T
HE
E
FFECTS OF
I
NCIDENTAL
E
MOTIONS ON
I
NTERTEMPORAL
D
ECISIONS
:
A STUDY ON ANGER
,
FEAR AND MEDIATORS
Master Thesis in MSc Economics
Behavioural Economics & Game Theory
University of Amsterdam
Louis Engesser – 11400250
Supervisor: Jan B. Engelmann
Second Reader: Frans van Winden
August 2017
A
BSTRACTEmotions, even if they are completely irrelevant to a specific decision can carry a strong impact on perception, judgment, and behaviour and can lead to detrimental outcomes. This piece of research investigates the effects of incidental anger and fear on intertemporal decision making between two in time separated rewards. The results showed that fearful but not angry participants increased their preference for immediate and smaller rewards and were willing to forego larger, delayed ones. This increased level of impatience is found to be mediated with not only risk attitudes but also different types of impulsiveness which appeared to be influenced by these specific emotions as well. In line with previous research, a gender specific effect was observed as women were more strongly influenced by incidental fear than men. That might be explained by gender differences in the way situational appraisals are processed. This research uses a novel online experimental approach which generalises previous findings further and may help future research in testing new experimental environments. Ultimately, many helpful insights can be gained from studying intertemporal choice and incidental emotions which might help to improve decision making across a diverse range of fields.
S
TATEMENT OF
O
RIGINALITY
This document is written by Student Louis Engesser 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. Louis Engesser 14th of August 2017
T
ABLE OF
C
ONTENT
1 Introduction ... - 1 - 2 Literature Review ... - 3 - 2.1 Emotion and Decision Making ... - 3 - 2.2 Models of Emotional Bias from Incidental Emotions ... - 4 - 2.2.1 Valence-based Models of Emotional Bias ... - 4 - 2.2.2 Appraisal-Tendency Framework ... - 4 - 2.2.3 Other Beyond-Valence Models ... - 8 - 2.3 Intertemporal Decision Making ... - 8 - 2.3.1 Intertemporal Discounting ... - 9 - 2.4 Incidental Emotion and Intertemporal Decision making ... - 10 - 3 Methodology ... - 11 - 3.1 Hypothesis: ... - 11 - 3.1.1 Fear ... - 11 - 3.1.2 Anger ... - 12 - 3.1.3 Mediating Factors ... - 14 - 3.2 Experimental Design ... - 15 - 3.2.1 Procedure ... - 16 - 3.2.2 Sample and Setup ... - 17 - 3.2.3 Emotion Inducing Video Clips ... - 18 - 4 Results ... - 20 - 4.1 Manipulation check ... - 20 - 4.1.1 Intertemporal Discount Rates ... - 20 - 4.2 Analysis ... - 22 - 4.2.1 Additional Regression and Tests ... - 26 - 5 Discussion ... - 30 - 5.1 Limitations ... - 33 - 5.2 Conclusion ... - 34 - 6 Appendix ... - 35 - 6.1 Qualtrics Experimental Outline: ... - 35 - 6.2 Additional statistical tests ... - 55 - 7 Bibliography ... - 57 -
1 I
NTRODUCTION
“Patience is a virtue.” This old proverbial phrase may carry some truth as patience has been identified as one predictor of economic success, higher cognitive and academic competence and better physical and mental health (Shoda, Mischel et al. 1990, Moffitt, Arseneault et al. 2011). But what if patience is malleable and could be influenced?
Previous research suggests this to be the case and has established emotions as one central factor to influence the willingness to delay gratification. Countless studies have indicated that emotions can play a powerful role in tainting perception, altering judgment and changing behaviour. For example, the research by Hirshleifer and Shumway (2003) has implicated daily stock market returns with the amount of sunshine. Is it possible that something so mundane and rudimental as weather induced moods influences the supposedly highly rational and sophisticated stock traders in their decision to buy and sell shares? Patience can be assessed through intertemporal decision tasks which surround us in our everyday life. Surely, everyone has been confronted with an intertemporal dilemma of choosing two in time separated options. Children in elementary school might opt for another piece of candy even if they know they might regret this at their next dentist appointment. Adults are constantly weighing whether they should save more money for retirement instead of buying a new car or going on that exotic vacation they have always wanted to go. Also voting for a candidate with rather short-signed policies instead for someone that offers long-term solutions could be an example of an intertemporal decision. While one option delivers an instant benefit, the other symbolizes the prospect of a larger reward in the future. There are many determinants that factor into a decision in such a context: patience, self-control, impulsivity or simple preferences are just a few of the many variables that contribute to such a decision. Regarding matters of emotional influence, things become interesting if the emotion that affects the decision is normatively unrelated to that decision. These kinds of emotional carryover effects are insightful because they represent an irrational reliance on emotional heuristics that may often cause suboptimal results. Outcomes which the individual would normally never choose may seem more attractive than before under the influence of certain incidental emotions. As this suboptimal deviation of behaviour becomes potentially detrimental for intertemporal decisions which affect large parts of an individual’s life (such as to take up that job offer or live a healthier life), their study may help to not just understand these processes better but also to improve decision making in these
areas. This thesis specifically investigates the impact of incidental anger and fear which are two emotional states that are frequently experienced and extremely powerful affects. The intent of this research is to shed some light on two central pieces of human decision making. Incidental emotions and intertemporal decision making. Both topics are highly relevant to a wide variety of human interactions which is why understanding the interconnectedness of them may prove to be fruitful in improving some of the harmful carryover effects of incidental emotions. The research question of this study is formulated as follows: What are the effects of incidental emotions, such as fear and anger on the discount rate of intertemporal decision making? The experimental part of this research consisted of an online experiment with an emotion induction task which elicited either fear, anger or a neutral emotion. Later individual discounting rates of future rewards were approximated in order to determine participants’ willingness to delay gratification. To create a more complete model of incidental emotional influence a series of mediating factors such as risk preferences, impulsiveness and the ability to regulate experienced emotions were inquired.
The results show that the emotion induction task was successful and that individuals in the fear condition became more impatient as they tended to prefer more often immediate over delayed rewards compared to the neutral group. Participants in incidental anger condition also increased discount rates, but the difference between the fear and the neutral conditions was not fully significant. While testing the influence of a series of mediation factors, interesting findings of impulsiveness were reported with different sub-scores of impulsivity showing opposing effects. Risk preferences were as expected related to discount rates but the ability to regulate experienced emotions was not a significant mediator of incidental emotions and intertemporal decision making. An intriguing gender effect was observed as women were much more influenced by incidental fear than men, but this was not the case for incidental anger.
This study contributes to relevant research in numerous ways. First, the combination of studying incidental anger and fear in the context of intertemporal decisions creates the opportunity to face an existing gap in the literature to this author’s knowledge. Furthermore, introducing a novel experimental design in the form of an online experiment for this specific field of research may prove to give some insights into the reliability and feasibility of this method for future studies. And lastly, testing a comprehensive series of mediating factors might help to close some gaps in the way we understand the effect of incidental emotions and how they shape behaviour.
The remaining of this thesis is structured as follows: In section 2 a brief literature of relevant academic research is presented and critically examined. Section 3 describes the methodology of the experimental research and derives the different hypotheses. In the next chapter, the results are reported, hypotheses tested and an additional regression model presented. Chapter 6 discusses and interprets the findings and its limitations and derives some useful applications for the gained insights. In section 7, the appendix comprises of some additional information such as the instructions for the experiment and more elaborate statistical tests. The last section will list all references that are used throughout this thesis.
2 L
ITERATURE
R
EVIEW
2.1 Emotion and Decision Making
The study of the impact of emotion on decision making requires distinguishing between two forms of emotional influences: integral and incidental. Integral emotional responses are triggered by a situation which itself is related to a decision and therefore may contain helpful information for that particular decision process. On the contrary, incidental emotions are caused by an unrelated event and are carried over to an object of judgment or decision and thus, should have no bearing on these unrelated incidents (Pham 2007). The difference between these two types of emotion can be illustrated by an everyday example. Suppose your car gets scratched by another vehicle on your way to work which makes you feel angry. This feeling of anger is normatively relevant to a related decision of changing your car insurance or deciding to use public transport the next day, but it is incidental to any unrelated task, such as hiring a new employee or taking on a risky project at work. While the study of both these emotional influences can be enlightening and fruitful, the fundamental implications of incidental emotions have particularly caught the interest of researchers. A constantly growing field of research indicates that emotion can highly influence beliefs (Fehr-Duda, Epper et al. 2011), perception (Ariely and Loewenstein 2006; Lerner and Keltner 2001) and behaviour (DeSteno, Li et al. 2014; Lerner, Small et al. 2004). A number of researchers have made a case for showing the detrimental effects of emotional bias on sound reasoning and the list of such emotional biases is numerous and keeps on growing. Some, for this thesis most relevant studies, are presented and critically examined in this section.2.2 Models of Emotional Bias from Incidental Emotions
2.2.1 Valence-based Models of Emotional Bias
Early research on the influential effects of emotion on decision making was grounded on a valence-based1 approach. This approach simply divides emotions into pleasant and unpleasant ones and
predicts that emotions of the same valence are infusing similar carryover effects (Lerner, Li et al. 2015). Applying this rather simplistic model, various studies revealed the powerful influence of moods and emotions on the decision making process. In an early study by Schwarz and Clore (1983), it was shown that subjects’ life satisfaction ratings were contingent on weather induced moods. Years later Hirshleifer and Shumway (2003) found a positive correlation between the amount of sunshine and the daily stock market return in 26 different countries, a finding that is difficult to explain with the rationale of price setting models. Forgas (1998) suggested that mood can alter the fundamental attribution error by which other people’s actions are overly explained by internal characteristics and not situational ones. Lerner and Keltner (2000) as well as Elster (1998) questioned these models for relying solely on the valence of the emotion as the predictor of the influential effect. Raghunathan and Pham (1999) criticised the valence-based models as too crude and simplistic. These two scientists rejected the assumption that all positive (all negative) moods are essentially equivalent but instead proposed the idea that different emotions prime district behavioural goals. This model resembles some elements of the later proposed Appraisal-Tendency Framework by Lerner and Keltner (2000).
2.2.2 Appraisal-Tendency Framework
Among others, Lerner and Keltner (2000) rejected the valence-based approach as one that lacks the specificity of different emotions of the same valence. They found it logically incoherent to expect the same behaviour patterns for emotions such as fear, anger and sadness solely due to their shared dimension of valence. These two researchers proposed and tested an alternative model they name Appraisal-Tendency Framework (ATF) in which not only valence but a series of attributes of a specific emotion shape the prediction of behaviour. This model starts with the assumption that a stimulus evokes different cognitive appraisals which are associated and therefore trigger certain emotions. 1 The term valence is borrowed from the field of psychologists where it is used to denoted the fact that emotions can be placed on a pleasure-pain scale with a neutral zero point of emotional indifference (Elster (1998) p. 51.)With the additional assumption that emotions are associated with specific appraisals, Lerner and Keltner predicted these appraisals as the primary determinant of an emotional influence on judgment and behaviour. Those two assumptions gave rise to the notion of an appraisal tendency which means that “(…) each emotion activates a disposition to appraise future events in line with the central appraisal dimensions that triggered the emotions”2. The appraisal patterns that underlie each emotion are based on the work of Smith and Ellsworth (1985) who identified and tested six different cognitive dimensions of classifying emotions. These dimensions are pleasantness, anticipated effort, certainty, attentional activity, responsibility, and control. Smith and Ellsworth (1985) noted that each emotion is characterised by a unique pattern3 of cognitive appraisals and
can, therefore, be distinguished. Based on this model it is possible to create different scores for emotions such as anger and fear.
The ATF is consistent with early experiments by Keltner, Ellsworth et al. (1993) which indicated that sadness and anger, although having both negative valance, exert different influences on causal judgments. In their study participants who experienced incidental sadness displayed an increased tendency to attribute situational factors as responsible for subsequent events. In contrary, incidental anger led subjects to perceive other individuals as more responsible.
The emotion induction was accomplished by randomly assigning participants to one of two groups and giving each group a different hypothetical emotional scenario in which they should picture themselves. As both, sadness and anger, are rated low4 on the valence scale (Smith and Ellsworth 1985), this finding stands antagonistically to the classical valence-based and inspired Lerner and Keltner (2000) to create the ATF. The notion of specific emotions being associated with distinct 2 Citation from Lerner and Keltner (2001) p. 147 3 There are two exceptions: shame and guilt as well as anger and contempt show somewhat similar pattern of appraisal. 4 Here, low valance is defined as being very unpleasant Table 1: Summary of the appraisal dimensions for Anger and Fear Table 1: Appraisal Tendency Framework for anger and fear - adopted from Lerner and Keltner (2000) Certainty Pleasantness (Valence) Attentional Activity Anticipated
Effort Control Responsibility Appraisal Tendency
Fear Low Low Medium High Low Medium Perceive negative events as unpredictable & under situational control
Anger High Low Medium Medium High High
Perceive negative events as predictable, under human control and brought about by others
appraisals is supported by many previous works such as Ortony, Clore et al. (1990), Raghunathan and Pham (1999) or Weiner, Graham et al. (1982), the latter found that pity, anger, and guilt, despite all being negative emotions evoked different perceptual tendencies.
In a later study by Lerner and Tiedens (2006) the original framework was refined by adding an additional concept regarding the depth of thought of emotion effects. Based on research by Tiedens and Linton (2001) and Small and Lerner (2008), the researchers proposed that the appraisal tendencies not only influence the content but also the process of thought. For example, the appraisal tendency regarding certainty for the emotional states of anxiety (low certainty) and anger (high certainty) can mediate the way individuals engage in heuristic or systematic thinking.
Since its proposal, the ATF has been widely used as a compelling conceptual framework for the prediction of emotional influence on judgment and decision making, but has also attracted some criticism. In the article by Shiv (2007), the Stanford Professor praises the ATF as a fundamental shift in science as it paved the way for a richer and more nuanced research on emotion and decision making. However, he also finds room for improving the framework by suggesting to further break down the part of “Content and Depth of Thought” into cognitive and motivational mediators. Shiv (2007) demonstrates this on the emotion of sadness, which action tendencies convey the need to change the present circumstances (Lerner, Small et al. 2004). This could be considered to belong more strongly to the motivational rather than the cognitive route. As sadness also modifies the perception of ambiguous events and causes a sense of loss of control by attributing events to situational factors, this perceptual element could be ascribed more to a cognitive route. Sharpening the ATF in this way and determining all relevant factors could enable more pointed prediction. In another argument, Shiv (2007) suggests to improving the ATF by adding the component of arousal, which could have an impact on multiple levels. For example, by affecting the likelihood that a particular set of appraisal themes and dimensions will be activated, arousal could influence the strength of the affect-decision link. Because the level of arousal might be greater for some individuals than for others as a result of individual tenancies or due to different emotional histories, that difference in arousal could then explain the variation of emotion-influenced judgment and behaviour between individuals. Specific Emotion Content and Depth of Though Judgment or Decision Appraisal Tendencies Appraisal Dimensions Appraisal Themes Graph adopted from Han, Lerner et al. (2007) Graph 1: Main construct of the ATF
Further evidence on the distinct nature of different negative emotions is presented in a paper by Lerner, Small et al. (2004) which showed that different negative emotions can impact the endowment effect in distinct ways. The endowment effect is an important phenomenon in behavioural economics and can be described as the impulse to demand a higher selling price for an object that is currently in possession than the buying price for that same object. Lerner and her colleagues hypothesised, based on the appraisal-tendency framework that incidental disgust reduced the endowment effect while incidental sadness caused a “reverse endowment effect”, where buying prices exceed selling prices. They explain these results of reduced selling prices for disgust with for that emotion-specific “(…)appraisal theme of being too close to an indigestible object or idea.”5 which results in the implicit need of removing oneself from currents objects. Shiv (2007) acknowledges that this finding on one of the most robust phenomena in decision making literature has proven that the robustness of the endowment effect may not be as irrefutable as previously assumed but more importantly that it was the ATF which laid the ground for this discovery. The study by Ferrer, Maclay et al. (2017) focused on gender differences in experiencing anger and sadness in a risk taking task. Regarding the gender effect, it was shown that men had a greater tendency to engage in risky behaviour when angry, compared to women. Subsequently, as the effect size of anger differed within the two genders, the argument could be made to alter the Appraisal-Tendency Framework to account for this. Ferrer and colleagues proposed that men and women may differ in their appraisals of certain situations. Compared to women, men may experience more certainty and control when under the influence of anger. This claim is supported by evidence from Tiedens and Linton (2001) who documented that even in situations in which women felt similar levels of anger they reported to fell less in personal control than men do. Kugler, Connolly et al. (2012) results are consistent with Ferrer, Maclay et al. (2017) findings that fearful individuals are more risk averse than angry individuals. Interestingly, Kugler and colleagues found that the perception of risk is less distorted by the incidental emotion than the willingness to take risk. This is an important distinction which previous studies have not made. Furthermore, they reported that the level of influence of incidental emotions on risk-taking is depended on the type of uncertainty. This suggests that the ATF could be improved by accounting for the class of uncertainty.
The robustness of the ATF and its predictions have been demonstrated through countless replications and the use of various experimental setups and different methods of inducing incidental emotions. Imagining hypothetical emotional scenarios (Raghunathan and Pham 1999), watching
emotional video clips (Lerner, Li et al. 2013, Lerner, Small et al. 2004), reporting on current moods (Fehr-Duda, Epper et al. 2011), dispositional affects (Lerner and Keltner 2001, Zhao, Cheng et al. 2015) and autobiographic induction of emotional distress (Ferrer, Maclay et al. 2017) have largely confirmed the notion of distinct carryover effects of different emotions.
2.2.3 Other Beyond-Valence Models
Raghunathan and Pham (1999) use a somewhat different framework to show that sad individuals become more risk seeking while anxious emotions prime more risk averse tendencies to behave in a risky environment. They explain these biases as a result of the fact that different emotions foster a selective focus on available information. According to the researchers, this selective information focus primes a distinct goal which may differ between emotions, such as anxiety and sadness, which were used in their experiment. They argue that sadness primes an implicit goal or attaining a reward for compensation and anxiety galvanizes a reduction of uncertainty. In a later study Raghunathan, Pham et al. (2006) address some shortcoming of their original study by extending it to a wider range of emotions and replicating their original findings. Furthermore, the authors provide evidence for the basis of an affect-as-information6 process as driver for their observed emotional carryover effects.Forgas (1995) expresses concern about this model of the almost binary character where the decision process is either influenced heavily or not at all, depending on whether or not the emotions are perceived to be relevant at the time.
2.3 Intertemporal Decision Making
Countless everyday decisions involve the choice between receiving a reward now and a larger reward in the future. These choice pairs of smaller immediate and larger delayed rewards are known as intertemporal decisions (Strotz 1955). In economic theory, the rational utility maximizer follows the discounted utility model (Samuelson 1937) which describes utility as the weighted sum of all discounted values over a given time span. Numerous studies have rejected this model as an accurate description of the processes that describe actual human decision making and more realistic models have been proposed. 6 The affect-as-information model states that, as long as an experienced emotion is perceived to be relevant to the context of the decision at hand then that emotion serves as a heuristic (Raghunathan and Pham (1999) p. 60)
2.3.1 Intertemporal Discounting
There is wide consensus in the literature on the fact that most species discount the future, however, there is some disagreement on the exact way and the process on how this is done (Soman, Ainslie et al. 2005). A large body of research focuses on the way in which subjective values of different, in time separated rewards are created by determining individual discount functions in dependence of reward and delay. This is typically done by giving subjects a series of intertemporal decisions and then calculating the indifference point for each participant. While theoretical economists proved mathematically that only exponential discounting leads to consistent behaviour over time (Strotz 1955, Fishburn and Rubinstein 1982), experimental results point to alternative models. Ainslie and Haendel (1983) for example found hyperbolic discount functions to be more predictive of actual behaviour. Hyperbolic discounting models have a relatively high discount rate for short periods of time but relatively low rates for long time horizons (Laibson 1997). As Thaler (1981) illustrates, the choice between “one apple today” over “two apples tomorrow” is often found in peoples revealed preferences, but preferring “one apple in 365 days” over “2 apples in 266 days” is not commonly observed in individuals’ choices. In economic theory, this reversal of preferences symbolizes a time-inconsistent discounting model. However observed in individuals’ choices. In economic theory, this reversal of preferences symbolizes a time-inconsistent with theoretical models, hyperbolic discounting has been established as one of the best ways of describing actual behaviour. Mazur (1987) proposed a hyperbolic discount function which fits the observed data very well and has often been used in subsequent studies.V =
A
1 + kD
The most interest lies on the discount parameter k, which can be interpreted as a measure for impulsiveness or impatience with higher values indicating a greater preference for the present. Laibson (1997) proposed an improvement over this functions which he calls a quasi-hyperbolic model. This is an approximation of the original hyperbolic function in discrete time but fulfils some of the mathematical properties that the original function does not. V = Present Value A = Delayed Reward D = Delay (in Days) k = Free Discount Parameter
2.4 Incidental Emotion and Intertemporal Decision making
A number of researchers have started to test and understand the influence of carryover effects of emotions on decisions in an intertemporal context. Lerner, Li et al. (2013) discovered a phenomenon they titled myopic misery in which incidental sadness is causally related to impatience, resulting in financial costs in intertemporal decision making. In a series of three experiments, the group of researchers induce sadness, disgust, and a neutral emotion into participants and measured their discount factor for delayed rewards. Here the results indicated that while disgust did not impact discount factors, sadness indeed decreased the willingness to wait for a larger delayed reward by 13-34 %, compared to the control condition In their study on positive emotions and intertemporal discounting DeSteno, Li et al. (2014) found that only incidental gratitude but not happiness decreased the discount factor of participants and made them more patient. The authors interpret this finding as an argument against the general idea that all affective responses should be concealed through self-regulation. Moreover, by reporting opposing effects of different positive emotions this paper offered further support for generalizing the AFT to previously rather understudied positive emotional states. Left: The preference order for two rewards (red: smaller and sooner, blue: delayed and larger) with an exponential discounting function do not change over time. The delayed and larger reward (blue) is always preferred over sooner and smaller one (red). Middle: For Hyperbolic discounting the subjective value of the delayed reward (blue) is initially greater (t)) than for the sooner one (red) but this reverses as the arrival time of the smaller reward comes closer (t+). Right: Comparative illustration of all three discount functions. The subjective value in dependence on the delay. - Graph adopted from Kable (2013) Su bj ec tiv e Va lu e Time Hyperbolic ,) ,+ ,) ,+ Exponential: δ/U(x) Hyperbolic: )56/3(4) Quasi-Hyperbolic: βδ/U(x)Exponential Comparison Exponential, Hyperbolic and Quasi-Hyperbolic
Graph 2: Illustration of the different discount functions Time Delay Su bj ec tiv e Va lu e
3 M
ETHODOLOGY
This section derives the hypotheses of this experimental research. Firstly, the specific emotions that were chosen are presented, and then each emotion’s hypothesis is formulated and explained. Secondly, the mediating factors of emotional influence on ITC are outlined, and predictions on their impact on intertemporal discounting are derived. And lastly, the experimental setup is laid out and justified.
3.1 Hypothesis:
3.1.1 Fear
The emotional state of fear is dominated by uncertainty and a sense of helplessness. Making important, sometimes even life changing decisions are accompanied or sometimes even triggered by the distinct experience of fear (Lerner and Keltner 2001). Investigating the exact implications of this emotional state can help to gain fundamental insights into our mind and may reveal ways to reduce some of the detrimental effects of incidental fear. Building on the Appraisal-Tendency Framework the emotional state of fear can be summarized by a few characteristics and its appraisal tendency which help to predict the influence of perception, judgment, and behaviour. The most relevant qualities associated with fear are: high uncertainty, high unpleasantness, low personal control and a medium level of attributing responsibility to others7. This portfolio of qualities leads Lerner, Li et al. (2015) to expect that individuals experiencing fear are perceiving risks as higher as compared to being in a neutral state. On the basis of these appraisal tendencies, the notion of high uncertainty and a sense of lack of control, the following hypothesis for incidental fear is built: Hypothesis I. Participant’s intertemporal discount rate in the incidental fear condition is higher as compared to the neutral condition.
The hypothesis for FEAR is directed and states that fearful individuals become more impatient, compared to individuals in a neutral emotional state. The fear induce lack of control and certainty can be compensated to some part by attaining a certain and immediate reward. Fearful individuals
may be prepared to compromise the size of the reward for an increased certainty of receiving it. Thus, creating a more certain and controllable environment, the ground for this hypothesis can be supported not only by the ATF but also by other models, such as the one proposed by Raghunathan and Pham (1999), presented earlier, which states that fear primes an implicit goal of increasing certainty.
3.1.2 Anger
The reasons for studying anger are numerous. Firstly, anger is one of the most frequently experienced emotions in everyday life (Frost and Averill 1982, Fischhoff, Gonzalez et al. 2005). Secondly, apart from being regularly experienced, it also draws individuals’ attention like no other emotions. Hansen and Hansen (1988) found that angry faces are most easily spotted in a crowd, a phenomenon they call the anger superiority effect. In fact, the ability to distinguish between angry and non-angry faces seems to be so deeply ingrained in humans that even 10-week old infants show distinct responses to different emotional faces (Haviland and Lelwica 1987). Lastly, as previously presented studies on anger and its influence have demonstrated, anger is exceptionally powerful in colouring our perception and judgment. Lerner and Tiedens (2006) summarised the emotional state of anger as one that is “associated with a sense of certainty or confidence about what has happened and about what the cause of the event was”8. The combination of certainty, being in control and making others responsible led Lerner, Li et al. (2015) to predict that anger makes individuals perceiving risks as lower than usual and causes a notion of optimism. The appraisal dimensions for anger are: high certainty, very unpleasant, high personal control and high in assigning responsibility to others. The motivational component of anger prompts an implicit goal of re-establishing the situation prior to the event that caused the emotion by removing the problematic components (Lerner and Tiedens 2006). Roseman, Wiest et al. (1994) found that angry subjects tended to focus on aspects regarding the fairness of a situation and had thoughts of violence against others. This has not only been documented in hypothetical questionnaire studies but also on a biological level. Carver and Harmon-Jones (2009) reviewed the academic literature regarding anger and cerebral activity and concluded that „The body of research
as a whole fairly consistently links anger to left anterior activation, suggesting further that left
anterior activation reflects approach motivation. It also suggests thereby that anger is an approach-
related affect.“9. In studies on the effect of different emotions on the peripheral nervous system, anger has also been linked to increased blood flow toward the hands and elevated heart rates, which has been interpreted as a physiologic preparation of engaging in a fight (Ekman, Levenson et al. 1983). Building on the ATF and previous studies of anger, the non-directed hypothesis for incidental anger is formulated. Hypothesis II. Participants’ ITC discount factor in the incidental anger condition differs from the discount rates of participants in the neutral condition. As laid out earlier, the appraisal tendency for anger predicts that negative events are perceived as predictable, under control and brought about by others (Smith and Ellsworth 1985). Consequently, individuals who experience anger show a decreased perceived risk (Lerner and Keltner (2001, 2000)) which could cause increased optimism and a higher tendency to indulge in risky behaviour. Considering this argument in isolation, one would expect incidental anger to have a decreasing influence on participants’ discount factor if incidental anger would solely induce risk seeking behaviour, thus preferring delayed (riskier) rewards over immediate (safer) ones. However, impulsivity, another fundamental driver of behaviour in the context of intertemporal decisions that is highly influenced by anger must be accounted for. Deffenbacher, Lynch et al. (2003) indicated that anger and aggression are correlated with increased impulsive behaviour. Dahlen, Martin et al. (2005) found in their study on unsafe driving that impulsiveness was positively correlated with trait driving anger and increased the likelihood that participants would use their vehicle to express anger. Following this argument, it can be expected that the experience of anger may increase the tendency to act more impulsively. If impulsive behaviour becomes more relevant as the level of experienced anger increases, it makes sense to expect an elevation of intertemporal discount rates as the prospect of immediate rewards becomes more attractive. Nonetheless, it may also be conceivable that an increased impulsivity score (Barratt Impulsiveness Score, BIS-11) leads to a decrease of impatience as the impulse to acquire the highest reward (which is always the delayed reward) and disregarding the time factor becomes more salient. As the underlying mechanisms of driving behaviour are not clear and possibly tearing in opposite sides, the direction of the anger-bias is not ex-ante predictable.
3.1.3 Mediating Factors
Based on some of the arguments made in the previous two hypotheses descriptions, it is possible to predict some mediating factors that may drive the degree of emotional influence.
People differ in their ability to regulate emotions. This emotion self-regulation is a deliberate and often effort demanding task to control and change an initial emotional response. The ways and techniques in which emotional states are controlled can broadly be distinguished into two categories. Cognitive Reappraisal tendencies reflect the ability to reinterpret and reflect on an initial emotional state in different ways. Physical Suppression techniques describe the degree at which one can control and contain the physical expression of an experienced emotion. These different coping mechanisms can be assessed by the Emotional Response Questionnaire (ERQ, Gross and John 2003). Previous studies have found that individuals who score high for reappraisal tendencies are more successful in mediating stressful events by opting for a more optimistic view on a situation and therefore, experience less negative emotions than individuals who score low on that scale. As far as suppression is concerned, the opposite has been documented. Individuals who suppress their emotions more strongly are generally experiencing negative emotions more harshly (Gross and John 2003). These findings lay the ground for the hypotheses for mediating effects of ERQ-scores. Hypothesis IIIa. Individuals with high reappraisal scores are less affected by the emotional bias (in FEAR and ANGER condition) and show lower deviations of discount factors from the NEUTRAL condition as compared to low-scoring individuals.
Hypothesis IIIb. High suppression-scoring participants who are in the treatment conditions have higher discount rates for future rewards than low-scoring participants. Hypothesis IIIa. states that high reappraisal leads to a less strong emotional experience of the induced emotion and will, therefore, impact the discount rates less severely. Hypothesis IIIb predicts that individuals who suppress their emotions more resolutely will also experience that emotion more intensively. This will increase the carryover effects of these emotions to the discount task and increase their impatience as compared to individuals who do not suppress their emotions. As previously explained, delayed rewards are associated with a greater sense of uncertainty and risk. Since individuals differ in their preferences of risks, it follows logically to assume a correlation between risk attitudes and preferences for intertemporal choices. Thus, this study predicts that risk
aversive individuals will have a higher preference for immediate (certain) rewards as compared to risk seeking individuals who may prefer higher, delayed rewards (risky).
Hypothesis IV. Risk aversive subjects discount the future more strongly than risk seeking ones. Lastly, it is hypothesised that impulsivity is related to higher discount rates as the prospect of an immediate reward may be more valued by an impulsive individual compared to a higher but delayed reward. Impulsivity is defined as the propensity to act without forethought and the inability to account for existing circumstances (Mathias and Stanford 2003). This lack of forethought gives rise to the hypothesis that the level of interpersonal impulsivity may explain some differences in delaying gratification and discounting the future between high- and low- impulsivity individuals.
Hypothesis V. Subjects with high scores of impulsiveness (BIS-11 Score) have a greater discount factor than low-scoring individuals.
Based on Personality Theory and experimental research it has been found that highly impulsive individuals generally exhibit lower resting arousal and higher increase of arousal if challenged compared to low-impulsivity individuals (Mathias and Stanford 2003). This forms the basis of the last hypothesis, stating that there are differences in impulsivity scores and discount rates between treatments and control conditions as the level of arousal should be higher in the treatment conditions.
Hypothesis Vb. Discount values are affected by impulsivity scores and differ between treatments and control conditions.
3.2 Experimental Design
The experimental design had the form of a one-factor, 3-level design (Emotion: ANGER, FEAR, NEUTRAL). In order to gather data and test the earlier presented hypotheses, an online experiment was conducted. A total of 405 participants were recruited using the online platform Amazon Mechanical Turk (MTurk). 94 Participants were excluded from the analysis due to failure to answer control questions correctly (65) or because of too long response times (2 times above the mean), leaving 311 (188 Male) participants in the sample. The majority of subjects were American (USA: 259; India: 31; other: 21) with a mean age of 35.4 years (SD = 10.1). It took participants on average 14.8 minutes to complete the experiment and they received a fixed participation fee of 1.00$ as well as the chance of being selected for payout in the incentivized ICT task.
3.2.1 Procedure
The online experiment consisted of three parts. Before the experiment started, subjects had to read the instruction which gave a brief overview of the different parts and prompted to answer the experiment in one sitting and pay close attention to all instructions and tasks. Furthermore, participants were informed that incorrect answers to control questions or logically inconsistent answers would lead to a rejection of their response and meant that they would not be compensated for their participation. After giving consent to the experiment, part one, the emotion inducement task started. Participants were randomly divided into three groups, each watching a different short video clip, which had been previously tested for inducing the target emotion (see section 3.2.3). Before the clip played, subjects were reminded again to pay close attention, make sure they could hear the audio and were not distracted as the control questions that followed would require their full attention. The control questions were carefully developed through repeated pre-tests beforehand as it was crucial to accurately filter out participants who did not watch the emotion induction clips as their answers from subsequent parts would bear no meaningful information for the analysis later on. This first part ended with a manipulation check by letting participants rate their current mood on a nine-point Likert scale for valence (positive/negative) and calmness (excitement/calmness). Part two consisted of an intertemporal discounting task to approximate each participant’s individual discount rate. This section started with detailed instructions on the task and included two examples. The validity of this part was ensured by incentivizing participants with the prospect of earning a substantial bonus (up to 85$) as one decision pair of four participants would randomly be chosen for playout (chance 1:100). This enforced subjects to choose according to their true preferences as real money was potentially at stake. A total of 27 intertemporal choice pairs of different reward and time instances (ranging from 1 week to 6 months) were presented. The entire ITC part was adopted from a widely cited and reproduced paper by Kirby, Petry et al. (1999). Part three consisted of a set of brief questionnaires to gain some background knowledge about participants’ preferences, personality traits and qualities. Firstly, the Emotional Response Questionnaire (ERQ), adopted from Gross and John (2003) was presented to measure in ten short questions each individual’s ability to overcome emotional responses and stimuli. Six out of these ten questions targeted cognitive reappraisal tendencies and the remaining four aimed at physical suppression techniques. Subsequently, the Holt and Laury (2002) risk task was filled out to assess individual risk preferences through the hypothetical prospect of winning one out of 10 different
lotteries. Next, the Barratt Impulsiveness Scale (BIS-11, Patton and Stanford 1995) was presented to estimate impulsive tendencies in participants. The BIS-11 is a revised form of the original Barratt Impulsivity Scale from 1959 which evaluates the level of impulsivity through a 30-item questionnaire. This part was included because previous studies have shown that impulsivity is one factor that influences the shape of hyperbolic discount functions (Ainslie 1975) and specifically, the emotion of anger has been linked to impulsive behaviour. This part ended by asking participant’s age, gender, educational background and whether or not they were familiar with the film their video clip originated from.
3.2.2 Sample and Setup
The decision to use an online experiment in combination with recruiting participants through Amazon Mechanical Turk was the result of a deliberate consideration of the research question, previous studies and the financial and institutional restrictions. As the induction of a specific emotion was central to the outline of the research question, the experimental setup was thoroughly developed around the most feasible method of successfully inducing the desired emotions. Using emotionally loaded video clips has been used in numerous studies on similar topics and seemed appropriate given all constraints of this thesis (time, budget, and resources). Ultimately, using MTurk simplified the payout mechanism for the ITC-task substantially as it was possible to reward bonuses to individual participants at any time after the experiment. This is paramount as any experiment involving intertemporal decision making can only plausibly incentivize true behaviour if the mechanism for paying out potential future rewards is credibly ensured. Another advantage of the online experiment is that it allowed measuring individual differences in a between-subject design with multiple treatments and did not require as much planning as in the laboratory. Lastly, recruiting many participants in a short period of time with relatively low amounts of financial means is another strong point of this experimental environment.
Limitations such as a partial loss of control over additional influences in participants’ environment that arise from this kind of experimental setting are reduced by acquiring a large sample size in order to address random noise. Moreover, multiple control questions were included as a means of screening out subjects who were inattentive or who’s only goal was to finish the task in the minimal amount of time without reading any instructions. Sourcing respondents from MTurk offers a more socio-economical and ethnical diverse subject pool, compared to the typical university sample
(Casler, Bickel et al. 2013) increasing external validity of any results. And limitations such as self-selection bias, as MTurk participants choose themselves which survey or task they want to take, are also present in laboratory experiments.
3.2.3 Emotion Inducing Video Clips
In order to draw inferences about causal relationships between experienced emotions and intertemporal discount rates, this experiment was designed around the emotion induction procedure. As the emotion induction was the most critical part of the experiment, the selection of used video clips was done carefully and tested in multiple small test trials beforehand. Based on the appraisal-tendency framework, one clip was used to induce anger, one for fear and one neutral clip that served as a baseline for later comparisons. Schaefer, Nils et al. (2010) conducted a study on the effectiveness of emotion-eliciting films in which a large sample of different clips was tested on their ability to induce specific emotions. The two films with the highest scores for anger and fear were adopted for this study. The anger-inducing clip was a 2 min long scene from Stephen Spielberg’s “Schindler’s List”, a film about the Holocaust in the 1940’s. The clip shows the commander of a concentration camp randomly shooting Jewish prisoners from his balcony. The cruel injustice of this commander’s actions triggered a set of negative emotions with anger being the most salient one in Schaefer and colleagues’ study. The clip for the fear condition was taken from the film “The Blair Witch Project”, in which a group of people explores an old, abandoned house at night in the search for a young boy. The entire clip is filmed by the protagonists using armature video recorders which increases the level of uncertainty that recipients feel when watching that scene. While this 3-minute clip was also rated for eliciting highly negative emotions, the most potent emotion that participants reported in Schaefer, Nils et al. (2010) was fear. The clip for the neutral condition was taken from a National Geographic documentary about the Great Barrier Reef and has previously been used as a control treatment in similar studies by Lerner, Li et al. (2013), Lerner, Small et al. (2004) and Ferrer, Maclay et al. (2017). This 4-minute clip depicts a colourful underwater scenery with a wide variety of sea animals.
As some of these clips are relatively old and the experimental environment of using MTurk participants is different from Schaefer and colleague’s study it seemed appropriate to test the clips effectiveness in triggering the desired emotions in a series of small trials. These trials also turned into a good opportunity for creating and testing adequate control questions to assess subjects’ cooperation and attention to instructions. After selecting five potential clips, the elementary parts of Schaefer, Nils et al. (2010) experimental setup were replicated in a between-subject design on
Qualtrics.com. After reading the instructions participants were randomly assigned to one of five treatments and watched one emotion-inducing clip. Next, their attention and cooperation were checked by three control questions and the Differential Emotional Scale (DES) was presented. The DES, adopted from Izard et al. (1974) measures through a 30-item self-report scale each discrete emotional state on a scale from 1 to 7. The descriptive results of these trials are presented in graph 3 for the three selected video clips. While the emotion induction procedure, offered the opportunity to draw causal inferences, it also comes with some limitations worth mentioning. One overt drawback of this procedure is the lasting effect, which states that early responses following the emotion-inducing part are likely to be affected more strongly by the target emotion than later ones (Fehr-Duda, Epper et al. 2011). This creates a trade-off between adding additional questions to more precisely measure preferences and traits and limiting the extent of the experiment to the most essential ones as each additional part will decrease the power of the treatment. This aspect led to the decision to place the ITC questionnaire directly after emotion induction part as the discount factor was the most relevant variable for the hypotheses. A previous test trial with a small sample (N=65) indicated that the average respondent needed 3.3 minutes to complete the entire ITC part, consisting of 27 questions. This was judged to be within the window of sufficient emotional influence by the previous part so that the original questionnaire by Kirby, Petry et al. 1999 was adopted in its entire form. 1 2 3 4 5 6 7
Anger Fear Happiness Calmness
Me an ra tin gs o f e m ot io na l a ro us al ( S ca le 1
-7) ANGER FEAR NEUTRAL
Graph 3: Mean ratings of emotions for each video clip. Error bars indicate standard errors from the mean.
4 R
ESULTS
4.1 Manipulation check
The emotion-inducing procedure using the video clips was effective in all conditions. Reported levels of experienced unpleasantness were significantly higher for FEAR (mean 6.57; SD 1.51) compared to the control condition (mean 2.92, SD 1.51), t(202)=–17.163; p<.001. Furthermore, participants in the FEAR group reported significantly lower scores for calmness (mean 3.99; SD 1.74) than those in the control group (mean 6.48; SD 224), t(192.036)=8.860, p<.00110. Likewise, for ANGER the ratings of valence (mean 8.03, SD 1.71) and calmness (mean 3.67, SD 2.11) differed significantly from the control condition, t(208)=-22.765, p<.001 and t(208)=9.348, p<.001. This is an indication that the emotion induction worked as participants whowatched either the FEAR or the ANGER video clip reported significantly different scores for both categories.
4.1.1 Intertemporal Discount Rates
An estimate of participants’ discount factors (k-value) for future rewards was calculated by approximating the point of indifference between choice pairs of different rewards/delays. This was done by calculating the geometric mean11 over choice pairs in which participants changed their 10 As the Levene’s Test (p<.05) indicated a violation of the equal variance assumption the adjusted t-test was used. 11 The geometric mean was used in order to avoid underweighting of the smaller of the two discount rates. Valence Mean SD Calmness Mean SD N ANGER 8.028 1.713 3.673 2.105 107 FEAR 6.574 1.506 3.990 1.741 101 NEUTRAL 2.961 1.501 6.476 2.240 103 Table 2: Mean and standard deviations of valence and calmness scores separated into treatment conditions. The scores were rated on a scale from 1-9, with 9 being extremely unpleasant/extremely calm. Graph 4: Mean levels of valence and calmness ratings(from 1-9) for all condition. Error bars indicate standard errors of the mean. Graph 4: Summary of Valence and Calmness ratings
preferences of receiving the delayed over the immediate reward. For example, question 19 offered the choice between “33$ today” and “80$ in 14 days” (a discount factor of .10 would make a subject indifferent between the two options). Question 5 asked participants to choose between “41$ today” and “75$ in 20 days” (participants with a discount rate of .041 would be indifferent here). An individual who chose the delayed option in question 19 and the immediate one in question 4 would be assigned a discount rate of .064 (geometric mean over .1 and .041). Since not all responses were perfectly consistent with any single k-value, the geometric mean for those individuals who indicated more than one indifference point was taken over the points that yielded the highest consistency. The ITC questions did not only differ in the distance between, but also in the magnitude of rewards. There were three categories of rewards: small (25$-35$), medium (50$-60$) and large (75$-85$), which gave category specific discount factors. The final discount rate was calculated as the mean of these domain specific geometric means of the most consistent choice indications. The entire procedure and the calculation of time discount factors were adopted from Kirby, Petry et al. (1999) to ensure consistent and objective estimations. Table 3 summarizes the mean discount rates and standard deviations for the three categories of rewards and grouped into the three treatment conditions. Graph 5 illustrates these different discount rates (small, medium, large) for the three treatment conditions. K-values for the three reward domains show the same trend of higher mean discount rates in the two emotion induction condition as compared to the control group. REWARD CATEGORY MEAN K-VALUE SD N FEAR Small 0.046 0.056 101 Medium 0.040 0.054 Large 0.040 0.054 ANGER Small 0.044 0.050 107 Medium 0.036 0.053 Large 0.036 0.053 NEUTRAL Small 0.031 0.037 103 Medium 0.029 0.047 Large 0.029 0.047
Table 3: Mean and SD of discount rates for the three categories of rewards and separated into treatment conditions
Graph 5: Separated (geometric) mean discount rates for the three different reward classes (small, medium,
large) and the three different treatment conditions. Error bars indicate standard error of the mean.
Graph 5: Overview of the category-specific discount rates in the three treatments
but even more for the small k ANOVA.
4.2 Analysis
A series of planned comparisons was constructed in order to systematically test all a-priori hypotheses. As the mean k-value (the mean over the geometric mean discount rates of small, medium and high reward categories) very closely resembled the different category specific discount factors, and in order to simplify the analysis and keep the hypothesis testing within the bounds of a master thesis, only this mean k-value was used in the analysis.
To test hypothesis I – that fear induced subjects discount the future more strongly than neutral
participants – a t-test indicated that discount rates of fearful participants (mean=.040, SD=.046) were indeed significantly higher than of those of individuals in the neutral condition (mean=.029, SD=.039), t(202)= 1.867, p=.032. Testing hypothesis II revealed that angry participants’ discount rates (mean=.0377, SD=.0483) did not significantly differ from neutral participants (mean=.02870, SD=.03944), t(208)=-1.470,p=.143. In order to test hypotheses IIIa and IIIb, the data was split into control (NEUTRAL) and treatment condition (FEAR and ANGER) to test for effects of ERQ-scores on discounting rates. The idea was to see if binned scores for suppression and/or reappraisal had an impact on the discount rates for individuals in the two treatment conditions. Hypothesis IIIa – stating that individuals with high reappraisal scores in the treatment conditions would have lower discount factors than individuals with high scores – was not supported, t(206)=.280,p=.610. The same was found for hypothesis IIIb, which means that individuals in the treatment conditions who had high scores for physical suppression of emotions did not have higher mean k-values than individuals with low suppression scores, t(206)=.244, p=.404.
However, if the same test was carried out with comparing separately FEAR with NEUTRAL and ANGER with NEUTRAL the results of this analysis partially changed, as displayed in graph 7. While NEUTRAL and ANGER display similar patterns of decreasing k-value for high suppression scores, as compared to low scores, this did not seem to be the case for FEAR. Testing for the significance for each individual treatment with comparison to the NEURAL conditions showed indeed that there was no significant difference in discount rates between ANGER and NEUTRAL for high (t(83)=.817,
Graph 6: Mean discount rates for the different conditions. Error Bars indicate 95%-confidence interval.
Graph 6: Mean discount rates in the different treatments
better done as continuous variable - also ERQ should be moderator.
p=.862) and low (t(90)=-1.014, p=.313) suppression bins. The comparison of FEAR and NEUTRAL revealed that there was equally no difference in k-values for individuals in the low suppression bin (t(75)=.404, p=.687) but for the high suppression bin the scores differ significantly between those two conditions, t(125)=1.977, p=.05012. This finding might be
surprising at first, but considering that the Appraisal Tendency Framework relies on a diverse set of appraisal patterns which differ enormously between fear and anger, it is reasonable to expect different mediating effects of two so different emotions.
The examination of hypothesis IV, regarding a relationship between risk aversion and higher discount factors, demanded to divide all risk scores into two bins of risk seeking and risk averse individuals. As predicted, risk averse individuals (mean k-value= .037, SD= .046) showed a (marginally) significantly higher discount factors than risk seeking ones (mean k-value=.028, SD= .039), t(227)=1.470, p= .071. Testing for a difference in discount factors of risk aversive subjects between the treatments and control condition revealed to be not significant, t(162)=.312, p=.756. However, there was a significantly higher discount rate for risk neutral participants, t(44.400)=-2.26013, p=.029 and for risk seeking individuals in the treatment conditions as compared to the control condition, t(63.266)=-2.534, p=.014. As illustrated in graph 8, the discount rates in the treatment conditions were similar across individuals with different risk preferences which were binned into three risk categories. The expectation that as individuals become more risk seeking they 12 The same test of comparison between individual treatment conditions and the control condition was done for the reappraisal bins. None of the comparisons was significant. These tests are reported in the appendix in section 7.2. 13 As the Levene’s Test (p=.035) indicated a violation of the equal variance assumption the adjusted t-test was used. Graph 7: Illustration of discount rates for binned Suppression scores Graph 7: Mean discount rates for high and low Suppression scores and separated conditions. Error bars show 95%-confidence intervals. Graph 8: Mean K-values for binned risk scores with separate lines for treatment (FEAR and ANGER) and control groups. Error bars indicate 95%-confidence intervals. Graph 8: Overview of discount rates for risk classes
are more willing to delay rewards (which are riskier), holds only for the control group. This trend was not observed in the treatment conditions. As discount rates did not differ between different risk groups, this finding may signal that the risk preferences become less important as incidental fear/anger become more predominant in driving preferences in the context of intertemporal decision making.
The scores for impulsiveness were also divided into two bins in order to analyse hypotheses Va and Vb. Hypothesis Va, stating that high impulsiveness is related to higher discount rates was not supported since the discount rates of high-impulsivity individuals were not significantly greater than those of low scoring individuals, t(309)=.650, p=.258.
In order to test hypothesis Vb all six sub-groups of the Barratt Impulsiveness Scale were also split into low and high groups to see if they differed in their discount rates and if this was related to the treatments, the individuals of these groups were in.
The First-Order BIS-11 score for Attention measures the level of impulsiveness related to the ability to concentrate and pay attention. However, the hypothesis was only partially supported by the data as only individuals with low BIS-Attention scores differed between treatment and control conditions, t(132.476)=-2.039, p=.04314 but this was not the case for high scoring individuals, t(158)=-.909, p=.365. Neither the treatment nor the control conditions differed individually between high and low attention bins.15 This finding was somewhat the opposite from the results of the First-Order BIS score for Cognitive Instability. While there was no significant difference between treatment and control groups for low scoring participants (t(109)=-.124, p=.902), individuals in the high scoring group who were in the treatment condition had a significantly higher discount 14 As the Levene’s Test (p=.023) indicated a violation of the equal variance assumption the adjusted t-test was used. The significance of the Levene’s Test can be explained by the fact that the treatment group which consists of FEAR and ANGER is about twice as large as the control group (NEUTRAL).
15 Testing for a k-value difference between high and low bins of attention impulsivity for in the treatment group:
t(206)=.458, p=.648. The same test for the control condition yielded: t(101)-.607, p=.545 Graph 9: Mean discount factors for binned Attention and Cog. Instability Impulsivity sub-score. Error bars indicate 95%- confidence intervals. Graph 9: Illustration of discount rates for impulsiveness sub-scores
value than individuals in the control group, t(167.271)=-2.271, p=.0034. This indicates that individuals who are impulsive regarding cognitive instability, for example having “racing thoughts” are more prone to be influenced by incidental anger/fear which increases their impatience for delayed rewards.
Regarding the Second-Order score for Motor-Impulsivity, the tendency to make up one’s mind and act without thinking it is found that participants with a high
motor impulsivity score differed significantly between conditions. Participants who had a high motor-impulsivity score and who were either in the FEAR or ANGER condition had significantly higher discount rates compared to participants in the NEUTRAL condition, t155.762)=-2.227, p=.027. At the same time, participants with low motor scores did not differ in their discounting of delayed rewards, t(141)=-.245, p=.803. This could indicate that having a low score in motor-impulsivity, thus not being very impulsive could have made participants more resistant to emotional influence from the video clips as their scores did not differ from those of the control condition.
Testing for differences in discount rates for impulsivity in Perseverance (the tendency to not focus on the future and often change jobs, residences, etc.) reveals that the treatment group differed only among low scoring individuals, t(94.869)=-3.191, p=.002 but not among high scoring individuals, t(212)=-.459, p=.646.
The Self-Control impulsiveness score shows that participants scoring low in self-control impulsivity differed in their discount rates between treatment and control groups. Statistically, the difference in discount rates of low-scoring participant between treatment and control group was significant at t(94.762)=-3.647, p=.000, but the difference between the conditions for high scoring individuals was not, t(212)=-.713, p=.477. This may indicate that the mood induction for participants who saw either the FEAR or the
Graph 10: Mean discount factors for binned Impulsiveness sub-scores. Error bars indicate 95%-confidence intervals.
ANGER clip did not affect their discount rate. On the other hand, the self-control score was predictive of discount rates for participants in the control group with significantly higher k-values for high-impulsive participants.
Analysing the last BIS sub-score of Cognitive Complexity, which measures the willingness to engage in challenging mental tasks, showed a similar picture as for
self-control. Participants who scored high in this impulsivity sub-score had comparable discount rates regardless of being in the treatment or the control group, t(169)=-.432, p=.666. However, participants in the control condition who scored low in this trait exhibited significant differences in discounting the future compared to those in the FEAR and ANGER group with low scores, t(136.494)=-3.435, p=.001. At the same time, the NEUTRAL individuals had significantly lower discount rates if they scored low in cognitive complexity than if they scored high, t(90,612)= -2.313, p=.023. This is in line with the prediction of hypothesis Va.
