Intervention(s)
We conduct a laboratory experiment at the Renmin University of China in Beijing, China, with 256 subjects recruited through the bulletin board system and posters. The majority of the subjects are students from Renmin University and nearby universities.
The experiment consisted of 18 sessions, which were all computerized using the z-Tree software package (Fischbacher, 2007). Both the instructions (see Appendix A) and the game information shown on the computer screen were in Chinese. In each session, we randomly assigned subjects to groups of four; our sample consisted of 64 groups in total.
The experiment included two stages: The first stage comprised twelve periods, whereas the second one comprised eight periods. Group members did not change within each stage but subjects were randomly reshuffled into groups of four after the first stage; this rule was made to be common information. The subjects were not told the exact number of periods in each stage. Instead, the subjects were told that the experiment would last from 30 minutes to one hour, including the time for sign-up, reading of instructions, taking a quiz designed to ensure that subjects understood the experimental rule, and final payment. We note two features of this design. First, we wanted to reduce the possibility of backward induction. Second, our study design approximated features of real-world situations. In many real-world cases, people do not know the exact number of coordination opportunities.
At the beginning of each period, subjects knew the stake of the current period but not those of future periods. This condition replicates the circumstances of many real-world cases, in which people do not know what is at stake in future interactions.
In each period, we endowed each subject with 20 points and asked each to give a certain number of points to the common pool of his/her assigned group. The required number could vary across periods, and each subject could only choose either “to give the exact points required” (we use the natural term “give” rather than “contribute” in the instruction), which we refer to as stake, or “not to give” at all. If all members in a group contributed, then each member not only received the stake back, but also gained an extra return, which equaled the stake. If not all group members contributed, then each member finished the period with only his/her remaining points (i.e., the initial endowment in each period minus the contribution of the subject).
At the end of each period, each subject knew whether all four group members (including himself/herself) contributed the stake for that period but did not know the total number of group members who contributed (in case fewer than four members contributed). This design is consistent with the standard design in the literature of minimum-effort coordination games (e.g., Van Huyck et al., 1990), in which the only commonly available historical data to players is the minimum contribution of group members. , By adopting this design we can also increase the difficulty of coordination given other aspects of the experiment, and study whether gradualism can help overcome such a difficulty.
The final total payment to each player equaled the accumulated earnings over all periods plus a show-up fee. The exchange rate was 40 points per CNY 1. An average subject earned CNY 21–22 (around USD 3) including the show-up fee for the whole experiment, which covered ordinary meals for one to two days on campus. With regard to the purchasing power, this payment was comparable to those experiments conducted in other countries.
Our experiment comprised four treatments: (1) Big Bang, (2) Semi-Gradualism, (3) Gradualism, and (4) a variant of the Big Bang treatment, which we call the High Show-up Fee treatment. All groups in the three main treatments faced the same stake in the second half (Periods 7–12) of the first stage, but stake paths differed for each treatment in the first half (Periods 1–6). The first half of the first stage featured different stake paths for each treatment. The different stake paths might yield potential earning differences and could potentially lead to a wealth effect across treatments. To isolate the wealth effect on the contribution of participants from the effect of the three main treatments in the second half of the first stage, we introduced the High Show-up Fee treatment, which is a variant of the Big Bang treatment. We describe the High Show-up Fee treatment in detail later in this Section. We randomly assigned 12 subjects into the three main treatments for 8 of the 18 sessions. In the remaining 10 sessions, we randomly assigned 16 subjects into the four treatments (three main treatments and one supplementary High Show-up Fee treatment). In total, we had 18, 18, 18, and 10 groups in Big Bang, Semi-Gradualism, Gradualism, and High Show-up Fee treatments, respectively. Unreported randomization checks show that the randomization of treatment assignment worked well (available upon request).