«IZA DP No. 6388 PAPER Experimental Evidence of Self-Image Concerns as Motivation for Giving Mirco Tonin DISCUSSION Michael Vlassopoulos February 2012 ...»
“You have just made three decisions on how to allocate £10 and one of these has been randomly selected to be implemented. Before carrying out the payment associated with your choice you are given an opportunity to opt out. This will imply that the decision you made will not be implemented and you will instead receive £10 at the end of the session. If you decide not to opt out then you will receive the payment associated with the decision that was previously selected.” Note that subjects were not aware of this option when taking the three decisions on how to allocate £10. At the end, participants completed a short questionnaire while we arranged the payments. A session lasted approximately one hour.
2.2 Treatments In the three sequential decisions labelled as DA, DB, and DC, we asked participants to decide how to allocate £10 between themselves and a recipient in three diﬀerent conditions in which we either vary the recipient (experimenter, charity) or amount received by the recipient (ﬁxed, varying).
In particular, in one condition that we will label T1, the experimenters were the recipient and the amount received corresponded to the one passed by the participant. The other two conditions involved a charity that the participant could in each case choose from a list of ten.5 In the condition that we label T2, the amount that the charity would receive was ﬁxed at £10, regardless of the experimental subject’s choice,6 while in condition T3, the amount received by the charity was given by the subject’s allocation. Subjects underwent the three conditions in a randomized order, with 5 out of 6 unique orders implemented twice and one implemented three times. Consequently, in some sessions the ﬁrst decision faced by participants, decision A or DA, corresponded to T1, in others number the monitor drew.
In the questionnaire at the end of the experiment, subjects indicated that they understood well this random implementation. In particular, the question was “when I took my decisions I understood that only one would be implemented”, with possible answers ranging between 1 (Strongly Disagree) and 5 (Strongly Agree). The average response for the sample that is used in the paper is 4.7.
Among the ten charities, the most selected one is Cancer Research UK, followed by Doctors without Borders and National Society for the Prevention of Cruelty to Children.
In particular, in this treatment we informed subject that “the experimenters will pay your selected charity a top-up (the diﬀerence between £10 and what you choose to pass) so that in total the charity receives £10” and that “in total your selected charity will receive neither more nor less than £10”.
to T2, in others to T3. The same for the second decision, DB, and the third, DC. These decisions have been analyzed in a companion paper (Tonin and Vlassopoulos, 2011) aimed at distinguishing and quantifying the two types of intrinsic motivation for giving that have been underlined in the literature: pure altruism and warm glow (Andreoni, 1989, 1990). The focus of this paper is instead on the opt-out decision that takes place after the selection of which of the three decisions on how to allocate £10 to implement.
As mentioned earlier, a total of 251 subjects participated in the experiment. Of these, 13 participants acted as monitors, leaving 238 subjects who made decisions. To check for understanding of the instructions, we asked participants to respond to questions about hypothetical allocation decisions before making each of the three sequential decisions. 133 answered all questions testing understanding of the treatments correctly, while most mistakes occurred in T2. However, out of 98 subjects making a mistake in T2, 63 provided the correct answers regarding the amounts the charity and the subject receive, while making a mistake regarding the experimenter contribution to the charity. Considering that all these subjects understood the essential parts of T2, we conduct the analysis including them. At the end, we have a sample of 192 subjects (81% of the original sample) who answered correctly to questions regarding T1 and T3 and at least answered correctly the questions about the amounts received by the charity and the subject for T2. We have also conducted the analysis using the smaller sample of 133 subjects who answered all questions correctly and the whole sample of 238 participants. The results (available upon request) are very similar.
The average donation decision among the 192 participants in our sample was £1.77 for T1 (with 55% giving 0), £1.84 for T2 (with 57% giving 0) and £4.29 for T3 (with 19% giving 0).
Almost a quarter of our sample (46 out of 192) decided to opt out. Notice, however, that subjects for whom the selected decision implied a £10 payment to themselves, opting out or not has no implications whatsoever in terms of the payment they receive at the end of the experiment. For instance, if I decided to pass nothing to charities in T3 and that decision is selected for implementation, whether or not I opt out, I will still receive £10. Indeed, the decision to opt out from a donation is not very meaningful if no donation was made in the ﬁrst instance. For this reason, from now on we restrict attention to those for whom the decision to opt out has implications in terms of personal pay oﬀ, i.e. those who passed something in the decision selected for implementation. This leaves us with a sample of 109 participants;7 in this sample more than one third of the subjects (37 out of 109) decided to opt out.
If we look at the decision to opt out by treatment (see the left-hand side of Table 1) what we see is that people are more likely to opt out when T1 is implemented, namely, when the experimenters are the recipients. In this case, more than half of those who had decided to give something, when given the opportunity withdraw their donation and keep £10 instead. On the other hand, when a charity is involved, only around a quarter of those who had decided to give something opt out (21% for T2, 29% for T3). The diﬀerence between the two treatments involving charities (T2 and T3) is not statistically signiﬁcant, while the diﬀerences between the treatment involving the experimenters and the treatments involving charities are. The fact that subjects are more likely to opt-out when the recipient is the experimenters is probably related to the fact that the moral cost associated with opting out is lower in this case compared to the case when the recipient is a charity.
Allowing participants to opt out had important quantitative implications in our experiment. In particular, participants in the sample we use for the analysis would have donated a total of £511 (an average of £4.7 each) if the experiment had stopped after the three sequential decisions, while they actually donated a total of £350 (£3.2 each on average). Thus, giving the opportunity to opt out reduced donations by one third. Donations to experimenters are particularly aﬀected, dropping by half from £124 to £58, but also donations that participants made to charities dropped by a remarkable 25%, from £388 to £292.
Looking at the decision to opt out by position (see the right-hand side of Table 1), we do not ﬁnd a particular trend, with 34% of subjects opting out when their ﬁrst decision is implemented, while the ﬁgures for the second and third decision are 40% and 28% respectively, with these diﬀerences not being statistically signiﬁcant.8 As the left hand side of Table 2 shows, the “stakes” at hand were not diﬀerent between those opting out and whose standing behind their initial decision. For instance, when T3 was the decision implemented, those opting out had given on average £4.86, while those not opting out had given on average £5.28. Both t-tests and tests on the equality of distributions, such as, the WilcoxonMann-Whitney test and the Kolmogorov-Smirnov test, fail to reject the null that mean giving, or the distribution of giving, was the same between the two groups across the diﬀerent treatments.
One could have expected that those that opted out did so because they had given much more in the Thus, 43% of our original sample gives nothing in the decision that is selected for implementation.
We cannot reject the null that the implemented treatment and the implemented position within the sequence are independent (p-values for Pearson chi2 test=0.979, for likelihood-ratio chi2 test=0.978, for Fisher’s exact test=0.977).
The same is true if we only consider implemented treatment and implemented position for those for whom the decision to opt out has implications in terms of personal pay oﬀ (p-values for Pearson chi2 test=0.748, for likelihood-ratio chi2 test=0.746, for Fisher’s exact test=0.764). Thus, randomization was successful and it makes sense to look at the decision to opt out by treatment and by position separately.
decision selected for implementation and so had more to gain from the decision to opt out of their previous allocation decision and receive £10 instead. On the other hand, the fact that somebody has given a lot means that he or she cared about donations, thus, making opting out less likely. As it turned out, these two contrasting forces cancel out and those opting out are as generous as those not opting out, when generosity is measured by their allocation decisions before opting out. The right-hand side of Table 2, reports giving in all three treatments separately for those that opted out and those that did not. From this table it is evident, for instance, that regardless of the decision that was randomly selected for implementation, those opting out gave on average £5.04 to a charity in T3, while those not opting out gave £5.47, and, once again, we fail to reject the null that mean giving, or the distribution of giving, was the same between the two groups.
These results are conﬁrmed by a regression analysis (see Table 3), showing that the probability of opting out is unrelated to the position in which the implemented decision is taken, while opting out is more likely if the implemented decision concerns the donation to the experimenters (T1) instead of donation to a charity (T2 and T3). This is the case even when we control for the amount that the subject originally donated in the implemented decision, a variable that, consistently with the left-hand side of Table 2, is unrelated to the probability of opting out.
As mentioned earlier, we invited an equal number of males and females in each session, with the purpose of testing for possible gender eﬀects. Out of the 109 participants for whom the decision to opt out has implications in terms of personal pay oﬀ, 59 are males and 50 are females. Among the 37 who decided to opt out, 18 are males (31% of all males) and 19 are females (38% of all females).
We fail to reject the null hypothesis that the decision of opting out and gender are unrelated.9
Why is it the case that a sizeable share of subjects in our experiment decided to give something either to a charity or to the experimenters, but then withdrew their donation and kept everything for themselves? One possible explanation, given that in our design decisions are taken sequentially, would be that participants acquire additional information as the experiment goes on and this induces them to reconsider their choice at the opt out stage. For instance, subjects may give generously to the experimenters when this is the ﬁrst decision they face. However, later on, once they see treatments involving charities, they may reconsider the “worthiness” of the experimenters and regret their initial donation. Considering that in our design there is random implementation of one of the three decisions, giving in any one treatment does not aﬀect in any way the material payoﬀs associated with the other treatments. Still, learning about the experiment may induce some P-values for Pearson chi2 test=0.410, for likelihood-ratio chi2 test=0.411, for Fisher’s exact test=0.425. Also, the coeﬃcient of a gender dummy added to the regressions in Table 3 is always insigniﬁcant (results not reported).
people to reconsider their initial decisions. If this were the case, then we would expect a strong trend in opting out, with a high incidence for decisions taken early on, and basically no incidence at all for the decision taken just before the opt out option was presented, as no additional information about the experiment could be acquired in this stage. As outlined above, no trend emerges from the data, with opting out having a strong incidence also for the third decision in the sequence. We can thus exclude the possibility that the acquisition of information about the experiment is what is driving our results.
Our claim is that this pattern of revision of the decision to give in the ﬁrst three rounds of the experiment suggests that giving in the ﬁrst instance was not motivated exclusively by a desire to improve the payoﬀ of the recipient, but was also driven by the decision-maker’s desire to selfsignal her altruistic inclination. The reversal being attributed to self-signaling is consistent with the theoretical framework of Bodner and Prelec (2003) who develop a model of choice in which the decision maker has a utility function with two components: outcome value, which is the beneﬁt derived from helping the charitable cause per se and diagnostic value, which is the value derived from becoming informed about one’s level of altruism from the action taken.10 In this framework, Bodner and Prelec suggest that “A self-signaling person will be more likely to reveal discrepancies between resolutions and actions, when resolutions pertain to actions that are contingent or delayed.
Thus she might honestly commit to do some worthy action if the circumstances requiring the action were remote (temporally or probabilistically), but would in fact regret the commitment if those circumstances were obtained.” (Bodner and Prelec, 2003, page 107). In our experiment, the payoﬀ consequence of the three sharing decisions that subjects make is both uncertain (each will be implemented with equal probability of one-third) and will be revealed with delay at the end of the experiment. If this aspect leads subjects to discount the utility they obtain from payoﬀs when they make the sharing decisions and not when they decide to exit, while the self-signaling component of utility remains the same, then this would explain the reversal of choices we observe.