The effect of awareness on
overconfident sports bettors
Abstract
People are overconfident. They overestimate their knowledge, skills and ability to control the
outcome of events which is problematic when it comes to gambling and sports betting. This
research tries to reduce the overconfidence bias by informing gamblers about the bias. I
performed an experiment in which participants needed to predict twenty-four football games
in which only half of the respondents was made aware of the bias. The forewarned people did
not calibrate their performance better and therefore awareness did not have a significant
impact. However, the results showed that the most incompetent gamblers were the most
overconfident.
Name: Mick Fruytier
Student number: 10553770
Bachelor program: Economics & Business
Specialization: Business Administration
Supervisor: Rob van Hemert
Statement of Originality
This document is written by Student Mick Fruytier who declares to take full responsibility for
the contents of this document.
I declare that the text and the work presented in this document is original and that no sources
other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of
completion of the work, not for the contents.
INDEX
1. Introduction
4
2. Theoretical framework
5
2.1 What is overconfidence?
5
2.2 Gambling and overconfidence
8
2.3 How to influence overconfidence?
11
3. Hypothesis
13
4. Methodology
15
4.1 Participants
15
4.2 Materials and procedure
15
4.3 Randomness
17
5. Results
17
5.1 Accuracy
17
5.2 Certainty
21
5.3 Percentage of correct responses
23
5.4 Overplacement
24
5.5 Demographics
26
5.6 Draw
27
6. Discussion
27
6.1 Limitations and remarks
27
6.2 Additional research
28
7. Conclusion
28
8. References
30
1
.
IntroductionA couple of months ago, my sister and I were talking on a Sunday evening back in my hometown at my parents’ place. We were having a simple conversation regarding everyday life. I do not know how the subject came up but suddenly she looks at me and says: ‘’I’ll still beat you in a hundred-meter sprint contest’’. I was surprised by her statement, started to laugh and thought; is she crazy? My sister used to be quite a good sprinter when she was younger but nowadays she is working in an office from nine till five and she has not played any sports over the last ten years. For me it was obvious that I was going to win a hundred-meter sprint because I do play sports every week and I get compliments for my speed during games. This conversation got me thinking again about the concept overconfidence bias. Why does my sister believe that she is a better sprinter then I am? Especially since we did a little sprint during our holiday in Ibiza last year where I won easily. She did not seem to remember this race anymore. This was the second time the overestimation bias came to my interest. Overconfidence bias refers to the systematic error of the judgements made by individuals when they are evaluating their own performance and looking at the correctness of their responses (Pallier et al., 2002, p. 258). People who make errors in the evaluation of their own performance can be under- and overconfident. One is overconfident when the actual performance is worse then what he or she predicted, and someone is underconfident when the actual performance exceeds one’s expectations.
The first time that I was really intrigued by this phenomenon was during my first year of college when me and my friend started to make sports bets. During this one year I spend over 600 euros on tennis, football and ice hockey matches. Fortunately, I won 70 percent of my deposits back, but it was still quite a big loss. Now, two years later I am not playing anymore. As I started living on my own and started paying for my own food everyday, I had other priorities. Although I stopped playing when money became an issue, I still played for a long time even though I was not performing well. Why did I do this? Obviously, the factor ‘fun’ played a role since it was nice to go to the store together and be there for half an hour deciding on which games to bet. Yet, was I really thinking I was able to predict all those games correctly? Or, was I was just overconfident with my skills.
As a result of my gambling ‘addiction’, I wanted to focus this research on the overconfidence bias in sports betting and especially focus on how to prevent someone of being overconfident during gambling. I want to see whether I am able to prevent people from being overoptimistic and therefore have an influence on one’s behaviour. Resulting in the research question: To what extend is it possible to influence someone’s overconfidence with sports betting with awareness? According to the Dutch government, becoming addicted is one of the risks of gambling and entails negative personal and social influences. Physical, mental and financial problems are the results of an addiction and therefore the government sees the prevention of addiction as one of the top priorities of their gambling policy (Statistics Netherlands, 2015, p. 13). In the same analysis, CBS (Statistics Netherlands) stresses out interesting facts regarding online gambling. The CBS found that during the last decade the online gambling market in The Netherlands grew nearly twenty-six percent on a yearly basis. Which means
the total proportion of the online market relative to the total market is now thirteen percent (2015, p. 44). Since, sports betting takes account for almost half of the activity of the Dutch online market, it seems to be that there is enough evidence that sports betting is an interesting topic to study because it could become a growing problem for the Dutch society.
To find an answer to the research question the following things need to be clarified. First of all, it is important to find out exactly what overconfidence means and when it accrues. Second of all, it is necessary to know whether sports betters are overconfident or not. Alongside the literature review, this question will also be answered trough a simple experiment. This experiment is designed to see if sports betters think they perform better then they actually do. The control group of the experiment will be asked to predict twenty-four games of football and afterwards they will be asked to evaluate their own performance. And the experimental group will also be asked to predict the same twenty-four games. However, this group will be given a text with information regarding the overestimation bias and overconfidence. The last question that needs to be answered is whether it is possible to influence overconfident people or not. Based on the results of the experiment and based on the literature review I will give answer to the research question.
In the next chapter, the theoretical framework, all the concepts and theories, necessary to answer the research question will be discussed. Also, the most important studies and the status quo of the topic will be addressed. In the third chapter, the hypotheses will be mentioned. After the
hypotheses the methodology will be discussed. In the methodology section, the way the experiment was executed will be clarified. In the fifth chapter, the results of the experiment will be analysed and visualised. In chapter six, the limitations of this research and possibilities for follow-up studies will be the discussed. Finally, there will be a conclusion in chapter seven.
2. Theoretical framework
2.1 What is overconfidence?The definition of overconfidence that derives from literature has three different aspects. First of all, overconfidence is the overestimation of someone’s actual performance and therefore called
overestimation. Second of all, overconfidence is the belief that one’s performance is good relative to others. Which is called overplacement. Last, the inordinate precision in one’s belief (Moore & Healy, 2008, p. 502). Thus, overprecision.
2.1.1 Overestimation
Overestimation is the first variety of overconfidence and refers to the fact that people overestimate their own actual ability and performance. People might believe that they performed better then they
actually did. Overestimation derives from the fact that individuals miscalibrate the chance of success and they have some sort of sense of control (Moore & Healy, 2008, p. 503)
2.1.2 Overplacement
The second aspect of overconfidence occurs when people believe they did a better job performing a task then their peers. This is called overplacement (Moore & Healy, 2008, p. 503). For example, when students believe they are better then their peers and think they belong to the best performing twenty percent of the class but they are actually in the bottom half, those students overplaced themselves. This is something different than overestimation. Overestimation only concerns one’s actual ability. Namely, the amount of right answers one gave to a test. Not the relative ability in comparison with others.
As already been stated regarding overconfidence, people who are skilled often undervalue their own performance. But, what happens with capable individuals concerning overplacement? The top performers often overestimate the performance of their peers and therefore underrate themselves. They had higher expectations of what the others were capable of and therefore underestimate their own performance. Although people do this, they can still believe that they are relatively good (2008, p. 502).
2.1.3 Overprecision
Overprecision is the third facet of overconfidence. ‘Overprecision is the excessive certainty regarding the accuracy of one’s believes’ (Moore & Healy, 2008, p. 503). In other words, how sure is someone that one knows the outcome or answer (whether its outcome or answer depends on the kind of task one needs to perform). Overprecision is often measured through a confidence interval. Participants are regularly asked to give a percentage that reflects their certainty that they provided the correct answers (Haran et al., 2010, p. 467).
2.1.4 Examples of overconfidence
As already stated in the introduction; overconfidence is a classic problem in daily live. Several psychologists, meteorologists, statisticians and economist have studied the concept. Those researchers have shown that people are overconfident with their own performance in a wide scale of abilities (Hvide, 2002, p. 15). However, few individuals are overconfident in everything they do. But still, individuals make the wrong judgment concerning their accuracy of their performance (Pallier et al., 2002, p. 258). For example, with the ability; driving. Many drivers continue to take risks on the road despite the possible consequences. For instance, people still drink and drive and maintain speed through stop signs (Job, 1990, p. 97). When it comes to driving, human beings tend to suffer from two biases. They see themselves as relative good drivers and think they are able to identify hazardous
situations. Therefore, they also believe that they are more cautious then their fellow road users (Sibley & Harré, 2009, p. 165). Obviously, driving is not the only task where overconfidence accrues. In an experiment with students, in which they were asked to predict how well they had done on an exam, overconfidence was also noticed. Right before the students left the class they need to answer the following questions: how well have you performed relatively to your peers and, what do you think your raw score will be. Especially the lowest percentile of students was the most overconfident. They overestimated their raw score with almost ten points (out of forty-five). They believed they scored thirty-three points on average when actually scored twenty-five. The worst performers also believed they had better mastery over the content of the course material then their peers. The worst students thought their performance would rank them in the 57th percentile while they were actually around the 25th percentile (Dunning et al., 2003, p. 85).
2.1.5 To what extent does overconfidence occur?
The extent to which overconfidence occurs depends on several different aspects. One of these aspects is; the difficulty of the task one’s performing. Different studies have revealed a difference in
overconfidence between hard and easy questions. The difference between an easy and a hard question or task is simple: easy questions are questions to which many participants know the correct answer and hard questions are questions to which a lot of participants guess the wrong answer. When agents perform a simple task, overconfidence seems to disappear and even undervaluation is detected. But, when agents perform a difficult task, overconfidence is increasing (Kleyman et al., 1999, pp. 217-218). However, this does not hold for every tough task. For instance, swimming. People who have never done this in their lives will not think they are better than average. Researchers Moore and Cain give riding a unicycle and computer programming as examples for this phenomenon (2007, p. 197). It seems to be that one has to cross a competence border. If people have not crossed this border and are incapable of performing a task at all they also won’t suffer from overconfidence.
Next to the difficulty of the task, the amount of overestimation is also determined by the extend of one’s knowledge or the ability to perform a particular task. For most domains in life, you need some kind of skill, knowledge, wisdom or strategy to be successful. Many people whom excel in a certain domain are aware of the fact they perform good but still undervalue their own performance. I suppose this does not hold for someone with a negative self image. But, according to Kruger and Dunning (1999, p. 1121), the ones that are not capable/successful will not undervalue their
performance. They suffer from a dual burden. Not only do these people perform bad but they are also not aware of their incompetence. They argue that they are not conscious of this because, the skills necessary to perform good are often the same skills needed to recognize a proper performance. For example, consider the ability to write a grammatically correct sentence in English. The skills needed to write correctly are the same skills needed to identify mistakes (1999, p. 1121). Kruger and Dunning
overestimate their ability and performance relative to others (overplacement), incompetent individuals will not recognize a solid performance when they see one and will not gain insight of their true level of performance trough social comparison (overconfidence) (1999, p. 1122). All three hypotheses were accepted.
The third reason why overconfidence occurs is because people believe they have greater sense of control then they actually do. sense of control is the degree to which someone believes that
everything that happens to him or her is due to their own actions. People can either have an internal locus of control or an external locus. Individuals who have an internal locus of control are convinced that they control their own success and failure and people who have an external locus of control believe the opposite. Individuals are more likely to link their own performance with its outcome when it is a positive relationship then when it is not (Bradley, 1978, p.68). In other words, people tend to ascribe success to their own capabilities and failure to external factors.
2.2 Gambling and overconfidence
This paragraph is focussed on finding out whether sports bettors are overconfident or not. To find the answer to that question I need to examine how people gamble and if it differs from other skill required activities. Therefore, I will first find motives for why people gamble and give insight in their gambling behaviour. Secondly, I am going to display the difference in overconfidence between gambling and other activities.
2.2.1 Why do we gamble?
Why do we (human beings) actually gamble? The possibility that one wins a lottery is often one to a lot of millions and the possibility that one loses a lot of money in the casino is quite big. But despite this, Americans spend over 70 billion dollars on a yearly basis on lotteries which is on average more then they spend on books, video games and movie theatres (Thompson, 2015). This means the gambling industry is quite big. People play because it is fun, but their basic motive is just to win money (Rachlin, 1990, p. 294). However, the downside of gambling, like downside alcohol and drugs, is that one can become addicted to it. The essential part of being addicted to something is that this activity becomes a major part of one’s life to which one fully commits and lose themselves. Most often, when someone is addicted it leads to a negative life outcome (Peele, 2001, pp. 3-4). One of the explanations that people still gamble despite the negative consequences is that individuals seem to remember wins more than they do remember losses (Rachlin, 1990, p. 294). Therefore, people can start playing again while thinking that they were successful the last time, since they weigh more value in their minds to the wins they had then to the losses.
2.2.2 Chasing
People go to a casino not just to play one or two rounds of roulette; people play multiple bets in a row. Since gambling costs money, gamblers need to keep track of their activities and tend to evaluate their outcomes/revenues in strings. A single string may consist of one ore multiple bets but the string only consists of one win. After this win, the monetary outcome is evaluated (Rachlin, 1990, p. 296). For example, if someone loses the first two bets and wins the third, the string will look like this: LLW. After winning the third game the person will evaluate his or her performance. So, it seems to be the case that people will continue playing until they have won once and can only stop after a win. When playing, it is possible that one losses several games in a row. When someone is in a losing streak, people tend to take riskier bets to get ‘that’ win (1990, p. 296). This phenomenon is called chasing. For sports gamblers, it works approximately the same. According to Rachlin, sports betters also reset their account and evaluate their performance right after a win which indicates that those gamblers will also continue playing until they have won a bet (1990, p. 297).
2.2.3 Difference in overconfidence between activities
There is a difference in overconfidence between gambling and other activities. For example, when making a test, the required knowledge to answer the questions correctly is a skill. People either have the skill or not and the relationship between the skill and giving the right answer is clear. But this gets interesting when one is performing a bet. There is a bit of an overlap between skill and luck when someone is gambling. Gambling is an activity in which individuals believe they have a sense of control. They have an unconscious believe that they can have some sort of influence on the outcome of a game because, during chance games they behave as though the game is subject to control (Langer, 1975, p. 311). This is where overconfidence accrues. Sense of control is one of the mediators of overconfidence. Where does the illusion of control comes from? In a series of studies that was executed by Langer, she found that there were several factors that had influence on people which made them ‘inappropriately confident’ (1975, p. 311). Factors that made them believe they were in control, such as competition, choice, familiarity and involvement had, as she hypothesised, an influence.
In one of the six experiments, Langer examined that involvement resulted in a greater sense of control. Prior to her study, she argued that when someone gives a lot of thought to a particular skill game, someone will come up with a strategy that will increase the likelihood of success (1975, p. 320). Therefore, she predicted that this would also be the case in a chance game and that this increased involvement would create some kind of illusion of control. In her experiment she measured
involvement on the basis of time. She believed that when one is occupied with a particular game for a longer period of time, one will get a higher perceived sense of control. The results of her study
thought they were going to win the lottery (that she created) or not. Later on at night, closer to the draw of the winning ticket, people started to believe that they had better chance of winning it.
2.2.4 Experience and overconfidence
It is often considered that one does not need skills to win a bet. You do not need to be an expert to put a five euro chip on red or black with roulette or to pick a winning team in a sports contest. However, some studies have revealed that sports experts do have some kind of skill and are able to defeat chance. In a study by Cantinotti, Ladouceur and Jacques (2004), it was examined whether expert betters could get a higher rate of correct predicted games and if those experts could get a higher monetary gain then random selection could do. For their study Cantinotti et al. asked thirty-five (hockey) experts to predict several games from the 2002/2003 National Hockey League season. A total of 1963 predictions were made and from these games, 47.3 percent were predicted correctly by the gamblers. In comparison, random selection merely came to a 33.3 percent of correct games. The percentage of correct predicted games that the random selection got is not surprisingly since the chances that a team wins a game is in essence 33.3 percent. A team either wins, loses or draws. However, sports contests are dependent to a lot of factors which influence the nature of the game. For example, which teams are facing each other, who is playing at home. But, because experts defeated the computer by 14 percent, the first hypothesis from Cantinotti et al, that experts can beat randomness, was accepted (2004, p. 145). Cantinotti et al. also tried to find evidence for their second hypothesis but that did not happen. In their experiment, the computer got a higher return rate then the hockey experts. However, the results were affected by an outlier and were not significant (2004, p. 145).
Despite the fact that the hockey experts were better predictors than the computer, Cantinotti et al. still argue that sports betters are overconfident. The hockey experts that participated in their
experiment indicated that they believed they have a better chance of winning a bet when they relied on recent results and other hockey knowledge. However, according to Cantinotti et al. knowing the past performances of the contesting teams does not lead to efficient wagers. Knowing this would only lead to a greater sense of control and their accuracy would only reinforce this sense of control. This is one of the three reasons why these researchers argue that the experts are still overconfident. (2004. P. 145).
The second argument from Cantinotti et al. is a quote from a research from the early nineties from Clotfelter and Cook. They cite these researchers to stress out why a teams past performance is not a good indicator for future results. ’The gambler’s fallacy is the belief that the probability of an event is lowered when that event has recently occurred, even though the probability of this event is objectively known to be independent from one trial to the next’ (1993, p. 1521).
The third argument of Cantinotti et al. for why sports bettors are still overconfident despite of their accuracy is the nature of the game. When a gambler is placing a bet, he or she might choose to bet on one game only. However, this is often not possible. Gamblers are encouraged/it is mandatory for gamblers to combine multiple games. In this way, people will get a higher pay-out because the
odds will be multiplied. But, the chance that you win something is smaller when you bet on two games together (1/3*1/3) than on two games separately (1/3+1/3). Maybe humans are, to a certain extend able to predict games but are misled to play in a way that is unfavourable for them.
2.2.5 Gambler’s fallacy vs hot hand
Based on the gambler’s fallacy, it makes sense to argue the following: when one team is leading the league table and is on a winning streak, you should bet they will lose their next game because they have already won a couple. This is because the probability of an event (winning) is lowered because it already recently occurred. Burns and Corpus find this contradictory and therefore argue the opposite. When someone believes the team who is in first position and on a winning streak will win another time, one believes in the hot hand (2004, p. 179). The hot hand is contrary to the gambler’s fallacy because believing in the hot hand is the belief that the probability of an event is increased when that event has recently occurred. However, when the team who is in first position but lost two games in a row, you are almost going to think they have to get back to winning ways so the gambler’s fallacy might also hold for sports betters too. Burns and Corpus have found a difference between the gambler’s fallacy and the hot hand which might explain this contradiction. In their study, they
distinguished a dissimilarity between random and non-random games. They argue that when playing a non-random game (such as sports betting) gamblers tend to believe in the hot hand and when playing a random game (such as roulette) individuals tend to believe in the gambler’s fallacy (2004, p. 182). So, one could say that the fallacy does not hold for sports betters.
2.2.6 Chances on a win versus a draw
Earlier on in section 2.2.4 I mentioned that in essence the chance of winning a match is 33.3 percent. However, previous research has indicated this is not true. In a study by Dixon and Coles (1997), the results of three seasons Barclays Premier League football were investigated (1993-1996). They found the ratio of frequencies of home wins, draws and away wins to be 46:27:27. So, the probability that a randomly selected match ends up in a win for the home team is 46 percent (pp. 267-268). During a European Championship, there is no home or away team (except for the hosting country) thus the chances on a draw might differ from the regular club seasons. In a study by Archontakis and Osborne (2007), results from World Championships from 1982 to 2002 were examined. They found that during the group stage the chance of a draw is 29 percent and during the knockout phase the chance on a draw is 32.3 percent (p. 298).
2.3 How to influence overconfidence?
Since online gambling and sports betting is a growing business it is important to know how players can be influenced to avoid the pitfall of becoming addicted. One of the things that can be done to
influence overconfident people is to give feedback. In the past, a lot of research has been done to the relationship between feedback and overconfidence. When one is calibrating his own performance (calibration is the process of assessing the accuracy of one’s knowledge), feedback plays a prominent role. For example, this is shown in a study by Winman and Juslin. The participants in their study answered several multiple choice questions with two possible answers. After answering a question, the participants needed to give an indication on how sure they were that they had given the correct
response on a 50 percent till 100 percent scale (1993, p. 143). The scale started at 50 percent because if someone gambles the chance that he or she gave the right answer is 50 percent. The difference between the experimental and the control group was that the control group received no feedback at all and the experimental group received feedback after answering a question. As a result of the
experiment they concluded that feedback significantly improved people’s calibration and even ruled out overconfidence (1993, p. 146). But this research is not the only one who affirms this relationship. In a comparable study, executed by Labuhn and Zimmerman, the role of feedback was also examined but this time, students were solving mathematical problems. Also, in this research they tried to find a relationship between self evaluation and calibration (2010, p. 179). However, the set of
self-evaluative measures did not produce any effect on students’ calibration (2010 p. 187).
Next, Researchers Russo and Schoemaker also found several ways to reduce overconfidence (1992). One of them was found in an experiment where they asked participants to answer multiple choice questions. One of the things they did was to let the participant think about a reason why the alternative answer (the answer they did not choose), could be correct. After they gave at least one counterargument, the participants were asked if they wanted to change their original answer or not. The control group was on average 72 percent sure they were right but only 54 percent picked the correct answer. This means they were overconfident by eighteen percent. The group that needed to give a counter argument thought they were 73 percent sure but gave the right answer more often. In this group, 62 percent was right which means they were overconfident by only eleven percent (1992, pp. 12-13).
A third measure to influence overconfident people is to let your participants do a scenario analysis, both positive (paths to success) and negative (paths to trouble) (Russo and Schoemaker, 1992, p. 13). If people are forced to have a look at all the possible routes and problems, one might calibrate his or her chance to success more accurately. Asking managers to construct different
pathways makes them better evaluate the uncertainty (1992. p. 13). This scenario analysis differs from the previous measure (counterarguments) because during a scenario analysis the participant start from scratch while with the counterarguments method the participant already chooses a first answer and is further in the decision process.
Finally, Russo and Schoemaker also give awareness as an affective measure for reducing overconfidence. They acknowledge that awareness may be all one needs to reduce overconfidence (1992, p. 13). In a negotiation experiment, the role of awareness was further examined. The
researchers took one half of the participants and made them aware of the phenomenon overconfidence. After this, the forewarned people were 30 percent more likely to come to an agreement then the unawared individuals (1992, p. 14).
If you want awareness to be effective, I believe it should be focussed around introducing the concept of overconfidence and not just on the dangers of gambling. In the past, awareness campaigns with the goal to inform people on the dangers of gambling were not successful. In a study that was done in Indiana (USA) the effect of social marketing campaign on gambling was examined. In this study, they asked the inhabitants from Indiana some questions before the campaign started, and afterwards and it appeared that the campaign did not have a significant impact (Najavits & Grymala2003, pp. 324-326).
2.3.1 How to influence overconfident sports betters?
For this research on the calibration of sports betters, it is difficult to give feedback to the participants because of the nature of the game. Placing a bet is not like a solving a mathematical problem. With math, there is a possibility to give feedback on how one is attempting to get to the right answer but with sports betting, that is not really possible since there is not a correct answer till the game has been played.
The second method for reducing overconfidence that was mentioned in section 2.3 was to ask participants whether they could give a counterargument for their answer. If this method is used on people who are performing a sports bet, the answers could become a bit unilateral. When one needs to predict twenty-four games (as it is in my experiment) and needs to give a counterargument for every match the easiest thing to say for a participant will be: ‘’Well, team B is also good’’. However, this could be ruled out by just forbidding one to answer in this way.
A third way to influence overconfident people is to let them do scenario planning. However, this measure might not work for sports betting since there is not really a possibility to do a scenario analysis other then just predicting how the game will elapse.
Finally, the fourth way to influence overconfident people is to make them aware of the fact that they are overconfident. For the arguments, mentioned in the last three paragraphs, there were several factors why these methods would not work for sports betters. However, I think it could be interesting to make sports betters aware of overconfidence and therefore I will try to find evidence that awareness reduces overconfidence.
3. Hypotheses
One thing that has not been done by Cantinotti et al. is to examine whether sports betters are really overconfident or not. These researchers give three arguments for why their respondents are overconfident but do not have absolute numbers to support their statements. They did not ask the
participants how many games they thought they would predict correctly and compared this with their actual score. In my experiment I will do this to see whether sports bettors are overconfident or not.
Hypothesis 1
Sport betters are overconfident and believe they will predict more games correctly then they will actually do
In section 2.3.1, al the possible options to reduce overconfidence were mentioned. I believe awareness is the only useful measure to influence sports bettors. In this research, awareness will be focussed around making the participants cautious about the overconfidence bias. Thus, I want to reduce the bias by making the participants aware of the bias. To see whether the awareness had an influence, the following hypotheses will be tested.
Hypothesis 2
Forewarned students will have a lower expected number of correct predicted games Hypothesis 3
Forewarned students will be less sure with their predictions Hypothesis 4
Forewarned students will rank themselves in a lower quartile
I do not think the awareness (intervention in experiment) will influence the absolute score that the participants in both groups score. The information the respondents will receive about the bias is namely not created with the intention to chance the gambling patterns of the participants.
Hypothesis 5
The means of actual correct predicted games will not differ between the experimental and control group.
The question whether sports betting is a skill or just luck has not been answered yet. Personally, I do think that knowledge could lead to more efficient wagers and therefore someone with more knowledge would predict more games correctly.
Hypothesis 6
The results from a previous studies show that the chance of a draw in the group stage of a final tournament is is 29 percent. I expect that the people who have football knowledge, know that a draw happens more often than you might think because underdogs surprise and favourites fail.
Hypothesis 7
The more football knowledge someone has, the higher the number of predicted draws
4. Methodology
4.1 Participants
Sixty-three individuals participated in this research by filling out the survey. The participants are mostly male students with an average age of 21.94 years (SD 2.01). Fifty-three of the sixty-three students were male and ten of them were female. Three fifty year olds filled in my questionnaire too but I decided to leave them out of the analysis because there is a discrepancy between their
characteristics and the characteristics of the other participants. To see whether the respondents could be used for my analysis, the participants were asked whether they have the idea that they posses some kind of football knowledge. This was asked at the beginning of the survey. Several students who wanted to take part in the experiment in the first place but stopped after they found out what they were supposed to do. This was because the first question about football knowledge scared them. The respondents were obtained via Facebook and at the University of Amsterdam. I started gathering participants at the university but when the start of the European Championships was approaching I also asked my friends via social media if they wanted to participate in my experiment. The
respondents did not receive anything for participating other then my appreciation.
4.2 Materials and procedure
For this research, two different questionnaires were used. One for the control group and another one for the the experimental group. I will first describe the control group survey. This survey contained two demographic questions: are you a male or female and how old are you. These questions were asked to make sure both the experimental and control group have the same characteristics. Secondly, the respondents were asked to give an indication of their football knowledge on a scale of 1 to 7. I asked this question for two reasons. The first reason was already mentioned in section 3.1 (to make sure the people without any football knowledge would directly leave the survey). And the second reason was because I wanted to see whether there is a difference in scores between someone who thinks he has a lot of knowledge and someone who believes he does not have.
So far an example, the first respondent was a male, 22 years old and rated his football knowledge with a 5. Next, the participants needed to predict the first twenty-four games of the European Championships 2016 in France. For each game, there were three options; a home win, an away win or a draw. After the participant chose one of these three options, he or she needed to give an indication of how sure they were they had given the right answer on a 30% to 100% scale. The 22-year-old male (the first respondent) believed France was going to win the opening match against Romania and he said he was 80% percent sure of his prediction. After all the participants predicted all twenty-four games they were asked to answer to three more questions. Firstly, they needed to give an indication of how many games they predicted correctly in total (overconfidence). Secondly, they were asked to give an indication of how well they performed relative to others. So, whether they were in the bottom, the second, third or top quartile (overplacement). Finally, they were asked how sure they were they gave the correct answer to the question concerning the relative performance.
The experimental survey was almost exactly the same as the control survey however, there was one major difference. Instead of starting predicting right away, the participants were asked to read a short text regarding the overconfidence bias. In the first paragraph of this introduction, the
phenomenon of the overconfidence bias was elaborated and possible pitfalls were mentioned. Also, the result of the experiment done by Cantinotti et al. was cited (Experts are only able to predict 47,3 percent correctly). I wanted to mention the 47,3 percent because I wanted to see whether this statement would lead to a better calibration of the question; how many games did you predict correctly? at the end of the survey. I choose not to mention that the participants in the research of Cantinotti needed to predict matches from a different sport. In the second paragraph, I gave two explanations of why the overconfidence bias accrues. I did this because argumentation is the best tool for shared understanding (Schwarz, 2009, p. 91). And Finally, in the third paragraph I wanted to mention the problem of overplacement since the worst performers are always the most overconfident.
4.2.1 The introductory text
We are overconfident. ‘’Psychologists have determined that overconfidence causes people to overestimate their knowledge, underestimate risks, and exaggerate their ability to control events’’ [1]. So, people believe they can perform better than they will actually do and think they have influence on the outcome of certain events. People are overconfident in different aspects of life but especially when they gamble and more specifically when they are betting on sports games. We think we can predict the outcome of a game while in fact we cannot. In a recent study it was found that experts can only predict almost half of the games correctly (47.3%). In this experiment the experts predicted almost 2000 games [2].
There are a few explanations for why people think they are good gamblers. Firstly, humans seem to remember wins more then they do remember loses, so when they think about gambling, the first things that will come up to their minds is a thought of a victory. And Secondly, people will ascribe a success to their own abilities and will ascribe failure to external factors. Humans easily make a relationship between their actions and a positive outcome but not between their actions and a negative outcome.
Not only do people think they are better then they actually are, people also overestimate their performance relative to others. In a recent study executed by Kruger and Dunning, it was found that incompetent people are
the most overconfident and are unaware of this [3]. In several experiments that they have done, the worst performing participants thought they were performing average while in fact they were the worst participants.
4.3 Randomness
Participants who participated in this research could either do the experimental survey or control group survey. In order to assure validity, the students were distributed on a random basis. They needed to choose between the letters A, B, C & D in order to proceed to rest of the questionnaire. Both the letters B and D were linked to the control group and A and C were connected to the experimental group (29 students choose for A or C and 34 for B or D).
5. Results
5.1 Accuracy
As already been mentioned in this research a couple of times, the sports betters from the research of Cantinotti et al. had a 47,3 percent accuracy. The participants were called experts by the researchers because they bet on a regular basis. However, you might question their expertise since they score below 50 percent. The respondents in this research have predicted 11,68 (SD 1.522) out of 24 games right on average which is 48,7 percent and is higher than the accuracy of the experts. The worst performing participant in this research predicted eight games correct and the best scoring participant predicted sixteen games correctly. I would not call the respondents who participated in my research experts because most of them do not gamble on a regular basis.
Between the experimental and control group, there is no significant difference in accuracy because both groups performed almost exactly the same. The control group scored 11.68 (SD 1.319) comparing to the 11.69 (SD 1.754) from the experimental group. These results confirm hypothesis 5: The means of actual correct predicted games will not differ between the experimental and control
group. The histogram of figure 1 presents the deviation of participants with on the X-axis - number of
correct predicted games and on the Y-axis – frequency.
I also checked whether the sample(s) is(are) normally distributed. If we combine the two samples and perform both the Kolmogorov-Smirnov and Shapiro-Wilk tests, the results confirm that the sample is normally distributed. However, if we split them into the experimental and control group, only the Kolmogorov-Smirnov test confirms normality. According to Razali and Wah (2011, p. 21) the Shapiro-Wilk formula is the most powerful for testing normality and those results are not significant if we split up the data set into the two groups.
5.1.2 Accuracy and self assessed expertise
At the beginning of the survey, respondents needed to rate their own football knowledge on a 1 to 7 scale. On average, the 63 students rated their expertise with 4.94 (SD 1,243). However, there is a difference between the two surveys since the control group rated themselves slightly better than the experimental group; 5.18 (SD 0,936) and 4.66 (SD 1.495), respectively. I performed both the Welch and Brown-Forsythe test to see whether the difference between the two samples was significant or not. Both tests give a significance of .111 (p > 0.05) which indicates there is no significant difference between both means (Figure 2). Welch and Brown-Forsythe are used to test if the averages of two samples with a different size significantly distinct or not.
Figure 2
If we have a look at the sixty-three respondents, it is evident that it does not matter which amount of football knowledge one believes to have since the all the students score between 10,67 and 12 on average in every group (Figure 3). Therefore, hypothesis 3: The more football knowledge someone
has, the higher their correct predicted number of games, will not be accepted.
5.1.3 Accuracy of self assessment
At the end of the survey I asked the respondents to give an indication of how many games they thought they would have predicted correctly. On average, the sixty-three respondents believed they would predict 14.76 (SD 3.339) games correctly. This means the participants thought they would predict three more questions (games) correctly then they actually have (14.76 expected versus 11,68 actual). These results confirm hypothesis 1; Sport betters are overconfident and believe they will
predict more games correctly then they will actually do.
There is also a difference between the two surveys. The control group believed they would predict 14.94 (SD 2.707) games correctly and the experimental group only 14.55 (SD 3.996). However, this difference between theses two samples is not significant. Again, I performed both the Welch and Brown-Forsythe tests and the outcome was a not significant of .658 (p > 0.05) (Figure 2). The results do not support hypothesis 2: Forewarned students will have a lower expected number
of correct predicted games.
Figure 4
As we have seen in section 4.1.2, it does not matter if people rate their football knowledge with a 2 or a 6, the respondents scored between 12.25 and 13.25 on average. However, the more knowledge one believes to have, the higher the mean of the variable I predicted __ games correctly will be. Figure 5 shows the results of the sixty-three participants together. Furthermore, I wanted to see whether the variable Football Knowledge is a predictor of the variable I predicted __ games correctly. To examine this, I executed a linear regression finding a R2 of .099 with a significance of .012 (Figure 6.1 & 6.2).
Figure 6.1
Figure 6.2
Figure 7
This means that someone’ self-assessed football knowledge explains for almost ten percent of how many games one thinks one will predict correctly. I also calculated the Pearson correlation which is .315 (Significance .012). This indicates there is a moderate positive relationship between the two variables.
Figure 7 shows the difference between the actual performance of the respondents and the expected performance. The result show that the people who believe they possess a lot of knowledge are the most overconfident. The difference between actual performance and expected performance is the greatest among students who rate themselves with a four, six or seven.
The Pearson correlation between football knowledge and expected number of correct predicted games differs between the two surveys. The correlation in the experimental survey is .326 and the correlation in the control survey is .279. However, if we split the sample into these two groups, the correlation is no longer significant.
5.1.3.1 Controlling for the effect of continuous variable
We have seen that there is a positive relationship (.315) between football knowledge and expected number of correct predicted games. Because the average football knowledge between the control and experimental group differs (5.18 versus 4.66 respectively) the effect of variable football knowledge needs to be controlled for. The average expected number of correct predicted games was 14,94 for the control group versus 14.55 for the experimental group but these values could be influenced by the difference in football knowledge between both groups. Therefore, I executed a Pearson correlation and had a look whether the variable group influenced the means of the expected number of correct
predicted games. The coefficient B from football knowledge only changed with .005 and therefore the different level of football knowledge between the two groups has no impact.
5.1.4 Self-assessment and relative performance
We have already seen that football knowledge has a positive relationship with the variable expected number of correct predicted games. The more football knowledge one thinks to have the higher the expected number of correct predicted games. However, this is not the only positive relationship the variable football knowledge has. There is also a positive relationship between football knowledge and the expected relative position. The variable expected ranking was divided into four options (I am among the worst students (1), second worst (2), second best (3) or best (4)). I executed a linear regression to find a R2 of 39.5 % (significance .000) and a Pearson correlation test of .629
(significance .000). This means there is a strong positive relationship between football knowledge and expected ranking. The more self-assessed football knowledge the higher the expected ranking.
5.2 Certainty
5.2.1 Average Certainty
After every predicted game, the participants needed to give an indication of how sure they were they had given the right answer to that specific match. The answers were given on a 30 percent to 100 percent scale. On average the sixty-three respondents, were 59,7 percent (SD 6,4) sure they gave the right answer. If we compare both groups, the control group was slightly more confident than the
Figure 8.2
Figure 9.1
Figure 9.2
Because of the big standard deviation, the difference between the two means is not significant. This was confirmed by the outcome of the Welch test; .452. These results do not support hypothesis 3:
Forewarned students will be less sure with their predictions since the difference between the two
groups is not significant.
Not only does the variable football knowledge have a significant correlation with the variables; relative score and expected number of correct predicted games, but also with average certainty. This relationship has a Pearson correlation of .285 (significance .023) and a R2 of .081. If we split up the sample into the two groups, we see that the correlation and the R2 stay around the same values. However, the results of the separate group are not significant.
5.2.2 Certainty for each game
The average certainty of all matches may not be the most interesting result. Because, we can not see the patterns of how both groups behaved during the whole survey and it is also not possible to spot differences between matches. Figure 10, shows all the discrepancies and similarities in cerntainty between the two groups for each game. Prior to the tournament, France, Germany and Spain were the favourites to win the Euros. If we have a closer look on the matches of those three teams we see that the sureness of these matches are the highest. The games of France (FRA) and Spain (ESP) have around 80 percent of sureness while Germany (GER) reaches almost 90 percent certainty in their game versus Ukraine.
For all the twenty-four games, I executed an independent T-test (equal variances not assumed) to see whether the differences between certainty of both groups were significant. Only four games were close to being significant. These were the following matches: Spain versus Czech Republic (.121), Germany
versus Ukraine (.121), Ukraine versus Northern-Ireland (.067) and Germany versus Poland (.064). The remaining eighteen games had a significance value of over .2 (P > 0.2). It is interesting to see that both Germany and Ukraine are involved twice which indicates that the two groups have different ideas about the strengths of those teams. However, I do not have evidence to support this statement.
5.3 Percentage of correct responses
Figure 11 represents the percentage of respondents who predicted the right outcome for each game. The matches with the highest correctness rate are won by the teams with the the most status (like France, Germany, Spain, England and Belgium). Also, the first match from Switzerland and Poland were predicted quite good but this is because they played unknown teams such as Albania and Northern-Ireland. In the second games of Switzerland and Poland, the correctness rates dropped. For games who ended up in a draw (eight out of twenty-four = 33.3%), a smaller percentage of students predicted the right outcome. Except for the games Romania – Switzerland and Czech Republic – Croatia (those games are predicted correctly around 32 percent and 28 percent respectively).
Figure 11
5.3.1 Hard versus easy questions
In section 2.1.5 the difference in overconfidence between a hard and an easy question was discussed and in that section I cited a research by Kleymann et al (1999). Those researchers argue that
overconfidence disappears when people perform a simple task and overconfidence increases as the tasks get more difficult. The results from Figure 12 confirms their thoughts.
Game Hard/Eas y
Correctness rate Confidence rate Underconfident or overconfident?
France - Romania Easy 98,4% 76,4% Underconfident
Germany - Ukraine Easy 100% 87,0% Underconfident
England – Russia Hard 17,5% 65,6% Overconfident
Portugal – Iceland Hard 11,1% 69,4% Overconfident
5.4 Overplacement
Besides the fact that people overestimate their own performance and skills they also overestimate their performance relative to others. This is called overplacement and was already discussed in section 2.1.2. At the end of the survey, I asked the participants whether they thought they would end up being one of the worst performing students (1), second worst performing students (2), second best
performing students (3) or best performing students (4).
The results of figure 13.1 shows that on average, all the respondents believe they belong to the second best performing group (in both the experimental and control survey). These results do not support hypothesis 4: Forewarned people will rank themselves in a lower quartile.
However, the students in the experimental and control group were distributed differently. The respondents in the control group were quite similar since 2/3 of the people chose for option 3, while the respondents in the experimental group were more divided. In the experimental group, a bigger percentage of people thought they would end up being one of the best (option 4).
Figure 14.1 shows the average distance between actual performance and expected performance. For the worst participants (first quartile, 0-25%), the mean difference between their expectations and actual ranking was 2.00 (SD ,843). This means they expected themselves to be in the third quartile, two quartiles higher than they actually are, making them the most overconfident people. The second worst group (second quartile, 25-50%) estimated on average that they would also be in the third best group, however their mean difference is only 1.06 (SD .680) making them less
overconfident then the worst participants. The second best participants (third quartile, 50-75%) know
Figure 12
exactly how well they’ve performed. Their mean difference is only .06, which indicates almost no difference and therefore no overconfidence. The best performing students (top quartile, 75-100%) have a mean difference of -0.106 (SD.772) which means they thought they would be in a lower quartile then they actually are making them underconfident.
5.4.1 Overplacement and expected performance
The results from figure 14.1 and 14.2 confirm what researchers Dunning et al. (2003) already argued in a previous study. The argued that the worst students are the most overconfident while the best students are underconfident. My results strengthen these findings. However, I do not find them that valuable. It seems to be the case that every student (in this experiment and in the research from dunning et al), thinks he or she is scoring above average (third quartile, 50-75%). So, isn’t it just coincidence that the people in the third best quartile actually are in that quartile? And will every quartile on average thinks they performed above average?
However, it is interesting to have a look at the absolute expected values of the different quartiles to see who is overconfident or not. If we have a look at figure 14 we see that the student who actually performed the worst, had the highest expectations of their performance. The worst two quartiles believed they would predict (more then) 15 games correctly while the best performing students believe they would score less (13,9 games and 14.6). These results do confirm that the worst students are the most overconfident. Not only do they actually perform the worst, they also have the highest expectations of their performance. Contradictory to the study of Dunning et al. the best students are not underconfident as they suggested. Since they are also overconfident. But just a little bit. (The numbers on the X-axis of the graph represent the four quartiles).
0 1 2 3 4 Worst
participants Second worst participants Second best participants Best participants
Actual ranking versus expected ranking
Actual ranking Expected ranking
5.5 Demographics
5.5.1 Age
The average age of the students was 21.94 years with the youngest participants being nineteen years old and the oldest twenty-seven years old. I examined whether age is a moderator for the variables; number of correct predicted games (R2 .028), self-assessed football knowledge (R2 .005) and expected number of correct predicted games (R2 .001). For all these variables, age had no significant impact and a small R2.
5.5.2 Men versus women
As already been said in section 3.1, fifty-three participants were male and ten were female. Although this research is not focussed on gender differences, I performed the same analyses (as I did for the variable age) to tests the differences between men and women. On average the male students
performed slightly better then the female students. 11.74 (SD 1.53) and 11.4 (SD 1.51), respectively. But these results are not significant due to a combination of a big standard deviation and a small difference. Furthermore, male respondents believed they have more football knowledge rating themselves with a 5.1 (SD 1.2) on average comparing to the 4.0 (SD .9) women gave themselves. I tested this difference with an independent t-test (equal variances not assumed) resulting in a .005 significance. Finally, the male respondents expected to predict 14.79 (SD 3.48) games correctly and their female counterparts 14.60 (SD 2.63). To conclude, the only gender difference is that men rate their knowledge higher than women do but this does not result in a significantly better performance.
5.6 Draws
During the first twenty-four games of the Euros, eight matches ended up in a draw (33.3 percent). This is above the final tournament average (29 percent) but not surprisingly because this European
Championship had the least number of goals per match ever. The respondents in my study thought on average there would be 4.6 draws in the first 24 matches (19.2 percent). If we compare the two surveys, the control group expected 4.8 (20 percent) draws versus the 4.4 (18.3 percent) the
experimental group expected. Prior to this experiment, I expected that football knowledge would lead to a higher number of predicted draws. The Pearson correlation in the control group is .039 which is very low. However, for the experimental group, the correlation between knowledge and number of expected draws was .339 (significance .067). The difference between the correlation of the two groups is remarkable but I can not find a reasonable explanation for it. It could be that the intervention made the students with a lot of knowledge choose for a draw more often then the students without the intervention. But, this seems farfetched. Because the results were not significant. Hypothesis 7: The
more football knowledge someone has, the higher the number of predicted draws will not be
accepted.
6. Discussion
6.1 Limitations and Remarks
The first remark on this research is that the sample sizes of both the experimental and control group are not sufficient enough. When I executed tests for two groups together, most of the results were significant. However, when I did the same tests for both groups separately, most results became insignificant. If both groups were bigger and more respondents filled in the survey, the correlations and R2’s of both groups independently would also be significant.
The second remark is concerning the self-assessment of football knowledge. The participants ranked their football knowledge on a one to seven scale but I do not know whether theses results are reliable. Someone could easily rank his knowledge with a five while in fact he knows less than someone else who rated his knowledge with a four. If I had asked the participants some questions and then ranked them based on the answers they had given, then that would be a better measure for someone’s football knowledge. In this experiment, there is no relationship between knowledge and actual performance but that could change if people would have been ranked based on actual knowledge. Nevertheless, the variable football knowledge may not be representing someone’s real knowledge but it is a good indicator for the expected value of correct predicted games.
The third remark is that the intervention in the experiment should have been designed differently. Like this, the awareness did not have a significant impact on almost all the variables. Possible explanations are: it could be the case that the text was too long. The introduction regarding
English while the most of the participants are Dutch. However, I expect that my respondents are able to understand the English language. Thirdly, it might be the case that the text was too boring and did not capture the imagination of the respondents. The text was purely informative and did not have interesting examples. So, respondents might just have read it without any interest. A fourth
explanations is that it could also be the case that I was not persuasive enough. I let them read the text but I was not really trying to convince them to be more cautious and aware of the overconfidence bias. Personally, I think awareness could have bigger influence on the calibration of sports bettors if it was designed differently.
6.2 Additional research
One of the other things that could be interesting to further investigate is whether people always think they are performing above average or not. On average the respondents in my experiment and in the experiment from Dunning et al. believed they performed above average. Personally, I believe that the people who actually perform above average and predicted their relative score correctly are just lucky they are performing above average. Especially with gambling, if two games had gone differently, other people would have predicted their rank correctly. On the other hand, you could argue this for everything. If two questions on an exam would have been different, the worst performers could suddenly be not that bad anymore and others would have been the worst.
The always above average effect could be tested by doing an experiment in which the participants need to do a lot of different tasks and let them give an indication of their relative performance for each task.
It could also be interesting to further investigate the differences between men and women. This research was not focussed on the discrepancies between men and women but the results showed some interesting facts. I found that the male respondents believe they have more football knowledge while in fact men did not perform significantly better than women. However, the sample size of women was quite small. The difference in self-assessed knowledge may solely exist because this experiment is about football but it could also be the case that man believe they have more knowledge then women on more occasions. For the whole sample, the intervention in the experiment did not have a significant influence on people’s calibration. However, it might be though that awareness has an impact on women but due to the small sample that was not visible.
7. Conclusion
Overconfidence is primary problem amongst sports bettors. Therefore, I wanted to see whether I could prevent gamblers from being overconfident. I wanted to see to what extent it is possible to influence someone’s overconfidence with sports betting with awareness? The results from my experiment show that people are overconfident when it comes to predicting soccer matches. On average, people believe they predict three more games correctly then they actually do and specially the worst performing
individuals are the most overconfident. These individuals have the highest expectations and the worst score. The same applies to the group of student with the highest self-rated football knowledge. They are overconfident too. There were also positive relationships between football knowledge and expected relative performance, expected number of correct predicted games and certainty.
Nevertheless, awareness did not have a significant impact on the calibration of the sports bettors. The forewarned people were slightly less confident, certain and more cautious but the results were
8. References
Bradley, G. (1978). Self-serving biases in the attribution process: A reexamination of the fact or fiction question. Journal of Personality and Social Psychology, 36(1), pp.56-71.
Burns, B. D., & Corpus, B. (2004). Randomness and inductions from streaks:“Gambler’s fallacy” versus” hot hand “. Psychonomic Bulletin & Review, 11(1), 179-184.
Cantinotti, M., Ladouceur, R. and Jacques, C. (2004). Sports Betting: Can Gamblers Beat Randomness?. Psychology of Addictive Behaviors, 18(2), pp.143-147.
Clotfelter, C. and Cook, P. (1993). Notes: The “Gambler's Fallacy” in Lottery Play. Management Science, 39(12), pp.1521-1525.
Dixon, M. J., & Coles, S. G. (1997). Modelling association football scores and inefficiencies in the football betting market. Journal of the Royal Statistical Society: Series C (Applied Statistics), 46(2), 265-280.
Dunning, D., Johnson, K., Ehrlinger, J. and Kruger, J. (2003). Why people fail to recognize their own incompetence. Current Directions in Psychological Science, 12(3), pp.83-87.
Haran, U., Moore, D. A., & Morewedge, C. K. (2010). A simple remedy for overprecision in judgment. Judgment and Decision Making, 5(7), 467.
Harré, N., Foster, S. and O'Neill, M. (2005). Self-enhancement, crash-risk optimism and the impact of safety advertisements on young drivers. British Journal of Psychology, 96(2), pp.215-230.
Harré, N. and Sibley, C. (2007). Explicit and implicit self-enhancement biases in drivers and their relationship to driving violations and crash-risk optimism. Accident Analysis & Prevention, 39(6), pp.1155-1161.
Hvide, H. (2002). Pragmatic beliefs and overconfidence. Journal of Economic Behavior & Organization, 48(1), pp.15-28.
Job, R. (1990). The application of learning theory to driving confidence: The effect of age and the impact of random breath testing. Accident Analysis & Prevention, 22(2), pp.97-107.
Klayman, J., Soll, J., González-Vallejo, C. and Barlas, S. (1999). Overconfidence: It Depends on How, What, and Whom You Ask. Organizational Behavior and Human Decision Processes, 79(3), pp.216-247.
Kruger, J. (1999). Lake Wobegon be gone! The "below-average effect" and the egocentric nature of comparative ability judgments. Journal of Personality and Social Psychology, 77(2), pp.221-232.
Kruger, J. and 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), pp.1121-1134.
Labuhn, A. S., Zimmerman, B. J., & Hasselhorn, M. (2010). Enhancing students’ self-regulation and mathematics performance: The influence of feedback and
self-Langer, E. (1975). The illusion of control. Journal of Personality and Social Psychology, 32(2), pp.311-328.
Moore, D. and Cain, D. (2007). Overconfidence and underconfidence: When and why people underestimate (and overestimate) the competition. Organizational Behavior and Human Decision Processes, 103(2), pp.197-213.
Moore, D. and Healy, P. (2008). The trouble with overconfidence. Psychological Review, 115(2), pp.502-517
Najavits, L. M., Grymala, L. D., & George, B. (2003). Can advertising increase awareness of problem gambling? A statewide survey of impact. Psychology of addictive behaviors, 17(4), 324.
Pallier, G., Wilkinson, R., Danthiir, V., Kleitman, S., Knezevic, G., Stankov, L. and Roberts, R. (2002). The Role of Individual Differences in the Accuracy of Confidence Judgments. The Journal of General Psychology, 129(3), pp.257-299.
Peele, S. (2001). Is Gambling an Addiction Like Drug and Alcohol Addiction?: Developing Realistic and Useful Conceptions of Compulsive Gambling. Journal of Gambling Issues.
Pfeifer, P. E. (1994). Are we overconfident in the belief that probability forecasters are overconfident?. Organizational Behavior and Human Decision Processes, 58(2), 203-213.
Rachlin, H. (1990). Why Do People Gamble And Keep Gambling Despite Heavy Losses?. Psychological Science, 1(5), pp.294-297.
Ramalingam, V., Palaniappan, B., Panchanatham, N. and Palanivel, S. (2006). Measuring
advertisement effectiveness—a neural network approach. Expert Systems with Applications, 31(1), pp.159-163.
Razali, N. M., & Wah, Y. B. (2011). Power comparisons of shapiro-wilk, kolmogorov-smirnov, lilliefors and anderson-darling tests. Journal of statistical modeling and analytics, 2(1), 21-33.
Russo, J. E., & Schoemaker, P. J. (1992). Managing overconfidence. Sloan management review, 33(2), 7.
Scheinkman, J. and Xiong, W. (2003). Overconfidence and Speculative Bubbles. Journal of Political Economy, 111(6), pp.1183-1220.
Sibley, C. and Harré, N. (2009). The impact of different styles of traffic safety advertisement on young drivers’ explicit and implicit self-enhancement biases. Transportation Research Part F: Traffic
Psychology and Behaviour, 12(2), pp.159-167.
Schwarz, B. B. (2009). Argumentation and learning. In Argumentation and education (pp. 91-126). Springer US.
Winman, A., & Juslin, P. (1993). Calibration of sensory and cognitive judgments: Two different accounts. Scandinavian Journal of Psychology, 34(2),
8.1 Sites
Centraal Bureau voor de Statistiek. (2015, August 18). Marktscan online kansspelen 2015. From http://www.kansspelautoriteit.nl
