Turkers of the World Unite: Multilevel In-Group Bias Amongst Crowdworkers on Amazon Mechanical Turk
35 Pages Posted: 7 Mar 2018
Date Written: February 12, 2018
Crowdsourcing has become an indispensable tool for researchers conducting behavioral experiments. Often, the “crowd” is thought of as a black box for gathering data from an idealized “psychology simulator.” Yet crowdworkers are people, and their psychology exhibits the behavioral peculiarities that characterize humankind. Here we focus on the possibility that merely being a crowdworker is sufficient to induce a social identity that will then drive in-group favoritism. Any such bias would limit the appropriateness of assuming that crowdworkers treat one another as strangers and would suggest that effect-size estimates for effects that can be amplified by in-group bias are upwardly biased. To address this possibility we investigate the existence and relative strength of in-group bias on Amazon’s Mechanical Turk (MTurk) by measuring how MTurk crowdworkers (N=2337, pre-registered) share a real monetary endowment in a Dictator Game with a counterpart who is also an MTurk worker, a worker from another crowdworking platform, or a randomly selected stranger. Results indicate preferential in-group treatment for MTurk workers in particular (15% advantage as compared to workers from another platform) and for crowdworkers in general (35% advantage as compared to a stranger). These results suggest that cooperation levels from typical one-shot anonymous economic games run on MTurk might not be a good proxy for truly anonymous interactions, and that results from MTurk studies on cooperation, generosity, and ingroup bias may generalize most readily only to contexts where interaction and cooperation take place within the in-group context.
Keywords: In-Group, Dictator Game, Economic Games, Amazon Mechanical Turk, Crowdworkers, Behavioral Economics
Suggested Citation: Suggested Citation