Author
Camille Ras s4530020
November 21, 2016 Master thesis
Faculty of Management Radboud University Nijmegen
Supervisor
MASTER THESIS
INFLUENCE OF EMOTIONS
ON LOSS AVERSION
AN EMPIRICAL ANALYSIS ON THE
HOUSING MARKET
Dr. O. Dijk - Radboud University Nijmegen
Contents
1. Introduction...3
2. Literature review...5
2.1 Defining affect, mood and emotion...5
2.2 Emotion and moods surrounding moving motives...7
2.3 Emotions, moods and real estate...9
2.4 Loss aversion and real estate...12
2.5 Effects of emotions on loss aversion in real estate...14
2.6 Amateurs and experts...17
3. Methodology...19
3.1 Ways to induce affect...19
3.2 Checks for affect...19
3.3 Pilots...20
3.4 Final experiment with loss aversion task...21
4. Results & Analysis...25
4.1 Variables...25 4.2 Data...27 4.3 Results...32 4.3.1 Mood induction...32 4.3.2 Overall satisfaction...34 4.3.3 Corrected ratio...37
4.3.4 Corrected gain domain...39
4.3.5 Corrected loss domain...42
5. Robustness checks...51
6. Discussion...52
7. Conclusion...56
References...58
Appendix A : Survey...67
Appendix B: Tables & Figures...77
Appendix C: Robustness - regression tables...83
Appendix D: Additional readings...86
1. Introduction
For most people buying a house is the biggest and most important purchase financial transaction in their lives. Households are in the housing context inexperienced and financially unsophisticated (Hong, 2007). Therefore most people in a housing transaction are inexperienced and have little information about the market. The households are also not able to process all information about their ability to pay the mortgage if the economic situation does not materialize as they expected (Ackert, Church & Jayaraman, 2011). Therefore Northcraft & Neale (1987) argue that those inexperienced households are vulnerable to behavioral biases. This leads to anxiety and trepidation when households are acquiring a new property (Brooks & Schweitzer, 2011; Conner, 2010).
In addition, changing houses often coincides with major life event like marriage, divorce, changing jobs, starting a family or death. These event induce emotions or moods surrounding the decision to sell or buy a property. There is evidence that these emotions do influence the decision to buy or sell (Cryder, Lerner, Gross & Dahl, 2008; Lerner, Small & Loewenstein, 2004; Van Acker, Kerselaers, Pantophlet & IJzerman, 2015)
Households, unlike professional investors, mostly serve both sides of the market at the same time. They decide to sell and buy, more or the less, at the same time since they are mostly changing from one house to another. This thesis will focus mainly on the decision to sell.
The decision to sell, mostly called the willingness to sell or WTA, is known to be influenced by loss aversion, especially if you consider the endowment effect (Brown, 2005; Gächter, Johnson & Herrman, 2007; Kahneman, Knetsch & Thaler, 1990; Tversky & Kahneman, 1991). Even in a real estate context nominal loss aversion is known to be present (Bokhari & Geltner, 2011; Einiö, Kaustia & Puttonen, 2008; Engelhardt, 2003; Genesove & Mayer, 2001).
Since the time surrounding the decision to sell is likely to be especially emotionally charged, the amount of loss aversion might be influenced. However there is little known about the influence of emotions on loss aversion (Camerer, 2005; Andersson, Holm, Tyran & Wengström, 2014). This paper will thus try to close the gap between emotions and loss aversion on the housing market since both phenomena are present at the housing market. This master thesis will thus provide theoretical and empirical evidence on the main research question:
How do emotions influence nominal loss aversion in a housing market setting?
Households are mostly represented by professional real estate agents. It is commonly accepted that real estate agents are less emotional and can therefore help households in their decision to sell. Vereniging eigen huis (2016) even argues that a good real estate agent should not be guided by emotions. These real estate agents and other professional institutions have experience, sophistication and knowledge of the market and this might reduce emotional effects. However, there is empirical evidence that those expectations might not be true (Bokhari & Geltner, 2011; Coval & Shumway, 2005; Haigh & List, 2005). There is some evidence that deciding for others may reduce loss aversion due to less influence of emotions (Andersson, Holm, Tyran & Wengström, 2014). However it is still unknown if real estate agents are not, or less, guided by emotions. Therefore this thesis will also focus on the difference between amateurs and professional real estate agents regarding their ability to turn off the effect of emotions on loss aversion.
The rest of the thesis will be organized as follows. In the second chapter the existing literature is described about the effects of emotions, effects of loss aversion and concludes with the effects of emotions on loss aversion at the housing market. The third chapter will consists of methodological and empirical implications of this research. In the fourth chapter the results will be shown and empirical tests are conducted. The thesis will end with a conclusion and a brief discussion about findings and used methodology.
2. Literature review
Just as loss aversion alters choices (e.g. through the so-called disposition effect), affective states shape decision making (Cryder, Lerner, Gross & Dahl, 2008; Feng & Seasholes, 2005; Lerner, Small & Loewenstein, 2004). Even if the affective state is not primarily germane to the decision being considered your decision making can be altered due to spurious affective circumstances surrounding the decision. Decision makers often take experienced affective states as relevant information to ongoing mental processes which eventually leads to colored decision preferences (Clore, Gasper & Garvin, 2001).
2.1 Defining affect, mood and emotion
Differences between emotion, mood and affect are not as clear cut as one might think. Therefore literature has little consensus about the definition of those terms. Therefore this paragraph will give the definition for emotion, mood and affect relevant for this thesis.
Affect can be seen as the term which is the least problematic and is used merely as a generic label to refer to both moods and emotions (Forgas, 1995a). Affect is thus a broader term which is not specific about the characteristics of the cognitive processes. This term can be defined as a “neurophysiological state consciously accessible as a simple primitive non-reflective feeling most evident in mood and emotion but always available to consciousness” (Russel & Feldman Barrett, 2009, p. 104). Affect is therefore a broad general label and academia merely use more precise definitions in their research to describe certain phenomena. Therefore they make distinctions between emotion and mood.
However academic literature gives little consensus about the distinction between emotions and moods. As Beedie, Terry & Lane (2005) mentions the majority of articles make 2 or 3 distinctions between emotions and moods. On the other hand, the literature aggregately makes 8 distinctions between moods and emotions. Those distinctions are based on intensity, duration, physiology, cause, awareness of cause, consequences, function and intentionality (Beedie, Terry & Lane, 2005). The only consensus those authors find is that academic literature argues that the duration of moods endures longer than the duration of emotions.
Ekkekakis (2012) argues that emotion is composed by different things namely, core affect, overt behavioral congruent with the emotion (facial expression like a smile), attention directed toward the eliciting stimulus, cognitive appraisal of the meaning and the possible implications of the stimulus, attribution of the genesis of the episode to the stimulus, the experience of the
particular emotion and neural and endocrine changes consistent with the emotion. For example a person who just sold their house far above the asking price would smile and feel happy. In addition they have a slight increase in heartrate and feel some arousal. The cause is clear and the intensity is probably high. Discovering the fact that your partner is cheating would make you feel angry and sad. You can imagine, and some maybe know, that the intensity is high, your heartrate raises and probably have a clear facial expression. These are typical examples of short-lived emotions. Ekkekakis (2012) also notes that emotions are elicited by something, are reaction to something and are generally about something. This contrasts with moods where these “somethings” are not always that clear.
Moods are, as earlier mentioned, more long term than emotions. Frijda (2009) argues that mood is “the appropriate designation for affective states that are about nothing specific or about everything-about the world in general” (p.258). This basically means that mood is temporally remote from the cause and therefore the cause of the mood is not always that easy to identify. Taken separately things as sunshine and amount of daylight can influence your overall mood and therefore your decision (Hirshleifer & Shumway, 2003). A clear example for real estate would be a person which sells his/her house during the winter (low duration of daylight) with barely any interest of potential buyers and not really a good relationship with the neighbors. Although not one of the points is clearly a cause, the person might suffer from a bad mood (e.g. feeling down/sad). Generally speaking moods have lower intensity, have a longer duration, do not always have a clear cause (combination of things) and the person is not always aware of the cause. Whereas emotions have a higher intensity, shorter duration and the person is aware of the clear cause (Forgas, 1995a). This is in line with the findings of Beedie, Terry & Lane (2005) which can be found in table 1.
Table 1
From: Distinctions between emotion and mood , p. 871, by Beedie, C., Terry, P., & Lane, A., 2005, Cognition & Emotion , 19(6)
2.2 Emotion and moods surrounding moving motives
If we take a look at table 2, we can see that the three major reasons to move are house, family or work related. There are three clear reasons for moving in house-related motives namely new/better house or apartment, better neighborhood and cheaper housing. I presume planning to move with this kind of reasons can induce happiness since one can look forward into buying another property which is for example bigger. This might be more a mood than an emotion since the person is probably not fully aware of his happiness, the cause is somewhat cloudy and the duration will be somewhat longer. On the other hand, family-related reasons can be either positive or negative. For example one could marry or divorce. Marriage should make a person happier and can be an emotion (e.g. just after proposing) or a mood (e.g. foresight of wedding, being with the person you love etc.). For a divorce the opposite might be true, emotions like fear, anger or sadness might be of influence on one’s decision. This can explain significant positive correlations between divorce rates and house prices found by Farnham, Schmidt & Sevak (2011) and Chowdhury (2013).
Another family-related reason is to establish your own household (e.g. have kids). Since the foresight of establishing an own household will give you an anticipated positive emotion namely happiness. Therefore I argue that this leads to a positive mood due to a
general feeling of happiness due to the long duration and multiple unclear causes like good job, happy relationship and the ability to move.
Furthermore, work-related motives are primarily based upon getting a new job, closer to work, looking for new job, lost job and retirement. The first two reasons are reasons which generate a happy emotion since both are favorable for the owner. A new job on itself will, on average, give a happy feeling and I presume this effect gets larger if a person was unemployed for a longer period. A new job will thus have direct influence on one’s decision whereas closer to work induces anticipated happy emotion at the moment of decision. The same goes for retirement. However looking for a new job is quite the opposite. I presume that a person at first will look at jobs near his residence and then after a long period of being unemployed move to another city. His current emotion will thus be sadness since he has to leave his current environment. In addition the anticipated emotion will be fear of not getting a job even if the person moved to a new area. As explained above we can argue multiple emotions and moods which can be elicited by the circumstances surrounding the decision to move.
Table 2: Reasons for moving
The fact that incidental emotions influence a person’s related decision (related to the cause of the emotion) might be logical. However, if the emotion is totally unrelated to the decision, will the emotion still influence your decision? According to the literature we can, without any doubt, say that this is still the case. According to Lerner, Li, Valdesolo & Kassam (2015) emotions carry over from one situation to another. Therefore emotions often influence decisions in a non-conscious and unwanted way. Although you would, from a normative perspective, expect that the unrelated emotion does not influence the decision. For example, judgement of unrelated topics and even objects have been proven to be influenced by affect through movies, sunny weather, stressful exams and other factors (Bodenhausen, 1993; Clore, Schwartz & Conway, 1994; Forgas 1995a; Forgas & Bower, 1988; Schwarz, 1990; Schwarz & Clore, 1996). Therefore even emotions and moods which are unrelated to real estate can influence the decision to sell a property or not. We can thus conclude that related and unrelated affective states, either emotions or moods, can influence the decision to sell a property.
However there are no papers that link emotions to real estate except a theoretical paper on housing bubbles and the meaning of homes (Christie, Smith & Munro, 2008), and therefore no direct empirical evidence is provided. As explained above, both happiness and sadness have potentially multiple sources of elicitation in the moving motives. Therefore this thesis will focus on these affective states. In addition, happiness and sadness have opposite valance and it is easier to imagine someone happy or sad for a longer duration than for example someone who is angry. However I suspect that these other emotions like fear, surprise, etc. will also influence decision making in real estate.
Since no papers link affective states to real estate this thesis will address the effects of those effective states on decision making and argue them towards the decisions to sell in real estate.
2.3 Emotions, moods and real estate
One highly researched topic in economics is the influence of emotions on risk perception. Of course acquiring or owning a property is a risky decision and therefore emotion influence one’s decision making. Emotions like fear and anger have different impact on people’s decision making due to the fact that they do differ in control and certainty. Lerner & Kelner (2000) asked participants induced with either fear or anger to estimate the number of annual deaths by accidents due to 12 events that lead to those deaths or estimating the likelihood of
that specific positive/negative event would occur in their lifetime. The empirical results showed that fearful people made pessimistic assessments while angry people made optimistic risk assessments. Fearful people have the feeling that they are not in control and are not certain and will therefore increase their risk assessment of the new situation. On the other hand, angry people do feel in control and certain and will therefore decrease their risk assessment. The same effect was found in a field experiment right after September 11th when fear increased perceived risk of terrorism while anger did the opposite. The effects of fear and anger also hold for other (non-terrorism) risks (Lerner, Gonzalez, Small & Fischhoff, 2003).
More of interest for this paper is the research which is focused on the influence of emotions on monetary value. Emotions influence the willingness to purchase (WTP) and the willingness to accept (WTA) in general. Lerner, Small & Loewenstein (2004) did an experiment and induced subjects with a neutral, sad or disgust emotion via movie clips. Then they conducted an experiment to see if the emotions influenced the subject’s behavior in the WTA-WTP gap assignment. They found that there was indeed some significant differences for disgust and sadness in comparison to the baseline neutral emotion. For disgust the selling condition dropped below the purchase condition which decreased the gap significantly. For the sadness emotion the purchase condition was increased and the selling condition was decreased. Therefore the WTA-WTP gap was increased.
The underlying foundation can be found in the behavioral elements of the emotions. The disgust emotion will give you the intuition to get rid of things or to be less willing to purchase something. Whereas the sadness will induce desire to change the situation because you do not like to feel sad. Therefore the sadness will increase the willingness to purchase and to increase willingness to sell (Lerner, Small & Loewenstein, 2004). These results could also apply to real estate since the underlying process of buying or selling a property is also dependent on the willingness to purchase and willingness to accept. It is not hard to imagine that if you feel disgusted about the property you just visited, your offer to acquire the property would drop in comparison to the neutral position. On the other hand, if you are the owner of the property and you feel disgusted about the place (e.g. your wife cheated on you in this house), you are willing to accept a lower offer to get rid of the property. On the other hand, if you are sad you may like to change your current situation. Therefore you are more willing to acquire or sell the property. Your (unconscious) mind will think, maybe a new home will make me happy again instead of sad.
The willingness to purchase a property could even be increased by unconscious aspects (i.e. like physical conditions) which relate closely to emotions. A house is, seen from an evolutionary perspective, a novel solution for shielding from the cold. Actual lower temperatures induce people to find a house more communal and increase their need to affiliate with other people or ‘homes’. More importantly, people feel more attracted to the house and this increases their willingness to purchase the house. The underlying process is based upon the evolutionary aspect to survive. When it is cold your body wants to warm itself and search for shelter for any thermoregulatory threats. Therefore you will affiliate more with other people since body temperature can be increased by staying close to each other (e.g. like penguins in a snow storm). Our attachments towards others extrapolates to our relationship with houses (and brand names etc.). In addition the house will also shield from the cold which increases our attachments directly. Therefore even physical conditions of a person can influence one’s cognitive systems and reasoning unconsciously and eventually make this person willing to pay more for a certain house (Van Acker, Kerselaers, Pantophlet & IJzerman, 2015).
Willingness to pay does not always equal actually paying a higher price for a product. To see if willingness to pay really translates to paying a higher price Cryder, Lerner, Gross & Dahl (2008) did a study. They induced participants with sadness or a neutral emotion and then checked if there was a significant difference in the price people payed for a bottle of water. Indeed sadness influenced participant’s willingness to purchase but also their real expenses on the bottle of water, even if they had to take the additional amount from their own pocket.
Emotions can also influence other behavioral effects like framing effects. Happiness makes people evaluate decisions more favorable and less willing to commit to the task at hand. Therefore happy people are, when the stakes are rather low, more prone to make intuitive, system 11, decisions while they are actually being framed. Sadness on the other hand will increase people’s willingness to commit to the task. Speaking in terms of Kahneman, these sad individuals are more willing to activate system 2 to debias the intuitive response of system 1. Stanton, Reeck, Huettel & LaBar (2014) tested whether happiness, sadness and neutral emotions induced participants difference in the processing and decision making with a framing effect. They induced their participants with one of the 3 emotions via a video clip
1 Kahneman makes a distinction in people’s processing by using the terms system one and system two. System one operates automatically and fast without any effort. System two can do effortful mental activities that are demanded from it. However system one is more prone to make intuitive estimations and those are therefore more biased or faulty. System two is often associated with concentration and experience of agency. For more
and then presented them a task. The task consisted of a series of questions. First the participants were given an initial endowment($75) and 2 seconds later they were given the question whether they would take guaranteed amount of money at a discount ($30) or a gamble were they would either win some amount(range of $25-$100) or lose the entire endowment. The guaranteed option is presented in 2 ways, either as a loss (lose $45) or as a gain (keep $30). As we can see the guaranteed option does not differ in actual amount and thus the only difference is the way of framing the question. Overall the loss framing of the certain part leads to a higher proportion of gambles.
In addition they find that happiness leads to a higher propensity to gamble and a higher magnitude of the framing effect. This means: if you are happy you will be more influenced by the framing effect in comparison to the sad and neutral group. However the induction of sadness did not influenced participant’s propensity to gamble nor the magnitude of the framing effect (Stanton et al., 2014). Hence, we can conclude that emotions also influence other behavioral aspects, like framing, rather than only the normative decision making process. Therefore the paper of Stanton et al. (2014) is closely related to this thesis which focuses on the influence of emotions on loss aversion rather than framing effects. Both are behavioral anomalies. However this paper is still highly relevant to real estate since framing effects can be used by sellers, buyers and real estate agents. When sellers face a potential loss, the real estate agent can frame this loss into a gain but even inflation will hide losses from people’s awareness. I will elaborate on this in the next paragraph.
2.4 Loss aversion and real estate
People tend to think of transactions in nominal terms rather than real terms. Therefore inflations can mimic (nominal) gains while actually real values are at a loss. Thus real losses can be hidden from people’s awareness by inflation. Stephens & Tyran (2012) conducted a survey experiment where people had to evaluate different hypothetical housing transactions. Stephens & Tyran argue that the housing market is ideal for investigating this kind of research question since holding periods are relatively long and therefore substantial inflation can accumulate. They find that people are indeed prone to valuation biases due to inflationary effects. The same results are found by Brunningmeier & Julliard (2008). We can therefore conclude that decisions on real estate are mainly influenced by nominal loss aversion rather than real loss aversion since most people are unable to debias the inflationary effect2.
The main ability to reduce this inflationary bias lies in people’s intelligence. People which are higher educated and score better on IQ-tests can do a better job in debiasing the inflationary effect but only if they are willing to commit themselves to do so. The intelligence part is thus mainly driven by the ability to overrule the intuitive response of system 1 and enable system 2. Therefore Stephens & Tyran (2012) argue that loss aversion is more a result of the system 1 in terms of the two-system theory (Stanovich & West, 2000). Camerer (2005) argued before that loss aversion is often a mistake due to emotional overreaction rather than a genuine preference.
Nominal loss aversion influences the housing market from multiple perspectives. One of the most important effect of loss aversion can be illustrated with the development of the housing market in Boston. Prices of properties in Boston rose 170% from 1982-1989 and fell 40% during 1990-1994. From 1997 the market starts to rise strongly again. These swings in prices are accompanied by swing in sales and listing prices and trade volume. In the downturn in 1992 the asking prices exceeded the expected selling prices by 35% while there was a low trading volume in the market. Therefore housing inventory was high during this period. When the market turned into a boom market the gap between listing prices and sales prices declined to 12% and trading volume became high. Therefore housing inventory was low during this period (Genesove & Mayer, 2001).
The fact that in a bust period the gap between listing prices and selling prices is bigger can be explained by prospect theory. The sellers might be reluctant to sell for the market price and stick to the asking price which is anchored by the acquisition price. This can be justified by the fact that those sellers bought their property before the market turned down. Therefore their nominal purchase amount is above the expected market value. The nominal purchase price is the reference point for most of the sellers. This means that the sellers face a potential loss when they sell for the expected market price. Since they dislike the loss more than a potential gain they will increase their asking prices, do not accept reasonable offers from potential buyers and have a higher reserve price. Therefore properties will also be longer on the market and potentially receive a higher sale price but may also be withdrawn from the market. Agents anchor on both asking prices and realized transaction prices. Somewhat surprisingly people who set higher asking prices because they face a nominal loss do get higher selling prices but only at the cost of longer time to sale. Of course, this argument only holds for the sellers who do not withdraw from the market early (Bokhari & Geltner, 2011; Einiö, Kaustia & Puttonen, 2008; Engelhardt, 2003; Genesove & Mayer, 2001).
The nominal loss aversion is not equally important in every moving situation. People who move intra-metropolitan (within the city/neighborhood) are more loss averse than people moving inter-metropolitan or people moving intra-metropolitan but change to renting a house (Engelhardt, 2003). Both investors and owner-occupants suffer from nominal loss aversion but the intensity differs. Owner-occupants set significantly higher asking prices and the effect is roughly twice as big as the effect on (professional) investors (Einiö, Kaustia & Puttonen, 2008; Genesove & Mayer, 2001). By contrast, nominal loss aversion is larger for more experienced investors and among more professional institutions such as funds in comparison to the smaller private investors (Bokhari & Geltner, 2011). There are no signs that sophistication and experience of sellers reduces nominal loss aversion but being an owner-occupant seems to increase loss aversion probably due to the endowment effect (Einiö, Kaustia & Puttonen, 2008). Evidence, not related to real estate, suggests that substantial amounts of loss aversion can be explained by the decision maker’s knowledge of the product, the age of the decision maker and the importance of the decision to the decision maker (Johnson, Gächter & Herrmann, 2006).
Another effect of nominal loss aversion in real estate is the reduction of household mobility. House owners have their house mainly financed with debt and therefore housing is a highly leveraged asset. Nominal price declines can thus cause equity constrained household who cannot move. Therefore market demand is reduced and this results in more price declines which again leads to lower household mobility (Engelhardt, 2003; Genesove & Mayer, 2001).
2.5 Effects of emotions on loss aversion in real estate
Emotions and moods can be of influence on one’s behavior and economic decision making through 2 channels (Loewenstein, Weber, Hsee & Welch, 2001). At first a person can take expected emotion after a particular decision into account at the moment of taking the decision. In this case the individual would like to optimize his utility by balancing positive and negative emotions (Loewenstein & Lerner, 2003). For example, you know if you sell your house you will feel sad and regretful afterwards. Thus you decide that you are not willing to sell the house.
The second way is the immediate emotions on one’s decision making. One’s immediate emotions can influence one’s decision making in a direct and an indirect manner. Direct immediate emotions have directly influence on the decision making process. This means that feeling the emotion changes your decision making whether you are aware of this
or not. Loewenstein & Lerner (2003) give a good illustration about an investor. This investor might experience immediate anxiety at the prospect of shifting his saving into a risky asset (like a property). Therefore his immediate anxiety leads to a change in his decision making. The indirect immediate emotion can influence ones decision making in 2 ways namely (1) influence people’s expectations of the emotion they will experience in the future (expected emotions) and (2) changes the quality and quantity of the information processing capacity (Loewenstein & Lerner, 2003). The indirect influence can be illustrated by the same investor. If the investor is in a good mood this might change her judgment about future prospect and her ability to shrug off the regret if the asset would drop in value. Thus this investor would think that the future prospects of housing prices as well as the economy are good thus the house is a good investment.
As explained earlier, incidental emotions influence a person’s unrelated decision (unrelated to the cause of the emotion). Therefore emotions often influences decisions in a non-conscious and unwanted way (Lerner, Li, Valdesolo & Kassam, 2015). Emotions carry over from one situation to another. For example, judgement of unrelated topics and even objects have been proven to be influenced by affect through movies, sunny weather, stressful exams and other factors (Bodenhausen, 1993; Clore, Schwartz & Conway, 1994; Forgas 1995a; Forgas & Bower, 1988; Schwarz, 1990; Schwarz & Clore, 1996). The literature also notes that when the emotional influences are unwanted, it is difficult to reduce their effects through effort alone (Lerner, Small & Loewenstein, 2004)3.
Therefore there are several core theories about the role incidental affective states play in the process of decision making. The Appraisal Tendency Framework is one of the most recent within the decision making field of research. With this framework emotions with the same valance can have opposite effects on decision making. This framework links the appraisal processes associated with specific emotions to different judgment and choice outcomes. The general approach predicts that emotions of the same valance can have different effects on decision making, while emotions of opposite valance can have similar influences. This links the affect infusion approach4 with the appraisal processes of emotions. The framework is based upon three broad assumptions: 1. A discrete set of cognitive dimensions changes emotional experience, 2. Emotions trigger a set of concomitant responses (Psychology, behavior, etc.) that enables the decision maker to quickly deal with the problems or opportunities, 3. Specific emotions carry specific action tendencies (Lerner, Li, Valdesolo
& Kassam, 2015). Thus this framework emphasizes that different affective states evoke certain cognitive styles and appraisals. These appraisals will shape how people prioritize and integrate information in the decision making process. For example, sadness can make information about losses more salient and make people want to avoid those losses more than in a neutral state (Lazarus, 1991). In addition appraisals may motivate individuals to seek action to maintain their mood if they find it pleasant or action to alter their mood if they find it undesirable. This is called the mood maintenance and mood repair hypothesis which is empirically validated by Forgas (1995b, 1998, 1999). In addition, there is evidence that people in a good mood increase their heuristic decision making (Bless, Schwarz & Kemmelmeier, 1996). To be more specific, house owners which are happy would like to stay happy and therefore maintain their mood. They will not sell their house at a loss because this transaction would make them lose some of the happiness they experience. On the other hand, a sad individual which is selling his house is more willing to accept an offer which leads to a loss. This is due to the fact he would like to change the situation of sadness into a situation of happiness (mood repair). The unconscious mind will argue like: maybe selling the house and moving on will increase my happiness (or at least decrease the sadness). Many empirical studies have been conducted to test this framework and generally this framework is supported (Bagneux, Bollon & Dantzer, 2012; Cavanaugh, Bettman, Luce & Payne, 2007; Han, Lerner & Keltner, 2007).
Conversely, other authors argue that happiness can induce people to increase their willingness to engage in deep and effortful cognitive processing and therefore increasing accuracy or more elaborate thinking. Therefore these authors argue that positive moods, like happiness, reduce heuristic decision making (Isen, Rosenzweig, & Young, 1991; Wegener, Smith, & Petty, 1995). The decision to sell at a loss will be evaluated more on facts rather than heuristics and therefore the amount of loss aversion will be lower. In addition there is evidence that losses are more salient for sad people in comparison to neutral people. These sad individuals would thus be more aware of the loss and thus their loss aversion could increase (Lazarus, 1991). This argumentation is generally called the hedonic contingency theory.
Another line of reasoning comes from the Affect as Information hypothesis which highlights that individuals judgment of unrelated decisions will be influenced by the cognitive effects of affective states of the individual. Therefore this model argues that the affect is taken directly as information in the cognitive processing of the decision at hand. (Schwarz & Clore,
1983). In our ‘property selling individual’ example this would mean that a positive (negative) affective state will lead a more positive (negative) view of the transaction. Therefore the decision to sell the property at a loss could be influenced. According to this hypothesis happiness will induce the individual to look more positive at the transaction with the loss and thus loss aversion is lower. For a sad individual the same transaction will be evaluated with more negativity and therefore the loss will be valued more important. The loss aversion for this sad individual is thus higher (Han, Lerner & Keltner, 2007).
Table 3 summarizes the different theories and their theoretically predicted effect on loss aversion. As we can see there is some discrepancy in the theories and therefore the effects can be opposite. However effects of loss aversion and emotions differ from one person to another. One argument made by real estate agents is that they can sell properties better (i.e. faster or higher transaction price) due to more trading experience, sophistication and not being related to the property (e.g. not living in the house). I will elaborate on this in next paragraph.
Table 3: Summary
Hypothesis theory Happiness Sadness
Appraisal tendency Higher loss aversion (mood maintenance with more heuristic processing)
Lower loss aversion (mood repair, losses more salient)
Hedonic contingency theory Lower loss aversion (Less heuristic processing)
Higher loss aversion
(More heuristic processing) Affect as information Lower loss aversion Higher loss aversion
2.6 Amateurs and experts
Empirical literature on amateurs and experts shows mixed results and there is so far little consensus. The one exception is the fact that owner-occupants show significant higher amount of loss aversion than investors or other parties (Einiö, Kaustia & Puttonen, 2008; Genesove & Mayer, 2001). This might be due to practical endowment of the product which influences loss aversion. However this goes beyond the scope of this thesis and therefore I won’t elaborate on this. Surprisingly, empirical evidence shows that loss aversion in real estate is larger among more experienced investors and among more professional institutions such as funds in comparison to the smaller private investors (Bokhari & Geltner, 2011).
In addition these authors argue that loss aversion is not reduced by sophistication nor experience of sellers. This is in line with Haigh & List (2005), List (2003) and Coval & Shumway (2005). They find that experienced traders (of futures and options) exhibit greater
behavioral anomalies than students. In line with this argument Johnson, Gächter & Herrmann (2006) argue that respondents who are older and have less education show more loss aversion, suggesting that research based on students may underestimate the importance of loss aversion and other behavioral biases. A recent study in golfing context (including Tiger Woods) shows that professional and experienced agents are not able to reduce the effects of loss aversion in a situation where the stakes are high (Pope & Schweitzer, 2011). Therefore these papers argue that loss aversion is not reduced by sophistication or experience of sellers.
Controversially, other empirical work of Feng & Seasholes (2005) on the disposition effect in financial markets is showing opposite results. They find that neither sophistication nor trading experience alone can eliminate biases but together they can eliminate the reluctance of investors to realize losses. However sophistication and trading experience does not totally eliminate the propensity to realize the gains. This is in line with the results of Kaustia, Alho & Puttonen (2008) which found that financial market professionals are showing lower levels of behavioral biases in their long-term future stock return estimations in comparison to the student group (unexperienced in trading). However they could not find any significant effect of experience within the professional group and thus conclude that there is a limit to debiasing. This is in line with many other papers in the field of finance (Anderson & Sunders, 1995).
We can conclude that literature is mixed about experience and sophistication in reducing biases and loss aversion. However owner-occupants seem to have a higher amount of loss aversion than investors. The bigger and more experienced investors such as funds show higher amount of loss aversion than smaller private investors.
3. Methodology
To empirically investigate the hypotheses there has to be sufficient data but there is no sufficient data available which links emotions with actual decision making on the housing market. Therefore data gathering is needed for this empirical research. The easiest way to gather sufficient data is by conducting an online survey or lab experiment. Since this research is bounded by both time and money constraints the online survey will be the best fitted methodology for this thesis. The online survey will be set up as follows.
3.1 Ways to induce affect
At first the task is to induce a participant with a certain emotion which we would like to test. This can be done via multiple ways. Lerner, Small & Loewenstein (2004) used video clips to induce neutral, sadness and disgust emotions for their participants. Fessler, Pillsworth & Flamson (2004) induced their participants by giving them just a task to write something about a scary situation which they had or something they were really disgusted about. This is in line with Inbar, Pizarro, Gilovich & Ariely (2013) with used this method to induce guilt and Dijk (2015) which induced happiness, sadness and fear. Another method would be via newspaper stories like Johnson and Tversky (1983) did to induce fear for their analysis of risk perception.
Since this research will be based upon an online survey the most important factor is that the proposed emotion is truly induced for the participant. Therefore writing something about an emotional event in your life will be a useful method. In addition the newspaper method could be used but then the participants need to be aware that they are going to be checked on their knowledge of the article. Nevertheless, both methods need to be complemented with some control variables to check to what degree the participants were subject to the emotion.
3.2 Checks for affect
To ensure the affect induction is done properly the survey needs to be complemented with some kind of measurement for affect. In this way we can be sure the correct affect is elicited and to what degree. As an experimenter you would like to actually observe the influence of affect yourself since self-assessment of the participants might lead to more awareness of their emotion and biased answers. Therefore an autobiography writing task can be assessed by the experimenter on how emotional the story is. For example, differencing between someone who wrote about a glass of milk that he/she dropped and someone who wrote about his/her car accident. However for other affect elicitation methods in combination with the online survey a subjective experience measurement is needed. From the psychological literature we get lots of affect score systems but those are mainly focused on mental health (depression and such) rather than affect. Therefore behavioral economics which uses emotions mainly uses a few scales5.
5 Examples are the current mood questionnaire (Barrett & Russell, 1998), Positive Affect and Negative affect (PANAS) (Watson & Clark, 1994), differential emotions scale (DES) (Izard, Dougherty, Bloxom & Kotsch,
This experiment will take the Brief Mood Introspection Scale (hereafter BMIS) as measurement scale in the online experiment. As the name suggests this is a shorter questionnaire (16 questions on a 7 point Likert scale). I will use the translations into Dutch of Erk, Toet & van Erp (2015) which found significant results and good reliability of the measurement. The BMIS consists of 4 scores namely, Pleasant-Unpleasant (BMIS1), Arousal-Calm (BMIS2), Positive-Tired (BMIS3) and Negative-Relaxed (BMIS4). Eight out of the 16 questions are positive valance words while the other are of negative valance. Therefore the scale for the negative items needs to be reversed (Mayer & Gaschke, 1988).
3.3 Pilots
In the initial pilot we tested the combination of two affective state induction methods namely, an autobiography writing task in combination with several emotion inducing images as a follow up. This pilot failed due to various reasons.
The second pilot was a video clip emotion elicitation based upon the findings of Quigley, Lindquist & Barrett (2014) which gave a clear summary of all methods and their efficiency. They argued that video clips are the second best method which can also be easily replicated by other researchers. The best method is based upon images which are commonly used and verified by many researchers. Although the use of images was a big problem in the first pilot. For more detailed information on the pilots see appendix E.
The results of the second pilot (table 4) show significant the differences for the sad and happy movies and have correct signs as expected for all BMIS scales expect BMIS3: Positive-tired. Participants induced with happiness showed higher scores for pleasure and positivity and lower scores for arousal and negativity in comparison to the participants induced with sadness. To see which videos performed best for the happiness and sadness induction we can take a look at the bottom of table 4. As we can see the BMIS1 score does not differ significantly between the sad movies although “The boy in the striped pajamas” performed better by having an insignificant lower mean and standard deviation. Therefore this video is chosen for the main survey. In addition all other BMIS scores showed no significant difference. For the happiness induction the video clip of “Najib Amhali” was chosen because the two-sided t-test shows us that this video performed significantly better in raising the mean of BMIS1 and BMIS3 without increasing the variance.
Affect score N Mean Stnd. Dev Pr(T>t)
BMIS 1 Sad videos 6 55,50 15,55 0,0045***
BMIS 1 Happy videos 6 81,67 8,26
BMIS 2 Sad videos 6 46,33 4,84 0,0003***
BMIS 2 Happy videos 6 30,33 5,35
BMIS 3 Sad videos 6 26,33 6,28 0,1907
BMIS 3 Happy videos 6 29,50 5,68
BMIS 4 Sad videos 6 23,83 6,88 0,0006***
BMIS 4 Happy videos 6 6,50 2,07
BMIS 1 SAD1 "The boy in the striped
pajamas" 2 48,50 6,36 0,4983
BMIS 1 SAD2 "Schindler list" 4 59,00 18,45
BMIS 1 HAPPY "Komt een man bij de dokter" 2 73,00 7,07 0,0494**
BMIS 1 HAPPY "Najib Amhali" 4 86,00 4,69
Note: BMIS 1 is score for pleasant-unpleasant mood, BMIS 2 for Arousal-Calm mood, BMIS 3 for Positive-Tired mood, BMIS 4 for Negative-relaxed mood. Higher scores means a more pleasant, aroused, positive or negative mood.
3.4 Final experiment with loss aversion task
The final experiment is based upon three video clips which should induce a neutral mood, happiness or sadness. The happy and sad inducing videos are “Najib Amhali” (3.11 min.) and “The boy with the striped pajamas” (1.22 min.) which were tested earlier in this paper. The neutral mood inducing video is a validated clip of the movie “Blue” (0.40 min.). This video is made publicly available by Schaefer, Nils, Sanchez & Philippot (2010) for experimental research.
The subjects are given a nominal loss aversion test for the housing market which is in line with Stephens & Tyran (2012). In this test the participant will get a small story about the original purchase price and the price for which the property was sold. The price for which the property was sold can vary between 20% above and 20% below the original purchase price. For each repetition the participant has to note how advantageous the purchase and sale of the property was. In addition the original property value is fixed at the amount of €200.000 which is a nice round number which comes close to the average amount of a regular house. An example of one repetition is given by figure 1. The configuration of the loss aversion task can be found in table 5 while the total survey can be found in the appendix of this thesis.
Table 5: Questions loss aversion task
Situation Initial purchase price Current selling price Nominal Percentage 1 €200,000 €240,000 €40,000 20% 2 €200,000 €225,000 €25,000 12.5% 3 €200,000 €210,000 €10,000 5% 4 €200,000 €200,000 €0 0% 5 €200,000 €190,000 -€10,000 -5% 6 €200,000 €175,000 -€25,000 -12.5% 7 €200,000 €160,000 -€40,000 -20%
Participants were rewarded with an ‘author made’ lottery ticket(s) where the main jackpot is €150,- cash. The lucky participant is randomly drawn from the lottery. Therefore participants are induced to participate in the survey.
The experiment on regular participants was published on a personal Facebook page as well as emailed to direct family members and friends. In addition, the company Ras Makelaars & Hypotheken6 placed an ad on Facebook to promote their followers to participate in the experiment. Most of their followers, and targeted persons, are possible real estate buyers and/or home owners. For the experts survey a private collection of email addresses of real estate brokers was used. These real estate brokers got emailed with the survey link included. Since membership of the professional body requires an extensive training of multiple years of both practical and theoretical aspects the real estate agents can be seen as professionals. Therefore 4800+ companies were emailed, with each employing one or more real estate agents.
The hedonic contingency theory argues that happiness leads to less heuristic processing of information. Therefore decisions are based more on facts rather than on
feelings. A loss situation is therefore only judged by the loss and not by the situational emotions surrounding them. Therefore this theory argues that loss aversion will actually be lower. This is in line with the Affect as information theory which suggests that individuals take their affective state directly as information in their decision making process. Therefore participants induced with happiness are evaluating the losses more “happy” and this reduces their loss aversion. On the other hand, happiness can also increase loss aversion by mood maintenance and more heuristic processing according to the appraisal tendency theory. Therefore hypothesis one will be as follows:
Hypothesis 1:
A. Happiness reduces nominal loss aversion in a housing market setting in line with the affect as information and the hedonic contingency theory.
B. Happiness increases nominal loss aversion in a housing market setting in line with the appraisal tendency framework.
The Affect as information theory argues that sadness leads to a sad view on the decision at hand. Therefore it predicts higher loss aversion. The hedonic contingency theory argues the same conclusion but motivates this with the fact that sadness increases people’s awareness for losses. Therefore the losses are more salient for the participants leading to decreased satisfaction for the loss domain which increases the overall loss aversion. Controversially, the Appraisal tendency theory argues that participants will try to repair their sadness and therefore predict that they will increase their satisfaction by taking the loss to move forward in life. In line with these theories hypothesis two will be as follows:
Hypothesis 2:
A. Sadness increases nominal loss aversion in a housing market setting in line with the affect as information and the hedonic contingency theory.
B. Sadness reduces nominal loss aversion in a housing market setting in line with the appraisal tendency framework.
As discussed in the literature review professionals are seen to be less emotional and better in holding back from behavioral anomalies. However empirical research discussed in the literature review showed mixed results. Therefore hypothesis three will be as follows:
Hypothesis 3:
A. The emotional effect on the nominal loss aversion coefficient is less pronounced for professional real estate agents in comparison to the general public.
B. The emotional effect on the nominal loss aversion coefficient is more pronounced for professional real estate agents in comparison to the general public.
4. Results & Analysis 4.1 Variables
Dependent variable
From the loss aversion task the points 1 to 11 are saved for every question. These points represent the satisfaction score of the participant for the given situation. It must be noted that 6 is the “neutral” position. The task consisted of three gain domain situations, three loss domain situations and one neutral (e.g. break-even) position. Therefore we can construct multiple dependent variables for further analysis.
The first dependent variable is the summation on the satisfaction scores in the gain domain (Table 5: situation 1, 2, 3), break-even position (Table 5: situation 4) and the loss domain (Table 5: situation 5, 6, 7). This can be seen as an overall satisfaction score.
Satisfaction=
∑
s=1 7
Satisfaction scoreis
Where s isthe givensituation∈table 5
From this point we can correct the given satisfaction scores for satisfaction level in the break-even situation. Therefore the scores will only represent the additional satisfaction of the gain or loss domain.
Gaincorri=
∑
s=1 3
(Satisfactionscoreis−Satisfaction scorei s=4)
Losscorri=
∑
s=5 7
(Satisfaction scoreis−Satisfactionscorei s=4)
From those “two” variables we can construct a ratio of point given in the gain domain in comparison to the loss domain. This ratio is valid due to matching of situations in both gain and loss domain which leads to a cumulative sum of the situations in the gain and loss domain of zero. However the Losscorr variable is reverse scored so the difference between satisfaction at gains and “dissatisfaction” at losses can be compared. Therefore the ratio will be one if the additional corrected satisfaction in the gain domain equals the additional corrected dissatisfaction in the loss domain.
Ratiocorri= Gaincorri
Independent affect variables
Since the experiment divided the sample into three groups (i.e. three different videos) a treatment dummy can be created to indicate in which group the certain participant was. Therefore a sad, neutral and happy video dummy was created with binary coding. In the models one of the dummies should be omitted and serve as a reference category. In this thesis this will be the neutral dummy since this is the base affective state of the experiment.
The affective state and the effectiveness of the videos is checked with the brief mood introspection scale (hereafter BMIS). The affective states are thus captured by the BMIS scales. There can be four scales constructed from the results of the BMIS questions. I will refer to those scales with BMIS1, BMIS2 and so forth. A short explanation on the BMIS scales is provided below.
BMIS1
: Pleasant-Unpleasant The score is measured for the pleasant-unpleasant mood. Thisincludes all scores on positive and negative mood states. BMIS2
: Arousal-Calm For the arousal-calm dimension of moods. The adjectivesassociated with arousal in this dimension are active, caring, fed up, gloomy, jittery, lively, loving, nervous, peppy and sad while calm and tired are associated with calmness.
BMIS3 :
Positive-Tired The positive-tired dimension of moods. The positive adjectives are active, caring, lively, loving and peppy while drowsy and tired are adjectives for the tired dimension.
BMIS4 :
Negative-Relaxed The negative-relaxed dimension of moods. The negative adjectives are fed up, gloomy, jittery, nervous and sad, while the relaxed dimension is measured by the adjective calm. Other variables
Due to the fact that some of the participants were professionals and the others were amateurs the sample is also divided into two groups. To see if the professionals significantly differ from the amateurs a professional dummy is used. Since both samples are not equally matched in size, the regressions have to be carefully done to account for differences as much as possible. In addition the ordinal control variables age, education and income are used as well as the dummy control variables female and married. Where the last two dummies indicate whether a person is married or whether this person is a female. As showed in the literature review, these control variables could influence one’s decision making and therefore the amount of satisfaction in any of the given situations.
4.2 Data
Participants
The survey was distributed online which has led to some participants not completing the entire survey. In total 485 persons participated the survey and 328 of them finished the survey. The gender ratio is not evenly balanced, with 37.5% female (N=123) and 62.5% male (N=205) respondents. This is due to the combination of amateurs (N=117) and professional (N=211) respondents. Females (61.5%) dominate the amateur pool while men (75.8%) dominate the professional pool. The age of the average participants lies within the category 35 till 45 years old. However the amateur pool is on average younger (25-35 years) while the average professional is older (45-55 years). Most participant, in both pools, have a HBO or WO degree but the mean of the education variable is lower for amateurs (3.9) in comparison to the professionals (4.77). This indicates that the professional group is more skewed to the left (i.e. higher education) while the amateurs are more skewed to the right (i.e. lower education). This can be justified by the fact that all professional real estate agents should have at least a HBO degree. The mean net salary of the participants is €2000 to €2500 while most observation can be found in the €3000 or more category. The mean net salary of amateurs is €1500-€2000 while the mean for the professionals lies within the category €2500-€3000. A narrow majority (53%) of the participants is married. However the majority of the amateurs is not married (58%) while professionals are mostly married (60%).
Most of the professionals work full time (83%) and sell on average between 30-40 properties a year. The majority of the participants in the amateur pool have never sold a property in their life. Half of this pool is owner of a property and their average property value is in the range of €200.000 till €300.000. Table 6 will provide a clear summary about the variables in the dataset.
The plot in figure 2, which consists the mean values of the satisfaction score, is hardly any different for any of the treatments in the gain domain. In the loss domain both treatments show mean values above the neutral position. The overall ratio between the loss domain and the gain domain could therefore be influenced. In this case the amount of loss aversion could be lowered for the treated participants. These facts hold even after correcting for participant’s own satisfaction in the break-even situation.
In addition we can see that the means of the break-even situation for the uncorrected satisfaction are not really different and have values above six which was the button with the label “Neutral”. Therefore we can conclude that on general participants were more than
neutral satisfied with the even situation. For the corrected satisfaction scores the break-even score is always zero since this is the corrected reference point. Uncorrected and corrected satisfaction scores as well as figures with error bars for amateurs and professionals separately can be found in appendix B. If we look further into the constructed corrected ratio we can see some clear differences between amateurs and professionals.
As shown in figure 3, the neutral treated participants in the total sample are loss averse. This is indicated with the corrected ratio being less than one, which basically tells us the participants disliked the losses more than they liked the equivalent amount in gains. The mean value for professionals (0.98) is far above the mean value for amateurs (0.74) in the neutral treatment. This could suggest that professionals are less loss averse in comparison to amateurs.
The sad treatment resulted into lower ratio scores for the total sample as well as the professional sample. Surprisingly the amateurs decreased their loss aversion in comparison to the neutral position but still remain loss averse. On the other hand, the happy treatment, have led to an increase in corrected ratio levels for all samples (total, amateurs and professional). All samples with happy treatment show mean values above one indicating that the participants induced with happiness liked the gains more than they disliked the equivalent amount of losses. Therefore participants induced with happiness must have scored higher additional satisfaction points in the gain or loss domain or a combination of both. As discussed earlier figure 2 suggest that the biggest difference can be found in the loss domain.
0 2 4 6 8 10 12
Satisfaction s cor e s
Sad Neutral Happy
-5 -4 -3 -2 -1 0 1 2 3 4
Cor re cte d Satisfaction s core s
Sad Neutral Happy
Graphical representation of (corrected) satisfaction scores differentiated by treatment effect. From lef to right: Gains-Losses
Figure 3: Corrected ratio per treatment
S a d N e u t r a l H a p p y 0 0.2 0.4 0.6 0.8 1 1.2 1.4
Corrected ratioper treatment
Amateur Prof Total
Graphical representation of the mean corrected ratio for the amateur or professional pool or the total dataset differentiated by treatment effect.
Table 6: Summary of variables
VARIABLE OBSERVATIONS MEAN STANDARD
DEVIATION MINIMUM MAXIMUM
PROF 328 0.64 0.48 0 1 FEMALE 328 0.38 0.48 0 1 MARRIED 328 0.53 0.50 0 1 EDUC 328 3.48 0.91 0 4 AGE 328 2.17 1.35 0 4 INCOME 328 4.53 2.03 0 7 NSOLD 117 1.68 1.09 1 6 OWNER 117 0.50 0.50 0 1 VALUE 59 3.19 0.99 2 5 SALES 211 3.49 2.29 1 10 FTE 211 0.84 0.37 0 1 DHAPPY 328 0.33 0.47 0 1 DSAD 328 0.33 0.47 0 1 DNEUTRAL 328 0.33 0.47 0 1 BMIS1 328 62.81 14.51 28 98 BMIS2 328 39.81 6.95 22 60 BMIS3 328 26.48 5.94 11 47 BMIS4 328 17.35 7.20 5 39 NORM. BMIS1 322 0.57 0.15 0.22 0.93 NORM. BMIS2 322 0.41 0.10 0.17 0.69 NORM. BMIS3 322 0.51 0.14 0.14 0.86 NORM. BMIS4 322 0.34 0.20 0 0.94 TASK1 328 9.41 1.39 2 11 TASK2 328 8.63 1.53 2 11 TASK3 328 7.71 1.72 1 11 TASK4 328 6.59 1.91 1 11 TASK5 328 5.06 1.86 1 11 TASK6 328 3.79 1.82 1 11 TASK7 328 2.91 1.92 1 11 SATISFACTION 322 44.34 8.26 15 73 GAINCORR 322 6.06 4.28 -7 22 LOSSCORR 322 -8.14 4.92 -23 5 RATIOCORR 322 0.94 1.96 -12 13
Missing data & outliers
The dataset has no missing entries due to forced entry of answers. However some more private questions were answered with the “unknown or I do not want to answer” button. Therefore these participants could not be placed within the certain dummy category. To solve this problem the answer was replaced with the mean value. In total there were 55 replacements of which 48 were in the net income variable and the other in the sales, fte and value variables.
Due to the fact that the dataset was gathered in an online survey, one should be especially careful when looking at the answers. There were six entries in the dataset which showed both high leverage and high residual. Therefore these entries are dropped to insure the
robustness of the OLS regression method since we use a dataset which is smaller than 500 observations. If we do not delete those observations the found regression line would be skewed towards those “faulty” observations instead of the best fitted line.
4.3 Results
4.3.1 Mood induction
To check whether the video actually increased the participant’s moods, the brief mood introspection scale was added after the video. Differences in this scale for the different groups should indicate if the mood induction succeeded or failed. The comparison is done for all of the four items of the BMIS and for both amateurs and professionals with a two-sided t-test and OLS regressions.
Result 1. Subjects induced with sadness are feeling less pleasant, more aroused and more
negative compared to the neutral treatment.
Result 2. Subjects induced with happiness are feeling more pleasant, more positive and less
negative compared to the neutral treatment.
The participants which saw the sad video showed significant lower and higher scores on the BMIS1, BMIS2 and BMIS4 scales in comparison to the neutral group. The participants which viewed the happy video showed significant higher and lower scores on the BMIS1, BMIS3 and BMIS4 scales in comparison to the neutral group7. The participants which were induced with happiness showed no significant difference in the BMIS2 scale, which is the arousal-calm scale, while participants induced with sadness showed no significant difference in the BMIS3: Positive-Tired scale.
The treatment and professional dummy interaction term is always insignificant indicating that professionals do not differ significantly from amateurs with the effect of the treatment. The results generally hold for regressions done with a split sample of amateurs or professionals which can be found in appendix B. However the happy treatment dummy for BMIS3 and BMIS4 turns insignificant for the split sample of amateurs. Regarding the mood induction we can conclude that the videos were successful in inducing different affective states.
Table 7: OLS regression with treatment dummies and control variables
BMIS1 BMIS2 BMIS3 BMIS4
b/se b/se b/se b/se
Sad video dummy -0.0638*** 0.0404*** -0.0113 0.101***
(0.018) (0.013) (0.018) (0.025)
Happy video dummy 0.105*** -0.0130 0.0785*** -0.115***
(0.018) (0.013) (0.017) (0.024) Sad*Prof interaction 0.00532 -0.0257 -0.0173 -0.0398 (0.037) (0.026) (0.037) (0.051) Happy*Prof interaction 0.0468 -0.00767 0.0465 -0.0525 (0.036) (0.026) (0.036) (0.050) Professional dummy -0.0394* 0.0529*** 0.0224 0.0938*** (0.021) (0.015) (0.021) (0.029) Female dummy -0.0215 -0.0161 -0.0288* 0.00628 (0.017) (0.012) (0.016) (0.023) Age 0.0190*** -0.000155 0.0188*** -0.0152 (0.007) (0.005) (0.007) (0.009) Education 0.00354 -0.0143** -0.0115 -0.0166 (0.009) (0.007) (0.009) (0.013) Net income -0.000391 -0.00231 -0.00128 -0.00481 (0.006) (0.004) (0.006) (0.008) Married dummy 0.00780 0.00128 0.00904 0.000464 (0.016) (0.011) (0.016) (0.022) Constant 0.537*** 0.436*** 0.483*** 0.397*** (0.038) (0.027) (0.038) (0.053) Observations 322 322 322 322 R2 0.261 0.128 0.166 0.229
Dependent variable: normalized BMIS1, 2, 3 or 4 scores. Note: The bmis values are normalized on a 0-1 scale. * p<0.10, ** p<0.05, *** p<0.01
4.3.2 Overall satisfaction
By looking at table 8 we can see that the treatment dummies for the sad and happy video and the control variables are never significant for the total satisfaction score. This result holds for amateurs and professionals. This indicates that amateurs nor professionals show significant higher or lower satisfaction scores when induced with sadness or happiness in comparison to the neutral group.
Table 8: OLS with treatment dummies and control variables Satisfaction Ama Satisfaction Prof Satisfaction Ama Satisfaction Prof
b/se b/se b/se b/se
Sad video dummy 0.536 0.543 0.611 0.310
(1.941) (1.398) (1.968) (1.420)
Happy video dummy -0.187 2.256 0.347 2.023
(1.900) (1.403) (1.917) (1.419) Female dummy 2.077 -2.101 (1.688) (1.431) Age -1.219 -0.0705 (0.768) (0.573) Education -0.888 0.680 (0.733) (1.040) Net income 0.150 -0.582 (0.566) (0.541) Married dummy -0.514 0.575 (2.027) (1.206) Constant 44.29*** 43.36*** 46.87*** 44.14*** (1.317) (1.003) (3.158) (5.147) Observations 114 208 114 208 R2 0.001 0.014 0.068 0.029 Adjusted R2 -0.017 0.004 0.006 -0.005 F 0.0717 1.422 1.102 0.853
Dependent variable: Satisfaction
Note: Ama/Prof stands for Amateur/Professional sample * p<0.10, ** p<0.05, *** p<0.01
Standard errors in parentheses
Result 3. Amateur subjects with more pleasant feelings, higher arousal or higher levels of
Result 4. Professional subjects with more pleasant feelings have a significantly higher overall
uncorrected satisfaction score.
Table 9 indicates that both amateurs and professionals were influenced by BMIS1 while amateurs were also influenced by BMIS2 and BMIS3. The effect of BMIS1: Pleasant-Unpleasant was bigger for amateurs than for professionals. An increase of 10 percent in the BMIS1 score would result in a 1.448 increase in total satisfaction score which is roughly two percent for amateurs while professionals would only increase 1.023 or 1.5 percent. In addition amateurs increased their satisfaction scores by 3.7 percent and 2.9 percent for a 10 percent increase in arousal or positivity. Respectively the BMIS2 and BMIS3 scores.
The control variables female and age are (weakly) significant for amateurs while none of the control variables are significant for professionals. Female amateurs were more satisfied than amateur males by roughly 4.5 percent. Being 10 years older as an amateur decreased satisfaction scores with 2.4 percent. So the difference between a 20 year old amateur and a 50 year old amateur is, ceteris paribus, 7.2 percent.
Table 9: OLS with BMIS scores and control variables Satis. Ama Satis. Prof Satis. Ama Satis. Prof Satis. Ama Satis. Prof Satis. Ama Satis. Prof
b/se b/se b/se b/se b/se b/se b/se b/se
BMIS1 6.548 10.35** 14.48** 10.23** (5.591) (4.058) (6.001) (4.101) BMIS2 22.37*** 1.242 25.78*** 0.517 (7.787) (6.907) (7.869) (6.990) BMIS3 14.00** 5.521 20.57*** 5.061 (5.903) (5.106) (6.206) (5.172) BMIS4 7.722* -5.215 5.905 -5.560 (3.931) (3.562) (3.918) (3.591) Female 3.232* -2.128 2.953* -2.128 dummy (1.638) (1.410) (1.628) (1.408) Age -1.695** -0.179 -1.791** -0.136 (0.748) (0.568) (0.756) (0.566) Education -0.429 0.732 -0.441 0.782 (0.710) (1.022) (0.714) (1.020) Net income 0.459 -0.647 0.436 -0.664 (0.529) (0.535) (0.528) (0.535) Married -0.922 0.612 -0.898 0.668 dummy (1.914) (1.191) (1.915) (1.190) Constant 31.99*** 37.85*** 35.28*** 43.29*** 26.79*** 39.25*** 33.77*** 44.42*** (5.229) (4.365) (3.558) (3.618) (6.609) (6.492) (4.981) (6.154) N 114 208 114 208 114 208 114 208 R2 0.069 0.033 0.059 0.035 0.163 0.050 0.156 0.052 Adj. R2 0.053 0.024 0.042 0.025 0.108 0.017 0.100 0.019 F 4.131 3.501 3.495 3.692 2.947 1.508 2.801 1.582
Dependent variable: Overall satisfaction score of sample amateur or professional Note: The bmis values are normalized on a scale from 0-1
* p<0.10, ** p<0.05, *** p<0.01 Standard errors in parentheses
