Measuring sensitive outcomes through online ballots

How to cite

Gazeaud, Jules; Crepon, Bruno; Elsayed, Ahmed. 2025. "Measuring sensitive outcomes through online ballots." AEA RCT Registry. AEARCTR-0017301. https://doi.org/10.1257/rct.17301-1.0

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

Jules Gazeaud* , Bruno Crepon , Ahmed Elsayed

* Primary investigator

Registry ID
AEARCTR-0017301
DOI
https://doi.org/10.1257/rct.17301-1.0
First registered
Last updated
Status
In development
Trial dates
2025-11-22 – 2025-11-28
Countries
Egypt ()
JEL codes
C81, C83
Registry record
http://www.socialscienceregistry.org/trials/17301

Abstract

A persistent challenge in empirical research is how to obtain truthful answers to sensitive questions, as misreporting can systematically bias estimates and misinform policy. In Crépon et al. (2025), we introduce the ballot-bag, a novel technique where respondents privately report their answers to one or more sensitive questions on anonymous paper sheets, which they then fold and place into a transparent bag held by an enumerator. We provide encouraging evidence on the method’s potential in the context of in-person surveys in Egypt. This add-on project seeks to adapt and test the method’s effectiveness in an online survey setting.

Experimental design

Respondents are recruited on the global online survey platform CINT. Given the nature of the sensitive outcomes, we focus on Egyptian males aged 18–35, residing in Egypt. We randomize respondents into two groups. In the first group, outcomes 1) and 4) are asked directly (DQ), while outcomes 2) and 3) are elicited using the ballot method (BB). In the second group, the assignment is reversed. We plan to estimate a basic linear probability model, regressing each outcome on a dummy indicating whether it is measured using BB. We expect BB to yield higher prevalence for sensitive outcomes (outcomes 1–3), given prior evidence that data on sensitive outcomes elicited using DQ often suffer from underreporting. To maximize statistical power, we will conduct one-sided tests with the alternative hypothesis that the BB coefficient is positive.

Design details

Randomization method
Randomization is done by computer directly within Qualtrics.
Randomization unit
The unit of randomization is individual.
Clusters
The experiment will run for one week, or until 2,000 eligible individuals complete the survey.
Observations
The experiment will run for one week or until 2,000 eligible individuals complete the survey.
Arms
Targets: - 1,000 in Group 1 (DQ for outcomes 1–4, BB for outcomes 2–3) - 1,000 in Group 2 (DQ for outcomes 2–3, BB for outcomes 1–4)
Primary outcomes
We aim to assess measurement differences between direct questioning and online ballots for four outcomes: 1) whether respondents plan to migrate to Europe without official papers; 2) whether they have ever consumed alcohol, even once; 3) whether they have ever used drugs, even once; and 4) whether they met any of their friends yesterday.

Cited by

No citation data available for this registration yet.

Keywords