Individual and group overconfidence in
relation with group performance, an
exploratory study
Author: Maureen Koek
Student Number: 10593667
Supervisor: Rob van Hemert
Date: 27-06-2016
Bachelor’s Thesis Economics and Business
Faculty of Economics and Business
1
Statement of Originality
This document is written by Student Maureen Koek 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.
2
Acknowledgements
I want to thank Rob van Hemert for his enthusiasm and help during the process of writing this thesis. This thesis would not have been
3
Abstract
This research explored individual overconfidence and its effect on group performance, and group overconfidence. This was done by means of an experiment, which was carried out at the University of Amsterdam. The main findings were as follows. Individuals severely overestimated and overplaced their performance. Moreover, individual overplacement negatively affected levels of group performance. Lastly, group performance was severly overestimated and overplaced, and this overconfidence was negatively related to both individual- and group performance. This research is relevant as group work has become the norm in many problem-solving situations in today’s business environment. Therefore, it is of importance that knowledge about overconfidence and its possible implications are explored and this research aimed at achieving that.
4
Table of Contents
Introduction 5 Literature Review 6 Methodology 13 Results 18 General Discussion 32Limitations and Suggestions for Future Research 34
5
1 Introduction
Many studies have shown that people generally overestimate their abilities, as well as their abilities relative to those of others (Kruger & Dunning, 1999; Mayer, et al., 2007). This overestimation of one’s abilities is generally known as overconfidence and is proven to be a common trait among the public and important economic agents, such as managers, investors, politicians and employees (Mertins & Hoffeld, 2015; Libby & Rennekamp, 2012). Overconfidence is predominantly present in society as a whole, as well as in businesses. Therefore, assessing the possible effects of such overconfidence for businesses is of great importance. Several papers have explored the effects of overconfidence on certain areas of business, such as investment policy (Malmendier & Tate, 2005), acquisition behavior (Malmendier & Tate, 2008), and speculative bubbles (Schlenker & Miller, 1977). These studies offer great insights. However, the primary focus of these papers has been the effect of overconfidence in individual decision making. Considering the growing role of team work and group projects in today’s business environment, previous research merely focused on strictly individual behavior may not tell the whole story. In light of the movement towards team work, several aspects of overconfidence in relation to team work are of significant interest. For example, does individual overconfidence affect team performance, and to what extent does overconfidence exist in group settings?
Several papers have started exploring the first question (Donald & Doria (1981); Mertins & Hoffeld (2015)), but have either focused solely on self-serving and group-serving attribution biases, or have focused on cooperation without interaction. Moreover, none of these studies have connected overconfidence to group performance in terms of quality. This is of great importance, as work teams should provide the best possible quality of work. Moreover, research has focused on attributions in team settings (Schlenker & Miller, 1977). However, no research has focused on overconfidence in group work and its relation with individual overconfidence and quality.
Therefore, this paper aims at filling these research gaps by exploring individual overconfidence, its effect on group performance, and group level overconfidence. This will be done by means of an experiment, conducted at the University of Amsterdam. This is an exploratory study that aims at relating individual overconfidence to group performance, as well as exploring group overconfidence and its possible implications for business.
This paper will be structured in the following way. In the second section of this paper, previous research on the discussed topics will be reviewed. The third section will explain the experiment that was carried out in this study. The fourth section will supply results and discussions. Lastly, this paper
6
will offer conclusions of the performed study, limitations will be discussed, and suggestions for future research will be made.
2 Literature Review
This literature review aims at constructing hypotheses about individual overconfidence and its relation to group performance, and about group overconfidence. An overview of several constructs will be provided, which will lead to a number of hypotheses. The structure of the review will be as follows. Firstly, to avoid ambiguity in used terminology, the concept of overconfidence will be divided into three sub-concepts: overestimation, overplacement, and overprecision. Once satisfactory terminology is defined, this literature review will focus on individual overconfidence. Literature on the sub-concepts within overconfidence will be reviewed to provide evidence for the existence of an overestimation- and overplacement bias. Additionally, possible explanations for the existence of such biases will be given, to provide more insight into the concepts and their causes, and possible factors that would diminish these biases. This will lead up to the first three hypotheses. Secondly, two theories are provided about the influence of individual overconfidence on group performance. Lastly, theories on group-serving biases will be provided in order to predict levels of overconfidence in group settings. This will lead up to the last two hypotheses. All hypotheses, then, will be empirically tested in this research.
Before starting the analysis of overconfidence, terminology has to be defined to avoid confusion. Overconfidence has been used in the literature to describe several constructs. Moore and Healy (2008) find that three distinct concepts are described by the existing literature on overconfidence. They redefine these concepts to generate a better construct of general overconfidence and its related concepts. The first concept they describe is overestimation. Overestimation relates to the overestimation of one’s actual ability, performance, level of control, or chance of success. It entails the absolute overestimation of one’s performance. The second concept, overplacement, describes the relative overestimation of one’s performance. It is also generally referred to as the better-than-average effect. The last concept defined by Moore and Healy (2008) is that of overprecision, which relates to one’s excessive certainty regarding the accuracy of one’s beliefs. Individuals tend to wrongfully believe they are very precise in their estimates of personal performance.
In this paper, the focus will be on both overestimation and overplacement. Overprecision, in this research, will not be analyzed, due to the fact that overestimation and overprecision in many cases overlap to a great extent. In this research, this would also be the case. Subjects will be asked to rate
7
their personal performance. If subjects overestimate their personal performance and are then asked to assess how precise these estimates are, overprecision would also display overestimation, as being excessively sure one chose the best answers reflects both overestimation of one’s performance and excessive confidence in the precision of one’s knowledge, and/or ability (Moore & Healy, 2008).
Both overestimation and overplacement have been found to exist to a large extent in previous research. Mayer et al. (2007) for example, find that students systematically reported higher SAT scores than they actually had obtained. Kruger and Dunning (1999) report the same phenomenon when they asked individuals to rate their own performance, relative to objective criteria, in several intellectual areas. Overplacement, moreover, is found in several fields; individuals tend to believe that they are better than average drivers (Svenson, 1981), that they have greater intellectual abilities than the average peer (Kruger & Dunning, 1999), and that they possess more favorable behavioral traits (Alicke, Klotz, Breitenbecher, Yurak, & Vredenburg, 1995). Both overestimation and overplacement have thus been found in previous research. Therefore, it is reasonable to assume both concepts will be found in this research as well.
H1a: Individuals will overestimate their individual performance. Thus, Perceived individual performance > Actual individual performance
H1b: Individuals will overplace their individual performance relative to the performance of others. Thus,
Perceived relative individual performance > Actual relative individual performance
Research is conclusive about the existence of individual overestimation and overplacement effects. However, agreement does not exist as to how these concepts behave in relation to actual performance. Moore and Healy (2008) theorize that people’s post-task estimates of themselves are regressive, and their estimates of others are even more regressive. This due to the fact that people often have imperfect information about their own performance, but even worse information about the performance of others. Thus, when performance is high, individuals will underestimate themselves, while underestimating others even more, thereby believing they are better than their peers. Consistently, individuals that perform poorly will overestimate themselves, but others even more so, and thus believe they are worse than others. This theory is depicted in figure 1. This is in contrast to the findings of Kruger and Dunning (1999), who argue that low performing individuals will overestimate themselves most relative to the
8
performance of their peers, while high performing individuals underestimate themselves. They argue that incompetent individuals suffer a dual burden; not only are they incompetent, they are also unaware of their incompetence. In contrast, highly competent individuals suffer the burden of expertise; they tend to believe their peers are at least as competent as they are, while this often is not the case. This paper will try to analyze how these concepts are related, to support either one, or none, of these two theories.
Fig 1. Depiction of the proposed regression effect, as constructed by Moore and Healy (2008)
Although disagreement exists as to how actual performance is related to overestimation and overplacement, agreement does seem to exist on the cause of the better-than-average effect (e.g. overplacement). Several papers point to a self-serving bias as the main creator of overconfidence. Langer and Roth (1975) argue that individuals tend to believe they control certain outcomes, even if those outcomes are completely generated by chance. They tested this by a coin flipping game, in which participants had to guess whether the outcome of the coin flip would be heads or tails. However, the sequence of outcomes was manipulated to portray an ascending, descending, or random pattern. A descending pattern meant that in the beginning throws, the outcome was heads multiple times in a row. The frequency as to which heads would be the outcome multiple times in a row then declined. In the ascending pattern, the outcome differed between heads and tails in the beginning, and moved towards more frequent outcomes of heads towards the end of the experiment. The motivation for subjects to see events as controllable was so apparent, that the introduction of just one cue, a fairly consistent
9
sequence of wins for the subjects in the descending condition (e.g. a correct guess of head or tails) was enough to induce an illusion of control in a chance-dominated task. Thus, when subjects were correct more frequently, they believed this was due to their control over the task; that it was their skill that lead to more success.
Libby and Rennekamp (2012) elaborate on this idea by constructing a theory on self-serving attributions and their effect on overconfidence in a managerial setting; the issuance of management forecasts. As earlier constructed, they argue that individuals tend to attribute success to internal factors, while attributing failure externally. Thus, when succeeding, individuals believe their personal capabilities are the reason for their success, while they tend to blame external factors when failure occurs. Therefore, the favorability of beliefs of initial performance will lead to self-serving internal attributions for this initial performance. Increased self-serving internal attributions will, moderated by stable individual differences in overconfidence, affect the level of confidence in the improvement of future performance.
Hilary and Menzly (2006) find the same pattern when analyzing the effect of self-serving attributions of analysts in relation to overconfidence in future forecasts. Their research shows that analysts who have predicted earnings more accurately than the median analyst during a certain time period, tend to be less accurate and further from the consensus forecast in their subsequent earnings prediction. Thus, past success appears to be attributed internally, which causes overconfidence in future performance.
More evidence for the self-serving attributions as the cause of overconfidence is provided by Billet and Qian (2008) who show that successful acquisitions lead to overconfidence in CEOs in subsequent acquisitions. Their results show that higher-order deals have more negative announcement effects than first deals. They explain this result by stating that CEOs are generally unbiased when realizing their first acquisition. Thus, the announcement effects will be nonnegative. However, CEOs subjected to self-serving attributions tend to attribute ex-post success (which is unpredictable, and thus due to chance) to their own ability. This internally attributed success then leads to overconfidence, which will cause the CEO to pursue additional acquisitions. These subsequent acquisitions, however, will exhibit overconfidence and will thus be value destructive for the acquirer. Moreover, the same process is found in financial trading. Traders tend to excessively attribute success internally, thus becoming overconfident, which leads to decreased performance (Gervais & Odean, 2001).
It may be argued that overconfidence is decreased in group settings relative to purely individual settings. Specifically, it is argued that the better-than-average effect thrives under ambiguity. For
10
example, individuals tend to overestimate their abilities less on ability dimensions than on emotions, as ability dimensions are more objectively defined and thus less ambiguous (Alicke, Klotz, Breitenbecher, Yurak, & Vredenburg, 1995). More importantly, Alicke and colleagues (1995) argue that the nature of the comparison target provides another source of ambiguity in the social comparison process, which might even be more fundamental than the comparison dimension. They state that in most literature on the better-than-average effect, students are asked to compare themselves with the average peer (e.g. the average college/university student). This comparison is fairly ambiguous, and will therefore enhance one’s overestimation. Therefore, the authors tested the effect of both individuation and personalization on the better-than-average effect. Their main finding was that the better-than-average effect is a pervasive and robust phenomenon. However, it can be decreased by means of individuation and personalization of the comparison target. Firstly, the better-than-average effect decreased significantly when participants compared themselves with a randomly selected, same-sex peer, as opposed to comparing themselves with the average university student. The bias decreased even though no specific information was available about the comparison target; the mere presence of a live comparison target with whom minimal social contact was established was sufficient to reduce the better-than-average effect. The bias further decreased when more specific information was provided about the comparison target. Thus, the better-than-average effect is pervasive, but is decreased when the ambiguity of the comparison situation decreases by for example individuation and personalization of the comparison target. In groups, this would mean that personal overestimation may be decreased, as individuals get to work on a personal level with peers, thus individuating and personalizing comparison targets.
Moreover, individuals may be less willing to publicly offer too boastful self-presentations in groups, as their actual performances will become evident in a later stage. On In general, on top of overestimating themselves, individuals provide even more favorable estimates of their performance to others, in order to create an as self-esteem enhancing and approval gaining public image as possible. However, research shows that this tendency decreases when anonymity decreases and future results will actually be observable by others. (Schlenker, 1975). According to this, individuals would offer more accurate presentations of their own predictions of future performance when working in groups relative to when working individually.
As individuation and personalization of the comparison target, as well as diminished desire to offer too boastful representations of future performance are assumed to occur in groups, it is predicted that individuals in this study will overplace themselves less when comparing themselves to their group members instead of to the average student.
11
H2: Overplacement will be decreased as the level of ambiguity of the comparison target is diminished by personalization and individuation. Thus,
Overplacement relative to the average group member < Overplacement relative to the average student.
In summary, it is expected that individuals will overestimate, as well as overplace, their individual performance. Overplacement is expected to decrease when ambiguity of comparison target is decreased, which is assumed to be the case when working in groups. The next topic of interest is group behavior. How does individual overconfidence influence group performance? And does overconfidence (e.g. overestimation and overplacement) exist at the group level. The next section will explore this topic.
Firstly, some research argues that individual overconfidence may not influence team work negatively. Mertins and Hoffeld (2015) argue that the prevalent social norm of team production is conditional operation. That is, the more an individual believes team members will contribute to the team cause, the higher this individual’s input to this team cause. The authors find that overconfident men are more optimistic about the expected input of team members than men with less confidence, and that they therefore cooperate more. This is a gender-specific link which is thus not found in women. According to the authors, this implies that overconfident males are “ideal” team players. Mertins and Hoffeld (2015) are not the first ones to argue a favorable link between overconfidence and performance. If overconfidence is sufficiently small, this overconfidence might induce high efforts of the overconfident agent, which may lead to a comparative pay-off advantage, caused by an increased probability of success. Therefore, the overconfident individual may actually have a comparative advantage (Ludwig, Wichardt, & Wickhorst, 2011). However, these studies do not link overconfidence to quality of group performance. They merely link overconfidence to productivity levels. It may also be argued that individual overconfidence leads to decreased levels of quality of group outcomes, due to the fact that overconfident individuals may push their ideas, believing these are better than the ideas of others, even if this is not the case. If more capable individuals suffer a burden of expertise, they might wrongfully believe the pushed answers of less quality are as good as their own, thereby choosing these outcomes as the right ones. However, this theory has not yet been tested in previous research. Thus, it is not certain which effect individual overconfidence has on group performance. However, an
12
effect of overconfidence on group performance is expected. This research will therefore analyze whether individual overconfidence affects group performance, either positively or negatively.
Moreover, regardless of the effect of individual overconfidence on group performance, it is interesting to explore the extent to which individuals overestimate and overplace the performance of their group. This is interesting, as overestimation and overplacement of group performance may have detrimental effects, similar to the effects of individual overconfidence. As stated earlier, individual overconfidence leads to decreased performance in many cases, such as acquisitions, financial trading, and creating accurate management forecasts (Libby & Rennekamp, 2012; Hilary & Menzly, 2006; Billett & Qian, 2008; Gervais & Odean, 2001) Therefore, if overconfidence in group output exists as well, it is highly likely that this has negative effects on future performance, similar to those of individual overconfidence.
Research on group overconfidence has focused mostly on self-serving attributions within groups. Schlenker and Miller (1977) find that members of successful groups perceive themselves to be more responsible for the group outcome, whereas members of unsuccessful groups tend to believe they are somewhat less responsible than the average group member. The authors explain this phenomenon by pointing to the general human tendency to associate one’s behavior with positive outcomes, while limiting the association with negative outcomes, in order to sustain self-confidence. Thus, individuals rarely attribute failure internally. Wolosin, Sherman, and Till (1973) elaborate on the factors that are used to blame failure externally. They examined the allocations of responsibility to three possible sources; the self, the other person, and the situation, as a function of outcome and whether the task is cooperative or competitive. They found that strongly expected, average, outcomes were attributed to the situation in both cooperative as well as competitive settings. Success was attributed to the self (e.g. personal capabilities/skills) in both settings. However, the attributions when failure occurred differed. In the cooperative setting, failure was attributed to the partner. Whereas failure in the competitive setting was attributed to the situation (Wolosin, Sherman, & Till, 1973). These findings are in line with theories on sustaining self-confidence. When competing, attributing failure to the competitor implies that the competitor was better than the self. As individuals prefer to believe they are, at least, better than average, it is more suitable to attribute failure in this instance to the situation. However, when blaming failure in a cooperative setting to a group member, the self-confidence is sustained, as the lower outcome is now actually less attributable to the self. This tendency was also found by Schlenkler and Miller (1977), who analyzed attributions in larger groups. Their findings suggest that members of failing groups do prefer the self-protective attributions as proposed by Wolosin et al. (1973). However, they prefer self-protective attributions that do not involve direct social comparisons with most other
13
group members. Often, it is preferred to blame one person within a group; a phenomenon generally known as scapegoating.
Self-serving biases are thus predominantly present in cooperative settings. However, Taiylor and Doria (1981) suggest that self-serving biases coexist with another bias: a group-serving bias. In their research, they examined how male and female intercollegiale athletes involved in teamsports rated the performance of themselves and their teammates. The results provide evidence for both self-serving biases and a team-self-serving bias. Firstly, results of earlier studies pointing to internal attribution of success and external attribution of failure are replicated; individuals were more likely to attribute personal failure externally than they did personal success. Moreover, individuals who performed well were more likely to claim that their success affected team performance positively than poorly performing individuals were to claim that their bad performance affected team performance negatively. A group-serving bias was indicated by the fact that players were more likely to attribute team success to effective team play than they were to attribute failure to generally bad team play. Additionally, individuals who experienced individual success were more likely to attribute this success to the helpful contribution of teammates than individuals who experienced personal failure were to claim that their personal goal was hindered by team mates. This, according to the authors, is a manifestation of the group-serving bias as it serves as protection of the group’s reputation. Nevertheless, this result contrasts the findings of Wolosin et al. (1973) as failure in this cooperative setting is not attributed to the partner, but to external factors. This may be explained by the fact that general team play was pointed out as a possible source of responsibility for failure, as opposed to individuals within the team. This makes the abovementioned scapegoating impossible and may cause individuals to refrain from blaming the team in general in fear of retaliation. More evidence for a group-serving bias is found in the phenomenon of in-group favoritism. Brewer (1979) found that members of an in-group are more positively evaluated than members of the out-group. However, this was only the case when individuals were grouped on the basis of a common (fictitious) characteristic. This phenomenon may be explained by social identity theory, which states that individuals increase their self-confidence, by identifying with a group they view positively. Therefore, it is in one’s best interest to view people in the in-group as positively as possible, in order to view oneself positively (Taifel, 1982). Thus creating a preference for in-group members.
Group-serving biases have thus been found in previous research. As self-serving biases are conclusively offered as the cause of overconfidence (e.g. overestimation and overplacement), it seems reasonable to assume that group-serving biases cause overconfidence at the group level, similar to individual settings. Therefore, due to the fact that group-serving biases have been proved to exist within
14
groups, it is expected that overestimation and overplacement will consequently be found in group settings. This leads to the following hypotheses:
H4a: Individuals will overestimate the performance of their group. Thus, Perceived group performance > Actual group performance
H4b: Individuals will overplace the performance of their group relative to the performance of other groups. Thus,
Perceived relative group performance > Actual relative group performance
In summary, this research aims at exploring several topics, which have not yet been related to one another in previous research. Firstly, results of previous studies on individual overconfidence will most likely be replicated, when researching levels of individual overestimation and overplacement. Secondly, this study aims at analyzing whether decreased levels of ambiguity in relation to the comparison target decreases levels of individual overplacement. This will be done as it is expected that ambiguity of the comparison target is decreased in group settings. Therefore, overplacement within groups may be decreased. Moreover, this study will analyze whether individual overconfidence affects the quality of group outcome. Lastly, group overconfidence will be analyzed on its own, as well as in relation with individual overconfidence. This has never been done before by previous research and may offer interesting insights, as team-work has proved to be more predominantly present in today’s business environment.
3 Methodology
This study was designed to measure several constructs. First of all, individual overconfidence was tested, by means of testing individual overestimation and overplacement. Moreover, effects of individuation and personalization on individual overplacement were tested. Secondly, individual overconfidence was related to group outcomes, to test for possible effects. Lastly, group overconfidence was measured, by means of testing group overestimation and overplacement.
15
Additional variables, such as contribution to and satisfaction with group answers were measured, to serve as control and/or explanatory variables.
3.1 Participants
Two groups of subjects were used to collect data. The first group consisted of 64 prospective students of the university of Amsterdam, divided over 4 seminars. They were recruited during the UvA matching day for the bachelor’s program “Economics and Business”. The second group consisted of 175 students taking a course in organizational psychology, also at the University of Amsterdam, divided over nine workgroups.
3.2 Methods
In the first study, subjects were given literature to read. Subsequently, they were asked to answer several questions about this literature. They were asked to do this individually first, and to later discuss the answers in groups of 3 to 5 participants. All groups were asked to come to an agreement about the answers. Following this, subjects were asked to fill in a paper-based questionnaire, measuring several constructs, as is explained in section 3.3. Unfortunately, it was not possible to match actual results to the answers on the questionnaire in this group. Therefore, this subject group was mostly used as a pilot study. Due to this study, mistakes and ambiguities in the questionnaire were resolved before the second study was conducted.
In the second study, participants were asked to participate in a survival challenge. The exact exercise can be found in the appendix. In short, the exercise consisted of instructions describing a situation in which the subjects had just experienced a plane crash. The plane had crashed in northern Canada in winter time, with daily temperatures averaging at 25 below zero, and night temperatures averaging at 40 below zero, and the nearest town being approximately 30 miles away from the crash site. Subjects then were asked to select five items from a list of twelve which they assumed would be most helpful in this situation.
This particular challenge was chosen for several reasons. Firstly, it was expected that most subjects would not have any previous experience with such a challenge and that estimates of performance, therefore, would become harder to make. This was done, because this sort of ambiguity exists in business decision making- and problem solving teams as well. Problems within business can often be solved in many ways, and it is often unclear what the best solution is. Generated solutions will
16
only prove successful after they are chosen. The survival challenge offers a similar situation. However, for this study, it was of significant importance that it was possible to measure actual performance on this fairly ambiguous task. The survival challenge offers this opportunity, as all twelve items that were listed as choices were ranked in order of importance by Mark Wanvig, a former instructor in survival training for the Reconnaissance School of the 101st division of the U.S. Army Therefore, actual quality
of the set of five items that were selected by subjects could be measured. This specific survival exercise is currently used in wilderness survival training programs in the U.S. military, which further strengthens the validity of the ranking as a measure of quality.
Subjects were asked to complete the survival exercise individually. Interpersonal discussion was not allowed. When all subjects had completed their individual answers, they were asked to form groups of 3 to 6 participants. They were asked to discuss their individual answers and to eventually reach a set of five items as a group. Intergroup discussion was not allowed. Subsequently, subjects were asked to fill in a paper-based questionnaire, which was the same questionnaire as was used in the first (pilot) study. Questionnaires and individual answers were collected together, so that they could be matched to one another in subsequent analysis. Questionnaires and individual answers were collected per group, together with the group answer, in order to match these outcomes as well.
3.3 Measurements
Firstly, this study aimed at replicating earlier studies on individual overconfidence. Therefore, individual overestimation and overplacement were tested. To test for individual overestimation, subjects were asked to ranked their personal performance on a scale from 1 to 100. Actual individual scores were collected and matched to each individual answer to this question. However, scores were measured on a scale from 15 to 50, where 15 was the optimal score. Therefore, the actual scores had to be recalculated to match the scale that was used in the questionnaire. The following formula was used to calculate the grades:
𝑆𝑐𝑜𝑟𝑒2 = (1 −𝑆𝑐𝑜𝑟𝑒1− 15
17 Where Score 1 represents the score measured on a 15-50 scale and Score 2 represents the score measured on a 0-100 scale1.
The difference between perceived performance, as indicated on the questionnaire, and actual scores yielded a measure of individual overestimation. Individual overplacement was measured by asking subjects to rate their individual performance relative to the performance of the average student on a scale from 0 to 99. Actual percentile rankings were calculated, based on actual scores of all subjects. This yielded a measure of overplacement. Moreover, the effect of individuation and personalization was tested by asking subjects to rate their performance relative to their group members, in addition to asking them to rate their performance relative to the average student. It was assumed that individuation and personalization would occur to an increased extent when comparing performance to a group member as opposed to comparing performance to the average student. Therefore, comparing levels of overplacement in comparisons with both the average student, and the average group member, would show whether individuation and personalization decrease overestimation.
The second aim in this study was to measure the effect of individual overconfidence on group performance. Measures of individual overconfidence have already been constructed, as described above. However, to test for influence on group performance, measures of group performance were constructed. Group outcomes were collected, together with individual scores per group. Thus, measures of average individual outcomes were constructed, as well as measures of group outcome. The average individual outcomes per group were calculated to serve as a control variable, as it was expected that these would have a profound effect on group performance (measured as group score). Average individual overestimation and overplacement were calculated per group to serve as predicting variables for group outcome.
Lastly, this study aimed at measuring overconfidence in groups. Overconfidence was again divided in overestimation and overplacement. Subjects were firstly asked to estimate the absolute score of their group on a scale from 1 to 100. The difference between this estimate and the actual group score yielded the group overestimation measure. Actual group scores were calculated using the same formula that was used to calculate individual scores. Secondly, subjects were asked to rate the performance of their group relative to the performance of other groups on a scale from 0 to 99. Actual percentile
1 The scores were measured on a 0-100 scale instead of the 1-100 scale that was used in the questionnaire. However, these differences are assumed to be negligibly small.
18
rankings for each group were calculated using actual group scores. The difference between estimated relative performance and actual relative performance yielded the measure of overplacement.
4 Results 4.1 Participants
Participants were 175 students taking a course in organizational psychology at the University of Amsterdam. Their mean age was 21,03 years and 61 percent of the subjects was female. Moreover, 64 prospective students of the University of Amsterdam were used as a pilot study. The mean age of this subject group was 17.03 years old and 39 percent was female. However, some results from this study were used to provide additional support for the findings of the main study.
4.2 Individual Overconfidence
The first concept that was tested in this study was individual overconfidence. This measure was divided into two sub-concepts: overestimation and overplacement, which were both tested.
4.2.1 Individual overestimation
As expected, subjects overestimated their performance to a large extent. On average, participants estimated their performance at a value of 68.13 Actual performance, however, averaged at 48.23. On average, participants thus overestimated performance by 75.82 percent. This difference is significant at a 1% significance level, paired-sample t(173) = 10,719, p < 0.0001. Worse performing individuals tended to overestimate their performance more than better performing individuals did. Overestimation is significantly negatively correlated with actual performance, r(175) = -.738, p < 0.0001. Elaborating on this correlation, subjects were divided into quartiles, based on their relative performance. Estimates of performance did not differ significantly between the four quartiles. Actual performance, however, differed significantly below the 0.01 significance level, one-way ANOVA, p < 0.0001. Thus, subjects across all performance quartiles believed they performed about the same, while performance differed significantly. Hence resulting in the negative relationship between perceived and actual performance. Interestingly, top-quartile subjects actually tended to underestimate their performance. They underestimated their performance by an average of 2.55 percent. Contrastingly, bottom-quartile subjects overestimated their performance by an average of 218.59 percent. The differences between quartiles are illustrated in figure 2. Perceived ability is fairly stable across quartiles. Actual scores
19
however, decline sharply. The difference between perceived performance and actual scores represents the amount of overestimation, which thus increased when moving from the top- to the bottom quartile.
Figure 2. Perceived individual performance in contrast with actual individual scores per quartile (based on individual performance).
4.2.2 Individual Overplacement
Subjects overplaced themselves when rating their performance relative to the average student. On average, subjects placed themselves at the 64th percentile, which exceeded the actual mean (50, by
definition), one-sample t(173) = 12,120, p < 0,0001 2.
Estimations of relative performance did not differ significantly between quartiles, one-way ANOVA, p = 0.076, which indicates that subjects uniformly believed they performed around the 64th
percentile. However, actual relative performance did differ significantly between the quartiles, one-way ANOVA, p < 0.0001.
As can be seen in table 1, subjects believed their performance on the survival exercise was better than average across all quartiles. Subjects in the top-, 2nd, and 3rd quartiles rated their relative performance at the 68th, 65th, and 61st percentile, respectively. All estimates significantly differed from the defined average of 50, p < 0.0001.
As can be seen in figure 3, subjects in the top- and 2nd quartile tended to underplace themselves, by 19 and 4 percent points on average, respectively. In contrast, subjects in the 3rd- and bottom quartile
2 This result was also found in a separate study performed on 64 prospective University of Amsterdam students at the University of Amsterdam matchingday. Subjects, on average, rated their relative performance on a business exercise at the 63rd percentile. This estimate differs significantly from the average of 50, one-sample t(62) = 7.333 , p < 0.0001.
0 10 20 30 40 50 60 70 80 T O P Q U A R T I L E 3 R D Q U A R T I L E 2 N D Q U A R T I L E B O T T O M Q U A R T I L E Perceived ability Actual score
20
overplaced themselves, by 23 and 48 percent points, respectively. Subjects in the bottom-quartile did not only overplace themselves, they actually believed they performed better than average. As stated above, they placed themselves at the 60th percentile, which significantly exceeded the average, one-sample t(42) = 4,997, p < 0,0001.
These results imply a negative relationship between actual individual performance and overplacement. When tested, these variables proved to be strongly correlated to a significant degree, r(175) = -.844 , p < 0.0001. Thus, as individual performance decreased, overplacement increased. This relationship is illustrated in figure 3.
Figure 3. Perceived individual relative performance in contrast with actual individual relative performance per quartile (based on individual performance).
0 10 20 30 40 50 60 70 80 90 100 T O P Q U A R T I L E 2 N D Q U A R T I L E 3 R D Q U A R T I L E B O T T O M Q U A R T I L E
Actual relative performance (in percentiles) Perceived relative performance (in percentiles)
Table 1
Please compare your personal performance on the exercise with the average student taking this course and indicate your performance ranking by indicating a number between 0 and 99
ActualPersRank Mean N Std. Deviation
Top quartile 67,52 44 12,905
2nd quartile 65,25 44 17,573
3rd quartile 61,28 43 14,146
Bottom quartile 60,28 43 13,488
21
4.2.3 Individuation and personalization
To analyze the effect of individuation and personalization, subjects were asked to rate their individual performance relative to the performance of their group members. Overplacement was again apparent. Subjects, on average, placed their relative performance at the 63rd percentile. This, once more, differed
significantly from the average, one-sample t(172) = 10.965 , p < 0.0001 3.
A slight decrease in overplacement occurred when subjects compared themselves to group members as opposed to the average student taking the course (estimates decreased from the 64th to the
63rd quartile). However, these results were not significant, paired-sample t(172) = ,120 , p = ,905. It was expected that individuation and personalization of the comparison target would occur when subjects compared themselves with group members, and that therefore overestimation would decrease. However, this effect was thus not found.
4.2.4 Gender in relation to individual overestimation and overplacement
Gender and its effect on overconfidence was not specifically mentioned in the research design. However, gender differences in overestimation behavior were so profound in this research that this factor was not to be neglected. Female subjects overestimated their performance less than did their male counterparts to a significant degree, one-way ANOVA p < 0.05. As can be seen in table 2, female subjects overestimated their performance on average by 53.39 percent. For male subjects, overestimation averaged at 110.29 percent. Differences between female and male subjects in overplacement behavior, contrastingly, were not found. As can be seen in table 3, female subjects did overplace themselves to a lesser extent than male subjects did, both when comparing their performance relative to the average student and the average group member. However, neither of these differences were significant, one-way ANOVA, p = 0.663, one-way ANOVA, p = 0.731, respectively.
Table 2
Individual percentage overestimation
Gender Mean (in percentages) N Std. Deviation
Male 110,287668 69 220,2082196
Female 53,388487 106 77,3651994
Total 75,823021 175 152,7752248
3 The same result was found in the separate study done at the University of Amsterdam matchingday. In that study, subjects placed their performance relative to the performance of group members at the 62nd percentile, which additionally differed significantly from the general mean, one-sample t(59) = 5.771 , p < 0.0001.
22
Table 3
Individual overplacement in percentile points
Gender
Individual overplacement (compared to average student)
Individual overplacement (compared to average group member) Male Mean 14,9869 12,001449 N 69 69 Std. Deviation 33,34306 41,1091552 Female Mean 12,9472 14,098113 N 106 106 Std. Deviation 27,99013 38,3045123 Total Mean 13,7514 13,271429 N 175 175 Std. Deviation 30,13718 39,3306739 4.2.5 Summary
Subjects in this study heavily overestimated their abilities. On average, participants overestimated their performance on the survival challenge by 75.82 percent. Hence, hypothesis 1a was supported. Moreover, this study found that individual performance and individual overestimation are heavily negatively correlated. Which means that worse performance leads to higher levels of overestimation.
These results replicate earlier studies on overestimation. Previous research has shown that individuals who perform poorly overestimate themselves most, whereas slight underestimation tends to occur when performance is high (Mayer, et al., 2007;Kruger & Dunning, 1999;Moore & Healy, 2008).
Moreover, subjects in this study overplaced themselves to a significant degree. On average, participants overplaced themselves by 14 percentile points. Thus, hypothesis 1b was supported. Similar to overestimation, overplacement is negatively correlated with actual performance; worse performance increases levels of overplacement.
These results support the findings of Kruger and Dunning (1999). Low-performing individuals overplace their performance most, relative to others, whereas individuals with high-performance levels underplace their relative performance. Low-performing individuals thus not only overestimate their performance, they are also not able to recognize their poor performance when they compare themselves in group work to more capable individuals, as was proposed by Kruger and Dunning (1999). However,
23
individuals that are performing at the top level underplace themselves relative to their peers, even when they have seen the work of less capable individuals within their groups. Kruger and Dunning (1999) proposed that high-performing individuals would adjust their relative judgments to a more accurate level when they were able to perceive the actual performance of their peers. However, in this study, all subjects were able to perceive the performance of their group members. Nevertheless, high-performing individuals still underplaced themselves. This implies that the burden of expertise, as proposed by Kruger and Dunning (1999) may still be apparent when actual performance is perceived. This has important implications for businesses, as this supports the theory proposed in section 2, which argues that more capable individuals may accept ideas of worse quality as they believe less performing individuals are as competent as they are. This may lead to decreased levels of group performance.
However, over- and underestimation, as well as over- and underplacement may have occurred to a larger extent in this study, due to the fact that the task was fairly ambiguous. Most subjects had never done a similar exercise before. Therefore, it is likely that they were less able to judge their absolute and relative performance. However, relatively high levels of ambiguity were intentionally applied in this study. In problem solving group work within organizations, high levels of ambiguity exist as well. Therefore, to increase the generalizability of this study, ambiguity was intentionally created and the results are therefore specifically valuable for organizational settings which involve ambiguous problem solving.
When testing whether individual overplacement would be decreased by decreasing levels of ambiguity regarding the comparison target by asking subjects to compare their performance to the performance of group members instead of comparing their performance to the average student, a slight decrease of overplacement was observed. However, this decrease was not significant Thus, hypothesis 2 was not supported. A possible explanation for this result may be that subjects had interacted with other subjects before taking part in this experiment, as most subject were part of the Bsc Economics and Business at the University of Amsterdam. Therefore, due to previous contact, individuation and personalization may already have occurred. Hence, comparing themselves to the average group member as opposed to the average student may not have increased personalization and individuation, as those processes had already occurred in earlier interactions. However, similar to ambiguity, this effect is likely to be generalizable to organizational settings. When temporary teams are formed, it is not unlikely that individuals within this team have met each other before, and that personalization and individuation have already occurred.
Lastly, this study showed gender differences in overestimation behavior. Male subjects tended to overestimate their abilities more than twice as much as female subjects did. However, differences between male and female subjects were not found for overplacement behavior. Gender differences in
24
overestimation behavior may have been caused by the nature of the task. Subjects were asked to perform a task based on survival skills. It may have been that male subjects believed that they would perform better at such a task than female subjects would, as survival may have been perceived as a somewhat masculin task. However, if this had been the case, it would have been logical for females to overplace themselves less than men as well, as they would believe men would be better at the task. This however, was not found. Moreover, male subjects overestimated their abilities more than female subjects did to such a large extent (twice as much), that it is unlikely that this effect is caused merely by the nature of the performed task. Therefore, it seems reasonable to conclude that gender does affect individual overestimation behavior.
4.3 The Effect of Individual Overestimation and Overconfidence on Group Performance
In addition to testing for levels of individual overconfidence, this study aimed at testing the effect of individual overconfidence on group performance. Therefore, levels of individual overconfidence (e.g. individual overestimation and individual overplacement) within teams were analyzed in relation to group performance. The sample consisted of forty-four groups, consisting of 3-6 individuals each. Several variables were used to test the effect of individual overconfidence on group performance. Actual group scores were used as a measure of group performance. As can be seen in table 4, these scores averaged at 56.75, with a standard deviation of 10.61. Secondly, average individual scores per group were calculated. This variable was calculated to control for the effect of individual quality on group quality. On average, individual scores per group were 47.87. The average group scores significantly improved relative to the average individual scores per group, paired-sample t(43) = 4.091, p < 0.0001. Thus, in general groups do perform better than the average individual. To be able to test the effect of individual overconfidence on group performance, average individual overestimation (in percentages) and individual overplacement (in percentile points) were calculated per group. Average individual overestimation per group was 80.36 percent4. Subjects overplaced themselves in comparison
with their group members by 13.62 percentile points, on average.
Firstly, the effect of the control variable (average individual scores per group) on group performance was tested. OLS-regression analysis shows that average individual scores strongly affect group score, β = 0.643, p < 0.0001, adjusted R² = 0.400.
4 This number differs slightly from the average individual overestimation mentioned in section 4.2.1 as this estimate was calculated per group instead of per individual.
25
It was predicted that individual overconfidence would have an effect on group outcome. Therefore, overestimation and overplacement were both added to the OLS-regression as described above, separately. Moreover, an extra control variable was added. Group size was used as an additional control variable, as larger groups may have more information, assuming that each group members contributes a certain amount of information to a group. Therefore, larger groups may achieve better outcomes.
Firstly, the effect of individual overestimation on group performance was tested by means of an OLS-regression, controlled for average individual performance and group size. Individual overestimation did not significantly affect group performance, p = 0.475. Secondly, the effect of individual overplacement within groups on group performance was tested in the same manner. Individual overplacement within groups was used, instead over general individual overplacement (e.g. comparison with the average student), as it was expected that overplacement when comparing to group members, would have a larger effect on group processes, and therefore on group performance. Individual overplacement within groups negatively affected group performance and this effect approached significance, β= -.198, p = 0.095.
Table 4
N Minimum Maximum Mean Std. Deviation Average overestimation of
individual performance per group
44 -15,54 50,76 19,7029 13,71661
Average individual scores
per group 44 25,71 74,29 47,8680 10,61490 Actual group performance 44 22,8571 94,2857 56,753241 18,7263273 Valid N (listwise) 44
In summary, individual overestimation does not affect group performance. However, individual overplacement within groups does seem to affect group performance negatively. Thus, when individuals within workgroups overplace themselves to sufficiently high degrees, this may harm group performance.
It seems intuitive that individual overplacement has a stronger effect on group performance than individual overestimation does, as overplacement measures relative performance in comparison with group members. Therefore, when overplacing themselves, individuals falsely believe they are better than their group members . This may lead those individuals to push their own ideas and not consider ideas of others well enough. This effect, nevertheless, may be partly offset by more capable
26
individuals noticing overconfidence more than less capable individuals do. This study shows that competence (measured as personal performance on the exercise) is positively correlated with perceptions of overconfidence of group members, r(167) = 0.170 , p < 0.05. So, more competent individuals perceive their group members as more overconfident about the quality of their own ideas and information than less competent individuals do. This, consequently, may slightly decrease the negative effect of individual overplacement on group performance.
Nevertheless, if overplacement affects group performance in a negative manner, this would have great consequences for group work, as this study has shown that overplacement levels are considerably high. If group performance is decreased by individual overplacement, and group performance, subsequently, is overestimated and/or overplaced, this could have detrimental effects for businesses. Even if overplacement and overestimation do not effect group performance negatively, it would still be of significant interest to analyze how group performance is evaluated, and whether overconfidence exists at the group level. Especially since group performance in some cases in this study was worse than the worst individual performance. If members of these groups are overconfident in the performance of their group, this may have severe negative effects for business.
4.4 Group level overconfidence
Overestimation and overplacement at the group level could potentially be harmful. As shown in previous literature, overconfidence at the individual level, in many cases, leads to decreased levels of future performance. At the individual level, research has conclusively shown that overconfidence exists, and that caution for individual overconfidence is necessary. However, overconfidence in group outcomes has never been tested. Therefore, this research will continue by analyzing overestimation and overplacement at the group level and how these forms of overestimation and overplacement are related to individual overestimation and overplacement.
4.4.1 Group level overestimation
Similar to individual overestimation, subjects overestimated the performance of their group. Subjects rated group performance on average at a level of 76.06. Actual performance, however, averaged at 57.53. Thus, on average subjects overestimated the abilities of their group by 49,42 percent. This difference is significant at the 1% significance level, paired-sample t(173) = 10.553, p < 0.0001. Group performance is negatively related to overestimation of group performance, r(175) = -.811 , p < 0.0001.
27
Thus, overestimation of group performance increased as group performance declined; poorly performing groups overestimated themselves to a larger extent than more capable groups did. Individual performance is also negatively related to overestimation of group performance, r(175) = -.348 , p < 0.0001. Hence, poorly performing individuals, as well as individuals within poorly performing groups, tended to overestimate group performance more than more capable individuals did. Overestimation thus persists in judgments of individual performance, as well as in judgments of group performance. However, estimates of group performance were characterized by overestimation to a lesser extent than were estimates of individual performance, paired-samples t(174) = 2.206 , p < 0.05. Thus, overestimation seems to be more present in estimates of individual performance than in estimates of group performance.
However, this relationship between individual- and group overestimation depends on individual performance. When subjects were clustered in quartiles based on individual performance, these differences became visible, as is depicted in figure 5. In the top quartile, group overestimation was more predominant than individual overestimation. On average, top performing subjects overestimated group performance by 17.93 percent point more than they overestimated individual performance (which was actually underestimated), paired-sample t(43) = -2.804 , p < 0.01. Contrastingly, the subjects in the bottom-quartile overestimated individual overestimation more than group performance by 139.04 percent points, paired-sample t(42) = 3.362 , p < 0.01. Thus, overestimation of performance decreased when work was done in a group for subjects in the bottom- and 3rd quartiles, but increased for subjects in the top- and 2nd quartiles.
Nevertheless, when subjects were clustered in quartiles based on group performance, similarities between individual- and group overestimation became apparent. Similar to the individual setting, no significant differences were found in estimates of group performance between the quartiles, one-way ANOVA, p = 0.256. However, significant differences in terms of actual group performance (p < 0.0001) were apparent between quartiles. Members of groups in the bottom-quartile, on average, overestimated their performance by 157.19 percent. Members of the groups in the top-performing quartile, however, tended to underestimate themselves (on average by 2.27 percent). These differences are shown in figure 4. Individual- and groups overestimation follow the same pattern; individual (group) overestimation decreases when individual (group) performance increases.
In total, concepts of actual individual- and group performance, and individual- and group overestimation were all related to one another. Low individual scores increased overestimation of one’s abilities and were the main cause of low group scores. Low group scores were again followed by high levels of group-overestimation. In this framework, levels of individual overestimation predicted subsequent levels of group overestimation; high levels of individual overestimation lead to high levels
28
of group overestimation. Low performing individuals overestimated themselves most, both in individual- and group settings, and this overestimation may be harmful.
Figure 4. Perceived group performance in contrast with actual group scores per quartile (based on group performance).
Fig 5. Individual overestimation and group overestimation per quartile (based on individual performance).r.
0,00 10,00 20,00 30,00 40,00 50,00 60,00 70,00 80,00 90,00 T O P Q U A R T I L E 2 N D Q U A R T I L E 3 R D Q U A R T I L E B O T T O M Q U A R T I L E
Please think about the performance of your group in the exercise and indicate how well you think your group did on a scale from 1 to 100
ActualGroupRaw -50,000000 0,000000 50,000000 100,000000 150,000000 200,000000 250,000000 T O P Q U A R T I L E 2 N D Q U A R T I L E 3 R D Q U A R T I L E B O T T O M Q U A R T I L E
29
4.4.2 Group level overplacement
Similar to individual overplacement, subjects overplaced group performance relative to other groups. On average, subjects placed group performance at the 70th percentile, which significantly exceeded the actual mean (50, by definition), one-sample t(161) = 16.536, p < 0.0001 5. Moreover, group
overplacement was significantly stronger than individual overplacement, by 7 percent points, paired-sample t(161) = 4.457, p < 0.00016.
No significant difference between estimates of relative group performance was found, p = ,098, although it approached significance. This implies that subjects across all four quartiles believed they performed at a similar level (e.g. the 70th percentile). This, however, was not the case as actual performance differed significantly, p < 0.0001.
Individuals were again divided into quartiles, now based on relative group performance. As can be seen in table 5, individuals in all four quartiles believed their group performed better than average. As would be expected, individuals that were part of groups in the top-performing quartile placed themselves highest in terms of relative performance (at the 75th percentile). However, members of groups in the bottom-quartile rated themselves second highest on relative performance, at the 70th percentile, which is significantly better than average, one-sample t(35) = 7.093, p < 0.0001. This shows that overplacement is stronger for individuals in poorly performing groups. Subjects in the second and third quartiles overplace themselves as well, at the 69th and 67th percentile, respectively. The small decrease in estimates between these groups further illustrate that overplacement increases when group performance decreases.
This relationship is shown in the strong negative relationship between actual group performance and group overplacement, r(175) = -.782 , p < 0.0001. This relationship is illustrated in figure 6. Again, the difference between actual relative group performance and perceived relative group performance equals the level of group overplacement. Similar to the individual setting, perceived relative group performance is fairly stable across quartiles. However, actual relative performance decreases. Therefore, as actual performance decreased, overplacement increased. Note, again, that individuals in the top percentile actually underplaced relative group performance.
5 This result was also found in the study done at the University of Amsterdam matching day. Subjects in that study rated group performance at the 71st percentile, on average. This estimate exceeded the average significantly, one-sample t(48) = 10.517, p < 0.0001.
6 These results were also found in the pilot study. Group overplacement was significantly stronger than individual overplacement, paired-sample t(48) = 2.722, p < 0.01.
30
Moreover, individual performance was negatively related to group overplacement, r(175) = -.328, p < 0.0001. Thus, as performance at the individual level decreased, overplacement of group performance increased.
Table 5
Please compare your group performance in the exercise with the performance of the other groups and indicate your group performance ranking by indicating a number between 0 and 99
Actual Group Rank Mean N Std. Deviation
1 74,62 40 17,000
2 68,56 43 12,916
3 66,51 43 13,530
4 70,39 36 17,247
Total 69,92 162 15,333
Fig. 6 Perceived relative group performance in contrast with actual relative group performance per quartile (based on group performance).
4.4.3 Gender and group level overestimation and overplacement
Additionally, gender did seem to affect group overestimation, however not significantly. Female subjects did overestimate group performance more than male subjects did, by 18.31 percent point. This difference approached significance, one-way ANOVA, p = 0.087. To analyze gender effects in more depth, the relationship between individual- and group overestimation were analyzed per gender category. For female subjects, no significant difference was found between their overestimation levels
0,0000 10,0000 20,0000 30,0000 40,0000 50,0000 60,0000 70,0000 80,0000 90,0000 100,0000 T O P Q U A R T I L E 2 N D Q U A R T I L E 3 R D Q U A R T I L E B O T T O M Q U A R T I L E
31
of individual- and group performance, paired-sample t(105) = -,372 , p = 0.711. However, male subjects significantly overestimated individual performance more than they overestimated group performance, by 71.96 percent point, paired-sample t(68) = 2.725, p < 0.01.
Similar to the individual setting, differences in overplacement behavior at the group level were not found, one-way ANOVA, p = 0.135.
4.4.4 Summary
In addition to overestimating individual performance, subjects in this study overestimated group performance. Hypothesis 4a was hence supported. This group overestimation is negatively related to group-, as well as individual performance. Thus, poor performance in both individual as well as group settings, leads individuals to overestimate group performance more. However, overestimation occurred to a lesser extent when group performance was judged as opposed to individual performance. This implies that overestimation is reduced when individuals estimate the performance of a group instead of their personal performance. However, other explanations for this reduced bias are possible. Firstly, the bias could have been reduced by the so-called anchoring effect. The anchoring effect describes people’s tendency to anchor estimations to certain numbers. In many situations, individuals make estimations by adjusting a certain initial value. Wherever this initial value originates from, the adjustments made to those values are usually insufficient to yield a correct estimate. Hence, individuals make estimations that are biased towards such initial values. In the questionnaire, subjects were asked to rate absolute and relative individual performance first, before rating absolute and relative group performance. Therefore, ratings of absolute individual performance may have functioned as the initial value (e.g. the anchor) and the estimations of group performance therefore may have been biased towards the estimates of personal performance. Group performance was significantly better than average individual performance. However, if estimations of group performance were anchored by the beliefs of individual performance, those estimations would not have increased enough to match the actual increase in performance. Overestimation of group performance then, would be lower than overestimation of individual performance. The anchoring effect would also explain why top-performing individuals overestimated group performance more than they overestimated individual performance. Because they did not adjust group performance ratings downwards enough, due to the anchoring effect, and probably the burden of expertise (e.g. believing all other subjects performed at at least a similar level), their overestimations of group performance were higher than those of individual performance.
