Policy Research Working Paper 10701 Gender-Specific Transportation Costs and Female Time Use Evidence from India’s Pink Slip Program Yutong Chen Kerem Cosar Devaki Ghose Shirish Mahendru Sheetal Sekhri Development Economics Development Research Group February 2024 Policy Research Working Paper 10701 Abstract This paper estimates a synthetic difference-in-differences time on household chores. Low-skilled married women specification on the roll-out of a program providing free increase time on household activities and reduce labor bus transit for women in several Indian states, to examine supply. Unemployed women increase job search with no the impact on women’s time allocation and labor supply. effect on employment. The findings show that gender roles Household expenditures on buses fall and women save within households undermine the effect of gender-specific time on travel. However, there is substantial heterogeneity. travel subsidies on female labor supply. Skilled employed women increase labor supply and reduce This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at dghose@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team G ENDER -S PECIFIC T RANSPORTATION C OSTS AND F EMALE T IME U SE : E VIDENCE FROM I NDIA’ S P INK S LIP P ROGRAM* Yutong Chen† sar‡ Kerem Co¸ Devaki Ghose§ Shirish Mahendru¶ Sheetal Sekhri‖ Keywords: Transport, Gender, Time-Use, Female Labor Force Participation JEL codes: J16, J22, R41 * We thank the UVA department of economics for financial support. For helpful comments and feedback, we thank S Anukriti and the Office of the Chief Economist, South Asia Region. † yc3jk@virginia.edu (University of Virginia) ‡ kerem.cosar@virginia.edu (University of Virginia, CEPR, CESifo, NBER) § dghose@worldbank.org (Development Economics Research Group, The World Bank). The views expressed in this paper do not represent the views of the World Bank or its partner organizations and solely represent the authors’ personal views. ¶ shirish.mahendru@giz.de (Gesellschaft für Internationale Zusammenarbeit (GIZ) India) ‖ ssekhri@virginia.edu (University of Virginia) 1 Introduction Women face significant commuting barriers not only in developing countries but also in the developed world. Social norms on household chores and traveling alone affect female labor supply through commuting decisions (Fanning Madden, 1981; Turner and Niemeier, 1997; Lee and McDonald, 2003; Abe, 2011; McQuaid and Chen, 2012). A recent OECD (2016) report finds that men in OECD countries have an average commuting time of 33.4 minutes per day, while women have an average of 21.9 minutes, resulting in a gender commuting gap of 31.1%. The patterns are even starker in developing countries. For example, according to data from the National Time-use Survey (2020), on average, Indian women spend only 8 minutes per day on employment-related travel, while men spend 36 minutes. ILO (2017) reports that lack of transportation decreases women’s probability of participating in the labor market by 16.5 percentage points among developing countries. In short, there is ample evidence that commuting barriers distort women’s labor supply. This naturally begets the question of whether reducing commuting barriers can increase women’s labor supply. In this paper, we first examine if women are responsive to the cost of transportation in determining their travel demand, or whether gender norms are so entrenched that demand is inelastic. Second, if women travel more frequently or opt for faster modes of transportation due to decreased commuting costs, how does this affect the allocation of women’s time among household chores, commuting, and labor supply? To this end, we exploit the rollout of a free busing scheme, the Pink Slip Program, in two states of India. India is a pertinent setting to study this question. There is an overwhelming gender gap in commuting time and modes used. The 2011 Census of India reveals that 30.2% of women travel to work on foot, and only 24.6% use any kind of transportation, highlighting the limited access to transportation options for many women.1 There is also evidence that women use slower modes of transportation to commute to work as faster modes are usually more expensive 1 In contrast, only 20% of men travel on foot and more than 50% of men use any kind of transportation. 1 (Anand and Tiwari, 2006). Concurrently, female labor force participation is low in India with a labor force participation rate of 24% in 2019, significantly lower than the average of 46% in low- and middle-income countries (World Bank, 2019).2 We leverage the Pink Slip programs’ state-wide roll-out in Punjab, and Tamil Nadu in April and May 2021, respectively. While Delhi also introduced this program in November 2019, data limitations preclude us from including it in our sample.3 Reports from the ground indicate that women’s response to the initiative was overwhelmingly positive. For example, from July 2021 to March 2022, the percentage of women travelling by bus in Tamil Nadu increased from 40% to 61% (Sundaram, 2022). According to Goswami (2021), women made up the majority of riders on Delhi Transport Corporation buses by March 2021. To identify the causal impact of the program on women’s labor outcomes, we collated data from a number of sources. Our main empirical analysis is based on the Consumer Pyramids Household Survey (CPHS) data maintained by the Centre for Monitoring the Indian Econ- omy (CMIE). The rich CPHS data is a panel of about 160,000 households across all Indian major states after 2014. It includes comprehensive information not only on household ex- penditures and members’ demographic characteristics and employment status but also on their time use patterns and allocation of time on various activities. We bolster the findings with a large primary survey of women conducted in Delhi. Our identification approach compares women in treated states (i.e., Punjab and Tamil Nadu) to those in their geographical neighbors which we will refer to as control states henceforth. The implementation of the policy in 2021 provides temporal variation. To address concerns about endogeneity, we implement a synthetic differences-in-differences strategy (SDID) pro- posed by Arkhangelsky et al. (2021) at the state level. This approach combines the synthetic 2 The female labor force participation rate has increased to 37% according to the most recent female labor force survey by the Ministry of Statistics and Program Implementation (Source: http://tinyurl.com/mu2vx7m3) 3 The CMIE, our data source in this paper, started to collect information on individual time usage after Delhi started to implement the Pink Slip scheme. So, we are unable to include Delhi in our sample directly. 2 control method (Abadie and Gardeazabal, 2003) with the difference-in-differences strategy to take advantage of the benefits of both. Event study models reveal parallel pre-trends. The weights from the SDID approach are used in all survey data-based analyses at the indi- vidual level. We find evidence consistent with elastic demand for transportation for women using the CPHS data. Overall expenditures on travel, specifically on public buses, trams, and ferries, were reduced for households with women in treated states compared to control states. Findings from the Delhi survey also corroborate these results at the individual level. New female users of buses report negligible transportation costs after the policy change as opposed to substantial transport costs prior to it using other modes. We find heterogeneous effects of the policy change on employed and unemployed women which are in opposite directions. Consequently, these cancel each other and the policy ap- 4 pears to have a null overall effect. We focus on time spent on household chores, traveling, and labor supply. Skilled employed women use the time saved from commuting to increase hours of labor supplied plausibly due to the substitution from slower modes of transporta- tion to faster free buses. An interesting finding of our analysis is that providing free buses to women has the most considerable effect on the time-use patterns of unemployed women, especially those who are not married. These women spend more time outside the house, and more time traveling, partially spending this time searching for jobs more intensively and farther away from home. However, we do not find an increase in the likelihood of them find- ing employment up to four months after the scheme, indicating that women face additional hurdles to finding employment in the short run. In sharp contrast to skilled employed women, low-skilled employed, especially married women, spend the time saved from commuting to substitute for household chores. In fact, married women with low skills reduced their hours of labor supplied. This reduction is consistent 4 We show evidence that the program does not lead to changes in the employment or marital status of women, as a result of which the employment and marital status are pre-determined relative to the policy change. Thus, heterogeneity along these margins is not confounded by changes in the margins themselves. 3 with our findings of intrahousehold substitution in time use and its allocation to activities. While low-skilled employed married women increased the time spent on household chores and reduced their labor supply, married men’s behavior changed in the opposite direction: employed married men reduced their time spent on household chores and increased their work hours. Unemployed married men increased the time spent searching for jobs. This reduced labor supply from low-skilled married women could be driven by a shift from men to women of household chores that require an indivisible, discrete amount of time. For exam- ple, a long commute by mothers may have previously made it optimal for the father to take their kids to school, perhaps on a bus. After the roll-out of free bus rides for women, this responsibility could be reallocated to the mother who could avail of a potentially faster mode of transport (bus) for free, allowing the husband’s work hours to increase at the expense of the mother’s time. The gender wage gap increases the likelihood of such re-optimization at the household level. In sum, while the free bus provision benefited skilled and unmarried women by improving their labor market outcomes, low-skilled married women responded by reducing their work hours and doing more household chores. Our paper complements a growing body of work studying the effects of barriers to women’s mobility and access to public transport systems on female labor force participation (Field and Vyborny, 2022; Martinez et al., 2020; Alam et al., 2021; Lei et al., 2019; ILO, 2017; Petrongolo and Ronchi, 2020). Farré et al. (2022) show that a 10-minute increase in com- muting decreases the likelihood of married women participating in the labor market by 4.6 percentage points. Using a job search model where commute matters, Le Barbanchon et al. (2021) estimated that approximately 10% of the wage gap between men and women in re-employment in France could be attributed to differences in the willingness to commute between genders. Black et al. (2014) show that metropolitan areas’ commuting times are one explanation for the large variation across US cities in married women’s labor force par- ticipation. In a closely related paper, Field and Vyborny (2022) show that women-only buses 4 increase female job search in Pakistan.5 We extend this literature by highlighting that de- mand for transportation is elastic for women but there is heterogeneity by skill and marital status. Free buses do incentivize skilled, unmarried and employed to increase their labor supply, and unemployed women to search more intensively for jobs. But intrahousehold gender norms seem to preclude unskilled employed married women from directly improving labor market outcomes. Surprisingly, low-skilled married women use the time saved from commuting using free buses to do more household work, replacing some of the household work previously done by their spouses, who in turn increase their work hours. Thus, in the presence of restrictive gender norms, reducing commuting costs alone may not be enough for women to increase work hours or increase participation in the labor markets. The second strand of literature we connect to analyzes the effects of reductions in commute times, often by providing transit subsidies in randomized control experiments, on job search and employment creation (Franklin, 2018; Abebe et al., 2016; Phillips, 2014; Moreno-Monroy and Posada, 2018). The general consensus in this literature is that reductions in commuting costs increase job search intensity and employment. We demonstrate that social norms in developing countries can undermine the effects of policies that reduce commuting costs for women. While unmarried skilled and unemployed women increase job search efforts, we do not detect an increase in employment possibly due to other forms of discrimination and disparities, consistent with restrictive gender norms discussed by Jayachandran (2021) and Dinkelman and Ngai (2022). Low-skill married women’s labor outcomes become worse if anything. Our findings have important policy implications: If the goal of providing free transporta- tion to women is to increase women’s labor force participation, only a small share of the 5 Dasgupta and Datta (2023) use a cross-sectional time-use survey and compare men and women across states to assess how Pink Slip scheme in Delhi affected women’s time use patterns. They document an increase of 30 to 50 minutes in the time women spent on work during the first two months after the introduction of the scheme. We find a null causal effect on time use in our panel estimation masked by heterogeneity by employment status. Borker et al. (2020) have an ongoing experiment where they want to compare the partial equilibrium results of free bus passes to the general equilibrium effects of Delhi’s policy. 5 workforce, primarily skilled employed women, benefit from this program in the immediate short-run. The rest of the paper is organized as follows. Section 2 describes the study setting. Section 3 describes the datasets. Section 4 outlines the empirical methodology to estimate the im- pacts of reductions in commuting costs. Section 5 presents the results. Section 6 provides concluding remarks. 2 Background In India, limited infrastructure and transport services restrict mobility for both men and women, but women frequently experience extra socio-cultural and economic factors that negatively affect their commute patterns (Srinivasan and Rogers, 2005; Tripathi et al., 2017; Alam et al., 2021). Given the sizeable gender wage gap (Duraisamy and Duraisamy, 2016; Deshpande et al., 2018) and the additional barriers in accessing the financial system com- pared to men (Khera, 2018), paid access to transportation is plausibly harder for Indian women. Besides, the low rate of female usage of public transport might raise a perception problem since more female presence in public transportation makes women feel safer (Saj- jad et al., 2017). In a survey of 3,800 students at Delhi University, Borker (2021) found that women are willing to travel 27 minutes more per day or 40% more than their daily travel time if they can use a perceived safer transport route. These factors put women at a disadvantage regarding access to transport services and infrastructure (Astrop et al., 1996; Dominguez Gonzalez et al., 2020), potentially affecting their participation in labor markets (Patacchini and Zenou, 2005; ILO, 2017; Sajjad et al., 2017; Martinez et al., 2020). In light of these challenges, the Delhi government introduced a scheme offering free bus rides to all women in the city from November 2019 onward (Kejriwal, 2019). The scheme makes bus travel free for women in all Delhi Transport Corporation (DTC) and Cluster buses. On each ride, bus operators provide a pink ticket to each woman. Afterward, Delhi’s government 6 compensates the bus operators with |10—equivalent of $0.14, all currency conversions use the 11/2019 exchange rate—per pink ticket ride (The Economic Times, 2019; Durai, 2021). The program showed an early response: just 20 days after the scheme’s launch, female daily ridership in DTC and cluster buses increased from 33% to 44% (Sengar, 2019). Spurred by the good reception of the initiative in Delhi, on April 1 and May 7, 2021, the states of Punjab and Tamil Nadu, respectively, implemented free bus ride schemes for women in their states, allowing free travel in government-owned public buses.6 From July 2021 to March 2022, the percentage of women commuting by bus in Tamil Nadu increased from 40% to 61% (Sundaram, 2022). This increment led the Government to increase the budget of the program from |12 million ($168 mn), allocated in the first year, to |15.2 million ($212.8 mn) in the second year (Durai, 2021; Sundaram, 2022). 3 Data 3.1 Main Data: Consumer Pyramids Household Survey Our main source of data is the Consumer Pyramids Household Survey (CPHS). It is a household-level longitudinal survey conducted by the Centre for Monitoring Indian Econ- omy (CMIE). Starting with the first wave in January-April 2014, the CMIE runs surveys three rounds a year (January-April, May-August, and September-December). Each wave covers about 160,000 households from all major Indian states, maintaining a consistently high household response rate of over 80%. A multi-stage stratified survey design is de- ployed. The broadest level of stratification is a homogeneous region (HR), which is defined as a set of neighboring districts within a state that is comparable in the following charac- teristics: climate, urbanization, female literacy rate, and population. In Appendix Table C1, 6 In Punjab, these include PEPSU Road Transport Corporation (PRTC), PUNBUS, Punjab Roadways Buses, and City Bus Services, but did not include AC buses, Volvo Buses, and HVAC Buses (Express, 2021) In Tamil Nadu, the free ride scheme includes tickets for the Tamil Nadu State Transport Corporation (TNSTC) ordinary city buses. 7 we list the two treated states of Punjab and Tamil Nadu—which implemented free bus ride schemes for women—and their control states, i.e., the states that are adjacent to the treated states and did not distribute free bus tickets to women.7 Since the CPHS data is represen- tative at the level of HRs, our analysis only includes neighboring HRs in the control states, i.e., we exclude HRs in control states that are not adjacent to treatment states. Appendix ?? displays the map of treated and control HRs. In total, we have 20 HRs in two treatment and seven control states. The CPHS has four sections: Consumption Pyramids (CP), People of India (PoI), Aspiration India (AsI), and Income Pyramids (InP). In this study, we use CP, InP, and PoI data. The CP is a household-level monthly survey reporting household expenditures on various kinds of goods and services, without a breakdown of individual members of multi-person households. Specifically, it asks households about their monthly expenditure on all kinds of transport including expenses on a combined category of “buses, trains, and ferries” (BTF). Our study period for the CP is from November 2020 to September 2021. The InP is a monthly survey that tracks the income of each household member. We use the same study period as in the CP data. The PoI is an individual-level survey conducted every four months. There are three waves in a year: January-April, May-August, and September-December. The PoI data has information on one’s employment status, time usage, and demographic characteristics like gender, education level, and marital status. From the PoI data, we know how much time a person spends on household activities, at work, and traveling. Reported time on travel is the time spent by a person traveling from one place to another for all kinds of purposes including work-related activities. The CP survey does not ask specific questions about time spent commuting to work, searching for a job, or on leisure. We use six waves of PoI from May-August 2020 to January-April 2022 (or from the 20th 7 The CMIE started to collect household members’ time usage information in the wave of September- December 2019, while the government of Delhi started the pink slip scheme in November 2019. We do not include Delhi in the analysis since we do not have pre-period information on time usage there. 8 wave to the 25th wave). We also match the households that appeared in the PoI data to households in the CP data. Appendix Table C2 lists the study periods for the two sectional data sets. We restrict our sample to women (or households having women) aged between 15 and 65 at their first appearance in the data. Appendix Table C3 lists the variables used in the analysis and their definitions. Appendix Table C4 displays the summary statistics of our study sample. Panel A displays the household characteristics in December 2020. Differences between households in treated vs. control HRs in terms of rural residence, number of people, and per-capita income and expenditures are relatively small. In panel B, we compare women in treated HRs to those in control HRs in May-August 2020. The distributions of age, marital status, and education are comparable for the two groups of women. Women in treated areas are less likely to participate in the labor market but conditional on participation, they are more likely to be employed. They also tend to spend more time on household activities and work but less time on travel than women in control areas. 3.2 Delhi Primary Survey The CMIE started to collect household members’ time usage information in the wave of September-December 2019, while the government of Delhi started the Pink Slip scheme in November 2019. Since we do not have pre-period information on time usage in Delhi, we do not include it in the analysis. To complement and bolster our baseline analysis, however, we use primary data collected via a survey in Delhi in February 2020 by the Gesellschaft für Internationale Zusammenarbeit (GIZ) India (Mahendru, 2022). The survey collected data from 1,525 female bus users (1,294 continuous users and 231 new users) and 500 nonusers.8 The sample is randomly selected at major attractions and generation points across Delhi (Appendix Figure B1).9 To construct a comparable sample of nonusers, new 8 Continuous users were women who took buses both before and after the implementation of the Pink Slip Program. New users were women who began using buses after the program’s implementation. Nonusers were women who did not use buses before or after the program. 9 The generation points are all major locations within the city where trips either originate or are attracted to. These include major work areas, shopping districts, major schools, and others. 9 users, and continuous users, we employ propensity score matching. We match women on the following variables: age, occupation,10 total average monthly household income, total average monthly household expenditure on travel, ownership of private vehicles, and knowl- edge of how to drive. Our study sample of the Delhi primary survey is the matched sample ( n = 1, 290) consisting of 184 never users, 182 new users, and 924 always users. Compared to the treated sample in the CPHS data, the matched sample is characterized by higher levels of education and annual household income. Specifically, approximately 68% of the bus users in the matched sample hold a graduate degree, and 85% of them report an annual household income of over |240,000. We compare perceptions of buses between nonusers and users (panel A of Table 1), as well as between new users and continuous users (panel B of Table 1) in the post-scheme period. We specifically examine their differences in the perceptions of five aspects: 1) affordability and availability of bus transit; 2) safety regarding accidents, crashes, threats, and thefts; 3) connectivity; 4) bus frequency, waiting time, travel duration, and unnecessary stops; 5) accessibility to bus stops. In panel A, we can see that nonusers consistently give lower rat- ings across all five perspectives; that is, they find bus travel less satisfactory across all five dimensions compared to users. They are particularly concerned about safety issues. The average rating concerning safety among never users is 1.63, indicating a level of satisfac- tion that falls between highly unsatisfactory and unsatisfactory. In panel B, we find that compared to continuous users, new users tend to give a lower rating on safety but a higher rating regarding the affordability and availability of free commutes. These results suggest that when it comes to transportation, women prioritize not only safety concerns but also affordability. Reducing costs in public transportation has the potential to encourage women to choose buses as a viable option for job search and commuting. 10 The occupation variable consists of the following categories: service, business, informal worker, daily wager, homemaker, and student. 10 3.3 Auxiliary Data Since our study period overlaps with the Covid-19 pandemic period, we obtain Covid-19 case data from Google Health. We also get the Indian Population Census of 2011 to obtain district-level characteristics like population. We combine the two data sets and construct average new confirmed daily cases as a share of the population in a district. We include this variable in our regressions to account for the potential effects of Covid-19 on outcomes like labor market participation. To address further concerns that different states experienced different levels of mobility restrictions, we use data from the Oxford Covid Tracker on the share of days each month that state governments imposed a shutdown on public transporta- tion. 4 Empirical Strategy States that implemented the Pink Slip program may differ from other states. Besides includ- ing only neighboring states of the treated state as controls, we employ a synthetic difference- in-differences strategy (SDID) proposed by Arkhangelsky et al. (2021) to further allay this concern. The SDID re-weights the control units to make their time trend parallel to the treated units and then applies a DID analysis to the re-weighted panel. This method con- structs a synthetic counterfactual for causal estimation. Since in our setting the treatment is applied at the state level, we construct a state-level panel using the CPHS data. With this panel, we are then able to derive the SDID unit and time weights for each control state. A detailed explanation is provided in Appendix A.1. To estimate the impact of the Pink Slip Program on women (or households with women), we employ two regression specifications for individual (or household) i in state j and time t weighted by the individual (or household) sampling weight, along with the SDID time and 11 unit weights. Starting with a standard stacked event-study design b Yi jt = α i + α t + β t · T reated j × 1(T ime = t) + ε i jt , (1) t= a where Yi jt are a set of outcome variables such as household monthly expenditure on trans- port and individual time usage on travel. The dummy variable T reated j takes the value 1 if a state implemented the Pink Slip program and 0 otherwise. 1(T ime = t) takes the value 1 in time period t after the event and 0 otherwise. Since the time units differ between the CP and PoI data, we set t as 1 month for the household consumption regression and as 4 months (equivalent to one wave) for the individual-level time usage regression. For example, in the household-level data (CP data), T ime = 0 corresponds to 1 month before the event; in the individual-level data (PoI data), T ime = 0 corresponds to 1 wave (equivalent to four months) before the event. The omitted base time period (or T ime = 0) is the wave (or month) of the survey before the start of the program. We then pool the pre- and post-treatment periods together to estimate Yi jt = α i + α t + β · T reated j × P ost t + ε i jt . (2) In both specifications, α i are individual (or household) fixed effects that control for all time- invariant characteristics of an individual (or a household). α t represents the time fixed ef- fects, which encompass month and year fixed effects in the regression that examines changes in household consumption. In the analysis of PoI data, α t comprises wave and year-fixed ef- fects. Appendix Table C2 provides a detailed definition for T ime t and P ost t . To account for the effect of the pandemic on people’s behavior, we also control for average daily confirmed Covid-19 cases as a share of the population. Because the treatment is at the state level, we cluster standard errors at the state level for statistical inference. Our parameters of interest are β t and β. 12 5 Main Results 5.1 Household Expenditure 5.1.1 Evidence from the CPHS Data We begin by examining the impacts of the Pink Slip Program on monthly household ex- ˆ t from the event-study specification (1) in Fig- penses. We report the estimated coefficient β ure 1. We do not detect any pre-trends. However, we see a substantial decrease in household expenditure on transportation in the treated states after the implementation of the program. Treated households report 10.6-37.2 log points lower expenses on all kinds of transport rel- ative to control households after the program was implemented (Figure 1(a)). We also find that the share of transport expense in total expenditure decreases by 0.3-0.6 percentage points (Figure 1(b)). Consistently, we find a reduction in the expenditure on Bus/tram/ferries category (BTF) as shown in Figure 1(c) and the share of BTF in total transport expenditure also fell as seen in Figure 1(d). In Table 2, we present the average treatment effects of the scheme on household expendi- tures, as estimated using equation (2).11 Across columns 1 to 5, it is evident that households in Punjab and Tamil Nadu spend considerably less on transportation compared to their coun- terparts in neighboring states during the post-treatment periods. Specifically, following the scheme implementation, expenditure on BTF as a proportion of total transport expenses for treated households decreased by an average of 6.6 percentage points in comparison to the control group (column 5). Even though we have controlled for Covid-19 cases in our main specification, Appendix Table C6 presents results from a robustness check where we additionally control for the share of days the state government had either recommended or required closing public transport in a month, and the share of days the state government 11 In Appendix Table C5, we cluster the standard errors at the district level. The significance levels of the estimated coefficients are similar to those in Table 2. 13 had either recommended or required individuals not to leave the house in a month. Results remain virtually the same, addressing any concern that changes in household expenditures could be driven by differences in stringency measures across treatment and control groups. 5.1.2 Evidence from the Delhi Primary Survey Next, we utilize the sample from the Delhi Primary survey data to examine changes in monthly transportation expenditure, including bus expenses, at the individual level.12 Of users (new and continuous) , 55% did not spend any money on transportation, including buses, per month in the period after the launch. The remaining 45% spent between |1 and |1,000 on transportation. Among non-bus-users, 64% spent |1∼1,000 per month on transportation and the remaining spent more than |1,000. In Figure 2, we present the changes in monthly travel expenditure among new users before and after the policy in Delhi. As shown, about half of these users transitioned from spending a certain amount of money on transportation to spending 0 rupees. To the extent that the two groups of users (new and continuous) are comparable, there is a notable shift in users’ travel expenditure patterns. 5.2 Women’s Time Use and Labor Supply Given that they face different decisions regarding time spent working and job searching, we present separate results for employed and unemployed (actively seeking jobs) women. Table 3 shows the results for time use based on using women’s contemporaneous work status without considering flows in and out of employment (Panel A) and restricting the sample to women who were continuously employed or unemployed throughout the study period (Panel B).13 The categories are time spent on household chores, travel time, and for those employed, 12 The survey inquires about the monthly travel expenditures of all respondents after the launch of the Pink Slip Program. However, it only queries new bus users about their monthly travel expenditures before the program’s launch. 13 In Appendix Table C7, we cluster the standard errors at the district level. The significance levels of the estimated coefficients are similar to those in Table 3. 14 time spent on work. We observe reverse patterns for change in time spent traveling for the employed and unemployed (Columns 2 and 5 of Panel A). While employed women spend less time traveling, unemployed spend more time. This is consistent with potential substitution from slower modes of travel (cheaper but slower buses or walking) to faster buses for the employed and increased time searching for a job for the unemployed.14 Since there are two time-use categories for the unemployed, their reported allocation to household chores decreases. For employed women, time savings from faster commuting can be reallocated to household chores or to additional work or a combination of both. Surprisingly, we find that it is the former: their hours spent on household work increase marginally. We will investigate this result in further detail below. To isolate changes in employment status, Panel B reports the results for women who were continuously employed or unemployed. For this sample, the same patterns persist, and are in fact, starker in terms of differences in time use for employed and unemployed women.15 Now, all four coefficients in columns 1-2 and 4-5 in panel B are significant at the 5 percent significance level. Comparing the results in panel B to panel A suggests that the changes in women’s time allocation patterns are driven by the intensive margin of those who have the same pre- and post-period labor market status.16 In order to unpack why employed women spend their saved travel time on household chores, we turn to behavior differences by skill levels of employed women. We define low-skilled 14 Unfortunately, we are limited in our CPHS data to shed direct light on this substitution in the mode of commuting but our Delhi Survey will enable us to do so in section 5.4.2. 15 In Appendix Table C8, we present a matrix of women’s labor market status before and after the program. Among those employed in the post-treatment periods, 65.8% had also been employed before the program de- ployment. Similarly, among those who were unemployed in the post-treatment periods, 58.3% were without employment in the pre-period. 16 We present the event study graph of travel time for currently unemployed women in Appendix Figure B2(a) and for always-unemployed women in Appendix Figure B2(b). In both figures, we observe that unemployed women in treated states spent more time on travel after the scheme implementation compared to those in the control states. 15 women as individuals who have not received education beyond primary school.17 We doc- ument the results by skill levels in Table 4.18 While all women save time on travel and use this saved time to do work household work (columns 1 and 2), low-skilled women, in addition, reduce their time towards work (column 3). In the following section 5.3, we pro- vide additional insights into this somewhat unexpected result that low-skilled women tend to allocate more time towards household chores at the cost of work hours when their bus commute becomes free. Specifically, we ask the question whether ascribed gender roles in developing countries, particularly for married women, can explain our result. 5.3 Intrahousehold Behavior We now explore whether marital status plays any role in inducing low-skilled women to spend more time on household chores and reducing travel time after the free bus scheme. In India, as in many developing countries, gender norms ascribe child rearing, grocery shop- ping, and cooking as roles to women. These expectations could impact low and high-skilled employed women differently as high-skilled employed women are likely to have higher wages and hence, higher bargaining power in the household or may simply marry later. To shed light on this, we examine how married employed women change their time use after the policy is implemented. Subsequently, we also examine the behavior of husbands and the heterogeneity of the women’s results by skill levels. 17 Goel (2017) and Thomas (2011) classify skill levels by education levels. Asuyama (2012) categorizes “people who were illiterate or have only received education below the primary level” as individuals with the lowest skill level. Mehrotra, Gandhi and Sahoo (2013) also group people who have below primary or only primary level of education as individuals with an extremely low level of skill. In this paper, we follow previous literature and group skill level by education level, setting individuals who have not received education or have only attended primary school as low-skill individuals. 18 These results are based on women’s skill levels before the scheme, as shown in Appendix Table C9. The time period is too short for any changes in skill status to occur from the pre- to the post-period. Appendix Table C11 verifies that this is the case. 16 Table 5 documents the results for the employed women by marital status.19 Married em- ployed women save time on commuting. However, this is invested in an increase in time spent on household chores. At the same time, it is only the married women who reduce the time spent at work: the reductions in work hours we observed in Table 3 for employed women is driven by the married employed women. Given the richness of our data, we can further parse out how these results vary by skill level and marital status. Appendix Table C12 shows the results by skill levels of married employed women (Panel A) and single employed women (Panel B). For unmarried women, especially with higher skills, there is no change in time spent on travel. We find this natural since this group is likely to have a low elasticity of demand for public transit: earning higher wages, they are less likely to ride public transit or change their behaviour even if it is very cheap or free. Especially among the married women we find heterogeneity by skill levels. It is the low-skilled married women who experience time savings from commuting. They significantly increase time spent on household chores and reduce labor hours. In fact, it is only the low-skilled married women who reduce their work hours. This reduced labor supply could be driven by two mechanisms related to household-level decision-making to maximize joint utility: first, there may be a discrete shift from men to women in household chores that require a discrete amount of time. For example, a long commute by mothers may have previously made it optimal for the father to take their kids to school, perhaps on a bus. After the roll-out of free bus rides for women, this responsibility may be reallocated to the mother. The gender wage gap increases the likelihood of such re-optimization at the household level. A second mechanism could be due to the nature of the work likely to be performed by low- skilled women. Let’s consider the following scenario. A low-skilled woman works as a house cleaner, which requires her to commute to a neighborhood with a high socioeconomic status. 19 The current results are based on women’s contemporaneous marital status but do not change if we use the pre-period status. As shown in Appendix Table C10, around 95% of women have the same status in the post-period as they do in the pre-period. In Appendix Table C11, we verify that the program did not have any compositional effect on women’s marital status. 17 Before the free rides, she has a time-consuming commute by foot. This induces an indivisible labor supply in that she finds it optimal to stay in the work neighborhood for the entire day and clean multiple homes. When buses become free and the travel time between her residen- tial and work neighborhoods shrinks, she finds it optimal to skip one of the houses she was previously cleaning, take a ride back to her neighborhood to run a household errand during the day, and later return to the work neighborhood for another job. This time reallocation simultaneously reduces both her work and travel hours, while increasing her time spent on household chores.20 Appendix D provides a simple model of optimal time allocation within the household to formalize these mechanisms and to explicitly state the assumptions needed for it to be consistent with the empirical results obtained above. To check the consistency of the above-stated conjectures within the household, we report the results of equation (2) for the time use of married men in Table 6. If the proposed mechanisms are potentially at work, we expect to find opposite patterns for married men. We find that, although imprecisely measured, there is a 3.7% reduction in the amount of time devoted to household work by employed married men compared to their unmarried counterparts (column 1 in panel B). At the same time, employed married men use the time saved from doing household chores to increase the time devoted to work by 5.4% with a statistical significance at the 1 percent level (column 3 in panel B). We found that only the low-skilled married women reduce their work hours, while increasing household work. To specifically check whether the husbands of low-skilled women drive the results in Table 6, in Table C13 we explore how the time use of men married to low-skilled women changed, compared to their counterparts married to medium-skilled or high-skilled women in our 20 For unemployed married women with low skills, we observe a reduction in time spent on household chores and an increase in time spent on travel, which is consistent with a possible increase in job search intensity. Results are available on request. While we are not able to test this directly using our data, we present some descriptive evidence using the Delhi survey data in section 5.4.2 which suggests that new users are more likely to use the scheme for work and travel longer distances for work using the free buses. 18 21 study. We restrict the sample to only single couple households to alleviate concerns that the distribution of household activities can be influenced by the skill-levels and time use decisions of other couples. Restricting to single couple households containing women in our study, we are able to analyse how the behaviour of husbands married to low, medium, or high-skilled women change. Panel A of Table C13 replicates the results in Table 6 for single couple households, confirming that only the married men increase time for work after the scheme. From column (3), panel B, we see that it is specifically the husbands of the low- skilled women in our sample who increase their time towards work. In fact, the spouses of high-skilled women in our sample reduce their work time. Our analysis thus suggests that the disproportionately large burden of household work that falls on married women, especially the unskilled, precludes them from benefiting in the labor market to either search for jobs or work more. These women instead use the saved time from commuting to facilitate their spouses to increase their labor supply. By contrast, married skilled employed women use their saved time from traveling to work more and increase their labor supply. Evidently, gender roles within households and bargaining power undermine the effects of the policy. 5.4 Overall Effects for Women 5.4.1 Evidence from the CPHS The results for time use for all women, employed and unemployed, pooled together are sum- marized in Table 7. Not surprisingly, we find no effect on time for household chores or travel time (columns 1 and 2) as the opposite effects for employed and unemployed women cancel each other. 21 We define low-skilled women as those who have not received education beyond primary school. Skilled women have a bachelor’s degree or higher, medium-skilled women have attended middle, secondary, or senior secondary education. 19 Using the information on work status from the People of India (PoI) panel of the CPHS as the outcome of interest in columns 3 and 5, we find no impact on being in the labor market or being employed. However, consistent with the time use and travel times of unemployed women, we find a marginal increase in the job search for women in column 4, whereas we do not detect this for men (see column 4 of Appendix Table C14).22 5.4.2 Evidence from the Delhi Primary Survey Leveraging our rich primary data from Delhi, we compare the purpose of travel and the average travel distance between new vs. continuous users of bus transit and find supportive evidence that free passes enable women to travel greater distances and facilitate job search efforts.23 In column 1 of panel A in Table 8, we find that new bus users are 6.8% more likely to travel for work (i.e., work and job search) than continuous users. In columns 3 and 4, we find that new users also tend to travel longer distances. They also travel more by bus (columns 5-6). In panel B, we restrict the matched sample to users whose travel purpose is work. We see that new users on average travel 1.1 kilometers longer distances for work by buses compared to always users (column 6). On comparing just the new users over time in Table 9, the average travel distance exhibits a negligible change over time of 0.005 kilometers. But, for new users who travel for work, we observe an average increase of 0.176 kilometers in the post Pink Slip Program period though the difference is measured imprecisely. 22 We present the corresponding event study graph for women in Appendix Figure B3. It shows that the marginal increase persisted in the post periods. The coefficients in time periods 1 and 2 after the policy are statistically significant at 10% and 15% levels, respectively. In period 3, the magnitude is unchanged but the estimate is imprecise. 23 We acknowledge that the new users are induced into using the buses because they are free and hence are selected relative to incumbent users. The comparison group of continuous users is matched on observable characteristics to the new users. While this does not address the selection issue completely, this comparison is still revealing and useful. 20 5.5 Women’s Income Appendix Table C15 shows the effects on women’s income. While there is no overall effect (column 1), there is an imprecisely estimated 12.0% increase in the income of the employed women (column 4). However, once we separate women by their skill levels, we see a 56.2% and significant increase in income for women with the medium-skill level. For unmarried women, we find an increase in income though not statistically significant at conventional levels (columns 2 and 5). Married employed women experience the largest ad- verse income effects. Compared to unmarried employed women, married employed women in treated states experience a 46.4% drop in wages in the post-treatment periods. This is consistent with our findings in section 5.3 where we found that low-skilled married employ- ment women take advantage of the scheme by doing more household work at the cost of reducing their working hours, while their married counterparts increase their labor supply. So, in terms of employment and wages, this program benefits only a limited set of women who are already employed and unmarried, or employed and married but highly skilled. Gen- der norms within the households, the bargaining power of women, and other discrimination that impede the culmination of more intensive job searches into higher employment preclude all women from reaping the benefits of reduced commuting costs. 6 Conclusion Using the roll-out of free bus services for women in several states of India under the Pink Slip Program, this paper evaluates the consequences of a reduction in commuting costs on women’s time use, including work hours, travel times, and time on household work, as well as their labor force participation. Employing a synthetic difference-in-differences approach, we find that household expenditures on buses fall and travel time for women employed be- fore the program decreases, implying that demand for transportation by women is elastic. 21 In the face of a large literature documenting that increasing commuting costs often reduce women’s work hours and labor force participation, policy attention has moved to decreasing the commuting barriers for women. We show that reducing commuting costs alone may not be enough for women to increase work hours or participate more in the labor market. Hetero- geneous responses may reflect the re-optimization of agents with varying demographic sta- tus. Free buses do incentivize unmarried, skilled, and unemployed women to increase labor force participation and search more intensively for jobs. Skilled and unmarried women who are already employed increase their work hours. But low-skilled and married women use the time saved from commuting using free buses to do more household work, replacing some of the household work done by their spouses, who in turn increase their work hours. This suggests the potential importance of household-level decision-making and intra-household bargaining power in determining gender norms that impede all women from taking advan- tage of the free bus scheme to increase their labor supply. Our findings have important policy implications: If the goal of providing free transporta- tion to women is to increase women’s labor force participation, only a small share of the workforce, primarily skilled unmarried women, benefit from this policy in the immediate short-run. 22 References Abadie, Alberto and Javier Gardeazabal, “The economic costs of conflict: A case study of the Basque Country,” American Economic Review, 2003, 93 (1), 113–132. Abe, Yukiko, “Family labor supply, commuting time, and residential decisions: The case of the Tokyo Metropolitan Area,” Journal of Housing Economics, 2011, 20 (1), 49–63. Abebe, Girum, Stefano Caria, Marcel Fafchamps, Paolo Falco, Simon Franklin, and Simon Quinn, “Curse of Anonymity or Tyranny of Distance? The Impacts of Job-Search Support in Urban Ethiopia,” Technical Report, National Bureau of Economic Research 2016. Alam, Muneeza Mehmood, Maureen Cropper, Matias Herrera Dappe, and Palak Suri, “Closing the Gap: Gender, Transport, and Employment in Mumbai,” World Bank Policy Research Working Papers, 2021, (9596). Anand, Anvita and Geetam Tiwari, “A gendered perspective of the shelter–transport– livelihood link: the case of poor women in Delhi,” Transport Reviews, 2006, 26 (1), 63–80. Arkhangelsky, Dmitry, Susan Athey, David A Hirshberg, Guido W Imbens, and Ste- fan Wager, “Synthetic difference-in-differences,” American Economic Review, 2021, 111 (12), 4088–4118. Astrop, Angela, C Palmner, D Maunder, and DM Babu, “The urban travel behaviour and constraints of low income households and females in Pune, India,” in “Second National Conference on Women’s Travel Issues, Baltimore” Citeseer 1996, pp. 23–26. Asuyama, Yoko, “Skill distribution and comparative advantage: a comparison of China and India,” World Development, 2012, 40 (5), 956–969. Barbanchon, Thomas Le, Roland Rathelot, and Alexandra Roulet, “Gender differ- ences in job search: Trading off commute against wage,” The Quarterly Journal of Eco- nomics, 2021, 136 (1), 381–426. Black, Dan A, Natalia Kolesnikova, and Lowell J Taylor, “Why do so few women work 23 in New York (and so many in Minneapolis)? Labor supply of married women across US cities,” Journal of Urban Economics, 2014, 79, 59–71. Borker, Girija, Safety first: Perceived risk of street harassment and educational choices of women, World Bank, 2021. , Gabriel Kreindler, and Dev Patel, “Safety in Numbers? Free Public Transportation for Women and Women’s Safety,” AEA RCT Registry 2020. Dasgupta, Aparajita and Ashokankur Datta, “Gendered Transport Subsidy and its Short Run Effect on Female Employment: Evidence from Delhi’s Pink Pass Scheme,” Ashoka University Discussion Paper 2023. Deshpande, Ashwini, Deepti Goel, and Shantanu Khanna, “Bad karma or discrimi- nation? Male–female wage gaps among salaried workers in India,” World Development, 2018, 102, 331–344. Dinkelman, Taryn and L Rachel Ngai, “Time use and gender in Africa in times of struc- tural transformation,” Journal of Economic Perspectives, 2022, 36 (1), 57–80. Durai, Raja Sethu, “Tamil Nadu launches free bus travel for women’s empower- ment,” Frontline, July 2021. https://frontline.thehindu.com/social-issues/gender/ free-rides-women-empowerment-tamil-nadu-bus-ticket-scheme-critical-evaluation/ article34793039.ece, Visited 2022-11-16. Duraisamy, Malathy and P Duraisamy, “Gender wage gap across the wage distribution in different segments of the Indian labour market, 1983–2012: exploring the glass ceiling or sticky floor phenomenon,” Applied Economics, 2016, 48 (43), 4098–4111. Express, The Indian, “Free bus travel for women in Punjab from to- day,” April 2021. https://indianexpress.com/article/cities/chandigarh/ free-bus-travel-for-women-in-punjab-from-today-7253260/, Visited 2022-11-17. Farré, Lídia, Jordi Jofre-Monseny, and Juan Torrecillas, “Commuting time and the gender gap in labor market participation,” Journal of Economic Geography, 2022. Field, Erica and Kate Vyborny, “Women’s mobility and labor supply: experimental ev- 24 idence from Pakistan,” Asian Development Bank Economics Working Paper Series, 2022, (655). Franklin, Simon, “Location, search costs and youth unemployment: experimental evidence from transport subsidies,” The Economic Journal, 2018, 128 (614), 2353–2379. Goel, Manisha, “Inequality between and within skill groups: The curious case of India,” World Development, 2017, 93, 153–176. Gonzalez, Karla Dominguez, Ana Luiza Machado, Bianca Bianchi Alves, Veron- ica Raffo, Sofia Guerrero, and Irene Portabales, “Why does she move? A Study of Women’s Mobility in Latin American Cities,” World Bank, Washington, DC, 2020. Goswami, Sweta, “Pink tickets boost DTC women riders,” Hindus- tan Times, August 2021. https://www.hindustantimes.com/cities/others/ over-76-of-dtc-passengers-were-women-in-last-five-months-101627755797751.html, Visited 2022-11-15. ILO, “World Employment Social Outlook–Trends for Women,” Geneva, 2017, pp. 1–63. Jayachandran, Seema, “Social norms as a barrier to women’s employment in developing countries,” IMF Economic Review, 2021, 69 (3), 576–595. Kejriwal, Arvind, “Free transport: Delhi government’s landmark move to empower women, give them greater claim to public spaces,” The Times of India, Nov 2019. https: //tinyurl.com/47k93ytu, Visited 2022-11-16. Khera, Purva, Closing Gender Gaps in India: Does Increasing Womens’ Access to Finance Help?, International Monetary Fund, 2018. Lee, Bun Song and John F McDonald, “Determinants of commuting time and distance for Seoul residents: The impact of family status on the commuting of women,” Urban studies, 2003, 40 (7), 1283–1302. Lei, Lei, Sonalde Desai, and Reeve Vanneman, “The impact of transportation infras- tructure on women’s employment in India,” Feminist economics, 2019, 25 (4), 94–125. Madden, Janice Fanning, “Why women work closer to home,” Urban studies, 1981, 18 (2), 25 181–194. Mahendru, Shirish, “Impact of incentivizing public transport system for women case study pink slip scheme - delhi,” LinkedIn, January 2022. https://www.linkedin.com/pulse/ impact-incentivizing-public-transport-system-women-case-mahendru/?trk=articles_ directory, Visited 2023-5-2. Martinez, Daniel F, Oscar A Mitnik, Edgar Salgado, Lynn Scholl, and Patricia Yañez-Pagans, “Connecting to economic opportunity: The role of public transport in pro- moting women’s employment in Lima,” Journal of Economics, Race, and Policy, 2020, 3 (1), 1–23. McQuaid, Ronald W and Tao Chen, “Commuting times–The role of gender, children and part-time work,” Research in transportation economics, 2012, 34 (1), 66–73. Mehrotra, Santosh, Ankita Gandhi, and Bimal K Sahoo, “Estimating India’s Skill Gap: on a Realistic Basis for 2022,” Economic and political weekly, 2013, pp. 102–111. Moreno-Monroy, Ana I and Héctor M Posada, “The effect of commuting costs and trans- port subsidies on informality rates,” Journal of Development Economics, 2018, 130, 99– 112. OECD, “Time spent travelling to and from work,” Technical Report, OECD Social Policy Division, Directorate of Employment, Labour and Social Affairs 2016. Patacchini, Eleonora and Yves Zenou, “Spatial mismatch, transport mode and search decisions in England,” Journal of Urban Economics, 2005, 58 (1), 62–90. Petrongolo, Barbara and Maddalena Ronchi, “Gender gaps and the structure of local labor markets,” Labour Economics, 2020, 64, 101819. Phillips, David C, “Getting to work: Experimental evidence on job search and transporta- tion costs,” Labour Economics, 2014, 29, 72–82. Sajjad, Fizzah, Ghulam Abbas Anjum, Erica Field, and Kate Vyborny, “Gender Eq- uity in Transport Planning: Improving Women’s Access to Public Transport in Pakistan,” International Growth Center IGC Policy Brief, 2017, 89300. 26 Sengar, Shweta, “Delhi Govt Claims Free Public Transport For Women Will Ensure Safety, Will It Really Help?,” India Times, June 2019. https://www.indiatimes.com/news/india/ delhi-govt-claims-free-public-transport-for-women-will-ensure-safety-will-it-really-help-368564. html, Visited 2022-11-16. Srinivasan, Sumeeta and Peter Rogers, “Travel behavior of low-income residents: study- ing two contrasting locations in the city of Chennai, India,” Journal of Transport Geogra- phy, 2005, 13 (3), 265–274. Sundaram, Ram, “More Chennai women take buses for free ride,” The Times of India, March 2022. https://timesofindia.indiatimes.com/city/chennai/ more-chennai-women-take-buses-for-free-ride/articleshow/90315169.cms, Visited 2022- 11-17. The Economic Times, “With the pink ticket, women ride DTC & cluster buses free of cost,” Oct 2019. https://economictimes.indiatimes.com/news/ politics-and-nation/with-the-pink-ticket-women-ride-dtc-cluster-buses-free-of-cost/ free-travel-in-buses-for-women/slideshow/71803010.cms, Visited 2022-11-16. Thomas, Jayan Jose, “Locked in a low-skill equilibrium? Trends in labour supply and demand in India,” Indian Journal of Labour Economics, 2011, 54 (2), 195–218. Tripathi, Kartikeya, Hervé Borrion, and Jyoti Belur, “Sexual harassment of students on public transport: An exploratory study in Lucknow, India,” Crime Prevention and Com- munity Safety, 2017, 19 (3), 240–250. Turner, Tracy and Debbie Niemeier, “Travel to work and household responsibility: new evidence,” Transportation, 1997, 24, 397–419. 27 Figures (a) Ln(Transport) (b) Transport/Expenditure (c) Ln(BTF) (d) BTF/Transport F IGURE 1: Event Study Graphs: Monthly Household Expenditure Notes: “Expenditure” = Total monthly household expenditure. “Transport” = Total monthly household expenditure on transport. “BTF” = Monthly household expenditure on daily bus/train/ferry fare. We conduct analysis on the sample of households that are also presented in the POI data. The time unit of analysis is monthly and t=0 is one month before the scheme starts. The scheme started in April and May 2021 in Punjab and Tamil Nadu, respectively. The time period of analysis is November 2020-August 2021 for Punjab and December 2020-September 2021 for Tamil Nadu. All regressions include household, month, and year fixed effects, as well as control for the average number of daily new confirmed cases as a share of the population. Standard errors are clustered at the state level. Confidence intervals are at the 95 percent level. 28 F IGURE 2: Average Monthly (|) Expenditures on Transportation Before & After the Program Notes: The figure plots the distribution of average monthly expenditure on transportation of new users of bus transit (there are 180 of them) before and after the program. A new user is defined as a person who started using a bus after the program. The unit is Rupee. The blue bars represent the share of new users whose average monthly expenditure on all other modes of transportation before the program belongs to a particular category. The red bars represent the same for new users of bus transit after the introduction of the program. 29 Tables T ABLE 1: User Perceptions of Buses & Pink Slip Scheme (1) (2) t-test Variable N Mean/SD N Mean/SD (1)-(2)/SE Panel A. Nonusers vs Users Nonusers Users Difference Affordability and availability of 184 3.196 1,106 3.914 -0.718*** free commute (0.786) (0.820) (0.065) Safety against accidents, crashes, 184 1.630 1,106 4.052 -2.422*** threats, and thefts (0.640) (0.878) (0.068) Connectivity 184 2.772 1,106 4.046 -1.274*** (0.798) (0.919) (0.072) Bus frequency, waiting time, travel 184 2.484 1,106 3.238 -0.754*** time, and unnecessary stops (0.992) (1.197) (0.093) Accessibility to bus stops 184 2.424 1,106 3.167 -0.743*** (1.053) (1.091) (0.086) Panel B. New users vs continuous users New users Continuous users Difference Affordability and availability of 182 4.071 924 3.883 0.189*** free commute (0.828) (0.816) (0.066) Safety against accidents, crashes, 182 3.945 924 4.073 -0.128* threats, and thefts (0.884) (0.876) (0.071) Connectivity 182 4.187 924 4.018 0.168** (0.878) (0.924) (0.074) Bus frequency, waiting time, travel 182 3.214 924 3.242 -0.028 time, and unnecessary stops (1.177) (1.201) (0.097) Accessibility to bus stops 182 3.132 924 3.174 -0.042 (1.048) (1.100) (0.089) Notes: Respondents evaluate their perception of a particular aspect of buses on a scale ranging from 1 (highly unsatisfactory) to 5 (highly satisfactory). In the last column, we test the differences between treated and control areas using a t-test with equal variance. * is p<0.1, ** is p<0.05, and *** is p<0.01. 30 T ABLE 2: The Impact on Household Transportation Expenditures (1) (2) (3) (4) (5) Ln(Transport) Ln(BTF) Transport/Expenditure BTF/Expenditure BTF/Transport Treat × Post -0.197** -0.801*** -0.004** -0.003*** -0.066*** (0.084) (0.201) (0.001) (0.001) (0.016) Control Mean (Level) 349.97 100.56 0.03 0.01 0.28 R2 0.62 0.68 0.67 0.64 0.62 No. of HHs 22,791 22,791 22,791 22,791 22,791 N 150,233 150,233 150,233 150,233 150,233 HH FE Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Notes: “Expenditure” = Total monthly household expenditure. “Transport” = Total monthly household expenditure on transport. “BTF” = Monthly household expenditure on daily bus/train/ferry fare. We conduct analysis on the sample of households who are also represented in the individual data. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control mean (level) in columns 1 and 2 are in Rupees. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. T ABLE 3: The Impact on Time Use and Allocation for Employed and Unemployed Women (1) (2) (3) (4) (5) Panel A. By Current Employment Status Employed Women Unemployed Women Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Ln(Time for HH) Ln(Travel Time) Treat × Post 0.191*** -0.098*** 0.014 -0.437+ 0.102+ (0.053) (0.011) (0.051) (0.249) (0.056) Control Mean (Level) 3.01 0.60 6.72 2.58 0.24 R2 0.53 0.52 0.43 0.75 0.64 No. of Individuals 2,916 2,916 2,916 1,570 1,570 N 9,906 9,906 9,906 5,636 5,636 Panel B. By Always-Employment Status (Same Employment Status Over Time) Always-Employed Women Always-Unemployed Women Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Ln(Time for HH) Ln(Travel Time) Treat × Post 0.145*** -0.117*** 0.025 -0.596** 0.144*** (0.025) (0.012) (0.035) (0.236) (0.042) Control Mean (Level) 3.04 0.63 6.86 1.82 0.27 R2 0.60 0.58 0.38 0.73 0.99 No. of Individuals 1,687 1,687 1,687 776 776 N 6,408 6,408 6,408 3,197 3,197 Individual FE Yes Yes Yes Yes Yes Wave FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Notes: “Time for HH” = Time spent on household activities. “Travel Time” = Time spent on travel. “Time for Work” = Time spent on work done for the employer. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control mean (level) in columns 1-5 are hours per day and is the average for women in pre-periods and control states. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. 31 T ABLE 4: The Impact on Time Use and Allocation of Employed Women by Skill (1) (2) (3) Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Treat × Post (TP) 0.118* -0.078*** 0.077 (0.059) (0.009) (0.051) Treat × Post × Low-Skill (TPL) 0.211*** -0.064*** -0.205*** (0.061) (0.014) (0.049) TP+TPL 0.329*** -0.141*** -0.128* p-value (0.000) (0.000) (0.052) Individual FE Yes Yes Yes Wave FE Yes Yes Yes Year FE Yes Yes Yes Control Mean (Level) 3.03 0.61 6.78 R2 0.54 0.52 0.43 No. of Individuals 2,916 2,916 2,916 N 9,906 9,906 9,906 Notes: The omitted group is the skilled women who have bachelor’s degrees or above. “Low-Skill” = Women who went to primary schools or received no education. “Time for HH” = Time spent on household activities. “Travel Time” = Time spent on travel. “Time for Work” = Time spent on work done for the employer. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control mean (level) is hours per day. The control mean is the average for low-skill employed women in pre-periods and control states. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. T ABLE 5: The Impact on Time Use of Employed Women by Marital Status (1) (2) (3) Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Treat × Post (TP) 0.098* 0.017 0.133** (0.050) (0.027) (0.041) Treat × Post × Married (TPMa) 0.196** -0.215*** -0.248*** (0.059) (0.026) (0.061) TP+TPMa 0.294*** -0.198*** -0.116 p-value (0.003) (0.000) (0.188) Individual FE Yes Yes Yes Wave FE Yes Yes Yes Year FE Yes Yes Yes Control Mean (Level) 2.69 0.70 7.21 R2 0.54 0.52 0.43 No. of Individuals 2,916 2,916 2,916 N 9,906 9,906 9,906 Notes: “Married” = A dummy takes the value of one if a woman is married and zero if the woman is divorced, unmarried, or widowed. “Time for HH” = Time spent on household activities. “Travel Time” = Time spent on travel. “Time for Work” = Time spent on work done for the employer. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control mean (level) is hours per day. The control mean is the average for unmarried employed women in pre-periods and control states. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. 32 T ABLE 6: Time Usage Analysis for Men by Current Employment Status (1) (2) (3) (4) (5) Employed Men Unemployed Men Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Ln(Time for HH) Ln(Travel Time) Panel A. All Men Treat × Post -0.029 0.019 0.014 -0.195 0.182 (0.085) (0.046) (0.045) (0.389) (0.119) Control Mean (Level) 1.49 0.76 7.86 1.45 0.31 R2 0.76 0.62 0.31 0.83 0.67 No. of Individuals 31,237 31,237 31,237 3,641 3,641 N 131,586 131,586 131,586 14,334 14,334 Panel B. Married vs. Unmarried Men Treat × Post 0.003 0.012 -0.031 -0.174 0.170 (0.089) (0.033) (0.050) (0.390) (0.120) Treat × Post × Married -0.037 0.008 0.054*** -0.522 0.319+ (0.057) (0.039) (0.015) (0.407) (0.191) Control Mean (Level) 1.59 0.79 7.74 1.30 0.31 R2 0.76 0.62 0.31 0.83 0.67 No. of Individuals 31,237 31,237 31,237 3,641 3,641 N 131,586 131,586 131,586 14,334 14,334 Individual FE Yes Yes Yes Yes Yes Wave FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Notes: The sample is restricted to men residing in households with the studied women. A member is unemployed if he is willing to work and is looking for a job. “Time for HH” = Time spent on household activities. “Travel Time” = Time spent on travel. “Time for Work” = Time spent on work done for the employer. “Married” = A dummy variable takes the value of one if a man is married and zero otherwise. All regressions include individual, quarter/wave, and year fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control mean (level) in panel A represents the average daily hours for men in pre-periods and control states, while in panel B, they represent the average daily hours for unmarried men in pre-periods and control states. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. 33 T ABLE 7: The Overall Impact on Time Use and Labor Market Outcomes (1) (2) (3) (4) (5) Time Usage Labor Market Ln(Time for HH) Ln(Travel Time) in Labor Mkt. Job Search (Not Employed) Employed + Treat × Post -0.030 -0.009 0.003 0.011 -0.017 (0.074) (0.071) (0.011) (0.006) (0.016) Control Mean (Level) 4.86 0.24 0.14 0.07 0.51 R2 0.75 0.65 0.81 0.78 0.89 No. of Individuals 43,855 43,855 43,855 41,347 4,537 N 189,668 189,668 189,668 177,657 16,007 Individual FE Yes Yes Yes Yes Yes Wave FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Notes: “Time for HH” = Average daily time spent on household activities. “Travel Time” = Average daily time spent on travel. “in Labor Mkt.” = A dummy takes the value of one if a member is employed or is unemployed but is looking for a job; it takes the value of zero if a member is unemployed and is neither willing nor looking for a job. “Job Search (Not Employed)” = A dummy takes the value of one if a woman is searching for jobs and zero if a woman is out of the labor market. “Employed” = A dummy takes the value of one if a woman is employed and zero if a woman is unemployed and is looking for a job. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control mean (level) in columns 1 and 2 are hours per day and is the average for women in pre-periods and control states. Standard errors are clustered at the state level. + is p<0.15, * is p<0.1, ** is p<0.05, and *** is p<0.01. T ABLE 8: Travel Purposes & Distance: New vs. Always Users (After the Scheme) Travel for Work Travel Distance Travel Distance (Bus) (1) (2) (3) (4) (5) (6) Panel A. All Purposes of Travel New users 0.0683* 0.066* 0.320 0.415+ 0.337 0.416+ (0.0405) (0.039) (0.283) (0.277) (0.289) (0.283) Other Controls No Yes No Yes No Yes Control Mean 0.44 0.44 10.85 10.85 10.85 10.85 R2 0.00 0.07 0.001 0.08 0.00 0.07 N 1,106 1,106 1,106 1,106 1,106 1,106 Panel B. Travel for Work Only New users 1.102*** 1.241*** 1.021*** 1.114*** (0.342) (0.365) (0.357) (0.376) Other Controls No Yes No Yes Control Mean 10.66 10.66 10.72 10.72 R2 0.02 0.06 0.02 0.05 N 496 496 496 496 Notes: We conduct regression analysis on the matched sample. In panel A, we include all new users and always users; in panel B, we restrict the sample to users whose main travel purpose is work, which includes both work and job search, in the post-scheme period. “Travel for work” means the users’ main travel purpose is work. “Travel Distance” is a user’s average travel distance in kilometers after the free bus scheme. “Travel Distance (Bus)” is a user’s average travel distance by buses in kilometers after the free bus scheme. Other controls include an individual’s marital status and education level. Standard errors are clustered at the user level. + is p<0.15, * is p<0.1, ** is p<0.05, and *** is p<0.01. 34 T ABLE 9: Average Travel Distance of New Users (Kilometer) (1) (2) t-test Before the scheme After the scheme Difference Variable N Mean/SD N Mean/SD (1)-(2)/SE All travel purposes 181 11.177 181 11.182 -0.005 (3.675) (3.531) (0.113) Travel for work 91 11.626 91 11.802 -0.176 (3.177) (2.894) (0.213) Notes: The survey asked new users their average travel distances before and after the scheme in kilometers. “All travel purpose” includes all new users who travel for all kinds of purposes (e.g., work, education, healthcare, shopping, religion, and leisure) after the free bus scheme. “Travel for work” only includes new users whose main travel purpose is work which includes both work and job search after the free bus scheme. In the last column, we test the differences between treated and control areas using a t-test with equal variance. * is p<0.1, ** is p<0.05, and *** is p<0.01. 35 APPENDIX A Data A.1 Construction of a State Panel for Synthetic Difference-in- Difference States that implemented the Pink Slip program could exhibit dissimilarities from other states, including neighboring ones. To mitigate this concern, we utilize a synthetic difference-in-differences approach (SDID) proposed by Arkhangelsky et al. (2021). The SDID method re-weights the control units to align their time trends with those of the treated units and subsequently applies a DID analysis to the re-weighted panel. That is, this method establishes a synthetic counterfactual state for causal estimation. Since our treatment oc- curs at the state level rather than the household (or individual) level, we aggregate our household-level (or individual) data to the state level and then match states and their pre- trends using a set of variables discussed below We use six variables for the household-level data. The variable “size group of household” is a categorical variable that includes groups such as one member, three members, eight to ten members, and more than 15 members. It is based on the number of members in a household. The variable “age group of household” is a categorical variable that includes groups such as households dominated by children, households dominated by grown-ups, and balanced households with seniors. The variable “occupation group of household” is a cat- egorical variable that includes groups such as wage laborers, self-employed professionals, entrepreneurs, and farmers. The occupation group of a household is based on the distri- bution of members of a household by the nature of their occupation. The CMIE classifies households into different groups based on certain rules and not just based on the occupation of the head of the household. The variable “education group of household” is a categorical 36 variable that includes groups such as all graduates, graduates majority, and some literates. The education group of a household is based on the distribution of members of a household by their education level. The variable “gender” is a categorical variable that includes groups such as only males, only females, female majority, and balanced. The last variable we use is the average new confirmed daily cases as a share of the state population. For each categori- cal variable, we then generate corresponding dummy variables, and these dummy variables after aggregating to the state level can be interpreted as the share of households having a certain characteristic. For example, one of the dummy variables created from “gender” is the share of households only having male members. We use nine variables for the individual-level data: age, religion (e.g., Hindu and Muslim), caste (e.g., upper caste, scheduled castes, and scheduled tribes), discipline (e.g., law and medicine), marital status (e.g., married, unmarried, and divorced), literacy, and education level (e.g., primary school and middle school). Except for age, all other variables are categori- cal variables. Similarly, we generate the corresponding dummy variables before aggregating them to the state level. Again, we also use the average new confirmed daily cases as a share of the state population. Using this approach, we compare the outcomes of treated states with that of a weighted com- bination of control states, the “synthetic” treated states without the free bus ride scheme, which has similar pre-treatment trends as the treated group. The weights applied in our regression for the household-level analysis and individual-level analysis are defined as fol- lows: Household-Level: Weight i jt = State Weight j × Time t × Household Weight it , (3) Individual-Level: Weight i jt = State Weight j × Time t × Individual Weight it . (4) In equations (3) and (4), “State Weight j ” and “Time t ” are the state unit weights and time weights derived from SDID, respectively. “Household Weight it ” is the sampling weight of a 37 household in the CPHS data. “Individual Weight it " is the sampling weight of a member of a sample household in the CPHS data. A.2 Delhi Primary Survey Travel Purpose & Distance The survey inquires about respondents’ primary travel pur- pose during the post-scheme period, offering a selection of options including work (including work and job search), education, healthcare, shopping, religion, leisure, pick/drop off, and others. Additionally, the survey collects information regarding users’ average travel dis- tances and average travel distances specifically by buses. It should be noted that the survey only asked new users about their average travel distance prior to the implementation of the scheme. 38 B Additional Figures F IGURE B1: Map of Delhi Sample Locations Notes: Each red dot on the map indicates a sample location, with the size of the dot representing the number of women surveyed at that location. Bigger dots indicate a larger number of women surveyed in that particular location. (a) Ln(Travel Time), Currently Unemployed Women (b) Ln(Travel Time), Always Unemployed Women F IGURE B2: Event Study Graphs: Time Spent on Travel Notes: “Travel Time” = Average daily time spent on travel. The time unit of analysis is one wave (four-month). Here t=0 represents one wave (four months) before the scheme. The scheme started in April and May 2021 in Punjab and Tamil Nadu, respectively. The time period of analysis here is May-August 2020 to January-April 2022 for both Punjab and Tamil Nadu. All regressions include individual, wave, and year-fixed effects, as well as control for the average number of daily new confirmed cases as a share of the population. Standard errors are clustered at the state level. Confidence intervals are at the 95 percent level. 39 F IGURE B3: Event Study Graph: Job Search (Not Employed), All Women Notes: “Job Search (Not Employed)” = A dummy takes the value of one if a woman is searching for jobs and zero if a woman is out of the labor market. The time unit of analysis is one wave (four-month). Here t=0 represents one wave (four months) before the scheme. The scheme started in April and May 2021 in Punjab and Tamil Nadu, respectively. The time period of analysis here is May-August 2020 to January-April 2022 for both Punjab and Tamil Nadu. The unit is Rupee. The regression includes individual, wave, and year-fixed effects, as well as control for the average number of daily new confirmed cases as a share of the population. Standard errors are clustered at the state level. Confidence intervals are at the 95 percent level. In the figure, the coefficients in time periods 1 and 2 are statistically significant at 10% and 15% levels, respectively. 40 C Additional Tables T ABLE C1: Treated vs. Control States Group Treated State Control States Group A Punjab Rajasthan, Haryana, Jammu & Kashmir, Himachal Pradesh Group B Tamil Nadu Kerala, Karnataka, Andhra Pradesh T ABLE C2: Definition of Time Periods in CP, InP & PoI Panel A. Consumption Pyramids (CP) & Income Pyramids (InP) T ime t Group A Group B P ost t -4 2020/11 2020/12 0 -3 2020/12 2021/1 0 -2 2021/1 2021/2 0 -1 2021/2 2021/3 0 0 2021/3 2021/4 0 1 2021/4 2021/5 1 2 2021/5 2021/6 1 3 2021/6 2021/7 1 4 2021/7 2021/8 1 5 2021/8 2021/9 1 Panel B. People of India (PoI) No. of Wave T ime t Group A Group B P ost t 20 -2 May-Aug 2020 May-Aug 2020 0 21 -1 Sept-Dec 2020 Sept-Dec 2020 0 22 0 Jan-Apr 2021 Jan-Apr 2021 0 23 1 May-Aug 2021 May-Aug 2021 1 24 2 Sept-Dec 2021 Sept-Dec 2021 1 25 3 Jan-Apr 2022 Jan-Apr 2022 1 Notes: The time unit in the CP and InP data is a month, while in the PoI data, it is a wave. The boxed red time period is when the event (Pink Slip program) starts. 41 T ABLE C3: Definition of Variables Variable Name Definition Household expenditure Expenditure Total monthly household expenditure which includes expenditure on food, trans- port, entertainment, and others. Transport Monthly household expenditure on transport. It includes daily bus, train, and ferry fares, auto-rickshaw or taxi fares, outstation bus or train fares, parking fees, toll charges, and airfare. BTF Monthly household expenditure on daily bus, train, and ferry. It includes fares paid for by both public and private modes of transport. Time usage Time for HH Average daily time spent on household activities including cooking food for household members and taking care of children. Time for Work Average daily time spent on work done for the employer. The forms of employ- ment include self-employment and salaried jobs. Travel Time Average daily time spent on traveling from one place to another for shopping, working, school, and others via all kinds of transportation. Labor market participation & Employment in Labor Mkt. A dummy variable is equal to one if a woman is either employed or unemployed but is willing to work or is looking for a job and zero otherwise. This variable is defined separately for the pre and post periods. Employed A dummy variable is equal to one if a woman is employed and zero if a woman is in the labor market but unemployed. This variable is defined separately for the pre and post periods. Job Search (Not A dummy variable is equal to one if she is unemployed and is looking for a Employed) job and zero if a woman is out of the labor market. This variable is defined separately for the pre and post periods. Out of Labor Mkt. A sub-sample (360) of women who are out of the labor market in the pre-period → Employed and become employed in at least one post-period wave. Out of Labor Mkt. A sub-sample (405) of women who are out of the labor market in the pre-period → Unemployed and start searching for a job in at least one post-period wave. It should be noted that the sub-samples of "Out of Labor Mkt. → Employed" and "Out of Labor Mkt. → Unemployed" are mutually exclusive. We excluded 29 women who experienced both employment and unemployment in the post-period. Education Middle A dummy variable equal to one if a woman has gone to a middle school and zero otherwise. High A dummy variable equal to one if a woman has gone to a secondary school or a higher secondary school and zero otherwise. ≥Bachelor A dummy variable is equal to one if a woman has at least a bachelor’s degree and zero otherwise. 42 Table C3 continued from previous page Variable Name Definition Skill Low-Skill A dummy variable equal to one if a woman has gone to primary schools or re- ceived no education and zero otherwise. Medium-Skill A dummy variable equal to one if a woman has gone to a middle school, a sec- ondary school, or a higher secondary school and zero otherwise. Skilled A dummy variable is equal to one if a woman has at least a bachelor’s degree and zero otherwise. Marital status Married A dummy variable is equal to one if a woman is married and zero if a woman is divorced, unmarried, or widowed. T ABLE C4: Summary Statistics: Household & Individual (1) (2) t-test Control Treated Difference Variable N Mean/SD N Mean/SD (1)-(2)/SE Panel A. Household, December 2020 Rural 9,380 0.259 5,992 0.246 0.013* (0.438) (0.431) (0.007) Number of household members 1∼3 9,380 0.483 5,992 0.595 -0.112*** (0.500) (0.491) (0.008) 4∼6 9,380 0.179 5,992 0.120 0.060*** (0.384) (0.325) (0.006) ≥7 9,380 0.338 5,992 0.285 0.053*** (0.473) (0.451) (0.008) Annual income of households ≤200K 9,380 0.346 5,992 0.557 -0.211*** (0.476) (0.497) (0.008) 200K∼400K 9,380 0.474 5,992 0.299 0.175*** (0.499) (0.458) (0.008) ≥400K 9,380 0.180 5,992 0.144 0.036*** (0.384) (0.351) (0.006) Monthly household expenditure Expenditure 9,380 14,356.035 5,992 13,120.722 1,235.312*** (6,903.750) (5,669.422) (106.682) Transport 9,380 348.090 5,992 382.163 -34.073*** (262.673) (292.496) (4.542) BTF 9,380 107.191 5,992 110.559 -3.368* (115.407) (111.090) (1.881) 43 Table C4 continued from previous page (1) (2) t-test Control Treated Difference Variable N Mean/SD N Mean/SD (1)-(2)/SE Panel B. Individual, May-August 2020 Age 11,958 42.927 8,394 43.317 -0.390* (14.255) (14.309) (0.203) Married 11,958 0.751 8,394 0.736 0.015** (0.432) (0.441) (0.006) Education Primary School 11,958 0.217 8,394 0.233 -0.015*** (0.412) (0.422) (0.006) Middle School 11,958 0.174 8,394 0.225 -0.051*** (0.379) (0.418) (0.006) Secondary & Higher Secondary School 11,958 0.451 8,394 0.403 0.047*** (0.498) (0.491) (0.007) ≥ Undergraduate 11,958 0.113 8,394 0.131 -0.018*** (0.317) (0.338) (0.005) Labor market participation in Labor Mkt. 11,958 0.145 8,394 0.087 0.057*** (0.352) (0.283) (0.005) Employed 1,732 0.515 734 0.678 -0.163*** (0.500) (0.467) (0.022) Time usage Time for HH 11,958 4.930 8,394 6.491 -1.560*** (3.044) (2.721) (0.042) Travel Time 11,958 0.198 8,394 0.107 0.091*** (0.382) (0.258) (0.139) Time for Work 892 6.119 498 7.058 -0.939*** (2.602) (2.249) (0.005) Notes: “Expenditure” = Total monthly household expenditure. “Transport” = Total monthly household expenditure on transport. “BTF” = Monthly household expenditure on daily bus/train/ferry fare. “in Labor Mkt.” = A dummy takes the value of one if a member is employed or is unemployed but is looking for a job; it takes the value of zero if a member is unemployed and is neither willing nor looking for a job. “Employed” = A dummy takes the value of one if a member is employed and zero if a member is unemployed and is looking for a job. “Time for HH” = Time spent on household activities. “Time for Work” = Time spent on work done for the employer. “Travel Time” = Time spent on travel. In the last column, we test the differences between treated and control areas using a t-test with equal variance. * is p<0.1, ** is p<0.05, and *** is p<0.01. 44 T ABLE C5: The Impact on Household Transportation Expenditures (SE at the District Level) (1) (2) (3) (4) (5) Ln(Transport) Ln(BTF) Transport/Expenditure BTF/Expenditure BTF/Transport Treat × Post -0.197** -0.801*** -0.004*** -0.003*** -0.066** (0.079) (0.227) (0.001) (0.001) (0.029) Control Mean (Level) 349.97 100.56 0.03 0.01 0.28 R2 0.62 0.68 0.67 0.64 0.62 No. of HHs 22,791 22,791 22,791 22,791 22,791 N 150,233 150,233 150,233 150,233 150,233 HH FE Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Notes: “Expenditure” = Total monthly household expenditure. “Transport” = Total monthly household expenditure on transport. “BTF” = Monthly household expenditure on daily bus/train/ferry fare. We conduct analysis on the sample of households who are also represented in the individual data. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control mean (level) in columns 1 and 2 are in Rupees. Standard errors are clustered at the district level. * is p<0.1, ** is p<0.05, and *** is p<0.01. T ABLE C6: Robustness Check: The Impact on Household Transportation Expenditures (1) (2) (3) (4) (5) Ln(Transport) Ln(BTF) Transport/Expenditure BTF/Expenditure BTF/Transport Treat × Post -0.258** -0.883** -0.003* -0.003*** -0.068*** (0.081) (0.342) (0.002) (0.001) (0.017) Control Mean (Level) 349.97 100.56 0.03 0.01 0.28 R2 0.63 0.68 0.67 0.64 0.62 No. of HHs 22,791 22,791 22,791 22,791 22,791 N 150,233 150,233 150,233 150,233 150,233 HH FE Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Notes: “Expenditure” = Total monthly household expenditure. “Transport” = Total monthly household expenditure on transport. “BTF” = Monthly household expenditure on daily bus/train/ferry fare. We conduct analysis on the sample of households who are also represented in the individual data. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population, the share of days the state government had either recommended or required closing public transport in a month, and the share of days the state government had either recommended or required individuals not to leave the house in a month. The control mean (level) in columns 1 and 2 are in Rupees. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. 45 T ABLE C7: The Impact on Time Use and Allocation for Employed and Unemployed Women (SE at the District Level) (1) (2) (3) (4) (5) Panel A. By Current Employment Status Employed Women Unemployed Women Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Ln(Time for HH) Ln(Travel Time) Treat × Post 0.191*** -0.098** 0.014 -0.437*** 0.102* (0.062) (0.048) (0.070) (0.140) (0.058) Control Mean (Level) 3.01 0.60 6.72 2.58 0.24 R2 0.53 0.52 0.43 0.75 0.64 No. of Individuals 2,916 2,916 2,916 1,570 1,570 N 9,906 9,906 9,906 5,636 5,636 Panel B. By Always-Employment Status (Same Employment Status Over Time) Always-Employed Women Always-Unemployed Women Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Ln(Time for HH) Ln(Travel Time) Treat × Post 0.145*** -0.117* 0.025 -0.596*** 0.144 (0.050) (0.059) (0.062) (0.153) (0.114) Control Mean (Level) 3.04 0.63 6.86 1.82 0.27 R2 0.60 0.58 0.38 0.73 0.99 No. of Individuals 1,687 1,687 1,687 776 776 N 6,408 6,408 6,408 3,197 3,197 Individual FE Yes Yes Yes Yes Yes Wave FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Notes: “Time for HH” = Time spent on household activities. “Travel Time” = Time spent on travel. “Time for Work” = Time spent on work done for the employer. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control mean (level) in columns 1-5 are hours per day and is the average for women in pre-periods and control states. Standard errors are clustered at the district level. * is p<0.1, ** is p<0.05, and *** is p<0.01. T ABLE C8: Share of Women by Labor Market Status Post-periods Out Employed Unemployed Any Out 88.7% 6.2% 10.0% 57.1% Employed 0.6% 65.8% 1.3% 10.4% Pre-periods Unemployed 0.5% 1.4% 58.3% 8.2% Any 10.2% 26.6% 30.4% 24.3% Total 100% 100% 100% 100% Notes: “Out” = Out of the labor market. “Any” = The women experienced more than one labor market status during pre- or post-periods. 46 T ABLE C9: Share of Women by Skill Post-periods Low-skill Medium-skill Skilled Total Low-skill 99.3% 0.7% 0.0% 100% Pre-periods Medium-skill 0.0% 100.0% 0.0% 100% Skilled 0.0% 0.0% 100.0% 100% Notes: The table presents the proportion of women who belong to a specific skill group and continue to be part of the same group in the post-periods. For example, around 99% of women who were classified as low-skill during the pre-periods maintained their low- skill status in the post-periods. T ABLE C10: Share of Women by Marital Status Post-periods Unmarried Married Total Unmarried 94.5% 5.5% 100% Pre-periods Married 3.2% 96.8% 100% Notes: The table displays the proportion of women who were (un)married in the pre-periods and remained (un)married in the post- periods. For instance, approximately 97% of women who were married during the pre-periods remained married in the post-periods. T ABLE C11: The Impact on Women’s Marital Status and Skills (1) (2) (3) (4) All Women Employed Women Married Skill Married Skill Treat × Post 0.005 0.002 0.006 -0.002 (0.006) (0.005) (0.008) (0.006) Control Mean 0.75 0.86 0.60 0.90 R2 0.93 0.93 0.90 0.98 No. of Individuals 43,855 43,855 2,916 2,916 N 189,668 189,668 9,906 9,906 Individual FE Yes Yes Yes Yes Wave FE Yes Yes Yes Yes Year FE Yes Yes Yes Yes Notes: “Married” = A dummy takes the value of one if a woman is married and zero if the woman is divorced, unmarried, or widowed. “Skill” = A categorical variable assigns 0 to women classified as low-skill, 1 to those classified as medium-skill, and 2 to those classified as skilled. All regres- sions include individual, quarter/wave, and year-fixed effects. The control mean is the average of the outcome variable for women in pre-periods and control states. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. 47 T ABLE C12: The Impact on Time Use of Employed Women by Marital Status & Skill (1) (2) (3) Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Panel A. Married Women Treat × Post 0.371*** -0.210*** -0.213** (0.049) (0.014) (0.091) Treat × Post × Medium-Skill -0.099+ -0.007 0.114+ (0.060) (0.035) (0.066) Treat × Post × Skilled -0.350*** 0.007 0.398* (0.099) (0.093) (0.176) Control Mean 1.76 0.48 2.47 R2 0.54 0.53 0.43 No. of Individuals 1,602 1,602 1,602 N 5,038 5,038 5,038 Panel B. Unmarried Women Treat × Post 0.251*** -0.023 -0.029 (0.052) (0.031) (0.037) Treat × Post × Medium-Skill -0.154** 0.038 0.152+ (0.053) (0.028) (0.091) Treat × Post × Skilled -0.348*** 0.057 0.187+ (0.086) (0.043) (0.108) Control Mean 1.69 0.60 2.59 R2 0.58 0.52 0.45 No. of Individuals 1,318 1,318 1,318 N 4,552 4,552 4,552 Individual FE Yes Yes Yes Wave FE Yes Yes Yes Year FE Yes Yes Yes Notes: We restrict the sample to employed women. A woman is unmarried if she is divorced, unmarried, or widowed. In both panels, the omitted group is the low-skill women who went to primary schools or received no education. “Medium-Skill” = Women who went to middle school, secondary schools, or higher secondary schools. “Skilled” = Women who have bachelor’s degrees or above. “Time for HH” = Time spent on household activities. “Travel Time” = Time spent on travel. “Time for Work” = Time spent on work done for the employer. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. 48 T ABLE C13: The Impact on Time Use of Employed Men in Household with only One Couple by Marital Status and Household Composition (1) (2) (3) Ln(Time for HH) Ln(Travel Time) Ln(Time for Work) Panel A. Married vs. Unmarried Men Treat × Post 0.011 0.040 -0.064 (0.089) (0.040) (0.058) Treat × Post × Married -0.038 -0.027 0.094*** (0.081) (0.042) (0.028) Control Mean (Level) 1.55 0.80 7.66 R2 0.76 0.62 0.32 Panel B. If households have married low-skill/medium-skill/skilled women Treat × Post 0.011 0.049 -0.099 (0.092) (0.055) (0.064) Treat × Post × Married -0.001 -0.040 0.119+ (0.059) (0.087) (0.069) Treat × Post × Married -0.063 0.017 -0.036 × Has Med.-Skill Married W (0.101) (0.065) (0.054) Treat × Post × Married 0.126 -0.006 -0.497+ × Has Skilled Married W (0.143) (0.164) (0.291) Control Mean (Level) 1.74 0.77 7.65 R2 0.76 0.62 0.31 Individual FE Yes Yes Yes Wave FE Yes Yes Yes Year FE Yes Yes Yes No. of Individuals 23,962 23,962 23,962 N 100,951 100,951 100,951 Notes: We restrict the sample to employed men who are in households with only one couple containing the women in our study. A man is unmarried if he is divorced, unmarried, or widowed. “Time for HH” = Time spent on household activities. “Travel Time” = Time spent on travel. “Time for Work” = Time spent on work done for the employer. “Has Med.-Skill Married W” = A dummy variable that takes the value of one if a man belongs to a household with only married, medium-skill women, and zero otherwise. “Has Skilled Married W” = A dummy variable that takes the value of one if a man belongs to a household with only married, skilled women, and zero otherwise. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. 49 T ABLE C14: Impacts on Time Usage & Labor Market Participation for Men (1) (2) (3) (4) (5) Time Usage Labor Market Ln(Time for HH) Ln(Travel Time) in Labor Mkt. Job Search (Not Employed) Employed Treat × Post -0.076 -0.001 -0.007 -0.005 -0.005 (0.084) (0.032) (0.008) (0.006) (0.004) Control Mean (Level) 1.72 0.61 0.75 0.32 0.85 R2 0.77 0.70 0.94 0.88 0.81 No. of Individuals 47,480 47,480 47,480 16,977 34,825 N 207,772 207,772 207,772 71,633 148,514 Individual FE Yes Yes Yes Yes Yes Wave FE Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Notes: The sample is restricted to men residing in households with the studied women. “Time for HH” = Average daily time spent on household activities. “Travel Time” = Average daily time spent on travel. “in Labor Mkt.” = A dummy takes the value of one if a member is employed or is unemployed but is looking for a job; it takes the value of zero if a member is unemployed and is neither willing nor looking for a job. “Job Search (Not Employed)” = A dummy takes the value of one if a woman is searching for jobs and zero if a woman is out of the labor market. “Employed” = A dummy takes the value of one if a woman is employed and zero if a woman is unemployed and is looking for a job. All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. The control means (level) in columns 1 and 2 are hours per day for men in pre-periods and control states. The control means (level) in columns 3-5 are the average share of men in pre-periods and control states. Standard errors are clustered at the state level. + is p<0.15, * is p<0.1, ** is p<0.05, and *** is p<0.01. 50 T ABLE C15: Impacts on Wages for Women (1) (2) (3) (4) (5) (6) All All All Employed Employed Employed Treat × Post -0.006 0.025 -0.003 0.120 0.299 -0.235 (0.015) (0.039) (0.037) (0.239) (0.263) (0.349) Treat × Post × Married -0.041 -0.464** (0.040) (0.153) Treat × Post × Medium-Skill -0.009 0.562** (0.031) (0.249) Treat × Post × High-Skill 0.016 0.354 (0.054) (0.354) Control Mean 0.08 0.15 0.13 2.32 4.06 2.52 R2 0.93 0.93 0.93 0.86 0.86 0.86 No. of Individuals 36,245 36,245 36,245 1,865 1,865 1,865 N 270,340 270,340 270,340 9,141 9,141 9,141 Individual FE Yes Yes Yes Yes Yes Yes Month FE Yes Yes Yes Yes Yes Yes Year FE Yes Yes Yes Yes Yes Yes Notes: The outcome variable is the log of wages. In columns 1-3, we include all women who also appear in the People of India (PoI) data; in columns 4-6, we restrict to employed women. The Income Pyramids (InP) data is collected on a monthly basis at the individual level. To ensure consistency with our Consumption Pyramids (CP) data, we use the same sample period as in CP (November 2020 to September 2021), which is displayed in Table C2. The control mean represents the average log wage value across various categories: all women in control states during pre-periods (column 1), all unmarried women in control states during pre-periods (column 2), all low-skill women in control states during pre-periods (column 3) employed women in control states during pre-periods (column 4), employed unmarried women in control states during pre-periods (column 5), and employed low-skill women in control states during pre-periods (column 6). All regressions include individual, quarter/wave, and year-fixed effects and control for the average number of daily new confirmed cases as a share of the population. Standard errors are clustered at the state level. * is p<0.1, ** is p<0.05, and *** is p<0.01. 51 D A Model of Intrahousehold Time Allocation We now describe a simple model of optimal time allocation within the household to formalize the mechanisms discussed in section 5.3. The empirical results indicated a decrease in labor supply and an increase in household chores for married, low-skilled and employed women after the introduction of the pink slip program. Accordingly, we consider a different-sex household in which the man works salaried so that his income I m does not depend on hours worked. The woman, on the other hand, has a divisible labor supply at an hourly wage of w f where the f subscript stands in for female. Total female time endowment is E hours per day. There is an indispensable household chore requiring H hours. It is an indivisible activity, such as grocery shopping, which needs to be performed by either the man or the woman. The disutility of the chore is d m and d f for the man and woman, respectively. The household maximizes joint utility by allocating the chore to one of its members, captured by an indicator function 1H that takes the value of one if it is performed by the woman, and zero otherwise:   0  if household chore performed by man, 1H =  1  if household chore performed by woman. Note that, in most households in India, household work will be shared by both men and women with the bulk of the work done by women.24 The choice of either men or women performing household work is done for simplicity for exposition and the results hold when household work can be shared. Household surplus is increasing in total income net of the disutility from the household chore. Total income depends on hours worked by the woman, which is the residual left after commuting C hours and taking care of H if she is responsible 24 For example, in our data the average hours per day spent on household chores by married women and men are 3.20 and 1.47 hours respectively in control states before the scheme. 52 of the household chore: max I m − d m (1 − 1H ) H + w f (E − C − 1H · H ) − d f · 1H · H 1H =0,1 man’s surplus woman’s surplus s.t. I m + w f (E − C − 1H · H ) ≥ w where the constraint captures the subsistence threshold of income w that the household needs to attain. We assume that male salary alone is not high enough to meet subsis- tence, i.e., I m < w, so that the woman needs to work regardless of C subject to the time constraint E − C − 1H · H ≥ 0. We assume that the commute time C ∈ {C slow , C f ast } depends on whether the woman takes a slow or fast mode of transport, requiring a higher or lower travel time, C slow > C f ast , respectively. Therefore, there is a trade-off in female time alloca- tion between work versus commuting. We assume that before the program offering free bus rides to women is rolled out, the slower and more-time consuming commute is chosen. We do not model man’s commuting as his salary does not depend on hours worked. Next, we back out from the model the restrictions on the parameters needed for the optimal decisions to be compatible with our empirical results. If these restrictions are plausible, the model offers a useful lens through which these results can be interpreted. Recall that our objective is to rationalize the decrease in labor supply and the increase in household chores for women after the introduction of free bus rides. In our model, this corresponds to 1H = 0 when C = C slow , and 1H = 1 when C = C f ast . Assuming that fast commuting by free buses leaves enough time for the woman to work even if she undertakes the household chore, i.e., E − C f ast − H > 0, the latter implies I m + w f (E − C f ast − H ) − d f H > I m − d m H + w f (E − C f ast ) ⇒ dm − d f > w f . For these assumptions to be compatible with non-negative wages, gender-specific disutility of chores needs to satisfy d m > d f . Moreover, man’s disutility has to be high enough so that 53 hourly low-skilled female wages cannot compensate for it within the household. This is a plausible assumption for a low-skilled household in the Indian context. However, in the absence of another restriction, d m − d f > w f would imply that household work would be allocated to the woman even before the program when C = C slow . To ra- tionalize 1H = 0 when C = C slow , it must be the case that doing both the household chores and working—which necessitates commuting—violates the time constraint of the woman: E − C slow − H < 0. Hence, the two constraints faced by the household—to attain subsistence within the time endowment—leave no choice but for the man to do household chores and the woman to work. When commuting time drops, however, they find it optimal to reallocate the household work to the woman. For this to reduce her market work hours, model parameters have to satisfy E − C slow > E − C f ast − H ⇒ H > C slow − C f ast . work hours before free buses work hours after free buses That is, household chores are more time-consuming than the time saving from the free and faster commuting. The indivisibility of a high enough H is therefore essential for rationaliz- ing the empirical results. 54