Bringing Work Home: Flexible Work Arrangements as Gateway Jobs for Women in West Bengal2026

I participated in this paper as a research assistant.

This paper uses an experiment in West Bengal to answer many different questions relating to women's ability to work. From my understanding, the implications of this paper is to say that under a system of patriarchal hegemony that is given¹1Another approach would be to change the norm., what are the most helpful interventions that can help women who want jobs to obtain them. The intervention is to offer flexible jobs from inside the home. Additionally, under the stepping stone approach to norms, these interventions can weaken the hegemony. Once women work, they become less traditional in their gender attitudes and become more accepting of less flexible jobs outside of the home. 

The authors divide a sample of 1,670 women²2Contacted with the help of Calcutta Foundation., with an Android smartphone and who were out of the labor force³3Not employed and not looking for work. and not in a training course and could read Bangla or Hindi, in a control group of 420 women and a treatment group of 1250 women. The treatment. group is offered one of five possible jobs: Home: Fully Flexible, Home: Time Inflexible, Home: Child Inflexible, Home: Time and Child Inflexible, and Office. Then, 50% of the non-Home: Fully Flexible treatment group is surprised with an upgrade offer to the Home: Fully Flexible job after deciding to accept or refuse the first offer. After ten weeks an end-line survey is conducted with a second round of job offers is done. The paper then presents the following estimations:  

\[ \text{StartsWork}_i = \sum_{j=1}^{5} \beta_j \, \mathbb{1}\{\text{FinalWorkArrangement}_i = j\} + \mu_s + \varepsilon_i \]where \(\text{StartsWork}_i\) equals one if woman \(i\) starts the Round 1 job she was offered, and zero otherwise. \(\mathbb{1}\{\text{FinalWorkArrangement}_i = j\}\) equals one if her final job offer is \(j\), which is of the five job options. \(\beta_j\) is the take-up rate for job \(j\) compared to the no offer control group. \(\mu_s\) are strata fixed effects (area, smartphone ownership, child under eight). \(\varepsilon_i\) is the error term. Moving from the office job to the fully flexible home job raises the chance a woman starts work by 33 percentage points⁴4This is shown in Figure II (from 15% to 48%). The KEY result is that allowing for childcare is doing almost half of the work for this 33 point effect. Letting a woman work from home, with fixed hours and no children allowed, raises the chance she starts work by 15 percentage points (from 14% to 29%). However, letting her choose her own hours has no effect⁵5Table A.7 Column 1. on whether she starts work: 46% vs 48% when children are allowed (p = .69), and 29% vs 29% when they are not (p = .91). We then have the following estimation:

\[ y_i = \beta_1 \mathbb{1}[R1 > R2]_i + \beta_2 \mathbb{1}[R1 \leq R2]_i + \gamma_{r(i)} + X_i'\delta + \mu_s + \varepsilon_i\]where \(y_i\) equals one if woman \(i\) starts the Round 2 job she was offered. \(\mathbb{1}[R1 > R2]_i\) equals one if her Round 1 job was more flexible than her Round 2 job. \(\mathbb{1}[R1 \leq R2]_i\) equals one if her Round 1 job was less flexible than or as flexible as her Round 2 job. The control group here got no Round 1 offer. \(\beta_1\) is the effect of a more flexible first job on taking up the second job. \(\gamma_{r(i)}\) are fixed effects for which Round 2 job she was offered. \(X_i\) are controls picked by double post-LASSO. \(\mu_s\) are strata fixed effects. For women who never worked before⁶6Table III Columns 3-4., getting a first job that is more flexible than the second one raises the chance that she takes the second job by 9 percentage points, all else equal. For women who worked before⁷7Table III Columns 5-6., the effect is not statistically significant. The third equation is 

\[ y_i = \beta \mathbb{1}\{\text{JobOffer}_i = 1\} + X_i'\delta + \mu_s + \varepsilon_i \]where \(y_i\) is an outcome such as the gender attitudes index, the agency index, or the well-being index. These are measured at endline for woman \(i\). Then  \(\mathbb{1}\{\text{JobOffer}_i = 1\}\) is one if she was offered any Round 1 job. \(\beta\) is the effect of being offered a job compared to the control group. \(X_i\) are controls picked by double post-LASSO. \(\mu_s\) are strata fixed effects again. Here we have the result on gender attitues: women who were offered a job had their gender attitudes become 0.06 standard deviations less traditional. For women who started out more traditional, the shift is 0.11 standard deviations⁸8Figure IV Panel B. However, being offered a job has no effect on well-being (0.01, p = .86) or agency.

\[ y_{it} = \beta_T \mathbb{1}\{\text{FinalJob}_i = \text{Office}\} + X_{it}'\delta + \mu_s + \varepsilon_{it} \]

where \(y_{it}\) is the performance of worker \(i\) on task \(t\), measured as accuracy (0 to 2 scale) or seconds to finish the task. The sample is women who accepted the office job. \(\mathbb{1}\{\text{FinalJob}_i = \text{Office}\}\) equals one if she stayed in the office, and zero if she was randomly upgraded to Home: Fully Flexible. \(\beta_T\) is the effect of working from the office compared to working from home, for the same type of worker. Working from the office instead of home lowers time per task by 4.65 seconds⁹9Table IV Panel A Column 3. Finally, the last estimating equation is:

\[ y_{it} = \beta_S \mathbb{1}\{\text{InitialOffer}_i = \text{Home: Fully Flexible}\} + X_{it}'\delta + \mu_s + \varepsilon_{it} \]

where \(y_{it}\) is accuracy or seconds for worker \(i\) on task \(t\). The sample is everyone who ended up working in Home: Fully Flexible. \(\mathbb{1}\{\text{InitialOffer}_i = \text{Home: Fully Flexible}\}\) equals one if she was first offered the fully flexible job, and zero if she first accepted the office job and was then upgraded. \(\beta_S\) is the performance difference between the two types of workers doing the same job. These results¹⁰10Table V seem to suggest that women who would have worked at the office perform similarly to women who are drawn to work from home.

I like that the paper does not take a string stance of beliefs vs. preferences. This paper is also one of the few experimental papers that show how work changes gender attitudes. While the author's are humble in telling us that they cannot disentangle the gateway effect from trust with the employer, but I would suppose that they would move together in a general setting: if trust drives the effect, a home job from one employer may not lead women to less flexible jobs at a different employer. It would be interesting to see if the effect that we see in the paper concerning attitudes and a gateway effect persists in the long-run with job offers that are not "manufactured". But this is all minor, I thought this was an excellent paper and I find it very impressive how one RCT is able to answer so many questions. Most other papers I have seen end up having a lot of problems because the arms were not as carefully thought out as in this case, with the five different job options, the surprise upgrade, and a second round with hand-labor jobs. Very cool and important paper!