Survey Design for Sensitive Information in Organizations

How to cite

Gonzalez-Torres, Ada; Boudreau, Laura; Chassang, Sylvain; Heath, Rachel. 2021. "Survey Design for Sensitive Information in Organizations." AEA RCT Registry. AEARCTR-0007103. https://doi.org/10.1257/rct.7103-2.1

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

Ada Gonzalez-Torres* , Laura Boudreau , Sylvain Chassang , Rachel Heath

* Primary investigator

Registry ID
AEARCTR-0007103
DOI
https://doi.org/10.1257/rct.7103-2.1
First registered
Last updated
Status
Completed
Trial dates
2021-09-09 – 2021-11-15
Countries
Bangladesh (South Asia)
Registry record
http://www.socialscienceregistry.org/trials/7103

Abstract

This research studies how survey design affects transmission of sensitive information within organizations. We conduct a phone-based survey experiment with workers at two garment factories in Bangladesh to study how survey design affects their willingness to report misbehavior by managers, including threats, physical harassment, and sexual harassment. We experimentally vary whether the survey elicitation method provides plausible deniability when asking sensitive questions. In particular, building on Chassang and Padró i Miquel (2018) and Chassang and Zehnder (2019), we use hard garbling to provide plausible deniability by exogenously distorting survey responses. We also experimentally vary the extent to which the survey enumerator builds rapport with the surveyed individual and the level of identifiability of a surveyed worker.

Experimental design

We experimentally vary direct elicitation and a hard garbling design to ask sensitive questions on management misbehavior, as well as the extent to which the enumerator builds rapport or trust with the surveyed individual, and the level of identifiability of a surveyed worker through demographic questions.

In the survey, we also elicit workers’ primary deterrents to reporting, and use demographic and other less sensitive survey measures to predict their probability of being victimized.

Design details

Randomization method
The randomization is done by a computer.
Randomization unit
The unit of randomization is a worker, stratified by factory-production team and sex.
Clusters
Not applicable.
Observations
Approximately 2,620 workers.
Arms
We cross-randomize different treatment conditions and the sample size varies from 220 to 440 workers per treatment condition (see pre-analysis plan for details).
Primary outcomes
The key outcome variables are threats, physical harassment and sexual harassment (including gender-based violence) from direct supervisor.
Secondary outcomes
Additionally, as secondary outcome, if we find that one or more of our experimental methods increases reporting, we will study how increased reporting affects workers’ well-being in the short run.

Cited by

No citation data available for this registration yet.

Keywords