“Last-place aversion”: Evidence and redistributive implications
Kuziemko, I. / Buell, R.W. / Reich, T. / Norton, M.I. (2014)
Quarterly Journal of Economics, 129(1), 105-149
Adam Altmejd, Anna Dreber and Magnus Johannesson
Kuziemko et al. find that subjects who are randomly placed in the second-to-last place in terms of endowments are the least likely to give money to the person one rank below them. The paper includes two experiments where other things are also tested. Experiment 2 was chosen as it was the last experiment. In Experiment 2, there are two group sizes: 6 person groups and 8-person groups, out of which we randomly chose to study the 6-person groups.
The original p-value is reported as p<0.10; the exact p-value=0.070 based on a z-test of the marginal effect from a probit regression (probit regressions with round fixed effects and clustering on individuals and a dummy for subjects ranked 1st and last (as they don’t have parallel choice sets); the coefficient of “Second from last” in regression (1) of Table II, p. 129): “Column (1) shows that the second-to-last-place player is significantly less likely to give to the lower-ranked player relative to other players,”
The original sample size is 42 participants. To achieve 90% power the required sample size is 134 participants.
The sample for replication consists of 138 students (in order to assign 6 person groups) from Boston-area colleges and universities. Apart from having participated in the original experiment, there are no exclusion criteria.
We use the material of the original experiment (programmed in PHP with a MySQL database) provided by the authors, along with instructions available at the journal’s webpage.
We follow the procedure of the original article, with only slight but unavoidable deviations as outlined below. The following summary of the experimental procedure is therefore based on the section “IV.A. Experimental Design” (pp. 124–125) in the original study.
As participants enter the lab, they are seated sequentially in different groups of 6, to minimize the likelihood that people who know one another (and thus enter the lab together) are assigned to the same group. Each group will engage in the same game. Participants are seated in separate carrels surrounded by blinders. As all participants are seated, an experiment supervisor reads the instructions and subjects start by playing a test round before the actual experiment begins.
As a round begins the computer randomly ranks the 6 players in each group and allocates \$1, \$2, …, \$6 to them. Subsequently, players with rank 2-5 (i.e. that are allocated \$5-\$2) choose between giving the player directly above or directly below their rank, 2 dollars. The first player (with \$6) makes a choice between the second (\$5) and third (\$4) player, and the last player considers players in fourth and fifth place. All participants are clearly instructed that the additional \$2 that they are allocating come from a separate account. When the round is finished the computer randomly selects one participant. Two dollars are added to the allocated amount of the player chosen by that participant. The final allocation is not shown to participants, and neither is the decision of other players.
The game is repeated for 8 rounds. After all rounds are played, the computer randomly selects one round, and the final allocation from that round is paid out to all subjects. Subjects are further paid a show-up fee of \$15. Payments are distributed while the players are still at their seats actively engaged in answering a questionnaire, and without other players being able to see the amounts handed out. Show-up fees are paid to subjects when they exit the lab.
As in the original article, a probit regression is run on the data, with round and game fixed effects and with separate dummy variables for first- and last-place players. Standard errors are clustered by individual. The dependent variable is 1 if the participant has chosen to give money to a player with lower rank and 0 otherwise.
Differences from Original Study
The replication procedure is identical to that of the original study, with some unavoidable deviations. This replication will be performed at Harvard University in Cambridge MA, USA, in 2015, on students at Boston-area colleges and universities. The original data was gathered at Harvard University in Cambridge MA, USA, in September 2010, on subjects from the subject pool of the Harvard Business School Computer Lab for Experimental Research (CLER). The experiment will be in English as in the original study.
The paper contains two experiments and other treatments: for the replication the focus is only on Experiment 2 and 6 person groups. Experiment 2 is included as it is the last experiment in the paper; we randomly determined whether to replicate the 6 person or the 8 person redistribution game. In the original study, subjects stayed in the lab for approximately one hour, partaking in several other experiments after the main study. We will only conduct the main study.