GENDER DIFFERENCES IN PREFERENCES FOR NON-PECUNIARY BENEFITS IN THE LABOR MARKET: EXPERIMENTAL EVIDENCE FROM AN ONLINE FREELANCING PLATFORM

We conduct an experiment on a major international online freelancing labor market platform to study the impact of greater flexibility in choosing work hours within a day on female participation. We post identical job advertisements (for 320 jobs) covering a wide range of tasks (80 distinct tasks) that differ only in flexibility and the wage offered. Comparing the numbers of applicants for these jobs, we find that though both men and women prefer flexibility, the elasticity of response for women is twice that for the men. Flexible jobs receive 24 percent more female applications and 12 percent more male applications compared to inflexible jobs. Our findings have important implications for explaining gender differences in labor market outcomes and for firms interested in attracting more women employees.


Introduction
Do women value non-pecuniary job attributes more than men?The answer to this question has important implications.First, the value men and women attach to various nonpecuniary benefits could explain part of the gender wage gap (Petrongolo 2019).Women may have higher valuations for non-pecuniary benefits like flexibility, because of social norms around who should shoulder the responsibility of household work.Flexibility might allow women to balance household and wage work (Sullivan 2019). 1 Men and women may sort into different jobs based on non-pecuniary benefits, and firms may make lower wage offers to employees that demand expensive non-pecuniary benefits (Penner et al. 2022).Second, the limited provision of these non-pecuniary benefits may cause women to stay out of the labor market.Female labor force participation in many developing countries, particularly in Asia, remains low even after accounting for the level of economic development.One explanation could be that frictions in the labor market often lead to limited provision of the non-pecuniary benefits that women prefer (Gupta 1993;Macpherson and Hirsch 1995;DeLeire and Levy 2004;Grazier and Sloane 2008;Kleinjans 2009;Borker 2018).Third, the increase in the use of Internet technology, the rise of the gig economy, and the changes because of COVID-19 have led to an increased provision of flexibility, opening up new debates around flexible working arrangements.A key question is whether firms must provide flexible working arrangements to attract better employees or to retain existing ones.The answer depends on how strong the preference for flexible working arrangements is.Gender differences in these preferences will have implications for the composition and the diversity of the workforce.Thus, firms and policy-makers interested in the optimal response to these changes must therefore understand gender difference in preferences for these work arrangements (Cook et al. 2021;Gottlieb et al. 2021). 2   Despite the far-reaching implications, answering the question is empirically challeng-1 That said, a large literature has documented gender differences in various attributes, such as competitiveness, risk preference, and willingness to negotiate, that are relevant to wage determination (Croson and Gneezy 2009;Azmat and Petrongolo 2014;Exley and Kessler 2019).However, the exact sources of these differences are often unknown.These differences can result from social norms, like the difference in competitiveness between matrilineal and patrilineal societies, or have evolutionary roots.Gender differences in preference for flexibility could also be a product of social norms or have other roots.In this study, we remain agnostic about the sources of these differences.
2 Some studies argue that increased flexibility may increase gender disparities by reinforcing existing gender norms (Lott and Chung 2016;Chung 2019).Men might use it to work and earn more while women might be expected to contribute more to household work now that their work arrangements are flexible.
In this paper, we address these empirical challenges by using a randomized audit study that focuses on a specific non-pecuniary benefit: the flexibility in choosing work hours during the day.We conducted our experiment on a major online freelance labor market platform.We posted four otherwise identical job advertisements for each of 80 distinct tasks that vary only in their flexibility and the wage offered. 3Flexible jobs ("high-flexibility") allow the freelancer to choose any two-hour window during the day on a pre-specified date to complete the task.Inflexible jobs ("low-flexibility") require the work to be completed within a pre-specified two-hour period of our choosing on the pre-specified date.The jobs also differ in the wage offered: a "high-wage" job posting offers a lump-sum onetime payment of USD 40 and a "low-wage" job posting offers a lump-sum onetime payment of USD 30.Thus, we have 320 job postings for 80 distinct tasks. 4We collected information about the number of male and female applicants for each of the job postings, as well as several applicant-level characteristics.Since the job postings for each specific task varied only along the dimension of flexibility or wage offered, we can attribute any difference between male and female application responses to a difference in the value attached to these dimensions of pecuniary and non-pecuniary benefits.
We believe that the context of online freelance labor markets is particularly relevant for answering this question.First, the online freelance market generates sizable levels of employment.Estimates suggest that there are 14 million active online workers.A substantial amount of the recent growth has come from developing countries of South Asia (Stephany et al. 2021). 5Second, online labor markets are likely to become more important in the near future.Firms have made investments in adapting to remote working during the pandemic.These investments may have created new knowledge (possibly in management skills) in dealing with online remote working.The fixed nature of these investments along with the new knowledge is likely to create incentives for firms to work in an online remote environment particularly by hiring online freelancers (Umar et al. 2021). 6Third, despite the recent growth, the participation of women in online labor markets continues to lag behind.Data from the Online Labor Observatory shows that only 39 percent of the workers are female (Stephany et al. 2021).In addition, there are significant differences across countries and occupations.In the US, 41 percent of workers are females while only 28 percent of all online workers in India are females.
The results from the experiment suggest a gender difference in the preference for flexibility.Flexible jobs attract a higher number of applications from both men and women.However, compared to inflexible jobs, flexible jobs lead to a 24 percent rise in the number of female applicants as opposed to a 12 percent rise in the number of male applicants.Thus, compared to men, a larger proportion of women (of the workers in the platform) find flexibility a binding constraint.Flexibility also make the applicant pool more gender diverse, leading to a 2 percent rise in the proportion of female applicants.Women are also more likely to put effort into getting a flexible job.Compared to inflexible jobs, women are more likely to make an application before men and include their previous work sample in the application for a flexible job.Our results also suggests that the valuation of flexibility is sufficiently high: an increase in the wage by only 10 USD will not attract the same set of workers that value flexibility.
We contribute to the literature on gender and non-pecuniary benefits.A large literature has highlighted the importance of non-pecuniary benefits, particularly for women (Goldin and Katz 2011;Flabbi and Moro 2012;Goldin 2014;Sullivan and To 2014;Bronson 2014;Lavetti and Schmutte 2016;Sorkin 2018).However, most papers face the key challenge to empirically disentangle the role of preferences from other unobserved characteristics of the worker, firm and job level7 .In addition, many papers in the literature face the data challenge of identifying the role of a specific non-pecuniary benefit.In this paper, we overcome these challenges by using an experiment that allows causally identifying the role of preferences and at the same time we focus on a specific non-pecuniary benefit.
The more recent literature has used experiments that elicit stated preferences (and willingness to pay) for various job characteristics (Wiswall and Zafar 2018;Maestas et al. 2018;Mas and Pallais 2017), broadly finding that women have a higher willingness to pay for non-pecuniary job benefits.Wiswall and Zafar (2018) uses a sample of students from a top US university and finds that women are willing to give up a higher salary for job stability and job flexibility.Maestas et al. (2018) uses the American Working Conditions Survey and finds that women have a higher preference for jobs with less physical work and more paid leave.These papers validate the stated preferences by looking at real job attributes, so as to check that the stated preferences of the respondents match the actual job characteristics.Even if the stated preferences match the real job attributes, we do not observe the set of jobs from which the respondents are choosing in the real labor market.Thus, at least partially, the concern remains that the stated preferences are not incentive compatible.Our experiment adds to this by focusing on the revealed preferences of workers for flexibility.In this, our paper is most closely to related to He et al. (2021).They conduct a field experiment using a Chinese job board and find that married females have a stronger preference for flexible jobs than do married males.We add to the findings of He et al. (2021) by focusing on the worldwide online freelance labor market and on applications for a range of 80 distinct job types that vary across several dimensions, including being male or female dominated.
Our paper also contributes to a large literature that investigates differences in preferences between men and women, particularly their implications for the labor market (Croson and Gneezy 2009;Azmat and Petrongolo 2014;Exley and Kessler 2019).Broadly, the literature documents, using both field and lab experiments, that there are significant gender differences in various attributes, such as risk preferences and competitiveness, that have an impact on labor market outcomes.We add to that literature by documenting gender difference in preferences for flexibility in jobs.Lastly, our paper also adds to a recent and growing multidisciplinary literature that focuses on various aspects of the gig economy and the online freelance labor market (Stanton andThomas 2016, 2020;Cook et al. 2021;Stanton and Thomas 2021).In general, this literature notes that there is only limited data about online freelance workers.We add to this literature by collecting a rich set of data about applicants and their applications.In addition, we also focus on the role that flexibility may play in limiting the participation of women in online labor markets.8

Conceptual Framework
We begin with a simple conceptual framework to help interpret the results from the experiment.Assume that there are n two-hour time slots during the day during which a freelancer can complete the task we advertise.In our inflexible job ads, we specify the two-hour slot in which the hired freelancer must work.In the flexible jobs, the applicants can choose to work during any two-hour window during the day.Let us denote the set of possible time slots by S = (1, 2, 3, ...., n).
Workers have an opportunity cost of working during these time slots.Such an opportunity cost captures the pecuniary costs of working, such as forgone wages from alternative occupations, and non-pecuniary costs, such as delays in childcare or other family obligations.There is no uncertainty about the potential realization of these opportunity costs.Workers can fully and correctly predict these opportunity costs.We index workers by i ∈ I, where I is the universe of freelancers on the platform who see our advertisement.Let us denote the opportunity cost of working during time slot s ∈ S for worker i by c is .
For simplicity, we assume that the application costs are zero (or minimal) and workers apply to all jobs that they will take if offered.This is not an unrealistic assumption in our context.The workers usually add minor details (like a short cover letter) to their existing profile on the platform to make an application.There are also no interviews for these jobs.9Worker i will apply for an inflexible job offering a wage w to be done during time slot s if w − c is > 0.
However, if the same job with a wage w allows the worker to choose their work hours s ∈ S, then a worker i will apply if where c is = min(c i1 , c i2 , c i3 , ...., c in ).
Now, let us assume that the distribution of c is across individuals has a probability density function f (c is ) and a cumulative distribution function F (c is ).Next, assume the distribution of c is is given by the probability density function g(c is ) and a cumulative distribution function G(c is ).For a job that offers a wage wb but no flexibility in choosing work hours, the share of all applicants applying for the job will be given by: Similarly, for flexible jobs with a wage w, the share of all applicants who will apply for the job will be given by: Based on our findings from Tables 2 and 3, we have where w L = 30 and w H = 40 in our experiment.This implies that there must be at least one individual i such that c is < c is ≤ w.
Or, F (.) first-order stochastically dominates G(.).The estimated effect of flexibility in Table 2 is proportional to G(w) − F (w).In other words, the coefficient of 5.99 is proportional to the share of all applicants for whom c is < c is .The higher (lower) the number of applicants with c is < c is , the higher (lower) will be the estimated effect of flexibility.
Next, let us differentiate the distribution of c is and c is for males and females.For males, let us denote the cumulative distribution functions by F M (c is ) and G M (c is ).For females, we denote them by F F (c is ) and G F (c is ).To construct a mapping that will help us compare the effects of flexibility across the two genders, let us assume F M (c is ) = F F (c is ).That is, the distribution of minimum opportunity cost for the two genders is the same.10A larger effect of flexibility (in percentage terms) on women, as we observe in Tables 2 and 3, implies: In other words, our findings of a higher percentage effect of flexibility on females than males imply c is < c is ≤ w is true for a larger share of female applicants than male applicants.This means that the opportunity cost of working during the 8 to 10 am slot is, on average, higher for females than for males.

Experimental Design and Data Collection
We conducted our experiment on one of the largest online freelance labor market platform, which attracts clients and freelancers from around the world.The process of matching a freelancer with a client starts with a client posting a description of their job and a wage that they will pay a freelancer to complete it.The client may invite specific freelancers to apply for the job or post the job for any freelancer who may be interested.Candidates apply with a cover letter, their proposed wage (a counteroffer), and other details, such as past experience with similar work, that may indicate their competence and interest in the job.The client can then choose one or more freelancers to perform the task.Next, the client sends the chosen freelancer a contract specifying the agreed number of hours, a fixed or an hourly wage, and a deadline for the work to be completed by.At this stage, the chosen freelancer can accept the contract, renegotiate with the client, or reject the offer.
Our experiment entails posting several jobs on this platform as clients and studying the responses we receive from the freelancers.Specifically, we post four variations ('jobs') of 80 distinct tasks, which cover a wide range of activities.Our job advertisements resemble the job advertisements typically posted on the platform.With four variations for each of the 80 tasks, the experiment consists of 320 job postings.The jobs vary in terms of the wage offered and the flexibility they provide in choosing work hours.A "high-flexibility" (or just "flexible") job allowed the applicant to choose any two-hour window during the day on a pre-specified date to complete the task.A "low-flexibility" (or just "inflexible") job required the applicant to start the job at a specified time (8 AM in their local time) on a pre-specified date and finish it within two hours."High-wage" jobs offered 40 USD for the two hours of work while "low-wage" jobs offered 30 USD.Thus, the four types of jobs were 1) Low-wage, low-flexibility, 2) High-wage, low-flexibility 3) Low-wage, high-flexibility 4) High-wage, high-flexibility.
It was important to ensure the freelancers understood that they could not work outside the specified two-hour work window in the case of the low-flexibility jobs or outside the chosen two-hour window in the case of the high-flexibility jobs.We took several steps to make sure that applicants understood these requirements before applying.First, the job postings contained information such as the skills required and the expected time it might take to complete the job, but did not reveal any details that would have allowed the applicants to work on the job in advance.The job postings specified that the details required to finish the job would be shared at the start of the specified or chosen two-hour window.Second, for each job posting, we added a screening question that requires the applicant to respond with the specified two-hour window (in the case of a low flexible job) or enter their chosen two-hour window (in the case of a high flexible job) before they could start the application.This made the requirements regarding the work hours more salient.It is important to note that both the flexible and the inflexible job postings required the task to be completed in two hours.Thus, all four job variants for a task required the same skill set and the same amount of time commitment.The only differences were the flexibility in choosing the work hours or the wage.
We used five different accounts for posting and hiring freelancers for the 320 jobs.We randomly allocated each of the 80 tasks to one of the five accounts and to one of the days of the week.All four jobs for a task were posted from the same account on the same day of the week and, as much as possible, at the same time of the day, but in different weeks.This was in an attempt to keep other observed and unobserved factors the same across job postings within a task.All job postings were kept open for 24 hours.Once a job posting was closed, we randomly hired an applicant to complete the jobs and paid them the promised wage.The order of posting of the four jobs within a task was random for each task.The title, the skills required, and other attributes were kept the same across the four job postings within a task.Table A2 provides an example.All the jobs required the job to be completed two days after the posting.For example, a high-flexibility job posted on Monday required the applicant to complete the job on Wednesday at a chosen time of their convenience.The jobs were posted on all days of the week for four weeks between November, 2021 to December, 2021.
The data we use for our analysis are the number of applications and information from the applicant profiles and applications.Applicants do not state their gender on their profile or the application. 11We infer the gender of the applicant from the profile picture used in the profile.The platform verifies the identity of the freelancer against identity documents, like a passport, driver's license, or national ID, to ensure that the money goes into the correct freelancer account and no freelancer can operate more than one account on the platform.The platform withholds payments until the name and photograph of the freelancer on the platform matches their identity documents.This makes the pictures a reliable source of information.We manually classify applicants as male, female, or gender uncertain, using their profile pictures. 124 Empirical Specification Our experiment was pre-registered with the American Economic Association registry.The primary aim of our empirical exercise was to understand the causal effect of flexibility in choosing work hours on the number of applications.For this, we estimated the following specification: where Y j is one of the following dependent variables of interest for job posting j: the number of all applicants, of male applicants, of female applicants, and the share of female applicants.F lexible j takes a value of '1' if the job posting allows the freelancers to choose their work hours, '0' otherwise.X j denotes task fixed effects.ϵ s j is the error term.
The main coefficient of interest is β.Since, for every flexible job posting, we also have an otherwise identical job posting that only differs in the flexibility of choosing the work hours, β captures the causal effect of flexibility on the labor supply.Because of our interest in understanding the gender difference in demand for flexibility, we compare the estimates of β across male and female applicants.A higher β (as a percentage of the average number of male or female applicants) will indicate a higher elasticity of labor supply in response to flexibility.
To compare the marginal effects of flexibility between high and low wages and the trade-off between wages and flexibility, we estimated the causal effects of each type of job classified applicants for all jobs (flexible, inflexible, high wage, low wage) for each task, any person-specific bias is likely to affect both flexible and inflexible jobs in the same manner.We could have used an algorithm to infer the gender from the names of the applicants (Blevins and Mullen 2015).Though algorithms that predict gender from names work well for Western countries such as the US and the UK, they are not as accurate for predicting gender from names of applicants from a range of countries as wide as the one that we observed in our experiment.In addition, the accuracy of such algorithms depends on the size of the sample they are trained on.Since we had a manageable number of applicants, we believe that manual classification is less prone to error than other methods.However, one possibility is that there may be an unconscious bias on our part in inferring gender.We had two external research assistants reclassify the applicants for 72 jobs, chosen randomly, as male, female, and gender unclear.Of the 2, 824 applicants they categorized, only 45 applicants (1.6%) were classified as having a gender different from what was initially entered.Moreover, this mismatch was not different for inflexible jobs and flexible jobs. posting.
where Y j is one of the following dependent variables of interest for job posting j: the total number of all applicants, of male applicants, of female applicants, the share of female applicants.HW LF j is an indicator variable that indicates that a job has a high wage but no greater flexibility than a low-wage-low-flexibility job.LW HF j respectively, HW HF j , indicate that the job has a low wage and a high degree of flexibility, respectively, a high wage and a low degree of flexibility.X j denotes task fixed effects.ϵ s j is the error term.Since we have two wage offers and both flexible and inflexible jobs for each of the wage offers, we can compare the marginal effects of flexibility of at higher and lower wages.The marginal effect of flexibility at lower wages is given by β s 2 and the marginal effect of flexibility at higher wages is given by β s 3 − β s 2 .We can also infer the willingness to trade off flexibility and wage.To do this, we will need to compare the response to an increase in wage (β s 1 ) with the response to the provision of flexibility (β s 2 ).

Results
Table 1 presents a summary of the characteristics of the applicants to our job postings.As the results show, women make up only one-third of all applicants.This is despite the fact that our job postings covered a wide range of tasks (80 distinct tasks) that include both female-dominated tasks, such as translation and proofreading, and male-dominated tasks, such as financial consulting and coding.The Online Labour Observatory at the University of Oxford tracks projects across major online labor market platforms (including our platform) from across the world.Their estimates suggest that women make up 39 percent of the whole of the work force in online labor markets (Stephany et al. 2021).Further, we also compare the countrywise distribution of our applicants and the data from the Online Labour Observatory.We find that the distribution of country profiles in our data closely matches that from the Online Labour Observatory.Both these comparisons indicate that our sample from the experiment is representative of the gender composition and country profiles of online labor markets.Other takeaways are that female applicants (i) are less likely to make a counteroffer that is lower than the offered wage, (ii) write marginally longer cover letters and, (iii) are less experienced.Notes: ***, **, and * indicate that the difference in the means of a variable between the two groups, male and female applicants, is significant at 1%. 5%, and 10%, respectively.Position percentile is the application's chronological position, in percentile terms, among all applications for the job, with the first percentile indicating that it was the first application received for the job.A negative (positive) difference between the wage offered and the counteroffer made by an applicant implies that the freelancer's counteroffer was higher (lower) than the proposed wage.'Underbid' is an indicator variable that takes the value '1' when wage offered − counteroffer > 0, '0' otherwise.'Overbid' takes the value '1' when wage offered − counteroffer < 0, '0' otherwise.
Does flexibility lead to more job applications?Table 2 reports the findings for our primary outcome of interest, the number of applications.As Column 1 shows, jobs that offer flexibility attract more applications compared to a job with no flexibility.On average, flexible jobs received 5.98 more applicants than inflexible jobs.Comparing this effect with the average number of applications per job, this is about a 15.8 percent increase in the number of applications.Notes: ***p < 0.01, **p < 0.05, *p < 0.1.Of the 320 jobs posted, one did not have any applicants.
A flexible job was one where the freelancers could choose any two-hour window during which they wanted to work on the pre-specified date.The omitted category comprises inflexible jobs that required freelancers to complete the task at a designated time (8 am to 10 am) on a pre-specified date.
Is the effect of flexibility different across gender?Columns 2 and 3 of Table 2 present the effect of an increase in flexibility on the number of male and female applications, respectively.Compared to inflexible jobs, flexible jobs attract 2.92 more male applicants and 3.01 more female applicants.While the estimated magnitudes of the effect are similar for males and females, the percentage change with respect to the mean is significantly larger for females.Only one-third of all applicants are women.Compared to the average number of female applicants, an increase of three applicants translates to a 24 percent rise in the number of female applicants.For males, it translates to a 12 percent increase.Thus, of the pool of workers on the platform, a larger proportion of women respond to flexibility than men.One can interpret this as the elasticity of the labor supply with respect to flexibility being twice as high for women than for men.Our results complement the recent findings from a study of the gender wage gap in online labor markets by Adams-Prassl (2020).The study finds that women in online freelance labor markets earn less, as they need schedule flexibility (taking breaks between tasks) because of childcare responsibilities.Using experimental evidence, our results add to that by showing that women are more likely to select into jobs that allow such schedule flexibility.Does flexibility in jobs lead to a more gender-diverse work force?Since women have higher elasticity with respect to flexibility than men, flexible jobs can lead to a more gender-diverse workforce.Our results from Table 2 and Table 3 suggest that this is indeed the case.Column 4 of Table 2 suggests that flexible jobs lead to a 1.5 percentage point rise in the share of female applicants, amounting to a 5 percent improvement over the average share of women applicants. 13Column 4 of Table 3 reports similar results.High-flexibility jobs lead to a 3 percentage points rise in the proportion of female applicants (at a lower wage), a 10.4 percent rise over the average share of women applicants.Flexible jobs increase the gender diversity of the application pool.These results have implications for employers and policymakers interested in improving gender diversity in the online labor market.Using changes in the maximum work limit in medical residencies in the US, Wasserman (2019) shows that, relative to men, the participation of women increases in sub-fields that limit the maximum work time.Our results complement this finding by providing experimental evidence, from a wide range of fields, that schedule flexibility narrows the gender gap in participation.
How does the effect of flexibility compare at higher and lower wages?Table 3 presents the findings.Compared to men, the effects of flexibility are higher for women at all wages.Next, the effects of flexibility for men are similar at lower and higher wages (β male 2 = 2.92 compared to β male 3 − β male 1 = 2.89).For women, the effect of flexibility is slightly higher at lower wages, but the difference between the sizes of the effects at the two wages is statistically insignificant (β f emale 3 − β f emale 1 = 2.38 at the higher wage and β f emale 2 = 3.63 at the lower wage).A related question is whether there are gender differences in willingness to trade off higher wage for flexibility.To answer this question, we compare the change in the number of applicants in response to higher flexibility as opposed to higher wage.The results in Table 3 show that both men and women have a similar willingness to trade off higher wages and flexibility.Providing flexibility has an effect on the number of male applicants similar to that of a 10 USD rise in the wage offered (β male 1 = 3.06, β male 2 = 2.92).
For women, similarly, providing flexibility also has an effect similar to that of a 10 USD rise in wages (β f emale 1 = 3.25, β f emale 2 = 3.63).In percentage terms, this translates into a 12.2 percent increase in the number of male applicants because of a ten-dollar increase in the wage as opposed to a 11.6 percent rise in response to flexibility.For women, a similar 10 USD rise in wages leads to a 26 percent rise in the number of female applications as opposed to a 29 percent rise in applications when offered flexibility.Thus, for both men and women, a 10 USD rise in wages attracts the same number of applicants as the provision of more flexibility.How do we make sense of these results?One possibility is that, for both men and women, there is sufficient heterogeneity in preference for pecuniary and non-pecuniary benefits, and applicants are reluctant to substitute one for the other.Some applicants might have strong preferences for flexibility.They might apply to both highand low-wage jobs as long as they are flexible.Thus, the marginal effects of flexibility could be the same at high and low wages.However, a similar number of applicants might apply only to high-wage jobs regardless of flexibility.In such a scenario, we will find the gender difference in the trade-off between higher wages and flexibility to be the same.Notes: ***p < 0.01, **p < 0.05, *p < 0.1.The outcome variable is the application's chronological position, in percentile terms, among all applications for the job, with the first percentile indicating that it was the first application received for the job.A flexible job was one where the freelancers could choose any two-hour window during which they wanted to work on the pre-specified date.The omitted category comprises inflexible jobs that required freelancers to complete the task at a designated time (8 am to 10 am) on a pre-specified date.The number of total applicants is more than the sum of male and female applicants because we could not deduce the gender of a few applicants from their profile pictures and names.All job-level observations are weighted by the total number of applicants for each job.
Do women put more effort into getting selected for these flexible jobs?While we do not have a direct measure of effort, we look at several indirect measures that indicate effort and willingness to get these jobs.First, we examine how quickly applicants apply to our job advertisement.For this, we rank all applicants by their position in the application queue.Since all job advertisements were open for applications for the same amount of time, we can compare the proportion of female applicants among "early" applicants across flexible and inflexible jobs.14Table 4 reports the effect of providing flexibility on the proportion of female applicants among "early" applicants.We find a higher proportion of women among the earliest 25 th and 50 th percentiles of applicants.Note that the effect at the 25 th percentile is significantly higher than the effect of flexibility on the overall share of female applicants (1.47) reported in Table 2.The changes in the share of women among the earliest 10 th and 75 th percentiles are statistically indistinguishable from the overall increase in the share of female applicants.This suggests that women are not only more responsive to flexible jobs on the extensive margin, but they also respond by applying more quickly than men.15Notes: ***p < 0.01, **p < 0.05, *p < 0.1.'Cover letter length' is the number of characters in an applicant's cover letter, including spaces.'Work sample provided' is an indicator variable that takes a value of '1' if the applicant attached at least one work sample with their application, '0' otherwise.
A flexible job was one where the freelancers could choose any two-hour window during which they wanted to work on the pre-specified date.The omitted category comprises inflexible jobs that required freelancers to complete the task at a designated time (8 am to 10 am) on a pre-specified date.The number of total applicants is more than the sum of male and female applicants because we could not deduce the gender of a few applicants from their profile pictures and names.
Second, we look at whether the applicant attached a previous work sample with their application to indicate their ability or expertise to complete the job, and the length of their cover letter written as a part of their application.Attaching a work sample takes effort and time and also indicates the willingness of the applicant to signal their quality to the employer.The length of the cover letter may also signal effort.The findings, reported in Table 5, show that compared to an inflexible job, men are no more likely to attach a work sample or write longer cover letters in response to a flexible job.Women, in comparison, are more likely to attach a work sample when applying for a flexible job, indicating an increased effort.Women also write shorter cover letters for applications to flexible jobs.The results seem to suggest that women put more effort into some dimension of the application.The effect on the length of the cover letter is not straightforward to interpret.Perhaps women partially offset the increased effort required for attaching samples by writing shorter cover letters.Or, they spend more time to make the letter more concise and precise.However, it is important to note that the results may also reflect differences in composition of applicants rather than their effort.It is possible that marginal applicants to flexible jobs are of higher quality and thus provide a better application package.Notes: ***p < 0.01, **p < 0.05, *p < 0.1.'Total prior contracts' is the number of contracts an applicant had completed on the platform by the time of their application for our advertised job.'Total prior contracted hours' and 'Total prior earnings,' similarly, capture the number of hours they had worked on job contracts and the earnings they had had through the platform.A flexible job was one where the freelancers could choose any two-hour window during which they wanted to work on the pre-specified date.The omitted category comprises inflexible jobs that required freelancers to complete the task at a designated time (8 am to 10 am) on a pre-specified date.The number of total applicants is more than the sum of male and female applicants because we could not deduce the gender of a few applicants from their profile pictures and names.
One proxy of quality is experience.Experienced women might prefer showcasing their are less likely to underbid.The results show that this translates into a 29 percent reduction over the average.One explanation for men underbidding more than women could be that male applicants may have a lower reservation wage, on average.Unfortunately, we do not observe the reservation wage.In the columns that follow, we control for other characteristics of the task, job, and applicant, as well as our crude measures of effort, to proxy for their reservation wage.As the results indicate, the association between gender and underbidding persists, suggesting that the gender difference in underbidding is potentially a reflection of a lower willingness to negotiate, even if that involves underbidding.
A large literature has documented that men are more likely to negotiate wage offers than women (see Hernandez-Arenaz and Iriberri (2019) for a review of the literature).For example, Leibbrandt and List (2015) finds that women are less willing to negotiate if the job postings do not explicitly mention the possibility of negotiating, a setting similar to ours.
Our results provide a new insight from online labor markets: men are also more likely to undercut wages to secure a job.The negotiation, therefore, can happen in either direction.
That said, we cannot comment on the reasons behind the gender differences in willingness to negotiate.This is distinct from the results in Table 7, which indicate that there is no difference in underbidding behavior in response to flexible jobs for either men or women.
A key question is why do women prefer flexibility?Our experiment does not allow us to directly provide an answer to this question.However, understanding the differences in effects across countries may provide an insight to this question.In Table 8, we study the difference in effect of flexibility between countries of high and low fertility rate.One reason why women might prefer flexibility could be that it allows them to manage timesensitive child care responsibilities.If so, the value of flexibility is likely to be higher for those women that have more children.We do not know the number of children our applicants have.However, a coarse proxy is the average fertility rate of the country of the applicant.In Panel A, we report the effect of allowing flexibility in choosing work hours separately for countries that have high fertility and low fertility.We do not find any significant difference in effect for women (relative to the mean) across countries with high and low fertility.The elasticity of response is the same for both high and low fertility countries.For men, the elasticity is slightly higher in low fertility countries.How do we explain these results?One possibility is that the time commitments for child care responsibilities are fixed costs and do not vary much by number of children.For example, the time commitment to cook food or to take a child to school could be a fixed cost and thus could be same regardless of the number of children.However, it is important to keep in mind such country level averages are coarse measures and our results can reflect the coarseness of the measure rather than any mechanism.Does lack of flexibility limit female labor force participation?Though our results cannot directly speak to this, a comparison of the effect of flexibility between countries that have low and high female labor force participation rates can provide some insights into that question.In Panel B of Table 8, we present the results separately for countries with high and low female labor force participation rates.The effect of flexibility (relative to the mean) is higher for women in low female labor force participation countries.The results for men do not differ across these countries.One possibility is that the demand for flexible jobs is higher in these countries and the limited availability of such jobs leads to low female labor force participation.When we post such jobs, we find a large response.That said, all of our applicants are existing workers on the platform and the reasons for their high response to flexibility may be different from the reasons that stop women from participating in the labor market at the extensive margin.important implications.In light of the ever evolving nature of the work place, it important for firms to know the demand for various non-pecuniary benefits.In addition, any potential gender difference in the valuation of these non-pecuniary benefits has important implications in explaining gender inequalities in the labor market.For example, lack of supply of non-pecuniary benefits like flexibility can be a potential reason that limits the participation of women in the labor market in developing countries.
However, we argued that despite the importance of this question, several empirical challenges makes answering this question difficult.Studies that use observational data cannot causally disentangle the preference for various non-pecuniary benefits from other worker-, job-, and firm-specific factors.In addition, several non-pecuniary benefits are offered at the same time, making it challenging to isolate the effect of one non-pecuniary benefit, such as flexibility in choosing the work time.Though studies that use stated preferences overcome these problems, stated preferences are often not incentive compatible.
In this paper, we overcome these challenges by using an audit experiment.We posted matched pairs of jobs on a major online freelance labor market platform, which only differed in the flexibility (of the work time) offered.Since these jobs are identical in all other attributes, except flexibility, any difference in the applications for these jobs are a result of a preference for flexibility.We find that flexible jobs attract more applications.Though flexible jobs attract a higher number of applications from both men and women, the effects are twice as large for women in percentage terms.Flexible jobs lead to a 24 percent rise in the number of female applicants and a 12 percent rise in the number of male applicants.Overall, the results suggests that indeed workers value flexibility and the demand is higher for women than for men.
That said, it is important to interpret our results in the context of the limitations of the nature of the experiment.First, though our results are internally valid, we cannot speak to how our results would hold in a general population that includes the brick and mortar labor market.The effects can go either way.It is possible that workers in freelance labor markets prefer flexibility more than workers in the brick and mortar labor market and we are overestimating the demand for flexibility.Yet, it is also possible that since freelance labor markets already offer so much flexibility, the marginal valuation for flexibility is lower in this market.Moreover, since the contracts in the freelance labor market are short term, workers may care less about flexibility.Yet, on the other hand, since the contracts are short term, the benefits from giving up flexibility are also less.Second, our results do not fully speak to the underlying reason behind the gender difference in the demand for flexibility.It is possible that women prefer flexibility because of an extra burden due to household work and any changes in the structure of intra-household bargaining will lead to a change in such preferences for flexibility.Though we explore some directions, our results on this question are inconclusive.Thus, we remain agnostic about the sources of gender differences in preferences for flexibility.Lastly, we are only referring to a specific type of flexibility, and the preferences for other types of flexibility may be quite different.For example, our experiment does not speak to the gender differences in the preference for work from home.
Our experiment and the results still are significant for policy makers interested in increasing female labor force participation.The preference for non-pecuniary benefits such as flexibility is a somewhat ignored aspect in the policy discussions that aim to increase female labor force participation.Mostly, these discussions focus on issues like the development of skills, access to finance, social norms and networks, education, and the organization of the family.But the limited availability of non-pecuniary benefits like flexibility is often an important barrier to participation in the labor market.A strong preference for non-pecuniary benefits like flexibility and a limited supply of these benefits in the labor market may explain low female labor force participation.
If the lack of flexibility is indeed an explanation, what can policy makers do?Assuming that firms are aware of these differences in preferences but find it costly to provide these non-pecuniary benefits, policy makers could provide incentives such as tax breaks or cheap credit to firms that provide benefits like flexible working hours.Further, innovations in technology that reduce search cost and promote the gig economy may open up possibilities for jobs that provide flexibility and thus encourage the participation of women in the labor market.Firms that are unaware of worker preferences may invest in providing more nonpecuniary benefits.Thus, we believe that our result contributes significantly to the policy discussions on encouraging female labor force participation.

Figure A2 :
Figure A2: Wage, flexibility, and country of origin of male applicants

Table 1 :
Summary Statistics

Table 2 :
The Impact of Flexibility on the Number of Applicants

Table 3 :
The Impact of Wage and Flexibility on the Number of Applicants Of the 320 jobs posted, one did not have any applicants.A high-wage, low-flexibility job offered a fixed wage of USD 40 and required the freelancers to complete the task at a designated time (8 am to 10 am) on a pre-specified date.A low-wage, high-flexibility job offered a wage of USD 30 but allowed the freelancers to choose any two-hour window during which they wanted to work on the pre-specified date.A high-wage, high-flexibility job offered USD 40 and allowed the freelancers to choose any two-hour window on the pre-specified date.

Table 4 :
The Impact of Flexibility on the Positions of the Freelancers' Applications

Table 5 :
Cover letter length and work samples

Table 6 :
Flexibility and applicant experience

Table 8 :
Heterogeneity in the impact of flexibility **p < 0.05, *p < 0.1.A flexible job was one where the freelancers could choose any two-hour window during which they wanted to work on the pre-specified date.The omitted category comprises inflexible jobs that required freelancers to complete the task at a designated time (8 am to 10 am) on a pre-specified date.Countries are grouped into high-and low-TFR/FLPF/GDP categories by splitting them at the mean value.

Table A1 :
Full List of Tasks posted on the platform