Jobs of the Future and Gender Norms in India

This project explores whether a digital persuasion approach can shift entrenched gender and masculinity norms in India. Partnering with Karya – an organization that provides digital employment to economically disadvantaged communities across India – this study randomly assigns workers to read, transcribe, and annotate content designed either to subtly challenge traditional gender norms or remain gender neutral. By leveraging insights from social psychology, including motivated reasoning and subliminal priming, I investigate if implicit cues, rather than explicit messages, can more effectively unravel rigid patriarchal norms among adults. This randomized field experiment uniquely applies persuasion through AI-generated content, aiming to reveal mechanisms by which routine tasks can influence deeply held attitudes without resistance. Given India’s entrenched gender disparities in economic, social, and political spheres, this study offers evidence leveraging a scalable, light-touch approach toward reshaping gender norms.

2025-06-02

2025-12-01

Attitudes toward masculinity and gendered household norms, time use, agency of women in participant households, gender bias in the workplace, willingness to allow spouses to participate in the workforce

This study measures changes in gender attitudes and behaviors through the following key outcomes:
1. Women’s Agency: An index will be constructed following Jayachandran et al. (2022) by aggregating responses to gender attitudes-related questions. This index will be measured at baseline and endline to assess shifts in agency within participating households.
2. Self-Reported Gender Attitudes and Beliefs: These will be collected before and after the month-long intervention to track attitudinal shifts.
Recognizing the potential for experimenter demand effects in self-reported measures, particularly in a work setting, the study will also assess the following behavioral outcomes:
3. Time Use: Data on time allocation for male and female participants will be collected regarding a typical weekday in the prior week, enabling an assessment of gender disparities in household duties across treatment and control groups.
4. Gender-Based Discrimination: At endline, participants will complete a data validation task evaluating a colleague’s work. The gender of the original worker will be randomized in half the tasks, allowing measurement of bias in professional evaluations across treatment and control groups.
5. Willingness to Allow Wife to Join the Firm: Male respondents will be informed of the implementation collaborator’s plans to expand its worker pool and asked whether they would refer their wives, serving as a behavioral indicator of shifting gender norms regarding women's workforce participation. The implementation collaborator will also provide administrative data on whether the wives who were thus onboarded through these referrals actually worked when assigned projects in the long-run.

As part of my experiment, male and female employees will be randomly assigned to one of two treatment conditions where the content they engage with on the app to complete tasks will be interspersed with gender norms-targeted content or will remain gender-neutral. The content will be specifically designed to reverse male breadwinner and gendered household labor norms, harmful masculinity norms, and gender bias in the workplace. These specific norms have been selected to address the underlying beliefs driving low and stagnant female labor force participation in India.

Not available

Randomization will be implemented by the researcher (using a randomization script implemented on standard statistical software). Finally, randomization will be implemented at the individual level.

Individual (i.e. employee level)

No

N/A

630 men and 600 women

Latent treatment: 410
Explicit treatment: 410
Control: 410

The proposed sample size was chosen based on power calculations with the following parameters: 80% power, 100% treatment take-up, and 80% non-differential attrition across experimental groups, and at the 95% confidence level. The entire sample is split into the three experimental groups: latent persuasion, explicit persuasion, and control groups equally. Note that 100% treatment take-up is entirely feasible in the context of this experiment, since the treatment is embedded in the workday of workers that are part of the field experiment. I have drawn a back-up sample to minimize attrition from unanticipated dropouts in the selected employee pool during the month-long study. The randomization is being done at the individual level therefore there is no need to cluster standard errors in this experiment design.