• No results found

Decomposing the effect of social interaction on risk behaviour : discussion and social facilitation

N/A
N/A
Protected

Academic year: 2021

Share "Decomposing the effect of social interaction on risk behaviour : discussion and social facilitation"

Copied!
35
0
0

Bezig met laden.... (Bekijk nu de volledige tekst)

Hele tekst

(1)

Decomposing the Effect of Social

Interaction on Risk Behaviour

Discussion and Social facilitation

University of Amsterdam

Faculty of Business and Economics Master Thesis Business Economics Organisation Economics

Tessa Snels 6135552

(2)

1

Introduction

Social interaction is likely to influence risky decisions, but in many studies it is assumed to be constant in a social context. If your risk preference is complete, interaction should not influence your decisions. Someone who has completely made up his mind about all possible risky decisions, is not open to suggestions and other influences. He will never deviate from his preference, no matter the circumstances. Literature however shows that this is not always true, to which I will come back later. But why is this important?

There are many situations where you interact with others where your risk preference is important. For example an organisation is full of people that interact and take risky decisions on a daily basis. Previous research has shown that risk taking in organisations is affected by, for example, the extent of diversification(Hoskisson, Hitt, & Hill, 1991), ownership structure and firm size (García-Marco & Robles-Fernández, 2008). The social aspects of an organisation, such as interaction with colleagues or the mere presence of others, can also affect risk taking. The conclusions of this study can thus contribute to explaining risk taking behaviour in organisations.

From the literature I identify social facilitation and discussion as the two factors that may play a role in this process. Social facilitation refers to the effect of the mere presence of others. So a change in behaviour when someone is watched by an audience or when he does a task alongside another subject. Discussing a problem can help a subject to familiarize with the problem and provides him with the opinions of others. The two effects are translated into two treatments: a public decision in which two subjects see each other’s decisions and communication in which the pairs get 30 seconds to deliberate. The treatments are implemented in an experiment inspired on the card game blackjack. The subjects want to maximise the total value of their cards, without exceeding the critical value of 14. Risk taking is measured by the probability of exceeding the critical value based on the actions of the subject. In this experiment I hope to answer the following research question:

How does social interaction influence risk behaviour and how does it differ under communication and private or public decisions

Although there has been some research on the effect of a social context on risky behaviour, most of this has been done by psychologists. So in most studies there were no monetary incentives to motivate the subjects. In my experiment, subjects receive performance

(3)

2

pay, which should reveal their true preferences and behaviour. Furthermore, to my knowledge there is no previous study that tries to filter out the separate effects.

The results point out that communication has a significant, negative effect of about 7 percent points on the exceedance probability. A public decision has a negative, but insignificant effect of about 4.5 percent points. I find that different aspects of a subject and a partner are related to the results. Whereas friendship and gender of the subject matter for the effect of communication, the public decision effect is affected by the risk preference, popularity and the gender of the partner. The effect of all treatments is quite similar, so there was no significant difference between the actions in part two if you control for the actions in part one.

The rest of the study is structured as follows. First the most important literature is summarized in the next section. Then the setup of the experiment and the hypotheses are explained in the methodology section, section three. Then the results are presented in section four and the conclusions from these results are discussed in section five. The most important documents concerning the experiment can be found in the appendices.

(4)

3

Related Literature

The famous article by Kahneman and Tversky (1979) on prospect theory first theorized the effect of risk attitude on behaviour in risky situations and dilemmas. It relaxed the idea that people are completely rational and base their decisions on expected utility. They conclude that people tend to behave risk seeking for otherwise sure losses and tend to avoid risk when the alternative is a sure gain. They manage to construct a model that includes the effect of risk on people’s behaviour, but include this risk attitude as a “black box”, a factor that is not really explained.

The idea that people may have different attitudes concerning risk is quite obvious, but before you can incorporate this into a model, you should have a proper way to measure risk attitude. Different measures have been used to estimate risk attitude. Among those are lottery choices (Binswanger, 1980), overbidding in auctions (Smith and Walker, 1993) and the difference between buying and selling prices for lotteries (Kachelmeier and Shehata, 1992). Holt and Laury (2005) develop a method that objectively attaches an actual number to the level of risk aversion. They use a lottery choice menu where subjects can either choose the safe or the risky lottery. For each choice, the probabilities differ, shifting from a high probability of the lower outcome to a high probability of the higher outcome. When there is a high probability that you will win the low outcome, most people choose the safe lottery and when the probability of the high outcome is high, most people choose the risky lottery. So the point where one switches from the safe to the riskier lottery, determines his risk attitude.

Most estimates of risk attitude are measured in a socially neutral context. That is, subjects make their decisions privately and typically do not get any advice or feedback from other subjects. However, in everyday life, you are likely to experience a context that is everything but socially neutral. If you consider a rational person, this should not affect his behaviour, but there is evidence that it does.

Cettolin and Riedl (2010) investigate if preferences concerning decision making under uncertainty are incomplete. They ran an experiment in which subject have to choose between a risky and an ambiguous project, or state that they are indifferent. They find that half of the subjects delegate multiple decisions, which shows that preferences must be incomplete. Although the setup of this experiment is very different from mine, the conclusion that people can have incomplete preferences in risky decisions is interesting. If preferences were

(5)

4

complete, subjects would not change their behaviour once they get the chance to communicate. Therefore, it is interesting to see if subjects take advice to heart or not.

Bem, Wallach and Kogan (1965) were one of the first to discover that there is a shift toward greater risk taking when people act in a group as opposed to their behaviour as individuals. This is called the risky shift. They also find that discussion and reaching a consensus is important for this shift to occur. Otherwise, the subjects behave more conservatively and choose less risky options. The authors explain this finding as the result of responsibility diffusion, because people apparently need the approval before they take the extra risk. After all, their decision affects other group members as well. Wallach and Kogan (1965) further investigate information about risk-taking levels of peers, discussion and consensus as possible explanations for the risky shift. They find that discussion, either with or without consensus, is most important to establish the risky shift. Note that the risky shift phenomenon is about risk taking in a group, where subjects are partly responsible for each other’s outcome, which is not the case in my experiment. Since my subjects will only be responsible for their own outcomes, responsibility diffusion should not play a role. The same reasoning goes for reciprocity and social pressure.

Another possible explanation is complying with a perceived cultural norm, which means that people will take more risk if they think others will too, but will behave more conservatively when others are expected to do so as well (Eliaz, Ray & Razin, 2005). This explanation could apply to my experiment, since a subject might consider what others would do, if they are going to see his decisions.

Among the articles that assess group influence on individual risk taking, where subjects could only influence their own results, some find greater risk taking. For example Lamm (1967), who found that hearing a group discussion about a risky dilemma is enough to cause a risky shift. A problem with this article is that the dilemmas are hypothetical and that the subjects do not get paid based on performance. This may cause subjects to act more risk loving, because they don’t have real money at stake. Furthermore, just as in the experiment of Wallach and Kogan (1965), group discussion leads to increased risk taking. Bateson (1966) states that this effect could also be attributed to familiarization with the problem and finds evidence that this is probably true. Subjects got the opportunity to familiarize with the problem, either alone or in a group, and the authors find that the risky shift for both treatments is about the same. Roberts and Castore (1972) however find that the tone of the discussion is very important for predicting the level of risk taking. In their experiment, subjects had to

(6)

5

watch a videotape in which actors had a scripted discussion about the problems. There were four versions of the videotape, the tone of the discussion varying from in favour of risk to in

favour of caution. The authors find that subjects not only conform to the tone of the

discussion, but also are more certain about their choice than before watching the discussion. We can conclude from this that discussions are helpful to familiarize with the problem, but that this does not always lead to a risky shift. Because of conformity processes, people base their decisions on the tone of the discussion.

Zajonc, Wolosin, Wolosin and Wallace (1970) consider imitation and social facilitation as two other explanations. Especially the second term, social facilitation, requires some explanation. Zajonc (1965) introduces the term social facilitation for the tendency of people to perform better at simple tasks in the presence of others, but perform worse for more difficult tasks. The presence of others causes arousal and increases the drive to perform well. This works out well for simple tasks, but for more difficult tasks, it causes you to “try too hard”. Zajonc and his co-authors want to investigate what social facilitation does to risk taking. The idea is quite understandable, if others can see your decisions and results, you might want to make an impression by taking extra risk.

In the social facilitation treatments, subjects were responsible for their own results and thus were not influenced by others. They were for example viewed by an audience or they worked alongside other subjects. They find that, in those circumstances, subjects are not influenced by others. They do not take more risk and also do not imitate the other’s actions. In the group treatment however, subjects significantly take on more risk. This would mean that the public decision treatment in my experiment would not differ from the control treatment. I do have some problems with the setup of their experiment. The stakes per decisions were very low: ranging from 0 to 1,5 cents. But since there was an effect in the group treatment, it may have been enough. Furthermore, there was a self-selection problem in the audience treatment, because the subjects had to give permission for the presence of an audience. Therefore, in this group there may have been relatively more individuals who are not easily influenced by an audience. Finally, the social facilitation theory (Zajonc, 1965) shows that the effect of social facilitation can differ for different tasks, so this may also be the case for risky tasks. In the experiment of Zajonc et al (1970), the participants have to bet which light on the table will light up, which is not a very exciting task. If the effect of social facilitation is driven by the impression you make on others, the nature of the task can really make a difference. To name an extreme example, if you get the opportunity to go bungee jumping, you are probably more

(7)

6

likely to pass if you are alone than if you are surrounded by a group of people. Compared to this example, betting an amount of money on a light has less effect on your status. Therefore, for a more exciting task, the effect of social facilitation should be clearer.

Rockloff and Dyer (2007) study a more exciting risky task: gambling. They investigate if social facilitation can influence gamblers’ behaviour. Some players received information about the results and actions of other (fake) players through messages, bells and lights. These players place more bets and lose more money, which indicates that they are prone to social facilitation. This article serves as an example that this can make a difference for an exciting risky task as gambling.

Gardner and Steinberg (2005) let participants perform risky tasks on their own, but in the presence of a peer group. This peer group is allowed to give advice and can see every choice of the subject. They find that, especially adolescents, tend to take more risk in the presence of a peer group. A weak point of the article is that participants are not rewarded for performance, so again, there is less at stake. What is also noteworthy, is that the subjects within the peer groups know each other. The participants all brought two friends who served as their peer group. This is important, because friends can exert more pressure than an acquaintance (McPhee, 1996). Furthermore, friends and relatives trust each other more and therefore sort into the same risk pooling group, according to Attanasio, Barr, Cardenas, Genicot and Meghir (2012). The setup of their article is very different, because payoffs depend on other subjects as well, but it is likely that you rather take advice from someone you trust. In my experiment, all participants know each other well, but I also include data on friendships in the analysis.

A nice experiment by Blascovich, Ginsburg and Howe (1975) observes subjects in a casino setting where they play blackjack. All players start in an individual game, after which some proceed to a group game and some stay in the individual setting. They find that subjects take on more risk in the group setting then in the initial individual setting. They use the height of the bet as the risk measure, but do not use card choices in their analysis. A very strong point of the article is that the participants play with their own money. However, I think that blackjack has many strategic aspects for which the authors cannot and do not control, such as the hand of the dealer, counting cards and the opportunity to split the hand. That is why I chose to simplify the game, which I will explain in more detail in the experimental design.

There are a few aspects in which I hope to contribute to the existing literature. First of all, most of the research on this topic has been done by psychologists, who typically make less

(8)

7

use of monetary incentives. In my experiment, the participants are paid for performance and thus have real monetary incentives. Secondly, I identified social facilitation and discussion as two of the most interesting factors that could influence individual risk taking and it would be interesting to see the effects of both. To my knowledge, there is no article that uses both with the goal to filter out the separate effects. The first factor will be implemented in the experiment in the form of a public decision. In a treatment with a public decision, participants can see each other’s decisions and results and thus act as an audience and a co-actor. The second factor will be implemented in the form of 30 seconds to deliberate before making a decision. So in a treatment with communication, the participants get the opportunity to shine their light on the other’s dilemma and vice versa. Thirdly, to assess the effect of social facilitation correctly, I use a more exciting task, a game based on blackjack. Additionally, my subjects are all acquaintances or friends, which will probably affect the social facilitation effect as well.

There are also some factors that could influence my results. In the rounds with interaction, the subjects can see each other’s dilemmas and if this influences their behaviour, this could bias the results. Rohde and Rohde (2011) investigate if risk attitudes are affected by the risk the others face. They find that this is not the case, so that is good news. Furthermore, we should keep in mind my participants, who are adolescents, react differently than adults. They typically take more risk and are more sensitive to the presence of peer groups (Gardner & Steinberg, 2005).

(9)

8

Methodology

Subjects

I ran an experiment at my high school, the Alberdingk Thijm College in Hilversum. I got permission to work with three classes, a havo 4 class with 26 students, a vwo 4 class with 20 students and a vwo 5 class with 16 students. This makes a total of 64 subjects, who all follow an economics course.

The Game

The game is inspired on blackjack, but is much simpler. The subject receives a card and has to decide if he wants extra cards and if so, how many. The goal is to maximise the total value of the cards, but if this total exceeds 14, the value is 0. Table 2 summarizes the value of each card.

Card Ace 2 3 4 5 6 7 8 9 10 Jack Queen King

Value 1 2 3 4 5 6 7 8 9 10 11 12 13

Table 1 - Card values

After the decisions are made, the experimenters provide the requested number of cards, which reveals the result of their decisions. For example, a subject first receives a four, which makes him decide to request one additional card. This card turns out to be an eight, so the total score of this subject is twelve.

With this setup, for almost each card, there is both the chance to remain beneath or to exceed 14. Therefore, the extremely risk averse person would settle for even the lowest cards and the extremely risk loving person would request an extra card for even the highest cards.

Goal of the Experiment

In this experiment, I want to filter out the effects of separate group influences on individual risk taking. In the previous chapter, I define social facilitation and discussion as the concepts of interest, which are implemented in the form of a public decision and communication, respectively. First, all subjects define their full preference for all possible card values from Ace to King in the first round. In this round, there is no communication and the subjects will make private decisions. This will serve as the control treatment. In the second round, each class will participate in one of the tree different treatments: one with communication, one with a public decision and one with both communication and a public decision.

(10)

9

Private decision Public decision

No communication Control treatment – all classes Havo 4 (26 students)

Communication Vwo 5 (16 students) Vwo 4 (20 students)

Table 2 - Treatments

By doing a within-subjects comparison of the first and second round, I am able to see if the subjects deviate from their preference in the treatment round. By doing a between-subjects comparison of the three treatment groups I am able to see the effect of either a public decision or communication, keeping the other factor constant.

In the previous example, the subject only received one card at the beginning of the experiment. In the experiment, subjects always receive multiple cards at the beginning of the round. These are the initial cards, on which they base their decisions. In the control round, subjects receive all thirteen card values and receive three in the treatment rounds. The cards do not form a set however, the subject makes as many decisions as there are cards. So if he receives three cards, he decides for each individual card if he wants extra cards or not.

Next the experimenters provide the requested cards. These are the additional cards, which can also be more than one, depending on the decisions of the subject. The additional cards belong to the corresponding initial cards. For the analysis, the most important information is about the value of the initial cards and the number of additional cards.

Procedures

Before each class, I prepare the classroom such that there is enough space between the tables and all the necessary documents are distributed. After all subjects are seated randomly, I read the instructions (see Appendix 1) out loud and answer questions, if any. If everything is clear, the first round starts

In the first round, the subjects have to decide for all 13 possible card values if they want additional cards. There are a few versions of the decision form, each with the cards in a different order (see Appendix 2). This provides me with a complete overview of their preferences. After that, I provide the requested number of cards, which are dealt from a very large set of cards, and this concludes the first round.

In the second round, the subjects receive new instructions (see Appendix 3.1-3.3). After that, the subjects pair up and receive three cards each. In a treatment with communication, the pairs now get 30 seconds to deliberate on what to do next. The subjects then decide for each card if they want additional cards (see Appendix 4.1 & 4.2). Afterward,

(11)

10

the requested number of cards are distributed. When the treatment includes a public decision, the pairs face each other when they make their decisions and when the cards are distributed.

Throughout the experiment, the teacher and I took note of all cards and decisions, but I had to make sure that the presence of the experimenters would not influence the results. Therefore, the decision form in the first round can be folded such that the decisions are visible, but not the cards to which they belong. In the second round, we each first took note in one half of the class and then switched, so that neither of us ever saw both the cards and the decision of one subject.

After the second round, I collected extra data through a questionnaire (see Appendix 5). Some questions are about other types of risky behaviour in real life, such as dangerous sports, offences, smoking and drinking. Other questions regard friendships and popularity.

Monetary incentives

In each experiment, five students were selected for payment. For these students, I randomly selected one initial card from the experiment and calculated their payoff based on the decisions they made for this card. As mentioned before, the critical value which the students should not exceed is 14. Furthermore, each point is worth €1, so they could earn up to €14. For high school students, earning up to €14 in an hour is a lot. Unfortunately I could not pay all students, but I think the odds of being selected for payment were high enough for the students to be motivated. When the students left the classroom, they all received a closed envelope, either empty or with their wins. The payment procedure was included in the instructions, so they all knew this before the experiment started.

Hypotheses

Since the goal is to extract the separate group effects on individual risk taking, the hypotheses are formed for each step from a social neutral situation to a situation with both discussion and social facilitation. This structure is depicted in the figure below.

H1

Control treatment H2

Communication H3, H4,

Communication & Public decision

H3

Public decision

Figure 1 – Structure of the hypotheses

Hypothesis 1: The subjects who take a private decision, take more risk when they communicate before making a decision.

Even though the evidence on the risky shift is mixed, many articles find that discussion contributes to the risky shift (e.g. Wallach and Kogan, 1965). Discussing the cards can also

(12)

11

help to familiarize with the problem, which typically leads to increased risk taking Bateson, 1966). He found that subjects are cautious and play it safe at first, but dare to take more risk when they know more about the problem and understand the risks. Therefore, I expect more risk taking when subjects can communicate, given that they make a private decision. To test this hypothesis, a within-subjects comparison is used for the class that did the communication treatment.

Hypothesis 2: The subjects who do not communicate, take more risk when they make a public decision than when they make a private decision.

I think the game in my experiment is quite exciting and all subjects know each other quite well. Friends or acquaintances typically have more influence on someone’s actions than strangers (Gardner & Steinberg, 2005; Mcphee, 1996). This combined with a gambling game, which had an effect in Rockloff and Dyer (2007), will probably contribute to increased risk taking for public decisions. Again, a within-subjects comparison is used, but this time for the class that got the public decision treatment.

Hypothesis 3: The subjects who both communicate and take a public decision take more risk than the subjects who do either one or the other.

There is not a strong foundation of literature on which I can base this hypothesis, but I expect that the positive effects of the first two hypotheses to complement each other.

Hypothesis 4: The subjects who communicate and take a public decision, tend to follow advice more often than subjects who communicate and take a private decision.

While ignoring advice is no problem if someone takes a private decision, subjects might feel more pressure to follow to the advice, when the partner can see his actions. If this is true, than people are prone to letting that pressure influence their decisions. Whereas learning from a discussion might be a good reason to change your standpoint, pressure is not a good reason. If your preferences are incomplete, there is no problem with taking advice. Taking advice because you want to please the other, even though you do not think it is a good idea, can be problematic. This hypothesis is tested with a between-subject comparison with the classes that do the communication treatment and the treatment that includes communication and a public decision.

(13)

12

Hypothesis 5: The subjects who are affected by the treatments are also more likely to smoke, drink excessively, have committed an offence and to have gotten themselves into a hospital.

The questionnaire includes some questions about risky behaviour and the four listed in the hypothesis are all events or habits that are not sensible, but can occur because of peer pressure, because people want to fit in or want to make an impression. Those factors are related to how receptive someone is to act differently when they make decisions in a social context.

(14)

13

Results

In this section the most important results are summarized. First the main variable is explained and the basic results are presented to get a clear picture of the data. Then the hypotheses are tested, along with some additional tests, where necessary.

Variables

The variable of interest for this study is risk behaviour. The measure should incorporate some important features. First is the goal of the game for the subjects. In the game, the subjects want to maximise the total value of their cards, but it should not exceed 14, because then the payment drops to €0. The risky aspect of the game is thus not to exceed the critical value. Second is the fact that the cards differ in each round and for each subject in part two. Since they receive three initial cards from the thirteen possible card values (Ace to King), it is inevitable that they differ among the subjects. Therefore a simple measure that compares the number of cards requested between the two parts is not sufficient, because actions depend much on the initial cards.

This would not be a big problem if the distribution among the three treatments is about the same, but figure two suggests that this is not the case, even though the cards are distributed randomly. The bars show how many times the subjects received a certain initial card for all treatments. The blue bars represent the ideal situation where all cards have the same frequency. It is clear that quite a few cards deviate from that situation. For example, in the communication class, the low cards are overrepresented and the high cards are underrepresented. For the class with both, the cards oscillate somewhat around the ideal percentage, but have a large spread. The consequence of this is that the classes did not have the same opportunities to take more or less risk. For example, requesting extra cards for low card values is less “dangerous” than for high card values.

Figure 2 – Distribution of the initial cards in part two 0% 2% 4% 6% 8% 10% 12% A 2 3 4 5 6 7 8 9 10 J Q K

Card distribution in part 2

(15)

14

Exceedance probability

Considering the important features described before, the probability that a subject would exceed the critical value, given his initial cards and his decisions, is a suitable measure. In this game, risk can be seen as the probability that the total value of the set of cards will exceed 14, the critical value. Because this probability differs for card values and actions, it is more accurate in describing how much more or less risk the subject took. Another advantage of this measure is that it takes the magnitude of a deviation into account. The figure below depicts the probabilities for each initial card value and the corresponding action. What should be kept in mind is that people are not able to calculate all these figures on the spot and thus do not base their decisions perfectly on these probabilities.

Figure 3 – Probability that total card value will exceed 14 for each initial card value

Summary Statistics

Observations

The total number of participants is 62, but a few of them either completely misunderstood the game, or did not really care. This was clear from the decisions they made. The number of requested cards typically declines when the value of the cards increases, but these students picked random numbers. There was no clear order in the numbers and they were also much higher. The data from these participants was completely random and thus useless, so I had to exclude them from my analysis. This resulted in 58 remaining participants. The average age of the participants is 16.45 and exactly 50% is male.

As explained before, the risk measure is based on sets of cards. The subjects play one round in part one and three rounds in part two. Within the rounds, the subjects make multiple decisions, one for each initial card, which are 13 in part one and 3 in part two. These are no individual observations, so the risk measure is an average of the 13 or 3 observations in each

0 0,2 0,4 0,6 0,8 1

0 cards 1 card 2 cards 3 cards 4 cards 5 cards

Pr

ob

ab

ili

ty

Number of cards requested

Exceedance probability

Ace

2 3 4 5 6 7 8 9 10 Jack Queen King

(16)

15

round. The three rounds in part two are all separate observations, since subjects interact with other partners in each round. This results in four observations per subject in total, or three observations for the differences. Unfortunately I ran out of time with one class, so I could only do two rounds in part two. This results in 209 or 151 observations, depending on the variable used.

Risk taking in part one

The first thing that is interesting to see, is if the decisions made make any sense. You would expect to see a declining number of additional cards, as the value of the initial card increases. Furthermore, by calculating the expected value of one and zero additional cards, you can see what decision results in a higher expected value for each card. For cards higher than seven, the expected value of settling for your initial cards is higher than the expected value of requesting additional cards. The figure below depicts the average number of cards requested for each card value in part one. The number of cards indeed decreases with the value of the initial cards, but is typically higher than would be sensible.

Figure 4 – Average number of additional cards for each initial card value

The decisions of the subjects result in the measure for risk explained before. To give an impression of the variable, the main statistics of the risk attitude in the first part are summarized in the table below. In part one, the subjects had on average 30.63% chance to exceed 14. This was 25.42% on average in part two. This suggests that subjects took less risk in the second part, about 5 percent points.

Mean Standard

Deviation

Minimum Maximum

Exceedance probability – Part 1

0.3063 0.1536 0.0533 0.8313

Exceedance probability – Part 2 0.2542 0.1845 0 0.955 0 0,5 1 1,5 2 2,5

Ace 2 3 4 5 6 7 8 9 10 Jack Queen King

Average number of cards requested

Part one - Control

(17)

16

Table 3 – Main statistics for the exceedance probability measure in part one and two.

The main interest of this research is about the difference in risk taking between the treatments. Figure five summarizes the exceedance probability data for all three treatments in part one and part two. For the data in part one I only use the initial cards that were used in part two for each subject and partner to ensure a fair comparison. The difference in behaviour in part one is noteworthy, because the experiment was exactly the same for everyone in this part and thus should be the same. This difference is still present when all data is used. There must be some heterogeneity between the classes and I will control for this by using the difference between the two parts, instead of comparing the behaviour in part two. The measure shows that risk taking declined in all three treatments. The change does not seem to differ much, but we will look at this more closely when testing the hypotheses.

Figure 5 – Exceedance probability for all treatments separately

Finally I tested the distribution of the data and this was about normal, so I can use t-tests and OLS to test the hypotheses.

Testing the hypotheses

The previous results are useful to get a clear picture of the data. We have seen that the average difference between the risk attitude measures from part one and part two is negative, indicating that subjects take less risk in the treatments than in the control treatment. I use a two-sided t-test to see if the average is significantly different from zero. The exceedance probability in part one is significantly different from part two with a p-value of 0.011. This would mean that social interaction results in less risk taking, which is counterintuitive, so there may be something else at play. A learning effect could explain the results. The subjects were very enthusiastic while filling in the decision form for the first part, but when the cards

0 0,1 0,2 0,3 0,4 0,5

V5 - Communication H4 - Public decision V4 - Both

Ex ee da nc e pr oba bi lit y

Exceedance probability

Part 1 vs Part 2

Part 1 Part 2

(18)

17

were dealt, they could see that many of their choices resulted in totals that exceeded 14. I could see that subjects did not expect this and thought they were going to get a good score. This undesirable outcome might have caused them to be more cautious in the following rounds.

Figure five indicated that the effect of introducing the treatment was different for the three classes. By testing the hypotheses it is possible to see what these effects are. But since I cannot be sure that the findings are consistent with real life behaviour, the within-subject comparisons for the first two hypotheses should be interpreted with caution.

Hypothesis 1: The subjects who make a private decision, take more risk when they communicate before making a decision.

I compare the data of part one and part two for the class that did the communication treatment in part two. A t-test is used to see if the subjects take more risk in part two. From the exceedance probability difference we can conclude that this is very unlikely. The average level in part one is 0.3173 and is 0.2487 in part two, so the exposure to risk decreases with about 7 percent points. With a p-value of 0.9552 I can reject that this level in part two is actually higher than part one. I can therefore reject hypothesis one, subjects do not take more risk when they can communicate before making a decision.

Hypothesis 2: The subjects who do not communicate, take more risk when they make a public decision than when they make a private decision.

I compare the data of part one and part two for the class that did the public decision treatment in part two. Again, a t-test is used to see if subjects take more risk in part two. The averages of 0.3446 in part one and 0.2979 are not significantly different from each other with a p-value of 0.3161. The odds that this difference of about minus 4.5 percent points is positive (p=0.820) are still very low, but is less conclusive than in the communication treatment.

It is difficult to correctly assess these outcomes. I assume a learning effect, but I cannot test if this is actually true. Another possible explanation for the negative result for the first hypothesis is that the tone of the discussions was risk averse. I cannot control for this and as Roberts and Castore (1972) concluded, a risk averse tone of a discussion leads to more risk averse decisions. There is also a chance that the familiarization effect worked the other way around here. If the subjects experienced a learning effect after the first round, talking about it might have amplified that effect.

The within-subjects comparison for the class with both treatments is not used to test any of the hypotheses, but can be used to shine some extra light on the issue. The statistics

(19)

18

indicate that risk taking is significantly lower in the second part, just like in the communication treatment. It is thus possible that especially communication has contributed to the cautious shift. What should be noted is that some students could not resist to deliberate quietly when there was something that was unclear to them or if they had to wait, even though I specifically asked them not to. If communicating has a negative effect on risk taking, this could have shifted the results in the public decision treatment downward. All t-tests are tabulated below.

Table 4 – T-tests comparing the Exceedance Probability in Part 1 and Part 2

In the previous section, the plain differences were tested to answer the hypotheses. However, there may be other factors at play that influence risk taking in the second part. The risk attitude of the partner may influence the tone of the conversation and thereby influence your decisions (Roberts and Castore, 1972). We have also seen that friends have greater influence on risky decisions than acquaintances (McPhee, 1996). In high school, popularity is important to students. In the questionnaire, the subjects were asked who their five best friends in that class are. With the data I constructed a popularity measure, namely the number of times a subject was listed as a friend. A partner that was named many times, who is thus popular, may have a greater influence on the actions of a subject. Furthermore, men are known to take more risk than women (Byrnes, Miller, & Schafer, 1999), so they may also react differently to communication or a public decision than women. Alpha male feelings may be stronger when he interacts with classmates. Finally, the gender of the partner may have an effect on risky behaviour. Possibly because someone wants to impress a partner of the opposite sex. Or because the advice from a specific gender is more trusted, it has more effect. All these effects are included in two regressions on the difference in exceedance probability. Both regressions are tabulated in table 5.

Mean Standard Error P-value Mean or difference ≠ 0 P-value Mean or difference > 0 Vwo 5 Communication 1 0.3173 0.0319 0.0895 0.9552 2 0.2487 0.0240 Havo 4 Public decision 1 0.3446 0.0314 0.3161 0.8420 2 0.2979 0.0339 Vwo 4

Communication and Public

1 0.2791 0.0200

0.0721 0.9639 2 0.2234 0.0233

(20)

19

Communication treatment Public decision treatment

Variable Coefficient P-value Coefficient P-value

Exceedance Probability of the Partner

-0.0509 0.720 -0.5482 0.053*

Friendship between Subject and Partner

0.1008 0.040** 0.0375 0.606

Popularity of the Partner 0.0258 0.171 0.0627 0.062*

Male -0.1143 0.019** -0.1125 0.153

Male - Partner 0.0115 0.832 0.1652 0.073*

Constant -0.1265 0.085* -0.1283 0.479

R-squared 0.2535 0.2565

Adjusted R-squared 0.1646 0.1636

Table 5 – Regression of difference in exceedance probability for the communication treatment and public decision treatment. * = significant at the 10% level, ** = significant at the 5% level.

We know from the t-tests that the average subject in the communication treatment decreases his probability to exceed 14 by about 7 percent points. But if the partner was a friend of the subject, he takes 10 percent points more risk than the pairs that were not friends, on average. Being a man however contributes to a cautious shift. Maybe this happened because women were already more cautious than men, so to get to the same “safe haven”, men had a longer way to go. The other variables have no significant relation to the risk measure. For the risk attitude this is surprising, because I expected the subjects to learn from the advice. If this was the case then there would have been a relation between the risk attitude of the partner and the actions of the subject.

Next is the public decision treatment. Remember that the plain results in de public decision treatment were negative, but less conclusive. The first thing that stands out, is that other variables matter in this regression. The exceedance probability of the partner in part one has a large significant negative relation with the actions of the subject in part two. This should not be the case, since it should have been impossible for the partner to express his risk attitude to the subject in this treatment. This was not the case however, as I mentioned before. Some students could not resist talking to their partners. The next important variable is the popularity of the partner. The coefficient means that if the partner has one extra friend, this is related to a 6 percent points increase in risk taking, relative to the actions in part one. To me this outcome makes sense. If the partner is popular, it may be more important to the subject to make an impression, if he can see his actions. The same goes for a male partner. If you consider a female subject, she may want to impress the opposite sex. A male male partnership may cause some competition.

What we should learn from the previous section is that introducing communication probably has a negative effect on risk taking. Especially if there was some excessive risk

(21)

20

taking in the socially neutral situation and discussing the problem contributes to making more sensible choices. Furthermore, the gender of the subject and the relation between the partners affects the communication effect. For the public decision treatment, the effect is less conclusive. If we can assume a learning effect, the effect of a public decision may even be positive, but this remains unclear. In this treatment, the features of the partner are important for explaining the behaviour. Partners that are male and/or popular are related to increased risk taking.

Hypothesis 3: The subjects who both communicate and take a public decision take more risk than the subjects who do either one or the other.

Because the risk taking in part one differed for the three classes, it is possible that there is unobserved heterogeneity between the classes. To control for this, the differences between part one and two are compared, instead of simply comparing the behaviour in part two. So I use a measure that is the difference between the two differences. A two-sample t-test with unequal variances was used to test if the difference is significant.

First I compare the mean of the communication treatment class with the mean of the class that did both. Both treatments include communication, so the difference shows the effect of a public decision. The difference between the two is 0.0129. Remember that the differences between part one and two were all negative. The positive figure thus tells us that including the public decision caused the subjects to behave less cautious than the group that only communicated, provided that it is significant.

The p-value of 0.3565 tells us that there is weak evidence that this is the case and that it is not significant at the 5% level. The economic significance of the measure can be interpreted as follows. A subject that only communicates decreases his exposure to risk about 1.3 percent points more than a subject who both communicates and makes a public decision. If the result was significant, this would support the hypothesis, because subjects act less cautious when a public decision is included.

Mean Standard Error P-value Difference ≠ 0 P-value Difference >0 Communication vs Both 0.0129 0.0350 0.7130 0.3565

Public decision vs Both -0.0090 0.0445 0.8401 0.5800

Table 6 – Two-sample t test with unequal variances comparing the Difference in Exceedance Probability

The same is done to compare the means of the public decision treatment with the treatment with both. The mean difference between the two is -0.009. If this difference is

(22)

21

significant, this means that introducing communication once again causes more risk averse behaviour in the second part. A more precise interpretation is that a person that communicates and then makes a public decision, decreases his risk exposure about 0.9 percent points more than someone that does not have the chance to communicate. With a p-value of 0.8401, the two are not significantly different from each other. The odds that the real difference could be positive are better with a p-value of 0.5800, but this p-value is still by far not low enough to accept that idea. If this was a significant result, you could say it contradicts the hypothesis, because of the negative effect. But again the two effects complement each other, which was expected, so that would have support the hypothesis.

We have seen that there are certain factors that can influence the change in behaviour, such as the gender of the subject and the partner, the relationship, the popularity of the partner etc. I have ran some regressions that included these variables, but this changed little to the results above. The coefficients are comparable to the means and remain insignificant. So for brevity, I leave this out of the text.

The conclusion for this hypothesis is that adding one effect to the other probably does not really have an effect. None of the differences are significantly different from zero. A possible cause could be that the effects of communication and public decisions are similar, so combining them does not have an extra effect. To test this, I use another t-test, which tells us that the difference of 2.19 percent points is not significant (p-value = 0.6168). The similarity of the effects could thus explain the results.

Hypothesis 4: The subjects who communicate and take a public decision, tend to follow advice more often than subjects who communicate and take a private decision.

In the treatments with communication, apart from their own preferences, subjects also wrote down the advice they got from their partners. As can be seen in the graphs below, when the subjects made a private decision, 20% chose to ignore the partner’s advice. When the subjects made a public decision, so the partner could see the decision, only 9% chose to deviate from the advice. This difference is quite substantial, but what should be noted, is that the variance in the exceedance probability in part one is much higher in the first group (0.0335 vs 0.0193). If the variance is small, that means that the preferences are more aligned and then there is less reason to deviate from an advice. Finally, to follow advice in at least 80% of the cases is very often. This tells us that subjects value the advice they get from the partners, independent of the treatment.

(23)

22

20%

80%

Private decision

Number of deviating cards Cards according to advice

Figures 4 and 5 – Percentages of cards that deviated from or were according to the advice, both treatments

Hypothesis 5: The subjects who are affected by the treatments are also more likely to smoke, drink excessively, have committed an offence and to have gotten themselves into a hospital.

To test this, I regress the difference in exceedance probability on all questions about risky behaviour for all three classes. If a subject is affected by the treatment and also scores on the factors mentioned in the hypothesis, the coefficient on those variables should be positive and significant.

The data was collected through a questionnaire (Appendix 5) and concerned everyday risky behaviour. For Risk scale, the subjects graded their own risk on a scale from zero to ten, where ten is most risk loving. Smoking is a dummy variable, as is drinking for subjects that drink more than six consumptions per night. Offence is a dummy variable for people who have ever committed an offence, for example crossing a red light. Accident is also a dummy variable that is one if a subject ever got into the hospital because of his own fault. Dangerous sports is one if a subject plays a dangerous sport like rugby. Another dummy variable is entrepreneurship, which if one if the subject has experience in entrepreneurship. The subjects also had to make a choice between a certain payoff and a lottery, in which the latter had a higher expected value. This variable is one if the subject chose the lottery. Finally, the subjects had to report what percentage of their income they save.

I hypothesized that the variables smoking, drinking, offence and accident are positively related to the effect of the treatments, because people of that age typically do that under peer pressure, because they want to fit in or to make an impression.

9%

91%

Public decision

Number of deviating cards Cards according to advice

(24)

23

Smoking and drinking turn out not to be significant in all treatments, although smoking is positively correlated. For the public decision treatment this is a large coefficient. If it was significant, it would have meant that people who smoke increase risk taking by 20 percent points more than non-smokers. I also think that some subjects lied about their smoking habits in the questionnaire, because the percentage non-smokers was lower than I expected. What should be said about the offence variable, is that almost everyone said to have ever committed an offence. In the third treatment everyone did, which resulted in perfect collinearity. So maybe this is not the best measure to assess the vulnerability to such pressures, since almost everyone has ever crossed a red light or something. The significant and positive coefficients on the risk scale is surprising. It speaks to mind that the scale relates to the absolute risk preference, but not per se that it relates to the change in preference. Apparently someone that sees himself as a risk loving person is more receptive to increase risk exposure in a social context. But there were very few entrepreneurs in each class, so maybe it should be interpreted with caution.

Communication Public Decision Both

Risk scale 0.0707** 0.0296* 0.0439** Smoking 0.0334 0.2081 0.0389 Drinking 0.0555 -0.0980 -0.0675 Offence -0.4043** 0.0839 - Accident -0.0314 0.0514 0.1401* Dangerous sport 0.1200** -0.0923 -0.1272* Entrepreneur -0.2270* 0.0742 0.1088 Risky choice 0.0441 -0.0175 -0.0287 Saving 0.0184 0.0251 0.0105 Constant -0.1712 -0.2977 -0.3476** R-squared 0.4067 0.1362 0.2310 Adjusted R-squared 0.2584 0.1068 0.1028

(25)

24

Discussion and Conclusion

The main focus of this study is the effect of discussion, thus communication, and social facilitation, thus a public decision, on risk taking. The effect of communication on risk taking is negative. The difference is significant and the subjects choose to decrease their exposure to risk with about 7 percent points on average. The effect of a public decision is less conclusive. The deviation measure is significantly negative, but the difference of 4.5 percent points combined with a larger variance results in an insignificant effect. This rejects the first two hypothesis, which is strange. I suspect a learning effect after part one, after which the subjects chose to play it safe. To control for this, the experiment could be replicated in a different order, or by using four groups, one for each treatment and one for control.

Further analysis shows that the differences between part one and two are influenced by different factors. In the communication treatment the difference is negatively influenced if the subject is a man and positively by the friendship the subject has with the partner. The difference in the public decision treatment however depends on the risk preference of the partner, which is counterintuitive because the subject should not be able to know the preference, since there is no communication. Furthermore, male and popular partners have a positive effect on the difference.

The differences between the treatment with both and either one of the treatment are both not significant. This can possibly be explained by the very similar effects of communication and public decision. Those effects are not significantly different from each other. So we can conclude that in the end, the effects of all treatments are alike.

I also find that people tend to follow advice more often when the partner can see the decision, than when they cannot. Finally risky behaviour that is said to be prone to peer pressure, turned out not to be related to the measure. The reported risk scale however was significant and positive for all treatments.

One problem of my study has to do with the setup of the game. It would have been easier to use a game like inflating a balloon until it bursts and just count how many times the balloon was inflated. I chose not to use such a game, because I wanted a game that could be played rationally. To stay with the balloon inflating example, if you do this with an actual balloon, there would be a major learning effect. The subjects will simply increase the amount of air in every round if it went well in the previous round. There are also programs that simulate risky games, but for such games, the odds change in each round. This solves the learning problem, but subjects now have nothing to base their decisions on. With my setup, it is clear from the beginning what the rules are and what is the critical value, so that subjects can make an educated and rational decision. Theoretically they could calculate the exceedance probability prior to each decision. But since the subjects only receive three cards in each round in part two, the subjects do not provide their preference for all cards. Therefore the results

(26)

25

rely partly on the value of those cards. I have found that the distribution differs in the three treatments and this adds to the problem. Unfortunately, I did not have enough time to use all card values in each round, because this would have solved the problem. For future research, this would be something to improve the setup. For example a computerized game that automatically saves all data and decisions would be very useful, because distributing the cards and writing down the decisions was very time consuming.

I also needed to make a trade-off between easy interpretation and accuracy for the measure. I think the two measures used are a good solution. The first one, the deviation measure, is simple but helps to find out if social interaction affects risky behaviour. If not, there would be no deviation, so that is very easy to use. The second measure, exceedance probability helps to find out how much it affects behaviour. Things like the initial card value and the number of additional cards are incorporated in the measure, and this makes it more accurate. In an experiment where all 13 card values are included in the second part, this would also improve the usefulness of the measures.

The implications for everyday life can be easiest understood when applying it to business for example. Say you are a business man and you need to make a decision that involves some risk. The case is not too complicated, but it is new to you, so a learning effect may apply here as well. At first, you underestimate the risk, but after consulting your colleagues, you are able to make a more sensible decision. Another example would be that you cannot ask your colleagues for advice, but you have to give a presentation about your decision at the end of the day. You take another good look at the assignment, because you do not want to make a fool of yourself, and see that a more conservative choice may be wiser. The shift is not as large as when you would have consulted your colleagues and some other colleagues, who got the same assignment and also present their decision, event took on some extra risk. Even colleagues who got the chance to do both, react about the same as you.

Of course the experiment and the situation differ in some aspects. In the experiment, the subjects had full responsibility for their own choices, which is typically not the case in a firm. So behaviour in a firm will always be more risk loving. But in a firm, your actions also affect others, which was not the case in the experiment. This will cause you to act a bit more cautious, so that compensates for the effect of not having full responsibility for you own actions. Furthermore, the gambling game is not something you do in everyday life. The gambling setting with the cards could have positively contributed to the risk levels, so that should be taken into account as well.

In the end, the results did not turn out as I expected, but that is interesting as well. I enjoyed working on the experiment and I hope to have contributed to the existing literature with my study.

(27)

26

Bibliography

Attanasio, O., & Barr, A. (2012). Risk pooling, risk preferences, and social networ. American

Economic …. Retrieved from

http://www.ingentaconnect.com/content/aea/aejae/2012/00000004/00000002/art00005

Bateson, N. (1966). Familiarization, group discussion, and risk taking. Journal of Experimental Social

Psychology. Retrieved from http://www.sciencedirect.com/science/article/pii/0022103166900734

Bem, D., Wallach, M., & Kogan, N. (1965). Group decision making under risk of aversive consequences. Journal of Personality and Social …. Retrieved from

http://psycnet.apa.org/journals/psp/1/5/453/

Binswanger, H. (1980). Attitudes toward risk: Experimental measurement in rural India. American

Journal of Agricultural …. Retrieved from http://ajae.oxfordjournals.org/content/62/3/395.short

Byrnes, J. P., Miller, D. C., & Schafer, W. D. (1999). Gender Differences in Risk Taking: A Meta-Analysis. Psychological Bulletin, 125(3), 367–383. Retrieved from

file:///C:/Users/Tessa/Dropbox/Master Thesis/Artikelen/Male risk taking.pdf Cettolin, E. (2010). Delegation in decision making under Are preferences incomplete ?

Eliaz, K., Ray, D., & Razin, R. (2006). Choice shifts in groups: A decision-theoretic basis. The

American Economic Review. Retrieved from http://www.jstor.org/stable/30034343

García-Marco, T., & Robles-Fernández, M. D. (2008). Risk-taking behaviour and ownership in the banking industry: The Spanish evidence. Journal of Economics and Business, 60(4), 332–354. doi:10.1016/j.jeconbus.2007.04.008

Gardner, M., & Steinberg, L. (2005). Peer influence on risk taking, risk preference, and risky decision making in adolescence and adulthood: an experimental study. Developmental Psychology, 41(4), 625–635. doi:10.1037/0012-1649.41.4.625

Holt, C., & Laury, S. (2005). Risk aversion and incentive effects: New data without order effects.

American Economic Review. Retrieved from http://www.jstor.org/stable/4132749

Hoskisson, R. E., Hitt, M. A., & Hill, C. W. L. (1991). Managerial Risk Taking in Diversified Firms: An Evolutionary Perspective. Organization Science, 2(3), 296–314. doi:10.1287/orsc.2.3.296 Kachelmeier, S., & Shehata, M. (1992). Examining risk preferences under high monetary incentives:

Experimental evidence from the People’s Republic of China. The American Economic Review. Retrieved from http://www.jstor.org/stable/2117470

Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk.

Econometrica, 47, 263–292. doi:10.2307/1914185

Lamm, H. (1967). Will an observer advise higher risk taking after hearing a discussion of the decision problem? Journal of Personality and Social Psychology. Retrieved from

(28)

27

Mcphee, J. (1996). Influence strategies in young adolescent dyads. Retrieved from http://psycnet.apa.org/psycinfo/1996-95015-289

Roberts, J. C., & Castore, C. H. (1972). The effects of conformity, information, and confidence upon subjects’ willingness to take risk following a group discussion. Organizational Behavior and

Human Performance, 8(3), 384–394. doi:10.1016/0030-5073(72)90057-8

Rockloff, M. J., & Dyer, V. (2007). An experiment on the social facilitation of gambling behavior.

Journal of Gambling Studies / Co-Sponsored by the National Council on Problem Gambling and Institute for the Study of Gambling and Commercial Gaming, 23(1), 1–12.

doi:10.1007/s10899-006-9042-4

Smith, V., & Walker, J. (1993). Rewards, experience and decision costs in first price auctions.

Economic Inquiry. Retrieved from

http://onlinelibrary.wiley.com/doi/10.1111/j.1465-7295.1993.tb00880.x/abstract

Wallach, M. A., & Kogan, N. (1965). The roles of information, discussion, and consensus in group risk taking. Journal of Experimental Social Psychology, 1(1), 1–19. doi:10.1016/0022-1031(65)90034-X

Zajonc, R. (1965). Social facilitation. Retrieved from

http://www2.psych.ubc.ca/~schaller/Psyc591Readings/Zajonc1965.pdf

Zajonc, R. B., Wolosin, R. J., Wolosin, M. A., & Loh, W. D. (1970). Social facilitation and imitation in group risk-taking. Journal of Experimental Social Psychology, 6(1), 26–46. doi:10.1016/0022-1031(70)90074-0

(29)

28

Appendices

Appendix 1: Instructions part one

Instructies

Welkom bij dit experiment. Lees deze instructies goed door, alsjeblieft. Het experiment duurt 45 minuten in totaal en bestaat uit twee delen en een vragenlijst.

Jullie kunnen geld verdienen met dit experiment! Dit is deels afhankelijk van je eigen keuzes en deels van geluk. De keuzes van je klasgenoten hebben echter nooit invloed op jouw resultaten, en andersom ook niet. Daarnaast hebben je keuzes in deel 1 geen invloed op je resultaten in deel 2 en vice versa. Je keuzes, resultaten en uitbetaling blijven te allen tijden anoniem. Wegens een klein budget zal ik niet iedereen voor elke ronde kunnen uitbetalen. Er zullen willekeurig 5 mensen worden gekozen, waarvan de verdiensten bij 1 beginkaart uit het experiment worden uitbetaald. Je kan in het experiment maximaal €14 verdienen. De uitbetaling vindt direct na het experiment plaats in een gesloten envelop.

Het is heel belangrijk dat jullie tijdens het hele experiment stil zijn. Als iets niet duidelijk is of als je vragen hebt, steek dan gerust je hand op, maar vraag het alsjeblieft niet aan je buurman / vrouw. Wanneer jullie overleggen, beïnvloedt dit mijn resultaten en dat maakt ze onbruikbaar. Als het te rumoerig is in de klas, kan ik dan ook helaas niet tot uitbetaling overgaan.

Deel 1

De opzet van het experiment is enigszins geïnspireerd op het spel Black Jack, oftewel 21-en. De opzet is echter iets anders. Hierbij de regels op een rijtje:

• Eén spel begint met één kaart, waarbij je kan kiezen of je nog meer kaarten wil, en hoeveel dan wel.

• De kaarten worden getrokken uit een hele grote stapel kaarten, wat betekent dat de kans op alle kaarten ongeveer gelijk is, ook als er al een aantal kaarten zijn getrokken.

• De waarde van alle kaarten zijn hiernaast samengevat:

Let op: de Ace is 1 waard, de Jack, Queen en King zijn respectievelijk 11, 12 en 13 waard zijn. (Engels kaartspel)

• Bij het spel Black Jack is 21 de waarde die je wil behalen, maar niet wil overschrijden. Bij dit experiment is dat 14.

• Wanneer de totale waarde van je kaart of kaarten 14 of lager is, verdien je deze totale waarde in euro’s. Bijvoorbeeld, een totaal van 11 is €11 waard.

• Wanneer de totale waarde van je kaarten 14 overschrijdt, verdien je altijd €0,-

• Wanneer je meerdere kaarten krijgt, moet je voor elke afzonderlijke kaart bepalen of je nieuwe kaarten wil en hoeveel. Als je bijvoorbeeld 3 kaarten krijgt, vormen zij geen set, maar ga je dus 3 beslissingen maken.

Op je tafel ligt een formulier met 13 kaartafbeeldingen. Wanneer ik dat straks aangeef, kan je het formulier omdraaien en voor elke kaart invullen of je nog extra kaarten wil en zo ja, hoeveel kaarten. Elk van deze 13 kaarten is dus een apart “spel”, oftewel, voor elke kaart ga je apart invullen of en hoeveel kaarten je er nog bij wil. Zet ook je naam en je tafelnummer op het formulier.

Kaart Waarde Ace 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 Jack 11 Queen 12 King 13

(30)

29

Er zijn een aantal versies van het formulier, waarop de kaarten steeds op een andere volgorde liggen. Wij weten niet hoe deze verdeeld zijn over de tafels. Als je je formulier hebt ingevuld, vouw je het weer dubbel zoals je het hebt gevonden. Zo kunnen wij de waarden van je kaarten niet zien.

Na deel 1 volgen de instructies voor deel 2. Veel succes!

Appendix 2: Example of the decision form in part one

Naam: Tafelnummer:

Appendix 3: Instructions part two

Appendix 3.1 Communication

Deel 2

In deel 2 krijg je weer een aantal kaarten te zien, waarbij je steeds moet beslissen of je extra kaarten wil of niet. Hierbij gelden weer dezelfde regels als in deel 1 en ook de opzet per ronde zal ongeveer hetzelfde zijn. In dit deel krijg je echter steeds 30 seconden de tijd om met je buurman/vrouw te overleggen, voordat je je beslissing maakt.

In elke ronde krijg je 3 blinde kaarten, die je mag omdraaien als ik dit aangeef. Je krijgt nu 30 seconden de gelegenheid om te overleggen met je buurman/vrouw, maar tijdens het overleggen, vul je je formulier nog niet in. Na het overleg vul je het formulier in en draai je de kaarten weer om. Het is belangrijk dat je keuzes privé blijven, probeer dus te vermijden dat de ander je ingevulde keuzeformulier ziet. Je begint met degene die nu naast je zit en daarna zullen we nog 2 keer wisselen, zodat je in totaal met 3 personen bent gekoppeld.

Een samenvatting van de stappen: 1. Wij delen de kaarten blind uit.

2. 30 seconden om te overleggen met je buurman/vrouw. 3. Je vult je formulier in.

4. Wij delen het gewenste aantal kaarten rond.

Kaart 1 2 3 4 5 6 7 8 9 10 11 12 13 Nog een kaart? Zo ja, hoeveel?

(31)

30

5. Degenen aan de rechterkant van een rij schuiven door naar de volgende tafel, volgens het onderstaande systeem. Neem altijd je eigen formulier mee naar de volgende tafel! 6. Stappen 1 t/m 5 worden 2 keer herhaald.

Appendix 3.2 Public decision

Deel 2

In deel 2 krijg je weer een aantal kaarten, waarbij je steeds moet beslissen of je extra kaarten wil of niet. Hierbij gelden weer dezelfde regels als in deel 1.

We zullen zo eerst de tafels zo schuiven, zodat je tegenover je buurman/vrouw zit. Daarna krijgt iedereen 3 kaarten. Voor deze kaarten ga je invullen of je er nog kaarten bij wil, en hoeveel. Als de formulieren zijn ingevuld, deel ik het gewenste aantal kaarten uit. Tijdens dit deel mag je partner je kaarten en beslissingen zien, maar het is dus niet de bedoeling dat er overlegd wordt. Je begint met degene die nu naast je zit en daarna zullen we nog 2 keer wisselen, zodat je in totaal met 3 personen bent gekoppeld.

Een samenvatting van de stappen: 1. Wij delen de kaarten blind uit. 2. Je vult je formulier in.

3. Wij delen het gewenste aantal kaarten rond.

4. Degenen aan de rechterkant van een rij schuiven door naar de volgende tafel, volgens het onderstaande systeem. Neem altijd je eigen formulier mee naar de volgende tafel! 5. Stappen 1 t/m 4 worden 2 keer herhaald.

Appendix 3.3 Communication and Public decision

(32)

31

In deel 2 krijg je weer een aantal kaarten te zien, waarbij je steeds moet beslissen of je extra kaarten wil of niet. Hierbij gelden weer dezelfde regels als in deel 1 en ook de opzet per ronde zal ongeveer hetzelfde zijn. In dit deel krijg je echter steeds 30 seconden de tijd om met je buurman/vrouw te overleggen, voordat je je beslissing maakt.

We zullen zo eerst de tafels verschuiven, zodat je tegenover je buurman/vrouw zit. Daarna krijgt iedereen 3 blinde kaarten, die je mag omdraaien als ik dit aangeef. Voor deze kaarten ga je beslissen of je er nog kaarten bij wil, en hoeveel. Voordat je dat doet, krijg je 30 seconden de gelegenheid om te overleggen met degene tegenover je. Als je het formulier hebt ingevuld, delen we het gewenste aantal kaarten uit. Ondertussen zullen wij rondlopen om alles te noteren. Tijdens alle rondes mag je partner je kaarten en beslissingen zien. Je begint met degene die nu naast je zit en daarna zullen we nog 2 keer wisselen, zodat je in totaal met 3 personen bent gekoppeld.

Een samenvatting van de stappen: 1. Wij delen de kaarten blind uit.

2. 30 seconden om te overleggen met je buurman/vrouw. 3. Je vult je formulier in.

4. Wij delen het gewenste aantal kaarten rond.

5. Degenen aan de rechterkant van een rij schuiven door naar de volgende tafel, volgens het onderstaande systeem. Neem altijd je eigen formulier mee naar de volgende tafel! 6. Stappen 1 t/m 5 worden 2 keer herhaald.

Appendix 4: Decision forms part two

Appendix 4.1 Treatment with

communicati on

Naam:

Wat was het

advies van je partner?

Wil je nog een kaart? Zo ja, hoeveel? Tafel nr: Partner: ……… Kaart 1 Kaart 2 Kaart 3 Tafel nr: Partner Kaart 1 Kaart 2

(33)

32 ……… Kaart 3 Tafel nr: Partner: ……… Kaart 1 Kaart 2 Kaart 3

(34)

33

Appendix 4.2 Treatment without communication

Appendix 5: Questionnaire

Vragenlijst

Nummer tijdens experiment:

1. Hoe zie je jezelf van een schaal van 1 tot 10? Ben je iemand die graag risico neemt of probeer je risico’s te vermijden?

Ik vermijd risico’s zoveel mogelijk Ik zoek het risico graag op

0 1 2 3 4 5 6 7 8 9 10

2. Rook je?

3. Heb je wel eens een overtreding gemaakt? Door rood fietsen o.i.d.

4. Beoefen je een sport? Zo ja, wat voor sport?

5. Ben je wel eens door je eigen schuld in het ziekenhuis beland?

6. Spaar je? Zo ja, welk percentage van je inkomen ongeveer?

7. Drink je wel eens alcohol? Zo ja, hoeveel glazen gemiddeld?

Naam:

Wil je nog een

kaart? Zo ja, hoeveel? Tafel nr: Partner: ……… Kaart 1 Kaart 2 Kaart 3 Tafel nr: Partner ……… Kaart 1 Kaart 2 Kaart 3 Tafel nr: Partner: ……… Kaart 1 Kaart 2 Kaart 3

(35)

34 8. Heb je ervaring met ondernemerschap?

9. Kruis aan welke van de volgende opties jij zou kiezen: o €18 met zekerheid

o 50% kans op €40, 50% kans op €0

10. Met welke 5 personen uit de klas ben je het best bevriend?

11. Wie uit je klas haalt meestal de hoogste cijfers?

12. Wie uit je klas zou als eerste een miljoen winnen in een casino?

Referenties

GERELATEERDE DOCUMENTEN

In the long run, it will be to the distinct advantage of the South African society as a whole if a culture of respect for fundamental rights and the constitutional process of

TAB L E 2 (Continued) Approach Experimental paradigm Minimum ligand set necessary Minimum number of experimental conditions Results confirming the presence of high ‐affinity state

(2001), which is the final price paid per share minus the target share price four weeks before the announcement, divided by the target share price four weeks before the

De provincie Overijssel koos dus voor het stimuleren van burgerinitiatieven door middel van een wedstrijd om vervolgens de uitvoering van de meest kansrijke initiatieven

The model results reveal the existence of stable equilibrium states with more than one inlet open, and the number of inlets depends on the tidal range and basin width (section 3)..

Two conditions required to apply option theory are that the uncertainty associated with the project is market risk (the value-in‡uencing factors are liquidly traded) and that

At least we expect a moderating effect of AC characteristics on the relationship between auditor gender and the readability of KAMs due to the possible impact AC can have

Despite the fact that different platforms and classifiers are used, the method of creating a classifier is the same for every classification model, whether it be for Twitter data,