The effect of incidental and dispositional emotions (fear and anger) on ambiguity preferences

(1)

Master thesis

MSc Economics: Behavioral Economics and Game Theory

The effect of incidental and dispositional

emotions (fear and anger) on ambiguity

preferences

Name Lotte Bruinink Student Number 11397306

Thesis Supervisor Dr. J.B. Engelmann

ECTS 15

(2)

Statement of originality

This document is written by Student Lotte Bruinink 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)

Abstract

In recent years, researches have become increasingly interested in the effect of emotions on decision making. When evaluating the effects of emotions on decisions involving uncertainty, most studies have solely focused on risk. Therefore, the objective of this paper is to find the effect of two distinct emotions – anger and fear – on ambiguity preferences, and to establish whether the valance-based approach or the appraisal-tendency framework better predicts the effects. Data was collected using Amazon Mechanical Turk. The results indicate that both incidental and dispositional fear increase ambiguity aversion, while no effect for anger was found. Given these results, it is not possible to determine which of the two approaches to affective influences better predicts the effect.

(4)

1. Introduction

The majority of standard (micro)economic theories is built on rational choice theory. This theory, also referred to as rational action or choice theory, is based on the assumption that people make decisions in their best self-interest (Green, 2002). They determine the options available and choose the most preferred one according to their consistent preferences. An important but often implicit assumption is that there are no limits to the reasoning power of the decision maker (Sandri, 2008). Other assumptions include that all information is freely available and that there are no search- or decision costs (Sandri, 2008).

Herbert Simon believed this method lacked realism and started his search for a new theory of economic behavior in the 1950’s (Barros, 2010). In contrast to the development of the rational choice theory, which was a result of armchair economics, he studied decision making empirically using laboratory experiments and observation techniques. One of his major findings in that period is the theory of bounded rationality (Simon, 1957), which would later be considered a revolution in decision theory.

Bounded rationality is the notion that decision makers are limited in their thinking capacity, available information and time (Simon, 1982). In the years after the invention of this theory, existing models based on rational choice were adjusted to include these cognitive and situational constraints. Even though this was an improvement, Simon noted that one important aspect was still missing: the role of emotion (Simon, 1990).

The belief that emotions affect the decision-making process is supported by papers in different fields of research. For example, the notion that personality traits influence the choices a person makes is confirmed in neuroeconomic research. Miu et al., (2008) investigate the effects of trait anxiety on decision making using psychophysiological measures of emotional states and the Iowa Gambling test, and show that trait anxiety impairs decision making. Furthermore, research in finance demonstrates that a mood caused by an event unrelated to the decision at hand, such as sunshine or a soccer loss, can influence the decision-making process of investors, which in its turn affects their stock returns (Hirshleifer & Shumway, 2003; Edmans et al., 2007).

Over the years, research focusing on the effect of emotions on decision making has grown substantially, and many types of behavior have already been researched, such as the influence of emotions on the endowment effect (Lin et al., 2006) and time preferences (Ifcher & Zarghamee, 2011). When it comes to decisions involving uncertainty, research has

(6)

However, little attention has been devoted to ambiguity preferences, where those outcomes are unknown. This is remarkable, since ambiguity better describes decisions in daily life. Therefore, the aim of this thesis is to answer the following research question:

What is the effect of incidental and dispositional emotions (fear and anger) on ambiguity preferences?

Using data from a survey distributed on Amazon Mechanical Turk it is shown that incidental fear increases ambiguity avoiding choices. Furthermore, it is found that people high on dispositional fear make more ambiguity averse choices.

This thesis is structured as follows. In section 2, the related literature will be discussed. In section 3 the experimental design, followed by the hypotheses, is outlined. Section 4 contains the descriptive statistics of the data and the results of the statistical tests and regressions. Section 5 presents the findings of this thesis, a discussion of its limitations and provides recommendations for future research.

(7)

2. Literature review

2.1 Emotions in decision-making process

Lerner et al. (2015) propose one of the first decision-making models that includes emotions. They use findings of more than three decades of research into emotion and decision making to develop a decision-making model that includes both rational choice and emotional inputs. This model is called the emotion-imbued choice (EIC) model and is shown in Figure 1.

The black lines represent the processes of the rational choice model (Lerner et al., 2015). As shown in Figure 1, lines A, B and C directly influence the evaluation process. The decision maker weighs the different options by assessing how much utility he gets from choosing the specific options (line A). Doing this, he takes into account the characteristics of the options, for example the time delays and the probabilities (line C). Additionally, personal characteristics such as the person’s risk preferences and discount rate influence the evaluation (line B). These factors combined provide the decision maker with a complete (rational) evaluation of the choice options.

Emotions are added to the rational choice model in two ways (Lerner et al., 2015). First, expected emotions are included as part of expected outcomes. They can be described as emotions that a decision maker is expecting to feel as a result of the outcomes associated with the different choices. Hence, these emotions are not experienced at the moment of choice, but are merely a prediction. A key feature is that expected emotions enter the decisions process as rational input, and are processed in the same way as (expected) utility.

The second type of emotions in the model are current emotions: the emotions felt during the decision-making process. The green lines in model display the potential sources of current emotions. This thesis focuses on two of these sources: incidental emotions and dispositional emotions. Incidental emotions, shown at the bottom of the model, emerge from situational origins unrelated to the decision at hand (Rick & Louwenstein, 2007). This could be the weather, mood, or an unrelated event. These incidental emotions cause momentary fluctuations in the emotional state of a decision maker.

Dispositional emotions or traits are part of the personality of the decision maker and can be described as fairly stable individual variations in the strength of emotional experiences (Larsen & Diener, 1987). Dispositional emotions generate a baseline level of current emotions (Lerner et al., 2015), which can cause the decision maker to react with a certain emotion relatively independent of the decision at hand (Wang et al, 2017).

(8)

Dispositional emotions are related to current emotions (emotional states) in two ways. It is shown that people high on dispositional fear experience more fear across situations at different points in time and that these people report higher levels of fear when induced with a negative emotion (Gross et al., 1998). However, regardless of their dispositional emotions, individuals still can experience any emotional state/current emotions (Baillon et al., 2016). Therefore, incidental and dispositional emotions can have different effects on decision making.

2.2 Approaches to affective influences

The most commonly used method for studying the effect of emotions on decision making is the valence-based approach, which compares the effect of negative and positive emotions. The question whether different emotions with the same valence influence decision making in the same way, remained largely unaddressed. In an attempt to answer this question, Lerner and Keltner (2000, 2001) propose the appraisal-tendency framework, which can be used when researching specific emotions. Both methods are discussed below.

The valence-based approach is derived from the circumplex model of affect, which proposes that all emotional states emerge from two systems in the brain (Russell, 1980). One of these systems is related to valence; the pleasantness of the emotion. The other system is related to arousal; the intensity of the emotion. As the name implies, the valence-based approach compares the effect of positive and negative emotions on decision making. For example, it is found that people in good moods make optimistic judgements and people in bad mood make pessimistic judgements (Wright & Bower, 1992). The valence-based

(9)

approach is used in psychology in a similar way to describe the effect of emotions on information processing. In that field of research, negative emotions are associated with detail-oriented and systematic types of processing (Schwarz & Clore, 1996), whereas positive emotions are associated with intuitive and holistic processing (Isen, 2004).

In contrast to the valence-based approach, recent findings indicate that emotions with the same valence can have different effects on decision making. Therefore, Lerner and Keltner (2000, 2001) developed a new and richer model, called the appraisal-tendency framework (ATF). The model is based on the cognitive appraisal theory of Smith and Ellsworth (1985) who argue that how a person experiences an emotion is related to how he appraises his environment along six cognitive dimension: pleasantness, anticipated effort, certainty, attentional activity, self-other responsibility/control, and situational control. The appraisal-tendency framework predicts that if emotions with certain cognitive dimensions are activated, people will assess future decisions in line with these dimensions (Lerner et al., 2015). For example, anger scores high on the certainty and individual control dimensions, whereas fear scores low on those dimensions. According to the appraisal-tendency framework, angry people will make different decisions than fearful people when facing a choice related to those cognitive dimensions.

2.3 Emotions and risk preferences

Considerable research exists on the effect of emotion on risk preferences. To illustrate the theoretical concepts described above, some previous work on this topic will be reviewed. An overview of the findings is shown in Table 1.

Risk refers to events where the probabilities of the outcomes are known (Knight, 1921). Risk preferences can be described the willingness of a person to bear risk (Einav et al., 2012). Most people are (slightly) risk averse, choosing a sure option over a risky option when the expected outcomes are equal (or even when the expected outcome of the risky option is higher). This is a result of loss aversion: the tendency to experience the pain of a loss more severely than the pleasure of an equal sized gain (Rabin & Thaler, 2001).

Zhao (2006) investigated the effect of incidental emotions on risk preferences using the valence-based approach. To change incidental emotions, different emotional manipulation methods can be used. He induced positive and negative emotions using a mathematical false-feedback test, after which the subjects participated in a gambling task. The results show that

(10)

people induced with negative emotions show stronger preference for the high-risk deck, whereas people induced with positive emotions preferred the low-risk deck.

Similarly, Chuang & Lin (2007) find, using autobiographic recollection as emotion induction method, that people in a positive mood are less risk taking than people in a negative mood.

In contrast, when examining the effect of different incidental emotions with the same valence using the appraisal-tendency framework it is shown that not all negative emotions have the same effect on risk preferences. For example, in their research Habib et al. (2015) induced emotions by showing pictures of facial expressions to their subjects. Afterwards, the subjects were asked to choose between a gamble and a sure option. They find that fear increases the number of a choices, while anger decreases the number of risk averse choices.

The appraisal-tendency framework yields similar results when researching the effect of dispositional emotions on risk preferences. Lerner and Keltner (2001) show that people high on dispositional fear make more risk averse choices, whereas people high on dispositional fear make more risk seeking choices. The valence-based approach is hardly used when examining the effect of dispositional emotions (traits), as it is common to look at emotional deficits or specific traits. Yet, some research indicates that negative affectivity is one of the predictors of risky behavior (Desrichard & Denarié, 2005).

2.4 Emotions and ambiguity preferences

In his examination of choices under uncertainty, Knight (1921) distinguished apart from risk also ambiguity, where the probabilities of the outcomes are not available. Ellsberg (1961) used this concept and showed in his famous two-color problem a pattern of preference that violated the subjective expected utility theory (Savage, 1954). This systematic violation would later be referred to as ambiguity aversion.

In the two-color urn experiment, two urns are described: urn 1 is filled with 50% black and 50% white balls and urn 2 is filled with the same amount of balls but with

Valence-based approach Appraisal-tendency framework Positive Negative Fear Anger Incidental emotions Decrease Increase Decrease Increase Dispositional emotions Decrease Increase Decrease Increase

Table 1: Summary table of the effect of incidental and dispositional emotions on risk taking using the valence-based approach and the appraisal-tendency framework. Increase: increased risk taking. Decrease: decreased risk taking.

(11)

unknown proportions of black and white balls. Participants have to choose which urn has higher probability of drawing a ball of a particular color. At this point, participants are indifferent about betting on white or black. This is consistent with the expected utility theory, as there is no reason to believe that the urn with the unknown composition contains an unbalanced proportion of white and black balls.

However, when participants are asked on which urn they would bet if the winning color is white, they choose urn 1 (known urn). When the winning color is black, participants again choose urn 1. This is a violation of the probability axioms. Choosing the known urn when the winning color is white implies that the participants believe that there are more white balls in the known urn, and therefore more black balls in the unknown urn. However, when the winning color is black, participants still choose the known urn. This pattern of preferences is nowadays known as the Ellsberg paradox.

The Ellsberg experiment is originally a thought experiment, used to highlight a deficiency in the expected utility theory. Subsequently, more formal versions of the experiments were conducted. Fellner (1961) showed different versions of the two-urn problem to undergraduate students and revealed their general tendency to prefer a 50-50 option over an option with unknown probabilities. Later, the widespread tendency of ambiguity aversion was confirmed in experiments with both students and non-students (i.e. Curley & Yates, 1989; Ho et al., 2002). Today, different versions of the Ellsberg experiment are still used as a model-free measure of ambiguity preferences.

Whereas on average most people are ambiguity averse, there is some heterogeneity in ambiguity attitudes. Dimmock et al. (2016) use a large representative US household survey to measure ambiguity preferences. They find that 52% of their participants are ambiguity averse, while 38% are ambiguity loving and only 10% are ambiguity seeking. Likewise, Potamites and Zhang (2012) find in a field experiment with Chinese small-scale stock investors that while most investors were ambiguity averse (57%), 26% are ambiguity loving and 15% are ambiguity neutral.

Despite the fact that the concepts of risk and ambiguity go hand in hand, the effect of emotions on ambiguity preferences is not as extensively studied. Besides, most of the papers on this topic focus on dispositional emotions. For example, Potatmites and Zhang (2012) find that Chinese small-stock investors who report higher levels of anxiety during the day are more ambiguity averse. This coincides with the prediction that anxious people are more ambiguity averse because ambiguity aversion engages in a neural circuitry that is also

(12)

the lack of dispositional happiness is related to higher ambiguity aversion (Yesuf & Feinberg, 2016).

The effect of incidental emotions on ambiguity preferences is hardly studied. In a recent paper Baillon et al. (2016) find that induced sadness leads to less ambiguity aversion, but the effect of other specific incidental emotions on ambiguity preference is still unknown. This paper attempts to fill (part of) this research gap.

2.5 Selection of emotions

When selecting the specific incidental emotions of which the effects on ambiguity preferences will be researched, two requirements are important. First, the emotions have to be of the same valence. Second, the emotions have to score differently on the cognitive appraisal dimensions related to ambiguous choices. If these requirements are met, it is possible to verify whether the valence-based approach or the appraisal-tendency framework better predicts the effects of emotion on ambiguity preferences.

It is preferred to study two emotions with negative valence, because specific negative emotions generate more mixed results than specific positive emotions on decision making (Baillon et al., 2016). Furthermore, the effects of negative emotions on decision outcomes are more differentiated than those of positive emotions (Baillon et al., 2016).

To predict which appraisal dimensions relate to ambiguity preferences, results from previous research on risk preferences are used. Lerner and Keltner (2000) predicted that the dimensions of certainty and control influence risk preferences, and confirmed this prediction in their successive research (Lerner & Keltner, 2001). Likewise, Tiedens and Linton (2001) show that certainty appraisals are an important influence on judgement and decision making, as feeling certain or uncertain influences judgements of certainty in following situations. Because risk is closely related to ambiguity, this paper argues that certainty and control are the two appraisal dimensions that influence ambiguity preferences.

Lerner and Keltner (2001) recommend researchers to compare emotions that are highly distinguished in their appraisal themes when using the appraisal-tendency framework. In Figure 2 and 3 two of the location estimate graphs of Smith and Ellsworth (1985) are presented. It is shown that fear and anger are the emotions most dispersed on the certainty and control axes. Moreover, they are both emotions with negative valence, which corresponds with the preference for negative emotions made earlier. For these reasons, anger and fear are the two emotions of interest in this paper.

(13)

2.6 Emotion regulation

Important to note is that people are not powerless in their battle against the effect of emotions on decision making. Recent findings in emotion regulation research show that humans often try to control their emotional experiences (Gross, 2002). Two emotion regulation strategies that are often studied are the cognitive reappraisal and the expressive suppression strategy. The cognitive reappraisal strategy acts before the emotions are activated and changes the emotional responses by redefining the meaning of the situation (Gross & Thompson, 2007). Expressive suppression is a strategy that acts after the emotions arise, and does so by inhibiting behavior associated with emotional responding, such as gestures and facial expressions.

It is shown that both strategies decrease the experience of positive emotions (Gross & Levenson, 1997), but only cognitive reappraisal can effectively reduce the experience of negative emotions (Troy et al., 2010). In line with these findings, Heilman et al. (2010) show that risk taking is increased when people use acute cognitive reappraisal, because this reduces the experience of the induced negative emotions (in their study: fear and disgust). Yet, expressive suppression had no effect on risk taking, as it does not reduce the experience of negative emotions.

Research shows that emotion regulation is not only related to emotional states but also to dispositional emotions. For example, Garnefski and Kraaij (2007) show that reappraisal is

Figure 2 and 3: Location estimate graphs of Smith and Ellsworth (1985). Emotions of interest for this thesis, anger and fear, are circled by the author.

(14)

negatively correlated with trait anxiety. This finding is consistent with that of Borkovec et al. (2004), who show that people who score high on trait anxiety avoid processing of emotion.

2.7 Amazon Mechanical Truk

The data for this research is collected using an online survey. Respondents will be recruited using Amazon Mechanical Turk (mTurk), an online platform launched in 2005. The platform connects companies and individuals in need for human intelligence (requesters) to people (workers) who are willing to work on surveys (Human Intelligence Tasks, HITs) in exchange for money.

It is shown that workers on Amazon mTurk express the same biases and heuristics and pay about the same amount attention as subjects from traditional sources (Paolacci et al., 2010) and that the reliability of the data obtained using the platform is as least as high as data obtained using traditional methods (Buhrmester et al., 2011). Moreover, data can be collected quickly and inexpensively. Altogether, mTurk is assumed to become a major tool for research in social sciences.

(15)

3. Methodology and Hypotheses

3.1 Data collection

Participants were recruited on the mTurk online platform. The workers were asked to fill out a 10 to 15 minute survey about personality, decision making and film clips. It is recommended to set the reward per completed HIT around US minimum wage, which currently is $7.25 per hour. This would mean that for a survey of about 10 to 15 minutes a reward of $1.21 - $1.81 should be chosen. However, it is found that realistic compensation rates do not improve data, but only speed up the collection process (Buhrmester et al., 2011). This led to the decision to set the compensation rate to $1.00. Still, the 200 participants were collected within three hours. A detailed description of the mTurk settings is added in Appendix A.

Approximately 50% of the workers on mTurk come from the United States and 40% come from India (Ipeirotis, 2010). In order to reach as much native English speakers as possible, the survey was distributed Tuesday 13 June 21:26 Central European Summer Time (UTC +2). At that moment it was 12:26 Pacific Daylight Time (UTC -7), 14:26 Eastern Standard Time (UTC -5) and 00:56 (June 14) in Indian Standard Time (UTC +5:30). Since it was afternoon in the United States and around midnight in India, it was most likely that mainly US workers would be reached.

3.2 Participants

Two hundred participants started the survey. Thirteen participants were prevented from finishing the survey and receiving payment because they did not pass the three control questions (12) or were using a mobile device which was not compatible with the film clips (1). Of the 187 participants who finished the survey, 113 were male and 74 were female. IP-addresses revealed that the majority of the surveys were filled out in the US (88%) and only a small part in India (7%), which suggests that the distribution time of the HIT had an effect on the demographics of the participants. The rest of the surveys (5%) was filled out in Canada, Indonesia, Australia and Russia.

(16)

3.3 Experimental design

The survey was programmed using the Qualtrics survey software and started with an introduction and a short consent statement. In the introduction the participants were informed that there would be “hidden attention checks”. This was done to legitimize possible rejections if people would behave inconsistently in the ambiguity and risk task. Subsequently, Qualtrics assigned the participants evenly to one of the three treatment groups: fear, anger or neutral. The neutral group was added as a control group. Depending on the treatment group, the incidental emotions of that type were induced. This was done by showing a short film clip, which is one of the most effective emotion manipulation methods (Gerrards-Hesse et al., 1994).

The short film clips (1-3 minutes) are found to reliably induce the emotions of interest (Schaefer et al., 2010; Gross & Levenson, 1995). For the anger treatment, a fragment of the movie Schindler’s List (1994) was used where a camp commander randomly shoots concentration camp prisoners from a balcony. A clip from the movie The Blair Witch Project (1999) was used to induce fear. It shows the final scene where the characters are supposedly killed. The neutral treatment consisted of a fragment from a National Geographic documentary about the Great Barrier Reef. After watching the film clip, the participants were asked to answer three control questions in order to check if they paid attention. If they answered 2 or 3 questions wrong, they received an end of survey message and were not able to continue the survey.

Subsequently, two tasks were presented to the participant. Participants were incentivized to show their real preferences, as one participant per task was selected for payment according to the choices he or she made. In Task 1, ambiguity preferences of the participants were measured using a multiple price list (MPL) version of the original Ellsberg experiment (Gneezy et al., 2015). Participants had to choose between drawing a ball form an urn with a known distribution of black and white balls, or an urn with an unknown distribution. The distribution of the balls stayed the same in each of the 20 decisions, but the payoff of winning when choosing the urn with the unknown distribution increased. The point where the subject switches from the urn with known distribution to the urn with unknown distribution is a measure of his level of ambiguity aversion.

Research into emotion and decision making has so far never used the mTurk platform. In order to be able to test if the data obtained on mTurk is comparable to the data obtained in for example a laboratory experiment, a measure for risk preferences was added to the survey

(17)

to see if previous findings could be replicated. The risk task was positioned after the ambiguity task in order to prevent participants from using the probabilities of success in the risk task as an anchor for their choices in the ambiguity task.

So in Task 2, participants had to fill out the Holt and Laury risk measurement task (Holt & Laury, 2002). The decisions were also shown in a multiple price list format, where the switching point is an indication of the subjects’ risk preferences. The participants were asked to choose between lotteries where the probabilities of each outcome were known. The outcomes of Lottery 1 were $2 and $1.60, and the outcomes of Lottery 2 were $3.85 and $0.10. The outcomes of the lotteries were the same in all 10 decisions, but the probability of the higher outcome increased up to point where the participant had to choose between a lottery with 100% chance of earning $2 and 100% chance of earning $3.85.

To measure the dispositional emotions of the participants, the tasks were followed by two sets of questions measuring dispositional fear and anger. Twenty questions from the State-Trait Anxiety Inventory (STAI) which measure trait anxiety were used to measure dispositional fear (Spielberger et al., 1970). As a measure for dispositional anger, ten questions from the State-Trait Anger Expression Inventory-2 (STAXI-2) were used (Spielberger, 1999). Consequently, the ability of the participants to regulate emotions was measured using the Emotion Regulation Questionnaire (ERQ) (Gross & John, 2003). This 10-item questionnaire measures both the cognitive reappraisal and the expressive suppression ability of the participant.

In the last block, the subjects were asked how the film clip made them feel. This was done using the Self-Assessment Mannequin method, where people have to rank their feelings on a 9-point scale which is accompanied with pictures that represent each type of feeling (Bradley & Lang, 1994). Two scales were used: a valence scale (positive-negative) and an arousal scale (excited-calm). These manipulation checks were intentionally added at the end, as adding them right after the film clip could reveal the purpose of the experiment (Baillon et

al., 2016). Also, the conscious assessment of one’s personal state may significantly alter

emotional processes (Kassam & Mendes, 2013). Despite the fact that emotions are not measured right after the film clip, any effects of the treatments are causal effects of the content of the film clips since the control treatment is a neutral film clip.

The questionnaire concluded with demographic questions, a question whether the subject had seen the movie before and a question about what the participant thought the purpose of showing the film clip was. The complete survey can be found in Appendix B.

(18)

3.4 Statistical Analysis

For the statistical analyses STATA 14 was used. To compare the means of the treatment groups, one-way ANOVA was used if the variable was normally distributed, otherwise the non-parametric Kruskall-Wallis test was performed. For the regression analyses, linear regression models (OLS) were computed.

3.5 Hypotheses

In line with the two approaches to affective influences described above, competing hypotheses will be used to determine which approach better predicts the effect of emotion on ambiguity preferences. The hypotheses are described as a comparison to the neutral treatment.

Appraisal-tendency framework

Because ambiguity is closely related to risk, this thesis uses the same line of reasoning as Lerner and Keltner (2001) used in their research into the effect of anger and fear on risk preferences. First, this thesis argues that the appraisal dimensions that influence ambiguity preferences are certainty and control. Second, it assumes that high certainty and high individual control decrease ambiguity aversion, while low certainty and low individual control increase ambiguity aversion. Because anger is associated with high certainty and control and fear with low certainty and control, this leads to the hypotheses as described below. Note that the appraisal-tendency framework predicts the same effect for both incidental and dispositional emotions, because it assumes that those emotions are related (Lerner & Keltner, 2000).

H1a Incidental/dispositional anger makes people less ambiguity averse.

H1b Incidental/dispositional fear makes people more ambiguity averse.

Valence-based approach

In setting up the hypothesis for the valence-based approach, different arguments for

incidental and dispositional emotions are used. For the prediction of the effect of incidental emotions, the findings of Wright & Bower (1992) are used, who argue that people in good moods make optimistic judgements when facing a decision involving uncertainty, whereas people in bad moods make pessimistic judgements. The relation between optimism and

(19)

pessimism on ambiguity aversion was studied by Pulford (2009). He found that optimistic people are less ambiguity averse than pessimistic people. Therefore, it is most likely that negative incidental emotions studied in this thesis are related to higher levels of ambiguity aversion.

As for dispositional emotions, previous research is used for setting up the hypothesis. As mentioned before, research indicates that trait anxiety is associated with ambiguity

aversion (Potatmites & Zhang, 2012). According to valence-based approach, this would mean that all negative emotions would make people more ambiguity averse. These predictions about the effect of incidental and dispositional emotions combined lead to the following hypothesis:

H2 Negative incidental/dispositional (anger, fear) emotions make people more ambiguity

(20)

4. Results

4.1 Participants

As mentioned before, 187 participants finished the survey. However, 21 participants were excluded since they had multiple switching points in the ambiguity and/or risk task, which does not satisfy monotonicity. Furthermore, 23 participants were excluded because they chose the low payoff option for the last question of the risk task, choosing a sure gain of $2 over a sure gain of $3.85, which reflects irrational behavior. Finally, 5 participants were excluded because they chose all the risk (ambiguity) loving options but were very ambiguity (risk) averse, or were extreme risk and ambiguity loving. It is hard to believe this behavior reflects real preferences, and it is probably a consequence of careless responding or misunderstanding the task. After excluding those participants 138 useful observations are left, which is about 74% of the participants who completed the survey and 69% of the participants who started the survey.

About 66% of the participants are men. The average age is 36 (SD: 9.24), with the youngest subject being 22 years old and the oldest 71 years old. Most participants indicated to have finished university or college (65%), followed by secondary school (18%) as highest completed education. About 15% of the participants have a level between secondary school and university and the remaining 2% only finished high school or less. Despite the exclusion of participants, the treatment groups are roughly equally sized: N=46 for anger, N=47 for fear and N=45 for the neutral treatment. About 20% of the participants guessed that the purpose of showing the film clip was to changing their emotional state. Another 23% guessed the same, but added that the emotions were induced to test the effect on decision making.

4.2 Summary statistics

Ambiguity preferences

Ambiguity preferences can be determined by the point on the multiple price list in Task 1 where a subject switches from choosing the risky option to the ambiguous option. In the following analyses, the share of ambiguity avoiding choices is used as a measure for this switching point. For example, when a subject chooses the risky option 6 times and thereafter 14 times the ambiguous option, his share is 0.3 (6/20). In the ambiguity task used in the survey, a share lower than 0.25 indicates ambiguity loving behavior, a share of 0.25

(21)

ambiguity neutrality and a share higher than 0.25 ambiguity aversion. In the survey data, the mean was 0.5732 with a standard deviation of 0.2858.

About 72% of the participants was ambiguity averse, 26% was ambiguity neutral and only 2% was ambiguity loving. These percentages do not match with findings in the literature on heterogeneity in ambiguity attitudes as reported in the literature section (Dimmock et al., 2016; Potamites & Zhang, 2012). This experiment found a much higher share of ambiguity averse participants (72% compared to 52% and 57% in the literature), a higher share of ambiguity neutral participants (27% compared to 10% and 16% in the literature) and a much lower share of ambiguity loving participants (2% compared to 26% and 38% in the literature).

The ambiguity data is normally distributed at a 5% confidence level, but not at a 10% confidence level (Shapiro-Wilk test for normality, p-value = 0.0724). The frequency graph in Figure 4 shows three noticeable peaks. About 26% of the participants had a share of 0.25 ambiguity avoiding choices, which indicates ambiguity neutrality. Furthermore, 15% had a share of 0.55, which corresponds to mild ambiguity aversion. Finally, about 20% had a share of 1, indicating severe ambiguity aversion. The size of the latter group does not match with previous findings. Therefore, it could be argued that this peak is caused by careless responding (only choosing the first option).

This assumption is supported by the finding that this group filled out the ambiguity task significantly faster than the other participants (Kruskal-Wallis, p-value 0.0371). They were also significantly faster in finishing the whole survey (Kruskal- Wallis, 10% significance level, p-value: 0.0547). However, the dataset from the experiment does not provide other information that can be used to check for careless responding. Therefore, it is not possible to confirm this assumption. For this reason, the participants with a share of ambiguity avoiding choices of 1 are not excluded. Appendix C provides additional analyses on the dataset where those subjects are excluded.

Risk preferences

Risk preferences can be determined by the point on the multiple price list in Task 2 where a subject switches from choosing the lower payoff to the higher payoff option. In the following analyses, the share of risk avoiding choices is used as a measure for this switching point. For example, when a participant chose the lower payoff option 6 times and thereafter 4 times the higher payoff option, his score is 0.6 (6/10). In the risk task used in the survey, a score lower

(22)

0.4 risk aversion. The mean in the survey data was 0.5949 with a standard deviation of 0.1525.

As shown in Figure 5, the risk data is approximately normally distributed, which is confirmed by the Shapiro–Wilk test for normality (p-value: 0.9988). Moreover, the graph shows that most participants are risk averse, which coincides with previous findings in the literature.

4.3 Manipulation checks

To check if the film clips had the intended changes in emotional state, tests are carried out to see if subjects in the three treatment groups reported different levels of arousal and valence.

Valence was not normally distributed in the total sample (Sharpiro-Wilik test, p-value: 0.0077). Therefore the non-parametric Kruskal-Wallis test was used to test if there are significant differences among group means. The test indicated that there was a significant difference between two or more means (p-value: 0.0001), but the test does not show between which ones. Therefore, Dunn’s test was carried out, which results indicate that all means are significantly different from each other (p-values: <0.0034). As shown in Figure 6, subjects induced with anger reported on average the most negative emotions (highest valence scores), followed by fear. Participants in the neutral treatment group reported positive emotions (low valence scores). Note that the boxplots depicted in this thesis show the median instead of the mean.

Arousal was normally distributed (Sharpiro-Wilik test, p-value: 0.5730). Therefore, a one-way ANOVA test in combination with the Bonferroni multiple-comparison test was used to check if there are significant differences among group means. The results indicated that the means of the anger and fear treatment groups differed significantly from the neutral group (p-value: 0.0000) but that they were not significantly different from each other (p-(p-value:

(23)

1.0000). As shown in Figure 7, subjects in the anger and fear condition reported significantly more excitement than the subjects in the neutral treatment group.

No relation was found between the treatment groups and the dispositional anger and fear scores, indicating that the film clips had no effect on how the subjects reported their dispositional anger and fear.

4.4 Hypotheses tests

Incidental emotions

To test whether the mean ambiguity avoiding choices differs between the treatment groups, the non-parametric Kruskal-Wallis test is used because the ambiguity data is not normally distributed. This test indicates that there are no significant differences between the three groups (p-value: 0.3179).

The data is further explored using linear regression analysis. The results are presented in Table 2, the regression equations can be found in Appendix D. In model 2 and 3 the treatment group fear had a significant effect on ambiguity preferences (p-values: 0.032 and 0.024 respectively). The results indicated that the induction of incidental fear increases the share of ambiguity avoiding choices. Moreover, in model 2 the interaction term between the treatment group fear and the arousal score is significant (p-value: 0.043). This implies, as shown in Figure 8, that subjects in the treatment group fear who reported excitement after seeing the film clip made more ambiguity avoiding choices than subjects in the same group who reported calmness.

Figure 6: Reported valence scores per treatment group. Low valence indicates positive feelings, high valence indicates negative feelings.

Figure 7: Reported arousal scores per treatment group. Low arousal scores indicate excitement, while high scores indicate calmness.

(24)

One of the assumptions for linear regression is that the residuals of the regression have to be normally distributed. However, this assumption is not satisfied in any of the models reported in Table 2. Log, square-root or arcsine transforming the ambiguity data did not solve this problem. Therefore, the results in this section should be interpreted with caution.

The risk data was – in contrary to the ambiguity data – normally distributed. Therefore, an ANOVA test was used to test if the mean risk avoiding choices differed between the treatment groups. The output of this test showed that there was no significant difference between the groups (p-value: 0.3200).

The same linear regressions were performed for risk. The results of the regression and the regression equations can be found in Appendix E and . In model 2 and 3 the treatment group fear had a significant effect on risk preferences (p-values: 0.022 and 0.029, respectively). The results indicated that inducing incidental fear increases risk avoiding choices. Secondly, in model 2 arousal was significant (p-value: 0.078), indicating that participants who reported higher arousal scores (calmer) were more risk averse. Finally, the interaction term between the treatment group fear and the reported arousal was significant in models 2 and 3 (p-values: 0.017 and 0.010 respectively). This indicates that that people who reported more excitement in the fear group made more risk averse choices, see Figure 9.

Figure 8: Interaction effect treatmentgroupXarousal on ambiguity preferences. Low arousal scores indicate excitement, while high scores indicate calmness.

Figure 9: Interaction effect treatmentgroupXarousal on risk preferences. Low arousal scores indicate excitement, while high scores indicate calmness.

(25)

Table 2: Summary linear regression analysis for ambiguity preferences Model 1 Model 2 Model 3 Model 4 Treatment anger 0.0135 (0.0942) 0.207 (0.233) 0.262 (0.236) 0.0903 (0.214) Treatment fear 0.0799 (0.0826) 0.500** (0.231) 0.537** (0.235) -0.267 (0.242) Arousal 0.00263 (0.0140) 0.0350 (0.0260) 0.0561 (0.0446) 0.00976 (0.0148) Valence -0.00201 (0.0160) -0.00979 (0.0163) -0.0118 (0.0165) -0.0145 (0.0259) Treatment*Arousal Anger -0.0174 (0.0353) 0.0269 (0.0495) Fear -0.0709** (0.0347) -0.0490 (0.0454) Treatment*Valence Anger -0.000461 (0.0360) Fear 0.0621 (0.0440) Dispositional fear 0.00823 (0.00572) Treatment*Arousal* Dispositional fear Neutral -0.000558 (0.000901) Anger 0.00185 (0.00132) Fear -0.00124 (0.00132) N 138 138 138 138 AIC 53.10 52.04 55.93 54.44 R2 _0.013 _0.049 _0.077 _0.032

Standard errors in parentheses. * p<0.10, ** p<0.05, *** p<0.01

Dispositional emotions

To test if there is a relation between dispositional emotions and ambiguity preferences, the non-parametric Spearman rank correlation was used because the measures of ambiguity and dispositional anger and fear were not normally distributed. The results of the tests indicated that both dispositional anger (p-value: 0.2552) and fear (p-value: 0.3014) were not correlated with the share of ambiguity avoiding choices. Likewise, no relation was found between risk preferences and the anger and fear scores (p-values: 0.8432 and 0.6738 respectively).

However, when regressing the share of ambiguity avoiding choices on treatment group and dispositional fear and anger1, the effect of dispositional fear was significant at a 10% level (p-value: 0.072). This result indicates the higher a person’s dispositional fear, the

(26)

more ambiguity voiding choices he makes. However, the errors of this regression are again not normally distributed, violating one of the linear regression assumptions.

When risk was used as the dependent variable in the same regression analyses, no significant relation was found between risk and the two dispositional emotions.

4.5 Additional tests

Effect of ERQ

To test if there is a relationship between incidental emotions and the two emotion regulation strategies, suppression and reappraisal, correlations were calculated between the separate emotion regulation scores and the valence and arousal scores. The Spearman’s correlation tests – and Pearson correlation for suppression and arousal as those are normally distributed – turned out not to be significant.

The same method is used to test if there is a significant relationship between dispositional anger and fear and the two emotion regulation strategies. No relation is found between the dispositional emotions and the suppression strategy. However, both dispositional anger and fear are negatively correlated with the reappraisal score (Spearman’s test, p-values: 0.0070 and 0.0000 respectively), which implies that the better someone is at reappraising his emotions, the lower his dispositional anger and fear is.

Relation dispositional and incidental emotions

Linear regression analysis was used to determine whether dispositional emotions influence the valence and arousal scores the participants reported. When valence was regressed on the two dispositional emotions and treatment group2, dispositional anger turned out to be significant (p-value: 0.050). The results show that the higher someone is on dispositional anger, the more negatively he rated the film clip. No significant results were found for fear.

2

(27)

5. Discussion and conclusion

5.1 Summary of findings

The aim of this thesis was to find the effect of incidental and dispositional anger and fear on ambiguity preferences. Furthermore, risk preferences were included to test the reliability of the online Amazon Mechanical Turk platform.

The results of this thesis suggest that incidental fear, if induced properly, increases ambiguity aversion. Furthermore, the analyses show that people high on dispositional fear are more ambiguity averse. However, no relation was found between incidental and dispositional anger on ambiguity preferences. Therefore, it cannot be determined whether the valence-based approach or the appraisal-tendency framework better predicts the effect of emotions on ambiguity preferences, as the findings are in line with both hypothesis 1b and 2.

As for risk, the only significant finding was that incidental fear, if induced properly, increases risk aversion. However, other previous findings – as reported in Table 1 – could not be replicated. Therefore, this thesis cannot confirm that mTurk provides reliable data.

5.2 Limitations

Even though the focus of this paper was on ambiguity preferences, risk preferences were included to test if previous findings could be replicated using a different data collection method. However, the only result replicated was the effect of incidental fear. The fact that no significant effect of incidental anger was found could be a result of the order of the tasks in the survey. It is possible that the emotions induced by the film clips already faded out by the time the subjects reached the risk task, which can be caused by their focus on answering the ambiguity task or simply the time passed. Researching ambiguity and risk in two different experiments (between-subjects design) should counter this problem.

The results of the ambiguity task did not correspond to findings in previous research. Too many participants were ambiguity averse and neutral, while too few were ambiguity loving. Moreover, the frequency graph showed three noticeable peaks, see Figure 4. The one in the middle was expected: most people are mildly ambiguity averse. However, a large share of participants indicated they were ambiguity neutral, which does not match previous findings. This might be caused by the fact that people did not completely understand the task, or that the monetary incentives did not work properly. Moreover, about 20% of the

(28)

needs to be extended with more choice options, to capture the switch point of every participant, even if he is extremely ambiguity averse. However, this behavior could also be explained by misunderstanding of the task, careless responding (only choosing the first option) or the lack of trust in the experimenter, for example believing that the ambiguous urn will be intentionally filled with few winning color balls.

The assumption that the severe ambiguity averse behavior was caused by careless responding was supported by the fact that this group finished both the ambiguity task and the whole survey significantly faster. Therefore, in Appendix C analyses were performed on the dataset were the severe ambiguity averse participants were excluded. Remarkable is that these analyses provide significant results for anger instead of fear, showing – in line with the appraisal-tendency approach – that incidental anger decreases ambiguity aversion. Further research with a different ambiguity measure is needed to check which of the findings can be replicated.

About 25% of the data collected on mTurk had to be excluded. A group of 20 workers who obviously did not pay attention and switched from urns more than once, were rejected from receiving payment via the mTurk platform. After being rejected about 10 people sent angry e-mails, often including swear words, claiming the rejection was unfair. One participant also posted a negative review on Turkopticon.com, a website where workers review HITs and requesters. For illustration, two e-mails and a review are added in appendix F. It took quite some effort and time to explain the reason for rejection to these workers and to calm them down. Supposedly, most companies and researchers do not have time for this type of customer care activities, and will therefore decide not to exclude any worker in order to avoid the hassle. However, because workers who provide bad quality work will not be rejected anymore, they might end up with high HIT approval rates and master worker titles. This will have negative consequences for the quality of the platform.

One thing that stood out in the correspondence with the workers was their major focus on control questions. Presumably, there is a group of workers who first determine if a question can be used as a basis for rejection before answering it. If so, they fill it out attentively, like the control questions about the film clip in this experiment. If not, they just answer it quickly and inattentively, in order to finish the survey faster. In the case of this survey, some subjects could have figured that the dispositional emotion questions can never be used as a basis for rejection, since they merely measure preferences. They might therefore have chosen to fill it out randomly. This can be an explanation for the few results obtained with the data from the dispositional questionnaires.

(29)

Despite asking the subjects to rank their feelings about the film clip at the very end of the survey, a large share of the subjects guessed that the film clips were used as an emotion induction method, or even guessed the purpose of the study. Therefore, it cannot be ruled out that the findings are a result of the experimenter’s demand effect. Nevertheless, this is a limitation common to experiments using emotion induction techniques and is hard to overcome without using deception.

The manipulation check showed that the subjects in the fear or anger treatment group reported higher levels of arousal. Furthermore, people induced with anger reported the most negative emotions, followed by fear. However, using the data from this survey it cannot be determined if the participants actually felt angry or fearful. Therefore, in following research a more extensive emotion questionnaire should be used, for example the Differential Emotions Scale (Izard, 1977).

This thesis cannot determine if people induced with incidental fear made more ambiguity and risk avoiding choices because of the level of certainty and control they felt, as predicted by the appraisal-tendency framework. The finding that incidental fear increases ambiguity aversion can, for example, also be explained by the approach-withdrawal model (Davidson et al., 1990), which describes anger as an approach emotion and fear as a withdrawal emotion.

5.3 Conclusion and future research

This thesis shows that both incidental and dispositional fear increase ambiguity aversion. The data for this research was collected using Amazon Mechanical Turk. Although the data collection was relatively cheap and very fast, the reliability of the platform is questionable, as previous findings for risk preferences could not be replicated.

A recommendation for future research is to use a different data collection method with a higher level of control, for example a laboratory experiment. Secondly, other emotion induction techniques could be used. Thirdly, the effect of more specific emotions on ambiguity preferences could be tested, as well as the effect of these emotions in the loss domain. Finally, it could be interesting to determine why these emotions lead to different behavior (i.e. appraisal tendencies, approach-withdrawal tendencies).

(30)

References

Baillon, A., Koellinger, P. D., & Treffers, T. (2016). Sadder but wiser: The effects of emotional states on ambiguity attitudes. Journal of Economic Psychology, 53, 67-82.

Barros, G. (2010). Herbert A. Simon and the concept of rationality: boundaries and procedures. Revista de economia política, 30(3), 455-472.

Borkovec, T. D., Alcaine, O., & Behar, E. (2004). Avoidance theory of worry and generalized anxiety disorder. Generalized anxiety disorder: Advances in research and practice, 2004.

Bradley, M. M., & Lang, P. J. (1994). Measuring emotion: the self-assessment manikin and the semantic differential. Journal of behavior therapy and experimental psychiatry, 25(1), 49-59.

Buhrmester, M., Kwang, T., & Gosling, S. D. (2011). Amazon's Mechanical Turk: A new source of inexpensive, yet high-quality, data?. Perspectives on psychological science, 6(1), 3-5.

Chuang, S. C., & Lin, H. M. (2007). The effect of induced positive and negative emotion and openness-to-feeling in student’s consumer decision making. Journal of Business and Psychology, 22(1), 65-78.

Curley, S. P., & Yates, J. F. (1989). An empirical evaluation of descriptive models of ambiguity reactions in choice situations. Journal of Mathematical psychology, 33(4), 397-427.

Davidson, R. J., Ekman, P., Saron, C. D., Senulis, J. A., & Friesen, W. V. (1990). Approach-withdrawal and cerebral asymmetry: Emotional expression and brain physiology: I. Journal of personality and social

psychology, 58(2), 330.

Desrichard, O., & Denarié, V. (2005). Sensation seeking and negative affectivity as predictors of risky

behaviors: A distinction between occasional versus frequent risk-taking. Addictive behaviors, 30(7), 1449-1453

Dimmock, S. G., Kouwenberg, R., Mitchell, O. S., & Peijnenburg, K. (2015). Estimating ambiguity preferences and perceptions in multiple prior models: Evidence from the field. Journal of risk and uncertainty, 51(3), 219-244.

Dimmock, S. G., Kouwenberg, R., Mitchell, O. S., & Peijnenburg, K. (2016). Ambiguity aversion and household portfolio choice puzzles: Empirical evidence. Journal of Financial Economics, 119(3), 559-577. Edmans, A., Garcia, D., & Norli, Ø. (2007). Sports sentiment and stock returns. The Journal of Finance, 62(4),

1967-1998.

Einav, L., Finkelstein, A., Pascu, I., & Cullen, M. R. (2012). How general are risk preferences? Choices under uncertainty in different domains. The American economic review, 102(6), 2606-2638.

Ellsberg, D. (1961). Risk, ambiguity, and the Savage axioms. The quarterly journal of economics, 643-669. Fellner, W. (1961). Distortion of subjective probabilities as a reaction to uncertainty. The Quarterly Journal of Economics, 670-689.

Garnefski, N., & Kraaij, V. (2007). The cognitive emotion regulation questionnaire. European Journal of Psychological Assessment, 23(3), 141-149.

Gerrards-Hesse, A., Spies, K., & Hesse, F. W. (1994). Experimental inductions of emotional states and their effectiveness: A review. British journal of psychology, 85(1), 55-78.

Gneezy, U., Imas, A., & List, J. (2015). Estimating Individual Ambiguity Aversion: A Simple Approach (No. w20982). National Bureau of Economic Research.

Green, S. L. (2002, May). Rational choice theory: An overview. In Baylor University Faculty development seminar on rational choice theory.

(31)

Gross, J. J. (2002). Emotion regulation: Affective, cognitive, and social consequences. Psychophysiology, 39(3), 281-291.

Gross, J. J., & John, O. P. (2003). Individual differences in two emotion regulation processes: implications for affect, relationships, and well-being. Journal of personality and social psychology, 85(2), 348.

Gross, J. J., & Levenson, R. W. (1995). Emotion elicitation using films. Cognition & emotion, 9(1), 87-108. Gross, J. J., & Levenson, R. W. (1997). Hiding feelings: the acute effects of inhibiting negative and positive emotion. Journal of abnormal psychology, 106(1), 95.

Gross, J. J., & Thompson, R. A. (2007). Emotion regulation: Conceptual foundations.

Gross, J. J., Sutton, S. K., & Ketelaar, T. (1998). Relations between affect and personality: Support for the affect-level and affective-reactivity views. Personality and social psychology bulletin, 24(3), 279-288. Habib, M., Cassotti, M., Moutier, S., Houdé, O., & Borst, G. (2015). Fear and anger have opposite effects on risk seeking in the gain frame. Frontiers in psychology, 6.

Hartley, C. A., & Phelps, E. A. (2012). Anxiety and decision-making. Biological psychiatry, 72(2), 113-118. Heilman, R. M., Crişan, L. G., Houser, D., Miclea, M., & Miu, A. C. (2010). Emotion regulation and decision making under risk and uncertainty. Emotion, 10(2), 257.

Hirshleifer, D., & Shumway, T. (2003). Good day sunshine: Stock returns and the weather. The Journal of

Finance, 58(3), 1009-1032.

Ho, J. L., Keller, L. R., & Keltyka, P. (2002). Effects of outcome and probabilistic ambiguity on managerial choices. Journal of Risk and Uncertainty, 24(1), 47-74.

Holt, C. A., & Laury, S. (2002). Risk aversion and incentive effects.

Ifcher, J., & Zarghamee, H. (2011). Happiness and Time Preference: The Effect of Positive Affect in a Random-Assignment Exper. The American Economic Review, 101(7), 3109-3129.

Ipeirotis, P. G. (2010). Demographics of mechanical turk.

Isen, A. M. (2004, April). Positive affect facilitates thinking and problem solving. In Feelings and emotions: The Amsterdam symposium (pp. 263-281). Cambridge University Press.

Izard, Carroll E. (1977), Human Emotions, New York, NY: Plenum Press.

Kassam, K. S., & Mendes, W. B. (2013). The effects of measuring emotion: Physiological reactions to emotional situations depend on whether someone is asking. PloS one, 8(6), e64959.

Knight Frank, H. (1921). Risk, uncertainty and profit. книга.

Larsen, R. J., & Diener, E. (1987). Affect intensity as an individual difference characteristic: A review. Journal of Research in personality, 21(1), 1-39.

Lerner, J. S., & Keltner, D. (2000). Beyond valence: Toward a model of emotion-specific influences on judgement and choice. Cognition & Emotion, 14(4), 473-493.

Lerner, J. S., & Keltner, D. (2001). Fear, anger, and risk. Journal of personality and social psychology, 81(1), 146.

Lerner, J. S., Li, Y., Valdesolo, P., & Kassam, K. S. (2015). Emotion and decision making. Annual Review of Psychology, 66.

(32)

Lin, C. H., Chuang, S. C., Kao, D. T., & Kung, C. Y. (2006). The role of emotions in the endowment effect. Journal of Economic psychology, 27(4), 589-597.

Miu, A. C., Heilman, R. M., & Houser, D. (2008). Anxiety impairs decision-making: psychophysiological evidence from an Iowa Gambling Task. Biological psychology, 77(3), 353-358.

Paolacci, G., Chandler, J., & Ipeirotis, P. G. (2010). Running experiments on amazon mechanical turk. Peer, E., Vosgerau, J., & Acquisti, A. (2014). Reputation as a sufficient condition for data quality on Amazon Mechanical Turk. Behavior research methods, 46(4), 1023-1031.

Potamites, E., & Zhang, B. (2012). Heterogeneous ambiguity attitudes: a field experiment among small-scale stock investors in China. Review of Economic Design, 1-21.

Pulford, B. D. (2009). Is luck on my side? Optimism, pessimism, and ambiguity aversion. The Quarterly Journal of Experimental Psychology, 62(6), 1079-1087.

Rabin, M., & Thaler, R. H. (2001). Anomalies: risk aversion. The Journal of Economic Perspectives, 15(1), 219-232.

Rick, S., & Loewenstein, G. (2007). The role of emotion in economic behavior.

Russell, J. A. (1980). A circumplex model of affect. Journal of Personality and Social Psychology, 39, 1161-78. Sandri, S. (2008). Reflexivity in economics: An experimental examination on the self-referentiality of economic theories. Springer Science & Business Media.

Savage, L.J. (1954) The foundations of statistics. New York: John Wiley

Schaefer, A., Nils, F., Sanchez, X., & Philippot, P. (2010). Assessing the effectiveness of a large database of emotion-eliciting films: A new tool for emotion researchers. Cognition and Emotion, 24(7), 1153-1172. Schwarz, N., & Clore, G. L. (1996). Feelings and phenomenal experiences. Social psychology: Handbook of basic principles, 2, 385-407.

Simon, H. (1990). Reason in human affairs. Stanford University Press. Simon, H. A. (1957). Models of man; social and rational.

Simon, H. A. (1982). Models of bounded rationality: Empirically grounded economic reason (Vol. 3). MIT press.

Smith, C. A., & Ellsworth, P. C. (1985). Patterns of cognitive appraisal in emotion. Journal of personality and social psychology, 48(4), 813.

Spielberger, C. D. (1999). State-Trait anger expression inventory. John Wiley & Sons, Inc..

Spielberger, C. D., Gorsuch, R. L., & Lushene, R. E. (1970). Manual for the state-trait anxiety inventory. Tiedens, L. Z., & Linton, S. (2001). Judgment under emotional certainty and uncertainty: the effects of specific emotions on information processing. Journal of personality and social psychology, 81(6), 973.

Troy, A. S., Wilhelm, F. H., Shallcross, A. J., & Mauss, I. B. (2010). Seeing the silver lining: cognitive reappraisal ability moderates the relationship between stress and depressive symptoms. Emotion, 10(6), 783. Wang, Y., Gu, R., Luo, Y. J., & Zhou, C. (2017). The interaction between state and dispositional emotions in decision making: an ERP study. Biological psychology, 123, 126-135.

(33)

Wright, W. F., & Bower, G. H. (1992). Mood effects on subjective probability assessment. Organizational behavior and human decision processes, 52(2), 276-291.

Yesuf, M., & Feinberg, R. M. (2016). Ambiguity aversion among student subjects: the role of probability interval and emotional parameters. Applied Economics Letters, 23(4), 235-238.

Zhao, J. (2006, May). The effects of induced positive and negative emotions on risky decision making. In Talk presented at the 28th Annual Psychological Society of Ireland Student Congress, Maynooth, Ireland.

(34)

Appendix A

Description of HIT to workers

Title: Survey about personality, decision making and film clips

Description: Survey that will take 10-15 minutes with chance of earning a bonus. The survey is about personality, decision making and film clips. Audio required. No mobile devices due to compatibility issues.

Keywords: personality, film clips, decision making

Settings HIT

Reward per assignment3 _$1.00

Number of assignments per HIT 20 and 1604

Time allotted per assignment5 _{30 minutes}

Hit expires in 4 days

Auto-approve and pay workers6 _{3 days}

Worker requirements

Require that workers be Masters to do your HITs7 _Yes

Specify any additional qualifications workers must meet to work on your HITs8_None

Project contains adult content No

HIT Visibility Private

3_{See section Methodology}

4_{A test run of 20 and the final run of 160 participants. This does not add up to the 200 (total number of}

participants) because mTurk does not count the participants who did not finished the survey due to the control questions or wrong device. Moreover, some participants were rejected and a new observations were collected.

5_{The time allotted has to be sufficient for a participant to fill out the survey without rushing. However, workers}

see it as an indication of the length of the survey, and choose not to participate if the time allotted is very high.

6_{Time a requester gets to review the results and potentially reject HITs before all HITS accepted and have to be}

paid for. Workers are known to avoid HITs with long auto-approval times.

7_{Masters are workers who have shown to complete HITs with certain accuracy. Additional costs: 5% of the}

reward.

8_{The qualifications can be about a variety of things, for example income, handedness, social media account}

holder, and location. For most of the qualifications, an extra fee has to be paid. One of the most important qualifications is the HIT Approval Rate. It is shown that workers who are approved more than 95% of the time provide higher quality data (Peer et al., 2014). However, since the Master requirement was already selected, there was no use in selecting this requirement.

The effect of incidental and dispositional emotions (fear and anger) on ambiguity preferences

Master thesis

MSc Economics: Behavioral Economics and Game Theory