Communicating multiattribute preferences: An experiment

How to cite

Hakimov, Rustamdjan; Khanna, Manshu. 2024. "Communicating multiattribute preferences: An experiment." AEA RCT Registry. AEARCTR-0013135. https://doi.org/10.1257/rct.13135-1.0

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

Rustamdjan Hakimov* , Manshu Khanna

* Primary investigator

Registry ID
AEARCTR-0013135
DOI
https://doi.org/10.1257/rct.13135-1.0
First registered
Last updated
Status
In development
Trial dates
2024-03-05 – 2024-08-01
Countries
Private
JEL codes
D47
Registry record
http://www.socialscienceregistry.org/trials/13135

Abstract

The field of matching has recently made significant strides in the development of mechanisms for allocations, particularly where participants may express preferences over distinct attributes. In some applications, the preferences are multidimensional and often should be reported separately. For instance, in matching cadets to military branches in the US, participants report not only preferences over the branches but also over the length of the contract. In Chinese college admissions, applicants report the university, field of study and tuition waiver as separate attributes. This raises a practical issue of the potential complexity of preference reporting by participants. While theory ignores this complexity and assumes that agents could rank all possible combinations of attributes, practitioners challenge this view and often opt for separate ranking of attributes, which typically leads to welfare loss, as the prominent mechanism needs to operate on a single ranking. The loss can be especially high if the different attributes might exhibit complementarities and the payoff is not monotonic in quantity.

Our study aims to develop and test preference reporting methods that effectively balance theoretical richness with practical simplicity for participants. To this end, we examine six preference reporting methods under three levels of preference complexities. One method is the richest—ranking the smallest items, which we call "bundles," i.e., all combinations of attributes. It reaches full efficiency in theory under any complexity of preferences but is behaviorally complex. Two other methods are simpler and currently used in practice but limited in expressing complex preferences, leading to efficiency loss theoretically. The last three, our contributions, that we hypothesize will improve efficiency by simplifying the task of reporting.

Intervention

The study is laboratory experiment.
In our experiment, participants will partake in the college admission market and will be assigned to programs. Each program is characterized by University prestige, measure of fit of the field, and tuition discount. ”Bundle” refers to ranking each program, which might be hard as there are in total 24 programs. As an innovation, participants are assigned a utility function that determines the score of each program, which they should use to infer ordinal ranking over programs. The payoff of participants will depend on the rank of the program they will be matched. We run the following treatment in the complexity of the preference dimension within-subjects:
1. Lexicographic preferences (LEX): Participants will have additive preferences over the three features (university, field of study, and tuition discount) with the university being the most important, then the field, then tuition.
2. Separable Preferences (SEP): Participants will have additive preferences over the three features (university, field of study, and tuition discount), but the weights of each component might differ, thus violating the lexicographic structure.
3. Non-separable Preferences (COMP): Participants will experience a complementary component in the utility function, where the importance of the tuition discount is contingent on the field of study.

Experimental design

Our main interest is different reporting languages. We run three preference reporting treatments between-subjects.
• Separate ranking of attributes with exogenous weights: Participants will rank each attribute separately, and the designer will aggregate these rankings and assign weights externally, imposing the lexicographic structure. Prediction: Participants can precisely communicate their preferences only in LEX.
• Separate ranking of attributes with reported weights: Participants will rank each attribute separately and report the weights assigned to each attribute. Prediction: Participants can precisely communicate their preferences only in LEX and SEP.
• Bundles: Participants report a rank-ordered-list (ROL) of programs. Prediction: Participants can precisely communicate their preferences in all three preference treatments: LEX, SEP, and COMP.

Design details

Randomization method
Computerized randomization into treatments.
Randomization unit
Session of 27
Clusters
6 sessions per treatment
Observations
972
Arms
972
Primary outcomes
Efficiency, truthfulness

Cited by

No citation data available for this registration yet.

Keywords