Decision-Making under Algorithmic Recommendation

How to cite

Listo, Ariel. 2023. "Decision-Making under Algorithmic Recommendation." AEA RCT Registry. AEARCTR-0011532. https://doi.org/10.1257/rct.11532-1.2

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

Ariel Listo*

* Primary investigator

Registry ID
AEARCTR-0011532
DOI
https://doi.org/10.1257/rct.11532-1.2
First registered
Last updated
Status
On going
Trial dates
2023-08-07 – 2024-03-31
Countries
Private
JEL codes
D81, D90
Registry record
http://www.socialscienceregistry.org/trials/11532

Abstract

I study the role of ambiguity preferences in predicting take-up of algorithmic (i.e., personalized) recommendations. In a two-part experiment, I first elicit individuals's preferences towards risk and ambiguity. In Part 2, I ask participants to make choices over a menu of risky prospects after I provide a recommendation (or not) based on their treatment assignment. I assign individuals to one of two control groups (no recommendation or randomly created recommendation) or to one of two treatment groups (personalized recommendation with no information or personalized recommendation with information). The difference between the two treatments is only the information provided to individuals on how the recommendation is constructed; both recommendations are of similar quality (i.e., based on the individual's risk preferences elicited in Part I).

Experimental design

In a two-part experiment, I first elicit individuals's preferences towards risk and ambiguity. In Part 2, I ask participants to make choices over a menu of risky prospects after provide a recommendation (or not) based on their treatment assignment. I assign individuals to one of two control groups (no recommendation or randomly created recommendation) or to one of two treatment groups (personalized recommendation with no information or personalized recommendation with information). The difference between the two treatments is simply the information provided to individuals on how the recommendation is constructed, while both recommendations are of similar quality (i.e., based on the individual's risk preferences elicited in Part I).

Design details

Randomization method
Randomization done by Qualtrics
Randomization unit
Randomization conducted at the individual level
Clusters
600 Prolific participants
Observations
600 Prolific participants
Arms
150 participants in Control I, 150 participants in Control II, 150 participants in Treatment I, and 150 participants in Treatment II
Primary outcomes
Take-up of personalized recommendation

Cited by

No citation data available for this registration yet.

Keywords