Automating Automaticity: Online Lab Experiments

How to cite

Agan, Amanda; Ludwig, Jens; Mullainathan, Sendhil; Davenport, Diag. 2025. "Automating Automaticity: Online Lab Experiments." AEA RCT Registry. AEARCTR-0015526. https://doi.org/10.1257/rct.15526-1.2

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

Amanda Agan* , Jens Ludwig , Sendhil Mullainathan , Diag Davenport

* Primary investigator

Registry ID
AEARCTR-0015526
DOI
https://doi.org/10.1257/rct.15526-1.2
First registered
Last updated
Status
In development
Trial dates
2025-07-14 – 2025-12-31
Countries
Private
JEL codes
A16, D63, D83
Registry record
http://www.socialscienceregistry.org/trials/15526

Abstract

Consumer choices are increasingly mediated by algorithms, which use data on those past choices to infer consumer preferences and then curate future choice sets. Behavioral economics suggests one reason these algorithms may fail: choices can systematically deviate from preferences. For example, research shows that prejudice can arise not just from preferences and beliefs, but also from the context in which people choose. When people behave automatically, biases creep in; snap decisions are typically more prejudiced than slow, deliberate ones, and can lead to behaviors that users themselves do not consciously want or intend. As a result, algorithms trained on automatic behaviors can misunderstand the prejudice of users: the more automatic the behavior, the greater the error. This RCT is an online lab experiment where we test this idea.

Intervention

Hidden until after experiment takes place to preserve integrity of experiment.

Experimental design

Hidden until after experiment to preserve integrity

Design details

Randomization method
Randomization done by computer
Randomization unit
Rushed vs non-rushed randomized at the individual respondent level Race and gender perception of recommender is randomized at the movie level
Clusters
1600 respondents split between control (no rush) and rushed being 5 minutes. Each Choose 5 movies. 1600 respondents split between control (no rush) and rushed being 1 minute. Each Choose 5 movies
Observations
8000 observations (1600 x 5 movies clicked), or ~42,000 (if we consider all movies seen) in 5 minutes rushed (vs control) Same in 1 minute rushed vs control.
Arms
Equally split ~800 respondents in rushed, ~800 in non-rushed
Primary outcomes
Whether a particular movie was chosen/clicked. The proportion of movies seen that were chosen (everyone chooses 5, so this embeds variation in how many movies were seen). How that varies by race and/or gender of the recommender.
Secondary outcomes
How the proportion of movies seen that are by in group recommenders varies by treatment

Cited by

No citation data available for this registration yet.

Keywords