BRINGING WORK HOME: FLEXIBLE WORK ARRANGEMENTS AS GATEWAY JOBS FOR WOMEN IN WEST BENGAL

BRINGING WORK HOME: FLEXIBLE WORK ARRANGEMENTS AS GATEWAY JOBS FOR WOMEN IN WEST BENGAL

Abstract

There is a large latent workforce in developing countries that consists of hundreds of millions of women who prefer to have paid work and yet are out of the labor force. Often, available job opportunities are incompatible with traditional gender roles that encourage women to stay at home. In a randomized experiment with one thousand six hundred seventy households, we partner with a jobs platform to offer short-term data work to women who are out of the labor force. We find three main results. First, flexible work-from-home jobs are highly effective at bringing women into paid work. Job flexibility more than triples take up from fifteen percent for an office job to forty-eight percent for a job that women can do from home while multitasking with childcare. Second, these jobs can act as a stepping stone to less flexible work. Trying paid work from home increases take up of less flexible jobs two to three months later by five percentage points. "Gateway jobs" are especially important for women from more traditional households: their labor supply is more likely to be marginal to flexibility, and in turn, work experience shifts their attitudes to become less traditional. Third, from the labor demand side, remote work comes with trade-offs in terms of worker performance, causing a four percent decrease in accuracy and a twenty percent decrease in speed. However, these performance drawbacks may be outweighed by the increase in available workers associated with remote work.

One Introduction

There is a large latent workforce in developing countries that consists of women who would prefer to work for pay and yet are out of the labor force. Representative surveys estimate this latent labor force numbers in the hundreds of millions of women, largely concentrated in South Asia, the Middle East, and North Africa. This latent workforce has high-stakes consequences for both the well-being of women and girls as well as aggregate efficiency. In addition to better respecting individual preferences, increasing the share of women in the labor force could improve outcomes such as agency, health, and educational attainment for women and girls and lead to a more efficient allocation of men's and women's labor that could spur economic development. This misallocation and, consequently, potential gains are particularly pronounced in countries such as India where levels of female employment are low despite advancements in women's education.

In these countries, social norms are thought to be a key barrier to women's paid work. One widespread social norm in our context is that a woman's place is in the home. In India, many women spend limited time outside the home after marriage: the median working-age married woman reports spending just zero point five hours per day outside her home, with forty-five percent reporting not leaving the home at all on an ordinary day. This norm is in conflict with the types of jobs which are available: fewer than twenty percent of jobs in India are fully remote. This mismatch between jobs that are available and jobs that women could do without violating norms of appropriate behavior for women suggests two potential strategies: (one) change norms so that existing jobs become more acceptable for women, or (two) change jobs to be more compatible with existing norms. In this project, we start by taking the second approach: change jobs so that women can do them while incurring lower norms-related costs. However, the two levers may be mutually reinforcing. If a woman's earned income and actual labor supply affect people's attitudes about women and work, then an approach that starts by changing jobs might also in turn change gender norms.

In a randomized field experiment in West Bengal, we test the effects of offering an emerging type of work - digital gig work - that is relatively compatible with existing norms of women's behavior due to the ability to work from home at flexible hours. Our experiment is designed to speak to three main research questions. First, would offering at-home, flexible work arrangements increase female labor force participation, and if so, which dimensions of flexibility are important? Second, given that many existing jobs require in-person attendance, can women who are initially only able to work from home use flexible jobs as a stepping stone to jobs with less flexible work arrangements? Third, to understand whether employers have an incentive to introduce remote working arrangements, what are the effects of work-from-home on job performance, both for inframarginal workers, and due to any change in the composition of workers drawn to the firm?

We randomly assign women from one thousand six hundred seventy lower-middle-income households to a treatment group that receives a job offer of month-long digital gig work or to a control group that receives no job offer. Among those receiving job offers, we introduce variation along three dimensions of job flexibility: (one) the ability to choose one's work hours each day, (two) the ability to combine work with childcare, and (three) the ability to work from home. All jobs are part-time, last for one month, and are offered in partnership with Karya, a smartphone-based data tasks platform in India. To separately estimate the effects of work-from-home on job performance versus worker composition,

after participants have decided to accept or reject their job offer, we randomly select half of the participants who accepted a less flexible job to be surprised with an upgrade to the most flexible job. After the jobs are completed, we estimate the effects of work experience on outcomes including women's gender attitudes and agency, as well as spillover effects on their children. Two to three months later, we measure subsequent take-up of different work opportunities (Jobs Round Two), including the effects of the initial at-home work experience on take up of less flexible jobs.

Most study participants are married women with little previous work experience. To focus on the extensive margin of labor force participation, women are only eligible for the study if they are not currently in the labor force or enrolled in skills training. During study recruitment, to avoid selecting the sample based on interest in finding paid work, potential participants are not told that the baseline survey could lead to a job opportunity. This leads to a sample where sixty-nine percent have never worked for pay prior to the study. However, to ensure women have the skills necessary to do the job if assigned to it, they must be literate in Hindi or Bangla and have access to an Android smartphone. On average, participants are thirty years old and nearly all, ninety-three percent, are married. Husbands and parents-in-law play a large role in whether or not women work: only thirty-six percent of women report having the final say in their own labor supply decisions. Two-fifths of participants live with at least one of their in-laws, and forty-eight percent have a child under the age of eight.

We find three main sets of results. First, flexible work arrangements more than triple women's job take up. Compared to a job which requires working from an office, the most flexible job we offer - which includes the ability to choose work hours flexibly, combine work with childcare, and work from home - dramatically increases job take up from fifteen percent to forty-eight percent. This thirty-three percentage point effect on women's job take up is larger than the effects of previously-tested interventions to increase women's labor supply. For example, a promotional video shown to women's family members in rural Uttar Pradesh increased job take up by seventy-eight percent, and correcting Saudi men's second-order beliefs about women working outside the home increased job sign up rates by thirty-six percent. The effect is also large compared to women's wage elasticity of labor supply; in an experiment in Mumbai, we found that increasing women's wage rates by nearly five times only resulted in one-quarter of the increase achieved in women's job take up by bundled job flexibility.

To shed light on the mechanisms driving the effects of flexible work arrangements on women's labor supply, we randomly vary dimensions of flexibility across job offers. This allows us to separately estimate the contributions of choosing work hours, combining work with childcare, and working from home to the overall effect of flexibility on job take up. The ability to work from home, even without time flexibility or the ability to combine work with childcare, doubles job take up from fourteen point six percent to twenty-nine point two percent. The ability to multitask work with childcare is also important, increasing job take up by approximately sixty percent, from twenty-nine point two percent to forty-five point six percent without time flexibility and from twenty-eight point six percent to forty-seven point nine percent with time flexibility. The ability to choose work hours flexibly, however, does not make a significant difference to take up - it appears these out-of-labor-force women can, and choose to, set aside consistent hours for paid work. Therefore, decomposing the thirty-three percent effect of bundled flexibility, approximately one half comes from the ability to work from home and the other half from the ability to multitask work with childcare.

Second, we find that flexible work arrangements are not only effective at bringing out-of-labor-force women into paid work, but they can also act as a stepping stone to less flexible jobs. To test whether flexible work can act as a gateway job to less flexible work, we return to all study participants two to three months after the endline survey and offer them another randomly assigned job. The offers in Jobs Round Two vary in flexibility along the same dimensions as the initial jobs and also introduce variation in the type of work offered. While the original jobs all consisted of online gig work, jobs in the second round also include non-digital piece-rate work such as sewing masks or making jewelry to assess whether effects on interest in work operate through digital-specific mechanisms or apply to interest in paid work more broadly. Consistent with flexible paid work acting as a stepping stone from unpaid at-home production to less flexible paid work, women are five percent more likely to start the job randomly assigned to them during Jobs Round Two if they were first given the chance to experience a more flexible job during the initial intervention. Consistent with the initial work experience offering a low-cost opportunity to learn what it is like for women to earn income, the effect is concentrated on women who had no previous work experience before the study. Examining only women randomly assigned to an office job during the second round, those assigned to a gateway job sequence in which their first round job was more flexible are eight percent more likely to start work compared to the control group that received no job offer in the first round. The transition from unpaid home production as a full-time homemaker to working outside the home might be a large leap - both for a woman herself and for her family members - and our results suggest that short-term, flexible jobs can act as a bridge for women to take multiple, more manageable steps to outside-the-home work.

One mechanism consistent with this gateway effect is a mutually reinforcing relationship between pro-work gender attitudes and actual labor supply, in which work experience shifts attitudes about appropriate behavior for women, thus expanding the set of jobs within reach. Women with more traditional attitudes at baseline had a stronger labor supply response to job flexibility, and in turn were also more likely to shift their gender attitudes in response to work experience. The impact of flexibility on job take up is fifty percent higher in households where women's pre-intervention gender attitudes are more traditional than the median participant, even conditional on other characteristics such as education, age, religion, previous work experience, cohabitation with parents-in-law, and having a young child. Receiving a job offer in turn shifts women's gender attitudes to become less traditional by zero point zero five standard deviations on average, with the effect entirely concentrated on women who held more traditional pre-intervention attitudes.

In addition to the effects on the women participants themselves, there are also spillover effects on other family members' gender attitudes and participation in home production. When the intervention ends, our survey team asks children about their attitudes and their family members' behaviors during the last month. The gender attitudes of children over twelve shift to become zero point one standard deviations less traditional after their mother had a chance to do paid work, with no differential effects by child gender. Treatment group children are also marginally more likely to report that their fathers contribute to home production. They are nine percent more likely to say that their fathers helped at least once with cooking, cleaning, or childcare during the time period of the intervention, as compared to the forty-two percent of children in the control group who say their fathers never helped with these activities.

fects on other family members' gender attitudes and participation in home production.7 When the intervention ends, our survey team asks children about their attitudes and their family members' behaviors during the last month. The gender attitudes of children over twelve shift to become 0.1 SDs less traditional after their mother had a chance to do paid work <LATEX>\left( p = 0 . 0 3 \right) ,</LATEX> with no differ- ential effects by child gender. Treatment group children are also marginally more likely to report that their fathers contribute to home production. They are 9 pp more likely to say that their fathers helped at least once with cooking, cleaning, or childcare during the time period of the intervention <LATEX>\left( p = 0 . 0 5 \right) ,</LATEX> as compared to the 42% of children in the control group who say their fathers never helped with these activities.

Turning to the labor demand side, the effects of flexibility on women's job participation show that introducing work-from-home - even temporarily - might allow employers to dramatically increase their pool of potential workers. So why do firms not offer flexible work arrangements more often? Beyond fixed costs and feasibility, employers would likely want to understand how adopting home-based arrangements would affect worker performance. Introducing remote work could affect performance in two ways: by changing the performance of inframarginal workers, those who are willing to work in person but can now work from home, and by drawing a different type of worker into the firm once the job is advertised as a work-from-home job. We separately identify selection into and treatment effects of flexible work arrangements using randomly assigned surprise upgrades to participants' initial job offers. The main job performance outcomes we measure are accuracy and speed.

We find that work-from-home reduces worker performance, with the effect driven by treatment rather than selection. Workers who are willing to work from the office but are randomly assigned to work from home complete tasks twenty percent more slowly and four percent less accurately. These effects are present for both easier and more difficult tasks, but are more pronounced for tasks that require greater cognitive load. For these more difficult tasks, work-from-home causes an additional one point five times slowdown in speed and an additional six times decrease in accuracy. We do not find any significant differences on the selection margin, suggesting that work-from-home compliers - those whose job take up is marginal to the ability to work from home - are not systematically different in their performance from those who are willing to work from both home and office. In this context of piece-rate wages, the performance cost in speed is borne by the worker. However, this negative effect on performance could be exacerbated for a firm that pays according to time worked, or in any context in which productivity is more difficult to monitor and thus there are greater incentives to shirk.

Examining how work patterns differ between treatment arms, flow effects can explain the negative effect of work-from-home on productivity. Defining a work session as a period of uninterrupted work in which fewer than ten minutes passes between consecutive tasks, work-from-home causes workers to have work sessions that are twenty-five percent shorter. These fragmented work patterns are costly to performance because of flow effects: workers complete tasks more quickly, and more accurately, when they work for a longer stretch of time without pauses.

Our paper contributes to four literatures. An extensive body of work demonstrates that female workers value flexible work arrangements more highly than men do in high-income countries. These studies focus primarily on women who are in the labor force and ask to what degree compensating differentials can explain gender wage gaps. In many developing countries, however -- particularly in South Asia, the Middle East, and North Africa - the first-order issue is not the gender wage gap, but rather the gender gap in labor force participation. In our setting, labor market entry is the origin of the gender gap: by age thirty, only fourteen point five percent of women have ever undertaken paid work, and this accounts for ninety percent of women who will ever enter the labor market. A natural question for these countries, then, is to what extent women's preferences for flexibility explain low labor market entry of women. In this study, we focus on women who are not labor market participants and show that the availability of flexible work arrangements is the deciding factor in whether or not many women begin paid work. In addition, although women's greater preference for flexible work arrangements is well-documented in the literature, these preferences are taken as exogenous. In West Bengal, we show that this willingness-to-pay is malleable and endogenous to women's own labor supply: work experience and the resulting shift in gender attitudes can increase women's willingness to work in less flexible jobs.

Second, we contribute to a growing literature about the effects of flexible work arrangements on job performance. Many recent papers examine trends in and impacts of post-pandemic increases in remote and hybrid work. Consistent with most of the literature examining the effects of fully remote work, we find that work-from-home lowers average performance. The mechanisms proposed for the reduction in performance are often related to communication, but we show that there may be negative impacts on performance even when workers' tasks are entirely individual. Our project differs from most previous work in our focus on effects on worker performance due to changes in the extensive margin of labor force participation-who is newly brought into paid work as a result of work-from-home jobs-as well as our focus on secondary-school-educated women in a lower-middle-income country. While other studies largely focus on incumbent workers who were willing to work in person and now choose flexible work arrangements, we aim to characterize the workers who would be brought into the labor force by an increase in flexible work arrangements in India.

Third, we add to a literature studying the effects of economic behavior on gender norms. Observational studies show that different economic conditions-such as suitability to plough versus hoe agriculture-give rise to different gendered divisions of labor, and that the resulting economic practices have an effect on gender norms in the long run. However, few experimental interventions have changed gender attitudes about women and work. There is little prior evidence that norms change when women start work, with the exception of Field et al., who find that getting access to direct deposit and training increases women's labor supply and liberalizes women's own and perceived norms. In this project, we test experimentally if changing economic conditions to make them more favorable to women's employment, by increasing the flexibility of available jobs, changes gender attitudes as well as divisions of labor in the home to become more supportive of women working for pay. Our results are promising in that women's entry into paid work appears to kickstart a multiplier effect for female employment, in which women working and less traditional gender roles mutually reinforce each other.

Finally, we contribute to an active literature on barriers to and enablers of increased female labor force participation in low- and middle-income countries. This literature includes factors such as perceived and actual gender norms, internal psychological constraints, intrahousehold bargaining power, safety and mobility, early childbearing, as well as employer and client discrimination. In addition, a small but growing set of papers explore how digital technology might interact with these barriers. The review paper shows that many women who are out of the labor force say that they are interested in working, and that there is a mismatch between the types of jobs available and women's job preferences. Our study shows that one important mismatch is the desire for flexible work arrangements, particularly the ability to work from home or multitask work with childcare.

The rest of the paper proceeds as follows: Section Two describes the study population and experimental design. Section Three presents the first set of labor supply results, which include the effects of flexible work arrangements on the extensive margin of labor force participation. Section Four presents the second set of labor supply results, which examines the effects of short-term work experience on take up of future work. This section also explores mechanisms for the effect on future labor supply, as well as impacts of women's work experience on the household more generally. Section Five turns to the labor demand side and presents the impacts of introducing home-based arrangements on worker performance, separating effects that operate through treatment versus worker selection. Section Six concludes.