The Two-Sided Barrier to Women's Work: Experimental Evidence from Firms and Households in Pakistan

There are two interventions, one on each side of the labor market.

The firm intervention is a ten-minute documentary-style film shown to hiring managers. Two mid-sized garment firms from Punjab describe how they began hiring women onto their production lines, what it cost them, and what happened afterwards to defect rates, productivity, the range of products they make, and the markets they sell to. The film ends with practical advice on how to start, beginning with one line of women on the same contracts as men.

The household intervention has three parts, delivered together by an enumerator in a single visit. First, a structured exercise, called decisional balance sheet (DBS), in which the household head and the target woman lay out the decision as a ledger. For two situations, the woman working in a garment factory and the woman staying home, they name up to three benefits and three costs, both for the household and for the woman herself, and rate how much each one matters. The enumerator reads the balance back to them, asks what would make the costs that matter tolerable, and records where the household stands. Note that the DBS exercise argues for nothing and contains no persuasive material. Second, every household receives a printed sheet with the names, addresses, and phone numbers of up to three nearby garment factories. Third, in random half the treated blocks, households also watch a five-minute docuentary-style short film covering the things the exercise tends to raise as worries (or costs), such as the work environment, mixed-gender workplaces, commuting safety and employer transport, pay, hours, maternity and sick leave, and testimony from women already doing this work, and from husbands who support their wives working.

2026-04-20

2026-05-25

Firms (200 firms):
1. Female share of the workforce (admin data).
2. Number of women hired in the three months before the survey (admin data and self reported).
3. Revealed willingness to hire women, from the telephone audits via standardized prospective workers.

Households (4,000 households across 200 census blocks, with around 8,000 respondents):
4. Whether the woman searched for paid work outside the home in the last three months.
5. Whether she contacted or visited an employer.
6. Whether she is currently working for pay outside the home, and separately whether she works in a garment factory.

On the firm side: managers' beliefs about how female workers perform (retention, training time, absence, share of line tasks women could do); managers' beliefs about what other managers think and do; hiring intentions, vacancies, wages paid to women, and the amenities the firm offers women; hypothetical hiring decisions in vignettes where the applicant's gender varies; how many people approached the firm looking for work and how many were women; firm sales, defect rates, capacity use, products made and export markets entered.

On the household side: what households believe about garment work (pay, commute time and cost, hours, education needed, amenities, safety); the quality of the household's reasoning, measured through decision vignettes; attitudes, norms and beliefs about women working, and second-order beliefs about what neighbors think; whether the household says the woman may work and on what conditions; women's mobility, permission and decision-making within the household.

Spillovers: search, employment, beliefs and norms among one friend per target woman, named at baseline, living in the same block but deliberately drawn from outside the 20 sampled households; and search and employment among other (non-targeted) household members.

The study is a 2x2 factorial experiment at the level of the firm and its neighboring census block.

We recruited 200 garment manufacturers with at least 50 employees across three districts of Punjab: Sialkot, Faisalabad, and Lahore. Using census blocks list from the Punjab Bureau of Statistics, we matched each firm to a census block within a 30-minute commute, connected by road, and not a high-income neighborhood. A census block is the smallest unit used in the national census and holds 200 to 250 households. This gives 200 firm and block pairs, which we call dyads.

Within each block we selected 20 households containing at least one woman aged 16 to 45, using systematic random sampling with a fixed skip interval. In each household we interview the household head, the target woman, and, where the head is not himself an adult male, a secondary adult male. Within each firm, we sample at least one manager, giving priority to those with hiring authority.

After the baseline, the 200 dyads were randomly assigned in equal numbers to four groups of 50: no intervention (control); the demand-side treatment (film) only; the supply-side treatment (household package) only; and both. Within the two groups receiving the household package, half the blocks were randomly assigned to also receive the household film. Assignment was stratified by district.

We follow firms and households roughly three months after the intervention and again roughly eight months after. At the second follow-up, funding covers half the households in each block, so we will survey a random 50% of households per block and all 200 firms, keeping all 200 dyads intact.

Not available

Randomization was done in the office by a computer, using Stata.

There are two levels.

The main randomization is at the level of the firm and block dyad: one garment firm together with the census block matched to it. Both the firm treatment and the household treatment are assigned at this level, so every firm and every sampled household in a given dyad share the same assignment for both factors.

The second randomization, which decides whether households also see the information film, is at the census block level.

Yes

200 firm-block dyads, each made up of one garment firm and one adjacent census block.

200 firms. 4,000 households (20 per block), with roughly 8,000 individual respondents, since each household contributes the household head and the target woman. In case a household head is not an adult male, a secondary adult male. Up to 4,000 additional spillover women at endline, one friend per target woman, living in the same block but outside the sampled 20 households. 400 telephone audit calls, two per firm. At the second endline the household sample falls to 2,000 households, a random half in each block. The firm sample stays at 200.

50 dyads control (50 firms, 1,000 households)
50 dyads firm film only (50 firms, 1,000 households)
50 dyads household package (50 firms, 1,000 households) of which:
25 dyads decision exercise + contacts, no film (500 households)
25 dyads decision exercise + contacts + film (500 households)
50 dyads both treatments (50 firms, 1,000 households) of which:
25 dyads decision exercise + contacts, no film (500 households)
25 dyads decision exercise + contacts + film (500 households)

Calculated at 80% power and a 5% two-sided significance level. Firm outcome: female share of the workforce. The firm is the unit of assignment, so no clustering adjustment applies. Comparing one treatment arm of 50 firms with 50 control firms, the minimum detectable effect is 8.5 percentage points. The baseline mean is 13% and the standard deviation across firms is 15 percentage points. Pooling the two arms that contain the firm film against the two that do not, 100 firms against 100, the figure falls to 6 percentage points. Household outcome: woman works for pay outside the home. Assignment is at the block, with 50 blocks per arm and 20 sampled households per block, so 1,000 women per arm. The baseline rate is 18%, which fixes the standard deviation of the indicator at 0.384. Assuming an intracluster correlation of 0.05, the design effect is 1.95 and the minimum detectable effect is 6.7 percentage points, which is about 37% of the baseline rate. At an intracluster correlation of 0.02 the figure is 5.7 percentage points, and at 0.10 it is 8.2pp.