1
Overconfidence and the Dunning-Kruger
effect: a field experiment.
Nick Verkerk 05-07-2018
10359494 University of Amsterdam
Master Economics Giorgia Romagnoli
2 Abstract
The goal of this thesis is to further investigate the Dunning-Kruger effect by doing a field experiment. According to Dunning and Kruger (1999) overconfidence is mainly found in the unskilled because the skills needed to perform a specific task are the same skills needed to evaluate your own, or somebody elses skills. This study has three main findings. The first finding of this study is that individuals, on average, are indeed overconfident. Secondly, the field evidence presented indicates that overconfidence is mainly found in the unskilled. Finally, the results of this study show that males are more overconfident than females. Also, the Dunning-Kruger effect holds for both males and females when using the overestimation definition of overconfidence. When using the overplacement definition of overconfidence, the Dunning-Kruger effect only holds for males.
3 Contents Abstract………..2 1. Introduction………4 2. Literature review………...5 2.1 Overconfidence………..5
2.1.1 Three ways of defining overconfidence………5
2.1.2 Hard and easy tasks………6
2.2 The Dunning-Kruger effect……….7
2.2.1 Criqitue on the Dunning-Kruger effect……….9
2.3 Why study overconfidence?...10
2.4 The gender effect……….12
2.5 Conclusion……….12
3. Methodology………..14
3.1 Hypotheses………..14
3.2 Research method……….15
3.2.1 Tennis rating system………15
3.2.2 Measuring overconfidence and the Dunning-Kruger effect……….16
3.2.3 Data analysis………..17
4. Results……….18
4.1 Overconfidence……….18
4.2 The gender effect……….32
5 Discussion and conclusion………..34
6 Appendix……….37
4 1.Introduction
Overconfidence is one of the most influenctial biases in both people’s day to day lives and the economy. According to Dunning and Kruger (1999) overconfidence is mainly found in the unskilled because the skills needed to perform a specific task are the same skills needed to evaluate your own, or somebody elses skills. This effect is called the Dunning-Kruger effect. Although overconfidence and the Dunning-Dunning-Kruger effect have been broadly studied in the lab, there is a need for more field evidence. It is important to study
overconfidence in the field because this creates naturalistic results due to the fact that the subjects are familiar with the task they have to perform. These naturalistic results are more reliable than artificial results. The goal of this study is to further investigate the Dunning-Kruger effect by doing a field experiment in order to answer the following research questions.
Research question 1: Are individuals, on average, overconfident? Research question 2: Is overconfidence mainly found in the unskilled?
Research question 3: Is there a gender-based difference difference in overconfidence and the Dunning-Kruger effect?
The first finding of this study is that individuals, on average, are overconfident. Secondly, field evidence indicates that overconfidence is mainly found in the unskilled. Finally, the results of this study show that males are more overconfident than females. Also, the Dunning-Kruger effect holds for both males and females when using the overestimation definition of overconfidence. When using the overplacement definition of overconfidence, the Dunning-Kruger effect hold for males only.
Relevant literature will be reviewed in chapter two, starting with explaining the three ways of defining overconfidence. Furthermore, the Dunning-Kruger effect and the
importance of studying overconfidence will be discussed. The findings from the reviewed literature are the base of the research questions and the hypotheses, which are posed in chapter three. Furthermore, the research method will be discussed. In chapter four, the results will be presented with both graphs and tables. In chapter five, these results will be discussed in detail and compared with the current literature. To conclude, important implications for further study will be discussed.
5 2 Literature review
2.1 Overconfidence
‘Over their lifetime, people base thousands of decisions on impressions of their skill, knowledge, expertise, talent, personality, and moral character’(Dunning et al., 2004). These decisions may be based on wrong impressions of our skills or knowledge, resulting in poor decisions. Overconfidence is a behavioural bias that makes people overestimate and misjudge their skills and knowledge. Plous (1993) highlighted the importance of studying overconfidence by stating that ‘no problem in judgment and decision making is more prevalent and more potentially catastrophic than overconfidence’ (Plous, 1993, p. 217). Overconfidence has been studied in three different ways: overestimation, overplacement, and overprecision (Moore & Healy, 2008, p.3). In the following section, these three different ways will be explained.
2.1.1 Three ways of defining overconfidence
Moore and Healy (2008, p.3) describe the first definition of overconfidence as ‘the overestimation of one’s actual ability, performance, level of control, or chance of success’. Overestimation is the most common way of examining overconfidence and is used in roughly 64% of empirical papers on overconfidence (Moore and Healy, 2008, p.4). One example is the study of Kennedy, Lawton and Plumlee (2002), in which students were asked to estimate the grade they expected for their exam immediately following its completion. The results indicated that students, especially the least performing ones, significantly overestimated their performance. The opposite of overestimation is underestimation. Underestimation is most likely to occur on easy tasks or when an individual is highly skilled (Moore & Healy, 2008, p.8).
The second way of defining overconfidence is overplacement, which occurs when the majority of people rate themselves better than the median (Moore & Healy, 2008, p.4). Overplacement is equivalent to the better-than-average effect: ‘The tendency to evaluate oneself more favorably than an average peer.’ (Alicke & Govorun, 2005, p.85). Originally, the theoretical explanation for the so called comparative bias was that people rate themselves higher than others because it makes them feel good about themselves to believe that they are above average (Brown, 2012, p.209). Since then, alternative explanations have been
6 given. For instance, it is believed that informational differences produce a
better-than-average effect. Informational differences are the tendency to know more about oneself than others (Chambers & Windschitl, 2004, p.813). Other alternative explanations are focalism, which refers to the tendency to focus more on oneself than others when making
comparative judgments (Krizan & Suls, 2008, p.931), naïve realism, which is explained as the believe that others have the same perspective and perceptions on the world as oneself (Reed, Turiel and Brown, 2013) and egocentrism, which is the tendency to place undue weight on one’s own perceptions (Alicke & Govorun, 2005, p.91). The most famous example of a study on overplacement is the study conducted by Svenson (1981), in which 93% of a sample of American drivers and 69% of a sample of Swedish drivers believed they were better drivers than the median driver in their own country. The opposite of overplacement is underplacement. On hard tasks, people tend to rate themselves below average and
therefore underplace themselves (Moore & Healy, 2008, p.10).
The third definition of overconfidence is called overprecision, which is explained as the ‘excessive certainty regarding the accuracy of one’s beliefs’ (Moore & Healy, 2008, p.4). Overprecision is the most robust form of overconfidence, because in contrast to
overestimation and overplacement, it is very rare to find evidence of the opposite of
overprecision, called underprecision (Moore, Tenney and Haran, 2015, p.5). This is because it is rare for people to believe that they are less sure than they actually are. The most common way of examining overprecision is to ask participants questions with numerical answers, for instance: ‘How high is the Eiffel Tower?’ and let the subjects come up with a 90% confidence interval around the right answer. Generally, these confidence intervals are too narrow, indicating that people are too sure they knew the correct answer (Moore & Healy, 2008, p.6).
2.1.2 Hard and easy tasks
Typically, overestimation and overplacement are positively correlated. Individuals who overestimate themselves are also likely to overplace themselves. This is not always the case when tasks vary in difficulty. On average, overestimation occurs for hard tasks and underestimation occurs for easy tasks. When a subject performs well, is it hard for them to overestimate their performance. Similarly, when a subject performs less, it is hard for them to underestimate their performance (Moore, Small, 2007, p.974). For overplacement the
7 opposite is true. If subjects misjudge their own performance, they are likely to misjudge the performance of others even more (Moore, 2007, p.47). On a hard task, subjects
overestimate their own scores, but overestimate the scores of others even more. They therefore believe that they performed worse than others. Similarly, on easy tasks subjects tend to underestimate their own scores, but underestimate the scores of others even more. Therefore, they believe that they performed better than others (Prims, Data, 2017). This phenomenom, where overestimation is paired with underplacement and underestimation is paired with overplacement, is called the hard-easy effect.
For some tasks it is hard to argue that an individual is actually overconfident. It could be that people have a different opinion on what is ‘good’ or ‘bad’. As mentioned before, in the famous study conducted by Svenson (1981), 93% of a sample of American drivers and 69% of a sample of Swedish drivers believed they were better drivers than the median driver in their own country. Roy and Liersch (2013) argue that people have different opinions on what is a ‘good driver’. For some people being a good driver may mean driving below the speed limit and for some people it may mean driving as fast as possible. Because for some tasks there is no universal definition of ‘being good’, individuals construct their own perception of the skills needed to be defined as ‘above average’. Therefore, Roy & Liersch argue that individuals tend to be more overconfident in tasks in which the definition of ‘being good’ is uncertain.
2.2 The Dunning-Kruger effect
The article of Dunning and Kruger (1999) starts with a story of McArthur Wheeler who tried to rob a bank in broad daylight. When he was arrested he was surprised that the surveilance cameras video-taped him because he believed that rubbing your face with lemon juice would make you invisible for video cameras. Dunning and Kruger state that ‘when people are incompetent in the strategies they adopt to achieve succes and satisfaction, they suffer a dual burden: not only do they reach erroneous conclusions and make unfortunate choices, but their incompetence robs them of the the ability to realize it.’(Dunning & Kruger, 1993, p.1121). In other words, Dunning and Kruger believe that poor performers think that they are doing fine, while their actual perfomance is insufficient (Dunning, 2011, p.260). This dual burdan arises because the skills needed to perform a meta-cognitive task (e.g.
8 predicting if you answered the trivia question correctly). Thus, if individuals perform poorly on the meta-cognitive task, they also lack the skill to judge their performances. Therefore, they would not know that their responses were incorrect (Dunning, 2011, p.261).
It is not only the incompenent who misjudge their performance. The most competent or best performers tend to underestimate their performance. In contrast to the
incompetent, where the source of the misestimation is the lack of skill to perform both the cognitive and meta-cognitive task well, the competent misjudge the performance of the other subjects (Dunning, 2011, p.271). For the top performers, giving the correct answers is relatively easy. As a result, they think that for others it is easy aswell because they see their own behavioral choices as common and alternative responses as uncommon. This source of misestimation is called the false consensus effect (Ross, Greene & House, 1977, p.280). Thus, the most competent overestimate others by assuming that they possess the same skills or knowledge.
Individuals tend to choose what they think is the most optimal option. If an individual lacks the skills to recognize this option, the decision made is not optimal and therefore the individual performs poorly. Nevertheless, the individual assumes that he performed well. Therefore, as the incompetent individuals tend to be overconfident about their skills and abilities, overconfidence is mainly found in the unskilled (Dunning & Kruger, 1993, p.1122). In order to examine whether the incompetents are indeed less able to correctly evaluate their performance than the competents do, Dunning and Kruger performed four studies. In each of the four studies, the subjects were presented with tests that assessed their ability or knowledge. After each test, the subjects were asked to predict how well they performed. In all four studies, the subjects in general overestimated their performance. The subjects in the bottom quartile not only grossly overestimated their performance, but they also tended to believe that they were above average. Subjects in the other quartiles did not overestimate their performances to the same degree. The subjects that were in the top quartile
underestimated their performance. These results indicate that overconfidence is indeed mainly found in the incompetent or unskilled.
Subsequent studies on the Dunning-Kruger effect have found field evidence of the Dunning-Kruger effect in different subject areas. For instance, the Dunning-Kruger effect was
9 found for university students’ logical reasoning ability and salesman’s ability to sell (Hubka, 2015).
2.2.1 Critique on the Dunning-Kruger effect
The most common critique on the Dunning-Kruger phenomenon is the so-called regression effect. The incompetent subjects performed close to the bottom of the
distribution, making it hard for those subjects to underestimate their performance. Likewise, the most competent subjects perform close to the top of the distribution, making it hard for them to overestimate their performance (Schlösser, Dunning, Johnson & Kruger, 2013, p.87). Krueger and Mueller (2002) state that the results found in the article of Dunning and Kruger (1999) can be explained by this regression effect.
Burson, Larick and Klayman (2006) accepted the presence of regression effects and state that the perceived performance of subjects is dependent on how they experience the difficulty of the task. When a task is perceived as easy, most subjects will think they
performed well. As a result, low performers will overestimate their performance and
therefore produce the standard Dunning-Kruger effect. In contrast, when a task is perceived as hard, most subjects will think they did not perform well. As a result, low performers will accurately rate their performance low and high performers will underestimate their performance. This phenomenon is the opposite of the standard Dunning-Kruger effect because the high performers are misestimating their performances (Dunning, 2011, p.266). As a response to this alternative explanation, Dunning (2011) states that two main issues prevent this explanation to be more accurate than the Dunning-Kruger effect. The first one is that Burson et al. (2006) only used the performance on trivia questions, whereas Dunning and Kruger used four different tasks. One important condition of the Dunning-Kruger effect is that participants must believe that their answers are reasonable. The problem with trivia questions is that participants may not have enough intellectual skills to believe that their answers are correct. As a result, people tend to rate themselves as below average (Dunning, 2011, p.268).
The second issue that Dunning (2011) points out is that Burson et al. stated that it is difficult for subjects to rate how well they did compared to the performance of the other subjects without really knowing how well those others were doing. As a result, the estimates
10 of the subjects were biased by perceptions of overall task difficulty. According to this bias, the subjects should show more signs of the Dunning-Kruger effect when they rate
themselves along objective scales than when they rate themselves on a comparative scale. Dunning opposes this theory by arguing that the Dunning-Kruger effect arises on both objective and comparative scales as low performers overestimate their performance, even without social comparison.
2.3 Why study overconfidence?
It is important to study overconfidence because it can have several economic consequences and influences (economic) decision making. For instance, Odean (1999) argues that due to overconfidence, investors trade too much. Also, Odean states that investors who trade most frequently are often the worst performers. This is because overconfident investors trade even when expected gains of a trade are lower than the expected trading costs. According to Bailey, Kumar and Ng (2011), the reason that investors trade more often is because they overestimate their knowledge and abilities. When an investor has a few succesful periods, he begins to believe that he can predict the future better than he actually can (Barberis & Thaler, 2003, p.1066). Furthermore, self-attribution is a reason why investors get overconfident. Self-attribution occurs when investors attribute succes to their own abilities and failures to bad luck (Daniel, Hirshleifer and Subrahmanyam, 1998, p.1841). This overconfidence can have negative consequences. For instance,
overconfident investors tend to rely more on private information and ignore the importance of public information (Chuang & Lee, 2006, p.2490). In addition, Chuang and Lee (2006) argue that overconfident investors underestimate risk, which leads to excessive volatility in their trades. Odean (1998) states that overconfident investors choose to hold undiversified portfolios because the investors falsely believe that this stock will lead to higher profits.
Another consequence of overconfidence is business failures. A high proportion of businesses fail and Camerer and Lovallo (1999) have tried to find a relationship between overconfidence and business failures. They argue that most business failures are due to the optimism of managers. These managers choose to enter the market and overestimate their chance of success. In Camerer’s and Lovallo’s study, this phenomenom is studied by using an entry game in which the success of the business depends on the individual’s skills and the skills of other participants. It is found that most participants are indeed overconfident and
11 enter the market too often, resulting in monetary losses. This shows that market failures can indeed be a result of overconfidence. In contrast, Landier and Thesmar (2008) found that overconfidence can also have positive effects on the business because overconfident
managers work harder to find costumers and develop new technologies in the early stage of their business.
Another important effect of overconfidence is the effect of CEO overconfidence on the value of a firm. According to the literature, CEO overconfidence can have both positive and negative effect on the firm value. Fairchild (2005) states that overconfident CEO’s
overestimate the probability of good states and underestimate the probability of bad states. Fairchild also adds that an overconfident CEO perceives debt as more undervalued than equity, resulting in a higher level of debt than a rational manager would have done. This overconfidence leads to higher debt values and also higher costs of distress, which can be harmful (Fairchild, 2005, p.5). Furthermore, Malmendier and Tate (2005) argue that overconfident CEOs are more sensitive to cash flows. As a result, they underinvest with insufficient internal funds and overinvest with sufficient internal funds. Malmendier and Tate also state that overconfident CEOs have a higher chance of engaging in value destroying mergers than rational CEOs. In contrast, Gervais, Heaton and Odean (2003) believe that overconfidence can have a positive effect on the firm value since an overconfident CEO tends to take on more projects. This higher amount of projects results in the avoidance of underinvestment. They conclude that a high confidence level is optimal for maximizing firm value. Hirshleifer, Low and Teoh (2012) agree with Gervais et al. but also add that
overconfident CEOs tend to be more innovative and better identify growth opportunities. Finally, the engineering sector has accepted the importance of studying
overconfidence. When engineers calculate how much concrete is required to build a bridge or a house, or how much material is needed to build a wing of an airplane, any
miscalculation can have serious consequences. The engineers are required to multiply their calculation of material needed by a factor between three and eight. The engineering sector has thus developed a method to protect the public against engineers that overestimate their own abilities and therefore miscalculate important factors (Heath, Larrick & Klayman, 1998, p.4).
12 2.4 The gender effect
It is important to study gender-based differences in overconfidence because of the potential reason for lower women’s wages than men’s wages (Johansson Stenman & Nordblom, 2010, p.2). In addition, overconfidence can be an explanation for why only 2.5 percent of of the top five executives for a large group of U.S. firms are women (Bertrand, & Hallock, 2001; Johansson Stenman & Nordblom, 2010, p.2). According to the current
literature, both man and woman are overconfident but man tend to be more overconfident than woman (Soll & Klayman, 2004; Barber, Odean, 2001). In most of the studies on the gender-based differences in overconfidence the overconfidence is measured by individual survey responses. The empirical results on this topic are ambiguous (Johansson Stenman & Nordblom, 2010, p.3). For instance, Barber and Odean (2001) conducted an experiment to investigate the gender difference in overconfidence on the stock market. As Odean (1999) already stated, overconfident investors trade too much. Furthermore, Barber and Odean (2001) state that men trade significantly more than women because men are more overconfident. However, Johansson, Stenman and Nordblom (2010) conducted a natural field experiment on exam behavior in order to examine the gender-based difference in overconfidence. Their results show no significant difference in overconfidence between the male and female subjects. According to Lenney (1977) the reason for amibiguous results is the lack of clear feedback. When feedback is clear and directly available, women are not less confident than man. However, when feedback is not clear, woman seem to be less confident than man. An explanation for the gender-difference in overconfidence is given by Niederle and Vesterlund (2007). They state that man are more competitive than females. This results in men wanting to outperform more than females. In the article by Dunning and Kruger (1999) gender receives no mention because they failed to qualify any results. To the best of my knowledge, there is no other literature available on gender-differences in the Dunning-Kruger effect.
2.5 Conclusion
According to the current literature there are three different measures of
overconfidence, namely overestimation, overplacement and overprecision. Overestimation is the most common way of examining overconfidence and refers to individuals rating
13 second way of examining overconfidence and refers to people rating themselves better than the median (Moore & Healy, 2008, p.4). The third way of measuring overconfidence is overprecision and is explained as the ‘excessive certainty regarding the accuracy of one’s beliefs’ (Moore & Healy, 2008, p.4).
According to the Dunning-Kruger effect overconfidence is mainly found in the unskilled because the skills needed in a specific domain are often the same skills needed to evaluate competence in that same domain, one’s own and someone elses (Dunning & Kruger, 1993, p.1121). There has been critique on the Dunning-Kruger effect. The most common critique on the Dunning-Kruger phenomenon is the so-called regression effect. The explanation behind this effect is that no two variables are ever perfectly correlated, the perception and the reality of the performance included. They are both measured with some level of error that degrades the correlation between them (Schlösser, Dunning, Johnson & Kruger, 2013, p.87).
It is important to study overconfidence because it can have several economic
consequences and influences (economic) decision making. A few examples are overconfident traders who trade too much and therefore perform worse (Odean, 1999), overconfident managers who make wrong investment decisions leading to more business failures (Camerer and Lovallo , 1999) and CEO overconfidence influencing the value of a firm, both positively and negatively (Fairchild, 2005; Gervais, Heaton and Odean, 2003) Furthermore, it is
important to study gender-based differences in overconfidence because it can be a potential reason why, on average, women’s wages are significantly lower than man’s wages
(Johansson Stenman & Nordblom, 2010, p.2). It is also interesting to study the gender-based difference in the Dunning-Kruger effect because there is no existing literature on this topic.
Several studies have done field experiments investigating the Dunning-Kruger effect, but there is still a need for more field evidence. Furthermore, to the best of my knowledge, there is no existing literature on the gender-based difference in the Dunning-Kruger effect. Also, the most common critique on the Dunning-Kruger effect is the regression effect (Schlösser, Dunning, Johnson & Kruger, 2013). For this study, the regression effect is no problem. Both subjects in the bottom of the distribution and in the top of the distribution can be overconfident or underconfident. Therefore, this study is new and innovative.
14 3 Methodology
3.1 Hypotheses
The main goal of this thesis is to find field evidence to further investigate the Dunning-Kruger effect. The field experiment consists of a questionnaire on the perceived tennis abilities of individuals. These tennis abilities will be compared with their actual tennis abilities. The data will be used to answer the following research questions.
Research question 1: Are individuals, on average, overconfident? Research question 2: Is overconfidence mainly found in the unskilled?
Research question 3: Is there a gender-based difference difference in overconfidence and the Dunning-Kruger effect?
The above literature has shown that individuals are, on average, overconfident. Furthermore, the Dunning-Kruger effect tells us that as the skills needed to perform a meta-cognitive task) are the same skills needed to perform a meta-cognitive task, overconfidence is mainly found in the unskilled. Based on the above literature, the following hypothesis are examined:
Hypothesis 1: The participants, on average, will overestimate their rating that corresponds to their perceived tennis ability.
Hypothesis 2: Lower skilled participants will overestimate their rating that corresponds to their perceived tennis ability more compared to higher skilled participants.
Hypothesis 3: Lowest skilled participants, those in the lowest quartile on tennis ratings, will overstate their perceived rating compared to their actual rating, while the highest skilled participants will understate their perceived rating.
Because it is also important to study gender-based differences in overconfidence, the following hypothesis is examined:
Hypothesis 4: Both male and female participants will overestimate their tennis ability, but male participants tend to be more overconfident than female participants.
Because there is no existing literature on the gender-based difference in the Dunning-Kruger effect, it is also interesting to look at this difference.
15 3.2 Research method
First of all, I collected the single and double tennis ratings of 134 members of A.S.T.V Chip & Charge, of which 75 are male and 59 are female. This data was retrieved from the
https://mijnknltb.toernooi.nl/ website. Secondly, the subjects had to give themselves a rating (both a single and a double rating) that, according to their knowledge, most accurately reflects their tennis abilities. They were also asked to tell how many players of the same gender and within A.S.T.V Chip & Charge are, according to their best knowledge, better than them. This data has been collected with a general ‘Qualtrics’ survey. I have reached out to all 134 subjects and asked them to fill in this survey. The importance of using incentives when measuring measuring overconfidence is shown by Cesarini, Sandewall and Johannesson (2006). Therefore a deadline (31-08-2018) has been set on which the participants can reach the rating that, according to them, most accurately reflects their tennis abilities. Among all the participants that correctly predicted their rating, one of them wins a prize of 15 euros. Figure 1 (see appendix) shows the set of questions that was presented to the participants filling in the Qualtrics survey.
3.2.1 Tennis rating system
The Dutch tennis rating system is called the DSS (KNLTB, 2017, p.1). The rating of a tennis player is always between 0 and 10, in which a lower rating corresponds with a better tennis player. The rating of a tennis player changes throughout the year by playing matches and getting match results. Every player starts the year with 6 ‘basic match results’. The rating of a player is calculated by the weighted average of those 6 ‘basic match results’. A
beginning tennis player that has never played a match before starts with a rating of 8,999. The rating of this player is the weighted average of his ‘basic match results’, all 8,999, and therefore this player has a rating of 8,999. When for instance, this player plays his first match against a player with a 8,500 rating, his rating depends on him winning or losing that match. When a player wins a match, the player receives the rating of his opponent minus one. When a player loses a match, the player receives the rating of his opponent plus one. The new match result replaces one of the 6 ‘basic match results’ and the new weighted average of those 6 match results is the new rating of the player. So in this example: the player wins his first match against a player with a rating of 8,500. The player therefore receives a match result of 7,500. This 7,500 replaces one of the 6 ‘basic match results’, which
16 are all 8,999. The new rating of this player will therefore become (7,500 + 5* 8,999)/6= 8.749. After six matches, all ‘basic match results’ will be replaced by actual match results. After that, every additional match result will not replace ‘basic match results’ anymore, but will be added to the six actual match results. For instance, when a players has played 14 matches in total throughout the year, the rating of this player is the weighted average of those 14 match results. At the end of the year, the new rating of a player will become the ‘basis match results’ for the upcoming year. So if the player ends the year with a rating of 8,345, the ‘basic match results’ of the player for the upcoming year will be six times the 8,345 and the procedure starts all over again. In tennis, every player can play both single matches (solo) and double matches (with a partner) and has a specific rating for both type of matches. A double matched played therefore has no influence on the single rating and vice versa. In this thesis I ask the subjects to give themselves both a single and a double rating which according to them, most accurately reflects their tennis abilities. In the data analysis I calculate the average of these two and compare that perceived ‘average rating’ with the actual average rating that I collected.
3.2.2 Measuring overconfidence and the Dunning-Kruger effect
This thesis will focus on two ways of defining overconfidence, overestimation and overplacement. Benoît and Dubra (2011) argue that evidence of overconfidence, and
especially overplacement, are misleading because it reveals no true overconfidence. Instead, it reveals only an apparent one. Later studies showed that overconfidence is robust to the Benoît and Dubra critic, and this study will use a standard measure of overconfidence which does not take the critic into account. Figure 1 (see appendix) shows the set of questions that was presented to the participants filling in the Qualtrics survey. The first two questions are used to find evidence for the first overconfidence definition, called overestimation.
Overestimation refers to individuals rating themselves better than they actually are. In these two questions, subjects were asked to give themselves a single and a double rating that according to them, most accurately reflects their tennis abilities. The average of those two is the perceived ‘average rating’. The overconfidence is measured by comparing the average rating that the subjects gave themselves to their actual average rating. For instance, when a subject has an ‘average rating’ of 5,000 but the subject rated himself as 4,500, the subject believes that his rating should be 0,500 lower and is therefore overconfident.
17 The third question is used to let the subjects rank themselves relative to others. This measure of overconfidence is called overplacement. The subjects are asked to answer the question: ‘What percentage of players (same gender and within Chip & Charge) do you think is better than you?’. By answering this question, the subjects place themselves in a specific percentile. When a subject for instance thinks that 40% of the players are better than he is, while actually 60% of the players are better, the subject is overconfident. In tennis, male players only play against other male players and the same goes for females. Therefore the subjects compare their skills only with other subjects with the same gender.
According to the Dunning-Kruger effect, overconfidence is mainly found in the unskilled. In order for this effect to be true the data should provide evidence that subjects with a higher tennis rating, the incompetent, overestimate their rating more than subjects with a lower tennis rating, the competent. The Dunning-Kruger effect will be investigated by comparing the overconfidence of the subjects with their actual tennis rating. Furthermore, the gender-based difference in overconfidence and the Dunning-Kruger effect will be investigated by comparing the overconfidence between the two different genders. 3.2.3 Data analysis
First of all, the overestimation and overplacement of both the male and female subjects will be investigated by using two-sided t-tests. This will be done for the single rating, the double rating and the average rating. Secondly, both the male and female subjects will be separated into four different quartiles. By doing this, I will investigate whether subjects in the bottom quartile, the incompetent, overestimate and overplace their tennis abilities more than the subjects in the top quartile do. Furthermore, a similar graphical analysis will be done as in the article of Dunning and Kruger (1999). To investigate whether subjects with a high rating, the incompetent, overestimate and overplace their tennis abilities more than subjects with a low rating, I will do a regression analysis. This will be done for both the male subjects and female subjects separately. To investigate the gender-based difference in overconfidence, one subject pool will be made. Because the subjects can not compare their abilities with subjects of another gender, only the difference in overestimation will be
18 4 Results
4.1 Overconfidence
The average single rating of the male subjects is 6,450 and the average double rating of the male subjects is 6,199. The average of these two is the average rating of the male subjects, which is 6,324. The rated single rating of the male subject is 5,911 and the self-rated double rating of the male subjects is 5,704. The average of these two is the self-self-rated average rating, which is 5,811. The actual rating of the male subjects is higher than the self-rated rating. This means that the male subjects, on average, overestimated their single rating by 0,539 points, which is significantly different from zero (t= -7,82, p < 0.0001). Furthermore, the male subjects overestimated their double rating by 0,495 points, which is also significantly different from zero (t= -7,92, p < 0.0001). The male subjects overestimated their average rating by 0,513 points, which is also significantly different from zero (t= -8,92, p < 0.0001). The male subjects believed that 36,01% of the members with the same gender are better than they are. The male subjects thus, on average, put their tennis abilities in the 64th percentile, which exceeded the actual mean percentile, which is 50 by definition, by 14 percentile points. This difference is significantly different from zero (t= -9,77, p < 0.0001). The average single rating of the female subjects is 7,294 and the average double rating of the female subjects is 6,950. The average of these two is the average rating of the female subjects, which is 7,122. The self-rated single rating of the female subject is 6,750 and the self-rated double rating of the female subjects is 6,557. The average of these two is the self-rated average rating, which is 6,653 in this case. Similar to the findings of the male subjects, the actual rating is higher than the self-rated rating. This means that the female subjects, similarly to the male subjects, overestimated their single rating by 0,544 points, which is significantly different from zero (t= -7,08, p < 0.0001). Furthermore, the female subjects overestimated their double rating by 0,393 points, which is also significantly different from zero (t= -6,63, p < 0.0001). The female subjects overestimated their average rating by 0,469 points, which is also significantly different from zero (t= -7,85, p < 0.0001). The female subjects believed that 50,60% of the members with the same gender are better than they are. The female subjects thus, on average, put their tennis ability in the 49th percentile, which barely exceeded the actual mean percentile, which is 50 by definition. As
19 expected, this difference is not significantly different from zero (t= 0,57, p =0,72). Table 1 summarizes the descriptive statistics.
Table 1: Descriptive statistics for all subjects
Actual Self-rated Difference
Male Single rating 6,450 5,911 0,539** Double rating 6,199 5,704 0,495** Average rating 6,324 5,811 0,513** Percentile 50% 64% 14%** Female Single rating 7,294 6,750 0,544** Double rating 6,950 6,557 0,393** Average rating 7,122 6,653 0,469** Percentile 50% 49% 1% *Significant at 5% level ** Significant at 1% level
The subjects whose average rating fell in the bottom quartile, the incompetent, grossly overestimated their perceived tennis ratings. The average single rating of the male subjects in the bottom quartile is 8,957 and the average single rating of the female subjects in the bottom quartile is 9,313. The self-rated single rating of the male subjects in the
bottom quartile is 8,156 and the self-rated single rating of the female subjects in the bottom quartile is 8,687. The differences are both significantly different from zero (t=-5,56, p < 0.0001 for male, t= -4,72, p=0,0016 for female). The average double rating of the male subjects in the bottom quartile is 8,847 and the average double rating of the female subjects in the bottom quartile is 9,153. The self-rated double rating of the male subjects is 8,044 and the self-rated double rating of the female subjects is 8,533. The differences are both
significantly different from zero (t=-6,21, p < 0.0001 for male, t= -6,63, p < 0,0001 for female). The average rating of the male subjects in the bottom quartile is 8,902 and the
20 average rating of the female subjects in the bottom quartile is 9,233. The self-rated average rating of the male subjects is 8,117 and the self-rated average rating of the female subjects is 8,610. The differences are both significantly different from zero (t=-5,53, p < 0.0001 for male, t= -6,00, p < 0,0001 for female). The male subjects in the bottom quartile, on average, put their tennis abilities at the 23rd percentile, whereas female subjects in the bottom quartile put their tennis abilities at the 14th percentile. The subjects in the third and second quartile overestimate their tennis abilities to a similar degree as the subjects in the bottom quartile, as can be seen in Table 2. The average single rating of the male subjects in the top quartile is 3,927 and the average single rating of the female subjects in the top quartile is 4,582. The self-rated single rating of the male subjects is 3,889 and the self-rated single rating of the female subjects is 4,400. They are both not significantly different from zero (t=0,50, p=0.69 for male, t= -1,62, p=0,063 for female). The average double rating of the male subjects in the top quartile is 3,786 and the average double rating of the female subjects in the top quartile is 4,283. The self-rated double rating of the male subjects is 3,663 and the self-rated double rating of the female subjects is 4,081. They are both not significantly different from zero (t=-0,61, p=0.27 for male, t= -1,44, p=0,086 for female). The average rating of the male subjects in the top quartile is 3,856 and the average rating of the female subjects in the top quartile is 4,432. The self-rated average rating of the male
subjects is 3,776 and the self-rated average rating of the female subjects is 4,241. Only for female this is significantly different from zero (t=-0,49, p=0.32 for male, t= -2,69, p=0,009 for female). The male subjects in the top quartile, on average, put their tennis abilities at the 94th percentile, whereas female subjects in the top quartile put their tennis abilities at the 90th percentile. See table 2 for the overestimation and overplacement of both male and female subjects per quartile.
21 Table 2: Ratings per quartile
Rating Quartile 1 Quartile 2 Quartile 3 Quartile 4
Male
Actual single rating 3,927 5,565 7,431 8,957 Self-Rated single rating 3,889 5,058 6,620 8,156
Difference single rating 0,038 0,507** 0,811** 0,801** Actual double rating 3,786 5,371 6,893 8,847 Self-rated double rating 3,663 4,874 6,325 8,044 Difference double rating 0,123 0,497** 0,568** 0,803** Actual average rating 3,856 5,468 7,162 8,902 Self-rated average rating 3,776 4,966 6,473 8,117 Difference average rating 0,08 0,502** 0,689** 0,785** Actual percentile 87,5% 62,5% 37,5% 12,5% Self-rated percentile 94% 82% 55% 23% Difference percentile 6,5%** 19,5%** 17,5%** 10,5%**
Female
Actual single rating 4,582 6,883 8,673 9,313 Self-Rated single rating 4,400 6,507 7,621 8,687 Difference single rating 0,182 0,376** 1,052** 0,626** Actual double rating 4,283 6,504 8,115 9,153 Self-rated double rating 4,081 6,287 7,557 8,533 Difference double rating 0,202 0,217* 0,558** 0,620** Actual average rating 4,432 6,693 8,394 9,233 Self-rated average rating 4,241 6,397 7,589 8,610 Difference average rating 0,191** 0,296** 0,805** 0,623**
22 Actual percentile 87,5% 62,5% 37,5% 12,5% Self-rated percentile 90% 55% 35% 14% Difference percentile 2,5% -7,5% -2,5% 2,5% *Significant at 5% level **Significant at 1% level
From this moment, I will focus on the actual average rating, the perceived average rating and the actual and perceived percentiles. In figure 2 the overplacement and overplacement per quartile for the male subjects are shown. The left part of the figure shows that in the top quartile the actual rating and self-rated rating are quite similar. This means that the male subjects in the top quartile do not overestimate their tennis abilities. In the second quartile we see that the actual rating is higher than the self-rated rating. This means that the subjects in the second quartile overestimate their tennis abilities. In the third and bottom quartile we see a rise in overestimation by the male subjects. We see that the higher the actual rating is, the more the subjects overestimate their tennis abilities. In the right part of figure 2 we see the overplacement per quartile for the male subjects. In all four quartiles we see that the perceived ability is higher than the actual ability. This means that in all four quartiles, the male subjects overplace their tennis abilities when comparing themselves with other male subjects. The overplacement is the lowest for male subjects in the top quartile and the overplacement is the highest for male subjects in the second quartile.
Figure 1 Overestimation and overplacement per quartile, male
0 2 4 6 8 10 1 2 3 4 R A TIN G ABILITY LEVEL O V E R E S T I M A T I O N P E R Q U A R T I L E , M A L E Actual rating Self-rated rating 0 20 40 60 80 100 1 2 3 4
OVERPLACEMENT PER QUARTILE, MALE
23 Figure 3 shows the overplacement per quartile for the female subjects.The left part of the figure indicates that in all quartiles, the actual rating is higher than the self-rated rating. This means that in all quartiles, the female subjects overestimate their tennis abilities. The overestimation of female subjects is the highest in the third quartile. The right part of figure 2 shows the overplacement per quartile for the female subjects. We see that in the top quartile and bottom quartile, the perceived ability is higher than the actual ability. This means that the female subjects overplace their tennis abilities when comparing themselves to other female subjects. However, In the second and third quartile the actual ability is higher than the perceived ability. This means that the female subjects in the second and third quartile actually underplace their tennis abilities when comparing themselves to other female subjects.
To examine whether subjects with high ratings, the incompetent, overestimate their tennis abilities more than subjects with low ratings, a plot has been made for both the male and female subjects. Figure 4 is the plot of the overestimation of male subjects. On the x-axis the average rating is shown and on the y-axis the difference between the self-rated rating and actual rating is shown. A positive difference means that the subject is overconfident and a negative difference means that the subject is underconfident. We can see a positive relationship between the average rating and the difference. This means that as the rating of the subjects get higher, the more overconfident they are. This is in line with the Dunning-Kruger effect.
Figure 2 Overestimation and overplacement per quartile, female
0 2 4 6 8 10 1 2 3 4 R A TIN G O V E R E S T I M A T I O N P E R Q U A R T I L E , F E M A L E Actual rating Self-rated rating 0 20 40 60 80 100 1 2 3 4
OVERPLACEMENT PER QUARTILE, FEMALE
24
Figure 3 Plot of male overestimation
In figure 4 we see that male subjects with a high rating, the incompetent, are more overconfident than the male subjects with a lower rating, the competent. To see whether this relationship is significant, a regression has been done. The regression results are shown in table 3. Control variables Age, Yrsofplaying and Academiclevel have been added and Diffrating is the dependent variable. The coefficient of Average rating is 0,193 and is
significant at a 1% level. This means that when the rating of the subject goes up by one, the Diffrating goes up by 0,193. We can therefore conclude that male subjects with a high rating, the incompetent, overestimate their abilties more than the male subjects with a low rating, the competent. For male subjects the Dunning-Kruger effect thus holds. Furthermore, Yrsofplaying has a coefficient of 0,062 and is significant at a 5% level. This means that male subjects that have been playing tennis for more years, are more overconfident than male subjects that have been playing tennis for less years. An explanation for this may be that subjects that have been playing for many years, feel like they should be on a higher level than they actually are. They might think that they have a lot of experience and therefore overestimate their tennis abilities.
25 Table 3: Regression results
male overestimation Diffrating
Average rating 0.193*** (0.037) Age -0.035 (0.023) Yrsofplaying 0.062** (0.025) Academiclevel -0.044 (0.067) Constant -0.044 (0.067) Observations 76 R-squared 0.34
Robust t statistics in parentheses
* significant at 10%; ** significant at 5% ; *** significant at 1%
26
Figure 4 Plot of male overplacement
Figure 5 shows the overplacement of the male subjects. On the x-axis the average rating is shown and on the y-axis the difference between the self-rated and actual
percentiles is shown. A positive difference in percentile means that a subject overplaces his tennis abilities when comparing himself to the other male subjects. Figure 5 shows a positive relationship between the average rating and the difference in percentile. To see whether this relationship is significant, a regression analysis has been done. The regression results are shown in table 4. The same control variables are used as in table 3, but now Diffpercentile is the dependent variable. The averate rating variable now has a coefficient of 0,045 and is significant at a 1% level. This means that when the average rating goes up by one, the difference in percentiles goes up by 4,5%. Again, we can conclude that male subjects with a high rating, the incompetent, overplace themselves more, when comparing themselves to other male subjects, than the male subjects with a low rating, the competent. For male subjects comparing themselves to other male subjects, the findings are in line with what the Dunning-Kruger effect predicts. Furthermore, the Yrsofplaying has a coefficient of 0,03 and is significant at a 1% level. An explanation for this might be that subjects know that they have been playing tennis more than others and therefore conclude that they should be better than the other subjects as well.
27 Table 4: Regression results
male overplacement Diffpercentile
Average rating 0.045*** (0.009) Age -0.002 (0.006) Yrsofplaying 0.03*** (0.025) Academiclevel 0.015 (0.067) Constant -0.044 (0.067) Observations 76 R-squared 0.26
Robust t statistics in parentheses
* significant at 10%; ** significant at 5% ; *** significant at 1%
28
Figure 5 Plot of female overestimation
Figure 6 is the plot of the overestimation of female subjects. On the x-axis the average rating is shown and on the y-axis the difference between the self-rated rating and actual rating is shown. A positive difference means that the subject is overconfident and a negative difference means that the subject is underconfident. We can see a positive
relationship between the average rating and the difference. This means that as the rating of the subjects get higher, the more overconfident they are. This is in line with the Dunning-Kruger effect. To see whether this relationship is significant, the same regression analysis has been done for the female subjects as for the male subjects. The regression results are shown in table 5. The variable average rating has a coefficient of 0,169 and is significant at a 1% level. Furthermore, the variable Yrsofplaying has a coefficient of 0,048 and is significant at a 10% level. We can therefore conclude that female subjects with a high rating, the
incompetent, overestimate their abilties more than the female subjects with a low rating, the competent. The results are similar for the female subjects as for the male subjects and again, the Dunning-Kruger effect holds.
29 Table 5: Regression results
female overestimation Diffrating
Average rating 0.169*** (0.042) Age -0.039 (0.036) Yrsofplaying 0.048* (0.026) Academiclevel -0.057 (0.086) Constant -0.063 (0.751) Observations 60 R-squared 0.29
Robust t statistics in parentheses
* significant at 10%; ** significant at 5% ; *** significant at 1%
30
Figure 6 Plot of female overplacement
Figure 7 shows the overplacement of the female subjects. On the x-axis the average rating is shown and on the y-axis the difference between the self-rated and actual
percentiles is shown. For the male subjects, the regression line was upwards but for the female subjects, the line looks flat. This suggests that there is no difference in
overplacement for female subjects with a low rating and female subjects with a high rating. To see if this is true, a regression analysis has been done. The regression results are shown in table 6. The variable average rating has a coefficient of 0,004 and is not significant. We can therefore conclude that the average rating of female subjects has no effect on the
31 Table 6: Regression results
female overplacement Diffpercentile
Average rating 0.004 (0.014) Age 0.005 (0.012) Yrsofplaying 0.003 (0.008) Academiclevel -0.002 (0.027) Constant -0.17 (0.24) Observations 76 R-squared 0.26
Robust t statistics in parentheses
* significant at 10%; ** significant at 5% ; *** significant at 1%
32 4.2 The gender effect
To investigate whether there is a gender-based difference in overconfidence, a regression analysis has been done. The results of the regression analysis are shown in table 7. The variable of interest is the dummy-variable Male. The variable Male has a coefficient of 0,125 and is significant at a 10% level. This means that male subjects, on average
overestimate their rating with 0,125 points more than female subjects do. Therefore we can conclude that the male subjects are more overconfident than the female subjects.
Furthermore, the variables average rating and Yrsofplaying are also significant and have a coefficient of respectively 0,182 and 0,055. This is no surprise because we have already seen that when analyzing both the male and female subjects separately.
33 Table 7: regression results
one subject pool Diffrating
Average rating 0.182*** (0,027) Male 0.125* (0,075) Age -0.037* (0,019) Yrsofplaying 0.055*** (0.017) Academiclevel -0,05 (0,051) Constant -0,259 (0.464) Observations 136 R-squared 0.3217
Robust t statistics in parentheses
* significant at 10%; ** significant at 5% ; *** significant at 1%
34 5 Discussion and conclusion
Hypothesis 1: The participants, on average, will overestimate their rating that corresponds to their perceived tennis ability.
The current literature has shown that individuals are, on average, overconfident. The findings in this study are in line with what the literature predicts because both the male and female subjects, on average, overestimate and overplace their tennis abilities and therefore are overconfident.
Hypothesis 2: Lower skilled participants will overestimate their rating that corresponds to their perceived tennis ability more compared to higher skilled participants.
According to the Dunning-Kruger effect overconfidence is mainly found in the unskilled because the skills needed in a specific domain are often the same skills needed to evaluate competence in that same domain, one’s own and someone elses (Dunning &
Kruger, 1993, p.1121). In this study, lower skilled participants are the participants with a high rating and vice versa. To find evidence for this Dunning-Kruger effect, the subjects are
separated into four different quartiles. The two-sided t tests showed that males in the second, third and bottom quartile overestimated their tennis abilities and males in all four quartiles overplaced their tennis abilities when comparing themselves with other male participants. The female subjects overestimated their abilities in all four quartiles and did not overplace their tennis abilities in any of the four quartiles. The graphical analysis showed that the overestimation and overplacement of the male subjects increased when the rating increased. For females, only the overestimation increased when the rating increased. To prove if this is actually true, a regression analysis has been done. The regression results showed that an increase in the rating had a positive effect on the overestimation and overplacement of the male participants. For males, these findings are thus in line with what the Dunning-Kruger effect predicts. Furthermore, the regression results showed that an increase in the rating had a positive effect on the overestimation of the female subjects but did not have an effect on the overplacement of the female subjects. Therefore, we can conclude that the overestimation of female subjects is in line with what the Dunning-Kruger effect predicts but the overplacement is not.
35 Hypothesis 3: Lowest skilled participants, those in the lowest quartile on tennis ratings, will overstate their perceived rating compared to their actual rating, while the highest skilled participants will understate their perceived rating.
In the study of Dunning and Kruger (1999) the subjects that were in the top quartile actually underestimated their performance. The results in this study are not in line with the findings of Dunning and Kruger. In this study, subjects in all four quartiles were, on average, overconfident. One reason for this is that people tend to be optimistic about their sport abilities.
Hypothesis 4: Both male and female participants will overestimate their rating that corresponds to their perceived tennis ability, but male participants tend to be more overconfident than female participants.
According to the current literature, both man and woman are overconfident but man tend to be more overconfident than woman (Soll & Klayman, 2004; Barber, Odean, 2001). When looking at overestimation, both the male and female participants were overconfident. To investigate if there is a difference in overestimation, a regression analysis has been done with male as a dummy variable. The results showed that male participants do indeed overestimate their abilities more than female participants do. This is in line with what the literature predicted. When looking at overplacement, male participants are also more overconfident than the female participants. The male participants overplaced their tennis abilities in all four quartiles, the female participants did not overplace their tennis abilities at all. This is also in line with what the current literature predicts. An explanation for this gender-difference in overconfidence can be that males are more competitives than females and therefore want to believe they are better than other males more then females do.
Because, to the best of my knowledge, there is no existing literature on the gender-based difference in the Dunning-Kruger effect, it is also interesting to look at this difference. In this study, the Dunning-Kruger effect holds for both the male and female participants when looking at overestimation. When looking at overplacement, the Dunning-Kruger effect only holds for male participants. We can therefore conclude that there is only a gender-based difference in the Dunning-Kruger effect when looking at overplacement and male participants behave accordingly to what the Dunning-Kruger effect predicts more than
36 female participants do. For the female participants in this study, the findings are in line with the bias that Burson et al. (2006) predicted. According to this bias, the subjects should show more signs of the Dunning-Kruger effect when they rate themselves along objective scales compared to when they rate themselves on a comparative scale. Dunning opposed this theory by arguing that the Dunning-Kruger effect arises on both objective and comparative scales as low performers overestimate their performance, even without social comparison. In this study, this is true for the male participants.
This study has three main findings. The first finding of this study is that individuals, on average, are indeed overconfident. Secondly, field evidence indicates that overconfidence is mainly found in the unskilled. Finally, the results of this study show that males are more overconfident than females. Also, the Dunning-Kruger effect holds for both males and females when using the overestimation definition of overconfidence. When using the
overplacement definition of overconfidence, the Dunning-Kruger effect only holds for males. One limitation of this study is that the sample size is small and all participants are extracted from the same club. This study can be extended with a subject pool with tennis players from different clubs and age.
37 6 Appendix
38 7 References
Alicke, M. D., & Govorun, O. (2005). The better-than-average effect. The self in social judgment, 1, 85-106.
Bailey, W., Kumar, A., & Ng, D. (2011). Behavioral biases of mutual fund investors. Journal of Financial Economics, 102(1), 1-27.
Barber, B. M., & Odean, T. (2001). Boys will be boys: Gender, overconfidence, and common stock investment. The quarterly journal of economics, 116(1), 261-292.
Barberis, N. & Thaler, R., 2003. A Survey of Behavioural Finance. In G. M. Constantinides, M. Harris, & R. Stulz, eds. Handbook of the Economics of Finance. Elsevier Science B.V., pp. 1051–1121
Benoît, J. P., & Dubra, J. (2011). Apparent overconfidence. Econometrica, 79(5), 1591-1625.
Bertrand, M., & Hallock, K. F. (2001). The gender gap in top corporate jobs. ILR Review, 55(1), 3-21.
Brown, J. D. (2012). Understanding the better than average effect: Motives (still) matter. Personality and Social Psychology Bulletin, 38(2), 209-219.
Burson, K. A., Larrick, R. P., & Klayman, J. (2006). Skilled or unskilled, but still unaware of it: how perceptions of difficulty drive miscalibration in relative comparisons. Journal of personality and social psychology, 90(1), 60.
Camerer, C., & Lovallo, D. (1999). Overconfidence and excess entry: An experimental approach. American economic review, 89(1), 306-318.
Cesarini, D., Sandewall, Ö., & Johannesson, M. (2006). Confidence interval estimation tasks and the economics of overconfidence. Journal of Economic Behavior &
Organization, 61(3), 453-470.
Chambers, J. R., & Windschitl, P. D. (2004). Biases in social comparative judgments: the role of nonmotivated factors in above-average and comparative-optimism
effects. Psychological bulletin, 130(5), 813.
Chuang, W. I., & Lee, B. S. (2006). An empirical evaluation of the overconfidence hypothesis. Journal of Banking & Finance, 30(9), 2489-2515.
Dunning, D. (2011). The Dunning–Kruger effect: On being ignorant of one's own ignorance. In Advances in experimental social psychology (Vol. 44, pp. 247-296). Academic Press.
Fairchild, R. J. (2005). The effect of Managerial Overconfidence, asymmetric information, and moral hazard on capital structure decisions.
39 Gervais, S., Heaton, J. B., & Odean, T. (2003). Overconfidence, investment policy, and executive stock options.
Hubka, T. (2015). The capability to align estimated performance with actual
performance: Insights from physical & cognitive performance contexts (Master's thesis, University of Sydney).
Heath, C., Larrick, R. P., & Klayman, J. (1998). Cognitive repairs: How organizational practices can compensate for individual shortcomings. In Review of Organizational Behavior.
Hirshleifer, D., Low, A., & Teoh, S. H. (2012). Are overconfident CEOs better innovators?. The Journal of Finance, 67(4), 1457-1498.
Johansson Stenman, O., & Nordblom, K. (2010). Are Men Really More Overconfident than Women?-A Natural Field Experiment on Exam Behavior.
Kennedy, E. J., Lawton, L., & Plumlee, E. L. (2002). Blissful ignorance: The problem of unrecognized incompetence and academic performance. Journal of Marketing Education, 24(3), 243-252.
KNLTB (2017). Het DSS (Dynamische Speelsterkte Systeem). Retrieved from
https://www.knltb.nl/siteassets/1.-knltb.nl/downloads/tennissers/speelsterkte/dss_uitgebreide_uitleg.pdf
Krajc, M., & Ortmann, A. (2008). Are the unskilled really that unaware? An alternative explanation. Journal of Economic Psychology, 29(5), 724-738.
Krizan, Z., & Suls, J. (2008). Losing sight of oneself in the above-average effect: When egocentrism, focalism, and group diffuseness collide. Journal of Experimental Social Psychology, 44(4), 929-942.
Krueger, J., & Mueller, R. A. (2002). Unskilled, unaware, or both? The better-than-average heuristic and statistical regression predict errors in estimates of own performance. Journal of personality and social psychology, 82(2), 180.
Kruger, J., & Dunning, D. (1999). Unskilled and unaware of it: how difficulties in recognizing one's own incompetence lead to inflated self-assessments. Journal of personality and social psychology, 77(6), 1121.
Landier, A., & Thesmar, D. (2008). Financial contracting with optimistic entrepreneurs. The Review of Financial Studies, 22(1), 117-150.
Lenney, E. (1977). Women's self-confidence in achievement settings. Psychological bulletin, 84(1), 1.
Malmendier, U., & Tate, G. (2005). CEO overconfidence and corporate investment. The journal of finance, 60(6), 2661-2700.
40 Moore, D. A. (2007). Not so above average after all: When people believe they are worse than average and its implications for theories of bias in social comparison. Organizational Behavior and Human Decision Processes, 102(1), 42-58.
Moore, D. A., & Healy, P. J. (2008). The trouble with overconfidence. Psychological review, 115(2), 502.
Moore, D. A., & Small, D. A. (2007). Error and bias in comparative judgment: on being both better and worse than we think we are. Journal of personality and social
psychology, 92(6), 972.
Moore, D. A., Tenney, E. R., & Haran, U. (2015). Overprecision in judgment. The Wiley Blackwell handbook of judgment and decision making, 182-209.
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.
Odean, T. (1998). Volume, volatility, price, and profit when all traders are above average. The Journal of Finance, 53(6), 1887-1934.
Odean, T. (1999). Do investors trade too much?. American economic review, 89(5), 1279-1298.
Plous, S. (1993). The psychology of judgment and decision making. Mcgraw-Hill Book Company.
Prims, J. P., & Data, D. A. M. (2017). Overconfidence over the lifespan. Judgment and Decision Making, 12(1), 29.
Reed, E. S., Turiel, E., & Brown, T. (2013). Naive realism in everyday life: Implications for social conflict and misunderstanding. In Values and knowledge (pp. 113-146). Psychology Press.
Ross, L., Greene, D., & House, P. (1977). The “false consensus effect”: An egocentric bias in social perception and attribution processes. Journal of experimental social
psychology, 13(3), 279-301.
Roy, M. M., & Liersch, M. J. (2013). I am a better driver than you think: examining self‐ enhancement for driving ability. Journal of applied social psychology, 43(8), 1648-1659. Schlösser, T., Dunning, D., Johnson, K. L., & Kruger, J. (2013). How unaware are the unskilled? Empirical tests of the “signal extraction” counterexplanation for the Dunning– Kruger effect in self-evaluation of performance. Journal of Economic Psychology, 39, 85-100.
Soll, J. B., & Klayman, J. (2004). Overconfidence in interval estimates. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30(2), 299.
Svenson, O. (1981). Are we all less risky and more skillful than our fellow drivers?. Acta psychologica, 47(2), 143-148.
