Testing a condensed methodology to estimate distributional preferences à la Fisman et al. (2007) – Follow-up Study

Last registered on February 22, 2023

View Trial History

Pre-Trial

Trial Information

General Information

Title

Testing a condensed methodology to estimate distributional preferences à la Fisman et al. (2007) – Follow-up Study

RCT ID

AEARCTR-0007217

Initial registration date

February 16, 2021

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published

February 17, 2021, 10:37 AM EST

First published corresponds to when the trial was first made public on the Registry after being reviewed.

Last updated

February 22, 2023, 9:06 AM EST

Last updated is the most recent time when changes to the trial's registration were published.

Locations

Country

United Kingdom of Great Britain and Northern Ireland

Region

Primary Investigator

Name

Malte Baader

Affiliation

University of Zurich

Contact Primary Investigator

Other Primary Investigator(s)

PI Name

Fabio Tufano

PI Affiliation

University of Nottingham

PI Name

Chris Starmer

PI Affiliation

University of Nottingham

PI Name

Simon Gächter

PI Affiliation

University of Nottingham

Additional Trial Information

Status

Completed

Start date

2021-01-13

End date

2021-04-02

Keywords

Behavior, Lab

Additional Keywords

Distributional preferences

JEL code(s)

C81, C91

Secondary IDs

Prior work

This trial does not extend or rely on any prior RCTs.

Abstract

Estimating individual distributional preferences has become a widely used technique in behavioural economics research. A frequently utilised methodology to do so has been put forward by Fisman, Kariv and Markovits (2007). In their methodology, subjects make distributional choices in 50 modified dictator games, which are then used to estimate two distinct utility parameters. One captures the relative weight for the own payoff (α) and the other describes an efficiency-equity trade off (ρ). In this study, we use the original and new data to explore whether we can reduce the number of distributional choices made by subjects whilst still accurately estimating preference parameters. Simulation results, already obtained by us, show high accuracy in estimating both parameters when using as few as 20 dictator games. In order to experimentally confirm our simulations, we collected a first wave (Wave 1) of data, indicating good performance of our proposed methodology. However, we are going to extend the analysis with additional data to explore some aspects of the results based on Wave 1 data.

External Link(s)

Registration Citation

Citation

Baader, Malte et al. 2023. "Testing a condensed methodology to estimate distributional preferences à la Fisman et al. (2007) – Follow-up Study." AEA RCT Registry. February 22. https://doi.org/10.1257/rct.7217-2.0

Sponsors & Partners

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Experimental Details

Interventions

Intervention(s)

Intervention Start Date

2021-02-22

Intervention End Date

2021-04-02

Primary Outcomes

Primary Outcomes (end points)

Distributions of preference parameters across distinct methodologies

Primary Outcomes (explanation)

We will estimate two distinct preference parameters, alpha and rho, based on a CES utility function using Tobit maximum likelihood estimations. These will then be compared across treatments.

Secondary Outcomes

Secondary Outcomes (end points)

Secondary Outcomes (explanation)

Experimental Design

Our proposed methodology only utilised 20 modified dictator games, rather than 50 as in Fisman et al. (2007). However, the price ratios of the games are not randomly drawn but the draw follows a bimodal distribution.
In Wave 1 of data collection we compared the original (Fisman et al., 2007) with our proposed methodology across the following three treatments:

1. Control – Original design using 50 modified dictator games.
2. Treatment 20 + 30 – First 20 dictator games according to our proposed methodology, last 30 follow the original design.
3. Treatment 20 & 20 – Two independent sets of our proposed methodology spaced out by a filler task.

If our proposed methodology is as effective as the original one, we should observe comparable distributions of alpha and rho parameters across treatments. From our simulation exercise, we predicted very consistent estimates of alpha and slightly more noisy estimates of rho. Overall, Wave 1 confirmed our predictions that the estimated parameters across the methodologies are identical, except for our estimates of alpha in ‘Treatment 20 + 30’. This seems to be an anomaly, as we also fail to obtain identical distributions of alpha parameters when estimating the parameter based on 50 dictator games, as in the original methodology.
As we collected data for the different treatments in two distinct weekdays, we hypothesise that the inconsistent estimations of alpha are likely to be a result of sampling error and imperfect randomisation.
Thus, in order to address this concern in this follow-up study we want to collect additional data for our ‘Control’ and ‘Treatment 20 + 30’ but with randomisation of subjects to conditions within a session. Through this randomisation strategy, we hope to diagnose the anomalous results from Wave 1.

Reference
Fisman, R., Kariv, S., & Markovits, D. (2007). Individual preferences for giving. American Economic Review, 97(5), 1858-1876.

Experimental Design Details

Randomization Method

We will randomise subjects within the same session using a random number generator.

Randomization Unit

Individuals

Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters

300 MTurk subjects

Sample size: planned number of observations

300 MTurk subjects

Sample size (or number of clusters) by treatment arms

150 MTurk subjects

Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

The study will be conducted using Amazon MTurk and all sample sizes are determined as a result of a power analysis utilising the effect size (Cohen’s δ = 0.44) found in Wave 1. The sample size below provides 90% power for a Kolmogorov-Smirnov test and 95% power for Wilcoxon-Mann-Whitney test. Thus, if the inconsistency is alpha is persistent we ensure that we have a high probability of detecting it.

Supporting Documents and Materials

There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

IRB