Regret and preferences for randomization

How to cite

Loewenfeld, Moritz. 2025. "Regret and preferences for randomization." AEA RCT Registry. AEARCTR-0015997. https://doi.org/10.1257/rct.15997-1.1

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

Moritz Loewenfeld*

* Primary investigator

Registry ID
AEARCTR-0015997
DOI
https://doi.org/10.1257/rct.15997-1.1
First registered
Last updated
Status
Completed
Trial dates
2025-05-15 – 2025-05-30
Countries
Austria ()
JEL codes
D81
Registry record
http://www.socialscienceregistry.org/trials/15997

Abstract

A growing body of literature documents preferences for randomization in decision-making under risk. When faced with a choice between two lotteries, many experimental subjects prefer to delegate the decision to a coin flip. We propose a novel model in which individuals randomize to hedge against regret from unfavorable outcomes. This model explains puzzling findings in the literature, such as high rates of randomization even when one option first-order stochastically dominates the other. We test a distinct prediction in an online experiment: (1) subjects randomize more when prospects are more negatively correlated.

Experimental design

We study preferences for randomization in risky decision-making under different correlation structures. In our online experiment, participants choose probability mixtures between two risky lotteries. Each lottery has two possible outcomes with known probabilities. Participants are required to choose a probability with which to play each lottery, ranging from 0% to 100% (including corner solutions).
There are three within-subject conditions. 1) the lotteries are independent 2) the lotteries are perfectly positively correlated 3) the lotteries are perfectly negatively correlated. This allows us to test whether preferences for mixing depend on the correlation structure, as predicted by our model. Subjects encounter the three within-subject conditions in random order.
For details, see the document PAP.pdf.

Design details

Randomization method
Computer
Randomization unit
The unit of randomization is the individual participant. Each participant is exposed to all correlation structures (independent, positively correlated, negatively correlated) in randomized order. Within each structure, the order of decision tasks is also randomized at the individual level to control for order effects.
Clusters
140 participants. Who pass a predefined criterion. For details, see the document PAP.pdf.
Observations
18 choices by each of 140 participants.
Arms
All 140 individuals participate in all three treatment arms (within-subject design): 140 individuals: Independent correlation structure 140 individuals: Positively correlated structure 140 individuals: Negatively correlated structure
Primary outcomes
Probability of choosing a first-order stochastically dominant lottery under three different correlation structures: 1) state-wise 2) independent 3) maximally negatively correlated. For details, see the document PAP.pdf.

Cited by

No citation data available for this registration yet.

Keywords