Financial decision making in a social context:
a field experiment
Universiteit van Amsterdam
Faculty of Economics and Business
Master Business Economics: Organisation Economics
Thesis supervisor: Jeroen van der Ven
Boris van Moorsel
Studentnumber: 10431985
Abstract
In a field experiment, ninety high school students were tested on their financial risk behavior in a social context. They were asked to play a simple game consisting of gambling choices. All the participants played the game alone and in the presence of two peers. Additionally, in one of the in total four rounds, participants received information on the risk attitudes of the two peers accompanying them. No significant difference in risk behavior was found between the treatment played individually compared to those treatments played in the presence of two peers. In conclusion, adolescents are not influenced by peer presence when making financial decisions. Additionally, no significant difference was found between the treatment that provided information on peer risk attitudes and those that did not. The study contributes to existing literature by examining peer influence in the financial domain, a domain that was not yet examined. Additionally, a modified experimental method was developed through the combination of existing methods. The new method contributes to the existing methods by combining a good comprehensibility for participants and a wide range of risk attitudes that it is able to analyze.
Keywords: Risk aversion, Peer Influence, Behavioral Economics
Boris van Moorsel
Cornelis Trooststraat 33-1 1072JB, Amsterdam bmoorsel@hotmail.com Student number: 10431985
1 Introduction
Many financial choices that people make have uncertain outcomes. However, research in the field of economics has shown that, on average, people prefer to avoid uncertainty (Holt and Laury, 2002; Millner and Pratt, 1991). The concept of risk aversion was therefore incorporated in standard economic models. Prospect theory (Kahneman and Tversky, 1979) and expected utility theory (Arrow, 1971) established risk aversion as a vital element in the explanation of decision making behavior. Despite the increasing incorporation of risk aversion within standard economic theories, in real life, people make choices that expose them to considerable amounts of risk. Examples of such risk seeking behavior are stock investments, smoking, unprotected sex, gambling and drug abuse. The question remains as to why people engage in this kind of behavior, when they are in fact, risk averse.
One explanation could be that people make riskier choices in the presence of their peers than they would make alone. The adulation of risk taking behavior by modern society could be the driver of this influence. The influence of peers has proven to be strong in relation to risk seeking behavior. In the case of smoking for example, Biglan et al. (1995) confirmed that peers attributed to the consumption of cigarettes by fellow students.
Despite the vast amount of research on the subject, peer influence has not been studied in a financial context. This is interesting because the use of teams in organizations has increased considerably (Piña et al., 2007). In the firms of today, many financial decisions are thus made in a group composition. And while decisions are made in a group composition, they may not be taken by all the group members together. Instead, decisions can also be taken by one individual in the presence of other group members or peers. For example, a team leader has the power to decide over the faith of a particular project with important financial consequences for the him or herself. If such a decision is made in a team meeting, the decision maker will possibly be influenced by the presence of his or her team members. In the search to act conform the attitudes of the peers that are present, a decision maker could make a different decision than when the decision would be made alone. In this case, information on the risk attitudes could have a great influence on the decision that is made. Understanding the direction and the magnitude of peer influence can thus be important information for organizations. With full knowledge of the influence of peers on financial decisions, organizations can adjust the decision making process to suit their risk profile. To examine the influence of peers on financial decision making, the following research question was formulated:
Research question:
Do people make less risk averse financial decisions in the presence of their peers than they make alone?
A field experiment addressed the research question above. Using a French playing cardgame, a simple gambling game was developed. All participants were endowed with 200 eurocents every turn that was played. Each turn, one card was drawn from the deck. Participants were presented with the choice to bet on either the symbol or, the number or the image of the card that was drawn. Betting was financially risky, since the amount betted was lost if participants betted on a different symbol or number or image than that of the card that was drawn. Betting on a number was rewarded differently compared to betting on an image, taking into account the probability of choosing the correct card. The experiment consisted out of four rounds. In one of the rounds of the experiment, participants made decisions alone. In the remaining rounds of the experiment, decisions were still made by one individual, but in the presence of two peers. In two of these rounds, no information was transferred from the peers to the decision maker. Finally, in one round, information on risk attitudes of the peers was provided to the decision maker.
The sample group consisted out of ninety Dutch high school students. Due to financial
constraints only nine of the participants were paid their earnings. Adolescents were chosen as the subject pool for two reasons. Firstly, due to the fact that adolescents are especially vulnerable to peer influence (Gardner and Steinberg, 2005) and secondly because an adolescent sample group could be found easily.
Results showed no significant difference in risk aversion behavior between the rounds
played individually compared to those rounds played in the presence of two peers. Additionally, no significant difference was found between the rounds that provided information on peer risk attitude and those that did not. Due to the limitations discussed in section 6, replication of the experiment with similar and different sample groups is desirable. For these replicated experiments, it is advised that the payout amounts are increased and that every participant of the experiment is paid, because hypothetical payments might be the cause for the erratic behavior that was found in the results. Finally, the design of the experiment should be revised to contain well-defined risk classifications that correspond to a more extensive scope of risk attitudes than the present experiment.
The content of this paper is divided as follows. The second section will provide an overview of the related literature. The third section will explain the methodology of the experiment and will elaborate on the research question and the hypotheses. The fourth section will give the results of the experiment and the fifth section will conclude after discussing these results. Finally, section six includes the limitations of this study and the recommendations for further research.
2 Related Literature
While people are, on average, claimed to be risk averse (Holt and Laury, 2002; Baker et al., 2007; Binswanger 1980), they engage in various forms of risk taking behavior such as smoking, gambling and drug abuse. Risk seeking behavior is especially found in adolescents’ age groups (Burnett et al. 2010, Boyer 2006). In this age group, peers have been identified as a major influence on risk taking behavior (Gardner and Steinberg, 2005). Interestingly, no research has been performed on the influence of peers with respect to financial risk taking. The following paragraphs will give an overview of the literature with respect to this subject. The first section of this chapter will provide an overview of the studies with respect to influence of peers. Section 2.2 discusses the literature on financial risk. Section 2.3 covers the most important elicitation methods of risk that have been applied in previous experiments. Finally, Section 2.4 discusses the relevance of information exchange concerning the risk attitudes of peers.
2.1 Influence of Peers
The society of today has a positive association with risk taking. This positive association is clearly proclaimed in a statement by Wilson and Daly (1985): “Successful risk taking certainly evokes admiration”. The authors indicate that people often predict future winning probabilities by looking at past victories. Luck can thus be perceived as a personal quality, resulting in a prestigious status. In this way, accepting risk is interpreted as a positive phenomenon and prestige is attributed to those persons taking risks, even when it concerns games of pure chance. Taking this into consideration, it is likely that people will act less risk averse in the presence of their peers.
In an interesting experimental study by Gardner and Steinberg (2005), risk taking behavior was studied among three age groups: adolescents (13 16), college students (18 22), and adults (24 and older). Two questionnaires were used to assess risk preference and risky
decision making. Risk taking was assessed through a computer game which required a participant to stop a car or keep it moving at the moment when the traffic light jumped from green to yellow. The further the car would drive without crashing, the higher the reward that the respondent received. For this game, participants were randomly assigned to either the group, existing out of 3 participants, or the individual treatment.
The conclusions that were reached by Gardner and Steinberg are interesting. The authors found that risk taking decreased with age and that participants made riskier decisions when they were in peer groups than when they were alone. They concluded that “peer influence plays an important role in explaining risky behavior during adolescence”.
However, the authors used different procedures for the recruitment of adolescents compared to the college students and adults. This resulted into the fact that adolescents did not know their group members while group members of students and adults were familiar with one another, which could have led to biased results.
Additionally, the external validity of the results of Gardner and Steinberg’s
experiment may be limited, due to the specificity of the traffic light game. It is difficult to tell if the results by Gardner and Steinberg can, for instance, be applied to financial decision making.
Risk preference and risk decision making were examined by looking at the results from established scales that used hypothetical situations for their assessment. Both required respondents to complete questionnaires. The reliability of questionnaires has been debated since questionnaires require a fair amount of self-knowledge from the respondent and due to fact that the framing of questions has an influence on the answers that are given (Kahneman and Tversky, 1984). Also, questionnaires do not incorporate incentives, leading to less reliable data.
Despite the shortcomings of the study by Gardner and Steinberg, the division of age groups is interesting, because hereby the authors acknowledge that demographic characteristics can affect risk attitudes. The experiment of this paper builds upon the implications of the study by Gardner and Steinberg, but aims to specify the domain of risk behavior by relating peer influence to financial decision making. To overcome the unreliability of questionnaires used by Gardner and Steinberg, financial incentives were provided to the participants of the present experiment. Paragraph 2.3 will examine risk aversion in the financial domain in order to create a full understanding of the topic of this study.
The following paragraph addresses the influence of information exchange. In groups, previous research has indicated that information on the risk preference of group members significantly influences the risk behavior of a decision maker. It is most likely that this effect will also be present in the experiment of this paper.
2.2 Influence of Information
In the real world, communication leads to the sharing of information within teams. To test the influence of this information, one round was created in which the decision maker received information on the risk attitudes of the peers that were present. It has been proven that when forming a group, social pressure can make an individual change the position they would take individually to a more socially desirable one (Mcguire et al, 1987). This happens especially when the individual has information on the risk attitude of fellow group members.
Harrison et al. (2013) examined the role of individual preferences over social risk. Their experiment contained two different treatments. In the first treatment individuals were provided with information about the risk preferences of the other members of their group and the second treatment they did not have this information. The authors concluded that providing a person with information had a significant effect on their behavior with respect to risk. Subjects showed more risk aversion when they knew the risk preferences of the members belonging to their group. As was indicated in section 2.1, the next section will elaborate on the existing literature on financial risk behavior, examining both the situation of group and individual decision making under risk.
2.3 Financial Risk
Previous laboratory experiments have examined individuals dealing with financial risk. In the majority of these studies, individuals exhibited low to high levels of risk aversion. One of the most influential papers on this subject was written by Holt and Laury in 2002.
Holt and Laury examined risk aversion using a variety of payoff levels. In their study,
the authors addressed the influence of hypothetical versus real and low versus high payoff levels and hereby tested the concept of constant relative risk aversion. By developing a series of ten paired lottery choices they were able to assess the risk aversion levels of every individual in their experiment. Each choice consisted out two options: a safe and a risky option. The number of safe choices served as the measure of risk aversion to compare the payoff treatments with one another. They found that most people are risk averse, especially
when payoffs are real compared to hypothetical and when payoffs are high compared to low. Additionally, the authors discovered that a higher scale leads to greater risk aversion.
Harrison et al. (2005) pointed out that Holt and Laury did not control for order effects. Studying these effects, Harrison et al. found that order significantly affects risk aversion levels. Furthermore, while the scale effect was still present and significant, its influence decreased considerably when controlling for order effects. Holt and Laury reacted with a follow up study in 2005 in which order effects were taken into account and in which the conclusions they had drawn originally were supported by new data.
Holt and Laury examined the risk behavior of individuals. This paper focuses on the behavior of individuals in the presence of peers. It is likely that the influence on behavior is similar to the influence that is registered for individuals in a group composition. Therefore, it is worth taking a look at group mechanisms with respect to financial risk aversion. Previous economical experiments have indicated that in general, compared to isolated individuals, groups behave more risk averse (Masclet et al, 2009; Baker et al, 2007).
Masclet et al. (2009) used the same method as the one that was used by Holt and Laury (2002) and derived ten lottery choices which were presented to both individuals and groups. Decisions within groups were reached by means of voting. Participants were asked to reach a unanimous decision within the limit of five voting rounds. If the decision was not reached, a computer would randomly make the decision for the group. Results showed that groups chose safer lotteries than the average of all individuals belonging to that group. This finding did not change when socio demographic variables were taken into account.
Baker et al. (2007) concluded similarly after examining the results from their experiment: “the count of safe lottery choices submitted by three person groups is significantly greater than the mean of the group members”.
However, the elicitation method that was used to determine the risk attitude in the three papers discussed above, also known as the Multiple Price List (MPL) method, has been criticized for being too complex (Charness et al., 2013). The next paragraph discusses why and introduces an alternative elicitation method which overcomes the complexity of the MPL method.
2.4 Previous Methodologies
Over the past three decades many elicitation methods for risk aversion have been developed. This paragraph will discuss the MPL method and the method developed by Gneezy and
Potters. For this discussion, these two methods were chosen because they form the basis of the method that was applied in the present experiment.
The MPL method used by Holt and Laury, Baker et al., Masclet et al. is a widely used
method. Individuals are presented with a series of two lotteries. One choice is the safer choice, due to the fact that difference between the payoffs of the gamble is small. The second choice is risky due to the fact that the difference between the payoffs of the gamble is greater, leading to a larger standard deviation of the expected payoff. Subsequently, the probabilities of the occurrence of the high (and low) payoff are adjusted along the series, to make the riskier option more attractive. At a certain point, the risky option becomes so attractive that switching from the ‘safe’ option becomes rationally inevitable. The switching point is used to determine the risk attitude of an individual. Despite the fact that the method has often been used by experimenters, the method can be considered too complex (Charness et al. 2012) because it requires a degree of mathematical knowledge from the subjects of the experiment. This makes it difficult for some sample populations to thoroughly understand the procedure, leading to unreliable results and possibly inconsistent choices (Dave at al., 2010).
An alternative method used to determine risk attitudes was developed by Gneezy and
Potters. In their experiment, subjects were endowed with 200 cents. Participants could spend an amount X (0≤X≤200 cents) of their endowment in a lottery. The lottery comprised of a probability of 1/3 to win two and a half times the amount betted and a probability of 2/3 to lose the amount that was betted. Total earnings were calculated as follows: the starting amount of 200 eurocent plus the earnings in the lottery of that round (more specifically, plus 2,5 times the amount betted or minus the amount betted). The amount that was betted was used as a measure for risk aversion. Compared to the MPL method, the method by Gneezy and Potters is easier to comprehend. It requires little mathematical skill and a single simple choice, namely to choose the amount X to bet. But as Charness et al. point out, an elicitation measure in this form comes down to the fact that a risk neutral or a risk seeking individual bets the entire endowment and a risk averse individual bets a certain part of the endowment or no money at all, depending on the degree of risk aversion of the specific individual. This limits the scope of the elicitation measure with respect to risk attitude because it does not differentiate between individuals who are risk neutral and those who are risk seeking.
To overcome the shortcomings of the elicitation methods described above, the method
of Gneezy and Potters was modified for the experiment of this paper. Section 3 will go into more detail with respect to this modification and the method in general.
3 Methodology
As is indicated in the previous section, the method of this paper was derived from the experiment developed by Gneezy and Potters in 1997. This method was selected for its simplicity and reliability compared to alternative methods. Simplicity was especially important because the sample population consisted out of high school students, who were expected to possess less mathematical knowledge than the average college or university student most often employed in economical experiments.
The experiment took place at the high school Rijnlands Lyceum Wassenaar. All
participants in the experiment attended this school. All students followed courses from the same education level, namely the VWO, the standard education preceding Dutch university education. In total, 90 students participated in the experiment.
The following paragraph will explain the task that the participants of the experiment were required to perform. Paragraph 3.2 will discuss the measure that was used to rank each of the possible alternatives on their risk levels.
The experiment considered three treatments: 1. the Individual treatment, 2. the Peer
no information treatment and 3. the Peer information treatment. Section 3.3 to 3.5 thoroughly explain all treatments. Rounds were designed in line with these three treatments. The experiment consisted out of four rounds. To double the amount of observations, each round contained two successive turns by the same individual.
The game was explained by the experimenter who read the first part of the instructions (see Appendix Explanation 1 UITLEG). Two practice rounds were played to confirm the understanding of the game by the participants. After the experiment, each participant was asked to complete a questionnaire. The questionnaire included questions on the characteristics and risk behavior of the individuals taking part in the experiment (see Appendix Explanation 1 Vragenlijst). For a full overview of the chronology of the experiment please see the logbook in the Appendix (Explanation 2).
3.1 Task
As in the experiment by Gneezy and Potters, each turn, a participant was endowed with 200 eurocents. The participant was asked to bet an amount X (from 0 to 200 eurocents) of their choice on either one of the four symbols or on a number or image of the card, of a “regular” French playing card game. The amount that was not betted was kept by the participants and hereby risk free. After all participants had placed their bet, a card was drawn.
Symbols of choice were spades (♠), hearts (♥), diamonds (♦) and clubs (♣). A bet on a symbol implied that the participant had a 25 percent chance of choosing the correct symbol. If a person had chosen the symbol corresponding to the symbol of the card drawn, the amount betted was quintupled. If this was not the case, the amount betted was lost. Hereby, an expected payoff of 125 percent was realized when one betted on a symbol. Since the average individual is claimed to be risk averse (Holt and Laury 2002), an expected gain was necessary to attract individuals into betting. Only if individuals bet, it is possible to register a change in risk behavior.
To enhance the scope of risk aversion levels, in comparison with the original
experiment by Gneezy and Potters, a second lottery was included. Participants had the option to bet on a number or image of the card that was drawn: 2 10, Ace, Jack, Queen or King. If a person chose the number or image corresponding to the number or image of the card drawn, they would receive ten times the amount that was betted. The chance of choosing the correct number or image was 1 divided by 13, namely around 7,7%. A correct number or image bet was rewarded with ten times the amount betted. This comes down to an expected payoff of about 77 percent. Hereby, betting on number or image leads to a negative expected loss, which would only be attractive to extreme risk seekers. Table 1 stipulates the importance of this second lottery. The second lottery creates an extreme risk seeking alternative, one that was missing in Gneezy and Potters experiment.
All participants were asked to record their choices on the forms provided to them. To
overcome the possibility of cheating, each decision form was collected before the card of that particular turn was drawn. Cards were randomly drawn by assisting teachers and cards were shuffled with a shuffling machine. Completeness of choices was checked by the experimenter and the assisting teachers every turn.
Due to financial constraints not all participants were paid out. Three persons per class were paid out their earnings on the day following the experiment. All other participants did not receive payment.
3.2 Risk Aversion Measure
To make analysis possible, a risk level had to be assigned to every possible alternative available to the participants. This process involved calculating the ranges of relative risk aversion for every possible betting amount for a symbol and for an image or number. Similar calculations have been applied by Holt and Laury, to facilitate the classification of the
choices of their experiment in 2002. All relative risk aversion r’s are determined by maximizing utility. Holt and Laury used the following formula for utility: U(x) = x!!! 1 − 𝑟
where U(x) is utility and x is the constant relative risk aversion for money. Risk attitudes are defined as follows: r<0 for risk seeking, r=0 for risk neutrality and r>0 for risk aversion. However, to complete the calculations x should be specified. Therefore, the formula for expected payoff was developed.
µ = expected payoff
p = probability of correct choice m = reward multiplier
a = amount betted
µ = (p)(ma + 200-a)+(1-p)(200- a)
A bet placed on a symbol implied m=5 and p=1/4
A bet placed on a number or image implied m=10 and p=1/13
Specifying x in the utility formula, the following two formulas for bets on a symbol (1.1) and for bets on a number or image (1.2) were obtained.
1.1 U(a,r) = !! (4a+200) 1-r + ! ! (200-a) 1-r 1.2 U(a,r) = ! !" (9a+200) 1-r + !" !" (200-a) 1-r
To find the range of relative risk aversion for each betting option, the optimal betting amount for every possible r (with three decimals) was found, given that utility was maximized. This was done by simulating a wide range of r’s in Excel. Boundaries were indicated by a change in the optimal betting amount from one particular r to the next. The results of this process can be found in Table 1 below.
TABLE 1: RISK AVERSION CLASSIFICATIONS Symbol/ Number or image Betting amount
Range of relative risk aversion Risk level Risk preference classification Number or image
200 r < -0.700 r = -0.700 Highly risk loving
Symbol 200 -0.700 ≤ r <0.052 r = -0.324 Risk neutral
Symbol 190 0.052 ≤ r< 0.070 r = 0.061 Slightly risk averse
Symbol 180 0.070 ≤ r < 0.081 r = 0.076 Slightly risk averse
Symbol 170 0.081 ≤ r < 0.090 r = 0.086 Slightly risk averse
Symbol 160 0.090 ≤ r < 0.100 r = 0.095 Slightly risk averse
Symbol 150 0.100 ≤ r < 0.109 r = 0.105 Slightly risk averse
Symbol 140 0.109 ≤ r < 0.119 r = 0.114 Slightly risk averse
Symbol 130 0.119 ≤ r < 0.129 r = 0.124 Slightly risk averse
Symbol 120 0.129 ≤ r < 0.141 r = 0.135 Slightly risk averse
Symbol 110 0.141 ≤ r < 0.154 r = 0.148 Slightly risk averse
Symbol 100 0.154 ≤ r < 0.169 r = 0.162 Risk averse
Symbol 90 0.169 ≤ r < 0.186 r = 0.178 Risk averse
Symbol 80 0.186 ≤ r < 0.208 r = 0.197 Risk averse
Symbol 70 0.208 ≤ r < 0.235 r = 0.222 Risk averse
Symbol 60 0.235 ≤ r < 0.270 r = 0.253 Risk averse
Symbol 50 0.270 ≤ r < 0.321 r = 0.296 Risk averse
Symbol 40 0.321 ≤ r < 0.399 r = 0.360 Risk averse
Symbol 30 0.399 ≤ r < 0.535 r = 0.467 Risk averse
Symbol 20 0.535 ≤ r < 0.850 r = 0.693 Very risk averse
Symbol 10 0.850 ≤ r < 2.476 r = 1.663 Very risk averse
Symbol 0 2.476 ≤ r r = 2.476 Stay in bed
Table 1 does not provide the risk levels for bets of 10 to 190 eurocents on a number or image. For these alternatives, utility could not maximized for any r. As a consequence, it can be concluded that these choices do not coincide with any form of rational behavior towards risk. Due to their inability to represent a risk aversion level, these choices were defined as inconsistent and were therefore discarded. For the remaining choices, the mean of the ranges
displayed in column 3, were taken as the numeric indicator for risk aversion, as displayed in column 4. By choosing to use the mean of the ranges of relative aversion, an important assumption is made. Namely, for each option, the mean of the r range provides a sound representation of the risk attitudes of the average individual choosing that specific option. For two alternatives, the risk aversion level was determined differently. For the two most outer boundaries of all the relative risk aversion ranges, the starting point was used as the numeric indicator. This indicator is imperfect but provided a solution to the fact that both boundaries run from a certain point to infinity and means cannot be calculated. For the most risk seeking choice of a 200 eurocent bet on a number or image, the risk level was set at -0.700. For the most risk averse choice of betting nothing in either lottery, the risk level was set at 2.476. The drawback of this reasoning is that the model as such does not account for risk seekers with a lower r than -0.700 nor for risk averse individuals with a higher r than 2.476. This limits the scope of risk aversion levels of the experiment as individuals with more extreme attitudes than the outside boundaries are filtered out of the experiment. Additionally, there is a chance that the mean of the range does not represent the actual risk level of the decision makers who choose the alternative corresponding that range. This applies especially to a bet of 200 eurocent on symbol. The range of this alternative is extensive, comprising an r from -0.700 to 0.052. The mean is calculated at -0.324, implying a risk seeking attitude. Despite this negative mean, it is possible that all decision makers who choose this alternative are actually risk neutral (with an r of 0.000) or even slightly risk averse ( with an r of 0.001 to 0.052).
However, the main research question of this paper compares the risk level between two treatments. Since the same risk aversion measure was applied in both treatments, if a significant shift is found, this is still caused by a change in the choices that were made in the first treatment compared to the second, despite the possible inaccuracy of the measure.
The risk classifications displayed in the last column of Table 1 are partly based on those inferred by Holt and Laury. They are, however, slightly adjusted to fit the data that was produced by the experiment of this paper.
3.3 Individual Treatment
In the individual treatment, all participants played the game alone. The individual treatment provided insight into the risk behavior of individuals with no interference of any kind. This treatment serves as the control treatment and was compared with the other two treatments explained in the following paragraphs. In line with the results of the experiments by Holt and Laury (2002) and Binswanger (1980), it was expected that individuals would show some degree of risk aversion. Therefore, the following hypothesis was formulated:
Hypothesis 1
For the individual treatment, the mean risk level will be higher than 0.
As is explained above, a positive r implies risk aversion. Holt and Laury stated that, most commonly, the risk aversion level falls within the r range of 0.3 to 0.5. However, since the alternatives and payoff levels available to the participants of the present experiment are distinctively different, registered risk levels may differ from the experiment performed by Holt and Laury. Nonetheless, a positive mean risk level average is expected.
3.4 No Information Peer Treatment
Each class played three rounds in which the decision maker was in the presence of two peers. In two of these rounds, no information was exchanged between the decision maker and the accompanying peers. For this treatment, the class was split up into two groups. One third of the class was assigned the role of decision maker or player, the other two third of the class was assigned the role of spectator. Players played the game while spectators watched. Only players were allowed to make choices. Each player was accompanied by two spectators.
This treatment examined the choices participants made when they were in presence of
their peers without any form of communication or information exchange. For this treatment, it was hypothesized that players would behave less risk averse than in the individual round. This hypothesis follows from the results of the experiment by Gardner and Steinberg (2005). The authors concluded that the presence of peers leads to more risk taking behavior. Players may feel pressured by the presence of peers to play more aggressively. The source of this pressure lies in the fact that taking risks is generally perceived as an admirable course of action (William and Daly, 1985). In other words, being considered a risk taker by one’s peers
is desirable. This led to the formulation of Hypothesis 2 with respect to the no information peer treatment:
Hypothesis 2
Compared to the individual treatment, the no information peer treatment will produce a significantly lower mean risk level.
3.5 Information Peer Treatment
One round also considered a peer treatment and proceeded in similar matter as the two peer rounds described above. In this round, however, the two spectators were able to provide the player who they were assigned to with advice. Advice was provided to all players just prior to being asked to make their choices. Advice was presented to them in the form of an advice choice indication form (see Ronde 3, from Explanation 1 in the Appendix), which had to be completed by all spectators of that round.
Harrison et al. (2013) studied the effect of providing a person with information on the
preferences of others. This information led individuals to behave more risk averse (see section 2.2). Therefore, the following hypothesis concerning the information peer treatment was formed:
Hypothesis 3
Compared to the no information peer treatment, the information peer treatment will produce a significantly higher mean risk level.
To overcome the possibility of bias due to a set round sequence, the sequence of the rounds was different for every class. Table 9 in the Appendix provides an overview of the sequence of the rounds for all three classes.
3.6 Questionnaire
After all four rounds of the game were played, participants were asked to complete a questionnaire (see the Vragenlijst in the Appendix, Explanation 1). The questionnaire contained questions on the characteristics and the behavior of the students. The questionnaire was included to examine the influences of three aspects with respect to risk behavior in a
social context. The influence of the gender of the of spectators was addressed by question 8 and the presence of befriended spectators was addressed by question 9 of the questionnaire. Finally, question 10 addressed the popularity of the decision maker.
As one will see in section 4.6, linear regressions were performed to examine the influences of the aspects above. The expected relationship to and influence on risk behavior of all variables that were derived from the questionnaire will be discussed the remaining part of this section.
Previous research has provided evidence for age and gender as determinants of risk attitude. Risk aversion has been shown to increase with age (Halek & Eisenhower, 2001) and men, in general, behave less risk averse than women (Byrnes et al., 1999). Therefore, one can expect a positive relationship of age to the mean risk level and a lower mean risk level for men compared to women.
Questions 1 until 4 contained questions regarding risk seeking behavior. Smoking and
alcohol consumption are widely known as risk seeking behavior because these habits have been proven to be related with certain health issues. As was indicated by the World Cancer Report (Stewart and Kleihuis, 2003), “Smoking causes lung cancer”. In turn, alcohol consumption has been related to variety of problems, ranging from mental illness to liver cirrhosis (World Health Organization, 2004). Bone fractures can be related to risk seeking behavior since they often involve engagement in risky activities or recklessness . Extreme sports have been linked to risk attitude due to the high chance of dangerous, uncontrollable and sometimes fatal situations that can occur when practicing these sports (Baker and Simon, 2002). A general risk question was included in the form of question 5 to examine if students are able to accurately estimate their own risk attitude. The more risk seeking an individual indicates to be, the lower the expected risk level of that person. Questions 6 and 7 were included as control questions, to check if all participants understood the game and the payout structure.
The last three questions addressed aspects that could influence the decision maker due
to the presence of peers. The gender of the spectators might have an influence on the behavior of the players. Question 8 was included in the questionnaire to address this aspect. As pointed out above, males are generally less risk averse than females. Therefore, one can expect that male decision makers behave less risk averse when they are in the presence of male spectators. However, it is also possible that that men would like to behave especially tough to impress the opposite sex. For female subjects it is questionable how they will be influenced by the gender of their spectators. Thus, the direction of the relationship was predicted as ambiguous.
Friends might also influence the behavior of the players. To test for the influence of friends, question 9 was included in the questionnaire. It asked students if they were friends with the peers that were assigned to them while they were players. Most probably, the presence of friends will encourage players to behave more risk seeking, since they prefer not to lose face in front of their friends. Best friend influence in smoking proved significant in the study performed by Urberg in 1992.
Finally, question 10 addressed the popularity of players. Participants were asked if they had 5 or more, or less than 5 friends participating in the experiment. This element was included to examine if popular individuals would choose significantly different from less popular individuals. Popular individuals might be more concerned with their reputation. Taking this into account, they are expected to choose less risk averse than less popular individuals. By doing this, they hope to evoke admiration from their peers because they dare to take risks.
4. Results
In section 4 the results of the experiment of this paper will be examined. The section is organized as follows. Paragraph 4.1 discusses the general and most fundamental statistics of the experiment. Section 4.2 goes into more detail in dealing with the main research question of this paper by examining the differences between treatments. Section 4.3 4.4 cover the differences between classes and rounds. Section 4.5 focuses on the information peer treatment by examining the influence of information more closely. Finally, section 4.6 examines the results of the linear regressions that were performed with the variables generated by the questionnaire of the experiment.
4.1 Summary statistics
In total, three classes of 30 students participated in the experiment. 62 of the participants were male and 27 were female, one person did not indicate gender correctly. One person was of American nationality, one of Chinese nationality, five were of dual nationality (including the Dutch nationality) and all other students were of Dutch nationality. The average participant was 16.2 years of age. The students played 4 turns each, which accumulates to 360 turns in total. 277 turns, players placed a bet on a symbol. 81 turns, players betted on a number or image. Of these 81 turns, 28 were defined as inconsistent according to the process
developed in section 3.2 of the Methodology section. Two turns were inconclusive due to an incorrect indication of choice.
With respect to the risk levels, Table 2 below provides an indication per treatment. The standard errors are given in parentheses. Additionally, it gives the total amount of observations produced by each round. As discussed in the Methodology section, inconsistent choices were discarded and were not used for analysis.
TABLE 2: SUMMARY STATISTICS
Treatment Mean risk level Total # observations
Individual -0.154 (0.026) 165
No information peer -0.148 (0.034) 109
Information peer -0.192 (0.042) 56
First inspection of the risk levels above shows slight differences between the three treatments of this experiment. Individuals were found to be most risk seeking in the information peer treatment followed by the individual and the no information peer treatment. Contrary to what was hypothesized by Hypothesis 1, a negative mean risk level for the individual treatment was found. However, this result cannot be interpreted as black and white as stated above. The mean risk level of the individual lies within the relative risk range of a 200 eurocent bet on a symbol, where the risk level was set at -0.324. Since the range of this bet is extensive and includes a wide variety of r levels (negative, zero and positive), its inaccuracy makes it difficult to conclude on the exact risk attitude of the sample population in the individual treatment. The following table however, can provide an overview of the proportion of choices made for each risk classifications set out in Table 1 of the Methodology section.
TABLE 3: PROPORTION OF CHOICES PER RISK CLASSIFICATION Individual No information peer Information peer
Highly risk loving 0.15 0.17 0.18
Risk neutral – risk loving 0.39 0.37 0.38
Slightly risk averse 0.19 0.17 0.30
Risk averse 0.26 0.26 0.14
Highly risk averse 0.00 0.03 0.00
The proportions in Table 3 indicate great variety in risk attitudes throughout all three treatments. The proportion of highly risk lovers in each treatment differs by 2 to 3 percent. This classification includes all 200 eurocent bets on a number or image. A bet of 200 eurocent on a symbol was placed most frequently, indicated by the risk neutral to risk loving classification. The proportions across treatments are roughly equal, centered around 0.38. Slight differences are found in the risk averse segment. Especially in the info peer treatment, risk aversion attitudes seem to be less extreme than the other two treatments. However, taking all risk averse individuals together for each treatment, the total proportion of individuals that were classified as risk averse is reasonable constant throughout the three treatments. The proportions are 0.45, 0.46 and 0.44 for the individual, the no information peer and the information peer treatment, respectively. Finally, the results do clearly indicate that, contrary to the results by Holt and Laury, risk aversion levels are very varied and not centered around the 0.3 to 0.5 range. This can possibly be explained by the fact that only 10 percent of all participants were paid out their earnings. As is pointed out by Holt and Laury (2002), hypothetical payoff treatments can lead to erratic behavior. Another explanation could be a lack of understanding of the game. Control question 6 and 7 of the questionnaire provide a better insight into this issue. The results of question 6 indicated that 10 percent of the respondents had difficulty understanding how their earnings were calculated (understanding below 4 on the scale of 1 to 5). The results of question 7 indicated that around 11 percent of the individuals had difficulty understanding the difference in riskiness of betting on a symbol compared to betting on a number or image, indicated by a choice of answer B. Furthermore, around 8 percent of the individuals chose inconsistent alternatives, indicating a lack of understanding as well. On the basis of these numbers, there is no reason to conclude that the experiment was prone to serious lack of understanding. The coming paragraphs of this section will go into more detail to analyze the data produced by the present experiment.
4.2 Differences between treatments
To address the main research question of this paper, the significance of the differences in risk level between treatments should be determined. A skewness and kurtosis test for each treatment rejected the hypothesis that the risk level follows a normal distribution (at a 10% level). The results for these tests can be found in Table 10 of the Appendix. As a consequence, parametric t tests could not be used to analyze the data. Instead, as is done in the paper by Gneezy and Potters (1997), a non-parametric Mann Whitney test, also known as
the Wilcoxon rank-sum test, tested the differences between treatments. The results of these tests can be found in the Table 4.
TABLE 4: DIFFERENCES BETWEEN TREATMENTS
Treatments Wilcoxon rank-sum
z-value
p-value (one sided)
Individual vs No information peer 0.006 0.9955
Individual vs Information peer 1.109 0.2673
No information peer vs Information peer 1.023 0.3061
Examining the results above, no difference between treatments can be found. All z-values fall within the range of -1.96 and 1.96 as set by an alpha of 0.05. Additionally, all p-values exceed the 0.05 level. Due to the fact that no significant difference was found between the individual and the no information peer treatment (Wilcoxon rank-sum test, p=0.9955), Hypothesis 2 is rejected. When no information exchange is possible, the presence of peers does not influence behavior with respect to financial decision making under risk. Similarly, no difference was found between the individual and the peer treatment where information exchange did occur. When making financial choices under risk, individuals thus do not significantly alter their choices, when they are in the presence of peers and they are provided with information on the risk preferences of these peers (Wilcoxon rank-sum test, p=0.2673) compared to when they make these choices individually. Finally, the choices of the no information peer treatment were compared to those of the information peer treatment. Again, no significant difference was found between these two treatments. (Wilcoxon rank-sum test, p=3061). Contrary to previous research, the exchange of information did not produce a significant alteration in the behavior towards risk. This leads to the rejection of Hypothesis 3. In the presence of peers, individuals do not alter their risk behavior as a consequence of receiving information on the risk preference levels of these peers.
4.3 Differences between Classes
Three classes of thirty students participated in the experiment. It is possible that the average risk attitude per class differed. Some classes may include a higher proportion of risk loving or risk averse individuals than others. Therefore, the difference between classes will be studied in this paragraph.
The results of Wilcoxon rank-sum tests, testing the significance of the difference between classes, can be found below. Tests were performed per treatment. The numbers shown are the p values corresponding to each test.
TABLE 5: DIFFERENCES BETWEEN CLASSES
Treatment Class 1 2 Class 1 3 Class 2 3
Individual 0.0087 0.0000 0.0008
No information peer 0.0095 0.2326 0.1248
Information peer 0.2842 0.9117 0.2272
In the individual round, all classes show significantly different risk behavior compared to one another (Wilcoxon, p=0.0087, p=0.0000 and p=0.0008). Additionally, participants of class 1 and 2 chose significantly different in the no information peer round. For class 1, the risk level in was around -0.003 in the individual treatment and -0.262 in the no info peer treatment. For class 2, the risk level was -0.135 for the individual and -0.054 for the no info peer treatment. On the basis of these numbers it is not possible to conclude that there is a difference in risk attitude between class 1 and 2, since the differences for both rounds do not show a clear pattern. Also, significant differences in risk attitude between classes were not found between the other combinations of treatment and classes. Therefore, conclusions cannot be drawn with respect to the effect of a particular class. It is possible that classes behaved different initially, when decisions were made individually, but that decision makers were affected by the presence of peers in such a way that class differences were taken away. However, we have too little information to support this theory.
4.4 Round Sequence
For every class, rounds were played in a different sequence. An overview of the order of treatments per class can be found in Table 10 in the Appendix. The fact that the order of treatments can be very influential was pointed out in a paper by Harrison et al. (2005) concerning the experiment by Holt and Laury (2002), discussed in section 2.3. To test for order effects, a Wilcoxon rank sum test was performed for every combination of rounds. The results of these tests are displayed in Table 6 below.
TABLE 6: ORDER EFFECTS
Round 1 2 Round 1 3 Round 1 4 Round 2 3 Round 2 4 Round 3 4
p-value 0.0004 0.0001 0.0000 0.9128 0.0827 0.0564
Except for the difference between round 2 and 3, all rounds are proven to be significantly different from one and another. For the difference between round 1 and all other rounds, p-values are lower than 0.01, indicating strong significant differences. For the difference between round 4 and round 2 and 3, p values levels are higher, indicating weak but significant differences at a 10 percent level. Examining the means of all rounds, the risk level decreases from one preceding round to the next. The average risk level is 0.000 for round 1 and -0.180 for round 2, -0.182 for round 3, and -0.283 for round 4.Despite the control for order effects, Table 7 shows that order plays an important role in explaining the risk levels of the participants of the present experiment.
4.5 Influence of Information
In the information peer round, peers were asked to provide the decision maker with advice on what bets to place. It is interesting to examine if decision makers were influenced by the information that was provided to them in this treatment. This poses the question of how many times the decision maker followed the advice that was provided by one or both of the peers that were present during this particular round. To examine this, the following information was gathered. Column 2 of Table 7 displays the number of times that the advice of both peers was discarded by the decision maker. Column 3 displays the number of times that the advice of one peer was followed. Column 4 displays the number of times that the decision maker chose an alternative that took the advices of both peers into account. This can occur in two ways. Primarily, it is possible that both peers provided an advice to choose an alternative with the same risk level. Secondly, it is also possible that the peers provided an advice of two alternatives with different risk levels, but that the decision maker decided to choose an alternative corresponding to a risk level that lies in between the risk levels of the alternatives that were advised by the peers. The third row provides the proportion of the total choices in which the occurrence discussed above took place.
TABLE 7: FOLLOWING OF ADVICE
Not followed # times 1 followed # times 2 followed
# of choices 18 24 14
Proportion 0.32 0.43 0.25
The table above indicates that 68 percent of the decision makers followed at least one of the advises that was provided to them by their peers. And 25 percent followed both of the advises that were provided by their peers. Interestingly, this seems to indicate that a large proportion of individuals were influenced by the information on the risk preferences of their peers. However, as has been indicated in section 4.2, no difference was found between the peer treatment that did entail the sharing of information and the peer treatment that did not.
4.6 Linear Regression
To create a better understanding of the forces at work when an individual is in the presence of peers, three possible influences were examined, namely the influence of the gender of the spectators, the influence of friends as spectators and the influence of the popularity of the decision maker. These influences were examined through linear regression. The questionnaire provided the information necessary to perform these regressions. To begin with, information was gathered on two variables that have been established to have predictive power with respect to risk behavior, namely GENDER and AGE. GENDER and AGE were incorporated as control variables in the linear regression specification. Additionally, behavior towards risk was examined by the variables SMOKING and ALCOHOL consumption, the practice of EXTREMESPORTs and the number of FRACTUREs endured. For a full explanation of these variables, please see Explanation 3 of the Appendix. All of the variables above are predicted to be related to risk behavior in general (see Section 3.6).
The variables discussed below were predicted to influence those individuals that were in the presence of peers. Taking this into account, only those observations that were registered in the peer treatments were used for regression. MALESPEC refers to the amount of male spectators present during the round in which the decision maker was accompanied by two peers. Two dummy variables were created, namely MALESPEC1 and MALESPEC2. MALESPEC1 reads 0 for the presence of no or two male spectators and 1 for the presence of
one male spectator. MALESPEC2 reads 0 for the presence of no or one male spectator and 1 for two male spectators. Not being in the presence of any male spectators served as the reference category. Off course, the influence of males may be different for female compared to male decision makers. Therefore, the linear regression included an two additional interaction term, namely CGENDERxCMALESPEC1 and CGENDERxMALESPEC2. To overcome the possibility of multicollinearity issues arising, concerning the interaction term and the variables GENDER, MALESPEC1 and MALESPEC2, it was necessary to center, i.e. subtract the mean, from these variables before proceeding with linear regression. Uncentered interaction variables lead to an erroneous interpretation of regression results (Robinson & Schumacker, 2009). Therefore, the interaction terms are the products of the centered variables CGENDER and CMALESPEC1 and, CGENDER and CMALESPEC2.
FRIENDSPEC refers to question 9 of the questionnaire and indicates if the spectators that were present during the experiment were friends of the decision maker or not. FRIENDSPEC is a dummy variable and reads 0 for the presence of no friends and 1 for the presence of one or two friends.
POPULARITY is a dummy variable which reads 0 for those individuals who indicated to have less than five friends and 1 for those who indicated to have five or more than five friends among the other members of their class. The linear regression was specified as follows.
RISKi = α + β1GENDERi + β2AGEi + β3SMOKINGi + β4ALCOHOLi + β5EXTREMESPORTi
+ β6RISKWILLINGNESSi + β7MALESPEC1i + β8CGENDERiCMALESPEC1i +
β9MALESPEC2i + β10CGENDERiCMALESPEC1i+ β11FRIENDSPECi + β12POPULARITYi
+ εi
α = constant
β = beta corresponding to the variable i = ith observation
εi = error term
Prior to interpreting the results of this regression, a Variance Inflation Factor test was performed to check for multicollinearity. The results of the VIF test are given in Table 11 of the Appendix. As a rule of thumb, VIF values below 10 are acceptable for linear regression (Hair, 1995). Since the results of the VIF test displayed values lower than 10 for all variables, it could be concluded that no multicollinearity issues were present.
TABLE 8: LINEAR REGRESSION1 Variable α, β p CONSTANT 1.031** 0.016 GENDER2 0.181*** 0.002 AGE 0.086*** 0.001 SMOKING 0.132* 0.092 ALCOHOL 0.240*** 0.000 EXTREMESPORT 0.250 0.674 FRACTURE 0.010 0.553 RISKWIL 0.031 0.124 MALESPEC1 0.008 0.921 CGENDERxCMALESPEC1 0.288 0.133 MALESPEC2 0.071 0.336 CGENDERxCMALESPEC2 0.166 0.337 FRIENDSPEC 0.126** 0.018 POPULARITY 0.096 0.115 Number of observations 162 Adjusted R squared 0.2736 *** significant at a 1% level ** significant at a 5% level * significant at a 10% level
The linear regression results are displayed in Table 8. Both control variables GENDER and AGE produced the expected signs. As was expected, men are more risk seeking than women. On average, men choose alternatives with a risk level of 0.181 points lower than women. For every year increase in AGE, respondents choose alternatives with a 0.086 higher risk level. As was expected, smokers are generally more risk seeking than non-smokers. The SMOKING coefficient indicates that average risk level of a smoker was 0.132 point lower than a non- smoker. ALCOHOL also has predictive power concerning risk behavior. The more standard alcoholic beverages are consumed, the more risk seeking the individual will be, which can be derived from the negative sign of this coefficient, namely -0.240. Nor the practice of
1 I advise readers of this paper to take caution with the results of the linear regression performed in this
section, since the residuals of the linear regression were not normally distributed.
EXTREMESPORTs, nor the amount of FRACTUREs endured in a respondents lifetime proved to have predictive power concerning risk behavior. Respondents were not able to correctly estimate their own risk behavior by means of the general risk question (Question 5 of the questionnaire), indicated by the insignificance of the coefficient for RISKWIL.
The remaining part of this section will examine the possible aspects that may influence an individual when financial choices are made in the presence of peers. The coefficient for MALESPEC1 and MALESPEC2 are insignificant. Not taking the gender of the decision maker into account, the average respondent was not influenced by the presence of one or two males as spectators. Similarly, the coefficient of the interaction terms CGENDERxCMALESPEC1 and CGENDERxMALESPEC2 are insignificant. Taking gender into account, this means that neither female nor male decision makers alter their behavior in relation to the gender of the spectators who are present while they make decisions.
Finally, the influence of befriended spectators was examined by FRIENDSPEC. Interestingly, the variable is significant, indicating that friends influence the risk behavior of individuals. If there are one or two friends present during the decision making process, respondents choose alternatives with a 0.126 point lower risk level than when there were no friends present. Decision makers who were considered popular did not choose significantly different from less popular individuals, as can be evaluated from the insignificant coefficient of POPULARITY.
5. Discussion and Conclusion
This paper discusses the results of a field experiment which examines financial decision making under risk in a social context. A simple gambling game was developed, which provided a set of alternatives which were classified on the basis of their corresponding relative risk aversion ranges. Students were asked to make decisions alone and in the presence of two peers (or spectators). Additionally, in one of the in total four rounds, the risk preferences of the spectators were provided to the decision maker, to examine the influence of information.
No significant difference was found between the risk behavior of a decision maker who made decisions alone compared to a decision maker in the presence of two peers. According to the results produced by the present experiment, people do not behave significantly different in the presence of peers than individually, with respect to financial
decision making under risk. On average, participants behaved risk seeking. Information on the risk preference levels of peers did not significantly alter the behavior of the participants.
However, in light of the results above, a few remarks are in order. The risk seeking
average of the sample population requires additional analysis. Namely, this phenomenon can be the result of the wide range of relative risk aversion corresponding to a 200 eurocent bet on a symbol, the alternative that was most frequently chosen. Due to the wide range of this alternative and the assumption that the mean of this range represents the average risk attitude of the individual who chose this alternative, risk neutral or even slightly risk averse individuals could have been falsely classified as risk seeking. Secondly, it could also be the result of the repetitive, cumbersome task that participants were asked to perform which could have led to the order effects that were found be significant. Participants played increasingly less risk averse as the experiment proceeded. The data may thus not represent the true risk attitudes of the participants. Instead, the search for excitement could have contributed to a negative sign for the mean risk aversion level. Finally, the fact that not all participants were paid out their earnings could have played a role. Hypothetical and low payoff schemes have produced erratic decision making behavior (Holt and Laury, 2002). Recommendations to overcome these shortcomings are discussed in the next section.
The design of the experiment contributes to existing elicitation methods of risk aversion. Interestingly, it combined the use of the relative risk aversion ranges as in Holt and Laury (2002) and the simplicity of the method by Gneezy and Potters (1997). The experiment is simplified compared to the MPL method, by providing a more comprehendible choice, which requires little mathematical ability of its respondents. Furthermore, it broadens the scope of risk levels compared to the experiment by Gneezy and Potters by providing a choice between two lotteries. As a consequence, the amount of risk attitudes that can be analyzed includes a highly risk seeking segment.
Under the influence of peers, known predictors of risk behavior were confirmed by the linear regression performed in section 4.6. As has been found in previous experiments, men behave less risk averse than women. Age was positively related to the risk level, meaning that the older the individual, the more risk averse the individual will be. Smokers were found to be more risk seeking than non-smokers and alcohol consumption was negatively related to the risk level.
No significant relationship was found between financial risk and the practice of extreme sports or the amount of bone fractures endured by an individual. For the practice of extreme sports, it may be more important to take into account the safety precautions that an
individual takes before engaging in an extreme sport. Those taking sufficient safety precautions may not be as risk seeking as one would expect. Possibly, bone fractures are more related to the bone quality of an individual (Seeman et al. 2006) and less to the behavior of the individual.
The behavior of the decision maker was significantly affected by those spectators that the decision maker considered to be friends. This is an interesting result for organizations since it gives insight into the influence of friends within teams. Team composition may be adjusted accordingly, to achieve the risk attitude that an organization aims to pursue.
The gender of the spectator did not influence the behavior of participants nor did the popularity of the decision maker. This is also an interesting result for organizations since it can it contributes to the understanding of team mechanisms. Taking this result into account, organizations can focus their attention on the gender of the decision maker only, since the decision maker is not affected by the gender of the peers that will be present during the decision making process.
Finally, the popularity of a decision maker has no predictive power when it concerns decision making in the presence of peers. It is possible to that this variable would have produced a significant coefficient if one would propose a more specific question asking the exact number of classmates that the respondent considered friends. However, in my opinion, this poses ethical issues since it would require an respondent with no friends to note a confronting zero.
The final section includes some limitations of the present study and directions for further research to overcome them.
6. Limitations and Directions for Further Research
The study above is bound to a couple of limitations. Primarily, the alternatives available to the participants of the experiment in this paper are insufficient in facilitating a precise classification of risk attitudes. This applies especially to the risk seeking segment, in which the two possible alternatives, 1. 200 eurocent bet on a symbol, 2. 200 eurocent bet on a number or image, are comprised out of extensive ranges of relative risk aversion. To enhance the design of this experiment it is necessary to provide additional choices in this segment. These choices will have to correspond to narrow ranges of relative risk aversion, such as those that can be found in the risk aversion segment. Hereby, the scope of risk attitudes that
can be analyzed is increased and risk classification will become more precise. Additionally, it will enable the researcher to isolate risk neutral from risk seeking individuals.
The erratic behavior that was found throughout all treatments of the experiment (see
Table 3) may have been caused by the fact that not all individuals were paid their earnings. In order to take away the possible bias that may have been caused by not paying all individuals, it is recommended that the experiment is replicated. Payoffs should be real and sufficiently high to generate the seriousness towards the task, necessary for the participants to present their true risk attitudes.
A large percentage of the participants of the present experiment followed the advice of at least one peer in the information peer treatment. However, no difference was found between the treatment that did and the treatment that did not entail information exchange. Since the sample size of this treatment was fairly small, further research may investigate the issue by examining the importance of information in peer environments with a larger sample.
Furthermore, as is indicated in the conclusion above, the behavior of the decision maker was significantly affected by spectators that were considered friends of the decision maker. Further research could focus on the impact of friends specially to confirm this finding, taking into account that the gender of the decision maker possibly affects this impact.
Finally, I acknowledge the limited external validity of the experiment discussed in this paper due to the specificity of the sample population. For further research, it is interesting to carry out the experiment with different sample populations. Groups varying in social background, nationality, age and educational level might have different attitudes towards risk and react differently to the presence of peers.
Thank You
I would like to thank my thesis supervisor, Jereoen van de Ven, for his patience and guidance throughout the process of writing my thesis. Also, I would like to thank the Rijnlands Lyceum Wassenaar for allowing me to use their facilities and students to perform my experiment and, in particular, Mrs. Rietmeijer, Mrs. Slooter and Mr. Stuurman for their help prior to and during the experiment.
