Don’t make decisions when you’re angry, don’t make promises when you’re happy : how different emotional states affect your risk choices

(1)

Don’t make decisions when you’re angry,

Don’t make promises when you’re happy:

How different emotional states affect your risk choices

Victor C. Fernandez, 11085517

MSc Economics: Behavioural Economics and Game Theory

University of Amsterdam, Faculty of Economics and Business

15 ECTs August 2016

Supervisor: dhr. dr. J.B. (Jan) Engelmann Second examiner: dhr. dr. A. (Aljaz) Ule

Abstract

Emotions change people’s behavior. And day to day events have an effect on our emotions. We analyze the effect of a random income effect on our subjects’ emotions using a reduced PANAS questionnaire and the effect of these on their risk aversion as measured by Holt and Laury (2002) in an online laboratory using the oTree tool. We find that only the negative income shock had a significant effect on the subjects’ risk aversion, this caused by being upset due to the loss of income, which translated to a reduction in their risk aversion. Subjects’ naturally occurring emotion regulation strategies did not have an effect on these results. We recommend an increase in the subject pool and a physical laboratory setting for future development of this hypothesis.

(2)

Statement of Originality

This document is written by Student Victor C. Fernandez who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1. Introduction

Imagine the following setting: in one of your days off, you decide to go to a casino. You are not doing this out of need, merely due to entertainment, so you decide to just spend some time at the roulette table, as this depends merely on luck, and the only decision you make is the amount of money you will gamble in each round and the number you think the little white ball will land on. Are you the type of person that gambles all on one number, increasing the risk but also the payment? Or will you spread your money across different numbers, increasing the odds of winnings but a lower gain?

Imagine now that while you are there, you receive a phone call: You have been fired! You are sad, upset, who knows how you are feeling, but you’re not feeling the same. Do you continue gambling as if nothing had happened? Maybe you are so sad you decide to stop gambling, or so mad that you decide to put it all on one number out of anger at the universe. One thing is almost certain: your behavior changes, and this is clearly due to the emotional impact of the information you received.

Traditional economics would say that you would behave exactly in the same way with and without this information, the homo economicus, or “rational man” – consistent throughout time and all-knowing and discounting of the future – would have known the chance of this news arriving and would have already adjusted his behavior from the moment he set foot inside the casino. That is wrong.

The purpose of this thesis is to add to traditional economic theory some specific knowledge that has been understood in psychology for decades, specifically by answering the following question: How do economic shocks affect people’s emotions and risk-taking behavior?

The rest of this document is structured as follows: Section 2 discusses the related literature, which is mainly in the field of psychology; Section 3 presents the theoretical framework and original economic model that it is to follow; Section 4 describes the experimental design; Section 5 presents the results of the study; Section 6 discusses the results, further analysis and closing remarks.

(4)

2. Related Literature

2.1. Emotions and risk taking

The studies on how emotions alter decision making go back for decades, increasing exponentially in the past decade (Figure 1). The body of research is so vast that Lerner et al. (2015) have classified it in eight themes of knowledge. We do not intend to do an exhaustive review of all the related literature but rather focus on the general results that the literature gives rise to in the field of risk aversion.

Figure 1. Number of scholarly publications from 1970 to 2013 that refer to “emotion(s)/affect/mood and decision making” (green bars) and proportion of all scholarly publications referring to “decision making” that this number represents.

Source: Lerner et al. (2015)

Chuang and Lin (2007) use the method of asking their subjects to recall experiences in order to induce negative emotions, and then measure their willingness to take risks using the Personal Risk Inventory to find that subjects influenced by negative emotions are more likely to engage in risk taking than subjects influenced by positive emotions. Raghunathan et al. (1999) induce emotions in their subjects with a role playing

(5)

sadness are risk seeking but those induced with anxiousness are risk averse. This is an important conclusion as it shows that emotions with the same valence might have different effects, so emotion analysis must go deeper than just positive emotions and negative emotions.

Hockey et al. (2000) go one step further. Instead of measuring induced emotions during a laboratory experiment, they choose to measure the level of anxiety, depression and fatigue that a person experiences in their day to day life, and use a hypothetical

everyday choice scenario to measure their risk appetite. They conclude that only fatigue was a predictor of risk taking, and that it had a negative effect on risk aversion (that is, it increased risk taking).

Smith et al. (2005) extend this research more into the economics space. Using heart rate data as proxy for emotion, they conclude that in the setting of a Dutch auction, this information can act as a predictor for prices, however the same result was not found in the case of an English auction. This shows than more than just the outcome of the activity (as in this case both were auctions) or the framing (or in their terms, the institutions) can have an impact with regard to the importance of emotions.

The effect of emotions within framing is further confirmed in Druckman et al. (2008). In that paper they show that framing has an effect on risk taking, but the magnitude of this effect is dependent on the emotions and intensity of emotions that the subjects are experiencing at the moment of the framing. Moreover, Lee et al. (2015) show that the same emotion – in this case, fear – might elicit both risk aversion or risk seeking behavior based solely on the framing of the experiment – an investment experiment framed as a financial decision or a casino.

Kugler et al. (2012) find that the impact of framing is not restricted to just scenarios: if the uncertainty of the experiment is generated by a computer or by a human, the same emotion will generate different risk behavior. In their investment experiment Kugler et al. find that when comparing fear and anger, if the uncertainty is generated by a machine, fear will induce more risk aversion than anger, but if the uncertainty is generated by a human, the opposite will be true.

(6)

Finally, Campos-Vazquez et al. (2015) use the prospect theory model to estimate the effect of induced emotions on risk. They find that inducing sadness leads to an increase in risk aversion.

2.2. Emotion Regulation

Looking deeper into the field of emotion and its effect on decision making we find the field of emotion regulation, which is the control that people exert over the emotions that they feel (Gross and John, 2003). We will focus on two regulation processes: cognitive reappraisal (thinking something different) and expressive suppression (suppressing the emotions). While still at a nascent stage, current research in this area also shows promise in playing a role to explain risk-taking behavior.

Heilman et al. (2010) show that using a balloon analogue risk task, inducing cognitive reappraisal in subjects reduces their risk aversion. But research has also been done without inducing emotion regulation strategies. Panno et al (2013) find that usual use of the emotion regulation strategies predicts risk taking when the decisions require thinking and deliberation, as opposed to gut feeling.

2.3. Why is this document different?

As we can appreciate from the literature, the approach to measuring emotion and risk aversion in psychology is as diverse as its conclusions. One fact that remains is that most experiments induce specific emotions to be measured in the experiments. We plan to take a different approach by creating a positive and a negative income shock –

similar to that done by Haushofer et al. (2013) – to affect our subjects emotionally, allowing us to potentially analyze the underrepresented positive valence emotions with the positive shock. This also has the effect of grounding this research more to an

economic reality by setting the shock be an actual scenario instead of a fictitious one. In addition to this, we plan to measure naturally occurring emotion regulation, instead of induced emotions. In effect, we try to make the experiment a more naturally occurring scenario than those presented in the literature (within the boundaries of what can be considered “natural” in a laboratory experiment).

(7)

3. Theoretical framework

Our experiment follows the original model and experiment developed by Holt and Laury (2002). In this experiment subjects were presented with Table 1 and told to choose either Option A or Option B for each row, subsequently a row would be chosen and paid, so there was a clear incentive for subjects to report their preference correctly. The logic behind the table is as follows: beginning at the first row, the potential payoffs of Option A are less variable than Option B, and so with a probability of only 1/10 for the high payoff in both options, only an extreme risk seeker would choose Option B over Option A (this is further reinforced by the fact that the expected payoff of Option A is $1.17 higher than Option B). As subjects continue going down in the table, the probability of the higher

payoff goes up, until in the 10th_{row there’s a certain payoff of $2 for option A and $3.85}

for Option B, so even the most risk averse person will prefer Option B.

Table 1. The ten-paired lottery-choice decisions with low-payoffs

Source: Holt and Laury (2002)

This guarantees that no matter the risk profile of the subject, eventually they will switch from A to B, and once they switch they will keep this choice until the last row. These values are chosen in such a way that, assuming constant relative risk aversion, the utility

function takes the form of 𝑈(𝑥) = 𝑥1−𝑟_{for 𝑥 > 0. In this equation, risk seeking is}

represented by 𝑟 < 0, risk neutrality by 𝑟 = 0 and risk aversion by 𝑟 > 0. Subjects will arrive at one of the risk classifications shown in Table 2 based on the point where they

(8)

change from choosing Option A to Option B. These values are chosen so that 4 safe choices followed by 6 risky choices signifies risk neutral, and 6 safe choices followed by

4 risky choices signal a utility of around 𝑈(𝑥) = 𝑥0.5_{, which has been reported as a}

reasonably standard level of risk aversion in econometric analysis of auction data (Holt and Laury, 2002). In addition, this model allows all payoffs in Table 1 to be scaled up by a constant without the loss of generalization.

Table 2. Risk aversion classifications based on lottery choices

Source: Holt and Laury (2002) 4. Experimental design

4.1. The design

The experiment had a between-subjects design where subjects were randomly assigned to one of three treatments: positive shock, negative shock or no shock. Subjects read the instructions, where they were informed of their initial income during the experiment – €5 for the positive shock treatment, €15 for the negative shock treatment, and €10 on the no shock treatment - which would be part of their final winnings at the end of the experiment. Subjects then took part in three rounds of a modified Holt and Laury (2002) risk

(9)

measurement test (described below). The difference in treatments in the experiment consisted of the change from a subject’s given initial income to €10 at the end of the first round. This produced a positive income shock for the €5 treatment, a negative income shock for the €15 treatment and no income shock for the €10 treatment (control). Since the end income for every treatment was €10, this allowed us to measure the impact of the shock without worrying about a pure income effect on the subjects. In addition to this, subjects’ emotions were measured three times following a modified PANAS test: once at the beginning of the experiment before the first round, once at the end of the first round of the experiment after receiving the notification of the income shock and once at the end of the third round. Afterwards, subjects were asked to answer some control questions and the Emotion Regulation Questionnaire (ERQ).

4.2. The Modified Holt and Laury (2002) test

As discussed above, Holt and Laury’s (2002) original experiment presented subjects with a single table, where only the assigned probabilities changed from one pair of choices to the next. This experiment changes this approach in two important ways. First, it disposes of the single table in favor of individual choices between two lotteries, where each ten of these, then, represent a whole “table” and thus, a full measurement of the risk profile of a subject. This is done in order to gather data on the time each subject spends on each individual decision. It also allows us to force each subject to consider the lotteries in a vacuum, without the influence of previous decisions. The second change done to the original experiment is that the values do not remain the same throughout the different lottery choices. This is done as to control for mechanization on the subjects’ part, as there is a chance that if all the numbers were the same, subjects would not consider each choice individually. This is especially important in an online environment, as we have no control over the actual environment and distractions each subjects experiences, so it would be easier for subjects to miss information if most of what is presented to them were to be too similar. This, however, presents a change in the actual values from one lottery to the next, and as seen also in Holt and Laury (2002), people’s behavior varies depending on the magnitude of the lotteries. An assumption is then made in this experiment: since the conclusions of the base paper relies on big increases in magnitude

(10)

(15 times and 20 times the magnitude of the “low” treatment), we assume that a low variation in the magnitude of the lotteries will not have a significant impact. The sets are then created by creating increments of 5 cents on the expected value of the first row of choices of the lottery with less variability, from €3 to €5, and from these 40 different groups of values, we choose 30 of them to compose the three sets of 10 lottery pairs to be part of the experiment, which are shown in the following table. The order of these choices within each set is also randomized, as another means of controlling for subjects getting used to a particular order. A full step-by-step of the experiment from the perspective of the subject is available at Appendix C.

(11)

Table 3. Set of choices and rounds where they are presented Lottery A Lottery B 1/10 of 4.57, 9/10 of 3.66 1/10 of 8.8, 9/10 of 0.23 2/10 of 5, 8/10 of 4 2/10 of 9.63, 8/10 of 0.25 3/10 of 5.12, 7/10 of 4.1 3/10 of 9.86, 7/10 of 0.26 4/10 of 4.09, 6/10 of 3.27 4/10 of 7.86, 6/10 of 0.2 5/10 of 5.55, 5/10 of 4.44 5/10 of 10.68, 5/10 of 0.28 6/10 of 4.94, 4/10 of 3.95 6/10 of 9.51, 4/10 of 0.25 7/10 of 5.98, 3/10 of 4.78 7/10 of 11.5, 3/10 of 0.3 8/10 of 4.33, 2/10 of 3.46 8/10 of 8.33, 2/10 of 0.22 9/10 of 4.76, 1/10 of 3.8 9/10 of 9.16, 1/10 of 0.24 10/10 of 4.15, 0/10 of 3.32 10/10 of 7.98, 0/10 of 0.21 Set A (Choices 1-10) Lottery A Lottery B 1/10 of 3.66, 9/10 of 2.93 1/10 of 7.04, 9/10 of 0.18 2/10 of 3.84, 8/10 of 3.07 2/10 of 7.39, 8/10 of 0.19 3/10 of 5.49, 7/10 of 4.39 3/10 of 10.56, 7/10 of 0.27 4/10 of 4.7, 6/10 of 3.76 4/10 of 9.04, 6/10 of 0.23 5/10 of 4.45, 5/10 of 3.56 5/10 of 8.57, 5/10 of 0.22 6/10 of 4.88, 4/10 of 3.9 6/10 of 9.39, 4/10 of 0.24 7/10 of 3.9, 3/10 of 3.12 7/10 of 7.51, 3/10 of 0.2 8/10 of 4.39, 2/10 of 3.51 8/10 of 8.45, 2/10 of 0.22 9/10 of 5.91, 1/10 of 4.73 9/10 of 11.39, 1/10 of 0.3 10/10 of 5.73, 0/10 of 4.59 10/10 of 11.03, 0/10 of 0.29 Set B (Choices 11-20) Lottery A Lottery B 1/10 of 3.78, 9/10 of 3.02 1/10 of 7.28, 9/10 of 0.19 2/10 of 5.67, 8/10 of 4.54 2/10 of 10.92, 8/10 of 0.28 3/10 of 4.63, 7/10 of 3.71 3/10 of 8.92, 7/10 of 0.23 4/10 of 3.72, 6/10 of 2.98 4/10 of 7.16, 6/10 of 0.19 5/10 of 3.96, 5/10 of 3.17 5/10 of 7.63, 5/10 of 0.2 6/10 of 4.21, 4/10 of 3.37 6/10 of 8.1, 4/10 of 0.21 7/10 of 5.37, 3/10 of 4.29 7/10 of 10.33, 3/10 of 0.27 8/10 of 5.24, 2/10 of 4.2 8/10 of 10.09, 2/10 of 0.26 9/10 of 6.04, 1/10 of 4.83 9/10 of 11.62, 1/10 of 0.3 10/10 of 6.1, 0/10 of 4.88 10/10 of 11.74, 0/10 of 0.3 Set C (Choices 21-30)

(12)

4.3. Addition of the third round

The analysis of the impact of the income shocks on the decisions of subjects could have been completed with only the first two rounds of the experiment. The third round of the experiment was added to have a look beyond this analysis, into the territory of how emotional shocks dissipate and if these have a further impact on future behavior.

4.4. Procedure and online environment

The experiment was done in an online environment using the oTree software. A total of 68 subjects successfully finished every section of the experiment, with an additional two choosing to skip the ERQ. These two are used as part of the total sample in every analysis except in those that involve the ERQ. Regarding their assignment to the different treatments, 22 subjects were assigned to the positive shock, 22 to the negative shock, and 26 to the no shock treatment.

It is important to note that there are clear differences between an online laboratory and a physical one, including the compulsiveness of completion once subjects are present. As a matter of fact, the 70 subjects mentioned above were but 34% of the total potential pool of subjects that actually accessed the URL during the time the experiment ran. However, we do not expect this attrition rate to have a significant impact on the behavior that we are trying to observe. Hergueux and Jacquemet (2012) show that comparing online laboratories with offline laboratories, people behave substantially the same, being only less risk averse in the former environment. Since our question is based around the change in risk aversion of our subjects due to the income shock (and not based on the absolute level of risk aversion at any given time), the assumption we must make here is that even though initial levels of risk aversion may differ between online and offline environments, the effect of the shock will be the same regardless of environment.

4.5. Payment of the subjects

Each subject had a 10% chance of obtaining their final payment. This was done in order to reduce experimental costs but there is evidence that this method of payment does not bias results Laury (2005). The winnings of each subject were clear to them from the start, these being equal to their non-lottery dependent income (which was €10 for every subject

(13)

by the end of the experiment, independent of their treatment) plus their actual winnings of a randomly drawn lottery from the 30 choices they made during the experiment. This guaranteed that subjects acted with the same incentives throughout all the rounds, as each subject had the same chance of being chosen for payment.

5. Results

5.1. Consistency

One of the first things we need to check is the consistency of the subjects in our experiment. It is expected that a rational agent will only have one switching point from the lottery with less variance (Lottery A from our design) to the one with more variance (Lottery B in our design) as the probability that the bigger payoff is chosen increases. In essence, once any person starts choosing lottery B, they should continue to choose B as long as the probability assigned to the bigger payoff increases. This, however, is not always the case. Figure 2 shows the number of rounds where people showed consistent behavior vs. inconsistent behavior.

Figure 2. Consistency during rounds

As we can see, 30% of the rounds played by all subjects show signs of inconsistency. Moreover, Figure 3 shows that the inconsistencies are not accumulated within one person

30%

70%

(14)

(meaning either a person is either perfectly consistent or he is not), showing that the subject pool as a whole displayed different degrees of inconsistency.

Figure 3. Consistent rounds by subjects

As we can see from Figure 3, only 29 of the 70 subjects in the experiment (41%) are perfectly consistent throughout the rounds, the other 41 have some level of inconsistency: Five were inconsistent in every round, 11 were inconsistent in two rounds, and 25 were inconsistent in one round. The simplest fix in our dataset would be to eliminate subjects who show inconsistencies in their behavior, however, that would result in the loss of 59%

of our data, and leave us with too few data points to do any statistical inference.1_The

solution proposed (and used in the next section) is to calculate three different measurements of risk. The simplest one is to assume that every lottery A choice was meant to be sequential even if it was not, and so if a subjects chose lottery B in round 2, but kept choosing A until round 7, it is considered that they chose A until round 6 - we shall call this Sum of choices. An alternative solution would consider the same situation described before but consider rounds 3 to 6 to be mistakes, and so the real turning point would be round 2 – we call this First choice. A final potential solution considers everything before consecutive choices to be a mistake, which would consider round 7 to be the turning point in this scenario – we call this Last choice. The analysis in this paper chooses

1_{Nevertheless, for those curious enough to see the analysis in this scenario, this is reported in Appendix B.} 5 11 25 29 0 5 10 15 20 25 30 35 0 1 2 3

(15)

to focus mainly on Sum of choices, reporting the results of the analysis for First choice and Last choice in Appendix A.

5.2. Effect of the treatments on emotions

The first question we must ask is if the treatments – the income shocks - have the intended effect, i.e. whether they had an effect on subjects’ emotions.

As we can see from the results shown in Tables 4, 5 and 6, the treatments had the expected effect, influencing how interested, upset, enthusiastic, irritable nervous and attentive subjects were. The direction of the effects are also expected, and so we see that a positive income shock makes subjects less upset and irritable, and a negative income shock makes subjects less interested, more upset, less enthusiastic, less nervous and less attentive (Table 6). We choose to present three tables due to the fact that, while analyzing the variables through fixed effects – and so controlling for more unobservables than in the other two analyses – shows us the more promising results, these do not seem to survive many robustness checks. This may be either due to the size of the shock (only €5) or the relatively low number of observations (70 subjects, with only slightly more than 20 subjects in each treatment).

(16)

Table 4. Effect of the treatments on emotions in round 2

Table 5. Effect of the treatments on emotion change between rounds 1 and 2

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Variables Interested Distressed Excited Upset Scared Hostile Enthusiastic Irritable Alert Inspired Nervous Attentive Positive 0.0320 0.0854 -0.274 0.148 -0.0942 0.380 -0.261 0.235 -0.262 -0.0888 -0.0460 -0.213 (0.351) (0.266) (0.354) (0.250) (0.135) (0.230) (0.344) (0.285) (0.347) (0.368) (0.157) (0.348) Negative -0.114 0.507* -0.405 0.909*** -0.0882 0.260 -0.560 0.334 -0.173 -0.244 0.173 -0.664* (0.355) (0.269) (0.358) (0.252) (0.137) (0.233) (0.347) (0.288) (0.350) (0.372) (0.159) (0.352) Cognitive Reappraisal 0.0852 -0.106 0.335* 0.152 0.0458 0.134 0.143 0.150 0.309 0.373* 0.0971 0.139 (0.192) (0.146) (0.194) (0.137) (0.0741) (0.126) (0.188) (0.156) (0.190) (0.201) (0.0861) (0.191) Expressive Suppression -0.250 -0.0985 -0.107 0.0170 -0.0338 -0.199 0.121 -0.0646 0.310 0.0914 -0.0475 0.116 (0.218) (0.165) (0.220) (0.155) (0.0840) (0.143) (0.213) (0.177) (0.215) (0.228) (0.0977) (0.216) Constant 2.035*** 1.905*** 2.117*** 0.953** 1.319*** 1.006** 2.005*** 1.477*** 2.071*** 1.502** 1.270*** 2.966*** (0.643) (0.487) (0.648) (0.457) (0.248) (0.421) (0.629) (0.522) (0.635) (0.674) (0.288) (0.638)

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations 68 68 68 68 68 68 68 68 68 68 68 68

R-squared 0.168 0.148 0.141 0.304 0.060 0.082 0.131 0.057 0.175 0.097 0.229 0.116

Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Variables Interested Distressed Excited Upset Scared Hostile Enthusiastic Irritable Alert Inspired Nervous Attentive

Positive 0.320 0.0233 -0.278 -0.359 -0.106 0.260 -0.147 -0.587* 0.214 0.155 -0.0854 0.0523 (0.295) (0.254) (0.315) (0.320) (0.122) (0.188) (0.310) (0.298) (0.340) (0.300) (0.220) (0.273) Negative -0.0860 0.466* -0.303 0.455 -0.0527 0.223 -0.702** -0.160 -0.239 -0.240 -0.167 -0.463* (0.299) (0.256) (0.318) (0.324) (0.123) (0.190) (0.313) (0.301) (0.344) (0.303) (0.223) (0.276) Cognitive Reappraisal 0.109 0.0427 0.200 0.154 0.0425 0.0541 -0.0284 0.108 0.104 0.196 0.0623 -0.0205 (0.162) (0.139) (0.172) (0.175) (0.0668) (0.103) (0.170) (0.163) (0.186) (0.164) (0.121) (0.149) Expressive Suppression -0.524*** -0.151 -0.485** 0.0723 0.00135 -0.0743 -0.120 0.318* 0.100 0.0650 0.154 -0.0904 (0.183) (0.157) (0.195) (0.199) (0.0757) (0.117) (0.192) (0.185) (0.211) (0.186) (0.137) (0.169) Constant -0.761 -0.664 -1.119* -0.776 0.159 -0.854** 0.0673 -0.577 0.277 -0.101 0.400 -0.0571 (0.541) (0.464) (0.576) (0.586) (0.223) (0.344) (0.567) (0.545) (0.623) (0.549) (0.403) (0.499)

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations 68 68 68 68 68 68 68 68 68 68 68 68

R-squared 0.177 0.112 0.229 0.192 0.048 0.144 0.155 0.225 0.047 0.079 0.201 0.139

(17)

Table 6. Effects of the treatments on emotions – Fixed Effects

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Variables Interested Distressed Excited Upset Scared Hostile Enthusiastic Irritable Alert Inspired Nervous Attentive

Positive 0.0909 -0.273 -0.0455 -0.409* -0.0909 -0.0909 0 -0.636*** 0.182 0.0455 -0.182 0.0909 (0.198) (0.168) (0.220) (0.206) (0.0766) (0.129) (0.195) (0.208) (0.213) (0.194) (0.151) (0.176) Negative -0.364* 0.0455 -0.273 0.455** -0.0455 -0.0455 -0.636*** -0.0909 -0.0909 -0.273 -0.318** -0.500*** (0.198) (0.168) (0.220) (0.206) (0.0766) (0.129) (0.195) (0.208) (0.213) (0.194) (0.151) (0.176) Constant 3.386*** 1.779*** 2.579*** 1.450*** 1.157*** 1.329*** 2.829*** 1.757*** 2.707*** 2.279*** 1.400*** 3.064*** (0.0710) (0.0602) (0.0786) (0.0739) (0.0274) (0.0460) (0.0699) (0.0745) (0.0763) (0.0694) (0.0540) (0.0629) Observations 140 140 140 140 140 140 140 140 140 140 140 140 R-squared 0.050 0.038 0.023 0.114 0.025 0.009 0.135 0.123 0.013 0.029 0.080 0.109 Number of id 70 70 70 70 70 70 70 70 70 70 70 70

(18)

5.3. Effect of the treatments on behavior

Table 7. Effect of the treatments on risk taking

Table 7 partially confirms our hypothesis and shows that a negative income shock has an effect on subjects’ risk taking decisions. Having a negative income shock made our sample decrease their risk aversion - in terms of our 10 decision profile, the small

negative income shock was enough to make subjects switch from safe lotteries to riskier lotteries one decision earlier. On the other hand, the positive income shock had no effect on subject behavior. With respect to the emotion regulation strategies, we see here that the habitual strategies (as opposed to induced strategies) do not seem to have an effect on risk aversion. However, this could be misrepresented, as potential opposite effects in the different treatments could be cancelling out when analyzing the whole sample. Table 8 fixes this issue by analyzing the impact of the emotion regulation on the positive treatment subjects (columns 1 and 2), negative treatment subjects (columns 3 and 4) and neutral treatment subjects (columns 5 and 6). While it seems that cognitive reappraisal has a weakly significant effect on the negative treatment subjects, it does not seem to be robust and is not significant once we add the controls. It is important to

(1) (2) (3)

Variables Sum of choices Sum of choices Sum of choices

Positive 0.255 0.178 -0.455 (0.500) (0.575) (0.426) Negative -0.881* -0.772 -0.727* (0.500) (0.582) (0.426) Cognitive Reappraisal -0.0358 (0.315) Expressive Suppression -0.205 (0.357) Constant 5.654*** 4.215*** 5.800*** (0.338) (1.054) (0.152) Controls No Yes No

Fixed Effects No No Yes

Observations 70 68 140

R-squared 0.073 0.117 0.056

Note: Columns 1 and 2 refer to the data obtained on the second round, i.e. immediately after the income shock. Column 3 refers to data from rounds 1 and 2.

(19)

note, however, that these regressions are done with 20 observations, making it very hard to make statistical inference from them.

Table 8. Effect of Emotion Regulation Strategies on Risk taking by treatment

An additional question which arises is, following the negative income shock, which specific emotion is responsible for the change in risk appetite. As shown above, the negative treatment had an effect in several different emotions. Table 9 shows that of all the emotions recorded, the only one that is both affected by the negative treatment and affects behavior is Upset, which would lead us to conclude that being upset changes risk taking behavior in this scenario. Hostile also seems to be significant, making subjects more risk averse, though this is we do not consider this a relevant result as neither of our treatments had a direct effect on this emotion.

(1) (2) (3) (4) (5) (6)

Variables Sum of choices Sum of choices Sum of choices Sum of choices Sum of choices Sum of choices

Cognitive Reappraisal 0.301 0.464 -1.373* -0.991 0.364 0.620 (0.366) (0.437) (0.718) (0.837) (0.461) (0.488) Expressive Suppression -0.637 -1.038 0.167 0.324 -0.104 0.958 (0.457) (0.604) (0.561) (0.682) (0.493) (0.669) Constant 5.623*** 3.432** 5.584*** 2.051 5.488*** 5.277*** (0.365) (1.440) (0.593) (2.954) (0.415) (1.551)

Controls No Yes No Yes No Yes

Observations 22 21 22 22 26 25

R-squared 0.100 0.249 0.164 0.312 0.029 0.292

(20)

Table 9. Effect of the emotions on risk taking

5.4. Effect of the time on emotions and risk taking

A big unexplored (or at least under-reported) area of analysis in the literature is the effect that emotional changes can have on the time subjects take to make decisions. An investigation into the timeliness of decisions could shed light on why subjects change

(1) (2) (3)

Interested -0.227 -0.412 -0.0732 (0.315) (0.350) (0.361) Distressed 0.183 0.296 0.133 (0.289) (0.313) (0.349) Excited 0.132 0.314 0.303 (0.370) (0.408) (0.301) Upset -0.652** -0.631** -0.767** (0.272) (0.289) (0.306) Scared 0.757 0.710 0.294 (0.658) (0.730) (0.809) Hostile 0.326 0.284 1.124** (0.416) (0.437) (0.457) Enthusiastic 0.308 0.191 -0.0495 (0.355) (0.404) (0.330) Irritable -0.193 -0.132 0.0237 (0.333) (0.344) (0.261) Alert -0.136 -0.227 0.0316 (0.297) (0.322) (0.294) Inspired -0.263 -0.266 -0.149 (0.308) (0.324) (0.348) Nervous 0.356 0.364 -0.155 (0.485) (0.608) (0.357) Attentive 0.399 0.508 0.321 (0.290) (0.314) (0.356) Constant 4.065*** 2.071 3.763** (1.078) (1.404) (1.521) Controls No Yes No

R-squared 0.249 0.328 0.205

(21)

their behavior after a shock, and would allow us to see if riskier decisions are in fact correlated with a shorter or longer response time.

Table 10. Effect of response time on choice taken.

As we can see from Table 10 there does not seem to be any relation between the time taken to make a decision and if the decision is riskier or not (as denoted by choosing lottery B over lottery A). In addition, Table 11 confirms this by showing that the emotions

(1) (2) (3) (4)

Variables Choice Choice Choice Choice

Response time -0.000408 0.0112 0.0199* 0.0223 (0.000563) (0.00713) (0.0104) (0.0233) Interested 0.0775 0.0378 0.0434 0.0447 (0.0945) (0.102) (0.100) (0.114) Distressed -0.00788 0.0322 -0.0301 -0.0782 (0.0796) (0.0852) (0.0841) (0.0982) Excited -0.168 -0.146 -0.124 -0.147 (0.116) (0.128) (0.129) (0.157) Upset 0.287*** 0.295*** 0.347*** 0.323*** (0.0934) (0.0998) (0.0973) (0.108) Scared -0.749*** -0.835*** -0.741*** -0.806*** (0.220) (0.233) (0.232) (0.271) Hostile -0.580*** -0.607*** -0.591*** -0.711*** (0.108) (0.114) (0.113) (0.134) Enthusiastic -0.0300 -0.00582 -0.00655 0.0465 (0.113) (0.122) (0.120) (0.147) Irritable 0.415*** 0.426*** 0.419*** 0.427*** (0.0826) (0.0890) (0.0885) (0.103) Alert 0.194** 0.175* 0.141 0.00216 (0.0866) (0.0928) (0.0928) (0.112) Inspired 0.168* 0.149 0.134 0.234* (0.101) (0.109) (0.107) (0.127) Nervous -0.128 -0.236 -0.267* -0.260 (0.133) (0.145) (0.145) (0.180) Attentive -0.268*** -0.242** -0.238** -0.276** (0.0931) (0.0983) (0.0969) (0.112) Cognitive Reappraisal 0.107 0.131 0.123 0.305*** (0.0890) (0.0955) (0.0938) (0.112) Expressive Suppression 0.122 -0.00140 0.00303 -0.0921 (0.0960) (0.105) (0.103) (0.119) Constant -4.029*** -4.160*** -4.068*** -3.795*** (0.352) (0.378) (0.379) (0.449) Observations 2,070 1,876 1,887 1,472

(22)

that were impacted by the treatments (namely, upset) have no influence on the time taken for a subject to make a decision during the experiment. The different columns in these tables show the same results for different levels of strictness in the data. Since the experiment is online, distractions are a factor and there is a lot dispersion in the time taken in each decision. To correct for this, column 2 of Table 11 only looks at the decisions that are within 10 seconds of the decision time of each subject, column 3 only looks at decisions made within 30 seconds, and column 4 only looks at decisions made within 15 seconds. The results are not significant independent of treatment type.

(23)

Table 11. Effect of emotions on Response time

The addition of round 3 allows us to analyze the passage of time in a different way: it permits us to see if the impact created by the treatment lingers or if it is reduced as time goes by. Table 12 shows investigates this question by studying the case where it is relevant – that is, Table 12 includes only the subjects that were assigned to the negative income shock, as this was the only treatment group were an emotional impact was linked

(1) (2) (3) (4)

Variables Response time Response time Response time Response time

Interested 3.946 1.510*** 1.148*** 0.439*** (4.751) (0.352) (0.222) (0.128) Distressed -1.936 -0.471 -0.271 -0.259** (4.017) (0.299) (0.189) (0.111) Excited -13.75** -3.181*** -1.906*** -0.617*** (5.829) (0.437) (0.283) (0.174) Upset -1.171 -0.209 0.0673 0.283** (4.688) (0.346) (0.217) (0.122) Scared 17.15 0.837 0.185 0.587** (10.71) (0.787) (0.504) (0.295) Hostile -8.208 -0.766* -0.295 -0.268* (5.324) (0.392) (0.250) (0.147) Enthusiastic 2.187 0.654 0.601** 0.385** (5.686) (0.427) (0.268) (0.166) Irritable 5.696 0.476 -0.123 -0.145 (4.094) (0.303) (0.193) (0.114) Alert 0.734 0.180 0.271 -0.168 (4.337) (0.323) (0.206) (0.125) Inspired 8.695* 1.259*** 0.201 0.0457 (5.054) (0.376) (0.237) (0.142) Nervous -1.253 1.394*** 0.866*** 0.244 (6.660) (0.502) (0.325) (0.201) Attentive -0.418 0.567* 0.252 0.434*** (4.653) (0.342) (0.216) (0.126) Cognitive Reappraisal 7.129 0.823** 0.250 -0.0572 (4.489) (0.333) (0.210) (0.125) Expressive Suppression -1.761 1.215*** 0.976*** -0.0171 (4.847) (0.363) (0.229) (0.133) Constant 12.07 7.861*** 9.448*** 6.598*** (16.48) (1.202) (0.761) (0.446) Observations 2,070 1,876 1,887 1,472 R-squared 0.010 0.075 0.069 0.075

(24)

to statistically significant results. In this case, analyzing the whole sample of subjects would only under represent the significance of the impact of time, if found at all.

Table 12. Differences in risk taking between rounds 2 and 3

As we can see from the results in Table 12, there does not seem to be a decrease in the effect of the negative income shock from round 2 to round 3. This raises the

question of exactly how long it takes for the shock to wear off and for the risk level to go

(1) (2) (3)

Round 3 0.273 0.304 0.341 (0.583) (0.663) (0.494) Interested -0.0590 -0.0249 (0.501) (0.432) Distressed 0.238 0.382 (0.329) (0.289) Excited 0.178 -0.0791 (0.589) (0.518) Upset -0.445 -0.505* (0.308) (0.249) Scared 1.028 2.017* (1.404) (1.164) Hostile 0.580 0.435 (0.549) (0.484) Enthusiastic 0.437 0.736 (0.707) (0.726) Irritable -0.440 -0.0842 (0.341) (0.287) Alert -0.127 -0.762 (0.555) (0.486) Inspired 0.0787 -0.367 (0.536) (0.516) Nervous 1.031 1.024 (0.812) (0.739) Attentive -0.0578 1.035 (0.804) (0.732) Constant 4.773*** 1.767 -5.339* (0.412) (1.626) (2.607) Controls No No Yes Observations 44 44 44 R-squared 0.005 0.337 0.710

(25)

back to neutral levels. Unfortunately this falls outside of the scope and capacity of this study.

6. Discussion and closing remarks

6.1. Consistency

One issue surrounding the experiment was the consistency issues denoted above. It would be interesting to analyze what caused these issues and so Table 13 shows a simple analysis of the effect of the control variable on the consistency of the subjects (Consistency here being a variable that is 1 if the subject is consistent in all three rounds and 0 otherwise, and so it is estimated with a logit). It is interesting to find that older people are less consistent. One would be tempted to conclude that older subjects may be busier and face more distractions and so an online experiment would affect them more, however, this would contradict the result that a higher average response time actually correlates with a bigger consistency. To improve on this issue it would be interesting to compare these results with the same experiment run in a physical laboratory, as it would give us yet another point of comparison on the two experimental methods.

6.2. Effect of Emotion Regulation Strategies on Risk Aversion

The literature on this subject is still young and there is much to learn, however the results that we obtained do not accord with previous literature. Since each subject could deliberate as long as they wanted on each lottery pair, this was a “cold” experiment, however, there was no effect of the different emotion regulation strategies on the risk behavior of the subjects as Panno et al. (2013) concluded, even when analyzed on a treatment-by-treatment basis. This last analysis was only done with a subject pool of 20, and so we cannot completely refute the prior research done in this area, but we also cannot support it in any way.

(26)

Table 13. Effect of control variables on consistency of subjects

6.3. Effect of Emotions on Risk Aversion

We have found that a negative income shock will make subjects upset, making them less risk averse, successfully addressing our research question. We found no effect of a positive income shock in any emotion or change in risk aversion, leaving us unable to fill the gap in the literature which is mostly filled with analysis of negative emotions. This leads us to have two potential conclusions regarding the positive income shock: either it does not affect people emotionally at all, or a bigger shock is required to create this effect. Due to the fact that people are happy when they win the lottery, we are willing to give the benefit of the doubt to positive income shocks and recommend an increase in the positive shocks in any further studies.

(1) VARIABLES Consistency Age -0.0733** (0.0371) Gender -0.426 (0.572) Internet use -0.0435 (0.117)

Monthly income -2.47e-05 (4.51e-05) Positive 0.0422 (0.718) Negative -0.584 (0.717) Cognitive Reappraisal 0.458 (0.395) Expressive Suppression -0.0260 (0.456) Response time 0.0870* (0.0509) Constant 1.419 (1.431) Observations 68

(27)

6.4. Prospect theory and different models

The theoretical framework presented in this document is not the only economically valid one. In fact, following more modern approaches than the classical Bernoulli function, one should follow the Prospect Theory model and estimate it with methods such as those shown by Prelec (1998), which is more complex and more in line with modern behavioral economics findings. However such models require a larger sample to be estimated correctly, and cannot be estimated with the data collected from the Holt and Laury (2002)

model, and so it has to be left for future research to cover. 2

6.5. Closing remarks

If anything is to be taken from this document it is that even a small-scale shock can have an immediate effect on some of our emotions and subsequently on our behavior. The message that this sends is slightly different than that presented in the rest of the literature in that not only is it saying that being upset will make you less risk averse, but also that being notified that you will receive €5 less than expected in any given moment will significantly alter your behavior for at least the following 10 minutes. Given the random amount of shocks a normal person faces every day, it becomes increasingly harder to believe in a unified theory of normal human behavior, much less a perfectly rational homo

economicus.

2_{The Prelec (1998) model requires variation in both the odds of the gambles and the quantities, while the Holt and} Laury (2002) model only creates variation of probabilities, not quantities, not allowing enough degrees of freedom to estimate the parameters.

(28)

References

Campos-Vazquez, R. M., & Cuilty, E. (2014). The role of emotions on risk aversion: A Prospect Theory experiment. Journal of Behavioral and Experimental Economics, 50, 1-9.

Chen, D. L., Schonger, M., & Wickens, C. (2016). oTree—An open-source platform for laboratory, online, and field experiments. Journal of Behavioral and Experimental Finance, 9, 88-97.

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.

Druckman, J. N., & McDermott, R. (2008). Emotion and the framing of risky choice. Political Behavior, 30(3), 297-321.

Derks, D., Fischer, A. H., & Bos, A. E. (2008). The role of emotion in computer-mediated communication: A review. Computers in Human Behavior, 24(3), 766-785. Gorn, G. J., Goldberg, M. E., & Basu, K. (1993). Mood, awareness, and product evaluation. Journal of Consumer Psychology, 2, 237–256.

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, 348-362.

Haushofer, J., Schunk, D., & Fehr, E. (2013). Negative income shocks increase discount rates. Working Paper.

Hockey, G. Robert J., et al. (2000) "Effects of negative mood states on risk in everyday decision making." Cognition & Emotion 14.6: 823-855.

Keltner, D., Ellsworth, P. C., & Edwards, K. (1993). Beyond simple pessimism: Effect of sadness and anger on social perception. Journal of Personality and Social Psychology, 64, 740–752.

Kugler, Tamar, Terry Connolly, and Lisa D. Ordóñez. (2012). "Emotion, decision, and risk: Betting on gambles versus betting on people." Journal of Behavioral Decision Making 25.2: 123-134.

Lee, Chan Jean, and Eduardo B. Andrade. (2015). "Fear, excitement, and financial risk-taking." Cognition and Emotion 29.1 : 178-187.

(29)

Haushofer, J., Schunk, D., & Fehr, E. (2013). Negative income shocks increase discount rates. Working Paper.

Heilman, Renata M., et al. (2010) "Emotion regulation and decision making under risk and uncertainty." Emotion 10.2: 257.

Holt, C. A., & Laury, S. K. (2002). Risk aversion and incentive effects. American economic review, 92(5), 1644-1655.

Jérôme Hergueux, Nicolas Jacquemet (2012). Social preferences in the online

laboratory: A randomized experiment. Documents de travail du Centre d'Economie de la Sorbonne.

Kugler, T., Connolly, T., & Ordóñez, L. D. (2012). Emotion, decision, and risk: Betting on gambles versus betting on people. Journal of Behavioral Decision Making, 25(2), 123-134.

Laury, S. (2005). Pay one or pay all: Random selection of one choice for payment. Andrew Young School of Policy Studies Research Paper Series, (06-13).

Ordonez, Lisa, and Lehman Benson. (1997). "Decisions under time pressure: How time constraint affects risky decision making." Organizational Behavior and Human Decision Processes 71.2: 121-140.

Panno, Angelo, Marco Lauriola, and Bernd Figner. (2013). "Emotion regulation and risk taking: Predicting risky choice in deliberative decision making." Cognition & emotion 27.2: 326-334.

Prelec, Drazen.(1998) "The probability weighting function." Econometrica: 497-527. Raghunathan, R., & Pham, M. T. (1999). All negative moods are not equal: Motivational influences of anxiety and sadness on decision making. Organizational behavior and human decision processes, 79(1), 56-77.

Smith, Kip, and John Dickhaut. (2005). "Economics and emotion: Institutions matter." Games and Economic Behavior 52.2 :316-335.

Watson, D., Clark, L. A., & Tellegan, A. (1988). Development and validation of brief measures of positive and negative affect: The PANAS scales. Journal of Personality and Social Psychology, 54(6), 1063–1070.

(30)

Appendices

Appendix A – Other measurements of risk and regressions

Appendix A presents the results of the analysis done with the other two methods to measure risk proposed in section 5: First choice and Last choice. This to provide a full picture of the potential results of the experiment and to shed a light on the robustness of the exercise, providing information on where it might be improved. The regressions presented here d receive the same treatments as the ones reported in the main document, merely replacing the dependent variable from one measurement of risk to another.

I. First Choice Regressions

Table A1. Effect of the treatment in risk taking – First choice

(1) (2) (3)

Variables First choice First choice First choice

Positive 0.392 0.579 -0.0455 (0.612) (0.709) (0.555) Negative -1.063* -0.919 -0.636 (0.612) (0.717) (0.555) Cognitive Reappraisal 0.443 (0.388) Expressive Suppression -0.277 (0.440) Constant 5.154*** 3.832*** 5.064*** (0.414) (1.299) (0.199) Controls No Yes No

R-squared 0.078 0.115 0.019

(31)

Table A2. Effect of emotions on risk taking – First choice

(1) (2) (3)

Interested -0.595 -0.729 -0.178 (0.397) (0.451) (0.481) Distressed 0.205 0.417 -0.109 (0.364) (0.404) (0.465) Excited 0.342 0.450 0.415 (0.466) (0.526) (0.402) Upset -0.598* -0.583 -0.530 (0.343) (0.372) (0.407) Scared 0.459 0.507 0.155 (0.830) (0.941) (1.078) Hostile 0.334 0.306 0.661 (0.524) (0.563) (0.609) Enthusiastic 0.0579 -0.0406 -0.0368 (0.448) (0.521) (0.439) Irritable -0.107 -0.0781 -0.239 (0.419) (0.443) (0.347) Alert -0.367 -0.432 -0.0211 (0.374) (0.416) (0.392) Inspired 0.293 0.226 -0.239 (0.388) (0.418) (0.464) Nervous 0.182 -0.0344 -0.551 (0.611) (0.783) (0.475) Attentive 0.599 0.759* 0.513 (0.365) (0.405) (0.474) Constant 4.007*** 1.907 4.699** (1.360) (1.810) (2.026) Controls No Yes No

R-squared 0.206 0.263 0.135

(32)

Table A3. Differences in risk taking between rounds 2 and 3 – First choice

(1) (2) (3)

Round 3 0.545 0.355 0.399 (0.646) (0.766) (0.714) Interested -0.511 -0.495 (0.579) (0.625) Distressed 0.277 0.406 (0.380) (0.418) Excited 0.219 -0.0380 (0.682) (0.748) Upset -0.330 -0.538 (0.356) (0.360) Scared 1.088 2.519 (1.623) (1.683) Hostile 0.585 0.149 (0.635) (0.700) Enthusiastic 0.530 1.205 (0.817) (1.049) Irritable -0.327 0.0537 (0.395) (0.415) Alert -0.491 -0.917 (0.641) (0.702) Inspired 0.378 -0.414 (0.619) (0.746) Nervous 0.740 0.779 (0.939) (1.069) Attentive 0.545 1.188 (0.930) (1.058) Constant 4.091*** 0.787 -7.104* (0.457) (1.880) (3.769) Controls No No Yes Observations 44 44 44 R-squared 0.017 0.287 0.513

(33)

Table A4. Effect of Emotion Regulation Strategies on Risk taking by treatment – First Choice

II. Last choice regressions

Table A5. Effect of the treatment in risk taking – Last choice

(1) (2) (3) (4) (5) (6)

Variables First choice First choice First choice First choice First choice First choice

Cognitive Reappraisal 0.186 0.333 -0.845 -0.198 1.178** 1.556** (0.453) (0.568) (0.883) (1.009) (0.556) (0.629) Expressive Suppression -0.312 -0.601 0.417 0.387 -0.479 0.315 (0.565) (0.785) (0.690) (0.822) (0.595) (0.862) Constant 5.398*** 3.936* 4.590*** 0.299 4.562*** 4.223** (0.452) (1.873) (0.729) (3.559) (0.501) (1.999)

Observations 22 21 22 22 26 25

R-squared 0.019 0.091 0.063 0.260 0.187 0.343

(1) (2) (3)

Variables Last choice Last choice Last choice

Positive 0.357 0.0346 -0.773 (0.517) (0.582) (0.516) Negative -0.462 -0.344 -0.818 (0.517) (0.588) (0.516) Cognitive Reappraisal -0.258 (0.319) Expressive Suppression -0.199 (0.361) Constant 5.962*** 4.570*** 6.564*** (0.350) (1.066) (0.185) Controls No Yes No

R-squared 0.034 0.095 0.065

(34)

Table A6. Effect of emotions on risk taking – Last choice

(1) (2) (3)

Interested -0.0308 -0.225 0.275 (0.315) (0.341) (0.445) Distressed 0.310 0.416 0.303 (0.289) (0.305) (0.430) Excited -0.0551 0.132 0.243 (0.369) (0.397) (0.371) Upset -0.696** -0.662** -0.882** (0.272) (0.281) (0.377) Scared 0.988 0.805 -0.242 (0.657) (0.711) (0.997) Hostile -0.0122 -0.0773 1.102* (0.415) (0.425) (0.563) Enthusiastic 0.532 0.418 -0.326 (0.355) (0.394) (0.406) Irritable 0.114 0.164 0.137 (0.332) (0.335) (0.321) Alert -0.127 -0.270 -0.209 (0.296) (0.314) (0.363) Inspired -0.543* -0.500 -0.434 (0.307) (0.315) (0.429) Nervous 0.398 0.593 0.211 (0.484) (0.592) (0.439) Attentive 0.314 0.443 0.503 (0.289) (0.306) (0.438) Constant 4.130*** 2.086 4.788** (1.077) (1.367) (1.874) Controls No Yes No

R-squared 0.269 0.361 0.184

(35)

Table A7. Differences in risk taking between rounds 2 and 3 – Last choice

(1) (2) (3)

Round 3 0.136 0.408 0.427 (0.665) (0.757) (0.504) Interested 0.216 0.0422 (0.572) (0.441) Distressed 0.429 0.678** (0.375) (0.295) Excited 0.196 0.229 (0.673) (0.528) Upset -0.471 -0.345 (0.351) (0.254) Scared 0.923 0.808 (1.604) (1.188) Hostile 0.526 0.736 (0.627) (0.494) Enthusiastic 0.385 0.343 (0.807) (0.741) Irritable -0.413 -0.0406 (0.390) (0.293) Alert 0.406 -0.183 (0.633) (0.496) Inspired -0.318 -0.488 (0.612) (0.527) Nervous 1.185 0.993 (0.928) (0.754) Attentive -0.551 0.966 (0.919) (0.747) Constant 5.500*** 1.955 -2.095 (0.471) (1.857) (2.660) Controls No No Yes Observations 44 44 44 R-squared 0.001 0.334 0.768

(36)

Table A8. Effect of Emotion Regulation Strategies on Risk taking by treatment – Last Choice

(1) (2) (3) (4) (5) (6)

Variables Last choice Last choice Last choice Last choice Last choice Last choice

Cognitive Reappraisal 0.370 0.477 -1.708** -1.775* 0.174 0.386 (0.385) (0.403) (0.698) (0.856) (0.476) (0.502) Expressive Suppression -0.821 -1.332** -0.167 0.186 0.141 1.330* (0.481) (0.557) (0.545) (0.697) (0.509) (0.688) Constant 5.953*** 2.733* 6.509*** 5.748* 5.955*** 6.261*** (0.384) (1.328) (0.577) (3.020) (0.429) (1.597)

Observations 22 21 22 22 26 25

R-squared 0.140 0.415 0.242 0.310 0.009 0.271

(37)

Appendix B – Consistent subject analysis

In section 5 of the main document, it is stated that we ignore the inconsistencies of the subjects in order to have a big enough subject pool so that valid statistical inference can be done. In this appendix we present the same regression exercises done in the document, but taking into account only the subjects that were consistent during all 3 rounds of the experiment – reducing the subjects from 70 to 29, as shown before.

Table B1. Effect of the treatments on emotions in round 2

Table B2. Effect of the treatments on emotion change between rounds 1 and 2

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) Variables Interested Distressed Excited Upset Scared Hostile Enthusiastic Irritable Alert Inspired Nervous Attentive

Positive 0.0958 0.177 0.243 -0.00553 0.113 0.265 0.0240 -0.236 -0.516 -0.103 0.182 -0.271 (0.545) (0.406) (0.634) (0.338) (0.274) (0.321) (0.570) (0.399) (0.614) (0.665) (0.263) (0.633) Negative 0.0399 0.317 0.581 1.402*** 0.0401 0.514 -0.175 0.638 0.227 0.227 0.416 -0.264 (0.603) (0.449) (0.701) (0.374) (0.303) (0.355) (0.630) (0.441) (0.679) (0.735) (0.291) (0.700) Cognitive Reappraisal -0.0894 0.0487 0.218 0.158 0.245 0.240 0.0301 0.121 -0.0742 0.175 0.340* -0.334 (0.355) (0.264) (0.413) (0.220) (0.179) (0.209) (0.371) (0.260) (0.400) (0.433) (0.171) (0.412) Expressive Suppression 0.125 -0.157 -0.223 -0.200 -0.335* -0.474** 0.249 -0.302 0.433 0.215 -0.394** 0.0533 (0.370) (0.275) (0.430) (0.229) (0.186) (0.217) (0.386) (0.270) (0.416) (0.451) (0.178) (0.429) Constant 1.555 0.828 1.234 -0.300 1.042* 1.063 1.507 0.413 1.904 1.287 0.902 3.326** (1.199) (0.892) (1.393) (0.742) (0.602) (0.705) (1.252) (0.876) (1.349) (1.461) (0.578) (1.391) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations 29 29 29 29 29 29 29 29 29 29 29 29

R-squared 0.146 0.110 0.129 0.598 0.297 0.254 0.092 0.321 0.225 0.120 0.308 0.120 Standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.1

Positive 0.542 -0.148 0.351 -0.622* 0.130 0.196 0.593 -0.736 0.378 0.553 -0.180 0.320 (0.427) (0.464) (0.526) (0.317) (0.174) (0.197) (0.482) (0.511) (0.388) (0.354) (0.228) (0.399) Negative 0.121 0.534 0.196 0.331 0.308 0.167 -0.209 0.329 0.314 0.104 -0.239 -0.337 (0.472) (0.514) (0.582) (0.350) (0.193) (0.218) (0.534) (0.565) (0.429) (0.392) (0.252) (0.442) Cognitive Reappraisal -0.0948 -0.136 0.137 -0.184 0.0263 0.0665 0.0214 -0.0724 0.210 0.371 0.0369 0.240 (0.278) (0.302) (0.342) (0.206) (0.114) (0.128) (0.314) (0.333) (0.253) (0.231) (0.149) (0.260) Expressive Suppression -0.0739 -0.0300 -0.340 0.353 -0.170 0.00385 0.100 0.0418 -0.241 -0.0519 -0.0407 -0.365 (0.289) (0.315) (0.356) (0.215) (0.118) (0.134) (0.327) (0.346) (0.263) (0.240) (0.155) (0.271) Constant -2.818*** -2.094* -2.875** -0.746 0.390 -0.428 -0.576 -0.630 -1.941** -0.569 0.134 -0.584 (0.938) (1.020) (1.156) (0.696) (0.383) (0.433) (1.061) (1.123) (0.853) (0.778) (0.502) (0.878) Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations 29 29 29 29 29 29 29 29 29 29 29 29

R-squared 0.360 0.242 0.331 0.540 0.377 0.175 0.181 0.345 0.272 0.266 0.272 0.301 Standard errors in parentheses

(38)

Table B3. Effects of the treatments on emotions – Fixed Effects

Table B4. Effect of the treatment in risk taking – Consistent

Positive 0.250 -0.375 0.250 -0.375 0 0 0.250 -0.625* 0.250 0.250 -0.125 0.250 (0.330) (0.332) (0.391) (0.244) (0.134) (0.136) (0.317) (0.361) (0.275) (0.250) (0.163) (0.278) Negative -0.333 0.111 -0.222 0.667*** 0.111 0 -0.556* 0.444 0.111 0 -0.111 -0.444 (0.311) (0.313) (0.369) (0.230) (0.127) (0.128) (0.299) (0.340) (0.260) (0.236) (0.154) (0.262) Constant 3.483*** 1.707*** 2.759*** 1.431*** 1.190*** 1.276*** 2.931*** 1.621*** 2.862*** 2.414*** 1.379*** 3.328*** (0.109) (0.110) (0.129) (0.0808) (0.0444) (0.0450) (0.105) (0.119) (0.0911) (0.0827) (0.0540) (0.0920) Observations 58 58 58 58 58 58 58 58 58 58 58 58 R-squared 0.060 0.049 0.028 0.285 0.028 0.000 0.131 0.149 0.036 0.036 0.039 0.120 Number of id 29 29 29 29 29 29 29 29 29 29 29 29

(1) (2) (3)

Variables Risk taking Risk taking Risk taking

Positive 0.958 1.150 -0.125 (0.884) (0.990) (0.723) Negative -0.778 -0.162 -1 (0.854) (1.096) (0.682) Cognitive Reappraisal 0.387 (0.645) Expressive Suppression -0.741 (0.671) Constant 5.667*** 3.071 6.103*** (0.559) (2.176) (0.239) Controls No Yes No

R-squared 0.116 0.271 0.075

(39)

Table B5. Effect of emotions on risk taking – Consistent

(1) (2) (3)

Variables Risk taking Risk taking Risk taking

Interested -0.853 -1.215* 0.189 (0.632) (0.657) (0.591) Distressed 0.930 1.050 1.193* (0.616) (0.597) (0.593) Excited -0.527 -0.883 -0.372 (0.613) (0.664) (0.411) Upset 0.176 0.103 -0.153 (0.549) (0.579) (0.880) Scared 0.229 -1.011 -1.634 (0.902) (1.063) (1.162) Hostile -0.376 -0.942 -0.682 (0.905) (1.090) (1.147) Enthusiastic 1.792* 2.224* 0.326 (0.920) (1.008) (0.586) Irritable -1.085* -1.326* -0.531 (0.579) (0.630) (0.579) Alert -0.0805 0.271 0.571 (0.517) (0.540) (0.534) Inspired -0.897 -0.805 0.463 (0.590) (0.606) (0.633) Nervous 1.618 2.820 0.986 (1.145) (1.762) (0.982) Attentive 0.478 0.297 0.983 (0.394) (0.561) (0.608) Constant 3.957** 1.582 -0.0717 (1.684) (2.429) (3.074) Controls No Yes No

R-squared 0.544 0.749 0.565

(40)

Table B6. Effect of emotions on response time – Consistent subjects

(1) (2) (3) (4)

Variables Response time Response time Response time Response time

Interested 6.437 2.889*** 1.471*** 0.959*** (13.50) (0.881) (0.425) (0.232) Distressed -3.045 -1.375* -0.313 -0.357* (10.80) (0.704) (0.332) (0.188) Excited -19.20 -3.526*** -1.542*** -0.921*** (14.13) (0.933) (0.458) (0.268) Upset -9.821 -0.417 0.642 0.849*** (13.11) (0.851) (0.402) (0.241) Scared 15.23 2.107 1.210 1.315*** (23.87) (1.534) (0.736) (0.407) Hostile -21.16 -1.228 -0.735 0.331 (17.41) (1.129) (0.551) (0.342) Enthusiastic 6.257 0.288 -0.115 -0.0993 (16.44) (1.086) (0.520) (0.305) Irritable 20.49* 2.276*** 0.440 0.0449 (11.54) (0.747) (0.363) (0.209) Alert -1.473 -1.414** -1.150*** -0.322* (10.89) (0.712) (0.342) (0.190) Inspired 10.72 1.359* 0.308 -0.217 (12.06) (0.782) (0.376) (0.207) Nervous -1.466 0.843 0.811 -0.588 (20.36) (1.304) (0.623) (0.389) Attentive -0.726 0.323 -0.0658 0.444** (10.12) (0.654) (0.310) (0.175) Cognitive Reappraisal 18.55 1.244 -0.282 -0.717*** (13.08) (0.860) (0.411) (0.239) Expressive Suppression -11.63 2.530*** 1.951*** 0.530** (13.48) (0.883) (0.424) (0.244) Constant 26.54 10.35*** 12.90*** 7.139*** (37.90) (2.419) (1.154) (0.662) Observations 870 785 767 594 R-squared 0.017 0.083 0.107 0.097

(41)

Table B7. Effect of response time on decision taken

(1) (2) (3) (4)

Variables Choice Choice Choice Choice

Response time -0.000681 0.00592 -0.0271 -0.0639 (0.000933) (0.00671) (0.0206) (0.0506) Interested 0.227 0.190 0.218 -0.147 (0.211) (0.232) (0.238) (0.282) Distressed -1.040*** -1.069*** -1.156*** -1.538*** (0.185) (0.208) (0.208) (0.280) Excited 0.549** 0.621** 0.729*** 1.062*** (0.222) (0.250) (0.264) (0.372) Upset -0.0963 -0.107 -0.0328 0.332 (0.203) (0.223) (0.224) (0.307) Scared -0.344 -0.357 -0.336 -0.310 (0.376) (0.401) (0.410) (0.493) Hostile -0.131 -0.100 -0.0491 0.263 (0.272) (0.290) (0.306) (0.429) Enthusiastic -0.685*** -0.746*** -0.874*** -0.837** (0.258) (0.286) (0.295) (0.398) Irritable 1.457*** 1.467*** 1.545*** 1.887*** (0.204) (0.222) (0.231) (0.308) Alert 0.0906 0.0531 -0.0344 -0.332 (0.169) (0.189) (0.193) (0.243) Inspired 0.300 0.264 0.294 0.189 (0.188) (0.208) (0.212) (0.255) Nervous -1.135*** -1.309*** -1.353*** -2.159*** (0.330) (0.364) (0.369) (0.531) Attentive -0.111 -0.0242 -0.00916 0.146 (0.159) (0.172) (0.174) (0.216) Cognitive Reappraisal 0.120 0.109 0.0903 0.124 (0.202) (0.227) (0.233) (0.307) Expressive Suppression 0.641*** 0.532** 0.639*** 0.911*** (0.213) (0.237) (0.247) (0.324) Constant -6.628*** -6.743*** -6.301*** -5.516*** (0.723) (0.776) (0.817) (0.976) Observations 870 785 767 594

(42)

Table B8. Effect of Emotion Regulation Strategies on Risk taking by treatment – Consistent subjects

(1) (2) (3) (4) (5) (6)

Variables Risk taking Risk taking Risk taking Risk taking Risk taking Risk taking Cognitive Reappraisal 0.316 -0.498 -2.071 -4.895 0.690 1.650 (0.776) (1.377) (2.339) (4.959) (0.871) (1.007) Expressive Suppression-1.137 0.497 -1.214 -0.765 0.552 1.915 (0.751) (2.360) (1.169) (1.990) (1.116) (1.770) Constant 6.120*** -3.086 7** 0.975 5.655*** 11.30** (0.725) (11.30) (2.188) (7.048) (0.895) (4.359)

Observations 8 8 9 9 12 12

R-squared 0.315 0.786 0.262 0.679 0.099 0.550

(43)

Appendix C – Experiment material

This section shows the screenshots of what one session during the experiment looked like, each image being one page of the browser. Although not in the images, it is important to remember that the different treatments the initial income value was different and that the lotteries appeared in random order in each subset of 10, as described in section 4.

(44)

(45)

(46)

(47)

(48)

(49)

(50)

(51)

(52)

(53)

(54)

(55)

(56)