Risky behavior of entrepreneurs : the moderating role of the representativeness heuristic in the relationship between overprecision and explorative activity

(1)

RISKY BEHAVIOR OF

ENTREPRENEURS

The moderating role of the representativeness heuristic in

the relationship between overprecision and explorative

activity.

Björn Chin Fo Sieeuw (0125334)

Master Thesis (version 1) Supervisor: G. T. Vinig MSc Entrepreneurship Universiteit van Amsterdam/Vrije Universiteit 7-1-2018 “Information about the distribution is accumulated over time, but choices must be made between gaining new information about alternatives and thus improving future return […]”

(2)

1 Statement of Originality

This document is written by Student Björn Chin Fo Sieeuw who declares to take full responsibility 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)

2 Preface

Thank you for reading this master thesis, which I have written before graduating in the field of entrepreneurship. This thesis is about the role representativeness plays in the established negative relation between overprecision and explorative activity. The causality between the variables is explained by looking at the degree of risk someone perceives, as this is a personal characteristic.

The cover page contains a quote from James G. March, the originator of the famous

distinction between organizational exploration and exploitation. For the research that led to this thesis, a computer experiment is done to assess the explorative behavior of the subjects. Although March associates risk-taking with exploration, in this experiment a low exploration score points to taking much risk. This is because with little explorative activity, subjects start exploiting the bandit machine early on while they have little information about them.

This study is based on empirical quantitative research. I had a tough time for both implementing the computer experiment and finding enough respondents for my data collection. In the end I managed to get the program working and find the requested respondents. Most of the respondents are entrepreneurs in the surrounding area of Amsterdam. I would like to thank all the participants, many of which are a member of Business Network International. Finally, I would like to thank my supervisor, mr. Vinig, for the time and effort he dedicated in the process of writing my thesis.

I wish the reader the best during and after reading my work.

Björn Chin Fo Sieeuw

Master student Entrepreneurship

Joint Degree Universiteit van Amsterdam/Vrije Universiteit

The copyright rests with the author. The author is solely responsible for the content of the thesis, including mistakes. The university cannot be held liable for the content of the author’s thesis.

(4)

3

activity. Overprecision indicates a risk-taking personality, while representativeness indicates a low risk-perception. Entrepreneurs show a large amount of explorative activity, and this is associated with taking less risk. This is consistent with prior literature, which indicates that the lower risk-aversion of entrepreneurs is not an explanation for the pursuit of an

opportunity.

The results imply that the moderator representativeness strengthens the negative relationship between overprecision and explorative activity. Without the representativeness variable, entrepreneurs seem almost twice as sensitive for decreased explorative activity than non-entrepreneurs when overprecision is higher. An explanation could be that non-entrepreneurs perceive the risk of an opportunity as lower than non-entrepreneurs. Entrepreneurs scoring above-median on representativeness are slightly more sensitive for explorative activity than entrepreneurs scoring below-median, for increased overprecision levels. This strengthening effect of the representativeness moderator for entrepreneurs could be explained by their low risk perception when making decisions. Above-median representativeness causes non-entrepreneurs to be almost twice as sensitive for decreased explorative activity than entrepreneurs, for increased overprecision levels. Experiental learning, which takes place more rapid for entrepreneurs than for non-entrepreneurs, is suggested as a possible explanation.

Theoretically, representativeness could partly explain the negative relationship between overprecision and explorative activity. Practically, HR-managers can read the tolerance towards risk of future employees from this test, and nascent entrepreneurs can use this test as a self-assessment tool to get insight in their own risk perception.

(6)

5 1. Introduction

The current study will reveal whether there is a moderating role of the representativeness heuristic on the relationship between overprecision and explorative activity. These constructs will be measured using a series of computer experiments. This thesis can be perceived as a contribution to entrepreneurship research specifically, because of the way explorative activity is measured. When gambling using a one-armed bandit, people try to maximize their profit after they have explored the behavior of the slot-machine. The trade-off between exploration and exploitation is a proxy of the degree of entrepreneurial personality of people. When striving to an optimal financial performance, people should manage their investment and payout well, in order to finally win money.

Further, previous research has scrutinized the influence of the representativeness heuristic on exploitative behavior of entrepreneurs (De Carolis & Saparito, 2006). This heuristic is the tendency of people to categorize a sample to a resembling population. The influence of the representativeness heuristic on the main relationship between overprecision and financial performance is not yet proven irrefutably (Biais, Hilton, Mazurier & Pouget, 2005).

The contribution to the literature of the current study is to determine if there is a moderating role of the representativeness heuristic on the relationship between overprecision and explorative activity. Other previous research has indicated there is a negative relationship between overprecision and financial performance in experimental situations (Glaser, Langer & Weber, 2013; Herz, Schunk & Zehnder, 2014).

1.1 Research question

The research problem of this thesis is showing that there does exist a strengthening moderating role of the representativeness heuristic. This leads to the following research question:

How is the relationship between overprecision and explorative activity moderated by representativeness?

(7)

6 In order to answer the research question, three sub-questions are formulated:

• How are the three constructs measured in an experimental session?

• How does the main relationship between overprecision and explorative activity look like?

• Does the representativeness heuristic influence the main relationship positively or negatively as a moderating variable?

1.2 Theoretical and practical relevance

Anderson and Sunder (1995) have compared student traders and professional traders in which a laboratory market is used as an indicator of financial performance. Biais, Hilton, Mazurier and Pouget (2005) have scrutinized the influence of self-monitoring on the relationship between overprecision and financial performance. Camerer (1987) states that the winner’s curse in financial performance testing of auctions could be caused by the representativeness heuristic. The theoretical relevance of the current study is that the influence of the

representativeness heuristic as a moderating variable on the relationship between overprecision and explorative activity is understudied; at least it has not formerly been hypothesized in the context of the research domain of entrepreneurship. Skala (2008) argues searching for the origins of the miscalibration type of overconfidence is relevant in

economics; the representativeness heuristic might be one of the measures mitigating or spurring miscalibration.

The practical relevance of assessing an individual’s score on the exploration-exploitation continuum is that this score tells something about his personality traits. More exploring managers often have an open mind, while more exploiting managers often take well-thought decisions (Keller & Weibler, 2014). The openness or consciousness of someone’s personality is important to know for HR-managers before someone is given a certain job, and the

testability of these personality dimensions makes it a handy tool for them. The fact that overprecision is a negative indicator of explorative behavior (Glaser, Langer & Weber, 2013; Herz, Schunk & Zehnder, 2014) makes testing overprecision practically relevant for the same reason. For example, asking how many tennisballs fit in an airplane during an application

(8)

7 conversation not only assesses logical thinking of the applicant; it also tests whether s/he is overprecise, which is one of the dimensions of overconfidence.

The practical relevance of testing representativeness and determining its role towards the overprecision-financial performance relationship, is that HR-managers can conclude from that how someone makes decisions. Individuals showing a high representativeness bias choose for matching occurences of events with appearing similar representations (Tversky & Kahneman, 1972; Grether, 1980; Grether, 1990). This could have predictive value for someone’s

decisional behavior in a certain work-related situation. Besides using the tests in a corporate environment, nascent entrepreneurs can use the tests to obtain insight in their own risk perception. This can save them from chasing risky opportunities.

In the following chapter, the theoretical background of the current study is provided. Key terms are defined, and the literature review and hypothesis are given. The third chapter contains the methodology of the current study. Because of the nature of the hypothesis, a quantitative research method is chosen. Chapter 4 contains the results of the analysis,

including the outcomes of the moderated regression, and the reliability tests if applicable. The last chapter starts with a discussion of the results of the analysis, provides limitations and avenues for future research, and ends with a conclusion.

(9)

8 2. Theoretical background

2.1 Key terms

• Ambidexterity is the ability of an organization to simultaneously pursue both

explorative (discontinuous) and exploitative (incremental) innovation (Junni, Sarala, Taras & Tarba, 2013).

• Calibration is the measurement of the validity of probability assessments

(Lichtenstein, Fischhoff & Phillips, 1976). In the current study, the calibration for probabilities assigned to uncertain continuous quantities is determined.

• Experiental learning is the transformation process of (prior) knowledge into newly created and re-created knowledge structures (Holcomb, Ireland, Holmes Jr. & Hitt, 2009). It is provided as an alternative explanation for entrepreneurs’ early-stage exploitation, besides the representativeness heuristic.

• Explorative activity is the number of rounds a subject is exploring the behavior of the three slot machines in the multi-armed bandit game (McCaffrey, 2016). This variable is used as a proxy of entrepreneurial behavior.

• Financial performance is the profit obtained in Biais, Hilton, Mazurier and Pouget’s (2005) trading game, Ederer and Manso’s (2013) lemonade stand game, and Herz, Schunk and Zehnder’s (2014) ice-cream game. It is the result of the strategy of exploration or exploitation, which have a certain profitability. In the current study, financial performance is the profit obtained in the multi-armed bandit game (Audibert, Munos & Szepesvári, 2009).

• Overprecision (also called judgemental overconfidence (Herz, Schunk & Zehnder, 2014), or overconfidence in judgement (Biais, Hilton, Mazurier & Pouget, 2005) is “the excessive certainty regarding the accuracy of one’s beliefs” (Åstebro, Herz, Nanda & Weber, 2014: 58). This appears from the tendency to set narrow intervals for one’s own answers to trivial questions.

• Representativeness is the bias that subjects categorize occurrences according to the principle of similarity. Textual as well as numerical experiments have been conducted that point to the representativeness heuristic. Textual experiments include the decision subjects have to make whether a description of a person is a lawyer or an engineer.

(10)

9 Numerical experiments include the assessment of chance of repeatedly flipping a coin (Tversky & Kahneman, 1972).

• Risk perceptions are “the beliefs about the magnitude of potential losses associated with a particular business situation” (De Carolis & Saparito, 2006). In the current study, the particular business situation is simulated by offering a choice in the

representativeness test. High representativeness is explained by a low risk perception.

2.2 Literature review

Overprecision is the third form of overconfidence, along with overestimation and overplacement. Overestimation is a person’s tendency to estimate his own ability in an optimistic manner, while overplacement is an individual’s tendency to assess his own skills on a higher level than that of other people (Åstebro, Herz, Nanda & Weber, 2014). Biais, Hilton, Mazurier and Pouget (2005) show that miscalibration of the asset value in ‘the trading game’ leads to the ‘winners curse’, and thereby to lower profits. Here they define

miscalibration as the tendency to overestimate the precision of one’s own information. However, they do not propose an explanation for this phenomenon in terms of the trade-off between exploration and exploitation. Overprecision in ‘the ice-cream game’ seems to be negatively related to explorative activity. As such, it is negatively related to financial performance (Herz, Schunk & Zehnder, 2014).

Anderson and Sunder (1995) have conducted a trading experiment on professional traders and students. They conclude that trading experience influences the Bayesian behavior of the subjects. For instance, they have determined the representativeness heuristic described by Tversky and Kahneman (1974), and it appears that professional traders suffer less from this bias. In the current study, a widely used experiment, devised by Grether (1980), is used to measure the representativeness heuristic. This experiment is an operationalization of the representativeness heuristic described by Tversky and Kahneman (1972). The reason Grether’s (1980) experiment is used instead of Tversky and Kahneman’s (1972) decision experiment whether a person’s description concerns a lawyer or an engineer, is that the former can be perceived of as more objective than the latter.

Further, Taktak and Triki (2014) have assessed the level of all three overconfidence types, as well as the risk aversion, of students and expert entrepreneurs. They collected data from an

(11)

10 investing game which consisted out of three parts: identification, evaluation and exploitation. The outcome was that more overconfidence, for example by miscalibration, leads to less risk aversion in the investing game. In the current study, only two phases are distinguished: exploration and exploitation. It is expected that a higher overprecision level leads to less explorative behavior, which is an indication of less risk aversion. In contrast, less

overprecision is expected to be related to more explorative behavior, which can be perceived as a more risk averse attitude.

While the overprecision research originates from psychology, its application in the field of finance is becoming much more common from the 1990’s onward. In finance, the temporal delay of execution of stock options is used as a proxy of overconfidence (Skala, 2008). An example of an experiment about the relationship between overprecision and trading activity is Fellner-Röhling and Kruger (2014). These researchers showed that overprecision on general knowledge questions does not have an impact on trading activity. However, overprecision on the interpretation of signals does have impact, as subjects who were not given information about the distribution of the outcome variable showed more overprecision than subject who were given this information. Also, gender differences do play a role, as men do show more overconfidence when trading than women, for high levels of risk aversion.

The gap in the literature that will be addressed is that the effect of the representativeness heuristic on financial performance is never studied before in past entrepreneurial

overprecision research with psychometric valid results. Biais, Hilton, Mazurier and Pouget (2005) state this as a possible avenue for future research. The hypothesis to be tested is that there is a moderating relationship of representativeness between overprecision and financial performance. The more representative bias participants show, the lower the degree of exploration. This might be the case because these people are more rigorous in their decision making. Keller and Weibler (2014) provide evidence for the assumption that managers who are more explorative score higher on openness, whereas managers who are more into exploitation, have a more conscious personality.

2.3 Hypothesis

(12)

11

The representativeness heuristic has a strengthening moderating role on the negative relationship between overprecision and explorative activity.

The conceptual model is shown in Figure 1.

Figure 1: The moderating relationship of representativeness between overprecision and explorative activity.

Overprecision Explorative Activity

(13)

12 3. Methodology

3.1 Data collection

3.1.1 Research characteristics

This study is testing theory in a deductive way and can therefore be called a quantitative study. The used sampling method can be described as restricted systematic sampling

(Blumberg, Cooper & Schindler, 2014); most of the participants are active self-employed in the broadest sense of the word. The number of subjects in the sample is 50, so the results of the regression will be unbiased. The accuracy of the sample will be secured by drawing participants properly from the population; for example, their age will contain enough variance. The precision of the sample, measured by the variance of for instance gender, will be secured because the population will be reflected in the sample that contains approximately an equal number of men and women.

The relationship between overprecision, representativeness and explorative activity is more a variance-theory than a process-theory, as the strength of the relationship between the

constructs is assessed, and there is not sought for the deeper reasons why the relationship occurs (Van de Ven, 2007). The variables are measured on the individual level, as the behavior of individual people in the test is measured. The research design is explanatory, which is used to test the relationship between the variables using an experimental design. The goal is to validate the theory, by looking for similarities among participants. Theory is

generated by generalizing over their measured behavior.

Overprecision is a proximal variable because the ‘guestimates’ of the large quantities is close to the way the participant is calibrated. Also, miscalibration has been shown to fluctuate over time (Skala, 2008). Representativeness is a distal variable because the behavior in making the decisions is a proxy of the way the participant is influenced by the given information.

Explorative behavior is a proximate variable because the number of rounds a participant is exploring the multi-armed bandits is an exact measure of his behavior. All three variables are reflective indices, rather than formative ones, as the consequences of the participants’

behavior is measured, while the antecedents of their behaviors are not unveiled (Davidsson, 2004).

(14)

13 3.1.2 Assessment of overprecision

In order to assess the overprecision tendency of the subjects, they will be asked to fill in a questionnaire with estimation questions. These questions will be comparable to the

questionnaire used by Biais, Hilton, Mazurier and Pouget (2005), Alpert and Raiffa (2011), and Herz, Schunk and Zehnder (2014). To answer the estimation questions, a lower- and upperbound of the estimated interval will be filled in. Using these intervals, the miscalibration will be calculated by determining the number of incorrect answers (Alpert & Raiffa, 1982; Glaser, Langer & Weber, 2013). See Appendix 1 for the questions used in this experiment. Besides the overprecision questionnaire, two computer experiments (Blumberg, Cooper & Schindler, 2014) will be conducted: Grether’s (1980, 1990) bingo cage test to assess the representativeness bias of the subjects and a multi-armed bandit game to assess the

explorative activity of the subjects. While the bingo cage game is the easiest one described by Grether (1980, 1990), the multi-armed bandit game is a simplified version of the one

described by Audibert, Munos and Szepesvári (2009). The environment in which these experiments will be implemented is z-Tree (Fischbacher, 2007).

3.1.3 Assessment of representativeness

In the bingo cage test the representativeness scores of the subjects will be used as moderating variable (El-Gamal & Grether, 1995). Because Grether (1980) indicates people show less representativeness bias when they are incentivized for this test, in the current study people will be paid in the representativeness test. Further, Harrison (1994) has tested whether pay-for-performance in Grether’s (1080) representativeness test influences the test results. Using a fixed-pay and pay-for performance group, he had the participants choose the right cage 20 times. It turned out that pay-for-performance yields better results than fixed-pay in this experiment (Hertwig, 1998). In the current experiment, participants will choose the right cage 30 times, with randomly generated draws. The moderating role of the representativeness heuristic will be tested using SPSS (Aguinis, 2004).

(15)

14 3.1.4 Assessment of explorative activity

Whereas overprecision and the representativeness heuristic are assessed in the way older research has done this, the trade-off between exploration and exploitation is assessed using a multi-armed bandit game (Audibert, Munos & Szepesvári, 2009). The explorative activity is used as a proxy of profitability because the profit is dependent on the measure of exploration. The multi-armed bandit game is a game in which there are at maximum 20 rounds of

exploration, and after a number of times of exploration, the remaining rounds will be

exploited. Of the three bandits, the best one according to the available information until then is exploited. McCaffrey’s (2016) implementation is used specifically. In the multi-armed bandit game, the variances of the explorative activities (Junni, Sarala, Taras & Tarba, 2013) during the treatment will be used as the dependent variable. In fact, because the experiment is done twice, the average of these variance scores will be used in the regression.

Managers who are more engaged into exploration instead of exploitation, show a more open personality (Keller & Weibler, 2014). When measuring the exploration-exploitation trade-off in an experimental setting, the subject can be perceived as the manager of his own financial performance. The less explorative activity some subject exposes in a computer experiment, the less innovative activity s/he shows (Herz, Schunk & Zehnder, 2014). As such, explorative activity can be used as a proxy of entrepreneurship (Åstebro, Herz, Nanda & Weber, 2014). Concerning the insights of Suzuki (2016), the exploration-exploitation experiment in the current study can be thought of as both a and a stage-perspective. The choice-perspective applies because the resources of the remaining rounds can either be used to explore or exploit the bandits. As such, the exploration-exploitation trade-off can be seen as consisting out of a dichotomous choice. The stage-perspective applies because the possibility of exploitation arises after the subject has explored the bandits for a certain number of rounds. This implies the sequential staging of the exploration-exploitation decision.

(16)

15

3.2 Moderated regression

3.2.1 Independent variable

In the overprecision questionnaire, ten general knowledge questions will be asked which the subjects must answer with 90% confidence. The boundaries between which the answer lies must be provided in the unit given (for example hundreds of millions). When a subject is well-calibrated, s/he will not show overprecision, and have at maximum one wrong answer (10% of 10 answers = 1 wrong answer). This answer is said to fall outside the 90%

confidence interval. When a subject is wrongly calibrated, s/he will show overprecision, and have at maximum one right answer (Glaser, Langer & Weber, 2013). The number of wrong answers of each subject is taken as independent variable of the linear regression.

3.2.2 Moderation variable

In the representativeness experiment, there are 7 possible outcomes, which differ in the number of N’s: 0 to 6 N’s. Each of these outcomes can be gotten from cage A or cage B. The probability of using cage A (cage B) can be .67(.33), .5(.5) or .33(.67). This probability is called the prior. The posterior is the probability with which subjects choose for cage A or cage B, given the outcome and a given prior (Grether, 1980; Grether, 1990; El-Gamal & Grether, 1995). For the analysis of the representativeness heuristic, the value 1 is assigned to choices which reflect representativeness (if the draw looks like the content of cage A or B), and the value 0 is assigned to choices which resist representativeness (if the choice is led by the given base rates). The average of these values is calculated, and subtracted from the average of the posteriors, which yields the representativeness score that is used in the regression. The subjects will be asked to judge 30 draws. The moderating representativeness scores will be converted to a dummy whereby the scores above the median will have the value 1 and the scores under the median will have the value 0. This representativeness dummy will be used to calculate the moderating variable in the linear regression.

(17)

16 3.2.3 Dependent variable

The explorative behavior of the participant is tracked using a multi-armed bandit experiment, in which there are at maximum 20 rounds of choice between exploring or exploiting them. When the subject chooses another exploration round, the number of rounds left for

exploitation decreases with one. When the subject chooses exploitation, the number of rounds left will all be exploited; this means the exploration phase is over (McCaffrey, 2016). The number of exploration rounds of the exploration-exploitation trade-off will be used as the dependent variable in the linear regression.

3.3 Validity and reliability

In this section, the validity and reliability of the overprecision, representativeness and

explorative activity measures is discussed. Where possible, the exact value will be calculated; otherwise the literature about the specific topic will be treated.

3.3.1 The validity of the overprecision data

Gigerenzer, Hoffrage and Kleinbölting (1991) prompt participants to make a decision on questions that involve comparative considerations about numerical values. Their concept of cue validity helps the participants to reason about these quantities so the right decision is made more often. For example, when the question is ‘which city (city A or city B) has more inhabitants?’, it is possible that the right city is picked of all cities (the Brunswikian (1955) ‘reference class’) by help of a cue; this cue could be the consideration which of the two cities has a football club. Obviously, the city that does not have a football club is smaller in the number of inhabitants. Cue validity in the probabilistic mental model, helping humans with numerical considerations, is what ecological validity would be for a real-world problem (Özarslan, 2016). According to Gigerenzer (1984) ecological validity, the question whether laboratory results are real world problems, lies close to external validity, the question whether the results may be generalized for people in the real world. Unfortunately, the literature does not provide a method for ecological or external validity to assess it for Alpert and Raiffa’s (2011) overprecision measurement.

(18)

17 3.3.2 The reliability of the overprecision data

Alpert and Raiffa have repeatedly shown that their students are overprecise in providing 90% confidence intervals of ten general knowledge questions. Therefore, the reliability of the outcome that the percentage of right answers lies below 60% (and sometimes even below 30%) can be called quite high (Moore, Tenney & Haran, 2015). From another perspective, an exact measure of the split-half reliability of the overprecision data can be given by using the Kuder-Richardson’s formula 20 (Cureton, 1966). This measure is suitable for non-Likert scale datasets that contain only zero’s and one’s, and is related to Cronbach’s α. The value lies between 0 and 1, in which values closer to 1 indicate more reliable datasets.

3.3.3 The validity of the representativeness data

In the representativeness experiment, the given prior is an example of a base rate, while the percentage of right decisions indicates the hit rate. When choosing for cage A or cage B, participants tend to ignore the base rates, lowering the validity of their choices. Also, a high base rate is reducing the validity of the study because hit rate then also increases (Einhorn & Hogarth, 1978).

Multiple previous studies by Kahneman and Tversky have shown that the assumption of rational economic agents underlying economics research does not hold. Instead they propose that for instance the representativeness heuristic influences subjects’ performance in decision tasks. The reply of the experimental economy field is that people learn by doing, lowering the internal validity of Kahneman and Tversky's findings. Because people are smart beings, the deceit of the experimental design is well understood by the subjects, making the

representativeness measurement internally invalid; they might not measure representativeness as they wanted to measure it. Besides this, the external validity of Kahneman and Tversky's results is questionable. The results might not be generalizable outside the laboratory to the whole human population, because the representativeness experiment only measures the decisional behavior of the participants (Hertwig, 1998). In order to increase the validity of the results, the participants will be paid for their performance in the representativeness task. Directly after each choice they will be informed about whether they have taken the right decision.

(19)

18 The ecological validity of experiments using base-rates, such as Grether’s (1980, 1990) experiment, suffers from the fact that decision making in the laboratory is not a real-world problem. People exposed to such experiments might show different errors in their decision making when being out of the laboratory. This flaw causes the representativeness experiment, testing the specific cognitive illusion of representativeness, to be called not ecologically valid (Koehler, 1996).

3.3.4 The reliability of the representativeness data

Gilovich, Griffin and Kahneman, (2002) argue that participants are insensitive to base rates, which not only decreases the validity, but also the reliability of their choices. They tend to make a decision based on the impression of the evidence that a certain draw came from a certain cage, instead of thinking about the relationship between the evidence and the target event of the draw.

Grether’s (1980, 1990) bingo cage experiment is a variant of his earlier experiment in which cup A contained two red and one blue marbles, and cup B contained one red and two blue marbles (Grether, 1978). When representativeness suggested the same answer as the draw, participant’s scores of choosing the right cup was 80%, which dropped to 60% in case representativeness suggested a different answer than the draw. Hammerton’s (1973) similar test also provided some context compared to Grether’s (1978) experiment. Holt and Smith (2009) assert that the comparable drop from 80% to 60% in Hammerton’s (1973) experiment seemed to be due to providing two different contexts. Context is detrimental to the

participant’s perceived reliability of the test, making Grether’s (1978, 1980, 1990) experiment more reliable than Hammerton’s (1973) experiment.

3.3.5 The validity of the exploration data

In order to give the participants the opportunity to learn from their decision making (Hertwig, 1998), the multi-armed bandit test will be done twice. This will increase the validity of the test results, because participants will understand the test better the second time. There will be looked for a learning effect by looking at the payout of both attempts.

(20)

19 3.3.6 The reliability of the exploration data

The intra-rater reliability of the two datasets of explorative behavior will be determined (Gwet, 2008). The purpose of this is to validate that the two datasets are internally consistent, meaning that the first and second measure of explorative behavior do not significantly differ.

(21)

20 4. Results

4.1 Descriptive statistics

See appendix 2 for all descriptive statistics. The number of respondents was N = 50. Of the respondents, 60% was male and 40% female. Their mean age was 44.2 (SD = 13.3; N = 47), and the most respondents (frequency = 12) were between 55 and 60 years old. 42% of the respondents finished their education at the university level, and 26% did so at the level of the university of applied sciences. 12% of the respondents finished high school. At least 76.9% of the respondents were Dutch; the nationality of 13.5% of the participants is unknown, but it is possible that a large part of these people is also Dutch. There are a couple of non-Dutch nationalities represented in the dataset. The percentage of entrepreneurs is 74%, the percentage of employed people 18%, and the percentage of students 8%.

The average number of wrong answers out of 10 in the overprecision test is 8.8 (SD = 1.4). The fact that this score approaches 9 wrong answers indicates that most respondents were overprecise, because the intervals were given with 90% confidence. The average

representativeness score is 27.9 (SD = 15.4). There were only 4 respondents who had a negative representativeness score. This anti-representativeness bias represents the decision skill to ignore the fact that a sample appears to be drawn from a certain population (Wickham, 2003). The average number of rounds people used to explore until choosing exploit is 6.7 (SD = 4.9). This value indicates a low explorative personality of the respondents, so it cannot even be called ambidextrous behavior.

4.2 Moderated regression

The median representativeness score is 29.5; this means that half of the respondents shows representativeness below this score and half of the respondents shows representativeness above this score. Dummy variables were used for all the below-median entrepreneur group, the above-median entrepreneur group, the below-median non-entrepreneur group and the above-median non-entrepreneur group. Moderated regressions were carried out in SPSS, in which the interaction term was the product of these dummies and the overprecision score. Table 1 contain the results of the regressions of a direct relationship between overprecision and explorative activity for non-entrepreneurs and entrepreneurs (model [1]). This model does

(22)

21 not include representativeness as a variable. Table 1 also contains regressions with below- and above-median representativeness as the moderator, both for entrepreneurs (models [2.1] and [2.2]) and non-entrepreneurs (models [3.1] and [3.2]). See Appendix 2 for all SPSS regression output, including graphs.

Model Variable Coefficient Significance

Standard Error 1 Intercept 7.60 0.390 8.75 Overprecision -0.25 0.796 0.96 Entrepreneur (dummy) 3.62 0.723 10.15 Overprecision ˟ Entrepreneur (dummy) -0.21 0.85 1.13 2.1 Intercept 12.02 0.24 9.92 Overprecision ˟ Representativeness

below-median (for entrepreneurs) -0.47 0.68 1.11

2.2 Intercept 11.28* 0.07 5.90

Overprecision ˟

Representativeness

above-median (for entrepreneurs) -0.54 0.44 0.68

3.1 Intercept -2.92 0.80 10.94

Overprecision ˟

Representativeness

below-median (for non-entrepreneurs) 0.76 0.54 1.17

3.2 Intercept 15.54 0.23 11.04

Overprecision ˟

Representativeness

above-median (for non-entrepreneurs) -1.02 0.47 1.26 * p < 0.1

(23)

22 The values in Table 1 lead to the following regression equations:

Non-entrepreneurs without representativeness moderator:

Expl = 7.6 – 0.25 ˟ OP [1.1]

Entrepreneurs without representativeness moderator:

Expl = 11.22 – 0.46 ˟ OP [1.2]

Entrepreneurs scoring below-medial on representativeness:

Expl = 12.02 – 0.47 ˟ OP ˟ ReprBelowMed [2.1]

Entrepreneurs scoring above-medial on representativeness:

Expl = 11.28 – 0.54 ˟ OP ˟ ReprAboveMed [2.2]

Non-entrepreneurs scoring below-medial on representativeness:

Expl = –2.92 + 0.76 ˟ OP ˟ ReprBelowMed [3.1]

Non-entrepreneurs scoring above-medial on representativeness:

Expl = 15.54 – 1.02 ˟ OP ˟ ReprAboveMed [3.2]

The abbreviations in the formulas are: Expl = Explorative activity score OP = Overprecision score

Repr = Representativeness score

BelowMed = Representativeness score below the median AboveMed = Representativeness score above the median

Equations [1.1] and [1.2], derived from model [1] can be compared. Without testing for representativeness, both the non-entrepreneur 0.25, p = 0.80, N = 13) and entrepreneur (-0.46, p = 0.85, N = 37) group shows a negative relationship between overprecision and

(24)

23 explorative activity. This indicates that entrepreneurs are more sensitive for decreased

explorative activity when their overprecision level is higher. Proposition 1 can now be formulated as follows:

Entrepreneurs are almost twice as sensitive for decreased explorative activity than

non-entrepreneurs when overprecision is higher. [P1]

The R2_{of the linear regression for the entrepreneur group is 0.016, which means that only}

1.6% of the variance is explained by the regression line. The R2_{of the linear regression for the}

entrepreneur group is 0.008, indicating that only 0.8% of the variance is explained by the regression line. The F(3, 46)-value is 0.703, not indicating that model [1] is an exceptionally good way of describing the data.

In model [2.1] only 1.1% of the variance of explorative activity of entrepreneurs is explained by the overprecision score and a below-median representativeness score (R2_{= 0.011). The}

F(1, 16)-value is 0.182, meaning that model [2.1] is not an extremely good model of the relationship. In model [2.2] only 3.6% of the variance of explorative activity of entrepreneurs is explained by the overprecision score and an above-median representativeness score (R2₌

0.036). The F(1, 17)-value is 0.633, meaning that model [2.2] is a mediocre model of the relationship.

In model [3.1] the R2_{and F-values for non-entrepreneurs scoring below-median on}

representativeness are irrelevant because of the strange positive relationship between overprecision and explorative activity. In model [3.2], 14% of the variance of explorative activity of non-entrepreneurs is explained by the overprecision score and an above-median representativeness score (R2_{= 0.140). The F(1, 4)-value is 0.651, meaning that model [3.2] is}

a mediocre model of the relationship.

The equations derived from models [2] and [3] can be compared, yielding four comparisons: • For the entrepreneurs group, there can be made a distinction between participants

scoring below-median (N = 18) and above-median (N = 19) on representativeness; in this case the slopes of equations [2.1] and [2.2] are compared. For entrepreneurs, above-median representativeness scores (-0.54, p = 0.44) lead to a slightly larger coefficient for overprecision than for below-median representativeness scores (-0.47, p = 0.68). In other words, representativeness has a moderating influence on the

(25)

24 relationship between overprecision and explorative activity for entrepreneurs. This leads to a second proposition:

Entrepreneurs scoring above-median on representativeness are slightly more sensitive for explorative activity than entrepreneurs scoring below-median on

representativeness, for increased overprecision levels. [P2]

• Unfortunately, the same comparison for non-entrepreneurs cannot be done using equations [3.1] and [3.2], because the strange result of equation [3.1] makes this comparison impossible. Equation [3.1] has a positive slope coefficient for below-median representativeness (0.76, p = 0.54) for non-entrepreneurs. This outcome is obtained because of the small sample size of non-entrepreneurs scoring below-median on representativeness (N = 7).

• The strange result of equation [3.1] also makes it difficult to compare entrepreneurs and non-entrepreneurs scoring both below-median on representativeness; this would be to compare equation [2.1] with equation [3.1].

• The entrepreneur (N = 19) and non-entrepreneur (N = 6) groups scoring above-median on representativeness can be compared; this is to compare equations [2.2] and [3.2]. For above-median representativeness scores, the regression coefficients is twice as large for non-entrepreneurs (-1.02, p = 0.47) as for entrepreneurs (-0.54, p = 0.44). This means that non-entrepreneurs are more sensitive to lower exploration levels when their overprecision level increases, than entrepreneurs. Proposition 3 can now be formulated as follows:

Above-median representativeness causes non-entrepreneurs to be almost twice as sensitive for decreased explorative activity than entrepreneurs, for increased

overprecision levels. [P3]

Apparently, proposition [P1] and [P3] contradict each other; according to [P1], entrepreneurs are (almost) twice as sensitive to decreased explorative activity than non-entrepreneurs for increased overprecision levels, while according to [P3] this seems to hold for

non-entrepreneurs. However, both propositions cannot be compared because in model [1] representativeness is excluded from the regression, and in model [3] it is included.

(26)

25

4.3 Reliability

For the overprecision experiment, the split-half reliability calculated using the Kuder-Richardson’s formula 20 is rKR20 = 0.48 (Cureton, 1966). As this value lies somewhat under

0.5, the split-half reliability of this dataset is almost reasonable.

For the two attempts of the exploration-exploitation experiment, the intra-rater reliability is γ = 0.37 (Gwet, 2008). Given this value, which is closer to 0 than to 1, the dataset cannot be called internally consistent. In other words, the first and second measure of explorative behavior do significantly differ. Because one of these respondents only did the exploration-exploitation test once, the intra-rater reliability of this test is based on the 49 remaining respondents.

(27)

26 5. Discussion and conclusion

5.1 Discussion

For two reasons, the computer experiment of the current study is relevant for the field of entrepreneurship research. The first reason is that gambling behavior when using a slot

machine indicates the explorative or exploitative activity of a person. When someone is acting more explorative, s/he possesses a more entrepreneurial character. When gambling, the person must manage his/her payouts well, in that winning money in the end is the ultimate goal. The second reason the current study is relevant for the field of entrepreneurship research is that representativeness is a predictor for explorative activity, and therefore for

entrepreneurship. De Carolis and Saparito (2006) argue that cognitive biases of entrepreneurs, such as representativeness, are positively influenced by trust and strong network ties. Their claim is that representativeness of non-random samples is negatively associated with risk perception in a given situation. In the current study, the opportunity is the decision that has to be made in the representativeness test. In particular, random samples are tested, being

sequences of six characters ‘N’ and ‘G’.

Entrepreneurs founding new ventures after seeing an opportunity cannot be explained by their lower risk aversion (Åstebro, Herz, Nanda & Weber, 2014). However, it can be explained by their lower risk perception; when entrepreneurs perceive low risk in the market, they are more inclined to start a new venture. Risk perception of entrepreneurs is negatively related to their degree of exploitation (De Carolis & Saparito, 2006). Following the exploration-exploitation trade-off of March (1991), this means it would be positively related to entrepreneurs’ degree of exploration. In contrast, a low risk perception would lead to a high degree of exploitation and low degree of exploration. In the current study, choosing for the representative cages indicates a stronger negative relationship between overprecision and explorative activity. Thus, the decreased degree of exploration can be explained by the lower risk perception towards choosing the representative cages.

The fact that both the entrepreneur and non-entrepreneur group show a negative relationship between overprecision and explorative activity is consistent with prior literature (Glaser, Langer & Weber, 2013; Herz, Schunk & Zehnder, 2014). In this case, representativeness was excluded from the regression. When someone fills in small intervals in the overprecision

(28)

27 experiment, s/he is overprecise and takes more risk in answering the questions. His/her risk-taking personality is also reflected in a low exploration score, because going to exploit a certain machine early on is a risky endeavor. Although laymen (students) and experts (entrepreneurs) do have different risk behavior, overconfident subjects in general are more reluctant to show risky behavior (Taktak & Triki, 2014).

According to proposition [P1], entrepreneurs are almost twice as sensitive for decreased explorative activity than non-entrepreneurs when overprecision is higher. An explanation for this could be that entrepreneurs are used to take risky decisions more often. For the

entrepreneurs holds that overprecision is a stronger predictor for less explorative activity because they perceive the risk as lower than non-entrepreneurs.

As Wickham (2003) remembers, Knight (1921) makes a distinction between risk and uncertainty. With limited information, entrepreneurs and managers make decisions with varying amounts of risk. Risk can be insured against, leading people to make different

decisions in different contexts (Wickham, 2003). In the current study, three experiments have been conducted that are related as to how people cope with risk. In particular, overprecise entrepreneurs have less tolerance for uncertainty and tend to take more risk in the answering of the ten general knowledge questions. Entrepreneurs showing the representativeness bias choose representative cages because they perceive this choice with low risk. And people who explore a slot machine very little take more risk in exploitation of this machine. The results of the entrepreneur group can be interpreted in the light of the distinction between risk and uncertainty.

According to proposition [P2], entrepreneurs scoring above-median on representativeness are slightly more sensitive for explorative activity than entrepreneurs scoring below-median on representativeness, for increased overprecision levels. In other words, there is a strengthening effect of the representativeness moderator for entrepreneurs. This can be explained by the low risk perception that is involved when showing high representativeness levels; a low risk perception also leads to a low exploration score. In contrast, for the entrepreneurs showing low representativeness levels, the negative relationship between overprecision and explorative activity is less strong. This can be explained by a somewhat higher risk perception when encountering the choice for a representative cage, and when being prompted to exploit the best one-armed bandit right away. So, the risk perception is the underlying mechanism of the relationship between high/low representativeness and low/high explorative activity.

(29)

28 As stated in proposition [P3], above-median representativeness causes non-entrepreneurs to be almost twice as sensitive for decreased explorative activity than entrepreneurs, for increased overprecision levels. Holcomb, Ireland, Holmes Jr. and Hitt (2009) suggest that experiental learning shapes one's knowledge, and that both knowledge and for example the representativeness heuristic influences decision making. They state the possiblility that entrepreneurial learning takes place more rapid than non-entrepreneurial learning, as an avenue for future research; judgemental processes of entrepreneurs could lead earlier towards expert performance because experienced entrepreneurs develop dynamic capabilities with each choice they have to make. This would aid future discovery and exploitation, and would for the current study explain the exploitation of the best bandit in an earlier stage.

The measured behavior in the second and third experiment do not measure constant traits. There appears to be a learning effect in both the representativeness test and the exploration-exploitation test. In the representativeness test, the average score of correct answers of the first ten choices is 6.24, while it is 6.34 for the second set of ten choices, and even 6.54 for the third set of ten choices. For the exploration-exploitation test, the first attempt yielded a

negative average score of -0.08 euro and the second attempt yielded a positive average score of 0.52 euro. These results indicate people learn from the feedback they get about their payout.

The representativeness heuristic, like the anchoring and adjustment heuristics, can be perceived of as not being specified precisely, and as such not falsifiable (Gigerenzer, 1996). Multiple challenging heuristics play a role which undermine Kahneman and Tversky’s ones, for example the recognition and fluency heuristics (Fiedler & Von Sydow, 2015). When Goldstein and Gigerenzer’s (2002) heuristics are applied to Grether’s (1980, 1990) experiment, the results of this test can be questioned.

Following the recognition heuristic, certain draws resemble earlier draws, and people tend to choose for the familiar option. When there are much of the same draws, the ecological validity of the test is high. In this case the number of correct decisions goes up, due to the recognition heuristic. Also, the recognition heuristic explains why people who show representativeness might score higher than people who don’t: sticking to a certain strategy while making decisions can be beneficial because processing too much cues at the same time gets the subjects confused. According to the fluency heuristic, people are learning to cope

(30)

29 with the test, in that they are increasingly choosing the right answers. These new heuristics tell us that people are learning by doing while the test is taking place.

5.2 Theoretical contributions and practical implications

The theoretical relevance of the current study is that it is suggested that the representativeness heuristic might be one of the explanations for the negative relationship between overprecision and explorative activity. As such, being able to state that representativeness plays a

moderating role, would be a major contribution to this understudied field of research. Several authors confirm that the assumption that entrepreneurs are characterized by a low

risk-aversion, is false (Palich & Bagby, 1995; Åstebro, Herz, Nanda & Weber, 2014). De Carolis and Saparito (2006) suggest low risk perception as an alternative explanation for high representativeness.

The practical relevance of this study is that various people can use the tests and draw conclusion about their loss aversive personalities. First of all, HR-managers can apply the overprecision and representativeness tests in their procedures when selecting candidates for a vacancy. These variables can be used to predict the explorative activity of a candidate, and how s/he copes with uncertainty and risky situations. For example, for corporate

entrepreneurship positions, highly explorative personalities might be required, while low representativeness scores might indicate the suitable candidates. Second, the tests conducted in the current study can be used as self-assessment tool for nascent entrepreneurs. In order to save time, effort and money spent on failing ventures, nascent entrepreneurs can test their representativeness to assess their risk perceiving attitude. People have to cope with cognitive biases, such as representativeness, when making decisions with either limited information or information overload. As representativeness influences the evaluation of opportunities, it is instructive for entrepreneurs to be conscious of this heuristic (De Carolis & Saparito, 2006).

5.3 Limitations and future research

The results of the current study are based on non-significant results, because of the small sample size (N = 50). For a typical moderation hypothesis to be validated with 95%

(31)

30 was divided in two sub-groups (entrepreneurs and non-entrepreneurs), the sample size for some regressions was even as low as N = 6. This affected the outcomes of the regressions. The validity of none of the variables could be determined because it was not clear from the literature how to do so. For Grether’s (1980, 1990) experiment, also the reliability could not be determined because random draws are generated which must be decided upon by the respondents.

Ozarslan (2016) proposes an enhanced way of testing overprecision by giving the participants feedback on the overprecision intervals. Because of simplicity of the computer

implementation in the current study, this way of assessing overprecision was not chosen. For similar reasons, tracking the behavior of the subjects by filming them or using an eye-tracker, like Payne & Braunstein & Carroll (1978) suggest, was not possible. Even recording the spoken words of the participants, so verbal protocols could be made of their decisional behavior (Payne, 1976), was not realistic.

Taktak & Triki (2014) have used an extended experiment to assess explorative behavior. In the current study, the exploration-exploitation test was kept simple by keeping it in these two stages instead of more stages. The test was designed this way because of the time limit of 15 to 20 minutes; when the test would take too long, participants would terminate their help. Fellner-Röhling & Kruger (2014) have directly assessed risk-aversive behavior in their experiment while the current study does not. They propose a categorization of risk-aversion on a 1-9 scale, in which 5 is risk neutral. It would have been useful to make a categorization of ambidextrous behavior on 1-20 scale, in which 7-13 is ambidextrous behavior. However, there was not found a method in the literature to do it this way. Ambidexterity, the balancing of exploration and exploitation, can be applied to individual people. Following Mom et al.’s ideas, ambidexterity can be defined as “behavioral orientation toward combining exploration and exploitation related activities within a certain period of time” (Bonesso, Gerli &

Scapolan, 2014). Another limitation is that the quadrant model of Bonesso et al. (2014) could not be applied to the current research because only the ambidextrous behavior of the

participants is measured; the ambidextrous perceptions should have been traced in depth-interviews, a method lying beyond this quantitative study.

Future research should take more variables into account, like the anchoring and adjustment heuristics (Kahneman & Tversky, 1974), and even the new heuristics: recognition and fluency (Gigerenzer, 1996; Gigerenzer & Goldstein, 1996). These variables could have a moderating

(32)

31 or mediating role on the relationship between overprecision and explorative activity, and even on each other. This makes this field of study a complex one. When the effect of these biases has been validated, they can be placed in a broader social perspective along with the

measurement of overconfidence (De Carolis & Saparito, 2006).

5.3 Conclusion

In the current study, the entrepreneurial behavior of people playing on a bandit slot machine is assessed. In particular, overprecision and representativeness were tested as predictors of explorative activity. The participants were split in two groups: entrepreneurs and non-entrepreneurs. The research question was:

How is the relationship between overprecision and explorative activity moderated by representativeness?

Despite learning effects in the representativeness test (Fiedler & Von Sydow, 2015), people undertaking the test might be biased towards choosing the representative population (content of the cages A and B) when being confronted with a sample (the draw of six balls). The ecological validity of the outcomes of Grether’s (1980, 1990) experiment could not be

determined. As such, it could both be the case that the outcomes of the test do or do not tell us something about how entrepreneurs make decisions in the real world.

Wickham (2003) carried out representativeness research in the Kahneman and Tversky’s way using vignettes. Because this experiment can be argued to be close to real world situations, the ecological validity is high. He concludes entrepreneurs and managers are subject to the

representative heuristic in their everyday working life. The current study has replaced Kahneman and Tversky’s way of testing representativeness by Grether’s (1980, 1990) experiment; in contrast to Kahneman and Tversky’s verbal test (choose whether the person described is a lawyer or an engineer), Grether’s test guarantees random draws (Harrison, 1994).

Three main conclusions can be drawn from the current study. Firstly, entrepreneurs are almost twice as sensitive for decreased explorative activity than non-entrepreneurs when

overprecision is higher. This could be the case because entrepreneurs perceive the risk of an opportunity as lower than non-entrepreneurs. Secondly, for increased overprecision levels, entrepreneurs scoring above-median on representativeness are slightly more sensitive for

(33)

32 explorative activity than entrepreneurs scoring below-median on representativeness. This strengthening effect of the representativeness moderator for entrepreneurs can be explained by the low risk perception towards highly representative opportunities. Thirdly, above-median representativeness causes non-entrepreneurs to be almost twice as sensitive for decreased explorative activity than entrepreneurs, for increased overprecision levels. The explanation for this should be sought in the fact that entrepreneurs’ expertise increases more easily than non-entrepreneurs when taking decisions. Therefore, non-entrepreneurs show less explorative activity in the experiment with the bandit slot machines.

The negative relationship between overprecision and explorative activity from the literature is supported (Glaser, Langer & Weber, 2013; Herz, Schunk and Zehnder, 2014): filling in small intervals indicates an overprecise personality, which is explained by little tolerance for uncertainty, a low risk-perception, and therefore less explorative behavior. The role

representativeness plays, is that choosing more often for the content of the cages means more gambling, and therefore more risky behavior is shown by entrepreneurs. As such,

representativeness seems to have a slight strengthening moderation effect on the negative relation between overprecision and explorative behavior.

(34)

33 6. References

Aguinis, H. (2004). Regression analysis for categorical moderators. New York, NY: Guilford Press.

Alpert, M., & Raiffa, H. (1982). A progress report on the training of probability assessors. In D. Kahneman, P. Slavic, L A. Tversky (Eds.), Judgment under uncertainty: Heuristics and

biases (pp. 294-305). Cambridge, England: Cambridge University Press.

Alpert, M., & Raiffa, H. (2011). Alpert-Raiffa Experiment #2 Spring 2010. Competitive Decision-Making and Negotiation. MIT OpenCourseWare.

Anderson, M. J. & Sunder, S. (1995). Professional traders as intuitive Bayesians.

Organizational Behavior and Human Decision Processes, 64(2), 185-202.

Åstebro, T., Herz, H., Nanda, R. & Weber, R. A. (2014). Seeking the roots of

entrepreneurship: insights from behavioral economics. Journal of Economic Perspectives, 28(3), 49–70.

Audibert, J.-Y., Munos, R. & Szepesvári, C. (2009). Exploration-exploitation tradeoff using variance extimates in multi-armed bandits. Theoretical Computer Science, 420, 1876–1902. Biais, B., Hilton, D., Mazurier, K. & Pouget, S. (2005). Judgemental overconfidence, self-monitoring, and trading performance in an experimental financial market. Review of

Economic Studies, 72, 287-312.

Blumberg, B., Cooper, D. R. & Schindler, P. S. (2014). Business Research Methods (4th

European ed.). London: McGrawHill.

Bonesso, S., Gerli, F. & Scapolan, A. (2014). The individual side of ambidexterity: do individuals’ perceptions match actual behaviors in reconciling the exploration and exploitation trade-off? European Management Journal, 32, 392-405.

Brunswik, E. (1955). Representative design and probabilistic theory in a functional psychology. Psychological Review, 62, 193–217.

Camerer, C. (1987). Do biases in probability judgement matter in markets? Experimental evidence. American Economic Review, 77, 981-997.

(35)

34 Cureton, E. E. (1966). Kuder-Richardson reliabilities of classroom tests. Educational and

Psychological Measurement, 26(1), 13-14.

Davidsson, P. (2004). Researching Entrepreneurship. New York: Springer.

De Carolis, D. M. & Saparito, P. (2006). Social capital, cognition, and entrepreneurial opportunities: a theoretical framework. Entrepreneurship Theory and Practice, 30(1), 41-56. Einhorn, H. J. & Hogarth, R. M. (1978). Confidence in judgment: persistence of the illusion of validity. Psychological Review, 85(5), 395-416.

El-Gamal, M. A. & Grether, D. M. (1995). Are people Bayesian? Uncovering behavioral strategies. Journal of the American Statistical Association, forthcoming.

Fellner-Röhling, G. & Kruger, S. (2014). Judgmental overconfidence and trading activity.

Journal of Economic Behavior and Organization, 107, 827-842.

Fiedler, K. & Von Sydow, M. (2015). Heuristics and biases: beyond Tversky and Kahneman’s (1974) judgment under uncertainty. In M. W. Eysenck, D. Groome (Eds.),

Cognitive psychology: Revisiting the classic studies (pp. 146–161). SAGE Publications Ltd.

Fischbacher, U. (2007). z-Tree: Zurich toolbox for ready-made economic experiments.

Experimental Economics, 10(2), 171–178.

Gigerenzer, G. (1984). External validity of laboratory experiments: the frequency-validity relationship. The American Journal of Psychology, 97(2), 185–195.

Gigerenzer, G. (1996). On narrow norms and vague heuristics: a reply to Kahneman and Tversky. Psychological Review, 103 (3), 592–596.

Gigerenzer, G., & Goldstein, D. G. (1996). Reasoning the fast and frugal way: models of bounded rationality. Psychological Review, 103 (4), 650–669.

Gigerenzer, G., Hoffrage, U. & Kleinbölting, H. (1991). Probabilistic Mental Models: a Brunswikian theory of conﬁdence. Psychological Review, 98(4), 506–528.

Gilovich, T., Griffin, D. & Kahneman, D. (2002). Heuristics and biases: the psychology of

(36)

35 Glaser, M., Langer, T. & Weber, M. (2013). True overconfidence in interval estimates:

evidence based on a new measure of miscalibration. Journal of Behavioral Decision Making, 26, 405–417.

Grether, D. M. (1978). Recent psychological studies of behavior under uncertainty. American

Economic Review, 68, 70–77.

Grether, D. M. (1980). Bayes rule as a descriptive model: the representativeness heuristic. The

Quarterly Journal of Economics, 95(3), 537–557.

Grether, D. M. (1990). Testing Bayes rule and the representativeness heuristic: some experimental evidence. Social Science Working Paper, 724.

Gwet, K. L. (2008). Intrarater reliability. In: N. Balakrishnan, T. Colton, B. Everitt, W. Piegorsch, F. Ruggeri, J. L. Teugels (Eds.), Wiley Encyclopedia of Clinical Trials.

Hammerton, M. (1973). A case of radical probability estimation. Journal of Experimental

Psychology, 101, 252–254.

Harrison, G. W. (1994). Expected Utility Theory and the Experimentalists. Empirical

Economics, 19, 223–253.

Holt, C. A. & Smith, A. M. (2009). An update on Bayesian updating. Journal of Economic

Behavior & Organization, 69, 125–134.

Hertwig, R. (1998). Psychologie, experimentelle Ökonomie und die Frage, was gutes Experimentieren ist. Zeitschrift für Experimentelle Psychologie, 45(1), 2–19.

Herz, H., Schunk, D. & Zehnder, C. (2014). How do judgmental overconfidence and over-optimism shape innovative behavior? Games and Economic Behavior, 83, 1–23.

Holcomb, T. R., Ireland, R. D., Holmes Jr., R. M. & Hitt, M. A. Architecture of entrepreneurial learning: exploring the link among heuristics, knowledge, and action.

Entrepreneurship Theory and Practice, 33(1), 167–192.

Junni, P., Sarala, R. M., Taras, V. & Tarba, S. Y. (2013). Organizational Ambidexterity and performance: a meta-analysis. The Academy of Management Perspectives, 27(4), 299–312. Keller, T. & Weibler, J. (2014). Behind managers’ ambidexterity – studying personality traits, leadership, and environmental conditions associated with exploration and exploitation.

(37)

36 Knight, F. H. (1921). Risk, uncertainty and profit. Houghton Mifflin, Boston, MA.

Koehler, J. J. (1996). The base rate fallacy reconsidered: descriptive, normative, and methodological challenges. Behavioral and Brain Sciences, 19, 1–53.

Lichtenstein, S., Fischhoff, B. & Phillips, L. D. (1976). Calibration of probabilities. The state of the art. Advanced Decision Technology Program. Cybernetics Technology Office Defense,

Advanced Research Projects Agency, Office of Naval Research, Engineering Psychology Programs.

March, J. G. (1991). Exploration and Exploitation in Organizational Learning. Organization

Science, Special Issue: Organizational Learning: Papers in Honor of (and by) James G. March,

2(1), 71–87.

McCaffrey, J. (2016). Test Run - The Multi-Armed Bandit Problem. MSDN Magazine, May 2016. Retrieved from: https://msdn.microsoft.com/en-us/magazine/mt703439.aspx

Moore, D. A., Tenney, E. R. & Haran, U. (2015). Overprecision in judgment. In G. Wu and G. Keren (Eds.), Handbook of Judgment and Decision Making. New York: Wiley.

Özarslan, A. (2016). Confidence in advice taking – How over- or underconfident judges and advisors coincide? Proceedings e-Book II: Business and Economics, 212–219. In Y.

Kahraman & E. Demirbas (Eds.). International Turgut Özal Congress on business economics

and political science: Competitive management and economic growth sustainable governance and democracy.

Palich, L. E. & Bagby, D. R. (1995). Using cognitive theory to explain entrepreneurial risk-taking: challenging conventional wisdom. Journal of Business Venturing, 10, 425-438. Payne, J. W. (1976). Task complexity and contingent processing in decision making: an information search and protocol analysis. Organizational Behavior and Human Performance, 16, 366—387.

Payne, J. W., Braunstein, M. L. & Carroll, J. S. (1978). Exploring predecisional behavior: an alternative approach to decision research. Organizational Behavior and Human Performance, 22, 17–44.

Skala, D. (2008). Overconﬁdence in psychology and finance – an interdisciplinary literature review. Munich Personal RePEc Archive, 32–50.

(38)

37 Suzuki, O. (2016). Revisiting the construct validity of exploration and exploitation. Social

Sciences Review, 21, 11–26.

Taktak, S. & Triki, M. (2014). The impact of overconfidence on entrepreneurial process: entrepreneurs versus students. British Journal of Economics, Management & Trade, 4(12), 1889–1904.

Tversky, L. A. & Kahneman, D. (1974). Judgment under uncertainty: heuristics and biases.

Science, New Series, 185(4157), 1124–1131.

Van de Ven, A. H. (2007). Engaged scholarship: a guide for organizational and social research. New York: Oxford University Press.

Wickham, P. A. (2003). The representativeness heuristic in judgements involving entrepreneurial success and failure. Management Decision, 41(2), 156–167.

Risky behavior of entrepreneurs : the moderating role of the representativeness heuristic in the relationship between overprecision and explorative activity