• No results found

Overconfidence and the self-serving bias : is overconfidence in the poorest performers motivated by self-protection?

N/A
N/A
Protected

Academic year: 2021

Share "Overconfidence and the self-serving bias : is overconfidence in the poorest performers motivated by self-protection?"

Copied!
47
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Overconfidence and the Self-Serving Bias

Is Overconfidence in the Poorest Performers Motivated by Self-Protection?

Jelle Hin 10590781

University of Amsterdam

Msc Economics: Behavioural Economics & Game Theory 15 ECTS

Abstract

Dunning & Kruger (1999) find that it is the poorest performers in many domains also tend to overestimate their performance the most. These individuals would namely lack a certain meta-cognitive ability, the ability to know whether they are correct. It however remains unclear what causes the upward bias in their beliefs. If these individ-uals are not able to detect their mistakes, what makes them think they are correct? This thesis proposes that it is self-protection mechanisms that underlie such overconfi-dent judgments. Self-protection is one of the two main motivations of the self-serving bias. It is for this reason that a relationship between overconfidence and the self-serving bias could exist. 74 Subjects partook in an online experiment that attempted to find a relationship between both biases. The data however shows no explicit rela-tionship between the two phenomena. This implies that the poorest performers are not motivated by self-protection to overestimate. It should however be noted that such a result could have been caused by methodological limitations.

(2)

Statement of Originality

This document is written by Student Jelle Hin who declares to take full respon-sibility for the contents of this document. I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

Contents

1 Introduction 1

2 Literature Review 2

2.1 Overconfidence . . . 2

2.1.1 Three types of overconfidence . . . 2

2.1.2 Consequences of Overconfidence . . . 3

2.1.3 Dunning Kruger Effect . . . 4

2.1.4 Overconfidence in the Poorest Performers . . . 4

2.1.5 Self-Protection . . . 6

2.2 Self-Serving Bias . . . 7

2.2.1 Motivation of the Self-Serving Bias . . . 8

2.2.2 How to Research the Self-Serving Bias . . . 9

3 Methodology 11 4 Results 13 4.1 Primary Results . . . 13 4.2 Secondary results . . . 17 5 Discussion 20 5.1 Hypothesis 1 . . . 20 5.2 Hypothesis 2 . . . 22 5.3 Corroborating evidence . . . 22 5.4 Quartiles vs. Quantiles . . . 24

5.5 The Use of Different Domains . . . 24

5.6 Hard easy effect . . . 24

(4)

1

Introduction

Making sound decisions in everyday life seems to depend on an individual’s ability to detect one’s mistakes, since such a realization decreases the likelihood of the mistake repeating. Research by Dunning and Kruger (1999) shows, however, that individuals are particularly prone to misevaluating their performance. More specifically, individuals that perform the poorest, in many social and intellectual domains, do not seem to be aware of their mistakes. They tend to be overly confident in assessing their ability.

Confidence in one’s ability is an important element for success (Taylor, 1994; Kan-ter, 2004). Overconfidence however seems to have both positive and negative consequences. On the bright side overconfidence can send positive signals towards others concerning their own ability. This leads to the overconfident individual being more persuasive and it might serve as a self-fulfilling prophecy through which characteristics like dedication and ambi-tion can be increased (Burks et al, 2013; Schardmann and van der Weele, 2017). The other side of the coin is that overconfidence elicits behaviour and judgment based on inaccurate beliefs and thus leads to false assessments, overly positive expectations and hazardous deci-sions (Johnson & Fowler, 2011). Studying the underlying factors that cause overconfidence will lead to a better understanding of this bias and why it affects our behaviour.

The debate as to why it seems to be the poorest performers, that are also the most overconfident is not settled. Dunning and Kruger argue that these individuals suffer a dual-burden. Not only do they lack the cognitive ability to form correct conclusions; they also lack the meta-cognitive ability to detect their mistakes (1999). According to the authors, the cause of this overestimation is the same ignorance that causes them to make mistakes in the first place. Others argue that individuals are bad self-estimators regardless of competence. They argue that the finding by Dunning & Kruger is instead caused by statistical and methodological artifacts such as regression to the mean, the better-than-average effect and the difficulty level of the test (Krueger & Muller, 2002; Burson et al, 2006).

Another explanation involves the protection of the self-image. The biggest over-estimators have the most at stake when confronted with their actual ability. By creating a rosy view of their performance, they might protect themselves from a painful truth. Self-protection is one of the two main motivations of the self-serving bias (Sedikides & Alicke, 2011). It is a bias where individuals tend to view themselves in an overly favourable way by attributing their successes towards their own doing, but their failures towards external factors. Considering such self-protection mechanisms there is reason to believe there is a

(5)

relationship between overestimation of one’s performance and the self-serving bias. Such research will contribute to the literature a deeper understanding of what lies underneath the overestimation in the poorest performers. To the best of my knowledge there has been no studies that relate overconfidence towards the self-serving bias, this thesis aims to fill that gap.

Data gathered by an experiment finds that poor performers, the biggest overesti-mators, do engage in self-protective behaviour by attributing external causalities towards their performance. They are however not more subject to a self-serving bias as compared to better performers, which rejects a relationship between overconfidence and the self-serving bias. Furthermore subjects also are not more accurate in their self-estimation once they have gotten the opportunity to attribute causalities. This as well constitutes evidence against a relationship between overconfidence and the self-serving bias. The data does bring evidence that confirms for both the self-serving bias and the Dunning-Kruger effect (1999).

The setup of this thesis is as follows: First, the existing literature concerning overconfidence, the self-serving bias and the relation between the two is reviewed and discussed. Second, the design of the experiment is explained along with a formulation of the hypotheses. The results will then be reported and discussed in the last section of this thesis.

2

Literature Review

2.1 Overconfidence

Overconfidence is one of the numerous biases in the decision making of human beings. It is the bias where an individual tends to believe that they are better than they actually are. A definition that is often used is that individuals tend to have a propensity to overestimate the accuracy of the given information (Siwar, 2011). This will translate into errors in either judgment or decision making (Johnson & Fowler 2011). Moore & Healy (2008) provide three main types of overconfidence that would influence decision making.

2.1.1 Three types of overconfidence

First there is absolute overestimation: the tendency to overestimate one’s skill level given a certain domain, as well as relative overestimation or overplacement: the overestimation of one’s ability relative to a certain group. Dunning and Kruger (1999), for example,

(6)

found that most people tend to overestimate their ability in many domains, absolutely and relatively.

The second bias is coined the better-than- average effect. It is found that a majority of individuals tend to believe they possess greater ability than their average peer (Alicke & Govorun, 2005). This bias is a robust finding as it is found in many areas. For example: couples tend to rate the quality of their marriage above average (Rusbult et al. 2000). 93% of American drivers and 69% of Swedish drivers estimate themselves to be above the median when asked about their driving skills (Svenson & Ola, 1981). Zuckerman and Jost found that participants rated themselves more popular than a social network analysis showed (2001). Such overestimations are, of course, a statistical impossibility.

The last form of overconfidence is over precision. It concerns the higher than actual precision of an individual’s beliefs. People tend to be overconfident in their answer to questions that have numerical answers to them (Such as: what is the circumference of the earth?). The 90% confidence intervals that that subjects submit are generally too small, since they are incorrect 50% of the time. Such precision is also not necessary since the interval can be as large as the subject wishes. This indicates their overconfidence (Soll & Klayman, 2004).

2.1.2 Consequences of Overconfidence

The literature gives several consequences for overconfidence. Evidence shows that overcon-fidence is present in doctors and lawyers (Tracey et al., 1997, Loftus & Wagenaar, 1988). It is the advice of these agents that people tend to trust in making health related or financial decisions. Overestimation of one’s ability can also cause students to make poor decisions in their academic lifespan and would influence their studying behaviour negatively (Dunlosky & Rawson, 2012). Overconfident students namely will stop studying prematurely, engage in inefficient studying behaviours and will be more likely to make the same mistakes again. (De Bruin et al., 2015).

Johnson & Fowler (2011) in their evolutionary account of overconfidence state that overconfidence leads to incorrect assessments, miscalculation of expectations and haz-ardous situations. In their research they conceptualize how overconfidence could have evolved among other strategies that involve more accurate beliefs. It namely seems coun-terintuitive that behaviour or judgment based on false premises would survive, but since this bias is still present in humans today it must have had some evolutionary benefit. Ac-cording to Taylor and Brown (1994) the individuals with an inflated view of their skills

(7)

are the one’s more likely to achieve success when compared to individuals with more hum-ble self- perceptions. Haselton and Nettle (2006) add that when the benefit of success is relatively high as compared to the costs of trying but failing, having overly inflated self-perceptions is a better strategy than having accurate self-perceptions. There is also evidence that positive illusions has the potential to prolong and improve health (Goleman,1989; Dean & Surtees, 1989). For example Reed et al. (1999) have shown that HIV-positive and AIDS patients that show exaggerated optimism of their condition experienced a slower course of illness and survived longer.

Johnson and Fowler then hypothesize that if such benefits provided net advan-tages in the competition of behavioural strategies, then having an overly positive self-view will survive through natural selection. Their evolutionary model hints that it is possible for overly confident individuals to make right decisions whereas a more accurate self-view can lead to suboptimal ones.

2.1.3 Dunning Kruger Effect

Dunning and Kruger find that individuals tend to be overly confident when evaluating their performance regarding many intellectual and social disciplines. The poorest performers especially tend to overestimate their performance the most (1999). In their original paper the authors presented several studies, varying in discipline, where a subject pool were to make a test, subsequently the subjects were asked in what percentile relative to the subject pool they thought they would rank. The results show that the poorest performers (the individuals whose score was in the lowest quartile) consistently thought they were positioned above the 60th percentile. Such overconfidence was not to be found in better performers, showing that the most confident individuals are not necessarily the ones who should be.

2.1.4 Overconfidence in the Poorest Performers

But what is the reason for the increased overconfidence in the poorest performers, as com-pared to better performers? Dunning and Kruger argue that the ignorance that hinders them to perform well is the same ignorance that prevents them from correctly evaluating their performance. The authors use the following example: in order to construct a gram-matically correct English sentence one must also possess the meta-cognitive skills to spot grammatical mistakes. Without this ability this person will likely make mistakes and will not realize it, thus leading to a bad performance without the person realizing it. To find

(8)

evidence for this notion, the authors manipulate competence by training the subjects in the relevant domain. The findings show that after such training that the bottom perform-ers increased the accuracy of their test performance by 18.6%, a significant result. By increasing competence subjects also show an increase in meta-cognitive skills, implying a relation between the two.

This explanation, however, has been put at odds by certain criticism. Krueger and Mueller (2002) argue that people, regardless of competence, are bad estimators of their relative performance and that the results of Dunning and Kruger are biased by statistical effects. Firstly, poor performers are bound to overestimate since there are more options for over- as compared to underestimation. Similarly, a good performer (for example an individual that places in the 90th percentile) has a higher probability of underestimating its own score since there is more room for underestimation. This effect is called regression to the mean. Secondly, individuals tend to evaluate themselves above average, the above-average-effect. (Alicke et al. 1985). These effects would introduce measurement error and thus a significant portion of overconfidence should disappear.

Erhlinger et al (2008) control for these issues by examining the same subject pool twice. This enables them to control for any measurement error, resulting in a reduction of roughly 5 percentage points of overestimation on the part of the bottom performers. The slight change in overestimation was however an insignificant one and thus they conclude that the Dunning-Kruger explanation still holds.

Moreover, Burson et al. (2006) agree with Krueger and Muller’s proposition that one’s ability to self asses accurately does not differ between top and bottom performers, but instead argue that the Dunning-Kruger finding is caused by methodological artifacts. Their argument relies on the proposition that people overplace themselves when a task is perceived as easy, while they underplace themselves when the task is perceived as hard (Healy & Moore, 2008). Thus, if people’s ability to self-asses is independent of competence, then an easy test will have all participants think they have done well relative to their peers, but only the top performers will be accurate in their self-assessment. Similarly, a test perceived as hard will have all participants be underconfident and the poor performers will be more accurate. The criticism of the authors is that the Dunning Kruger effect is researched by handing the subjects easy tasks. Assuming that all subjects overestimate to the same degree, then the top performers will look the most accurate, while the bottom performers look vastly overconfident. The opposite should then hold in the case of a task that is perceived to be hard. In other words: perception of ability is not related to actual

(9)

skill, estimates of performance rather correlate with task difficulty. Burson et al. find support for this notion by varying the difficulty level of the task given to the subjects.

A claim in this critique should however be discussed. Burson et al. argue that overconfidence is only overrepresented in the poor performers whenever the task is per-ceived as easy. The average performance of the given tasks in Dunning & Kruger (1999) will however show that this is not the case. Namely, the average performance ranged from 49.1% to 66.4% across all studies. Tests that produce an average result that is barely a passing grade cannot be considered and easy test. Still it is the bottom performers that overplace themselves the most.

Considering both critiques; Erhlinger et al. (2011) sought to find the Dunning Kruger effect in a real world setting. Burson et al. namely picked task that lay on the boundaries of the difficulty spectrum but were not necessarily tasks that occur naturally. Results in general will have improved external validity when they are researched in a con-text that is more naturally occurring instead of tasks that people have little experience with and are imposed by an experimenter. Erhlinger et al find support for the Dunning Kruger explanation in a more natural setting by finding that is again the poor perform-ers that overestimate disproportionally in a class exam and a college debate tournament relative to better performers.

2.1.5 Self-Protection

Another potential explanation for the inaccurate self-evaluations of poor performers is that they overestimate in order to keep a positive self-image. Poor performers might have an incentive to overestimate their performance because it is this group that will be hurt the most if confronted with their actual score. In support of this idea: it has been shown that individuals that are known to avoid negative information also show significantly more overconfidence (Ehrlinger et al., 2016). Research by Alicke (1985) also shows that college students describe themselves with more desirable characteristics as compared to the average college student and that this confidence relates to providing a positive self-concept. Lastly, there is evidence that children inaccurately self-evaluate because of self-protective motivations (van Loon et al., 2016).

Erhlinger et al (2008) sought to rule out other explanations regarding the over-confidence found in the poorest performers. One of which being the explanation that the poorest performers overestimate to keep a positive self-view. The authors replicate the method employed by Dunning and Kruger (1999) but manipulate the experiment by

(10)

in-centivizing subjects with a monetary payoff may they correctly estimate their performance. The rationale being that if poor performing individuals willingly overestimate to keep a positive self-view, they might be persuaded to more accurately evaluate their performance if properly incentivized. Erhlinger et al. found no significant increase in self- assessment accuracy using this method and thus conclude that individuals do not overestimate to keep a positive self-view.

There is however reason to believe this method is not the appropriate way to test the said hypotheses. The method of Erhlinger et al. rather tests whether poor performing subjects intentionally overestimate their performance. While this turns out not to be the case, the self-protective explanation cannot yet be rejected. Self-deception namely occurs on the conscious and the subconscious level (Von Hippel & Triver, 2011). It is still plausible that these individuals overestimate themselves motivated by self-protection, but do so subconsciously. Note that such an explanation does not conflict with the Dunning-Kruger explanation of overconfidence in the poor performers: if individuals sub consciously deceive themselves into believing they are performing well, they are by definition not aware of their mistakes. What the explanation of Dunning and Kruger however lacks is the explanation for the upward bias: if poor performers are not aware of their mistakes, what makes them think they performed well instead of performing badly? What makes them overconfident? The motivation for such behaviour is potentially protecting one’s self-esteem.

2.2 Self-Serving Bias

A method of measuring self-esteem protection behaviour lies within the self-serving bias. It is the finding that individuals tend to attribute their successes towards internal factors (such as ability or effort) but their failures towards external factors (such as luck) (Miller & Ross, 1975). Individuals thus tend to rationalize an outcome in a way that it perceived favourably. For example an individual that is playing poker and wins a pot is likely to attribute its win towards its bluffing ability or profound strategy. However when a player loses a pot, the player will be inclined to blame its hand or an unfortunate flop. Similarly a student will refer to its dedication and intelligence when receiving a good grade, however when a bad grade is received the student will blame the abnormal difficulty of the test or the harsh grader. By attributing negative outcomes towards external causalities, the individual’s self-esteem remains protected, whereas internal attributions for successes will enhance one’s self-esteem.

(11)

2.2.1 Motivation of the Self-Serving Bias

The self-serving bias can be explained by two approaches. The cognitive approach intro-duced by Miller and Ross (1975) entails that individuals attribute causalities in a self-serving way whenever the outcome is not consistent with their expectations. Evidence for this notion lies in the studies of Feather (1968) which finds that individuals attribute exter-nal causalities for outcomes they did not expect, regardless of its valence. The motivatioexter-nal approach, on the other hand, suggests that individuals will assign causalities for a certain outcome in a way that maintains their self-esteem (Zuckerman, 1979). Zuckerman argues against the cognitive approach by citing Bradley’s work, which shows that individuals do not always expect success (1978). Moreover, for some attributional factors (such as effort) it is instead intention rather than expectations that drives the attributional bias in people (Ross et al., 1974; Sicoly & Ross, 1977).

Empirical evidence does suggest that self-esteem protection is indeed a modulator of the self-serving bias. Studies that compare individual’s level of self-esteem with their susceptibility towards the self-serving bias find that in the case of negative outcomes, high self-esteem individuals tend to be more subject to the self-serving bias when compared to low self-esteem individuals (Blaine & Crocker, 1993). Similarly high self-esteem individu-als ascribe the failures towards externindividu-als factors more so as compared to low self-esteem individuals. (Schlenker et al.,1976; Seligman et al., 1979). Research by Strube (1985) that compares attributional styles between Type A and Type B personalities, Type A being on average more prestigious, find that Type A individuals are more subject to a self-serving bias.

Moreover, according to Sedikides and Alicke (2012); when considering self-protection, individuals tend to act defensively when they feel threatened. They argue that when these individuals get the opportunity to assign the causality to external factors, they will do so. In that case the level of threat should correlate with the level of SSB. Studies by Camp-bell and Sedikides (1999) sought to find whether the self-serving bias is indeed modulated by self-threat. Support for this claim was found; the higher the level of threat a person experiences, the more subject they are to the self-serving bias.

In the context of this thesis, the threat is the poor performers’ confrontation with their low score. Their actual competence in the tested domain may not comply with how they see themselves. Following the logic of the previous argument, these poor performers will become defensive and deflect the dissonance in their self-view towards external factors, elements outside of their control, so that their self-esteem remains unharmed. As argued

(12)

previously, the higher the level of threat, the higher the level of self-serving bias. So it will be interesting to find whether the poor performers also are subject to self-protection, more so than better performers.

2.2.2 How to Research the Self-Serving Bias

The self-serving bias is mainly researched within two paradigms: the interpersonal influence method and the skill-oriented method. The interpersonal method entails that a subject tries to change another’s behaviour, after which the subject (the one that attempts to influence the other) will indicate in how far the observed change is due to the subject’s influence (Arkin, Cooper & Kolditz, 1980).

More relevant to this thesis, however, is the skill-oriented paradigm. It is an experimental design where subjects perform an experimental task but are given false feed-back on their performance. This paradigm has subjects perform a task after which they receive a randomly generated false outcome rating. Subsequently these subjects can then rate the influence of internal (Effort & Ability) or external (Luck & Difficulty) factors on their performance. Evidence for a self-serving bias is found when internal factors receive a higher causality rating when considering successes over failures. (Miller, 1976).

Regarding this thesis, there is a methodological constraint that needs to be ad-dressed. The faculty of Economics & Business advice against the use of deception in experiments related to the University of Amsterdam since this could make subjects expect they are being deceived in future UvA experiments. This would negatively affect the valid-ity its future research. Giving subjects false feedback on a task is unfortunately considered to be deception.

A study done by Reifenberg (1986) offers an alternative: students are divided into two different success/failure groups. The subjective group performance (success or fail) is determined by the self-reported outcome of the students. The objective group performance is determined by their actual performance. Results show that in both groups the students give external attributions for failures and internal attributions for successes. Furthermore a significant positive correlation between both measures of performance shows that both methods are suitable ways of discerning between successes and failures.

Not being able to give random false feedback signal brings forth another method-ological constraint. A false feedback signal allows finding external as well as internal attri-butions regardless of performance, a poor performer could for example randomly receive a success signal while a good performer could receive a failure signal. Without a random

(13)

false feedback signal, attributions are solely measured in one direction. A poor performer might attribute its poor performance to external factors once confronted with its actual score, if the subject however receives a false feedback of a good performance the subject might attribute externally in the same way. Such a subject would then not be attributing causalities in a self-serving way.

A workaround for this limitation would involve an additional treatment. The treatment would have subjects not receive their scores. The attributional scores of these subjects are then used as a default, and are subsequently compared to the attributional scores in the treatment 0 (where subjects get their scores) across quartiles of performance. If, for example, the poor performers (the lowest quartile) are significantly more self-serving in their attributions when comparing treatment 1 to treatment 2, there is evidence for self-protection in the poor performers. Subjects that do not receive their scores namely are not as aware of their performance and thus have less reason to attribute causalities in a self-serving manner. The difference in attributional scores across treatments can be seen as the amount of self-protection/self-enhancement a quartile of performers exhibit. If these levels are then compared across quartiles of performance we can see if poor performers show more a higher level of self-serving bias as compared to better performers.

Considering the said limitations of steering clear from using deception, another manipulation that seeks for a relationship between overconfidence and the self-serving bias seems appropriate. A study by Xiao & Houser (2005) researches the effect of express-ing emotions on ultimatum game results.1 The authors argue that from an evolutionary

standpoint we have been programmed to prefer expressing our negative emotions towards a proposer, may the proposed split be an unfair one. The authors argue that since the stan-dard way the ultimatum game is performed involves no direct opportunity to vent one’s emotions, it causes an increased likelihood of rejections by responders. As such a rejection is one way of expressing one’s negative emotion. To test this notion the authors added a treatment with the option for responders to vent their negative emotion in the form of several statements that would be sent towards the proposer. As a result, this treatment had significantly less rejections as compared to the treatment without the option to send a statement.

A similar method could be applied for the research question of this thesis. If subjects have a chance to attribute causalities for their performance, they might likewise

1In an ultimatum game a proposer starts with an endowment x and decides how much of x to hand to

the responder. The responder can either accept the proposed split so that both subjects get this payoff or the responder can reject the proposed split so that neither player gets any payoff

(14)

be more precise in their estimation of the quality of their performance.

3

Methodology

In order to test both hypotheses an online experiment is set up through qualtrics.2 Sub-jects will partake in a 13 item multiple choice task that will test their spatial-visualization abilities. Subsequently, they will be asked to give percentile placements considering their spatial visualization skills in general and their ranking among other participants, along with an absolute measure of how many questions they think they have answered correctly. Participants then receive their absolute scores for the test along with their percentile place-ments. This percentile placement is based on how well the participant compares to a subject pool that has performed the test beforehand. In the last part participants rate causalities (attributional scores) for their performance by choosing a number between 1 and 7 that reflects in how far these factors have influenced their performance. They are asked ‘To what extent do you think luck has affected your performance on the spatial visualization test?’ and ‘How difficult did you find the spatial visualization test?’ to measure their exter-nal attributions (Luck & Difficulty) considering their performance, whereas the questions: ‘To what degree do you think this test reflects your true degree of competency in spatial visualization?’ and ‘How hard did you try on the spatial visualization test?’ represent the internal attributions (Ability & Effort) that are used to explain their performance. As an overarching measure, the subjects are also asked: ‘In how far do you feel you were responsible for your own score?’. Lastly this part also involves a text box where subjects can type how they feel about the test.

In order to measure self-serving behaviour across quartiles of performance a treat-ment is be added where subjects will go through the same procedure but will not receive their test scores nor their relative percentile rankings. Subjects in the treatment namely have less reason to attribute causalities self-servingly since they have no precise informa-tion about their performance. Attribuinforma-tional scores of both groups will then be compared, across quartiles of performance, in order to see whether the poor performers (the biggest overestimators) behave more self-servingly as compared to better performers. This data will find evidence rejecting or supporting hypothesis 1:

• Hypothesis 1: The poorest performers (the most overconfident) will be more subject to a self-serving bias as compared to the individuals that perform better (are less overconfident).

2

(15)

A reason to believe such a hypothesis would hold is that it is the poorest performers that have most to lose when confronted with their actual scores and thus have more reason to attribute causalities in a more self-serving way.

The same treatment also involves an altering of the order of the experiment. The part where subjects give attributional scores is switched with the part where subjects estimate their performance. Thus, after the 13 item test, participants will engage in at-tributing causalities and afterwards estimate their performance. Note that the attribution part of the experiment has a text box where subjects can express how they felt about the test. This additional manipulation will measure whether subjects are more accurate in their estimation of their performance when they have had the chance to attribute causal-ities considering the test. Comparing estimation scores across treatments will either find supporting or rejecting evidence for hypothesis 2:

• Hypothesis 2: Individuals will show a significant decrease in overconfidence when they have had the chance to show their feelings considering a task.

Research on the ultimatum game namely shows that expressing one’s negative emotions in such a game gives way for more rational behaviour (Xiao & Houser, 2005). Similarly expressing one’s feelings about a test by attributing causalities for one’s perfor-mance might make a subject more honest self-evaluator.

In order to ensure effortful participation subjects are told that two individuals will receive a payment for this experiment. The first subject will be randomly selected out of the subject pool and his/her payment will depend on how many questions the subject answered correctly. The monetary payoff is the amount of correct answers multiplied by two in euros. The second subject will win a monetary payoff of 10 euros may he/she be the most accurate in estimating one’s own performance. This will only count for the relative estimate that compares the subject to a subject pool that has performed the test before as well as the absolute estimate of how many questions the subject answered correctly. The subject with the lowest sum of deviations of their actual performance will win this monetary payoff.

Attributional means, as well as self-estimate means, will be compared by using Student’s paired t tests with equal variances. The histogram of the test scores shows a bell shaped curve and the aggregate data has a Skewness of -0.527 and a Kurtosis of 2.78, which is close to 0 and 3 respectively.3 This implies a normal distribution of the sample.

Furthermore the sampling is done as randomly as possible by spreading the experiment

(16)

links through social media networks.

4

Results

A total of 74 subjects have participated in the online experiment and 22 subjects have made the spatial-visualization test beforehand (in order to give the subjects a relative ranking). The following section will present the results of the experiment.

4.1 Primary Results

Self-Serving Bias Across Quartiles of Performance

Each attributional factor (Luck, Ability, Effort, Difficulty & Responsibility) is compared between treatments for quartile one (the worst performers) and quartile four (the best performers). Chart 1.1 contains the mean attributional scores for the first quartile of performance. Paired t-tests show that there exists no significant difference in mean for all 5 attributional factors when comparing treatments. Poor performing subjects that have received their test scores thus do not attribute causalities differently compared to poor performers that do not receive their test scores.

As can be seen in chart 1.2, performers in quartile 4 attribute causalities significantly different across treatments only for the ‘Difficulty’ factor. Participants that received their test scores gave a significantly higher difficulty rating (M = 5.666667, SD = 0.7071068) as compared to the participants that did not receive their scores (M = 4.666667, SD = 1.013794); t(8) = 3.4641, p < 0.01. This difference however does not express itself in a self-serving way.

(17)

Both performance groups show no higher level of self-serving attributions when compared to their non-score receiving counterparts, thus no evidence is found supporting hypothesis 1. These results thus are not indicative of an increased level of subjectivity towards the self-serving bias in comparison to better performers.

Self-Serving Bias Across Quartiles of Overconfidence

Another way of searching for evidence for hypothesis 1 is partitioning participants into quartiles of overconfidence. There seems to be a clear negative correlation between per-formance and relative overconfidence (-0.71) which indicates that it is indeed the poor performers that tend to overestimate their relative performance. Dividing the participants into quartiles of overconfidence rather than performance might corroborate the rejection of hypothesis 1.

(18)

biggest over estimators. Paired t-tests give no significant difference for all attributional factors leading to a similar conclusion as in section 1.1, participants do not attribute causalities for their performance differently may they know, or not know, the quality of their performance. Regarding the lowest overestimators (this group contains solely

underestimators) in quartile 4, there are two attributional factors that are significantly different across treatments. As in section 1.1 the underestimating participants that received their scores found the test more difficult (M = 5.67, SD = 0.7071068) compared to the underestimating subjects that did not (M = 4.67, SD = .8660254); t(8) = 2.8, p < 0.01. Furthermore the overarching factor ‘Responsibility’ has received a significantly higher score (M= 6.22, SD= 1.936492) for underestimating participants that received their scores when compared to participants that did not (M = 5.44, SD = 1.013794); t(0.08) = 2.4, p < 0.05.

Difference in confidence across treatments

The treatment also allows for comparing whether subjects estimate their performance more accurately may they have the opportunity to express their attributions towards their performance first. For all three measures of confidence used (ranking relative to the general public, ranking relative to other participants and absolute ranking of cor-rect answers), the treatment group submitted lower estimates. As can be seen in ta-ble 1 & 2, Score receivers ranked themselves in higher quartiles relative to the gen-eral public (GP) (M = 65.51351, SD = 2.837602) and relative to other task perform-ers (OP) (M = 61.75676, SD = 3.209035) as compared to subjects that did not receive such information (M = 61.86486, SD = 3.462836) (GP) (M = 57.67568, SD = 3.4172) (OP). Likewise, tables 3 & 4 shows that the absolute estimate of performance was also higher for score receivers (M = 8.702703, SD = .3833331) than for non-score receivers

(19)

(M = 8.27027, SD = 0.4611786). None of these differences are however deemed statisti-cally significant by a paired t-test, as expressed by charts: 2.1, 2.2, 2.3.4

Difference in overconfidence across treatments

Subtracting the subject’s actual percentile placements from their self-assessed percentile placements gives a measure of relative overconfidence. Similarly subtracting their actual scores from their perceived scores will give a measure of absolute overconfidence. As can be seen in tables 1-4, both measures of overconfidence show a lower level of overconfidence in the group that has received their scores as compared to the group that did not. For relative overestimation, score receivers estimate themselves in the 62nd percentile on av-erage, while their actual average percentile placement is in the 49th percentile this gives an average overestimation of 13 percentile points. Non-score receivers on the other hand ranked themselves on average in the 58th percentile while they actually ranked within the 41st percentile, an overestimation score of 17 percentile points. Considering absolute overestimation, score receivers estimated their score to be 8.7/13 on average which is lower than their actual average score of 9.32/13. Score receivers are under confident in general. Non-score receivers estimate their absolute score to be 8.27/13 but actually score 8.73/13. Here also some slight underestimation occurs.

Table 1: Relative Overconfidence for Score Receivers

Quartile Relative Estimate (GP) Relative

Estimate (OP) Actual Percentile Relative Overconfidence

1 56.8 45.2 11.46 33.74

2 62.11 63.78 40.86 22.92

3 75.22 72.11 63.23 8.88

4 68.89 67.78 83.53 -15.76

Total 65.51 61.76 48.74 13.02

Table 2: Relative Overconfidence for Non-Score Receivers

Quartile Relative Estimate (GP) Relative

Estimate(OP) Actual Percentile Relative Overconfidence

1 48.6 48.3 11.19 37.11 2 58.67 50.56 26.24 24.32 3 69.44 65 47.86 17.14 4 72.22 67.89 80.18 -12.29 Total 61.86 57.68 41.37 16.31 4

(20)

Table 3: Absolute Overconfidence for Score-Receivers

Quartile Mean Actual Score Mean Score Estimate Absolute Overconfidence

1 5.4 7 1.6

2 9.22 9 -0.22

3 10.89 9 -1.56

4 12.22 9.9 -2.56

Total 9.32 8.7 -0.62

Table 4: Absolute Overconfidence for Non-Score Receivers

Quartile Mean Actual Score Mean Score Estimate Absolute Overconfidence

1 5.4 6 0.6 2 8.22 7.78 -0.44 3 9.67 9.22 -0.44 4 12 10.33 -1.67 Total 8.73 8.27 -0.46 4.2 Secondary results Self-Serving Bias

Evidence is found for a self-serving bias when internal factors (Ability & Effort) are given a higher attributional score considering successes over failures. Likewise, evidence for a self-serving bias arises when external factors (Luck & Difficulty) are attributed for one’s performance when considering failures over successes (Miller, 1976).

By this logic poor performers (quartile 1) should have reported a higher average attributional score for the factors Luck and Difficulty as compared to better performers. Charts 3.1 shows that this holds in all but one case where the best performers in the Score-Receiver treatment a give a higher score towards Difficulty compared to the poorest performers. This factor however does appear to be decreasing when solely considering quartiles 1, 2 and 3.

In addition, the best performers in quartile 4 appear to give higher mean scores towards internal factors Ability & Effort as compared to poorer performers. The overar-ching measure ’Responsibility’ also received a mean score that is increasing in quartiles of performance, showing that poorer performers are less willing to take responsibility for their poor performance. Since failures (poor performances) are attributed externally and successes (best performances) are attributed internally, evidence is reproduced for the self-serving bias in this subject pool.

(21)

Overconfidence

Charts 3.3 and 3.4 show the trend lines of the relative confidence data considering both treatments. Both charts show that quartile 1, 2 and most of quartile 3 report a perceived ranking that is higher than their actual ranking, whereas quartile 4 shows under confidence in both treatments.

Tables 1 & 2 contain the specific estimated and actual relative scores per quartiles of performance. The poorest performers estimate themselves in the 46th (score receivers) and the 49th percentile (non-score receivers) while they actually rank in the 12th per-centile in both treatments. The best performers estimate themselves in the 69th and 73rd percentile, for score and non-score receivers respectively, while their actual ranking is in the 84th (score receivers) and 81st (non-score receivers) percentile. This corroborates the findings of Dunning & Kruger (1999): it is the poorest performers in many domains that overestimate themselves the most. The data shows that it is indeed the poorest performers that overestimate their performance the most, at least when considering relative overcon-fidence.

In the case of absolute overconfidence, tables 3 & 4 indicate under confidence on average. The poorest performers estimate to have answered 7/13 (score receivers) and 6/13 (non-score receivers) questions correctly, while in reality they have answered 5.4/13 questions correctly in both treatments, on average. The best performers instead estimate themselves to have 9.6/13 and 10.333/13 questions correctly, while their actual number of correct answers is 12.22/13 and 12/13, on average. These results show absolute over confidence in the poor performers and absolute under confidence in the best performers.

(22)

per-centile, whereas the non-score receivers estimate themselves in the 56th percentile on av-erage. Since these estimates are higher than the 50th percentile, evidence for better-than-average effect is found as well.

(23)

5

Discussion

5.1 Hypothesis 1

Hypothesis 1 states that the poorest performers will be more subject to a self-serving bias as compared to better performers. The poorest performers did not attribute causalities for their performance differently to the comparison group (that is assumed to have less incentive to attribute self-servingly), whereas the best performers only gave the difficulty rating a higher attributional score compared to the comparison group. Since for both performance groups there is little to no difference in attributions compared to the group that has less incentive to attribute self-servingly, it can be concluded that the poorest performers are not more subject to a self-serving bias as compared to better performers. Segmenting subjects into quartiles of overconfidence confirmed this conclusion as there were little to no differences in comparison with the comparison group.

It should however be noted that this conclusion relies on the assumption that the comparison treatment (the subjects that did not receive their scores) had less of an incentive to attribute causalities self-servingly. Chart 3.2 shows however that non-score receivers attribute causalities for their performance in a self-serving way as well. The poorest performers namely give the highest attributional scores towards the external factors (Luck & Difficulty) while the best performers rank the internal factors (Ability & Effort) higher than their poorer performing peers. Also the overarching responsibility rating is increasing in performance, which is indicative of the self-serving bias.

A plausible explanation for this phenomenon is that non-score receivers were somewhat aware of their performance regardless of whether they received their test scores or not. The test consisted of 13 spatial-visualization items that were rather unambiguous in their answers. If a participant chose their answer, they were likely considerably aware

(24)

of whether they answered it correctly or not. This would give participants a moderately accurate idea of how well they eventually did, leading to these self-serving attributions.

This would also explain why subjects demonstrated relative overestimation on average (13 and 17 percentile points for score and non-score receivers respectively) but in absolute terms the participants were much more accurate (the average difference between actual number of correct answers and perceived number of correct answers were -0.62 and -0.76 for score and non-score receivers respectively). The accurate absolute estimation implies that subjects were quite accurately aware of how many questions they answered correctly, but in relative terms rather underestimated how well other participants would perform.

Another limitation of how this experimental design aims to reject or confirm hypothesis 1 is that self-serving attributions are only measured in one direction. Poor performers receive a negative signal while good performers receive a positive signal. For this reason the attributions will also go in opposite directions. The measure that would describe subjectivity towards the self-serving bias is the difference between the attributional scores between the two treatments. Comparing this measure between the best and worst performers might however not be the best comparison tool since internal attributions might express themselves in a different scale compared to external attributions.

The solution for this issue would involve the use of a false feedback signal. By giving poor performers a positive feedback and good performance a negative feedback it is possible to measure self-serving behaviour the other way around as well. In this case, a treatment where some subjects do not receive their scores is also not necessary. Data will then be gathered on self-protecting (external attributions) as well as self-enhancing (internal attributions) per quartile of performance and in this way subjectivity towards the self-serving bias per quartile can be compared more rigorously.

Alternatively a methodology could be adapted that involves a distorted feedback signal. Participants would then receive their actual results with a certain probability and would otherwise give a false feedback (as described in the previous paragraph). Using this methodology allows for measuring external and internal attributional behaviour regardless of performance as well. More notably it will also not involve any use of deception, since the participants know that their feedback could be false. Such a method would however require a large sample base as the most useful data points will involve the subjects that receive the false feedback signal.

(25)

5.2 Hypothesis 2

Hypothesis 2 states that individuals will show a more accurate self-evaluation once they were given the chance to express their feelings considering the test they made (mainly by attributing causalities towards their performance). The data showed that for each measure of confidence (relative to the general public, relative to other participants and number of correct answers) participants that had the chance to express their feelings towards the test showed lower levels of confidence. These differences were however not significant. Differ-ences between both measures of overconfidence (relative & absolute) between participants that expressed their feelings before self-evaluating and participants that expressed their feelings after self-evaluating show that expressing one’s feelings did not decrease overconfi-dence. Instead participants that expressed their feelings had an average overconfidence of 17 percentage points whereas participants who did not had an average overconfidence of 13 percentage points. This difference is however not statistically significant. These results thus reject hypothesis 2.

A potential limitation that should be discussed is that subject might not have had ample opportunity to express their feelings towards the test. They were given the opportunity to express in how far 4 different attributional factors (Luck, Ability, Effort and Difficulty) influenced their performance and were given the opportunity to share any other opinions they had regarding test in a text box. This text box was however used only 13.5% of the time and the most common remark concerned the high difficulty level of the test. It is possible that if the feeling expression section would have been more extensive that this could have changed the results. On the other hand, it is also possible that participants were on average positive about the test and did not have much to complain about, in which case a significant difference between overconfidence levels across treatments would not appear as well.

5.3 Corroborating evidence

This thesis has provided corroborating evidence for the self-serving bias as well as the Dunning Kruger effect. Poor performing participants namely attributed their performance towards factors outside of their control (Luck & Difficulty) more so than better performers. The good performers on the other hand attributed mainly internal factors (Effort & Ability) more so than worse performers, towards their performance. The overarching measure of responsibility, that measures how responsible subjects felt they were for their performance, also was increasing in performance.

(26)

The Dunning Kruger effect in relative terms is confirmed mainly by charts 3.3 and 3.4. It shows the difference between actual and perceived relative performance, in other words relative overconfidence. As Dunning & Kruger showed in 1999, the poorest performers are also the most overconfident when compared to better performing individ-uals. Charts 3.3 and 3.4 show the same phenomenon; the lowest quartile of performance shows the biggest gap between actual and perceived relative performance.

However much of this relative overconfidence is due to regression to the mean and the mean and the above-average effect, the argument proposed by Krueger & Muller (2002), remains to be seen. The above-average effect is confirmed in the data. Regression to the mean can however not be accounted for. This namely requires the same subject pool to be tested at least twice so that the results can be controlled for measurement error. Evidence by Erhlinger et al (2008) however indicates that if a similar task is performed twice by the same subject pool, the measurement error involved did not cause a significant difference in overconfidence levels.

In absolute terms, overconfidence also decreases when increasing performance. In contrast to the findings of Dunning & Kruger the subject pool shows, on average, no overconfidence in absolute terms. This suggests that people might actually be good estimators of their ability, but instead underestimate the performance of their peers.

It is however also the case that the lack of absolute overconfidence occurs because of the nature of the questions could already give a signal of whether the subject is doing well, a problem that is mentioned before. Further research could attempt to find whether relative overconfidence is caused by an overestimation of the self or instead an underestimation of their peers.

It should also be noted that the lowest quartile of performers in the non-score receiving treatment have attributed causalities self-servingly (see chart 3.2). As argued previously, subjects in all quartiles were somewhat aware of their performance because they report self-serving attributions for their performance. This implies that the poor performers were also somewhat aware of their performance. It is however the case that the Dunning-Kruger argument for overconfidence in the poorest performers relies on the premise that they are not aware of their performance. It would be interesting to see further research explore however much poor performers are aware of how well they are doing since these two findings seem to be in contrast.

(27)

5.4 Quartiles vs. Quantiles

Traditionally research on the Dunning Kruger effect partitions the subjects into quartiles of performance. This might introduce a problem since each treatment in this experiment solely has 34 subjects. Comparing the means of two quartiles then compares groups of 8/9 samples, a relatively low sample size. Therefore an analysis is also conducted comparing the treatments across quantiles rather than quartiles.5 The quantile comparison analysis, that compares groups of 18/19 participants, shows that there exists no significant differences in attributional scores across quantiles of performance. A results that is in agreement with the conclusions of the quartile analysis.

5.5 The Use of Different Domains

A variable that was not controlled for concerns the use of solely one domain namely: spatial-visualization. It is plausible to think that the magnitude of attributions of one’s performance also depend on how relevant the domain is to the individual. A person with poor acting skills and accepting of its incompetence will most likely not feel the need to attribute a bad acting task performance externally. Such a person doesn’t find acting important, so a bad performance will not be seen as a threat to its self-esteem. Domains that are considered to be generally more important, such as logical reasoning or math, likely would entice self-protective behaviour more in the case of a bad performance. Whether people consider spatial-visualization an important skill, could be answered by looking at self-serving bias scores by this logic. It would have however been wise to add an additional question about whether spatial visualization skills are considered important by them, so that this issue could have been somewhat controlled for.

5.6 Hard easy effect

Many participants reported that the test was too hard. Feedback that participants reported after the experiment seemed to confirm this. This could constitute a problem for the results since according to the literature; the difficulty of a task has an effect on overconfidence levels. A task that is too hard would produce absolute overconfidence but relative under confidence (Larrick et al., 2007; Moore & Small, 2007). In this case it would hard to draw conclusions on relative overconfidence if participants would demonstrate primarily relative under confidence. The results of this experiment rather show relative overconfidence but absolute underconfidence, which is indicative of a test that was too easy. As the Histogram

(28)

of test scores show is that approximately 8.85/13 questions were answered correctly on average, which is almost 70%, thus if anything the test turned out to be on the easy side.6

6

Conclusion

All in all, no evidence is found that supports a relationship between overconfidence and the self-serving bias. When comparing subjectivity towards the self-serving bias between the poorest quartile (biggest over estimators) and best performing quartiles (biggest under estimators), the poorest performers are not susceptible more so than the best performers. Furthermore, when subjects are given the chance to attribute causalities (in a self-serving way), they are also not more accurate in their self-evaluations afterwards.

The data produced by the experiment did however bring evidence that corrob-orates previous research. Poor performers attribute external factors towards their poor performance more than subjects that perform better. Similarly good performers attribute internal factors more so than subjects with a lower score. The Dunning Kruger effect was also validated. Poor performing subjects showed more relative and absolute overconfidence as compared to better performers. Note however that the results showed no absolute over confidence in general. Both treatments were rather accurate in guessing their scores. This seems to propose that people estimate themselves accurately, but instead underestimate their peers.

While these findings imply that the self-serving bias and overconfidence are not related and thus that self-protection may not be a motive of over-estimation. It should be noted that the proposition should not yet be rejected. Possibly, by using a false-feedback signal to research the self-serving bias, we would see different results. Such a method should be employed in order to receive more complete data on subject’s attributional behaviour, this would make its conclusions more rigid. Additionally, changing the domain of the task could have altered attributional behaviour as well since some domains are considered less shameful to be bad at than other domains.

Dunning & Kruger explain that the poorest performers overestimate because they lack the meta cognitive ability to realize they are incorrect. But what causes them to think they are correct? While this thesis failed to give a plausible explanation, future research should pursue to give an answer to this question.

6

(29)

References

Alicke, M. D., & Govorun, O. (2005). The better-than-average effect. The self in social judgment, Psychology Press, 1, 85-106.

Alicke, M., & Hogan, Robert. (1985). Global Self-Evaluation as Determined by the De-sirability and Controllability of Trait Adjectives. Journal of Personality and Social Psychology, 49(6), 1621-1630.

Arkin, R., Cooper, H., & Kolditz, T. (1980). A statistical review of the literature concern-ing the self servconcern-ing attribution bias in interpersonal influence situations 1. Journal of Personality, 48(4), 435-448.

Blaine, B., & Crocker, J. (1993). Self-esteem and self-serving biases in reactions to positive and negative events: An integrative review. In Self-esteem (pp. 55-85). Springer, Boston, MA.

Bradley, G. W. (1978). Self-serving biases in the attribution process: A reexamination of the fact or fiction question. Journal of personality and social psychology, 36(1), 56.

Burks, S. V., Carpenter, J. P., Goette, L., & Rustichini, A. (2013). Overconfidence and social signalling. Review of Economic Studies, 80 (3), 949-983.

Burson, K., Larrick, R., Klayman, J., & Devine, Patricia. (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), 60-77.

Campbell, W., Sedikides, C., & Salovey, Peter. (1999). Threat Magnifies the Self-Serving Bias: A Meta-Analytic Integration.Review of General Psychology, 3(1), 23-43.

Dean, C., & Surtees, P. G. (1989). Do psychological factors predict survival in breast cancer?. Journal of Psychosomatic Research, 33(5), 561-569.

Dominic D. P. Johnson, & James H. Fowler. (2011). The evolution of overconfidence. Nature, 477(7364), 317-20.

Dunlosky, & Rawson. (2012). Overconfidence produces underachievement: Inaccurate self evaluations undermine students’ learning and retention. Learning and Instruction,22(4),

(30)

271-280

Ehrlinger, Johnson, Banner, Dunning, & Kruger. (2008). Why the unskilled are unaware: Further explorations of (absent) self-insight among the incompetent. Organizational Behavior and Human Decision Processes,105(1), 98-121.

Ehrlinger, Mitchum, & Dweck. (2016). Understanding overconfidence: Theories of intel-ligence, preferential attention, and distorted self-assessment. Journal of Experimental Social Psychology, 63, 94-100.

Goleman, D. (1987). Who are you kidding. Psychology Today, 21(3), 24.

Feather, N. T. (1968). Change in confidence following success or failure as a predictor of subsequent performance. Journal of Personality and Social Psychology, 9(1), 38.

Haselton, M. G.,& Nettle, D. (2006). The paranoid optimist: An integrative evolutionary model of cognitive biases. Personality and social psychology Review, 10(1), 47-66.

Kanter, R. M. Confidence: How Winning Streaks and Losing Streaks Begin and End. (2005). International Journal of Productivity and Performance Management, 54(3), 233.

Kudo, E., & Numazaki, M. (2003). Explicit and Direct Self-Serving Bias in Japan. Journal of Cross-Cultural Psychology, 34(5), 511-521.

Krueger, J., & Mueller, R. (2002). Unskilled, unaware, or both? The better-than-average heuristic and statistical regression predict errors in estimates of own performance. Jour-nal of PersoJour-nality and Social Psychology, 82(2), 180-8.

Kruger, Justin, & Dunning, David. (1999). Unskilled and Unaware of It: How Difficulties in Recognizing One’s Own Incompetence Lead to Inflated Self-Assessments.(Statistical Data Included). Journal of Personality and Social Psychology, 77(6), 1121-34.

Larrick, R. P., Burson, K. A., & Soll, J. B. (2007). Social comparison and confidence: When thinking you’re better than average predicts overconfidence (and when it does not). Organizational Behavior and Human Decision Processes, 102(1), 76-94.

Loftus, E. F., & Wagenaar, W. A. (1988). Lawyers’ Prediction of Succes. Jurimetrics, 437-453.

(31)

Miller, D., Ross, M., & Hernstein, R .J. (1975). Self-serving biases in the attribution of causality: Fact or fiction? Psychological Bulletin, 82(2), 213-225.

Moore, D., Healy, P., & Rayner, Keith. (2008). The Trouble With Overconfidence. Psy-chological Review, 115(2), 502-517.

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.

Peterson, C. (2006). A primer in positive psychology. New York [etc].: Oxford University Press.

Ross, L., Bierbrauer, G., & Polly, S. (1974). Attribution of educational outcomes by pro-fessional and nonpropro-fessional instructors. Journal of Personality and Social Psychology, 29(5), 609.

Reed, G. M., Kemeny, M. E., Taylor, S. E., & Visscher, B. R. (1999). Negative HIV-specific expectancies and AIDS-related bereavement as predictors of symptom onset in asymptomatic HIV-positive gay men. Health Psychology, 18(4), 354.

Reifenberg, R. J. (1986). The Self-Serving Bias and the Use of Objective and Subjective Methods for Measuring Success and Failure. The Journal of social psychology, 126(5), 627-631.

Rusbult, C., Van Lange, P., Wildschut, T., Yovetich, N., Verette, J., & Insko, Chester A. (2000). Perceived Superiority in Close Relationships: Why It Exists and Persists. Journal of Personality and Social Psychology,79(4), 521-545.

Schlenker, B. R., Soraci Jr, S., & McCarthy, B. (1976). Self-esteem and group perfor-mance as determinants of egocentric perceptions in cooperative groups. Human Rela-tions, 29(12), 1163-1176.

Schwardmann, P., Van der Weele, J. Deception and Self-Deception. Working paper, Uni-versity of Munich, 2017.

(32)

Oxford Handbook of Human Motivation.

Seligman, M. E., Abramson, L. Y., Semmel, A., & Von Baeyer, C. (1979). Depressive attributional style. Journal of abnormal psychology, 88(3), 242.

Sicoly, F., & Ross, M. (1977). Facilitation of ego-biased attributions by means of self-serving observer feedback. Journal of Personality and Social Psychology, 35(10), 734.

Siwar, E. (2011). The Impact of Overconfidence Bias and Disposition effect on the Vol-ume of Transaction and the Volatility of the French Stock Martket. Journal of Applied Economic Sciences (JAES), (1(15)), 61-83.

Soll, J. B., Klayman, J. (2004). Overconfidence in interval estimates. Journal of Experi-mental Psychology: Learning, Memory, and Cognition, 30, 299–314.

Strube, M. J., & Boland, S. M. (1986). Postperformance attributions and task persistence among Type A and B individuals: A clarification. Journal of Personality and Social Psychology, 50(2), 413.

Svenson, O. (1981). Are we all less risky and more skillful than our fellow drivers? Acta Psychologica, 47(2), 143-148.

Taylor, S. E. (1990). Positive Illusions: Creative Self-Deception and the Healthy Mind. (1990). Nutrition Health Review, (53), 20.

Taylor, S., Brown, J., & Steinberg, Robert J. (1994). Positive Illusions and Well-Being Revisited: Separating Fact From Fiction. Psychological Bulletin, 116(1), 21-27.

Tracey, Jocelyn M., Arroll, Bruce, Richmond, David E., & Barham, Philip M. (1997). The validity of general practitioners’ self assessment of knowledge: Cross sectional study. British Medical Journal, 315(7120), 1426-8.

Van Loon, De Bruin, Leppink, & Roebers. (2017). Why are children overconfident? De-velopmental differences in the implementation of accessibility cues when judging concept learning. Journal of Experimental Child Psychology, 158, 77-94.

Von Hippel, W., & Trivers, R. (2011). The evolution and psychology of self-deception. Behavioral and Brain Sciences, 34(1), 1-16.

(33)

Wrangham, R. (1999). Is Military Incompetence Adaptive? Evolution and Human Behav-ior, 20(1), 3-17.

Erte Xiao, & Daniel Houser. (2005). Emotion expression in human punishment behav-ior. Proceedings of the National Academy of Sciences of the United States of America, 102(20), 7398-7401.

Zuckerman, E., & Jost, J. (2001). What Makes You Think You’re so Popular? Self-Evaluation Maintenance and the Subjective Side of the "Friendship Paradox". Social Psychology Quarterly, 64(3), 207-223.

Zuckerman, M. (1979). Attribution of success and failure revisited, or: The motivational bias is alive and well in attribution theory. Journal of personality, 47(2), 245-287.

(34)

Appendix

Appendix A

Instructions (Score receivers)

Welcome to this online experiment that will test your spatial-visualization abilities. Spatial-visualization is the ability to mentally manipulate 2 or 3 dimensional figures. Please read the instructions with care.

The experiment consists of three parts. In the first part you have to answer 13 multiple choice questions that will test your spatial-visualization skills. Parts 2 and 3 involve two short questionnaires with questions about the spatial-visualization test.

To ensure effortful participation, two individuals in this experiment will receive payment. The magnitude of which depends on how well you perform.

The experiment will take approximately 20 minutes, but is mainly dependent on how fast you are able to answer the questions.

Thank you for your participation

Part 1

In this section you will have to answer 13 questions that will test your spatial-visualization ability, there is no time limit.

One participant will win a monetary payoff in this part. The winner will be randomly selected when the experiment is over. The money you might win will depend on how well you perform in the test: for every question answered correctly you can earn 2 euro, no points will be deducted when answering a question incorrectly.

You will need to submit your e-mail address at the end of the experiment so that you can be contacted in case that you are selected.

The test is on the next page.

Part 2 - Self evaluation

In this part you receive questions about how well you think you have performed on the spatial-visualization test. The individual that answers questions 2 and 3 most accurately

will win 20 euro. Make sure you enter your email address at the end so you can be contacted. These questions require some basic knowledge about statistics. You will be

asked in what percentile you think you will rank relative to a certain group. If you have no prior knowledge of percentile ranking here is a short explanation:

• If my test score for a math test in high school is higher than 75% of the test scores of my fellow students, then my score is said to be in the 75th percentile. This means that 75% of my peers have got a lower score and 25% have obtained a higher score.

(35)

or

• If I believe that my debating skills in general rank in the 90th percentile, then I believe that I am a better debater as compared to 90% of the world, but a worse debater as compared to the rest (10%).

1. In what percentile do you think you place when you consider your spatial visualization skills in general? (so regardless of your performance in the test)

(Slider)

2. In what percentile do you think you place considering your test results compared to other people that did the test?

(Slider)

3. Out of the 13 questions, how many questions do you think you have answered correctly? (Slider)

The next page will display your result of the spatial-visualization test you made in part 1.

Your result:

You have answered x/13 questions correctly. This places you in the 0th percentile. This means that you have outperformed y% of the reference group and 100-y% of the reference group has outperformed you.

On the next page is the last part.

Part 3

Your score was: x/13

In the last part you will have to fill in a short questionnaire. These questions consider your thoughts about the test you just made. The first questions are answered using a 7 point scale.

Example: for the question: ’How hard did you try on the spatial visualization test?’, a 1 would indicate that you did not try hard whereas a 7 indicates that you tried your hardest.

1. To what extent do you think luck has affected your performance on the spatial visual-ization test?

2. To what degree do you think this test reflects your true degree of competency in spatial visualization?

(Slider)

3. How hard did you try on the spatial visualization test? (Slider)

4. How difficult did you find the spatial visualization test? (Slider)

(36)

5. In how far do you feel you were responsible for your own score? (Slider)

Are there any feelings or thoughts you would like to share concerning the spatial visual-ization test?

What is your gender?

What is your age?

What is your e-mail address (this is essential if you want to win the monetary payoff)

Referenties

GERELATEERDE DOCUMENTEN

Thus, the present study adopts a qualitative approach and explores psychology, science and engineering stu- dents’ conceptualizations of mental health through semi-

heterostructures grown on Si(001), employing a high temperature stable, sacrificial oxide template mask to obtain freestanding cantilever MEMS devices after substrate etching..

Two flexure hinge types are optimized for high support stiffness and high first unwanted eigenfrequency for two different working ranges, ±5.7° and ±20°.. We show how multiple

In contrast to the earlier reported TriboFit System using PCU liners bearing against metallic femoral heads, the Gra- dion Hip Total Cartilage Replacement (TCR) System (Biomi-

The measured 21st harmonic yield for the cluster jet (black circles), calculated 21st harmonic yield for pure monomers (blue line ) and the liquid mass fraction, g, (red circles)

binnenlandse zaken van Queensland.. De Yarrabah missie rond 1900. Halse, Gribble and Race Relations,78.. 113 Gribble’s rol in de verzameling van kinderen op de missie leidde tot

18–20 The properties of the resulting bers (Ti, Ti/TiC and Ti/TiN), including porosity, pore size distribution, bending strength and resistivity, are reported for a low (800  C)

We further showed that background light scatter- ing is the dominant source of variation in B, as for all illumination powers the standard deviation of the background photon noise