2.3.1 Design and Procedures
The first experiment consisted of two sequences. In each sequence, participants played ten consecutive periods of a standard public goods game. At the beginning of the first sequence, we assigned participants randomly into groups of four. The group composition remained the same during the entire experiment. Each group member received an endowment of 20 tokens at the beginning of every period. The marginal per capita return (a) was 0.4. A period consisted of the individual contribution decision followed by the feedback about the
individual earnings and the contributions of the other group members.5
We based our salience manipulation on results from research in cognitive psychology. The literature shows that attention is often biased towards what is currently in mind (for a review see Olivers, Peters, Houtkamp, and Roelfsema (2011)). Furthermore, according to that literature, recalling an item affects the perception about that item. For example, Dutch students reported using their bike more often after recalling instances of bicycle
use.6 In a similar vein, Janiszewski, Kuo, and Tavassoli (2013) show that paying repeated
attention to a product increased the probability of choosing this product in a subsequent choice. Hence, to manipulate the salience of the different contribution levels within a group, we introduced a question at the end of every period. The question implied to recall one of the previously displayed contributions of another group member. The goal was to draw the attention of the participant on a specific contribution level. This central feature of our experimental design is worth emphasizing. Participants memorize a contribution level such that it occupies parts of their working memory and shifts their attention to that level. We provided participants with two stimuli, (i) the minimum and (ii) the maximum stimulus. Under the minimum stimulus, as the name indicates, participants were asked to recall the lowest contribution in their group. On the contrary, under the maximum stimulus, participants had to remember and reproduce the highest contribution of the other group
5
Participants received the information about the exact contributions of the other group members; they did not get the mean contribution of their group. However, it was easy to calculate the mean from the available information. The screenshots of this experiment are in appendix 2.C.
6According to the research, the effect is only present when the recalling task is easy, i.e., when the
members. Subjects earned a point if they recalled the respective contribution correctly.7 Hence, in both treatments, we did not manipulate the information structure but only the salience of a certain piece of information.
Within a group, all participants received the same stimulus. We assigned the groups randomly to one of the two stimuli at the beginning of the experiment, and they stayed within the assigned stimulus for the first sequence. In the second sequence, all groups switched stimulus. Groups who started under the minimum stimulus switched to the maximum stimulus in the second phase and vice versa. In the instructions, we explained that participants will need to memorize a specific information and that they will receive more information about this task directly on the screen once the experiment started. A translated version of the instructions is in appendix 2.C.
2.3.2 Sample Size Calculation
To know how many independent observations we need to draw statistically meaningful conclusions, we calculate the sample size using the power analysis by Cohen (1988). We want to compare the means under the two stimuli, hence, to calculate the sample size, we need an estimate of the effect size, the standard deviation and to specify the significance level and power. Since our model does not give precise predictions about the size of the treatment effect, we rely on values reported in the related literature. The three papers which are most closely related to our design (Hoffmann et al., 2013; Engel et al., 2014; Samek & Sheremeta, 2014) report treatment effects in the range of 25 and 30 percentage points. Only Engel et al. (2014) report the standard deviation. For the significance level and power, we use the common levels of 0.05 and 0.8 respectively. Based on a treatment effect of 5.3 token and a standard deviation of 5 tokens, the formula indicates that our sample size should be 28 (14 in each treatment cell) to be able to compare the mean under the maximum stimulus to the mean under the minimum stimulus.
7We incentivized this question to make sure that participants had an incentive to answer the question
2.3.3 Results
We conducted the experiment at the LABEX of the University of Lausanne in Switzer- land. 108 individuals participated, and we used the software zTree to run the experiments (Fischbacher, 2007). For the analysis, the unit of independent observation is the group. Hence, in this experiment, we observe the behavior of 27 independent groups.
In Figure 2.2 we display the behavior of the participants in the first sequence graph- ically. The thin lines depict the average contributions, which participants had to recall under the two stimuli. We observe that the mean with the maximum stimulus is 12.76 (SD: 6.84) compared to an average of 4.75 (SD: 6.55) with the minimum stimulus. This difference is statistically significant (Wilcoxon rank-sum test, two-sided, N =27, p <0.01) and a prerequisite for our treatment manipulation to work. Because if participants had to recall numbers which are not substantially different, there would be no reason to expect a difference between the minimum and the maximum stimuli.
To our surprise, the two stimuli do not translate at all into significant differences in the contribution behavior. Participants under both stimuli contribute on average similarly (see the thick lines in Figure 2.2). The mean contribution with the maximum stimulus is 8.76 tokens (SD: 7.37) compared to an average of 8.41 tokens (SD: 7.76) with the minimum
stimulus (Wilcoxon rank-sum test, two-sided, N =27, p=0.662).8 Moreover, the variance
within the groups over time remains rather constant and is statistically not different be-
tween the two stimuli (Wilcoxon rank-sum test, two-sided, N =27, p=0.771).
8We elaborated additionally if participants who behave as conditional cooperators behave differently.
Conditional cooperators are participants which condition their contribution on the contribution of the other group members. To classify them, participants had to fill out a contribution table at the very beginning of the experiment. In this table, they had to indicate how much they contribute given the average contribution of the other group members. We followed the procedures of Fischbacher et al. (2001), which allow classifying participants into freeriders, conditional cooperators, hump shaped and others. The results are the same if we restrict the analysis at the subgroup of conditional cooperators. Furthermore, the analysis of the second sequence (of all participants) reveals very similar results as in the first sequence. The average contribution, which subjects recalled under the maximum stimulus is 8.71 (SD: 7.86) compared to 4.52 (SD: 6.27) under the minimum stimulus. The difference is not as pronounced as in the first sequence. Moreover, it is only weakly statistically different (Wilcoxon rank-sum test, two-sided, N=27, p=0.093). Again, we do not observe differences in the subsequent contribution behavior (see Figure B1 in the appendix). The average contribution with the minimum stimulus is 6.93 (SD: 7.30), which is even higher than the average of 5.80 (SD: 7.58) with the maximum stimulus. The difference is statistically not significant (Wilcoxon rank-sum test, two-sided, N =27, p=0.599).
0 5 10 15 20 1 2 3 4 5 6 7 8 9 10
Behavior in the Recall Experiment
Max Stimulus Min Stimulus Contribution Max Contribution Min Contribution Period
Figure 2.2: Behavior in the Recall Experiment, first sequence.
One possibility for not finding differences could be that participants did not recall the correct contribution and therefore had different numbers on top of their head. However,
in 93% of the cases, participants recalled the contribution correctly.9 Hence, it is fair to
assume that people focused on the correct number.
We further elaborate whether the stimuli might affect groups differently depending on the homogeneity of the contributions within the group. It could be that if the contributions in a group are very similar the stimulus leads to a different effect compared to the situation where the behavior within a group is very different. However, this seems not to be the case. We do not find a difference between the stimuli in neither homogeneous nor heterogeneous groups.
So far we discussed the between-subject effects. It might be that, even though we do not observe differences between subjects, subjects themselves react to the stimuli. Our design allows to evaluate whether the behavior of the subjects is different between the first and the second sequence. Based on a sign-rank test we do not find a significant within-subject difference (Signed-rank test, N =16, p=.918).
Thus, the results of this experiment lead us to conclude that either the contributions of subjects in the public goods game are more stable than we thought or that our salience stimuli were too weak. We suspected the second option and decided to manipulate salience more radically in a second experiment.
950 % of the participants answered the recall question correctly in all ten periods. 33% replied to the