Referral Protocols for Achieving Gender Representation in Random Digit Dial Phone Surveys

How to cite

Glazerman, Steven; Grepin, Karen; Mueller, Valerie. 2021. "Referral Protocols for Achieving Gender Representation in Random Digit Dial Phone Surveys." AEA RCT Registry. AEARCTR-0007765. https://doi.org/10.1257/rct.7765-1.0

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

Steven Glazerman* , Karen Grepin , Valerie Mueller

* Primary investigator

Registry ID
AEARCTR-0007765
DOI
https://doi.org/10.1257/rct.7765-1.0
First registered
Last updated
Status
On going
Trial dates
2020-11-23 – 2021-09-30
Countries
Private
JEL codes
C83, J16
Registry record
http://www.socialscienceregistry.org/trials/7765

Abstract

In many applications of empirical economics, sample frames are not available, so random digit dial (RDD) surveys are the only feasible method for obtaining telephone survey data. This creates a challenge when the study seeks to focus on or disaggregate data to a particular demographic group, such as adult women. In most Low- and Middle- Income Countries (LMIC), cellphone ownership is skewed towards men (Forenbacher et al., 2019). This impedes the ability to obtain gender-representative samples in RDD surveys. A survey that relies on women to own and use mobile phones may systematically exclude women who rely on other household members, such as a spouse, to use a mobile phone. On the other hand, a survey that relies on men's willingness to pass the phone to a female household member may systematically exclude women in households with specific cultural gender norms such as those that discourage women's self-expression. In both cases, there is insufficient published data to determine which method would yield the highest response rate, reduce coverage bias, and be most cost effective.

This trial seeks to test the hypothesis that rapport-building can increase the number of women and diversify the types of women who will take and complete an RDD survey when referred by a male household member. Specifically, we test how referral and screening designs produce samples of women. We consider the impacts of these referral protocol choices on not only completions and response rates, but on sample composition and on inferences of interest from the survey data themselves to balance potential bias-variance tradeoffs.

Intervention

We will test two protocols for reaching women respondents: (A) referral design vs. (B) simple screening based on eligibility criteria. Each case is randomly assigned into one of these groups.

Protocol A, Referral design:
-If a woman answers the phone:
o We determine eligibility and ask all eligible women to complete the survey.
-If a man answers the phone:
o We ask the man if there is an eligible woman in his household or nearby, ask him if he is willing to pass his phone, and if so, we speak to the woman, determine her eligibility, and ask her to complete the survey

Protocol B, Screen Design
-If a woman answers:
o We determine eligibility and if woman is eligible, ask her to complete the survey.
-If a man answers:
o We determine eligibility and if man is eligible, ask him to complete the survey (no woman survey attempted for this household).

Experimental design

Telephone numbers will be contacted to determine if there is a potential respondent at that number. If a respondent picks up the phone, a script will be read to introduce the survey and the goals of the research study, they will be screened for eligibility (and for passing the phone if necessary), consent will be obtained, and then they will be asked if they can complete the survey.

Respondents will be automatically randomized into each group by the survey software based on 50% probability of assignment. Based on this randomization, the survey software will automatically direct enumerators to either pass the phone (Protocol A) or screen the respondent for eligibility (Protocol B). Enumerators are trained in both protocols and are prompted to complete the respective by the survey software.

Design details

Randomization method
Survey software (SurveyCTO) randomizes survey branching behavior.
Randomization unit
Individual cases (phone numbers) are randomized
Clusters
6,929 cases to attempt
Observations
6,929 cases to attempt Experiment is run conditional on eligible respondents, so sample size will be closer to 1,500.
Arms
Eligibility is determined after randomization, so sample size by treatment arm cannot be pre-specified. Estimates: Protocol A: 1000 women Protocol B: 500 women
Primary outcomes
These protocols are expected to affects three sets of outcomes: -Productivity for survey implementation (consent and completion) -Sample composition (demographic characteristics) -Sample outcomes (Coronavirus knowledge and practices, employment, access to health services, government support, and time use)
Secondary outcomes
The referral design will also provide information on how successful referral to female respondents may be.
Partners
Name: Innovations for Poverty Action Partner_type: ngo Url: https://www.poverty-action.org

Cited by

No citation data available for this registration yet.

Keywords