstede was invited to participate in this research project. I was granted permission to perform the experiment in four classes. All the students in these classes were in their fourth year of high school, comparable to the 10th grade in the American school
system. The experiments were done during the economics class, so the students knew it was an economical experiment, though they did not know anything on the content of the experiment. All these students are studying at the pre-university level (Voortgezet Wetenschappelijk Onderwijs in the Dutch system). The experiments lasted a bit less than one class hour (50 minutes). The exact, detailed instructions and questionnaire can be found in appendix A.
In each class, all students agreed to participate in the experiment. In total, 112 students participated, among which 69 male students and 43 female students. The data collection in the classes all took place on one day, Thursday April 10, however not on the same time during the day. It is therefore possible that some information leaked from one class to another. Therefore, even after the experiment, no information on the content of the experiment was disclosed to the students. The selected participants were paid the Wednesday after the experiment through sealed envelopes. They earned an average of €1.94, with a minimum of zero and a maximum of €7.00.
21 Variables
Competitiveness
The measure of competitiveness is based on Niederle and Vesterlund (2007) and Buser et al. (forthcoming). In task 3, the participants have to make a choice between the non-competitive piece-rate compensation scheme and the non-competitive tournament compensation scheme. This choice serves as a measure of preference for competition. Since the participants first go through both the piece-rate payout scheme and the tournament payout scheme, they have experience with both compensation schemes and are able to make a well-considered decision.
Just like in Buser et al. (forthcoming), the participants do not receive feedback on their performance, but they know the number of problems they completed. Based on Niederle and Vesterlund (2007), in task 3, the performance of a participant is compared to the performance of the three other group members in task 2. Therefore, the decision made on the compensation scheme cannot affect the payments of other participants. Moreover, it is only a decision based on beliefs about performance and not on beliefs about the decision of other participants.
Concluding, the variable Choice is a dummy variable which is 1 when the participant chooses the tournament compensation scheme and 0 when the participant chooses the piece-rate compensation scheme.
Attractiveness
Earlier studies show multiple methods in measuring attractiveness. Most studies use third party facial ratings. Photos are shown to third party raters, who rate the attractiveness of the subject in the picture on a particular scale. Other studies use an objective measure of attractiveness like height, BMI (body mass index), WHR for women (waist hip ratio) and CWR for men (chest waist ratio). A third method is to let the participants rate themselves. In 2007, Weeden and Sabini compared the different measures. They concluded: “Self-rated attractiveness was moderately related to the physical measurements the literature calls attention to (BMI and WHR for women and BMI and CWR for men) and strongly related to ratings of facial attractiveness by third parties” (Weeden and Sabini, 2007, p. 85).
22
The high school did not give permission to take photos of the participants, and therefore third party ratings were impossible. However, since self-rated attractiveness is strongly related to third party ratings, this subjective measure is used for attractiveness. French (2002) also uses a self-assessed measure of physical appearance, and he finds evidence for the beauty premium. Moreover, BMI and height are used as an objective measure of attractiveness.
The participants have to state their height and weight. Using the formula for Body Mass Index �(ℎ𝑤𝑤𝑤𝑤𝑤𝑤ℎ𝑡𝑡)𝑤𝑤𝑤𝑤𝑤𝑤𝑤𝑤ℎ𝑡𝑡2�, BMI can be calculated. Moreover, the participants have to answer two questions on their attractiveness. The first question is: “On a scale of 1-7, where 4 is average, how attractive do you think you are?”. They give an answer on a 1-7 scale with 1 being “not attractive at all” and 7 being “really attractive”. Similarly, they have to answer the question: “On a scale of 1-7, where 4 is average, how attractive you think that others think you are?”.
The attractiveness is evaluated by using three categories. The reason for this is that most other studies also look at above-average, below-average and average attractiveness. Moreover, only a small amount of people choose specific numbers of attractiveness (e.g. only 1 person chose 1). The first category is below, which indicates that the participant chose 1-3 at the scale for attractiveness. Thus, this participant evaluates his or her attractiveness as below-average. The second category, average, indicates that the participant chose 4. The third and last category, above, indicates that the participants chose a number above average for his or her attractiveness, so 5-7.
Concluding, the variables BMI and height measure attractiveness in an objective way and the variables below, average and above measure subjective self-rated attractiveness for the first question and expected third party attractiveness for the second question. The variable below is 1 when the rating is below 4 and 0 otherwise. The variable above is 1 when the rating is above 4 and 0 otherwise. When both variables are 0, the participant rates him- or herself as average.
Performance
As studies before have shown, ability may have an effect on the beauty premium (see for example Fletcher, 2009). People can earn higher wages when they are more able to perform
23
a specific task. People with better looks may be better or worse in performing some tasks. Moreover, a higher performance will probably sooner lead to the choice for tournament in task 3. Thus, performance may affect both the preference for competition and be related to attractiveness. Therefore, it needs to be included in the analysis.
Performance will be evaluated in two ways. The first is performance in the tournament task, task 2. This performance will probably influence the choice made in task 3, and it can be related to looks. Moreover, the performance in task 1 may also have an influence, and particularly the growth in performance from task 1 to task 2. Thus, following Niederle and Vesterlund (2007), a variable measuring the difference in performance between task 1 and task 2 will also be included.
Concluding, two variables will be used to measure performance. The first is Task 2, which gives the number of correctly solved problems in task 2. The second is Task 2 – Task 1, which is the difference between correctly solved problems in task 2 and task 1.
Confidence
This study uses two different confidence measures. The first measure is taken from both Niederle and Vesterlund (2007) and Buser et al. (forthcoming). The choice the participants make in task 3 may depend on their expected performance, relative to the performance of the other participants. Therefore, students are asked to guess their rank in the task 2 tournament, from 1 (best) to 4 (worst). To make the guess serious, subjects can earn €1.00 if their guess is correct.
The second measure of confidence is a more general confidence measure. It is adjusted from Clifton and Gill (1994), who study the confidence of cheer leaders. The participants are asked “On a 1-7 scale, with 4 being average, how confident are you in your own mathematical ability?”. They give an answer on a 1-7 scale with 1 being “no confidence at all” and 7 being “a lot of confidence”. This wording is adapted from Clifton and Gill (1994, p. 153), who adapted it themselves from Eccles and Harold (1991, p. 19).
Concluding, two measures of confidence are used in this study. The first is an applied measure of confidence, based on participants’ beliefs, Belief. The second is a more general confidence measure, Confidence Math.
24 Risk attitude
The two measures for risk attitude used in this study are taken from Buser et al. (forthcoming). They based their measures on three other studies (for details, see Eckel and Grossman, 2002; Dohmen et al., 2011; Lonnqvist et al., 2010). In the first measure, participants have to choose a lottery. They choose an option between a sure payoff of €2.00 and four 50/50 options with increasing riskiness and expected payoffs. The options are: 3.50 or 1.50; 4.00 or 1.00; 5.00 or 0.50; 6.00 or 0. If the participant is chosen for payment, the outcome of the lottery will be decided by tossing.
The second measure of confidence asks participants the following question: “How do you see yourself: Are you a person that is willing to take a lot of risk, or do you try to avoid risk as much as possible?” The participants answer this question on a 1-10 scale with 1 being “do not want to take risk” and 10 being “want to take a lot of risk”.
Concluding, there are two measures of confidence. Lottery gives the choice of a participant for a lottery and Risk Level gives the risk attitude on a 1-10 scale.
Sport
We can also look at the influence of sport decisions. Sport can change the way how you feel about your attractiveness, but also how competitive you are. This variable is measured by asking the following question to the participants: “Write down below which sports you perform regularly (1x per week or more often) and how often”.
The type of sport performed by the participants is coded as followed: Team Sport is 1 if the participant plays one or more different team sports like soccer, hockey or basketball and 0 otherwise. Individual Competitive Sport is 1 if the participant plays a sport that is performed individually, however it is competitive and 0 otherwise. Examples are tennis and golf. Individual Noncompetitive Sport is 1 is the participant plays an individual, noncompetitive sport like running or fitness and 0 otherwise. Frequency Sport gives the number of times a participants performs sports per week.
Controls
Multiple controls are included in this study. The first is class; this is a dummy variable for the class in which the experiment was performed. The classes were at different hours, which may affect the choices made in the experiment. Moreover, it may affect how confident a
25
participant feels concerning ability. The second represents the nationality of the parents, nationality father and nationality mother. The nationality of the student may affect both the preference for competition and the perceived attractiveness. These variables are both dummy variables, and are 0 when the nationality is Dutch, and 1 when the nationality is different from Dutch. Gender is the last control. This factor influences the preference for competition (Niederle and Vesterlund, 2007) and might also influence perceived attractiveness. It is a dummy variable which is 0 when the participant is male and 1 when the participant is female.
Hypotheses
The literature review evaluated studies which have shown evidence of the existence of the beauty premium, and studies which have investigated the influence of preference for competition on the gender gap in wages. This study investigates whether preference for competition may also influence the beauty premium. It is possible that pretty people have innate characteristics that make them compete more often and win tournaments more often. Therefore the following hypothesis will be tested:
H1: People with above-average attractiveness will enter the tournament more often than average looking people.
Based on evidence of the plainness penalty, it can also be hypothesized that people who are less than average attractive will enter the tournament less often than average looking people. Therefore the following hypothesis will be tested:
H2: People with below-average attractiveness will enter the tournament less often than average looking people.
Moreover, it is possible that below-average attractive people will not enter the tournament, while their performance shows that they should. The same holds for above-average attractive people who should actually opt out of competition. Therefore, the following hypotheses will be tested:
H3: People with below-average attractiveness do not enter the tournament often enough.
26
Results
Summary Statistics
Table 1 shows the summary statistics of the data. Of all participants 37.5 percent chooses tournament payout rate in task 3. The average self-rating of attractiveness is close to average, namely 4.321. The minimum of self-rated attractiveness is 1 and the maximum is 7, so people actually choose the lower end and the top end of the scale. The average number of correct answers increases from task 1 to task 2 to task 3. Furthermore, 38.4 percent of the participants are female. The average age is around 16 years, the minimum is 14 and the maximum is 46. This maximum is because of the teacher. When excluding him, average age becomes 15.8 and the standard deviation gets lowered to 1.097. The other answers of the teacher are not outliers, and therefore he is included in the sample. The rest of table shows that average height is 1.783 meter, average BMI is 20.202, 12.5 percent of students has a mother with foreign nationality and 10.7 percent has a father with foreign nationality. The other summary statistics are listed in table 1 and will be used below.
Table 1. Summary Statistics
Variable Number of
observations Mean Standard Deviation Minimum Maximum
Choice 112 0.375 0.486 0 1 Attractive Self 112 4.321 1.195 1 7 Attractive Others 112 4.170 1.146 1 7 Task 1 112 8.304 3.087 2 17 Task 2 112 9.366 3.171 4 20 Task 3 112 9.866 3.447 4 25 Gender 112 0.384 0.489 0 1 Age 112 16.089 3.054 14 46 Height 112 1.783 0.098 1.50 2.02 BMI 112 20.202 2.371 15.397 31.111 Nationality father 112 0.107 0.310 0 1 Nationality mother 112 0.125 0.332 0 1 Belief 112 2.473 0.870 1 4 Confidence Math 112 4.951 1.381 1 7 Lottery 112 2.866 1.263 1 5 Risk Level 112 5.513 1.900 1 10 Team Sport 112 0.661 0.476 0 1 Individual Competitive Sport 112 0.214 0.412 0 1 Individual Noncompetitive Sport 112 0.446 0.499 0 1 Frequency Sport 112 3.411 1.748 0 8
27 Graph 1. Tournament Choice, conditional on attractiveness
Attractiveness and Preference for Competition
First, we are going to look at basic information on tournament choice to evaluate hypothesis 1 and 2. After having experienced both the piece-rate and the tournament compensation scheme, in task 3, the participants have to choose which scheme they want to apply to this task. Graph 1 shows the percentage of people choosing piece-rate payout and tournament payout before performing task 3, by categories of attractiveness. Out of all below-average looking people, 33 percent chooses tournament and out of all people who rate themselves as above-average attractive 37.5 percent chooses tournament. Out of all average looking people, 40 percent chooses tournament. The difference is only small. A Chi-Squared test can be used to assess whether there is a relation between attractiveness and the choice for competitiveness. This test gives a χ2 of 0.232, and a p-value of 0.890. Thus, we can conclude that the null-hypothesis that there is no relation, cannot be rejected. For details, see appendix B1.
However, since there are many factors that can influence this relationship, it is necessary to perform some regressions. For example, the control variables gender, class and the nationality of parents need to be included to see the effect of attractiveness on choice of competition. When we look at table 2, column 1, we can see that the influence of attractiveness on the choice of compensation scheme is only small and not significant. Here we do not control for any factors. However, as explained in the variables section, it is important to control for certain factors. Therefore, from now on, we control for gender, nationality and class. When we look at table 2, column 2, we can see that the influence of
33% 40% 38% 67% 60% 64% 0% 10% 20% 30% 40% 50% 60% 70% 80%
Below Average Above
Tournament Piece-Rate
28
attractiveness on the choice of compensation scheme is still not significant and small. This can have multiple reasons. It is possible that the relation exists, but that there is too much variance in the data. It can be seen that the standard error of the coefficient is very large, so the 95 percent Confidence Interval is also large. It is also possible that the relation between tournament choice and attractiveness does not exist. However, since the standard error is very large, we can conclude that the estimator is not precise. We cannot confirm nor reject hypothesis 1 and 2. Moreover, multiple factors like performance, confidence and risk aversion can influence the results. Therefore, the next sections will look at this influence.
To conclude, there is no significant relation between attractive and tournament choice found in this sample. Neither hypothesis 1 nor hypothesis 2 can be confirmed. The next sections will look into possible factors that influence this relation.
Performance
To investigate the effect of performance on compensation scheme choice, conditional on attractiveness, we first have to evaluate performance. Looking at graph 2, we can see that performance rises over the three tasks for all categories of attractiveness. Moreover, average attractive people perform better than below-average and above-average attractive people in all tasks. Below-average looking people perform worst in all three tasks. Performance could depend on the version of the problem sheet. However, table 3 shows that, for each task, there is no significant difference between mean performances over the different versions.
Graph 2. Average Performance, conditional on attractiveness and tasks
Average number of solved problems in all three tasks, conditional on attractiveness. 0 2 4 6 8 10 12
Task 1 Task 2 Task 3
Below Average Above
29 Table 2. Determinants of Tournament Entry
Dependent variable: Choice made in compensation scheme in Task 3.
(1) (2) (3) (4) (5) (6) Above -0.020 (0.103) -0.032 (0.104) 0.003 (0.095) 0.022 (0.086) -0.016 (0.084) -0.023 (0.086) Below -0.062 (0.130) -0.046 (0.131) 0.088 (0.121) 0.071 (0.120) 0.116 (0.118) 0.124 (0.122) Task 2 0.079*** (0.015) 0.032* (0.018) 0.035** (0.017) 0.030* (0.018) Task 2 – Task 1 -0.022 (0.019) -0.005 (0.018) -0.007 (0.018) -0.002 (0.018) Belief -0.247*** (0.056) -0.219*** (0.056) -0.222*** (0.057) Confidence Math 0.007 (0.034) 0.011 (0.034) 0.019 (0.036) Lottery 0.027 (0.034) 0.022 (0.035) Risk Level 0.050** (0.023) 0.051** (0.024) Team sport 0.076 (0.112) Individual competitive sport -0.068 (0.111) Individual noncompetitive sport 0.028 (0.094) Frequency sport -0.015 (0.029) Gender -0.035 (0.099) -0.082 (0.092) 0.029 (0.088) 0.030 (0.086) 0.018 (0.090) Hour 4 -1.159 (0.130) -0.102 (0.115) -0.124 (0.107) -0.074 (0.107) -0.090 (0.111) Hour 7 -0.345*** (0.126) -0.312*** (0.114) -0.316*** (0.106) -0.253** (0.106) -0.269** (0.109) Hour 8 -0.106 (0.128) 0.002 (0.117) -0.007 (0.109) 0.041 (0.110) 0.035 (0.121) Nat. Mom -0.183 (0.145) -0.088 (0.130) -0.106 (0.122) -0.083 (0.122) -0.054 (0.129) Nat. Dad -0.074 (0.156) -0.141 (0.140) -0.107 (0.129) -0.070 (0.126) -0.080 (0.130) Constant 0.395*** (0.075) 0.591*** (0.116) -0.204 (0.178) 0.760** (0.325) 0.252 (0.364) 0.284 (0.371) N 112 112 112 112 112 112 R-squared 0.0021 0.0910 0.2989 0.4152 0.4598 0.4686
Dependent variable: Dummy where 1=tournament and 0=piece-rate. Table presents OLS coefficients. All regressions control for class, nationality of parents and gender, except column 1. Standard errors are in parentheses; * p<0.10, ** p<0.05, ***p<0.01 of the underlying coefficient.
30
Table 4 provides us with more information on average performance, conditional on both attractiveness and compensation scheme choice. It gives us three performance measures; the average number of problems solved in task 1 (piece-rate), the average number of problems solved in task 2 (tournament) and the difference between the two. It also gives the p-value for this difference. For both average attractive people and above-average attractive people performance in task 2 (tournament) is significantly higher than in task 1 (piece-rate). People who rate themselves as below-average attractive and who choose piece-rate in task 3 actually perform worse in the tournament than in the piece-rate compensation scheme. Furthermore, we can see that people who rate themselves as below-average attractive and choose tournament do not perform better in task 1, but do perform better in task 2 than below-average attractive people that choose piece-rate payout. Both average and above-average looking people that choose tournament payout perform better in both task 1 and task 2 than those that choose piece-rate payout.
We can also run a regression to see the effect of attractiveness on tournament, conditional on performance. As some people perform better, they are more likely to choose tournament instead of piece-rate payout. As performance may be related to attractiveness, we need to control for performance. These results are given in table 2, column 3. We control for performance in task 2, and the increase in performance between task 2 and task 1. We see that the coefficients of above and below change from negative to positive. However, they still remain highly insignificant. Furthermore, we can see that the influence of the performance in the tournament (task 2) significantly influences the choice to enter the tournament in task 3.
To conclude, average looking people perform better than below- or above-average attractive people. They choose tournament a bit more often. Looking at a regression, the effect of attractiveness on the choice for compensation scheme is not significant.
Table 3. Comparison of different versions
Mean Performance
Task Version 1 Version 2 Version 2 - Version 1 p-value
1 8.54 8.09 -0.45 0.442
2 9.21 9.53 0.32 0.599
31
Table 4. Performance characteristics by choice of compensation scheme
Attractiveness
Choice Performance Piece-rate Performance Tournament Tournament-Piece-rate p-value Below Piece-rate 7.71 (0.59) 7.07 (0.55) -0.64 (0.32) 0.0695
Tournament 8.14 (2.02) 10 (1.35) 1.86 (1.52) 0.2672 Difference 0.43 (1.33)
p-value=0.748 2.93 (1.31) p-value=0.027 2.5 (1.13) p-value=0.029
Average Piece-rate 7.54 (0.50) 8.77 (0.57) 1.23 (0.48) 0.0170 Tournament 10.47 (0.85) 11.94 (0.81) 1.47 (0.58) 0.0213 Difference 2.93 (0.90)
p-value=0.002 3.17 (0.87) p-value=0.001 0.24 (0.78) p-value=0.760
Above Piece-rate 7.2 (0.40) 8.43 (0.43) 1.23 (0.45) 0.0102 Tournament 9.72 (0.71) 10.89 (0.73) 1.17 (0.61) 0.0712 Difference 2.25 (0.87)
p-value=0.004 2.46 (0.86) p-value=0.005 -0.07 (0.75) p-value=0.930
Averages with standard errors in parentheses. Sample is 112 participants. P-values given.
Confidence
From Mobius and Rosenblat (2006) we know that attractive people have more confidence. According to these authors, higher confidence can influence the level of wage. Niederle and Vesterlund (2007) show that higher confidence can explain why men choose the tournament compensation scheme more often than women. Therefore, it is possible that attractive people receive higher wages because they choose tournament more often, due to their higher confidence.
To see the influence of confidence on payout choice, and how this is related to attractiveness, we first look at the correlation between confidence and attractiveness. This study uses two measures of confidence. The first gives the beliefs the participants have on their performance in task 2. They have to state which rank they think they have. In the second, they have to state how confident they are about their mathematical ability. Graph 3 shows which percentage of participants has chosen which rank, conditional on attractiveness. We can see in this graph that people who rate themselves as above-average attractive choose rank 2 significantly more than the other participants (52% is not in the 95% confidence interval of below-average and average attractive people). Moreover, we can see that below-average attractive participants have slightly lower beliefs than average participants. They choose the ranks 1, 2, 3 less often and rank 4 more often. A Chi-Squared test shows with a χ2 of 7.706, and a p-value of 260, we cannot reject the null hypothesis that
32 Graph 3. Beliefs
Shows which percentage of people of specific attractiveness chooses which belief, where 1 shows a lot of confidence and 4 shows low confidence.
Table 5 shows us the means of confidence in math chosen by the participants. Below-average attractive participants choose a confidence level of 3.67 on Below-average. Average and above-average attractive people choose confidence levels of 5.15 and 5.33 on average. The means of above-average attractive people and average attractive people are similar, but the mean of below-average attractive people is significantly lower than that of the other participants (pairwise mean comparison gives a p-value of 0.000). This indicates that the confidence of below-average attractive people is indeed lower than that of average of above-average attractive people.
To study the effect of confidence on competition choice, we run a regression. We add the two variables for confidence to the regression. Looking at table 2, column 4, we see that the coefficients for above and below remain insignificant. However, the belief does enter significantly negative. If the participant beliefs that he or she has a higher rank (so a lower performance), the chance that he or she will choose tournament significantly decreases with 24.7 percentage point. Since on average, 37.5 percent of people choose tournament, this decrease is large. 0% 10% 20% 30% 40% 50% 60% 1 2 3 4 Below Average Above
