Evaluating the Impacts of Skills Training and Access to Digital Platforms on Female Domestic Workers in Bangladesh

This trial will evaluate the joint and separate effects of two interventions. The first intervention is a skills training program developed by our partner NGO that seeks to build female domestic workers' life and occupational skills. The training covers interpersonal skills, literacy/numeracy, issue-based life skills, financial literacy, as well as imparts training specific to the skillset needed to be a successful domestic worker (namely, cooking, cleaning, etc.). The training is administered in groups of up to 20 women and takes a total of 14 full (non-consecutive) days.

The second intervention will facilitate onboarding of female domestic workers to an existing digital job platform operational in Bangladesh. This platform allows households looking to hire a domestic worker to match with a female domestic worker searching for a job via a phone-based app in a manner similar to Handy or Uber. Domestic workers can use a feature phone or a smartphone to register on this platform and find domestic work. The app allows options for the domestic worker to take an instant job, in which an appointment is scheduled for an agreed-upon amount of time for a set of tasks. Domestic workers receive appointments through automated calls on their phone and are paid through digital phone-based banking. The onboarding process will comprise a 3-4 hour-long in-person orientation session, followed by an phone-based IVR training to improve workers' familiarity and comfort with the job platform. The app rewards workers with small payments for completing the IVR training.

2023-01-30

2023-11-05

a. Whether respondent is currently working as a domestic worker at endline
b. Respondent's monthly earnings from domestic work (in the last month before endline)
c. Respondent's monthly earnings from non-domestic work (in the last month before endline)
d. Number of houses worked for by the respondent as a domestic worker during the past week at endline
e. Respondent's hours spent on paid domestic work during the past week at endline
f. Respondent's hours spent searching for domestic work during the past week at endline
g. Whether respondent is registered on a digital job search platform at endline
h. Whether respondent is an active user of a digital job search platform at endline
i. Whether respondent found at least one job through a digital job search platform during the intervention period
j. Whether respondent has tried to negotiate (salary/ hours/ something else) with an employer during the intervention period
k. Respondent's bargaining power relative to employers / outside option

A respondent's/ domestic worker's bargaining power will be measured by her level of agreement with the following three survey questions with regard to her (current/ past) primary employer:
1.If I left this job or if this house let me go, it would be difficult for me to replace this job
2.The people I work for in this house have more say in my compensation than I do
3.The people I work for in this house have more say on my working conditions and hours than I do

The following variables for the respondent will be used as secondary outcomes:
a. Job satisfaction at endline
b. Unmet job needs at endline (i.e., gap between desired and actual attributes of a job)
c. Distance/ commuting time to job at endline
d. Workplace abuse experienced during the intervention period
e. Agency at endline
f. Experience of IPV and spousal control during the last 6 months or intervention period
g. Mental and physical health at endline
h. Social connections (job-related and others) at endline
i. Locus of control at endline
j. Mobile phone ownership at endline
k. Ability/ comfort with mobile phone use at endline
i . Whether respondent has a mobile money account in her name at endline
j. Whether respondent received any payment for domestic work through her mobile money account during the intervention period
m. Whether respondent has a skill certification
n. Whether respondent is a part of a women's group/ union

a. Job satisfaction will be measured by survey items: "How suitable do you feel to do the tasks asked of you by this employer on a scale of 1-10, with 1 being not suitable at all and 10 being very suitable;" agreement with "I am comfortable talking to this home about my professional needs, such as taking breaks and compensation rates;" and agreement with "I am comfortable talking to this home about my personal needs, such as family issues or sickness that interferes with my work", with regard to the primary employer for both current and past domestic workers.
b. Worker unmet need/ preferences refer to the participants' desired number of houses/ employers, desired commute time and work location, and desired number of weekly work hours. We will compare worker reported preferences to worker reported current circumstances in these areas.
c. Commute time measured in minutes
d. Workplace abuse is measured as the experience of verbal, sexual, physical, or economic abuse by an employer within the past 6 months. Economic abuse by an employer is reported as the experience of having salary withheld, asked to leave the job unexpectantly, forced to work more than agreed, or forced to do an activity not in the job description that she did not want to do.
e. Women's agency will be a standardized scale generated from a set of questions on decision-making participation in the home, such as participation in how to spend own earnings.
f. IPV is measured using the DHS intimate partner violence scale as well as a validated partner economic control scale, measuring any experience in the past 6 months.
g. Mental health is captured by a validated stress scale and validated anxiety scale. Physical health is captured by the questions: "In the past 6 months, how has your overall physical health been?" (very poor, poor, ok, good very good), "In the past 6 months, have you had persistent pain anywhere on your body?", and "In the past 6 months, have you experienced any workplace injury related to work activities?"
h. Social connections are measured as the number of friends a woman reports, both overall and in terms of friends she speaks to about job opportunities.
i. Validated 4-item Locus of Control Scale

We use a clustered RCT (randomized controlled trial) strategy to identify the causal link between the interventions and outcomes of interest. Our baseline sample comprises 4,391 domestic workers in 238 clusters or "training groups". The unit of observation will be a female domestic worker but randomization will take place at the cluster level. We divide our sample into 4 arms with approximately equal number of groups:

1. Skills training arm
2. App onboarding arm
3. Skills training + App onboarding arm
4. Control arm

Using individual-level regressions, we will estimate intent-to-treat effects as follows:
a. To determine the effect of skills training, we will compare endline outcomes of respondents assigned to the Skills training arm with respondents assigned to the control arm.
b. To determine the effect of app onboarding, we will compare endline outcomes of respondents assigned to the app onboarding arm with respondents assigned to the control arm.
c. To determine the joint effect of skills training and app onboarding, we will compare endline outcomes of respondents assigned to the skills training + app onboarding arm with respondents assigned to the control arm.

Randomization was conducted using STATA 17.0 in an office. Random assignment of training groups was stratified on neighborhood, training group size, and percent of the group that is currently working as a domestic worker.

Training groups

Yes

238 training groups

4,391 female domestic workers

61 training groups - Skills training only
59 training groups- Skills training plus app onboarding
57 training groups- App onboarding only
61 training groups- Control

We will be able to achieve MDEs of 0.50 standard deviations for continuous outcomes and 10% for most binary outcomes with a control group of 61 clusters and a treatment group of 57 clusters with a compliance rate of 80%. Given the treatment effects anticipated by our partners, we expect these MDEs to be small enough to capture moderate treatment effects. This power may be improved by using an ANCOVA specification which controls for baseline outcome variables and jointly estimating treatment effects.