Breaking Down Information and Inequality: Evidence from a Field Experiment in the Job Market

How to cite

Ierokomos, Surya; Choi, Jungho; Sterling, Adina. 2022. "Breaking Down Information and Inequality: Evidence from a Field Experiment in the Job Market." AEA RCT Registry. AEARCTR-0008779. https://doi.org/10.1257/rct.8779-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Surya Ierokomos* , Jungho Choi , Adina Sterling

* Primary investigator

Registry ID
AEARCTR-0008779
DOI
https://doi.org/10.1257/rct.8779-1.0
First registered
Last updated
Status
In development
Trial dates
2022-03-30 – 2022-06-30
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/8779

Abstract

Online review platforms, such as Glassdoor or Indeed, generally display the unweighted average of all existing reviews. However, a singular, unweighted average does not effectively capture heterogeneity in reviewer evaluations. For example, an unweighted average will typically be closer to the majority opinion due to its numeric properties. This can potentially be problematic in many labor market settings, where women systematically face different experiences than men, and are numerically the minority. In this research, we study how this unweighted aggregation affects different genders’ behaviors and labor market outcomes. We do so by designing an experiment on a job review platform geared toward high paying, high growth technology careers. We are able to observe the gender of reviewers in our setting, and calculate separate company ratings for each gender. We compare job search behavior, as well as subsequent reviewing behavior of users who were randomly exposed to the unweighted, aggregate ratings to users who were randomly exposed to their own gender rating.

Intervention

We plan to disaggregate the ranking of employers by gender and to present either the aggregated or the disaggreated employer ranking to randomly selected groups of users.

Experimental design

Our experiment is restricted to mobile users of the application, and has two components. The first component will be sent to users who have already left their own company ratings, as well as self-identified their gender as a man or woman. We assign these users into one of three conditions: one aggreagated and two disaggregated ranking conditions. The second component will be sent to existing users of the application who have not left a review for their company. Users will be randomly assigned into conditions that are similar to those above. They can click on a randomly assigned push notification to be directed to either the aggregate rankings or the gender-based rankings to study how the value of information influences its participation in a subsequent survey.

Design details

Randomization method
Randomly assigned by a random number generator on a computer.
Randomization unit
We randomize individual users.
Clusters
We anticipate approximately 55000 users in the first component. We anticipate approximately 12000 users in the second.
Observations
We anticipate approximately 55000 users in the first component. We anticipate approximately 12000 users in the second.
Arms
For the first component: we will randomly split each gender into approximate thirds for each arm. This means that approximately 40000 men will be in each of control, treatment 1, and treatment 2. We anticipate 15000 women in each of control, treatment 1, and treatment 2. We will randomly split the second component sample into thirds for each of the three relevant treatments as described previously.
Primary outcomes
We want to measure click through rate on the intervention page, as well as propensity to leave a company review.

Cited by

No citation data available for this registration yet.

Keywords