Attribution bias: Evidence from a lab experiment

How to cite

sharma, karmini; liu, lily; fenske, james. 2024. "Attribution bias: Evidence from a lab experiment." AEA RCT Registry. AEARCTR-0014820. https://doi.org/10.1257/rct.14820-1.0

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

karmini sharma* , lily liu , james fenske

* Primary investigator

Registry ID
AEARCTR-0014820
DOI
https://doi.org/10.1257/rct.14820-1.0
First registered
Last updated
Status
In development
Trial dates
2024-11-14 – 2025-04-30
Countries
Private
Registry record
http://www.socialscienceregistry.org/trials/14820

Abstract

In many settings, economic outcomes depend on the competence and effort of the
agents involved, and also on luck. When principals assess agents’ performance they can
suffer from attribution bias by gender: male agents may be assessed more favorably than
female agents because males will be rewarded for good luck, while women are punished for
bad luck. We conduct a laboratory experiment to test whether principals judge agents’
outcomes differently by gender. Agents perform tasks for the principals and the realized
outcomes depend on both the agents’ performance and luck. Principals then assess agents’
performance and decide what to pay the agents. We aim to vary the type of tasks performed and measure whether principals show attribution bias. We aim to extend this to study whether attribution bias exists against other minorities.

Intervention

In each round the principal gets the information about demographics of the agent (hypothetical name, age group and country of residence) and then asked to judge the performance of the agent. Intervention, thus, is the demographic characteristic of the agent that the principal is paired with and whether the principal is shown the information about the agent or not.

Experimental design

We recruit principals and agents for the lab experiment. Agents perform either a male stereotypical or female stereotypical task in two separate sessions.
Principals are then randomly matched with agents (i.e., their performance) and are told the domain of the task performed by the agent . They are also given common demographic information (age and country of residence ) in all the sessions, but those assigned to control arms are not told the gender congruent names but instead about the team that the agent belongs to (team 1 and team 2).

Design details

Randomization method
Randomization done in office by a computer program written in python.
Randomization unit
This is a 2 (male or female stereotypical task)X3(male agent/female agent/ no information). Participants who are recruited as principals are randomized to see the performance of agents in different rounds where the demographics of agents vary within the participant. Hence, the unit of randomization is a principal-round.
Clusters
NA
Observations
Up to 2000 principals and up to 400 agents
Arms
Principals are equally divided between male stereotypical, female stereotypical, and control experiments for male and female stereotypical. (hence 500 principals in each treatment arm). Agents are equally divided into male and female stereotypical tasks (200 in each task).
Primary outcomes
Prior beliefs about agents, posterior beliefs about agents, and wage payment made to the agents (hiring decision).

Cited by

No citation data available for this registration yet.

Keywords