The meta study covers 90 experimental studies containing 429 treatment conditions. We include all studies using the setup introduced by Fischbacher and Föllmi-Heusi (2013), i.e., in which subjects conduct a random draw and then report their outcome of the draw. We require that the true draw is unknown to the experimenter but that the experimenter knows the distribution of the random draw. We also include studies in which subjects report whether their prediction of a random draw was correct (as in Jiang 2013).
The payoff from reporting has to be independent of the actions of other subjects, but the reporting action can have an effect on other subjects. The expected payoff level must not be constant, e.g., no hypothetical studies, and subjects are not allowed to self-select into the reporting experiment after learning about the rules of the experiment. We only consider symmetric, single-peaked distributions (including uniform) but allow asymmetric two-state distributions. For more details on the selection process, see Appendix A of the research paper.
We searched in different ways for studies to include in the meta study, using Google Scholar for direct search of all keywords used in the early papers in the literature and to trace who cited those early papers, New Economic Papers (NEP) alerts and emails to professional email lists. We include all studies in which subjects conduct a random draw, then report their outcome of the draw (their “state”), and are paid according to their report. This excludes sender-receiver games as studied in Gneezy (2005) and the many subsequent papers which use this paradigm or promise games as in Charness and Dufwenberg (2006).
We require that the true state is unknown to the experimenter but that the experimenter knows the distribution of the random draw. The first requirement excludes studies in which the experimenter assigns the state to the subjects (e.g., Gibson et al. 2013) or learns the state (e.g., Gneezy et al. 2013). The second requirement excludes the many papers which use the matrix task introduced by Mazar et al. (2008) and comparable real-effort reporting tasks, e.g., Ruedy and Schweitzer (2010). We do include studies in which subjects report whether their prediction of a random draw was correct or not (as in Jiang 2013).
Moreover, we require that the payoff from reporting is independent of the actions of other subjects. This excludes games like Conrads et al. (2013) or d’Adda et al. (2014). We do allow that reporting has an effect on other subjects. We need to know the expected payoff level, i.e., the nominal reward and the likelihood that a subject actually receives this nominal reward. If the payoff is non-monetary, we translate the payoff as accurately as possible into a monetary equivalent. We further require that the expected payoff level is not constant, in particular not always zero, i.e., making different reports has to lead to different consequences.
We exclude studies in which subjects could self-select into the reporting experiment after learning about the rules of the experiment. This excludes the earliest examples of this class of experiments, Batson et al. (1997) and Batson et al. (1999). Finally, we exclude random draws with non-symmetric distributions, except if the draw has only two potential states. We exclude such distributions since the average report for asymmetric distributions with many states is difficult to compare to the average report of symmetric distributions. This only excludes Cojoc and Stoian (2014), a treatment of Gneezy et al. (2016) and two of the treatments reported in our research paper.
We contacted the authors of the identified papers and thus obtained the raw data of 55 studies. For the remaining studies, we extract the data from graphs and tables shown in the papers. This process does not allow to recover additional covariates for individual subjects, like age or gender, and we cannot trace repeated decisions by the same subject. However, for most of our analyses, we can reconstruct the relevant raw data entirely in this way. The resulting data set thus contains data for each individual subject.
Overall, we collect data on 270616 decisions by 44390 subjects. Experiments were run in 47 countries which cover 69 percent of world population and 82 percent of world GDP. A good half of the overall sample are students, the rest consists of representative samples or specific non-student samples like children, bankers or nuns.
Having access to the (potentially reconstructed) raw data is a major advantage over more standard meta studies. We can treat each subject as an independent observation, clustering over repeated decisions and analyzing the effect of individual-specific co-variates. We can separately use within-treatment variation (by controlling for treatment fixed effects), within-study variation (by controlling for study fixed effects) and across-study variation for identification. Most importantly, we can conduct analyses that the original authors did not conduct.