Inaccurate Beliefs & Bias-Motivated Updating
How to cite
Rackstraw, Emma. 2022. "Inaccurate Beliefs & Bias-Motivated Updating." AEA RCT Registry. AEARCTR-0008849. https://doi.org/10.1257/rct.8849-1.0
Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.
* Primary investigator
Abstract
An important question in understanding the dynamics of discriminatory behavior is why statistical discrimination based on inaccurate beliefs occurs and -- perhaps more importantly -- why it persists when correcting information is available. For example, Americans hold inaccurate beliefs about household income by race, even though accurate information is readily available through the Census Bureau. Aigner & Cain (1977) assume such ``mistaken behavior" will not persist in competitive markets. However, if inaccurate priors about group productivity are driven by bias rather than a lack of information, they may be resistant to updating in the face of accurate information that contradicts a stereotype about that group. A recent hiring experiment showed that not all participants fully updated their inaccurate beliefs about the productivity of stigmatized groups after receiving correcting information, leading to persistent discrimination. This proposed experiment investigates the source of this non-Bayesian updating.
Intervention
In Part I of the experiment, I gathered information about different groups’ productivity in a stereotyped domain task: a math test.
In Part II of the experiment, I have a different set of participants – heretofore referred to as “employers” – engage in a task where they decide how much they’d be willing to pay the Part I workers for their work.
Part II employers see 8 anonymized "resumes” containing information about a worker's group identities and they are be asked to make an incentive-compatible wage offer using a Becker-Degroot-Marschak mechanism for payment. Group identities include education, age, beverage preferences, and either gender or race. The “control” arm of the experiment will end here.
The main treatment arm of the experiment (the “motivated reasoning” arm) will then induce random variation in the direction of signals (i.e. whether a participant receives a high or low signal of each group’s productivity) by providing the average score for each group among a small subsample of Part I workers. Then I measure wage offers to 8 additional workers in a second hiring task. Comparing the updating process of employers who receive randomly positive versus randomly negative signals about stigmatized group productivity will isolate any asymmetric (bias-motivated) updating.
In addition, in a smaller treatment arm of the experiment (the “accurate information” arm), employers will receive an information intervention with the actual group-level productivity averages among all Part I workers. This will test whether (and for whom) motivated updating disappears when there is no uncertainty about average group productivity.
Finally, all Part II employers will complete a relevant Implicit Association Test to measure implicit biases, attitudinal questions to measure explicit biases, and a demographic survey. This will allow me to measure to what extent any asymmetric updating correlates with measured explicit and implicit biases. The location of the IAT within the survey is randomly assigned to test whether the intervention affects IAT scores.
Experimental design
Pre-intervention, individuals will make wage offers to individuals based on "resumes" that include group identity information. I will also elicit beliefs. In the treatment group, I will then provide either a noisy signal (with randomly generated noise) of group productivity or an accurate signal of group productivity. I will then measure how wage offers and beliefs change post-information.
Design details
Cited by
No citation data available for this registration yet.