Paying or Paying for It?

Published on February 2017 | Categories: Documents | Downloads: 53 | Comments: 0 | Views: 389
of 25
Download PDF   Embed   Report

Comments

Content

Paying, or Paying for It?
Sean McCoy, Alexander Egan, Jonathan Maesato, and Richard Winkeller* December 19, 2013

Abstract
We investigate the forces driving takers’ actions in a modified power-totake game in which takers must explain their actions or pay to avoid doing so. Three groups of takers are tested against one group of receivers; we vary both the timing of informing takers that they will need to explain themselves and the takers’ ability to pay to avoid an explanation. We find that takers’ willingness to pay (WTP) to avoid explanation stays roughly the same whether they are told they can pay to avoid explaining themselves before or during the experiment. We observe that Big 5 personality traits affect each treatment differently with respect to WTP, take rate, happiness, and the qualities of the messages sent. Specifically, takers who are told mid-experiment that they have to explain themselves report a higher level of neuroticism than those told prior. All participants reported feeling less extraverted after the experiment; this was largely brought about by changes in neuroticism and agreeableness during the experiment. WTP did not change over different income ranges, but happiness had a large, positive effect on WTP. For takers whose messages include the word “sorry,” WTP falls and take rate and length rise, all by a large margin. When “fairness” appears, WTP, take rate, and happiness fall while length rises to a similarly large extent. We assert that these findings extensively influence social behavior: Consideration of future explanations, or avoidance of such scenarios, appears to have a significant monetary value to takers in the present that can be exploited in politics or simply in everyday interactions with one’s own family and neighbors.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
*

All are undergraduates at Vassar College, Poughkeepsie, NY. Email at: [email protected], [email protected], [email protected], and [email protected], respectively.

1

Introduction
People often behave in ways that, at first, do not seem rational. In reality,

we make decisions contrary to the laws of economics, and we often attribute these deviations to emotion. For example, knowing that your actions may directly affect someone else may influence your behavior, and knowing that you may have to explain your actions in the future may cause you to change your behavior in the present. With these ideas in mind, people act differently in different situations, causing their behavior to seem irrational at times. How can we reconcile seemingly irrational behavior with seemingly rational thought processes? Where exactly do these irrationalities lie, and are they irrationalities at all? Imagine that you have done something bad that will negatively impact your spouse. This is not a good situation. You want to minimize the damage done while keeping yourself reasonably happy. How should you respond? A simple apology might not suffice. Often, people in this situation will turn to buying something, paying in some form, to make up for their behavior. So, how does dinner sound? In this paper, we explore the behavior of takers in a power-to-take game. Others (Grosskopf and López-Vargas, 2013) have researched receivers in a similar scenario. However, we do not study takers simply to complement their findings. Takers, unlike receivers, are inherently selfish. People often think to act to serve their own self-interest before considering how their actions will affect others.

Would buying dinner make you feel better about telling your spouse bad news? Probably. Did you think about how your spouse would feel at dinner? Probably not yet, but you might argue that you will worry about that soon enough. Either way, others’ feelings initially take a back seat so far as many people are concerned; in a way, we are takers before we are anything else. Thus, we attempt to determine whether takers are willing to pay to avoid explaining their actions, and, if so, we consider the factors that influence their decision-making. In addition, we study the messages they send and look for relationships between their words and their actions. In particular, we are interested in take rate, WTP to avoid explaining one’s actions, explanations of actions, and changes in happiness.

2

Literature Review
We draw our inspiration primarily from two papers. The first (Bosman

and van Winden, 2002), investigates responses to emotion in a classical power-totake game. The study focuses on the ability of emotions to generate behavior. Prior to the game, the subjects perform an effortful task to earn money. After this, in the first stage of the game, the taker elects to take a percentage of the responder’s income. In the second phase, the responder decides on some amount of the their total income to destroy. Thus, while the take rate remains the same, the actual amount the taker receives is decreased at the cost of the responder having a portion of his or her own income destroyed as well.

Using the results of the destruction rates as revealed preferences and selfreported emotional states, they find that allowing the responder to self-select a destruction rate, rather than having a choice between no money or all the money, gives a more precise measure of emotional response than a dictator game. In addition, the experiment requires that both players perform an effortful task to earn income and avoid problems caused by the endowment effect. The authors discuss the concept of emotional hazard, when a person retaliates against an action done to him or her. In their experiment, the emotional hazard is measured by the destruction rate, as it is an emotional response to money being taken from the responder. Our experiment embraces the concept of emotional hazard with the recognition that there can be more to the cost of causing emotional hazard than having money destroyed. There are nonmonetary factors in emotional hazard, and we try to provoke these by having takers explain their behavior in our experiment. We predict that takers will pay to avoid explaining themselves to receivers, as they may view the non-monetary emotional hazard of the other player’s judgment as a heavy cost that they wish to avoid. Even more important to our study was the work of Grosskopf and LópezVargas, who explore the often-overlooked topic of emotional expression and the value of ex-post verbal communication in a power-to-take game. The authors use a game in which one party (T) can take resources from the other party (P). Players receive a $3 endowment for their participation. They then perform effortful tasks that garner them additional income; these tasks are calibrated so

that the average profit is around $10.35, totaling $13.35. Next, a T player decides what percentages of a given P player’s income (what part of the $10.35) to take. Then, player P is told about T’s decision and asked to report the maximum amount he would be willing to pay to send a message to T. After stating this value, an “actual price” is randomly drawn from a uniform distribution from $0 to $3. If P’s WTP exceeds the actual price, the message is sent for the aforementioned cost; if not, it is not sent and there is no cost to P. The $3 endowment guarantees that all P players are able to respond to T players if they so desire. The message sent falls into one of four categories: OFU (Only for You; messages sent directly to T player for a cost), SLTM (Somebody Listen to Me; messages sent to T player through neutral third party for a cost), FM (Free Message; messages sent to T at no cost), and NM (No Message; no message sent). Note that T players are informed of their partner’s type before they decide what percentage of their partner’s income to take. Their main findings were that expressions of this sort have material value, that more drastic changes in mood result in a greater value placed on expression, and that the anticipation of expression induces less selfish behavior. There was a considerable difference between take rates in the NM and FM trials: On average, the T player took 68% of the P player’s income when they knew no message could be sent and only 48% when they knew a free message could be sent.

3

Experiment
This study has three main goals: First, we determine whether people will

pay to avoid explaining their actions. Second, we investigate the factors that drive their behavior. Third, we analyze the messages they send in hopes of understanding why they behave in the ways that they do and whether their messages offer any valuable information about the senders.

3.1

Experimental Design
We test our hypotheses, detailed below, in a modified power-to-take game

in which both takers and receivers complete an effortful task for which they are then rewarded. Takers are then able to take any amount of the receivers’ earnings. The receivers cannot respond; they must accept the division of earnings as divided by the takers. Takers must then explain their actions unless, depending on the trial, they have the option to pay to avoid doing so. We conduct our experiment in four treatments, labeled trial 1-4 respectively. The first involves takers who know beforehand that they will have to explain themselves later but cannot pay to get out of doing so, whereas takers in the second trial know beforehand that they can pay to avoid explaining themselves later on. The third trial does not know that they will have to explain themselves until they reach that part of the experiment, but they also learn at that point that they can pay to avoid an explanation. The fourth trial is one of receivers only; we want to concentrate on the takers in the experiment, and since the

receivers in our design are only present for the sake of comparison—to see, for example, how happiness, message qualities, and other factors change with respect to the amount taken—we feel that playing receivers against aggregate data is acceptable. By conducting matters in this way, we gather far more data than were we to run a standard one-taker-versus-one-receiver experiment. The power-to-take game part of the experiment is preceded and followed by a survey. The first gathers basic demographic information, a starting level of happiness, data on Big 5 personality traits, and contains simple filter questions that double as the effortful task. Participants all answer the same survey and are all paid for getting through this part of the experiment. The aforementioned power-to-take game follows; each treatment receives a slightly different set of instructions pertaining to their treatment’s specific limitations (can pay, cannot pay, uninformed, receivers). The follow-up survey is similar for all participants. Big 5 personality trait and happiness data is collected again, along with a handful of questions specific to the participants’ trials. In general, this data was poorly collected—often in hypothetical questions—so it goes largely unused in our analyses. The full surveys are available in the Appendix of this paper.

3.2

Hypotheses
We have five hypotheses to explain how takers will respond. First, we

believe that takers are willing to pay to avoid explaining their actions. The guilt that they incur from having to explain their selfish actions is burdensome enough that they will want to avoid explaining it.

Second, we believe that takers act less selfishly when they know ahead of time that they must explain themselves later. These two hypotheses are tested by regressing WTP, take rate, and other variables on message length and comparing results by trial. Third, we believe that takers’ WTP is higher when the take rate is higher. We suspect that, as people take more, they will be willing to pay more to assuage their guilt. We measure this by comparing WTP and take rate regressions between trials. Fourth, we believe that explanations will vary with take rate. More specifically, as takers take more, we expect their messages to be longer and more apologetic. We measure this through regressions of several variables on message length and content, again sorting responses by trial. Fifth, we believe that takers’ explanations of their actions will vary over time. This relates to the hot and cold states of expression that occur during and after the experiment. In short, we expect that explanations given in the exit questionnaire will not match mid-game explanations. Since messages are extremely difficult to quantify, we have not yet fully supported or rejected this hypothesis, but we have regressed data of respondents whose messages contained certain words against that of those who did not. Though this is not the most complete analysis, we demonstrate the presence of strong effects of message quality on actions in the game through regression.

3.3

Logistics
We ran our experiment online using MTurk and SurveyMonkey in an

effort to maintain anonymous conditions, speed up the process of paying participant, and minimize time spent sorting and collecting data. Our survey was available worldwide to all MTurk users with a HIT approval rate of over 80%. Further restrictions could have been used, but we argue that, as our experiment deals with social interactions, severely limiting the number of people able to respond was not in our best interests. Participants were paid $0.10 for completing the HIT and another $0.10 for completing the survey. Takers could earn an additional $0.10 by taking receivers’ money; receivers could gain only the $0.10 from the effortful task section less whatever takers took from them. Trial 1 had 104 participants, trial 2 95, trial 3 96, and trial 4 83. Of the 83 receivers, roughly an eleventh played in a scenario against each possible amount taken ($0.00, $0.01, … $0.10). Data was collected on October 28 and 29, 2013. Surveys ran until roughly 100-120 participants had completed each trial. Data was then sorted. Data of participants who completed the surveys in less than 120 seconds was ignored; a majority of these finished in under 30 seconds, and compared to the 180 seconds it took us, the experimenters, to answer the survey, we conclude that these data were likely not thoughtfully reported. Participants were paid with respect to their responses on October 30 and 31. In total, we gathered approximately 500 responses of which 378 were analyzed. Nearly two thirds of responses were from America, one third from India, and 4% from other nations in Europe, North and South America, and the Middle East.

4.

Results
To test our first hypothesis, that takers are, in fact, willing to pay to

avoid explaining themselves, a t-test in Stata suffices: (Note that we explore this in more detail later on.) Table 1. T-test of WTP Versus 0.

For any reasonable p-value, we can safely say that takers are willing to pay to avoid explaining themselves. One possible explanation for this is that there is a cost involved in physically typing out message or in thinking of a message to type out. However, participants in our study were MTurk users, individuals who receive micropayments for doing similar tasks. So, we contend that inherent costs are minimal and that the act of explaining their behavior was the costly part. As for the second hypothesis, that takers change their behavior when they know they must explain themselves beforehand, we call a regression table (note that maleDummy is a binary sex variable, open1 etc. are measures of Big5 personality traits, take1 is take rate, and Informed etc. are the different trials):

Table 2. OLS Regression of Sex, Personality, and Trial on WTP and Take Rate

Note that in the leftmost column, Informed is not significant in predicting take1, the take rate from the first (incentivized) game. Thus, the statement “Takers take the same amount regardless of their knowledge (or lack thereof) of the fact that they will be required to explain themselves” cannot be statistically disproven. However, stopping at this simple hypothesis test would be insufficient. We are also curious if any factors are significant in predicting take rate under our different treatments. For this, we provide a second regression table:

Table 3. OLS Regression on Take Rate by Trial

Here we find a few interesting trends. First, an increase in age elicits a decrease in take rate. This is logical given that older people are generally wealthier than those who are younger; additionally, they are likely more aware of the wellbeing of others and therefore act less selfishly. This effect is strongest in

the first trial, suggesting that people who are aware of their future act more decisively than those who are not. The nationbinary variable (1 if American respondent, 0 if not) positively affects take rate in the two trials in which the takers have the option to avoid explaining themselves for a cost. Americans in trial 3 were particularly selfish, a trend that reappears several times in our analysis. We attribute the effects of employment to the “knowing where you stand” mentality; those who are employed are less willing to take less since they have a source of income and therefore can afford to let others keep their earnings and want to do so to avoid retribution. Happiness increases take rate since trial 1 takers know that they have to explain themselves later and thus see no reason not to take more since they cannot avoid those future explanations. Additionally, more agreeable takers were inclined to take less, perhaps suggesting that people have a predisposition to behave in line with their self-reported personalities (also, agreeable people might simply be less selfish than others). Finally, we note that WTP affects trial 3 more than trial 2, indicating that those who were previously uninformed about the explanation took more whereas those who knew about the explanation acted more conservatively. This finding supports our third hypothesis. Next we look more carefully at WTP to avoid explanations:

Table 4. OLS Regression on WTP by Trial.

Again we note that Americans were more selfish than international respondents. Education’s high significance implies those who were more educated took less in scenarios where they understand their future options; perhaps these people behave in a more sophisticated manner than the more naïve trial 3 takers. Employment behaves similarly: trial 2 takers understand the future and can therefore assess their status in the present to make a split-second decision in the

present about how to behave to benefit them most in the future. Incomebracket’s negative effect on WTP suggests that the rich are greedier, and happiness’s effect reflects that those who are informed are in a content enough state that they can pay a bit extra to avoid unnecessary guilt in the future. Finally, we again note that take rate and WTP are interrelated. Next we look at changes in happiness (dhappy = happiness at end of experiment – happiness at its start): Table 5. OLS Regression of Happiness by Trial

We note that men in trial 3 (and also trial 2) were happier than women. Partnered with the “greedy American” effect, this finding would indicate that American men may get more pleasure out of greed than those in other nations were it not for the overpower nationbinary finding that works in the opposite direction; this suggests that Americans were less happy in all trials as a result of their actions and makes the sex finding inconclusive. Those who were originally more conscientious reported a large increase in happiness by the experiment’s end, likely as a result of not feeling guilt or punishment for their actions. Most importantly, we find that takers are greatly less happy at the end of the experiment as a result of losing money. This is consistent with prospect theory and general punishment aversion mechanisms. We now explore more specifics of the messages sent: Table 6. OLS Regression on Length by Trial

We observe that older participants generally wrote shorter messages than their younger counterparts, perhaps out of impatience. The particularly high nationbinary numbers are telling: American receivers wrote 20 characters less than internationals, while takers in trials 2 and 3 split the difference, writing 20 fewer characters when they knew their future and 20 more when they did not. Happier receivers wrote more for aforementioned reasons, but happier takers tended to write less, again most likely relating to their belief that they know what will happen in the future, and a simple message will not change that fact. Those reporting high levels of openness wrote more, as did extraverted trial 3 takers. This finding is striking as all the other trials, even receivers, reported drops in message length with extraversion; it is probably best attributed to a punishment avoidance mechanism triggered by the trial’s specific “surprise explanation” conditions. People also wrote their messages after reporting initial levels of extraversion, so it is possible that those who thought themselves to be extraverts decided to prove it by writing longer messages. Amounttaken and education also had weaker, but still generally positive, effects on message length for obvious reasons. Lastly, we look at data† with respect to messages involving the words “sorry” and “fair”; we find that those who say sorry have decidedly lower WTPs, higher take rates, similar levels of happiness, and longer messages than those who

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!


Reported in Appendix as data is clunky since we could not find a neat way to output it.

do not. Those who discuss fairness have much lower WTPs, lower take rates, lower happiness, and longer messages, lending support to our fourth hypothesis.‡

5. Discussion
From this data, we have come to the following conclusions. First, we confirm the findings in the Grosskopf paper for takers instead of receivers. Therefore, evidence now exists supporting the notion that people on both sides of the power-to-take game are willing to pay to send messages, and accordingly, these messages inherently hold monetary value. The fact that our data, which was gathered on MTurk, a community of people looking to make money, shows this trend suggests that the magnitude of this trend in real life may be even larger since people are not always interested in profit maximization when dealing with people that they know personally, as opposed to people many miles away. Second, we demonstrate that profit maximization is not the only factor people consider when making decisions similar to those we pose in our game; this is a relatively novel accomplishment as little on this topic exists in the realm of power-to-take games. Punishment aversion seems a more likely mechanism across all trials since takers take and pay in ways that lessen the retribution they might expect from receivers. However, guilt likely plays a role as well since WTP increased in trial 3 compared to trial 2, demonstrating that, once informed, people are willing to pay more to remedy their selfish behavior. Backwards !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!


Note that our fifth hypothesis is still unconfirmed. These data seem to support it, but further analysis is required to make conclusions about it.

induction factors in as well; trial 1 participants displayed a smaller change in happiness than those in other trials likely because they knew beforehand that they would have to explain themselves later; this explanation could not be avoided, and thus they accepted it as an event that must occur whether they want it to or not. On the other hand, participants in trials 2 and 3 had the option of paying to avoid explaining themselves. This caused a smaller change in happiness, large increases in neuroticism levels, and large decreases in extraversion levels. Third, people with different personalities behave differently and, specifically, send messages differently. Extraverts and highly open people send longer messages, and grateful receivers send longer messages as well. Takers who learn that they have to explain themselves send longer messages than those who knew they would have to explain themselves ahead of time. Those who mention the words “sorry” and “fair” behave drastically differently in the game from those who do not; perhaps they feel that key words carry more value than more conservative players and may therefore be more (or less) selfish. There is much more analysis to be done on messages of this sort, an endeavor we may undertake in a forthcoming paper or perhaps one that will be taken up by other researchers. On a social level, people do not behave as if in emotional hot and cold states. They process data very quickly and make assumptions about the future almost instantly; trial 2 participants did not differ that much from those in trial 3, especially given that trial 3 players were uninformed for the first half of the

experiment. Behavior does not change as much with respect to when information is learned but rather with respect to what sort of information is learned. Additionally, guilting people into thinking about their actions cannot enforce pro-social behavior; instead, threats that induce a punishment aversion mechanism may be more successful. It is not yet clear what this says for politics or keeping up with the Joneses on a larger scale, but our data suggests that is it fear of having to explain why something is not a certain way that prompts us to preemptively change our behavior to avoid these sort of explanations. Looking forwards, we plan to re-run our experiment under slightly different parameters to collect more compelling data. We will conduct a deeper analysis of messages sent, and questions will be posed in a slightly different way to MTurk participants, who will be selected more carefully. Ideally, an expansion on this study would investigate the changes between messages sent at different times (i.e., Does your story today differ from your story tomorrow?) to see if people are time-inconsistent when it comes to explaining their actions. It is entirely possible that explanations change on a sort of sliding, prospect-theoryinduced scale that produces a different, but always fitting, explanation given the circumstances of the situation (i.e., Will you tell your spouse and your parents the same story today? How about tomorrow? Will the stories converge from two different starting points to a singular, “end state” story that is remembered and retold in the future as “the” single story of past events?) There is much work to be done, but for now, the question of “Paying, or Paying for It?” seems to be best answered by “Paying!”

6. Conclusion
In opposition to the standard belief that communication mainly has an indirect cost and benefit to interactions, our experiment focuses on the direct cost and benefit of communication itself. We find that the taker’s willingness to pay to avoid an explanation in a power-to-take game stays similar regardless of whether or not they are told that they can pay to avoid an explanation prior to the experiment. We find that apologetic takers had a much lower willingness to pay and sent longer messages, implying that both greed and guilt are important mechanisms to consider. Finally, experiment subjects who had messages regarding fairness and equality were more willing to explain their actions and, unsurprisingly, had a lower take rate. These results suggest that the act of explaining oneself, even to a complete stranger, has real costs. We care about what others think of us, so guilt and punishment aversion factor into our decision-making processes. However, because guilt is not particularly effective at preventing people from being selfish, punishment mechanisms are a more effective method of encouraging pro-social behavior.

References
1. Bosman, Ronald, and Frans van Winden. 2002. Emotional Hazard in a Power-To-Take Experiment. Economic Journal, 112 (476): 147-169. 2. Grosskopf, Brit and Kristian López-Vargas. Forthcoming. The Demand for Expressing (Emotions).

Appendix
“Sorry” data: Table 7. WTP by Sorry and Take Rate by Sorry

Table 8. Happiness by Sorry and Message Length by Sorry

“Fair” data: Table 9. WTP by Fair and Take Rate by Fair

Table 10. Happiness by Fair and Message Length by Fair.

Surveys can be found at the links below. Note that you have to fill in answers for all questions to progress through the different pages of the survey. Responses do not matter; feel free to fill them in randomly: Trial 1: http://goo.gl/NhUKNB Trial 2: http://goo.gl/lqsxZX Trial 3: http://goo.gl/RIeux6 Trial 4: http://goo.gl/i51Ktp§ The page displayed on MTurk can be found here: http://goo.gl/btBk5z More information is available upon request. Please contact the authors. !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
§

This survey was given to the $0.05 sub-group of the receivers. All other Trial 4 surveys were exactly the same except for the “$0.05 (or another amount) has been taken from you” line.

Sponsor Documents

Or use your account on DocShare.tips

Hide

Forgot your password?

Or register your new account on DocShare.tips

Hide

Lost your password? Please enter your email address. You will receive a link to create a new password.

Back to log-in

Close