The Effect of Time Pressure on (pro)
Social Behavior in combination with a
Check for Rationality: Evidence from a
Lab Experiment
Abstract: Many economically relevant decisions are executed under notable time pressure. This study investigates what the effect of time pressure is on (pro) social behavior in a modified dictator game and generates evidences from a lab experiment. Furthermore, this study applies the axioms of revealed preferences to check if decisions are rational and consistent. The experiment consists of two treatments, one group where Decision-making is tested under time pressure and one group where there is no time pressure. Results show a significant difference between both treatments due to selfishness and found that subjects act more generously under time pressure. Moreover, there is a negative relationship between decision time and the level of selfishness, indicating that behaving more generously takes more time. Furthermore, more violations of axioms of revealed preferences are made under time pressure.
Name: Puck Hooijer
Student Number: 10579699
Master Economics; Track: Behavioral Economics and Game Theory Supervisor: Prof. dr. J.H. Sonnemans
Second Reader: Prof dr. C.M. van Veelen ECTS: 20
Statement of Originality
This document is written by Student Puck Hooijer (10579699) who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Table of Contents
1 Introduction ... 4
2 Literature Review ... 5
2.1 Two System Theory ... 5
2.2 Axioms of Revealed Preferences ... 9
3 Methodology ... 11
3.1 Objections, Goals and Tasks ... 11
3.1.1 Matrix Task ... 12 3.1.2 Decision-Making Task ... 13 3.2 Experimental Design ... 15 3.2.1 Experimental Setup ... 15 3.2.2 Treatments ... 16 3.2.3 Questionnaire ... 17 3.2.4 Payments ... 17 3.2.5 Subjects ... 18 3.3 Limitations Experiment ... 18 3.4 Hypotheses ... 18 3.4.1 Manipulation Checks ... 19 4 Results ... 20
4.1 Descriptive Statistics Questionnaire ... 20
4.2 Results Manipulation Checks ... 22
4.2.1 Problem Matrix Task ... 24
4.3 Decision-Making Task ... 24
4.3.1 Individual Preferences ... 26
4.4 Linear Regression ... 28
4.5 Interpretation of Results ... 31
4.6 Violations of Revealed Preferences ... 32
5 Discussion and Conclusion ... 36
6 References ... 39
7 Appendix ... 41
A Experimental Instructions ... 41
1 Introduction
A significant collection of studies suggest that time pressure affects decisions and decision-making processes (Cone & Rand, 2014; Young, Goodie, Hall & Wu, 2012; Sutter, Kocher and Strauß, 2003; Dror, Basola & Busmeyer 1999; Payne, Bettman & Luce, 1996). Multiple economically relevant activities including financial decisions, buying a house and bidding in auctions occur under notable time pressure. Hence, a lot of economists subscribe to a popular slogan “time is money”, suggesting that quicker decisions may imply higher profits (Kocher & Sutter, 2006). Moreover, payoff systems often depend on the speed and quality of decision-making, whenever people experience certain time pressure this can influence their decisions or their decision-making process. Laboratory economic experiments have shown that people behave differently under a time constraint (Sutter et al., 2003). On the one hand, this could mean that people have different individual preferences when facing time pressure. On the other hand, people could use different decision-making strategies that lead to other choices. Kahneman (2003) discusses the two system theory in which decisions in general are
influenced by two systems: the intuitive and the deliberative system. An unbounded rational decision-maker will have unlimited capabilities when making decisions and often operates in the deliberative or rational system. Conversely a bounded rational decision-maker is often moved by intuitive thinking when making decisions under a time constraint. It is interesting to investigate if intuitive decision-making stimulates typical social behavior and if this leads to different individual preferences.
Alternatively, researchers Recalde, Riedl and Vesterlund (2015) claim that people do not have different individual preferences or different decision-making strategies under time pressure but that the differences in behavior are due to irrational behavior. This study
investigates what the effect of time pressure is on (pro) social behavior, if it stimulates typical social behavior and if people behave more irrationally under time pressure. To answer the research question, a laboratory experiment has been conducted which includes two
treatments. In the first treatment group, decision-making is tested under time pressure and in the second treatment group, decision-making is tested without time pressure. Furthermore, this study adopts the research design from Andreoni & Miller (2002) by applying the axioms of revealed preferences on the decisions of subjects to test if (pro) social behavior can be explained in the language of a well-behaved preference ordering.
Results show that people tend to be more generous under time pressure. In addition, this study finds that subjects are more irrational when making decisions under time pressure, this result is significant at a 5% significance level.
This paper is organized as follows: section 2 contains related literature, section 3 provides the methodology, including the experimental design, procedures, hypotheses and
manipulation checks. Section 4 summarizes the main findings and analyses the empirical results. Section 5 includes the discussion and conclusion and discusses some shortcoming of the experiment. Section 6 list all the references and the last section contains the appendix. 2 Literature Review
The section outlines the relevant theoretical and experimental work and consists of two parts. The first part discusses the two system theory, additional relevant literature, economic
experiments and outline some limitations in both the literature and the research. The second part explains the axioms of revealed preferences which are necessary for testing the
rationality of decisions. 2.1 Two system theory
As mentioned earlier in the introduction, Kahneman (2003) explains that decisions in general are influenced by two systems. System 1 is the intuitive system, which is responsible for fast automated, effortless, associative, emotionally charged and rule-based decisions. System 2 is the rational or deliberative system, which is responsible for slower, serial, effortful,
deliberately controlled decisions and through which calculated reflective decisions are made. When individuals make social decisions, they tend to be more extreme, either close to selfish or generous. Some suggest that generosity is intuitive however there is opposing evidence in favor of the deliberate and calculated choice (Kinnunen & Windmann, 2013; Fiedler,
Glöckner, Nicklisch, & Dickert, 2013).
Kinnunen & Windmann (2013), studied the relationship between rational cognitive and emotional-intuitive thinking models in the context of (pro) social behavior. They found that System 1 predominantly promotes sharing and altruistic punishment. Altruistic
punishment is a costly form of punishment that individuals use to punish and yields no material gain (Fehr & Gächter (2002). Kinnunen & Windmann (2013) are in favor of the intuitive system, saying that generosity is mostly regulated by System 1.
Fiedler et al. (2012) investigate the relationship between Social Value Orientation (SVO) and cooperative behavior in social dilemmas. SVO is an individual difference variable that represents a person’s preferences regarding how to allocate money between themselves and another person. Figure 1 shows a graphical representation of the SVO. Looking at the top right part of the circle, an individual can decide how to allocate their money, where the 45-degree point shows an equal division of the money between themselves and the other. Fiedler et al. (2012) find that decision time gradually increases with absolute SVO deviation from a purely selfish orientation. So that the act of giving is more in favor of a deliberate and calculated choice instead of the impulsive and intuitive choice.
Figure 1: SVO (Fiedler et al., 2012)
Many laboratory experiments study economic decision-making under time pressure. A lot of these experiments focus on social behavior such as altruism, fairness and selfishness and are tested in many economic exchange games including prisoner’s dilemma, public goods game, ultimatum game and dictator game. There is mixed evidence regarding how people behave in different economic exchange games and how people behave under certain time pressures. Some findings suggest a positive effect of time pressure on cooperation in a competitively framed game (Cone & Rand, 2014), whereas others found selfishly inclined behavior under a time constraint (Sutter et al., 2003).
To build on the literature of Kahneman (2003) on the two system theory, Rand, Greene & Nowak (2012) explore the cognitive basis of cooperative decision-making in humans using a dual-process framework. The framework focusses on the distinction between automatic and controlled processing and Rand et al. (2012) investigate cooperative decisions under time pressure in two experiments with a one-shot public good game. In the two
experiments, noncompliant participants (those who took longer than 10 seconds in the time pressure condition or less that 10 seconds in the time delay condition), were excluded prior to analysis. In the Registered Replication Report: Rand, Greene, and Nowak (2012),
Bouwmeester et al. (2017) argued that these experiments conducted with time pressure are consistent with the presence of selection biases and the absence of a causal effect of time pressure on cooperation. The results of the experiment show that the mean contribution is higher under time pressure compared to the group with a time delay. However, when Rand et al. (2012) included the noncompliant participants (those who took longer than 10 seconds in the time pressure condition or less that 10 seconds in the time delay condition) in their analysis to avoid selection bias, the contribution did not differ significantly between the two experiments. This study experiment is created in a way that it avoids such selection bias, by using a fixed time limit of five minutes per block. Since it is not possible to exceed this limit, there are no exclusions of subjects prior to the analysis. It is possible that subjects take more time for a specific task, but this is no reason for subject exclusion.
Other economists that devoted more attention to the relation between decisions and decision-making time are Sutter et al. (2003). They examined the influence of time pressure on bargaining behavior in an ultimatum game. In this game, the proposer makes an offer on the split of a sum of money and if the responder accepts the offer, the offer is implemented, while rejecting the offer gives both sides nothing (Güth, Schmittberger & Schwarze, 1982). In the experiment of Sutter et al. (2003) the subjects had either 10 seconds to decide on the offer (fast treatment) or 100 seconds’ (slow treatment). They found that the rejection rates of responders are higher under a tight time constraint that under a very weak time constraint. A critique on this experimental research is that in the instructions of the experiment, the responder is already aware of the fact that the proposer will receive an initial endowment of 10 euros and has the possibility to divide the money between him/her and the responder. By presenting the instructions like this, the responder can already think about accepting/rejecting an offer made by the proposer. For instance, if the proposer offers 4 euros or less to the responder, the responder decides to reject the offer of the proposer. Consequently, it can be questioned if bargaining behavior in this setting differs under time pressure compared to when there is no time pressure.
Instead of using an ultimatum game, this study uses a modified dictator game to test (pro) social behavior under time pressure. In an original dictator game, developed by Forsythe, Horowitz, Savin & Sefton (1994), the dictator has the option to divide the money between him/her and a recipient. In that case the payoffs system look than like: ps + po = m,
where ‘s’ stands for self, ‘o’ stands for other and ‘m’ is the total money which can be divided (Andreoni & Miller, 2002, p. 739). The difference between the original and the modified dictator game is that in the modified version the dictator faces several dictator “budgets” that each have different incomes and different prices of transferring this income to another subject. Because each budget is different each time it is presented to the dictator, the option of
learning and thinking about what to decide is less prevalent. Using this dictator game, Andreoni & Miller (2002) tested whether subjects have preferences for altruism and they found evidence in favor of this. Correspondingly, Andreoni & Vesterlund (2001) studied gender differences in altruism also by performing a modified dictator game. They found mixed evidence for altruism in their experiment. Females tend to be more generous when altruism is expensive, whereas males tend to be more altruistic when altruism is cheap. Similarly, males are more extreme, demonstrating complete selfishness, as opposed to females who prefer an equal division of money (Andreoni & Vesterlund, 2011).
Recalde et al. (2015) tests how response time correlates with individual decision-making and criticizes the fact that decisions may not result in more generous or selfish behavior but rather that choices of individuals are irrational, which they present as mistakes. Recalde et al. (2015) use constant-return public good games to test the correlation between response times and mistakes. For a clear inference on preferences from response time, they examine environments where the error can be identified and where the correlation between response time and error can be evaluated. Their study shows that subjects behave more irrational under a tighter time constraint and thus conclude that these decisions can be explained as mistakes and not as individual preferences.
Andreoni & Miller (2002) also examine if the behavior of people in economic exchange games is rational and whether subjects tend towards altruism or fairness. To avoid the problem of mistakes presented in the paper from Recalde et al. (2015), Andreoni & Miller (2002) expressed the behavior of the subjects in economists’ language of a well-behaved preference ordering. They use the modified dictator game to test for altruism and applied the axiom of revealed preferences to the altruistic actions of the subjects in their experiment. They found that altruism is indeed rational and can be expressed in the economists’ language of well-behaved preference ordering.
2.2 Axioms of revealed preferences
To test whether subjects behave rational, Andreoni & Miller (2002) used different axioms of revealed preferences and focusses on the Weak Axiom of Revealed Preferences (WARP), Strong Axiom of Revealed Preference (SARP) and Generalized Axiom of Revealed
Preference (GARP). To explain the axioms of revealed preferences, first assume that there are bundles of choices which are determined by a maximum amount that someone can give to themselves and a maximum amount a subject can give to another. To clarify, there are two bundles of choices i and j, and each bundle is lying on a linear budget constraint.
Directly revealed preference is explained as: bundle i is directly preferred to j, if j was in the choice menu when i was chosen.
Indirectly revealed preference is explained as: bundle is i is directly revealed preferred to j, j is directly revealed preferred to k, then i is indirectly revealed preferred to k.
WARP: if i is directly revealed preferred to j, then j is not directly revealed preferred to i. SARP: if i is indirectly revealed preferred to j, then j is not directly revealed preferred to i. GARP: if i is indirectly revealed preferred to j, then j is not strictly directly revealed preferred to i, that is, i not strictly within the bundle set when j is chosen.
The theorem of Afriat (1967) was the first to establish necessary sufficient conditions on a dataset for rationalizability by utility maximization (Polisson & Renou, 2016). The finite bundle of prices and demand observations that are generated by an individual’s choice can be rationalized by this well-behaved utility function, only when the data satisfy GARP (Choi, Fisman, Gale & Kariv, 2007).
Figure 2 illustrates a violation of GARP involving two choices i and j. The x-as presents the maximum amount of money someone can assign to themselves and the y-as presents the maximum amount someone can give to the other. GARP requires that if choice i is revealed preferred to j, then j is not strictly revealed to i. “Note that if choices violate WARP they must also violate SARP, and if they violate GARP then they must also violate SARP, however the opposite is not true” (Andreoni & Miller, 2002, p.739).
Figure 2: violation of GARP
The axioms of revealed preferences are explicitly used in the research of Andreoni & Miller (2002) and together with the research of Recalde et al. (2015), decisions from individuals can be checked against consistency for individual preferences. By using the same modified dictator game, likewise social behavior can be tested. Several menus of choices that are lying on different linear budget constraints are presented to the subjects. The dictator faces every time a different budget line and has the option to divide the money between him/her and the recipient. When presenting different menus of choices to the subjects, it can be easily checked whether there is a violation of the axioms of revealed preferences and whether subjects
behave more inconsistent with rationality under time pressure. If the decisions of the subjects do not violate the axioms of revealed preferences, then according to Afriat’s (1967) theorem, it can be concluded that there exists a well-behaved utility function that rationalizes these choices.
Andreoni & Miller (2002) tests whether altruistic choices are consistent with the axioms and found that altruism can indeed be explained by rationality. They concluded that the choices of the subjects are made on individual preferences and thus explain the date. Applying the same menus of choices from their research on this study experiment and place the subjects in an environment where they also feel time pressure, the critique from Recalde et al. (2015) can be explained or may be refuted.
However, Bardsley (2006) criticizes dictator game experiments in general. Many of these dictator game experiments including the one from Andreoni & Miller (2002) have been used to explore altruistic behavior, conversely these experiments almost never allow subjects
j i 0,0 1,0 2,0 3,0 4,0 5,0 6,0 7,0 8,0 0,00 2,00 4,00 6,00 8,00 Ot he r Self
game and Bardsley (2006) argued that dictator game experiments would appear to score highly on conventional standards of ex-ante external validity. He argued that the results of dictator game experiments with only giving cannot be generalized to other situations and to other people, since in everyday life it is slightly unlikely to give anonymous donations to random strangers. Money transfers are often made to family members, friends, charity organizations or face-to-face to people.
Therefore, Bardsley (2006) conducted an experimental design which in the first place focuses on the subjects’ willingness to give, but also gives them the opportunity to take away money from another participant. His experiment consists two treatments where it compares a dictator game (first treatment) to a “taking game” (second treatment) and found indications of inconsistency between the two treatments. Most subjects are not so generous anymore by giving them also the opportunity to take away money. However, when Bardsley (2006) only examines behavior in one domain, for example where only giving was allowed, evidence showed that behavior is consistent with the research of Andreoni & Miller (2002). 3 Methodology
This section outlines the methodology regarding this study experiment and consists of four parts. Firstly, some objections regarding earlier experiments are being discussed.
Furthermore, the goal and tasks of the experiment are outlined. The second part describes the experimental design, including the experimental setup, treatments, questionnaire, payments and subjects. In the third part, some limitations with respect to this study experiment are outlined. The hypotheses and manipulation checks for this study are discussed in the fourth part.
3.1 Objections, Goal & Tasks
As mentioned earlier in the section literature review, there are some objections regarding earlier research with time pressure. Firstly, in everyday life it is unusual to experience a very tight time constraint when making decisions. It is normal to be in a rush or make decisions facing a deadline, however it almost never occur that people must make decisions within a fixed number of seconds and sequential, after crossing these seconds decisions cannot be made anymore. Hence, a tight time limit can lead to selection bias, which Rand et al. (2012) encountered in their research. Subjects who took longer than 10 seconds, where excluded
prior to the analysis. Hereby, the measurement of irrational behavior as in the research from Recalde et al. (2015) cannot be performed correctly. This study experiment care for this problem by placing subject in a time pressure treatment, without tight time limits on decision-making. This study measures the mistakes in consistency by looking how often subjects show irrational behavior.
Importantly, the main goal of the experiment is to analyze (pro) social behavior under time pressure. This study combines two kinds of tasks to test this, a real effort task and a modified version of the dictator game. Furthermore, this study involves a measurement check on the violations of revealed preferences. The experiment consists of two treatment groups, in both treatments participants are informed that their decisions can affect their payoffs and the payoff of another participant. The payoffs of subjects depend on speed and quality when performing the tasks.
3.1.1 Matrix task
The first task is used before in an experiment of Weber & Schram (2016) and is a real effort task. In this real effort task, each subject face two 10x10 matrices that are randomly filled with two-digit numbers. For each pair of matrices, the subjects are asked to find the highest number in the left matrix, the highest number in the right matrix and to calculate the sum of these two numbers. This answer must be filled-in in the empty box at the bottom of the page. The subjects are not allowed to use calculators, but can use pen and paper for summations. After answering the task there will no feedback provided, this mean that the subjects have only one attempt to provide the correct answer and do not know whether their answer is the right one. Figure 3 shows a screenshot of the Matrix task.
Figure 3: Screenshot of the Matrix task
The subjects must correctly complete at least four matrices within one block of five minutes to eligible for a bonus. The specific choice for the number of four matrices is made after a few testing rounds with friends and acquaintances. The testing rounds also included blocks were participants had six minutes instead of five minutes to complete at least four matrices. However, completing four matrices within six minutes was too easy feasible, therefore time within one block is reduced to five minutes.
3.1.2 Decision-Making task
The second task in this experiment is used before in the research of Andreoni & Miller (2002) and is a modified version of a dictator game. A modified version is different than the original dictator game in the way that in an original dictator game, the dictator decides how to allocate money between themselves and another person, so that ps + po = m, where in the modified dictator game each subject is given a menu of choices with different endowments and prices for payoffs, ps + Ppo = m. The ‘P’ can differ between rounds so that the menu of choices also differ.
In each block of five minutes, subjects face three menus of choices, lying on different linear budget constraints, so twelve budget lines in total. This experiment uses the same eleven budgets as in the experiment of Andreoni & Miller (2002), but includes one extra menu to create an even amount. This extra menu, associate with one linear budget constraint is one out of the same menus that falls within the other eleven budgets. The task differs from Andreoni & Miller (2002) in the way it is presented to the subjects. In their experiment the
subjects made their decision by filling in the blanks in a statement, for example, “Divide 60 tokens: Hold___ at 1 point each, and Pass ___ at 2 point each” (Andreoni & Miller, 2002, p. 739). In this experiment, the subjects face in each block three menus of choices that are lying on different linear budget constraints and are encountered to each subject. The x-as describes the maximum amount subjects can assign to themselves and the y-as describes the maximum amount subjects can give to the other. Subjects must choose a point on the line that indicates this. So, on the extreme of the y-as the subject gives everything to the other, where on the extreme of the x-as the subject is perfectly selfish and gives nothing to the other. The subjects can click as many times as they want on different points on the line before sending in their choice. Figure 4 shows a screenshot of the Decision-Making task.
The full menu of budget lines offered to the subjects are shown in Table 1. Table 1: Budget Menus
Note: Budget menu 12 is double and is the same as budget menu 9
3.2 Experimental Design
This experiment uses a between-subject design to examine decisions in two types of
treatments. Each subject is either assigned to the first or to the second treatment. Subjects are staying in the same treatment during the whole experiment, so when a subject is in first treatment, he or she does not switch to second treatment.
3.2.1 Experimental Setup
The subjects are welcomed in the CREED laboratory at the University of Amsterdam. Everyone takes place behind a computer and can read the instructions by themselves on the computer. The instruction part consists of general instruction, two training questions and instructions about payment. At the end of the experiment the subjects are asked to fill in a questionnaire with questions related and unrelated to the experiment.
Budget Intersection X-as Intersection Y-as
1 600 cents 600 cents 2 720 cents 240 cents 3 450 cents 900 cents 4 360 cents 720 cents 5 360 cents 360 cents 6 900 cents 450 cents 7 240 cents 960 cents 8 720 cents 360 cents 9 480 cents 480 cents 10 240 cents 720 cents 11 960 cents 240 cents 12 480 cents 480 cents
Before beginning the real experiment, each subject first gets two training questions related to the Matrix task and the Decision-Making task. To begin with, one example of the Matrix task is given. Subjects get instructions about the task and face two matrices, at the bottom of the page the correct answer is given. They cannot fill anything in and are only able to click at the “continue” button to go to the next training question. For the second training question subjects get instructions about the Decision-Making task. One menu with a budget line is presented to the subject. Subjects can click on the budget line and experience how much they can assign to themselves and how much they can give to another participant. In addition, they are asked to pick a specific point on the line so that “your” earnings are between 180-200 cents and the earnings of the other are between 142-174 cents. Decisions made in this training phase do not influence the payoffs of the subjects. After correctly picking the point on the line within these ranges, they can click on the “send” button. The training phase is over and the subjects can continue reading the instructions. After the last instruction page, the real experiment starts.
Each block of five minutes begins with first a Matrix task, followed by a Decision-Making task and then back to a Matrix task and so on. In each block of five minutes’ subjects are encountered with three Decision-Making tasks and after the three Decision-Making task, the subjects only face Matrix tasks so that the design of the experiment looks like:
Matrix task– Decision-Making task – Matrix task – Decision-Making task – Matrix task – Decision-Making task – Matrix task – Matrix task – Matrix task - …
3.2.2 Treatments
There are two treatments in this experiment. The Time treatment is the treatment where the Making task is tested under time pressure. In the No-time treatment, the Decision-Making task is tested without time pressure. The subjects perform the Matrix task always under time pressure to avoid that otherwise in the No-time treatment easier money can be earned. The difference between the treatments is that in the Time treatment the clock
continues ticking when subjects perform the Decision-Making task. In the No-time treatment, the clock stops and time freezes during this task, so that in this treatment the subjects
experience no time pressure when performing the modified dictator game.
3.2.3 Questionnaire
After the fourth block of five minutes, subjects are asked to fill in a questionnaire. The questionnaire consists of questions related and unrelated to the experimental tasks. The
unrelated questions are demographic questions, such as age, gender and study field. Questions related to the experimental task are about how they experiences the two tasks and if the subjects find both tasks difficult, fun or stressful. The questionnaire also contains one control question related to the treatments and asks if the clock continued or stopped ticking during their experiment. There is also room for comments on the two tasks.
3.2.4 Payments
In both treatments subjects can earn money with the Matrix task and with the Decision-Making task. The experiment consists of four blocks of five minutes. For the Matrix task one of the four blocks is randomly chosen for payment, with a four-sided die. When the subjects have correctly complete four or more matrices within that block of five minutes, the subjects receive a bonus of 10 euros. There are three Decision-Making tasks in each block, so in total there are twelve Decision-Making tasks. One out of the twelve tasks is randomly chosen for payment, with a twelve-sided die. Next, one subject (e.g. Subject A) is randomly match with another subject (e.g. Subject B) in this treatment, with a 50% probability Subjects A’ decision in this task will be executed and with 50% probability the decision of Subject B in this task will be executed. In the latter case, Subject A will receive the “other” payoff of the choice of Subject B.
In the Decision-Making task, the “cents” that the subjects could give to themselves and to the other participant are used as currency. These “cents” are exchange into euros at the end of the experiment, where 100 cents are equal to 1 euro. So, in total the subjects could maximum earn 19,60 euros, which is 10 euros in the Matrix task and a maximum of 960 cents, so 9,60 euros in the Decision-Making task. The minimum that the subject could earn is 0 euros, that is when the subject do not get the bonus in the Matrix task and when “Subject B” choice is executed and “Subject A” gives nothing to the “other”.
3.2.5 Subjects
The experiment was conducted in English at the CREED laboratory at the University of Amsterdam on 21 of June 2017 with a total of 42 subjects recruited by the CREED subject pool. In the first session 20 participants showed up and in second session 22 participants showed up. The average age of the participants is 22.8 years and 71.4% of the participants are female. Only 11 participants are Dutch, so 26.2 %. Most of the participants are within the study field economic and business, namely 71.4%. Out of the 42 participants, 80.6% participate more than once in a CREED experiment before, for 11.9% of the participants it was their second experiment. Only 7.1% of the subjects participated for the first time in a CREED experiment.
3.3 Limitations Experiment
During the first session, there were some technical problems regarding the experiment. The first problem happened in the Matrix task. The clock was re-setting itself during this task after filling in the answer, so the clock jumped back to a fixed number and the participant was unable to finish the Matrix task within that block. This happened with two subjects, luckily there were no missing data values for these two participants within this session.
The second problem also happened in the first session and again with the Matrix task. During one block of five minutes, the clock jumped to one hour instead of the maximum of five minutes. This specific subject could not continue with the experiment before finishing this block of one hour, therefore it was necessary to re-set the clock for this participant. Luckily this subject already completed enough matrices correctly to eligible for the bonus.
The last problem that was experienced in the first session of the experiment was that two participants only solved ten Decision-Making tasks instead of the total number of twelve. However, this happened in both treatments twice so that each treatment has two missing values. In the second session, there were no technical problems.
3.4 Hypotheses
The first expectation of interest is the effect of time pressure on (pro) social behavior. The second expectation of interest is if decisions violate more axioms of revealed preferences.
Since there is mixed evidence found on how individuals behave under time pressure, the first hypothesis is tested two-sided. The first expectation is that there is a difference in (pro) social behavior between the Time and No-time treatment. To verify these Social
Behavior Hypotheses, the following null hypothesis and alternative hypothesis are specified:
H0: Time pressure does not affect (pro) social behavior, i.e.
!"#$%, '()# '+#,-)#.- = !"#$%, 01 − -()# '+#,-)#.- !1-ℎ#+, '()# '+#,-)#.- = !1-ℎ#+, 01 − -()# '+#,-)#.- H1: Time pressure does affects (pro) social behavior, i.e.
!"#$%, '()# '+#,-)#.- ≠ !"#$%, 01 − -()# '+#,-)#.- !1-ℎ#+, '()# '+#,-)#.- ≠ !1-ℎ#+, 01 − -()# '+#,-)#.-
The second expectation and hypothesis is in line with the research of Recalde, Riedl and Vesterlund (2015), assuming that there are more violations of revealed preferences in the Time treatment compared to the No-time treatment. To verify these Consistency Hypotheses, the following null and alternative hypotheses are specified:
H0: No difference in the violations of revealed preferences between the two treatments. !5(1$,-(1.", '()# '+#,-)#.- = !5(1$,-(1.", 01 − -()# '+#,-)#.- H1: More violations of revealed preferences in the Time treatment.
!5(1$,-(1.", '()# '+#,-)#.- > !5(1$,-(1.", 01 − -()# '+#,-)#.- 3.4.1 Manipulation Checks
The first manipulation check, named Time manipulation check, expects that decisions in the modified dictator game are made faster in the Time treatment compared to the No-time treatment. The second manipulation check, named Click manipulation check, expects that there are less clicks on the budget line in the Time treatment compared to the No-time
treatment. For the last manipulations check, named Matrix manipulation check, it is expected that less matrices are solved in the Time treatment.
4 Results
This section outlines and discusses the main findings of the experiment and consists of five parts. This first part discusses the descriptive statistics contained from the questionnaire to ensure that the subjects in the Time treatment do not differ from the No-time treatment and includes an analysis relating to the control question. The second part outlines the
manipulation checks regarding this experiment. In the third part the results from the Decision-Making tasks are discussed, including differences in individual preferences and outlines the Social Behavior hypothesis. The fourth part discusses the linear regressions and the final part outlines the Consistency hypothesis.
4.1 Descriptive statistics questionnaire
Table 2 summarizes the characteristics of the subjects per treatment. Besides the
demographic questions such as age, gender, experience and nationality, the questionnaire also consists of one control question and questions relating to the tasks of the experiment. The table provides additional insight into questionnaire and as can been seen, there are no
statistical differences between the two treatments with respect to age, gender, schooling, prior experience with a CREED experiment and nationality. It can be concluded that for both treatments there are no significant differences in sample demographics.
In each treatment’s instructions, there is a sentence that indicates which treatment group the subjects belongs to. In the Time treatment, this sentence reads: “The time limit of 5 minutes does include the time you spend on the Decision-Making task; so the clock with the remaining time in that block (at the top right of your screen) continues when you are in the Decision-making task”, in the No-time treatment this sentence reads: “The time limit of 5 minutes does not include the time you spend on the Decision-Making task; so the clock with the remaining time in that block (at the top right of your screen) stops when you are in the Decision-making task, and will continue when you are back at the Matrix task”. These sentences are included in the instructions to make the subjects aware of how the clock reacts during the experiment. The questionnaire contains a control question to check whether the participants were aware of how the clock reacted during the Decision-Making part. Subjects could either answer the question with: “The clock continued during the Decision-Making task”, “The clock stopped during the Decision-Making task” or “I don’t know”. Out of the 42
participants in the Time treatment and three participants in the No-time treatment answered the question with “I don’t know”. One participant in the Time treatment and three participants in the No-time treatment incorrectly answered the question. An incorrect answer is, for
instance when a participant is in Time treatment but answered that the clock stopped ticking during the experiment. Since more than half of the participants answered the question with “I don’t know”, it can be questioned whether the time pressure really worked on the participants, which will be revisit in the discussion of this paper.
The second part of the questionnaire consists of questions related to the task of the experiment. The questions are presented on a scale to determine how difficult, fun or stressful the tasks were for the subjects. On a rating scale from one to seven, subjects could assign how they experienced the tasks, 1indictating “not at all” and 7 indicating “very much so”. The subjects rated the stressfulness of the Matrix task equally in both treatments. In the Time treatment, the subjects found the Matrix task slightly more difficult but less fun compared to the No-time treatment. These differences do not significantly differ between the two
treatments. For the Decision-Making task the subjects found it more difficult and stressful in the Time treatment compared to the No-time treatment. The subjects rated the level of fun for the Decision-Making task lower in the Time treatment, however all these results do not significantly differ between both treatments. It can be concluded that the subjects experience the same level of difficulty, stress and fun among both treatments.
Table 2: Descriptive statistics questionnaire. Time T N=21 No-time T N=21 Mann-Whitney test
Mean Min Max Mean Min Max p-value
Age 22.33 (0.67) 19 30 23.2 (0.84) 19 35 0.454 Female 0.81 (0.09) 0 1 0.62 (0.11) 0 1 0.306 Study 0.71 (0.10) 0 1 0.71 (0.10) 0 1 1.000 Yearbegin 2014.95 (0.48) 2007 2017 2014.91 (0.36) 2011 2017 0.627 YearEnd 2018 (0.230 2017 2021 2017.81 (0.21) 2017 2020 0.553 Creed 2.71 (0.12) 1 3 2.76 (0.14) 1 3 0.688 Internat 0.71 (0.10) 0 1 0.76 (0.10) 0 1 1.000 Clock 1.71 (0.21) 1 3 2.00 (0.12) 1 3 0.166 Mdiff 3.52 (0.25) 2 5 3.24 (0.31) 1 6 0.464 Mfun 4.14 (0.29) 1 6 4.71 (0.35) 1 7 0.172 MStress 4.29 (0.40) 1 7 4.29 (0.42) 0 7 0.945 Gdiff 2.33 (0.33) 1 6 1.86 (0.28) 1 5 0.333 Gfun 2.76 (0.36) 1 7 3.19 (0.31) 1 6 0.286 Gstress 2.19 (0.30) 1 6 2.05 (0.33) 0 6 0.554
Note: the rank for the rating variables are from 1 to 7, when there is a zero as ‘min’ this means a missing variable. Standard errors are in the parentheses. The p-value is two-tailed.
4.2 Manipulation Checks
This part discusses the manipulations checks, followed by further discussion of the results. The Time manipulation check reads that decisions in modified dictator games are made quicker in Time treatment compared to the No-time treatment. In the Time treatment, the clock continues ticking during the Decision-Making task and when subjects take a longer time to make their decisions, they have less time solving the matrices. It is therefore expected that subjects will minimize their decision time in the Decision-Making task. Table 3 presents the results necessary for this manipulation check.
Table 3: Descriptive statistics Decision-Making task
Time Treatment No-time Treatment Mann-Whitney test
Mean Min Max Mean Min Max p-value
Self 412.42 (12.74) 166 960 453.34 (14.68) 102 960 0.046 **
Other 100.42 (5.96) 0 392 79.04 (6.61) 0 516 0.000 ***
Time 6.34 (0.34) 2 38 9.82 (0.50) 2 69 0.000 ***
Clicks 2.56 (0.18) 1 21 2.78 (0.17) 1 16 0.069 *
Note: * indicates statistical significance at 10% level, ** at 5% level, and *** at 1% level. Standard errors are in the parentheses.
The average number of seconds that subjects took for the Decision-Making task in the Time treatment is 6.34 seconds and in the No-time treatment subjects took on average 9.82 seconds. The difference between the mean for the variable “time” is negative and significant with a p-value of 0.000, so that the Mann-Whitney non-parametric test can reject the null hypothesis of the Time manipulation check, showing that this first manipulation check is successful.
The Click manipulation check reads that the average number of clicks made by the subject would be less for the Time treatment compared to the No-time treatment. Considering that subjects take less time to make their decision in the modified dictator game, it is also expected that they try-out less points on the budget lines as trying-out different points on the budget line would take more time. In the Time treatment, the average number of clicks is 2.56 and in the No-time treatment the average number of clicks is 2.78. Results show that indeed in Time treatment, subjects try-out less points in the Decision-Making task compared to the No-time treatment. This result is weakly significant at a level of 10% significance level and has a p-value of 0.069.
The Matrix manipulation check reads that subjects solve less matrices in the Time treatment compared to the No-time treatment. This is expected because, in the Time treatment the clock continues ticking during the Decision-Making task, therefore there will be less time for completing the matrices. The average amount of completed matrices in the Time treatment is 5.76 and in the No-time treatment 6.07, both independently of whether the matrices were answered correctly. Table 4 presents the descriptive statistics of the Matrix task for both treatments.
The total number of observations is 84 for each treatment, that is four parts multiplied by 21 subjects for each treatment. Results show that the average number of completed
difference in the mean of the variable “Complete Matrices” does not significantly differ between both treatments and has a p-value of 0.990. The null hypotheses of the Matrix manipulation check cannot be rejected. Therefore, this manipulation check is not successful. Even though in the Time treatment on average less matrices are solved, the average number of matrices that are solved correctly is slightly higher in this treatment. This difference in the mean of the variable “Correct” is not significant between the two treatments (p-value of 0.669). The table also shows the average number of matrices that are incorrectly completed is not significant with a p-value of 0.241.
Table 4: Descriptive statistics Matrix Task
Time Treatment No-time Treatment Mann-Whitney test
Mean Min Max Mean Min Max p-value
Complete Matrices
5.76 (0.16) 1 10 6.07 (0.22) 4 15 0.990
Correct 3.77 (0.13 0 7 3.70 (0.16) 1 11 0.669
Wrong 1.99 (0.15) 0 7 2.37 (0.20) 0 9 0.241
Note: Standard errors are in the parentheses
4.2.1 Problem Matrix task
In the first session of the experiment, subject number 16 faced a technical problem with the Matrix task. During the second part of the Matrix task, the clock switched to one hour instead of the normal limit of five minutes per block. Therefore, subject number 16 could complete more matrices within that block that time normally allows. The average number of completed matrices within one block in the No-time treatment is 6.07, so changing the number of
completed matrices within block two for subject 16 to 6 matrices, with the Mann-Whitney test the difference between the two treatments will show better results. The difference for the variable “Completed Matrices” is still not significant with a p-value 0.896. The null hypothesis for the Matrix manipulation check cannot be rejected.
4.3 Decision-Making task
Given that the demographics and rating level variables for both treatments and tasks do not significantly differ, it is possible to compare both treatments with respect to the
Decision-Making task. Table 3 reports the main descriptive statistics of this task for both treatments and includes a Mann-Whitney non-parametric test to check if there is a difference between the two treatments. This table indicates that participants in the Time treatment act less selfish compared to the No-time treatment. Furthermore, it shows that participants use on average less time to pick their final point for the Decision-Making task in the Time treatment
compared to the No-time treatment. In addition, the number of tries in the Time treatment is again less than in the No-time treatment. Both treatments include 250 observations based on twelve budget lines for 21 subjects which give 252 observations, however each treatment has two missing values.
The results in Table 3 show a difference between the two treatments in the mean of the variable “self” and “other”. The variable “self” indicates the amount that subjects give to themselves and the variable “other” stands for the amount subjects gives away. The average amount subjects give to themselves in the Time treatment is 412.42 cents and the average amount they give to the other is 100.42 cents. For the No-time treatment, the average amount that subjects assigned to themselves is 453.34 cents and the total amount they give to the other is 79.04 cents.
Social Behavior Hypothesis: Time does pressure affects (pro) social behavior.
To test the first hypothesis, the difference between the two treatments for the variable “self” and “other” is analyzed with a Mann-Whitney non-parametric test. The Mann-Whitney non-parametric test supports the stated hypothesis for the variable “self” with a p-value of 0.046, thus the null hypothesis can be rejected at a significance level of 5%. Together with the variable “other”, the difference between both treatments is positive and significant with a p-value of 0.000 and hence the null hypothesis can be rejected at a significance level of 1%. This evidence shows that subjects are more (pro) social towards their fellow participants in the Time treatment compared to the No-time treatment and it can therefore be concluded that time pressure positively affects (pro) social behavior.
Both treatments have two missing variables. In the Time treatment, decisions two and three for subject number seven and in the No-time treatment decisions seven and twelve for subject number eight. The maximum amount that subject seven could assigned to him/herself or to the other with these two “missing” budget lines is 1170 and 1140 cents. The maximum amount that subject eight could assign to him/herself or to the other with these two budget lines is 720 and 1140 cents. Excluding these two subjects prior to the data set, so that the total
number of observations is 240 per treatment and applying the Mann-Whitney non-parametric test, the differences are still significant for both variables “self” and “other. With a p-value of 0.043 for the variable “self”, the first hypothesis can be rejected at a 5% significance level. For the variable “other”, the hypothesis can be rejected at a 1% significance level (p-value: 0.000).
4.3.1 Individual Preferences
There are different individual preferences of subjects demonstrated in the experiment. Participants acted either perfectly selfish, close to perfect selfishness, or more towards a fair division of the total amount, that could be divided between two participants. Some
participants also made decisions that characterizes altruistic behavior. Table 5 gives an overview of decisions that have a perfectly selfish, fair or altruistic character.
The definition of perfect selfishness is when a subject assigns the total amount that must be divided to themselves and gives nothing to the other. The definition of perfect fairness is when there is an exact equal division of the total amount that must be divided. Altruistic behavior is when subjects give more money to the other than they assign to themselves.
Table 5: Individual Preferences
Time Treatment No-Time Treatment
Completely Selfish 18 72
Completely Fair 10 15
Altruistic behavior 12 6
Other 200 147
Note: n=240, leaving subjects 7 & 8 out of the data.
The results in Table 5 shows a difference in individual preferences between the two
treatments. In the Time treatment, eighteen decisions were perfectly selfish, one participant made twelve perfectly selfish decisions and two other participants made one or more perfectly selfish decisions. Furthermore, ten decisions in the Time treatment and fifteen decision in the No-time treatment have a perfectly fair character, four participants in the Time treatment and four participants in the No-time treatment, respectively. Additionally, six more decisions with an altruistic character were made in the Time treatment compared to the No-time treatments,
representing four participants in the Time treatment and two participants in the No-time treatment.
Figure 5 shows three decision-making sets of three participants where these behavior characteristics are found. Decisions that are perfectly selfish are shown in Figure 5(a). All decisions are made on the extremes of the budget lines that intersect with the x-as. Here subject five of the No-time treatment showed perfectly selfish behavior. Decisions with an altruistic character are shown in Figure 5(b). This subject chooses points on budget lines three, four, seven and ten where he/she gave more money to the other than to him/herself. Figure 5(c) displays fair and close to fair decisions. Only budget line five shows a perfectly fair division of the total amount of 360 cents.
4.4 Linear Regression
In this part, the results of the regression analyses are discussed. In addition to the Mann-Whitney non-parametric tests, these analyses describe how (pro) social behavior is related to the demographic and rating level variables. Furthermore, this part estimates the effect of (pro) social behavior on decision time.
To check whether selfishness in both treatments could be predicted by demographic and rating level variables, an OLS regression is conducted. For these two regression analyses, the level of selfishness is taken as a dependent variable, that is the total amount that subjects assign to themselves, giving 40 observations. The variable “self” is divided by 100, the unit for this variable is euros instead of cents. Subject numbers seven and eight from the first experimental session are excluded prior to the regression analysis, since both subjects have two missing values. The two regressions in Table 6 estimate the effect of selfishness on the demographic and rating level variables. The explanatory variables that are included in this regression are defined in Table 10 (see Appendix B).
Table 6: OLS Regression: Level of selfishness Dependent = Self/100 Regression 1 Regression 2 Time Treatment -2.85 (3.87) -9.87 (6.01) Age -0.42 (0.81) Female -0.55 (4.44) Economics -2.00 (9.92) International -1.73 (3.79) Yearbegin -0.95 (1.31) Yearend 1.37 (4.02) CreedMore7 16.09 (2.88) *** 17.91 (4.82) *** CreedOnce7 0.14 (6.01) 3.54 (9.40) ClockCont8 3.38 (6.36) ClockStop8 -7.88 (6.01) Mdiff 1.94 (2.20) Mfun 2.27 (1.99) Mstress -1.68 (1.08) Gdiff -3.69 (1.11) *** -3.80 (1.81) ** Gfun -0.51 (1.10) Gstress -0.70 (1.75) Constant 47.62 (3.46) *** -800.87 (7290.30)
Note: ** indicates statistical significance at 5% level and *** at 1%. Standard errors are in the parentheses. a The Creed question is a nominal variable with three possible answers. For the regression, binary variables are created.
b The Clock question is a nominal variable with also three possible answers. For this regression, binary variables are created.
By leaving both Creed No; which indicates when subjects did not participate before in an experiment and Clock don’t know; which indicated the other possible answer, out of the regression omitted variable bias is being avoided.
As can be seen in Table 6, this regression exposes that no background demographic features influence the level of selfishness. The Social Behavior hypothesis implicates that time
pressure on affects (pro) social behavior. As can be seen in the previous section and in Table 3, the mean differences for variable “self” is negative and significant between both treatments at a 5% significance level. Table 6 shows, in line with previous results, that the regression Dit hangt nauw samen met de
treatment; kan je niet beiden in1 regressie stoppen
coefficient of the dummy variable Time treatment is negative. This indicates that participants in the No-time treatment acted more selfishly compared to the Time treatment. The
coefficient in both regression analyses do not significantly differ from zero. Consequently, the Social Behavior hypothesis cannot be confirmed based on the respective p-values, 0.465 and 0.115, respectively.
The estimation coefficients of the variable CreedMore are positive, 16.09 and 17.91, respectively. Both coefficients are significant at a 1% significance level, this indicates that subjects who participated more than once in CREED experiments acted more selfishly than subjects for whom this was the first time participating, or had only participated once before. Furthermore, the table shows that the difficulty of the Decision-Making task has a significant effect on the level of selfishness. This expresses that the harder the subject found this specific task, the more generously they acted. Other demographic and rating variables do not have a significantly effect on the level of selfishness.
Table 7 shows the estimated regression with the level of selfishness as the dependent variable. This regression is based upon panel data, with fixed effects for the twelve Decision-Making tasks per subject. As each subject face the same twelve Decision-Decision-Making tasks, this regression focusses on the average level of selfishness and test deviations from the mean differences for these tasks. There are twelve observations per participant, so the total number of observations equals 480. The variable Time is the number of seconds the subjects took for each Decision-making task. The dummy variable Time treatment indicates the treatment and the interaction variable TimexTreatment is the change in the effect of decision time on the specific treatment.
As can be seen in Table 7, the regression coefficient shows that the explanatory variable decision time is negative and significant at a 5% significance level. This indicates that the more time subjects took in the Decision-Making tasks, the more generous they seem to be. This result seems contradictory to the results previously found. Moreover, the
regression coefficient of the dummy variable Time treatment is positive, which indicates that subjects who are in the Time treatment, will act more selfishly. However, this result is not significant with a p-value of 0.710. The interaction variable decision time and Time treatment is negative and significant with a p-value of 0.000. Implying that there is a negative relation between decision time and the level of selfishness and a negative relation between decision time and Time treatment. This relation becomes stronger when the estimation coefficient of the Time treatment increases. There is a positive relationship between Time treatment and the
This indicates that when subjects take a longer time for the Decision-Making task, the estimation coefficient of decision time decreases.
Table 7: Regression: Level of selfishness Dependent = Self/100 Regression 3 Decision Time -0.030 (0.011) **
Time Treatment 0.074 (0.201)
DT x TT -0.095 (0.021) ***
Constant 4.812 (0.142) ***
Note: ** indicates statistical significance at 5% level and *** at 1%. Standard errors are in the parentheses.
4.5 Interpretation of results
The results in Table 7 contradict earlier results from the Decision-Making task. To clarify the differences between the results; in Table 3 and 7, Figure 6 shows a scatterplot for each persons’ decisions together with the number of seconds relating to these decisions. The horizontal axis shows the total amount of seconds a subject needed to make all decisions. The vertical axis indicates the total amount of money a subject assigned to him/herself, the blue points indicating the Time treatment and the grey points the No-time treatment. As can be seen in Figure 6, behaving more socially takes more time, this applies for both treatments. Furthermore, selfish decisions are made more quickly in the Time treatment compared to the No-time treatment. Besides, a participant that behaves less selfishly takes a longer time to decide. Moreover, in the No-time treatment more perfectly selfish decisions are made.
In conclusion, time pressure indeed leads to (pro) social behavior. This result is in line with results from Rand et al. (2012) where they show that more contributions are being made under high time pressure. Behaving more socially takes more time, because subjects must pay attention to their own payoffs and the payoffs of the other. In addition, subjects who behave more selfishly make faster decisions, since they only have to care about themselves.
Moreover, Kahneman (2003) explains, in the two system, theory that decisions in general are influenced by an intuitive and deliberative system. In the research from Rand et al. (2012), decisions are forced to be made within 10 seconds and subjects decide more intuitively. The results from this study show that decisions cannot be linked directly to one of the two systems, since this experiment does not use a strict time limit but creates a general time pressure environment. It cannot be directly said that decisions in the Time treatment are made intuitively and that decisions in the No-time treatment are more deliberative decisions.
Even if subjects prefer to be (pro) social and choose intuitively, behaving more (pro) social takes more time than behaving selfishly.
Figure 6: Scatterplot decisions for both treatments
4.6 Violations of Revealed Preferences
Table 8 shows the descriptive statistics needed for the second hypothesis.
Table 8: Violations of revealed preferences.
Time Treatment No-time Treatment Binominal test
Sum Mean Sum Mean p-value (2-tailed)
Violations 10 0.50 (0.12) 3 0.15 (0.08) 0.038**
Note: ** indicates statistical significance at 5% level, Standard errors are in the parentheses.
Hypothesis 2: More violations of axioms of revealed preferences under time pressure.
To check which subject violated the axioms of revealed preference it is first necessary to determine the maximum amount per budget set that subjects can assign to themselves and to the other, assuming that the choice that the subject makes falls within the budget set such that:
9:,;<=>+ @A9:,;<=>
@A9:,BCD<E9:,BCD<E = @A9:,;<=>
0 1000 2000 3000 4000 5000 6000 7000 0 50 100 150 200 250 300 To ta l S el f Time in seconds
Then bundle j is then revealed preferred to bundle i if:
9F,;<=> + @A9:,;<=>
@A9:,BCD<E9F,BCD<E ≤ @A9:,;<=>
And bundle j is strictly revealed preferred to bundle i if:
9F,;<=> + @A9:,;<=>
@A9:,BCD<E9F,BCD<E < @A9:,;<=>
Furthermore, knowing this for each combination of budget lines a matrix can be made. On the horizontal axis, the numbers one to twelve represents the twelve budget sets. Similarly on the vertical axis the numbers one to twelve representing the twelve budget sets. Two matrices can be made where:
@,-+(I J:F = 1,0, if ( revealed preferred to Wif not
@,-+(I Z:F = 1,0, if ( strictly revealed preferred to Wif not
Checking if bundle i has indirectly revealed combinations, matrix J:F must be squared till there is also a matrix J^ and J^_` , where “k” is the quadratic power which has a maximum of twelve, representing the twelve budget lines.
J^_` :F = 1 als J^ :=J=F ≥ 1 C =b` 0 als J^ :=J=F = 0 C =b`
There is a violation of revealed preferences if JC
:FZF: = 1, were n = 12.
Table 8 shows that in both sessions, dictators violated the axioms of revealed
preferences and thus make inconsistent decisions. For example, subject thirteen from session two behaves irrationally. It can be easily checked whether there is a violation by looking at the different point that this subject has chosen. Figure 7(a) shows such violations between budget set 2 (720,240), 3 (450,900), 6 (900,450) and 11 (960,240), where the first three-digit
number is the number on the x-as and the second three-digit number is the number on the y-as.1 There are four allocations, labelled A, B, C and D that are lying on four budget lines that fall within these budget sets. Figure 7(a) shows that A is revealed preferable to B, but also that B is revealed preferable to A. Also, C is revealed preferable to D, but D is also revealed preferable to C. The budget sets are directly revealed preferred to each other. Subject thirteen violates the axioms of revealed preferences.
An example of a clear violation of GARP can be seen in Figure 7(b). This figure shows that allocation A is revealed preferable to allocation B and that allocation B is revealed preferable to allocation C. By this reason, allocation A is indirectly revealed preferable to C, however budget C is strictly directly revealed preferable to budget A. Figure 7(b) shows a clear violation of Generalized Axiom of Revealed Preference.
Figure 7(a): Violations of Revealed Preferences
Figure 7(b): Violation of GARP
To test the second hypothesis, the two treatments are tested against each other on the number of violations that occur in each treatment. Each subject that has one or more violations receives a one on violations given if subjects make one “mistake” it is possible that more decisions also violate the axioms of revealed preferences. Knowing which subjects face a violation, a binominal test could be performed to check whether there is a difference between the two treatment with respect to the number of violations. The results can be seen in Table 8.
The second hypothesis states that there are more violations of axioms of revealed preferences in the time pressure treatment. In line with the research from Recalde et al. (2015) results show that subjects in the Time treatment are more prone to violations than subjects in the No-time treatment. This result is significant at a 5% significance level.
Table 9 shows a correlation check between the variables decision time, self and other with respect to the number of violations. As can be seen for the Time treatment, there is a negative correlation between decision time and the number of violations. This implies that when subjects take more time during the Decision-Making task, less violations are made in the Time treatment. Nevertheless, this result is not consistent for the No-time treatment group, here results show a positive correlation between the two variables. Both correlations are highly insignificant, which suggest that there is possibly no correlation between the two variables decision time and violations. As can be seen in the table, there is a negative correlation between the number of violations and the level of selfishness. Acting more
selfishly would lead to less violations of the axioms of revealed preferences coupled with the positive correlation for the generosity level. Only in the No-time treatment the correlations are significant at a 10% significance level, which indicates a weak correlation between the
variables. When someone acts perfectly selfish, as can be seen in Figure 5(a), he/she also behaves rationally and consistently. It could be concluded that perfectly selfish behavior leads to rationality. When subjects allow for generosity, there is greater chance for irrational
behavior. It could also be the case that generosity is indicated as a mistake in preferences. The latter seems to be unlikely since subjects’ show similarity between preferences within the budget sets.
Table 9: Correlations time, self and other regarding violations. Time Treatment No-time Treatment
Violations Violations Decision Time -0.032 (0.892) 0.301 (0.185)
Self -0.205 (0.373) -0.401 (0.072) *
Other 0.236 (0.303) 0.381 (0.088) *
Note: * indicates a significance level of 10%. The p-values are in the parentheses.
5 Discussion and Conclusion
This section outlines the summary of the results. Furthermore, extra interpretations and explanations of results are given and some limitations of this experiment are outlined. Finally, some thoughts are given for future research.
As can be seen in the section results, the Mann-Whitney non-parametric test shows no significant difference in demographic and rating level variables between both treatments. This indicates that there is homogeneity between treatments with respect to the subjects.
Furthermore, the two manipulation checks, time and click, are both successful, showing that in the Time treatment subjects decide significantly faster in the Decision-Making tasks when compared to the No-time treatment. The Click manipulation check is weakly significant, showing that less clicks in the Time treatment are made and hence it can be concluded that subjects who decide more quickly in the Time treatment also use less clicks when making their final decisions. The Matrix manipulation check was not successful and did not
for both treatments seem equal, a non-parametric test on the level of selfishness is performed. Results show that subjects tend to be more generous in the Time treatment compared to the No-time treatment. The Social Behavior hypothesis states that (pro) social behavior is affected by time pressure. Given that the Time manipulation check, is successful, it can be concluded that differences in the level of selfishness are due to time pressure. Therefore, the Social Behavior hypothesis is confirmed.
The Consistency hypothesis states that more violations of the axioms of revealed preferences are made under time pressure. Evidence from the experiment shows that more violations are made in the Time treatment, this result is significant at a 5% significance level. It can be concluded that decisions made under time pressure tend to be more irrational and inconsistent. The correlation check reveals a negative correlation between the level of selfishness and the number of violations, which implies that when subjects behave less selfish, the chances of behaving irrational are higher. Alternatively, subjects who act more selfishly show less violations of revealed preferences. Additionally, the third estimated regression demonstrated a negative correlation between the level of selfishness and decision time. This indicates that faster decisions will lead to more selfishly behavior. However, earlier results demonstrated that under time pressure subjects behave more generous. In conclusion, behaving more generous takes more time, namely subjects must pay attention to both their own payoffs and the payoffs of someone else. Therefore, it can be concluded that under time pressure subjects behave more generously, but that generosity takes principally more time. Different from the experiment of Rand et al. (2012), it cannot be easily said that people behave more intuitively under time pressure. Since, this experiment does not use a strict time limit in the Decision-Making task, but creates an overall time pressure environment.
Some limitations of the experiment are discussed. Firstly, the experiment consisted of one control question relating to the treatments. Subjects had to answer how the clock reacted during the Decisions-Making task. The answers given this question were “Continued”, “Stopped” or “I don’t know”. More than half of the subjects did not know how the clock operated during this specific task. In the No-time treatment most subjects were aware of the process of the clock and answered the control question correctly. An explanation for this could be that subjects only noticed the clock and how it operated in the No-time treatment because of the unusual behavior of time. Normally, time continuous ticking in everyday life, so when the clock suddenly stops during a task, this is more noticeable. Nonetheless, the instructions included a sentence that clearly mentioned how the clock operated during the Decision-Making task so the correct answer was already given to the subjects. To this extent,
