The relationship between unrealistic optimism
and risk preferences
An experimental study
Martina Sorbello 11387122
Master’s Thesis (15 ECTS)
MSc in Economics – Behavioural Economics and Game Theory University of Amsterdam
Supervisor: Dr. Ivan Soraperra
1 Statement of Originality
This document is written by the student Martina Sorbello 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.
Abstract
Unrealistic optimism has important consequences for economics since it leads to greater risk-taking. Similarly, the preferences toward risk determine the behaviour in risky situation. However, the relationship between these two notions has been rarely investigated in the literature and the results found are contrasting. This master thesis investigates this relationship experimentally, by eliciting risk preferences before and after a semantic priming of optimism. The results obtained do not find strong supporting evidence on the relationship between risk preferences and unrealistic optimism requiring further research. Nevertheless, this study highlights important weaknesses in the literature on unrealistic optimism with regards to the methodologies used to assess it providing helpful hints for researchers in the future.
2
Table of Contents
1. Introduction ... 3
2. Literature review ... 6
2.1 The definition and measurement of optimism... 6
2.2 The economics of optimism and risk preferences ... 8
2.3 Optimism and risk preferences: experimental evidence ... 9
3. Methodology ... 12
3.1 The Eckel and Grossman task ... 13
3.2 Priming optimism and pessimism ... 15
3.3 The investment task ... 16
3.4 Comparative optimism measure ... 17
3.5 Demographic information ... 18
3.6 Hypotheses ... 19
3.6.1 Primary Hypotheses ... 19
3.6.2 Secondary Hypotheses... 19
4. Results ... 20
4.1 Characteristics of participants and randomization check ... 20
4.2 Risk preferences in the E&G task ... 21
4.3 Priming optimism ... 22
4.4 Treatments effect and primary hypotheses ... 24
4.4.1 Regression analysis ... 25
4.5 The two measures of risk preferences ... 26
4.6 Secondary Hypotheses ... 28
5. Discussion ... 30
6. Conclusion ... 34
7. References ... 35
3
1. Introduction
Optimism is pervasive in nature and has been the focus of many areas of research from psychology to economics. A vast literature following the seminal paper of
Weinstein (1980), provides a variety of definitions of optimism1. The current study is
based on the definition of Weinstein (1980). Optimism is an unrealistic expectation of the likelihood of an event, denoting an erroneous judgment. Optimists believe to be likely to experience more positive events than negative events, the opposite holds for pessimistic people. This bias manifests even when people are given objective probabilities of events (Sharot et al., 2011). Unrealistic optimism is not found consistent over time, evidence provided by Helweg-Larsen and Shepperd (2001) shows that unrealistic optimism is effectively moderated by different moods and level of anxiety.
Assessing unrealistic optimism can be a challenging task since the measures used in the literature present several weaknesses and are often correlated with other biases such as overconfidence and wishful thinking. Overconfidence is defined as an overestimation of one’s abilities and is mainly observed in skill-related tasks (Tyszka and Roy, 2005). Wishful thinking or desirability bias refers to the idea that estimation of an event is driven by the desirability of that event. Wishful thinking is a motivational bias whereas unrealistic optimism does not arise only for motivational reasons (Krizan and Windschitl, 2009).
Optimism is an important topic of research because of its effects on behaviour and of its implication on economic context. Sharot (2011) believes unrealistic optimism to have positive implications. In particular, she shows that mentally-healthy individuals manifest unrealistic optimism whereas depressed individuals lack this bias. Unrealistic optimism in this sense is believed to reduce stress and anxiety because it creates positive expectations in people’s minds. In addition, biologists believe this unrealistic expectation to have evolved through human history in order to enhance the survival and well-being of human beings. This evolutionary account is explained by two arguments: firstly, overestimating the likelihood of positive events and success drove competition over valuable resources and therefore enhanced fitness
1
4 of species; secondly, the development of awareness over sickness and old age (conscious foresight) could have had overwhelming consequences on human daily activity, therefore unrealistic optimism helped the human race to cope with this awareness (Sharot, 2011).
On the other hand, researchers of unrealistic optimism unanimously believe that holding unrealistic expectations has large negative consequences and leads to efficiency loss in the economic domain (see Coelho, 2010; Harris and Hahn, 2011). People who are unrealistic about the actual likelihood of events tend to take greater risk in life such as investing in high-risk assets and starting new businesses doomed to fail on the account that they believe to succeed against the odds (Coelho, 2010). Moreover, unrealistic optimists engage more frequently in risky behaviours such as unprotected sex, smoking and reckless driving. In this sense, optimism works as a “barrier to protective health behaviour” (Geers et al., 2013, p.31).
Engaging in risky situations is similarly determined by the preferences toward risk, defined as the degree to which people are willing to take risks (Charness et al. 2013). Risk preference (or attitude) is a widespread topic in economics within the realm of decision making under risk and uncertainty and it is depicted by models of expected utility theory as well as prospect theory. Evidence of different risk attitudes has been provided via extensive experimental research which classifies people as risk-averse, risk neutral or risk-seeking mostly using the context of lotteries (see Eckel and Grossman, 2002; Holt and Laury, 2002; Dave et al, 2010; Charness et al., 2013).
The right assessment of individual risk preferences is necessary in order to comprehend economic behaviour. Similarly, a better understanding of unrealistic optimism is required. However, since optimism and risk preferences drive risky behaviour these two notions are extremely hard to disentangle. For instance, in lotteries’ context the willingness to accept risk of an individual with unbiased expectations and risk-seeking attitude is hard to distinguish from that of an individual who is instead risk neutral and an unrealistic optimist, similarly it is hard to separate a risk-averse individual with rational expectations from a pessimist risk neutral (Mansour et al., 2008). For these reasons, we believe it is important to study the
5 correlation between risk preferences and unrealistic optimism in order to disentangle their effects on risk-taking behaviour, and this is the purpose of this master thesis. We investigate this relationship experimentally, eliciting participants’ risk preferences before and after a semantic priming manipulation that aims at increasing optimism and pessimism. Manipulating optimism in order to investigate its consequences has been previously suggested in the literature as a solution to the problem of finding a conclusive and unequivocal measure of this phenomenon (Shepperd et al., 2017; Coelho, 2010). Moreover, this approach is particularly suited for this research given the difficulty in separating the concepts investigated.
The research on the correlation of unrealistic optimism and risk preferences is rather scant, besides previous experimental studies have exposed contrasting results (see Mansour et al., 2008; Weinstock and Sonsino, 2014). This study contributes to the existing literature by introducing a manipulation of optimism to explore its correlation with risk preferences. Additionally, the approach used in this research will not limit our investigation on the correlation between optimism and risk preferences but will allow us to explore the nature of this relationship. In particular, by manipulating optimism we can test whether increasing optimism leads to higher observed risk-seeking and therefore risk-taking (Astebro et al., 2015). After having reviewed the literature, we expect to observe a positive correlation between unrealistic optimism and risk-seeking preference.
The results obtained do not find supporting evidence of a correlation between optimism and risk attitude, we mostly ascribe this result to experimental weaknesses, in particular, the weak effect of the manipulation, however we cannot exclude that perhaps there is no actual correlation between optimism and risk preferences and what was previously found was due to measurement error which assessed other biases rather than optimism ( e.g. overconfidence).
This introduction is followed by a literature review (Section 2) providing an overview of the current status quo in research. In section 3, we present the methodology, covering the experimental design and hypotheses. Section 4 outlines the results, followed by a critical discussion of the limitations (Section 5). Finally, last section (Section 6) will present the conclusions and suggestions for future research.
6
2. Literature review
2.1 The definition and measurement of optimism
The literature, as mentioned before, introduce many different definitions of optimism (eight in total following Beazley, 2009). This often causes erroneous connection between studies that focus on different types of optimism, making a review of existing literature challenging. This research focuses on unrealistic optimism as defined by Weinstein (1980) denoting a bias in estimating the likelihood of events. However, to understand the critique we will put forward, it is essential to define the other main form of optimism, namely dispositional optimism. Dispositional optimism is described in the literature as a stable personality trait underlining a generally positive outlook on events and does not have to be unrealistic (Jefferson et al, 2017). Unrealistic optimism instead is not a stable trait, and its level has been shown to be affected by moods and anxiety (Helweg-Larsen and Shepperd, 2001). Indeed, researchers on optimism have repeatedly suggested a distinction between these two different classifications of optimism (Shepperd et al., 2002). However, since these definitions overlap in the sense that they both represent a positive expectation about future outcomes (Shepperd at al., 2017) a positive correlation between the two cannot be excluded, especially given the limited research focused on their relationship (Radcliffe and Klein, 2002).
Dispositional optimism is assessed throughout literature by the Life Orientation Task (LOT), whereas the measure of unrealistic optimism varies in different studies. Assessing the different measures of optimism, Shepperd et al. (2013) suggest the distinction between absolute unrealistic optimism and comparative unrealistic optimism.
The former is measured by eliciting expectations of events and comparing those expectations against absolute and objective likelihoods calculated by a risk algorithm; whereas the latter is assessed by asking participants to compare personal odds with that of other people. Several studies based on the research of Weinstein (1980) use a comparative approach to prove and measure unrealistic optimism given that it is easier to implement (Shepperd et al., 2017). In this method, the researcher
7 does not require any knowledge of a person’s actuarial risk of an event to measure unrealistic optimism. Instead, he/she can infer it by noting that not all, or even the majority of people in the group, can perceive the chances of a negative event to happen to them as less than average. Some of them may be correct, while others are definitely mistaken (Radcliffe and Klein, 2002). The measurement is assessed in this manner: the researcher asks participants to compare their chances of experiencing a particular event with that of another person of the same gender and age and report their estimation on a bipolar scale. If the chances reported are zero, the researcher does not observe optimism nor pessimism. Optimism is observed if the participant rates his/her chances higher than the comparing group, therefore reporting a number higher than zero. The reverse signals pessimism. In events
determined by individual’s skills and performances, this method may assess the
individual’s level of overconfidence rather than his/her level of optimism representing a shortcoming of this approach.
Another important flaw with this type of measure is the role of private information of participants. Although it is statistically impossible for the majority of people within a group to be below average risk some people might have private information about their likelihood of experience certain events which justify their estimation (Harris and Hahn, 2011). Essentially, if the probabilities’ distribution of a particular event, for instance, lung cancer, is not symmetric in the group but skewed to the left because the sample is rather healthy and all non-smokers, observing that the estimation of the majority of the group is skewed to the left does not represent strong evidence of optimism. In this sense, the comparative approach loses its assessment power (Coelho, 2010). Harris and Hahn (2011) report three other weaknesses in the comparative assessment which lead to measurement error of optimism. These weaknesses are scale attenuation, minority under-sampling and base rate regression. Scale attenuation occurs when the variance is restricted by a small scale. These scales do not allow worse off (or better off) minority to give an extreme estimation to counterbalance the majority. With regards to optimism, this problem
leads to an overestimation of the phenomenon2. Minority under-sampling is a
statistical artefact which causes an underrepresentation of a minority in the sample compared to the population and leads to an inaccurate estimation of the level of
2
8 optimism. Finally, base rate regression is due to limited knowledge of actuarial risks and lead to regressive people’ estimation. That is, people underestimate the likelihoods of common events and overestimate the likelihoods of rare events. Contributing to existing research, we try to circumvent these measurement problems by exogenously manipulating unrealistic optimism. Nevertheless, given that we apply a comparative measure of optimism in this experiment in order to check for the manipulation effects, the term comparative optimism will refer to unrealistic optimism from here onwards.
Although a study on the causes of optimism is beyond the scope of this master thesis we feel compelled to briefly outline some of them. Shepperd et al. (2002) distinguish between motivational and cognitive determinants of optimism. The motivational factors refer to two concepts: self-enhancement, defined as the wish of reducing anxiety by believing that negative events will not happen to us; and self-representation interpreted as the attempt to establish a desired personal image of oneself. The cognitive determinants include the illusion of control, a cognitive bias which lead us to believe that we have complete control over the outcome of events even when we do not; and the representativeness heuristic, a decision shortcut used to estimate the likelihood of an event by comparing it to a prototype that already exists in our minds.
2.2 The economics of optimism and risk preferences
The literature on optimism in the field of economics is coherent in treating optimism and pessimism as behavioural phenomena incompatible with Expected Utility theory (see Wakker, 2010; Abdellaoui et al., 2011; Astebro et al., 2015). These phenomena are instead captured by non-expected utility model such as Rank-dependent utility, by assuming that optimistic individuals transform objective probabilities into subjective and more favourable ones through a probability weighting function (Astebro et al., 2015; Mansour et al., 2008). Rank-dependent utility models have been used in the context of decision under risk (known probabilities) and decision under uncertainty (unknown probabilities), in these models decision makers assign a weight to outcomes based on their ranks, from good to bad. Therefore ranks of good
9 outcomes are smaller than those of bad outcomes. The weight assigned to each outcome shows the level of optimism or risk-seeking of the decision maker. Specifically, extra weight given to unfavourable outcomes signals either pessimism or risk aversion. Conversely, if more weight is given to favourable outcomes we observe either optimism or risk-seeking (Wakker, 2010).
In expected utility model, risk-taking is expressed by preferences over money depicted by changes in the utility function, concave utility curves for risk-aversion and convex utility curves for risk-seeking. In rank-dependent utility risk attitude are depicted both by the curvature of utility and by the curvature of the probability weighting function. Concave weighting functions characterize optimism and seeking. In the same way, convex weighting functions manifest pessimism and risk-aversion (Abdellaoui et al., 2009). We believe that this suggests a positive correlation between risk-seeking and optimism. Rank dependent utility allows expressing risk-taking by using both the preferences over money and the subjective transformation of the probability which captures optimism and pessimism.
Taking together these considerations, we notice a problem in separating the effects of optimism on risk-taking from those of attitude towards risk. We believe that more investigation on the correlation between optimism and risk-preferences could provide a better understanding on this matter, especially given that research on this correlation is scarce and has provided contrasting results. We analyse experimental evidence on this topic, in the next sub-section.
2.3 Optimism and risk preferences: experimental evidence
Weinstock and Sonsino (2014) (henceforth W&S) provide evidence of a positive correlation between risk attitude and optimism. They implement two experiments, using a lottery-based task to measure risk preferences and financial forecasting to assess the degree of optimism in participants. Although their results are in line with the prediction of the present study, we feel the need to point out some relevant limitations in their study.
10 W&S’s definition of optimism is not consistent throughout the paper and often appears confusing. The inconsistencies start with using Hey’s definition (1984) who defines optimism in term of subjective beliefs over events beyond the decision maker’s control. W&S underline how their definition differs from the stable personality trait representative of dispositional optimism, however, later they state that their results have important implications to support the cognitive account of dispositional optimism. Yet, they use the concept of forecast optimism in the remainder of the paper. Forecast optimism is defined as a positive expectation (optimistic) over financial indicators and is measured by asking to predict for instance future performances of financial stocks. It is sometimes termed as forecast positivity or financial optimism in the literature and treated as a bias (Balasuriya et al., 2013). We believe W&S’s definition of optimism to be more related to unrealistic optimism than to dispositional optimism.
W&S assess forecast optimism by asking participants to predict the expected return of stocks and key economic indicators. One problem with this type of measure may be the motivation behind the forecast. This indeed can influence the level of optimism in forecasting (Balasuriya et al, 2013), especially if subjects actually held investments on the same stocks they were asked to predict, something that the authors fail to acknowledge. In this sense optimism can be also mistaken with wishful thinking, that is the higher the desirability of the event the higher its chance estimation, and therefore inflating the measure of optimism (Krizan and Windschitl, 2009). Likewise, using financial forecasting as a measure of optimism could accidentally estimate overconfidence in the ability to forecast rather than optimism (Coelho, 2010), and this is relevant given that a significant percentage of the sample worked in the investment industry (18% and 49% respectively for the two samples). Tyszka and Roy (2005) indeed point out that optimism is mostly observed in pure-chance related risk rather than in skill-related tasks.
Finally, only two years after the original experiment but prior to the publication, following a referee comment, W&S (2014) decided to distribute the LOT-Revised to the participants of study 2 and included this measure. Unfortunately, only 50% of the participants filled out this survey. The result of this measure shows that dispositional optimism weakly correlates (ρ=0.10) with forecast optimism, however, the size of the sample (36 observations) is too small for any definite conclusion.
11 A noteworthy feature of their study is the inclusion of a “coin tossing game” borrowed from Mansour et al. (2008). In this task W&S ask participants to imagine ten independent fair coin tosses and estimate the number of times heads will occur, presuming that participants will be paid a certain amount every time heads is the outcome of the game. An estimation of the number of heads above 5 is defined as “win-chance optimism”. The authors asked participants to give an answer based on their belief in personal luck and used this answer to control for this belief. The results show a weak and no significant correlation between forecast optimism and this additional measure of optimism.
This finding is crucial, taking into consideration that Mansour et al. (2008) uses this measure to assess the level of optimism. Similarly, in Mansour et al. (2008) an optimistic individual is defined as someone characterized by an estimation of the number of heads (wins) above five and vice versa for a pessimistic individual. Risk attitude is measured by asking participants their maximum willingness to pay to participate in the coin tossing game. Notably, this approach allows them to determine risk preferences and optimism in the same context via a single task, easy to understand and implement but the results obtained are rather puzzling. They find a negative correlation between optimism and risk seeking, implying that optimistic
people are more risk-averse. This result seems counterintuitive. If a subject’s
estimation of the number of heads is above five, signalling an expectation to win more than losing, it appears somehow irrational that his/her willingness to pay to take part on this game is lower than subjects who believe to lose. We believe there must be some error in the assessment of optimism and risk preference used in their study. Indeed, as pointed out above, the measure implemented to assess optimism in their study shows only a weak and non-significant correlation with another definition of optimism, namely forecast optimism (Weinstock and Sonsino, 2014). Moreover, this measure may be considered simplistic since the objective estimation of heads (wins) is 5/10, people with some level of statistical training might simply apply this knowledge in answering the question without exhibit any optimism/pessimism (Tyszka and Roy, 2005). This measure is particularly not suitable for this research given that our sample pool comes mostly from economics students who are highly trained in statistics.
12 Mansour et al. (2008) additionally measure optimism and pessimism following Wenglert and Rosen (2000) approach, by asking participants to estimate likelihood of positive and negative events, both at a personal and global level and find a similar level of “personal optimism”, 0.536 (0.596 in Wenglert and Rosen, 2000). However, they do not assess this measure against their estimated risk aversion in order to confirm their results. They limit to check for consistency of the level of optimism (pessimism) obtained in the coin task with just three of the questions used in this task and find a positive relationship.
The literature presented above manifests several weaknesses in the assessment of optimism, in particular, the methodology used in the past are not always conclusive and accidentally measure other biases. Finally, disentangling the effects of optimism from those of risk preference is particularly difficult. In consideration of these problems, the current study investigates the correlation between optimism and risk preferences differently. In this research, instead of assessing optimism level as in previous studies we manipulate it, this manipulation occurs between two tasks which elicit participants’ risk preferences. By doing so we bypass these measurement problems as will be further discussed in the next section on the methodology. The hypothesis brought forward is that optimism is positively correlated with risk-seeking. Additionally, by measuring risk preferences before and after the manipulation of optimism we can test whether a causal relationship exists between these two concepts. This approach could help future research in better distinguishing the effect of these two notions.
3. Methodology
We conducted an online experiment via the software Qualtrics which allow us to collect a large number of observations, freely. We collected 88 observations in total. The experiment run for approximately 15-20 minutes and was distributed mostly via social media e.g. Facebook. There were no participation criteria for this experiment but most of the sample pool included students with a background in economics. The experiment consisted of five parts with three treatments in part 2 using a
between-13 subjects design. All participants completed the five parts of the experiment. The instructions for each part are provided in Appendix B.
The first part assessed risk preferences prior to the manipulation using the Eckel and Grossman task. This task was followed by a manipulation optimism and pessimism through semantic priming (part 2). We then reassessed risk attitudes using an investment task (part 3). Part 3 was followed by a manipulation check of the effect of semantic priming where optimism was measured using the comparative measure approach (part 4). Part five concluded the experiment with a demographic questionnaire. The graph below represents the sequence of each task in the experiment. We will present each part in detail in the following subsections.
3.1 The Eckel and Grossman task
The first part of the experiment dealt with a measurement of risk preferences prior to the manipulation of optimism and pessimism in order to assess the initial risk attitude of participants. In choosing the ideal risk preferences assessment we analysed a number of experimental papers looking for potential weaknesses of each approach (specifically Eckel and Grossman, 2002, 2008; Holt and Laury, 2002; Dave et al, 2010; Charness et al., 2013). We opted for the decision task created by Eckel and Grossman (2002, 2008). The Eckel and Grossman decision task (from here onwards E&G task) consists of six possible lotteries with two outcomes determined by the occurrence of two events with equal probabilities (Event A and Event B). The lotteries are increasing in expected return and risk, and permit to classify participants from risk-averse (Lotteries 1 to 4) to risk neutral (Lottery 5-6) to risk-seeking (Lottery 6). Consistent with the literature, (Dave et al., 2010) we define participants who chose lottery 6 as risk-seeking since lottery 5 and 6 have the same expected return and differ only on the level of risk, higher in lottery 6. Table 1 below shows the 6
Part 1: E&G task
Part 2: Priming (3 treatments) Part 3: Investment task Part 4: Comparative assessment Part 5: Demographic questionnaire
14 lottery choices as adapted in this study. Table 3 in the results section reports these lottery choices and their implied CRRA (coefficient of constant relative risk aversion), with the fraction of subjects who chose each lottery.
Table 1 – The Eckel and Grossman Risk Preferences
Lotter choice Low Payoff €
(Event A 50%) High Payoff € (Event B 50%) Expected Return Standard Deviation Lottery 1 14 14 14 0 Lottery 2 12 18 15 3 Lottery 3 10 22 16 6 Lottery 4 8 26 17 9 Lottery 5 6 30 18 12 Lottery 6 1 35 18 17
The E&G task classifies participants in fewer categories compared to the multiple- price list by Holt and Laury (2002) however, the latter presents often inconsistency of choice. The Holt and Laury task includes a set of 10 binary choices between low-risk lottery (lottery A) and high-risk lottery (lottery B) with the same probabilities. In the first binary choice, these probabilities are low for the high outcome and high for the lower outcome, but they gradually increase (for high outcomes) and decrease (for low outcomes) in the subsequent choices. The idea is that participants choose lottery A (the safe choice) in the first binary choice, and continue to do so until at some point they switch to lotteries B (the risky choices). The number of safe choices (lotteries A) represents the degree of risk aversion. However, experimental evidence has shown that participants often switch more than once between lottery A and B creating inconsistency of choice which causes much noise in the data (Dave et al., 2010). In addition, the E&G task is easier to implement and understand than the other methods reviewed.
We included real monetary incentives for this task by randomly selecting a participant at the end of the experiment and paid him/her according to his/her
15 decision and the event that has occurred. The selection of the event occurred was done using a die-roll: If either 1, 2 or 3 was rolled Event A has occurred if either 4,5 or 6 was rolled Event B has occurred. This method was made clear to participants in the instructions. The payment was made by transferring money to the participant’s bank account.
3.2 Priming optimism and pessimism
As outlined in the literature review, the methodologies previously used to measure unrealistic optimism present some relevant shortcomings and are often not conclusive (Coelho, 2010; Fosnaugh et al., 2009). To solve this problem in this study we attempt to exogenously manipulate optimism and pessimism, via priming. Priming is a long-standing practice in psychology used to mentally activate the primed concepts in participants. These methods enable researchers to measure the effect of the primed notion on behaviour in following tasks (Cohn et al., 2015). This method represents a novel approach in the context of optimism and risk attitude. Additionally, Shepperd et al., (2017) in assessing the weaknesses with existing measure of optimism, suggest manipulating this bias in order to investigate its consequences.
In choosing the right task we considered different manipulations of optimism present in the literature and selected a semantic priming task, namely the scrambled-sentence task designed by Fosnaugh et al. (2009) and replicated by In Den Bosch‐ Meevissen et al. (2014). Fosnaugh et al. (2009) succeed in increasing both dispositional and comparative optimism in participants as measured by the LOT-R and the comparative assessment. This finding not only is important for this research, given our assessment of unrealistic comparative optimism but also suggests a correlation between these two definitions of optimism, which was suggested in the literature review.
Semantic priming via the scrambled-sentence task has never been implemented in an online setting representing a novel feature of this study. However, this implementation might be less effective compared to a laboratory setting.
16 The task involved 15 items of five scrambled words, participants were asked to create grammatically correct sentences using only four of the five words presented in each item. We implemented three different treatments for this task, an optimism and a control treatment following the approach in Fosnaugh et al. (2009) with the addition of a pessimism treatment to check for consistency with the prediction of this study. We believe to be the first to implement a manipulation of pessimism via the scrambled sentence task.
The material used for the optimism and control treatments was kindly provided by Dr. Andrew L. Geers (co-author of the paper Fosnaugh et al., 2009) and it consisted of 30 items of scrambled-words, fifteen for each treatment. In the optimism treatment, words related to optimism (e.g., hope, confidence) were embedded in eleven of the fifteen items. In the control treatment, we had the same scrambled-sentences with eleven neutral words instead of the optimism prime words of the first treatment (e.g., conversation, magazine). We used the same sentences to develop the pessimism treatment and a dictionary of antonyms to substitute the optimism prime words with pessimism related words. The sentences used including the sentences created for the pessimism treatment are provided in Appendix B.
Before commencing the task, participants were presented with a completed example of the scrambled-sentence task to further clarify the instructions given. Participants were randomly assigned to one of the three treatments by the randomization function available on Qualtrics.
3.3 The investment task
The third part of the experiment dealt again with a measurement of risk preferences.
The participants’ preferences were assessed straight after the manipulation of
optimism and pessimism which allow us to make a comparison between risk attitude for the three different treatments. More so, this approach allows us to test whether a causal relationship between unrealistic optimism and risk preference exist by comparing risk preferences elicited in the first task with risk preferences obtained in this task.
17 In selecting the right task for this part, we were aware of the inconsistency problem across risk preferences elicitation methods (Reynaud and Couture, 2012), however, we refrained from reusing the E&G task for fear of mere consistency with the answers given in the first part. We, therefore, opted for a risk investment task, borrowed from Cohn et al., (2015) and originally designed by Gneezy and Potters (1997) which directly measure the willingness to take risk in a simple manner.
In this task, participants were initially endowed €10 and were asked to make an investment decision by deciding how much of their endowment invest in a risky asset. Participants could have potentially won or lost real money depending on the event that has occurred. Similarly, in part one, two events were possible with 50% of probability each. This was made salient to the participants by displaying a picture of a plastic box (see figure A2 in Appendix B) with two balls, a yellow ball representing the ‘good’ event and a red ball representing the ‘bad’ event. If the yellow ball was drawn from the box participants won 2.5 times the invested amount. If the red ball was drawn participants lost the invested amount. This procedure was clearly explained in the instruction, and calculation on how participants’ earnings were determined was provided. Furthermore, given that outcomes and earnings were less explicit in this task than in the E&G task, we implemented a set of control questions. In these questions, participants were asked to give a hypothetical investment amount and the respective earnings for each possible event (yellow or red ball drawn) in this way we tested whether participants understood the risk of investing in this task. At the end of the experiment, we used the toss of a coin to establish which event happened (head represented the yellow ball and tail represented the red ball) and randomly selected a participant to be paid according to the state occurred and his/her investment decision. The payment was made by transferring money to the participant’s bank account.
Some of the material used in this part of the experiment was kindly provided by Dr. Jan Engelmann (co-author of the paper Cohn et al., 2015).
18 The fourth part of the experiment used a comparative measure to assess the level of optimism in participants. The rationale behind this task is to assess the effects of the manipulation of optimism and pessimism. The task was an adapted version of the method used by Weinstein (1980). In creating this task, we followed the recommendations suggested by Harris and Hahn (2011) in their critique of the work of Weinstein. One of the main drawbacks with the approach used by Weinstein is the problem of scale attenuation which results from using small scales for instance 7-points and 15-7-points scales. Specifically, when estimating the likelihood of particularly rare events, these scales do not allow extreme estimations even though rarity of events would require them, this problem leads to overestimation of optimism as estimations remain to close to numerically counterbalance. Evidence of this effect is given by Otten and Van Der Pligt (1996) who use both a small scale (9-point scale) and a large scale (-100 to +100) and find stronger optimistic responses in the smaller scale relative to the larger scale. In Weinstein original paper, optimism is expressed in a 15-point scale, conversely, in this study, we use a continuous -100 to +100.
Participants were randomly presented with twenty events (10 positives and 10 negatives) and asked each time to estimate the probability of experiencing that event at some point in their life, compared to another person of the same gender and age. A response of 0 indicates that the participant believed his/her chance to be the same as the average person of the same gender and age. Negative and positive responses indicate that participants believed their chances to be lower and higher respectively, by the percentage indicated by the number chosen in a slider scale. This procedure was carefully explained in the instructions and an example was given to avoid confusion.
3.5 Demographic information
Finally, participants were asked to answer few short questions in order to gather demographic information on their nationality, age, gender, occupation and educational background.
19 This information was necessary in order to control for the interpretation of the results but also to link results with previous findings in the literature. In the last step of this section, participants were asked to provide their email addresses to be contacted, in case of selection for payment in the incentivised tasks.
3.6 Hypotheses
In this subsection, the primary and secondary set of hypotheses of this study are outlined. The primary hypotheses refer to the main focus of this master thesis, the relationship between optimism and risk preferences. The secondary hypotheses refer to interesting gender differences in the area of optimism and risk preferences which we aim to replicate.
3.6.1 Primary Hypotheses
Hypothesis 1: Optimism is positively correlated with risk-seeking preference.
This first hypothesis is related to a correlation between optimism and risk preferences. In line with the theory and the findings of Weinstock and Sonsino (2014) outlined in section 2, we expect a strong positive correlation between optimistic individuals and their level of risk-seeking behaviour. Moreover, the experimental design of this study allows us to test whether increasing optimism via priming leads to observed risk-seeking preference independently from the initial level of risk-seeking assessed in the E&G task. In other words, we control for risk preferences estimated in the E&G task in order to test whether a causal relationship between unrealistic optimism and risk preferences exist. Conversely, we assess whether increasing pessimism in the priming condition leads to higher risk-aversion in the investment task. This approach consents to further distinguish optimism from risk-seeking and pessimism from risk-aversion.
20
Hypothesis 2: Men are more risk-seeking than women.
In line with extensive evidence in the literature (see Eckel and Grossman, 2002; Charness and Gneezy, 2012) we expect women to be more risk-averse than men.
Hypothesis 3: Men are more optimistic than women.
Few studies have focused on gender difference in the level of unrealistic optimism (see DeJoy,1992; Lin and Raghubir, 2005; Jacobsen et al., 2014), in these studies men show a higher level of unrealistic optimism than women. Although gender difference in optimism has been rarely investigated, we expect to find a higher level of optimism in male participants compared to female participants. Testing this gender difference in this study allow us to investigate whether gender difference found in risk attitude is persistent after we control for optimism level. If gender difference in risk attitude disappears when we control for optimism it would suggest that this gender difference is due to differences in the level of optimism and not to differences in risk attitude.
4. Results
This section outlines the results. Firstly, we provide some information on the characteristic of the participants and the randomization checks. This is followed by the analysis of the effects of optimism and pessimism priming. Subsequently, we report non-parametric test statistics and regression analysis in relation to the treatment effects and the primary hypothesis. This section concludes with the analysis relevant to the secondary hypotheses. A critical discussion of the results obtained will be presented in the next section of this master thesis.
4.1 Characteristics of participants and randomization check
A total of 88 participants took part in this experiment, however, 11 observations are dropped due to incorrect answers to the control questions3, therefore the final sample consists of 77 observations. Table 2 summarizes the characteristic of the
3
21 subjects in the total sample and per treatments. The final sample comprises 31 male, participants’, average age is 27 years old and 56% of them is studying or has studied economics.
We test whether the randomization successfully resulted in a balanced sample. We use Kruskal-Wallis test for interval variables and χ2- tests in case of binary variables, the p-values of these tests are provided in Table 2. Based on conventional significance level we cannot reject the null hypothesis that age, gender and economic education are balanced across treatments. Similarly, we cannot reject the null hypothesis that the distribution of risk attitude of the subjects as measured by the E&G task is balanced across treatments. This excludes a significant difference in the level of risk-aversion, prior to the manipulation, between subjects of different treatments.
Table 1 –Table 2 -Characteristic of Participants and Randomization check
Variable Total Sample (N=77) Optimism (N=26) Control (N=26) Pessimism (N=25) p-value Male 31 (40.26%) 13 (50.0%) 7 (26.92%) 11 (44.0%) 0.213 Age (Mean) 27.09 (5.63) 25.54 (3.60) 26.38 (3.67) 29.44 (8.00) 0.176 Economic Student 43 (55.84%) 32 (41.55%) 14 (53.85%) 11 (42.31%) 16 (61.54%) 12 (46.15%) 13 (52.0%) 9 (36.0%) 0.766 0.700 EG Risk preferences (Mean) 3.714 (1.746) 3.885 (1.818) 3.692 (1.738) 3.56 (1.733) 0.771
Notes: This table reports means and standard deviations (in parenthesis) for interval variables in the total sample and in treatments Optimism, Control and Pessimism. It reports the number of participants and percentage (in parenthesis) for binary variables in the total sample and in the treatments Optimism, Control and Pessimism. The last column displays p-values for the null hypothesis of perfect randomization (χ2 tests in case of binary variables and Kruskal-Wallis tests in case of interval variables).
“Age” is the individual’s age in years. EG Risk Preferences is the individual chosen lottery in the E&G task. “Male” is a dummy variable indicating male subjects. “Economic” is a dummy variable indicating whether the participant had some level of education in Economics. “Student” was a dummy variable to indicate whether the participants were still studying or not.
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
22 Table 3 below reports the lottery choices, their CRRA and the fraction of subject who chose them, the frequency of choice is plotted in figure A1 in Appendix A. The most selected lottery is lottery 3, the second most selected lotteries are lottery 1 and lottery 6 these results are compatible with previous literature (See Eckel and Grossman, 2002, 2008; Dave et al., 2010). However, lottery 1 was selected by a smaller percentage compared to lottery 6 (17% and 22% respectively) this suggests slightly higher risk-seeking preference in our sample compared to Dave et al. (2010).
Table 3 – The Eckel and Grossman Risk Preferences Choice (50/50 Lottery) Low Payoff € (Event A) High Payoff € (Event B) Expected Return Standard Deviation Implied CRRA Range Fraction of Subjects (%) Frequency of Choice Lottery 1 14 14 14 0 3.46<r 17 15 Lottery 2 12 18 15 3 1.16<r<3.46 5.7 5 Lottery 3 10 22 16 6 0.71<r<1.16 33 29 Lottery 4 8 26 17 9 0.50<r<0.71 10.2 9 Lottery 5 6 30 18 12 0<r<0.50 12.5 11 Lottery 6 1 35 18 17 r<0 21.6 19 4.3 Priming optimism
We test whether the manipulation of optimism via priming succeeded in increasing (Optimism treatment) or decreasing (Pessimism treatment) optimism. To determine this, we use the comparative optimism measure derived from part four of the experiment. A total score of comparative unrealistic optimism for each participant is obtained by summing his/her estimation of the 20 events and reversing the score for negative events. The result is then averaged across events. Mean estimates for the single event are shown in Table A1 in Appendix A.
We find a mean of comparative optimism of 3.49 (SD= 15.03) across treatments that is significantly different from zero at 5% confidence level (p-value= 0.0453, t-test). In
23 line with previous literature (Shepperd et al., 2002) we find higher optimism for negative events (mean= 3.75, SD= 18.48) compared to positive events (mean= 2.85, SD= 19.35) the difference is statistically significant at 5% confidence level (Wilcoxon signed rank p-value= 0.0469).
Figure 1 displays the distribution of comparative optimism across treatments. The median is slightly higher for optimism. However, the level of optimism is overall higher in the control group, suggesting a non-significant effect of priming.
Figure 1– Comparative optimism across treatments
To confirm this result, the total scores of comparative optimism are submitted to Kruskal-Wallis tests. The results do not yield a significant main effect of priming (p-value= 0.5520)
Additionally, paired comparisons between treatments are performed. We do not observe an increase in optimism between the control and the optimism treatment (Wilcoxon rank sum p-value= 0.3097) at any significance level. Similarly, we do not
24 observe lower level of optimism in the pessimism treatment, compared with the control (Wilcoxon rank sum p-value= 0.4121) at any significance level.
4.4 Treatments effect and primary hypotheses
Figure 2- Investment decision across treatments
Figure 2 displays the amount invested in the investment task by treatments. The figure suggests a relationship between the level of optimism and risk seeking. Specifically, subjects seem to invest more in the optimism treatment (mean= 6.30, S.D. 2.97) and invest less in the pessimism treatment (mean= 5.8, S.D. 2.35) suggesting a positive correlation between the level of optimism and risk-seeking (Hypothesis 1). However, these differences are not statistically significant at any confidence level, a Kruskal-Wallis test yields a p-value equal to 0.6935. Similarly, pair comparisons between treatments using Wilcoxon Rank Sum test do not yield significant differences.
25
4.4.1 Regression analysis
We perform a regression analysis in order to further investigate treatment effects, while controlling for background variables, in particular, the risk preference of each participant as estimated by the E&G task. We, therefore, proceed by regressing the amount invested in the investment task on the risk preference expressed in the E&G task, the optimism and pessimism treatment, the total measure of comparative optimism and some other control variables. Table 3 summarises the results of this regression analysis.
Table 3 - Regression Analysis of Investment Decision
Y= Amount Invested € (0-10)
Regression 1 Regression 2 Regression 3
X Marginal effect (SE) Marginal effect (SE) Marginal effect (SE) E&G Risk preferences 0.754 (0.201)*** 0.751 (0.204)*** 0.748 (0.211)** Optimism -0.111 (0.697) -0.194 (0.709) -0.058 (0.739) Pessimism -0.392 (0.703) -0.447 (0.710) -0.519 (0.747) Comparative Optimism -0.014 (0.020) -0.013 (0.020) Male -0.327 (0.614) Age 0.048 (0.056) Economic 0.199 (0.633) Observations 77 77 77
Notes: This table reports OLS coefficient estimates (robust standard errors in parentheses). The dependent
variable is the amount invested in the risky asset. E&G Risk preferences represent the lottery chosen in the E&G task. “Optimism” is a dummy for treatment Optimism. “Pessimism” is a dummy for treatment pessimism. Comparative optimism refers to participants’ optimism as assessed in the comparative measure. “Male” is a gender dummy. “Age” is the participant’s age in years, “Economic” is a dummy for participants who had some level of economic education.
*** Significant at the 1 percent level. ** Significant at the 5 percent level. * Significant at the 10 percent level.
26 The estimation results reported in Table 3 reveal that controlling for risk attitude prior to the treatment yields no positive correlation between increased optimism and risk seeking (Hypothesis 1). The regression coefficients in regression 1-3 are indeed negatives for optimism and these results are not significant at any confidence level (p-values > 0.4, t-test). A weak and non-significant negative correlation is found between pessimism treatment and risk-seeking (p-value= 0.578, t-test).
Since the manipulation of optimism did not produce significant results, we regress the invested amount on the comparative assessment of optimism score of each participant, testing again hypothesis 1. The coefficients of this variable in regression 2 and 3 are negative and not statistically significant at any confidence level, therefore no positive correlation is found between optimism and risk-seeking (Hypothesis 1). Lastly, the control variables for gender, age and economic education have no significant correlation with investment decisions.
4.5 The two measures of risk preferences
One major problem with different estimation methods of risk preferences is that they lead to different degrees of risk aversion (Reynaud and Couture, 2012). When comparing different elicitation method higher consistency is obtained if the choices in the two methods have comparable CRRA (constant relative risk aversion). We calculate CRRA coefficient for the investment task (calculations are provided in Appendix A). Table 4 shows the CRRA coefficient for each possible investment choice and the corresponding lottery according to the implied CRRA coefficient. From the CRRA values reported below, it appears that the two measures of risk preferences are not perfectly comparable. For instance, investing between 4 and 9 euros implies a coefficient of CRRA between 0.41 and 0.13 which is comparable with the implied CRRA coefficient of lottery 5 in the E&G task, signalling risk neutral. (see also table 3). Additionally, the CRRA coefficient from lottery 1 (r >3.46) is not considered in the investment task. For this reason, we decide to further assess the consistency between the E&G task and the investment task by plotting the observations of each method and using the results of the regression analysis.
27
Table 4 –CRRA levels in the investment task
Invested amount CRRA Corresponding
lottery from E&G task x= € 1 r=1.65 Lottery 2 x= € 2 r=0.84 Lottery 3 x= € 3 r=0.55 Lottery 4 x= € 4 r=0.41 Lottery 5 x= € 5 r=0.32 Lottery 5 x= € 6 r=0.26 Lottery 5 x= € 7 r=0.21 Lottery 5 x= € 8 r=0.16 Lottery 5 x= € 9 r=0.13 Lottery 5 x= € 10 r<0 Lottery 6
Given the small number of observations for each lottery, we decide to group them. Lottery 1 is grouped with lottery 2 representing high risk aversion, lottery 3 with
28 lottery 4 representing medium to low risk aversion and lottery 5 with lottery 6 representing risk neutrality and risk-seeking preferences. Figure 3 shows discrete consistency between the means in the investment task and the lotteries chosen in E&G task.
Moreover, regressing the amount invested in the Investment task on the E&G task reveals (Table 3) a strong positive correlation with the investment task of part three of the experiment. The coefficients in all 3 regressions are above 0.74 and significant at 5% and 1% confidence levels.
4.6 Secondary Hypotheses
We then proceed to test the secondary hypotheses to link this experiment with findings in the literature of optimism and risk preferences with regards to gender differences. The sample from this study included 31 males and 46 females.
Hypothesis 2 states that men are more risk-seeking than women. We plot the distribution of lottery chosen over gender, figure 4.
29 At first glance, higher risk seeking is observed in men compared to women. We test this difference by performing two Wilcoxon rank sum tests, using first the E&G task and then the investment task. The results reveal no significant differences in the level of risk aversion between women and men for both measures of risk preferences. The p-values of these tests are 0.8316 and 0.5907 for the E&G task and the investment task respectively.
Hypothesis 3 states that men are more optimistic than women. Again, we test this hypothesis starting by plotting the distribution (Figure 5).
The box plot below shows similar medians in the level of comparative optimism between men and women, however, men’s estimations seem more extreme at both end of the spectrum, optimism and pessimism. We test whether there is a statistical difference in the level of comparative optimism within gender by performing a Wilcoxon rank sum test. The result does not support a significant gender difference (p-value = 0.8112).
30
5. Discussion
In this section, we discuss the results obtained in section 4 and the relevant hypotheses pointing out experimental weaknesses and potential confounding factors.
The first result presented was relative to the effect of the manipulation of optimism via priming. The graphical analysis shows a slightly higher median for optimism than for pessimism, however, the nonparametric tests show no significant differences on the level of optimism assessed via the comparative measure across treatments. The sample size might be too small to detect significant treatments effect in this study. The sample in this experiment is significantly smaller than those used by Fosnaugh et al., (2009), their study comprise 52-53 subjects per treatment whereas this study included 25-26 participants per treatment. Furthermore, we devise three potential explanation for the non-effect of treatments.
Firstly, semantic priming, although a well-established technique, can be very susceptible to external stimuli (Hänze and Hesse, 1993). This is particularly true in this research given that the experiment was run online, which limited the control of the experimenter on participants. As a result, we could not control for potential external stimuli, and ensure the attention and focus of participants. Therefore, we believe that a bigger sample and a laboratory setting could overcome this problem, reducing noise in the observations. In addition, the scrambled-sentence task has never been tested in online experiments, this represents a novel feature of our research. However, further research is required in order to assess the effectiveness of this priming method in online setting.
Secondly, the effect of priming was assessed via the comparative optimism measure, however prior to this assessment, participants went through the investment
task, therefore, it could be that by the time participants’ optimism was assessed the
effect of priming was not as strong. Indeed evidence on semantic priming shows the effect to be short-lived (Jordens and Becker, 1997). However, given that this study’s interest was the relationship between optimism and risk preferences, assessing the latter straight after the manipulation of the former was the right course of action,
31 nevertheless, we need to acknowledge that this could have lowered our estimation on the effect of priming.
Thirdly, the experiment was entirely presented in English and we clearly expressed this in the instruction that preceded the distribution link to allow people who were not confident with English to refrain from participating. However, a variety of nationalities participate in this experiment and only 17 participants out of 77 were from a country whose first language is English. Therefore, we cannot control for differences in understanding and interpreting the prime words. Additionally, it could be that the
mental activation of the primed word in a language different from one’s mother
tongue is less strong and automatic. This effect has been proven negligible in bilingual with high proficiency (Perea et al., 2008) and we do believe the majority of participants to have a medium to high proficiency level however as proficiency of English was not assessed in this experiment we need to recognize this potential issue.
In lieu of the problems presented above, future research using similar semantic priming should lean towards a laboratory setting, using a larger sample and should pay attention to the proficiency of participants of the language used for the manipulation.
Moreover, this type of manipulation was chosen because it has been proven to work with comparative unrealistic optimism, other manipulations for instance “Imaging a best possible self” (by Peter et al., 2010) have been developed since. These manipulations have been shown by repeated applications, to have greater effects than the semantic priming as developed by Fosnaugh et al. (2009). However, these manipulations have only been applied to dispositional optimism, therefore, we do not know whether they would also work better for unrealistic optimism. Future research should evaluate these manipulations with regards to comparative unrealistic optimism, by doing so they could also assess whether dispositional and unrealistic optimism are indeed correlated.
The second set of results presented were relative to the treatments effect and the primary hypothesis. From the graphical analysis, we observe the hypothesized effect of priming on investment choices. Participants in the optimism treatment invested more than participants in the control and pessimism treatments. However, these
32 results were non-significant, not supporting the hypothesis of a positive correlation between optimism and risk seeking. Additionally, from the regression analysis, almost no relationship (slightly negative and not significant) was found between the optimism treatment and risk-seeking in the investment task, and the negative correlation found between pessimism and risk-seeking was weak and non-significant. Initially, we ascribed these results to the absence of priming effects (see above) and concluded that bigger sample and higher control could correct these problems. However, when controlling for the comparative measure of optimism the results continue to be non-significant. These results indicates no relationship between unrealistic optimism and risk attitude. Nevertheless, several weaknesses with the comparative assessment (see literature review and methodology) could also account for these results.
The level of optimism found is of lower magnitude than the general findings in the literature on unrealistic comparative optimism. We expected this result given that we decided to use a larger scale to avoid overestimation. However, the mean of comparative optimism found in this study, although significant (M= 3.48 SD= 15.02), is smaller than the one found by Otten and Van Der Pligt (1996) who found a mean of 6.64 with smaller variation (SD=8.45) using the same scale range but measuring it only using negative events. Optimism, therefore, appears to be stronger for negative events compared to positive events (see Shepperd et al., 2002; Shepperd et al., 2013 ). This difference was statistically significant in our study. In addition, given that we estimated comparative optimism mean across treatments, including pessimism, we would expect a lower mean in the level of optimism, however, nonparametric tests resulted in non-significant differences across treatments in the level of comparative optimism. Finally, we believe that assessing comparative optimism using an online channel might report some level of noise in the data.
The consistency between the two measures of risk preferences namely the E&G task and the investment task was also assessed. Comparison between CRRA levels in the two different tasks showed the consistency across choices in the two task not perfect. However, the correlation between the two elicitation methods was found to be strong and significant (coefficients above 0.74), signalling a fairly good consistency. Additionally, the results obtained from the E&G task are in line with
33 previous research, this suggests that eliciting risk preference online via the E&G task, is effective.
Finally, the assessment of gender differences in risk aversion (Hypothesis 2) and optimism (Hypothesis 3) did not show significant differences between female and male participants. With regard to risk preferences, this finding is at odds with previous literature, which has generally found higher risk aversion in women compared to male. The fact that we did not find significant gender differences might suggest future research to systematically assess this finding across different studies and elicitation methods. Moreover, given that the majority of our sample (56%) are students of economics who are accustomed to decision making under risk, we do not know whether studying economics has an impact on the size of gender difference in risk aversion and this calls for future research.
In the context of optimism, only a few studies have focused on gender differences (see DeJoy,1992; Lin and Raghubir, 2005; Jacobsen et al., 2014) and have found higher optimism for men than women, our results do not confirm this finding. Therefore, we believe that more research is needed in order to thoroughly assess whether gender differences actually exist in the level of unrealistic optimism.
We ascribe the unproven correlation between optimism and risk attitude to weaknesses in the design and measurement problems and invite researchers interested in this relationship to find a better measure for assessing unrealistic optimism. Additionally, possible lack of control over participants’ understanding on the task might have an effect on our results. Our concerns were partially solved by the control questions we asked regarding to the investment task, additional control questions would provide a better verification of participants’ understanding. Likewise, although all participants had equal chances of being selected for payment our monetary incentives might have been too weak to infer that participant committed to responding scrupulously. Furthermore, some tasks lacked this monetary incentive, namely the priming and the comparative assessment of unrealistic optimism, therefore, reducing participant motivation in performing these tasks with the necessary level of commitment. Finally, the findings in literature on the correlation between optimism and risk preferences are rather limited and contrasting (see Weinstock and Sonsino, 2014; Mansour et al., 2008). Additionally, several potential
34
weaknesses in the researchers’ assessment of this correlation might have led to
wrong conclusions4. Therefore, we should consider the possibility that there is no correlation to observe between unrealistic optimism and risk preferences.
6. Conclusion
Investigating risk preferences and optimism is important in the economic domain given their effect on risk-taking, as a consequence exploring the relationship between these two concepts is required. This investigation has been highly neglected in the literature and the scant research is not conclusive, this is partly due to the difficulty in disentangling unrealistic optimism from risk-seeking.
We aimed at shedding more light on this relationship by running an experiment. The various approaches used in the literature to assess unrealistic optimism inspired us to work with a manipulation of this behavioural phenomenon. The results shown in this master thesis, however, do not allow us to draw strong conclusions on the relationship investigated. In addition, we did not find significant gender differences in the level of optimism and risk aversion. We proposed some reasons explaining the weak results: Measurement problems, design weaknesses, and small sample size. Nevertheless, this study highlights how more extensive and systematic research is needed on this topic, prompting researchers to precisely define unrealistic optimism and the methods used to assess it. Finally, manipulating optimism revealed to be a hard task, more research into priming this behavioural phenomenon is needed and testing existing methods across definitions of optimism is suggested in order to better understand these different definitions and their correlations. We believe that an improved replication of our experimental design can provide a better understanding of this topic in the future and assess whether an actual correlation between optimism and risk attitude exists.
