The Choices of Others: An Experiment on Social Search

The experiment aims to explore the impact of social learning on the information acquisition process and on the choices made by individuals who must choose from a set of alternative options with unknown values. In a sequential search framework, in which first and second movers face the same decision environment, we exogenously manipulate the availability of social information and the presence of reputational mechanisms, allowing second movers to identify which reliable first movers to follow.

2023-12-22

2026-02-28

Individual search behavior and frequency of optimal search choices, as defined by the theoretical predictions derived from the model

In a between-subjects design, we exogenously vary: i) whether second movers can observe the choices made by one of the first movers who faced the same decision environment, and ii) the presence of reputational mechanisms allowing second movers to identify which reliable first movers to follow.
More precisely, conditional on having access to some social information, we manipulate - in a 2x2 factorial design - whether second movers can choose which first mover to observe from and the availability of reputational information on first movers. Our Benchmark treatment with social information is characterized by exogenous matching and the absence of reputational information on matched first movers.
We include an additional treatment condition: the Benchmark-ROBOT condition. This condition is equivalent to our Benchmark (exogenous matching and no reputational information on first movers), with the only difference that matched first movers are not human peers but computerized agents. Computerized agents are programmed to always behave optimally, given their search cost and the quality of the first action sampled, as defined by our theoretical model.
Overall, the experiment counts 6 treatment conditions:
1) No social information
2)-5) Social information with human first movers
6) Social information with computerized first movers
Participants are evenly allocated across treatment conditions.

Not available

Experimental sessions' randomization into treatments is done in office by a computer.
Upon registration, participants are assigned to one of the experimental sessions: subjects enrolled in the first (scheduled) experimental session are assigned to the treatment in which participants make choices in isolation, in the absence of social information; subjects enrolled in any of the following experimental sessions are randomly assigned to one of the four main social information treatment conditions, or to the additional one with computerized first movers.

Randomization into treatments happens at the experimental session level.
Within each session (treatment), participants are further randomly assigned to a matching group: all participants belonging to the same matching group, irrespective of the treatment condition, are exposed to the same set of alternative options, in the same order. Each matching group is composed of four participants: for each treatment condition, there are twelve matching groups, numbered I to XII, in total.

Yes

72 Clusters (12 matching-groups of 4 participants each per treatment condition)

288 Participants

48 Participants (12 Clusters) per Treatment arm

This sample size (12 clusters of 4 participants each per treatment arm, for a total of n=48 participants per treatment) would allow us to detect a MDE of 0.125 (1/8) standard deviations on second movers' overall propensity to imitate first movers, in a cross-sectional analysis. This corresponds to an economically meaningful variation, equivalent to roughly less than 3 percentage points.