The effect of affordable daycare on women’s mental health: Evidence from a cluster randomized trial in rural India

Abstract

Access to affordable daycare might improve population mental health. However, evidence is sparse and restricted to middle- and high-income country settings. We conducted a cluster-randomized controlled trial in one low-income setting, rural Rajasthan, India. Communities lacking daycare facilities were identified (n = 160) and randomly selected for assistance in setting up a community-based daycare program (n = 80) or not (n = 80). Women eligible for the daycare program living in these communities completed structured interviews before the intervention (participation rate = 89%) and approximately one year after rollout of the intervention (participation rate = 96%), resulting in a final analytic sample of 3041. Mental distress was measured with the Hindi version of the 12-item General Health Questionnaire (score range: 0–12). We modeled the relation between access to daycare and number of mental distress symptoms (GHQ-12 score) with negative binomial regression using an intention-to-treat approach, which groups women according to if they lived in communities randomized to affordable daycare. We also evaluated the effect of access to daycare on secondary outcomes that may be related to mental distress, including women’s work burden, agency, and intimate partner violence (IPV). We found that access to daycare resulted in modest reductions in symptoms of mental distress (mean difference = 0.21, 95% CI: −0.43, 0.02). We found some evidence that daycare reduced IPV, but virtually no change in women’s work burden or agency. Our results provide some indication that access to affordable daycare might be one policy lever to improve population mental health.

1. Introduction

Common mental disorders (CMDs) encompass mood and anxiety disorders that are commonly experienced in both community and clinical settings (D. Goldberg and Huxley, 1992). CMDs affect a large proportion of women worldwide. A recent systematic review and meta-analysis of 157 studies conducted in 59 countries estimated that 14% of women will experience a mood disorder and 18% will experience an anxiety disorder in their lifetime (Steel et al., 2014). Structural factors that give rise to economic and social disadvantage play a key role in the development of mental health problems (Patel, 2015), and identifying interventions to confront these structural factors could greatly improve population mental health (Patel, 2015). However, intervention studies that target structural factors, such as poverty alleviation programs (Lund et al., 2011), rarely investigate mental health outcomes. Access to affordable daycare might be one structural factor that could reduce women’s risk of experiencing a CMD, but evidence is sparse and inconsistent (Ángeles et al., 2011; Baker et al., 2008; Rosero and Oosterbeek, 2011).

Using data from two waves of a cluster-randomized trial, the aim of this study was to evaluate the effect of providing access to an affordable daycare program on women’s mental distress in a lower income setting, rural Rajasthan, India. We also evaluated if the provision of daycare resulted in changes to secondary outcomes related to daycare and mental distress, including women’s work demands, intimate partner violence (IPV), and agency.

2. Background

2.1. Daycare and development of common mental disorders

In many societies, traditional gender roles relegate domestic and childcare work to women (Fisher et al., 2014). This work is largely invisible. When these duties are counted as work, women worldwide perform greater amounts of work than men (International Labour Organization, 2016). High work amounts are associated with worse mental health (Dinh et al., 2017; Kleiner and Pavalko, 2010; Richardson et al., 2017), and specific aspects of high work amounts may be especially detrimental to mental health, such as attempting to complete too many tasks without enough time (Roxburgh, 2004), having conflicting paid work and household demands (Chandola et al., 2004; Cooklin et al., 2016), and having high childcare demands (des Rivières-Pigeon et al., 2002; Matud et al., 2015; Ozer, 1995).

Access to affordable daycare might improve women’s mental health through many interrelated mechanisms. First, daycare might reduce women’s total work amount, which could reduce mental distress and risk of CMDs by reducing detrimental work aspects such as conflicting housework and paid work demands. Daycare may also free up women’s time to engage in mental health promoting activities such as relaxation, self-care, and adequate sleep, as well as creating occasions to strengthen social ties with neighbors, friends, and family members (Kawachi and Berkman, 2001).

Second, daycare may shift the composition of women’s work demands by making it easier to engage in work that can be difficult when caring for young children, such as farm work or paid work. While these changes in work composition may not reduce work burden, they might lead to improvements in mental health by allowing women to contribute more directly to the economic productivity of the household. Prior research links such activities, such as farm work, with slight reductions in mental distress (Richardson et al., 2017). Relatedly – and conversely – daycare could increase women’s total work burden if it creates opportunities to engage in activities such as paid work without commensurate reductions in caregiving responsibilities. Thus, daycare could shift activities in a manner that increases the total work amount, which could negatively affect mental health.

Third, changes in the composition of women’s work and leisure time may affect other aspects of women’s lives that are related to mental health. For example, becoming a breadwinner through an increase in paid work may give women more say in decisions related to herself and the household. These changes might increase a woman’s ability to make choices and act upon those choices (i.e., agency (Kabeer, 1999)) or reduce her exposure to IPV, which are two factor associated with higher prevalence of CMDs (Devries et al., 2013; Golding, 1999; Patel et al., 2006; Yount et al., 2014).

2.2. Daycare in India

Access to affordable, high-quality daycare is limited in India. Current government programs include the Integrated Child Development Scheme (ICDS), which provides pre-school education to children ages 3 to 6 through local facilities (anganwadis), and nurseries that provide care for young children (crèches). However, crèches are poorly regulated and are rarely functional, and the few operational crèches are characterized by poorly trained staff and substandard facilities (Palriwala and Neetha, 2009). Anganwadis only reach about half of eligible children (International Institute for Population Sciences (IIPS) and ICF, 2015/2016), and are marked by insufficient hours of operation, poorly trained workers, chronic staff absenteeism, and substandard facilities (Palriwala and Neetha, 2009).

3. Methods

3.1. Study design

This cluster-randomized controlled trial assessed the impact of providing affordable daycare on women and children’s health and wellbeing. We recruited mothers or guardians with children between the ages of one and six living in rural communities from five blocks (geographical areas) in the Udaipur District of Rajasthan, India. A total of 160 village hamlets (i.e., clusters of houses that constitute separate communities) were included in the study. The trial protocol is publicly available (A Nandi et al., 2016).

3.2. Participants

Potential village hamlets were identified by the non-governmental organization (NGO) Seva Mandir, which operates community development programs, including daycare centers called balwadis, in rural communities in the Udaipur District. Village hamlets where Seva Mandir had not previously established a daycare were selected between December 2014 and January 2015. These hamlets met the following criteria, established a priori: 1) there was no readily accessible government-operated daycare; 2) at least 25 children between the ages of one and six lived in the hamlet; 3) hamlets had an existing structure suitable for a daycare; 4) a woman qualified to operate the daycare lived in the study hamlet or nearby; and 5) the village council indicated adequate demand for daycare. To reduce potential spillover effects between treatment and control hamlets that might occur if women in control hamlets enrolled their children in balwadis in treatment hamlets, we selected control hamlets that were at least 1.5 km from treatment hamlets. Hamlets tended to be geographically isolated.

We conducted a household census in the 160 selected village hamlets to enumerate the population and identify eligible households, namely those with a mother (either biological or guardian) of at least one child between one and six years old. Trained interviewers randomly selected one eligible woman from each eligible household. Selected women were invited to participate in the study and underwent an informed consent process. Women who could read and write signed a written consent form, and women who could not read or write gave oral consent. Interviewers gave all women a copy of their consent form, which included contact details for the regional research manager. After consenting, women completed baseline interviews (described in the Procedures section) and were offered a small gift for participation.

3.3. Randomisation and masking

Treatment was assigned using a stratified randomisation procedure. Since there was substantial heterogeneity across blocks, we stratified by block (n = 5) and randomly selected hamlets to receive access to the affordable daycare program (n = 80) or serve as control hamlets (n = 80). Assignment to a treatment or control hamlet was conducted by one of the investigators (SH) using a random number generator in Stata. The treatment assignment was communicated to Seva Mandir, who implemented the daycare programs.

Village hamlets were assigned to treatment or control groups after completion of baseline interviews to minimize bias in recruitment of participants and to avoid biased baseline participant responses due to treatment assignment. Due to the nature of the intervention, it was not possible to mask treatment assignment to study participants or interviewers after implementation of the intervention. The research assistant who cleaned the data and the author conducting the analysis (RR) were not blinded to treatment group assignment.

3.4. Procedures

The intervention was an offer of full-time, community-run, affordable daycare (balwadi). Each balwadi provided childcare, nutritious meals, preschool education, and linkage to health services (e.g., immunizations) to children between one and six years of age. Balwadis were operated by local women, called sanchalikas, who were hired and trained by Seva Mandir. Any child between the ages of one and six living in these communities could use the balwadi, regardless of participation in the study. Families using the daycare facility were charged a small yearly fee per child (i.e., 150 rupees or about $2.30 USD). These fees were deposited into a collective fund, which was used to purchase items for the children attending the balwadi (e.g., shoes, sweaters).

Daycare services in treatment hamlets were promoted to encourage high utilization. After treatment assignment, community meetings in the treatment hamlets informed residents that a daycare program would be set up in their community and explained how to access the daycare. Once the daycare was set up, the sanchalika visited households of eligible children to encourage families to enrol their children in the program.

Two field workers visited each of the 80 balwadis each month to verify that the balwadis were operational and operating in the correct location, that children were receiving food, and that care was being provided by the sanchalika. Seva Mandir monitored the number of days the daycare centers were open with a camera monitoring system that required sanchalikas to take three self-timed pictures each day (i.e., at approximately 10am, 12:30pm, and 4pm). This system has been shown to improve teacher attendance (Duflo et al., 2012). These photographs were used to confirm the number of days the balwadis were open for at least 6 h, which is considered a full day of operation. Sanchalikas were encouraged to operate the daycare centers at least 5 days a week for 6 h each day, and they received a monthly salary that was based upon the number of full days the balwadi was open.

Eligible women living in the 160 study hamlets completed structured interviews administered by trained interviewers in their homes. Baseline interviews were conducted between January and June 2015. Village hamlets were assigned to treatment or control groups after baseline interviews were completed. Follow-up interviews were conducted between June and October 2016, approximately one year after implementation of the daycare centers.

3.5. Measures

We collected household composition and socio-demographic information. At follow-up, women’s utilization of the balwadi in the past year was measured by asking women if they made any use of the balwadi, as well as the number of days per week and hours per day each eligible child typically attended. We measured household wealth with 23 asset-based indicators that are commonly used to measure wealth in LMICs (Filmer and Pritchett, 1999) using a principle component analysis, which is described in Appendix 1.

We attempted to reduce errors in the measurement of our primary study variables (i.e., mental distress, agency, work burden, IPV). During the development of our questionnaire we assessed face validity by vetting variables for relevance with a local advisory committee and pilot testing variables in a sample of 200 women living in villages adjacent to our study communities to assess comprehension and suitability. Some changes were made to our study variables based on these activities, which are discussed below. Overall, results of these activities indicate the survey questions were relevant to our study population.

Mental distress was measured using the 12 item General Health Questionnaire (GHQ-12) (D. P. Goldberg, 1972), which was translated into Hindi by Gautam et al. (1987). The GHQ-12 is commonly used to detect mental health problems in India (Patel et al., 1998, 2008; Shidhaye and Patel, 2010). For each item, women were asked which of four responses corresponded most closely to how they had been feeling recently. For example, one item asks, “Have you recently felt capable of making decisions about things?”, and potential responses ranged from “much less than usual” to “more so than usual”. We dichotomized responses to indicate none versus some distress for each item using a scoring system commonly employed in India (Patel et al., 1998, 2008; Shidhaye and Patel, 2010) (i.e., the 0–0-1–1 scoring system). Using this system, distress scores could range from 0 to 12, with higher scores denoting more distress. Although the GHQ-12 is commonly used to classify individuals as having a CMD using pre-determined cut-points, we chose to keep scores continuous for two reasons. First, we are not aware of validation work conducted among women in tribal communities, and thus the optimal cut-point in this population is not known. Validation work in other populations indicates that specific cut-points for classifying a CMD can vary considerably, even within India (Patel et al., 2008; Shamasundar et al., 1986). Second, it is widely acknowledged that mental health disorders fall upon a continuum and schemes to classify people as having or not having a mental health disorder, such as depression (Patel, 2017), rely on arbitrary cut-points. Accordingly, many have argued against classification schemes in the study of mental health problems (Krishnan, 2008; A. Nandi et al., 2009; Parker, 2000, 2006).

Work demands were measured with a time use survey (Beaman et al., 2012) that asked respondents how much time they spent on paid work (e.g., agricultural labour) and unpaid work activities (e.g., laundry) in the past 24 h. The survey only captured the primary work activity; multitasking, when women performed two or more work activities at once (such as cleaning while caring for children), was not accounted for. We created variables for paid work, farm work (caring for animals, working in own field), housework (collecting water, cooking, cleaning, laundry, gathering fuel or firewood), and caring for children, the elderly or the disabled. We also created a summary measure of total work amount by summing together all of these work activities.

IPV was measured with questions from the Demographic and Health Survey’s Domestic Violence Module (United States Agency for International Development, 2014), which includes 6 questions about women’s experiences of physical abuse (e.g., slapped by partner) in the past year, 4 questions about psychological abuse (e.g., partner threatened to hurt you) in the past year, and 5 questions about partner’s controlling behaviour (e.g., partner limits contact with your family) that was not restricted to the past year. Although this module also includes questions about sexual abuse (e.g., does your husband physically force you to have sexual intercourse?), these questions were not included in our survey because the advisory board suggested it was not culturally appropriate to ask sex-related questions during survey interviews. Response categories include “not at all”, “sometimes”, and “often”. We classified women as experiencing abuse in each of these three categories if she answered “sometimes” or “often” to any item.

Women’s agency is a latent concept that presents many measurement challenges. We drew upon a robust body of theoretical and empirical literature that developed indicators and measurement approaches to measure agency. This work indicates that agency encompasses many life domains (Agarwala and Lynch, 2006; Ibrahim and Alkire, 2007; Malhotra and Schuler, 2005; Mason and Smith, 2003) – such as freedom of movement and decision-making in the home – and women may have high agency in some domains but not others (Gupta and Yesudian, 2006; Malhotra and Mather, 1997; Mason and Smith, 2000). In addition, this work indicates that items used to measure agency vary by context (Kabeer, 1999; Malhotra and Schuler, 2005; Mason and Smith, 2003). Thus, measuring many domains using context-specific indicators can provide a fuller picture of women’s agency.

Using this theoretical and empirical work as a starting-point, we developed a tool to measure women’s agency in rural India. Development of this measurement tool, composed of 23 indictors, is described elsewhere (Richardson et al., 2018). Briefly, we first conducted a review of the literature to identify relevant domains of agency, as well as indicators that measure these domains, in an Indian context. We identified 40 potential indicators that tapped into 5 distinct domains of agency. We assessed the relevance of these items in our study context and excluded some items that were not relevant. For example, participation in community groups, although a relevant indicator of women’s agency in South Asia, did not apply to our study population because most women lived in communities that did not have community groups to attend. Using the reduced list of indicators, we then tested a few competing hypothesized measurement models of women’s agency using confirmatory factor analysis (CFA), which is a method to test the construct validity of a proposed measurement model. Our final measurement model was composed of 23 indicators encompassing 4 domains. Model fit indices show that the measurement model fit the data well (Bentler Comparative Fit Index = 0.974, Tucker Lewis Index = 0.970, Root Mean Square Error of Approximation = 0.031).

Our measurement model summarized four agency domains: Household Decision-Making (9 questions; e.g., who makes decisions about health care for yourself?), Freedom of Movement (5 questions; e.g., are you allowed to go to the market in your village alone?), Participation in the Community (6 questions; e.g., do you feel comfortable speaking out against a man beating his wife?), and Attitudes and Perceptions (3 questions; e.g., a husband should help with chores if a wife is working). These questions, and the frequency of responses, are shown in Appendix 2. Because some items were dichotomous or ordinal, we used robust weighted least squares CFA, which is a type of CFA that can appropriately model this type of data (Kline, 2011). We accounted for correlated observations within hamlets by estimating standard errors clustered at the hamlet level. Using this model and fixing the variance of the latent variable Agency to 1, for each woman we calculated a summary score for each domain of agency, as well as an overall agency score (minimum score = -2.83, maximum score = 1.66). Appendix 3 shows summary statistics for these calculated scores. All measurement models were estimated in Mplus 7.4 (Muthen & Muthen, 1998–2015). Higher scores denote greater agency.

3.6. Statistical analyses

The primary analytic approach was intention-to-treat (ITT). ITT compares outcomes across village hamlets according to their randomized treatment assignment, regardless of how compliant hamlets were with their treatment. Additionally, the ITT analysis includes all women living in these village hamlets who completed baseline and follow-up interviews, regardless of if they used, or did not use, daycare. Thus, the ITT estimates the average effect of the offer of access to a community based daycare program on women’s mental health.

We used negative binomial regression to compare the mean difference in the number of mental distress symptoms (GHQ-12 score), measured at follow-up, among women living in village hamlets randomized to the affordable daycare program compared to women living in control villages. We used negative binomial models because the outcome (number of mental distress symptoms) is a count variable, and count variables are most appropriately modeled with Poisson or negative binomial regression (Hilbe, 2011). We chose negative binomial regression models over Poisson models because we found evidence of overdispersion, which can result in underestimated standard errors (Hilbe, 2011). We calculated the mean difference in the number of distress symptoms in models that adjusted for the stratification variable (block), as well as in models that adjusted for baseline covariates that may be predictors of mental distress (i.e., baseline mental distress score, age, household wealth, marital status) to increase statistical precision (Glennerster and Takavarasha, 2013).

We estimated the mean difference of our secondary outcomes with linear regression for work demands and agency, and with logistic regression for IPV. The partially adjusted models included an indicator for the stratification variable (block), and the fully adjusted models included baseline covariates that may be predictors of the outcomes. For models estimating work demands, baseline covariates included work amount, age, wealth, and number of daughters in the household; for models estimating IPV, baseline covariates included IPV exposure and educational attainment; and for models estimating women’s agency, baseline covariates included age, educational attainment, wealth, marital status, and age of marriage.

All models estimated robust standard errors to account for potential clustering of responses among women within the same hamlet. Analyses were conducted using Stata 14. A data monitoring committee did not oversee the study, which is standard practice in evaluations of social interventions with no clear risk of participant harm. The trial is registered at the IRSCTN trial registry, number IRSCTN45369145, and the American Economic Association’s registry, number AEARCTR-0000774.

4. Results

Fig. 1 shows participant selection and response rates. We identified 3899 potentially eligible women living in 160 village hamlets, and 343 women were interviewed and determined to be ineligible. Among the remaining 3557 potentially eligible women, 3177 women participated (response rate = 89%), and 3042 of enrolled women were re-interviewed approximately 1.5 years later (participation rate = 96%). Among the 4% of women who were lost to follow-up, we compared baseline differences in socio-demographic characteristics (age, marital status, age at first marriage, annual household income, wealth, education), number of hours worked, exposure to IPV, and mental distress score. We found some minor differences between the two groups; namely, women lost to follow-up were slightly younger (mean difference = 1.7 years, 95% CI: 0.50, 2.9) and were less likely to be married (mean difference = 2.2%, 95% CI: 0.0%, 4.3%).

Fig. 1.

Participant flow chart.

On average, women enrolled in the study were approximately 30 years old and had an average household income of 56,452 INR (approximately $880 USD). The majority of women were from scheduled tribes (93%), had never attended school (74%), were currently married or cohabitating (98%), and reported 2.1 out of a possible 12 distress symptoms. The majority of women reported some form of IPV (70%), most commonly controlling behaviour (60%), although many women reported psychological abuse (34%) or physical abuse (37%). Table 1 shows that the baseline characteristics were balanced by treatment assignment. Table 2 shows that daycare centers were open for an average of 16 days each month in the treatment communities. Approximately 41% of women in the treatment group utilized daycare, while 5% of women in the control group did. On average, women in the treatment group utilized daycare 2.2 days each week and women in the control group utilized daycare 0.2 days each week. Among women in the treatment group, women who utilized daycare were slightly younger, had more children in the household, and were more likely to be a member of a Scheduled Caste or Scheduled Tribe.

Table 1.

Selected baseline characteristics of 3177 rural Indian women.

	Control group (n = 1543)	Intervention group (n = 1634)
Socio-demographic
Age (years)	30.0 (6.8)	30.0 (6.8)
Caste
Scheduled Caste	22 (2%)	55 (4%)
Scheduled Tribe	1177 (92%)	1286 (94%)
Don’t know/none of them*	74 (6%)	29 (2%)
Married or cohabitating	1531 (99%)	1610 (99%)
Age of first marriage (years)	17.5 (2.7)	17.4 (3.0)
Annual household income in rupees	57214 (66485)	55732 (64067)
Wealth score	0.1 (1.4)	−0.1 (1.4)
Never attended school	1130 (73%)	1226 (75%)
Hours worked in past 24 h	9.7 (4.3)	9.7 (4.1)
Selected agency questions
Who makes decisions about health care for yourself?
Respondent only	211 (14%)	207 (13%)
Respondent involved in decision	715 (47%)	687 (43%)
Respondent not involved in decision	590 (39%)	691 (44%)
A husband should help with chores if his wife is working
Agree	1420 (93%)	1443 (89%)
Disagree	111 (7%)	180 (11%)
Can you go to a market in your village:
Alone	1366 (89%)	1452 (89%)
Not alone	158 (10%)	153 (9%)
Not at all	17 (1%)	26 (2%)
Who makes decisions about whether you can work?
Respondent only	228 (15%)	267 (17%)
Respondent involved in decision	747 (49%)	743 (47%)
Respondent not involved in decision	542 (36%)	584 (37%)
Reported any IPV	1060 (70%)	1094 (69%)
Mental distress
GHQ-12 score	2.2 (2.5)	2.1 (2.4)

Data are n, n (%), or mean (SD).

^*includes 46 women from an Other Backward Class.

Table 2.

Treatment utilization & compliance.

	Control group (n = 1486)	Intervention group (n = 1555
Balwadi use
No	1408 (94.8%)	917 (59.0%)
Yes	78 (5.2%)	638 (41.0%)
Number of days women typically used balwadi	0.2	2.2
Number of hours women typically used balwadi each day they used balwadi	0.3	2.5
Average number of days balwadi open each month	0 days	16 days

Access to the affordable daycare program resulted in modest reductions in mental distress among women living in these communities. Fig. 2 shows histograms of the unadjusted number of distress symptoms (GHQ-12 scores) stratified by treatment assignment. These graphs illustrate that the density of women reporting no distress symptoms was higher among women living in treatment communities. Table 3 reports the unadjusted and adjusted mean differences. In unadjusted models, treatment assignment resulted in a reduction of 0.22 (95% CI: −0.51, 0.07) distress symptoms. We observed virtually the same relationship in partially adjusted (−0.18, 95% CI: −0.40, 0.05) and fully adjusted models (−0.21, 95% CI: −0.43, 0.02), although adjustment increased statistical precision. In fully adjusted models the reduction in mental distress symptoms corresponded to an 11% decline (95% CI: −23.1%, 1.1%) relative to the mean.

Fig. 2.

Unadjusted GHQ-12 score by treatment assignment.

Table 3.

Intention-to-treat (ITT) estimates for the effect of offering affordable daycare.

	Control group	Intervention group	Mean difference	Mean difference in standard deviation units
Mental distress
Partially adjusteda	1.95 (1.81, 2.09)	1.77 (1.60, 1.95)	−0.18 (−0.40, 0.05)	−0.07 (−0.16, 0.02)
Fully adjustedb	1.98 (1.83, 2.13)	1.77 (1.60, 1.95)	−0.21 (−0.43, 0.02)	−0.08 (−0.18, 0.01)
Total work amount (hours)
Partially adjusteda	10.23 (9.96, 10.51)	10.12 (9.82, 10.42)	−0.11 (−0.52, 0.29)	−0.03 (−0.12, 0.07)
Fully adjustedc	10.24 (9.96, 10.52)	10.11 (9.82, 10.43)	−0.12 (−0.53, 0.30)	−0.03 (−0.12, 0.07)
Housework (hours)
Partially adjusteda	4.79 (4.64, 4.94)	4.78 (4.62, 4.93)	−0.01 (−0.23, 0.20)	0.01 (−0.09, 0.08)
Fully adjustedc	4.79 (4.65, 4.93)	4.78 (4.62, 4.93)	−0.01 (−0.22, 0.20)	0.01 (−0.09, 0.08)
Caring for children, elderly, disabled (hours)
Partially adjusteda	2.10 (1.96, 2.25)	1.95 (1.82, 2.08)	−0.16 (−0.35, 0.04)	−0.08 (−0.18, 0.02)
Fully adjustedc	2.11 (1.96, 2.26)	1.94 (1.82, 2.07)	−0.16 (−0.37, 0.03)	−0.08 (−0.18, 0.02)
Farm work (hours)
Partially adjusteda	3.00 (2.85, 3.15)	3.08 (2.91, 3.26)	0.08 (−0.15, 0.31)	0.04 (−0.07, 0.14)
Fully adjustedc	3.00 (2.85, 3.15)	3.10 (2.91, 3.27)	0.09 (−0.14, 0.33)	0.04 (−0.06, 0.14)
Paid work (hours)
Partially adjusteda	0.35 (0.24, 0.46)	0.33 (0.24, 0.42)	−0.02 (−0.16, 0.12)	−0.01 (−0.11, 0.08)
Fully adjustedc	0.35 (0.24, 0.46)	0.33 (0.24, 0.42)	−0.02 (−0.17, 0.12)	−0.02 (−0.11, 0.08)
Overall agency
Partially adjusteda	−0.13 (−0.17, −0.08)	−0.10 (−0.16, −0.05)	0.02 (−0.05, 0.10)	0.03 (−0.07, 0.13)
Fully adjustedd	−0.13 (−0.18, 0.08)	−0.10 (−0.16, −0.05)	0.02 (−0.05, 0.10)	0.04 (−0.06, 0.14)
Household Decision-Making
Partially adjusteda	−0.03 (−0.05, −0.01)	−0.01 (−0.04, 0.02)	0.03 (0.01, 0.05)	0.01 (−0.02, 0.04)
Fully adjustedd	−0.03 (−0.05, −0.01)	−0.01 (−0.04, 0.02)	0.02 (−0.02, 0.06)	0.05 (−0.04, 0.13)
Freedom of Movement
Partially adjusteda	−0.17 (−0.21, −0.12)	−0.15 (−0.20, −0.10)	0.02 (−0.05, 0.09)	0.03 (−0.08, 0.13)
Fully adjustedd	−0.17 (−0.22, −0.12)	−0.15 (−0.20, −0.10)	0.02 (−0.05, 0.09)	0.03 (−0.07, 0.14)
Participation in the Community
Partially adjusteda	−0.02 (−0.08, 0.04)	−0.02 (−0.09, 0.04)	0.00 (−0.10, 0.09)	0.00 (−0.10, 0.09)
Fully adjustedd	−0.03 (−0.09, 0.04)	−0.02 (−0.08, 0.05)	0.01 (−0.08, 0.10)	0.01 (−0.09, 0.10)
Attitudes and Perceptions
Partially adjusteda	−0.07 (−0.09, −0.05)	−0.06 (−0.07, −0.04)	0.01 (−0.01, 0.04)	0.05 (−0.04, 0.15)
Fully adjustedd	−0.07 (−0.09, −0.05)	−0.05 (−0.07, −0.04)	0.02 (−0.01, 0.04)	0.07 (−0.03, 0.16)
Any intimate partner violence
Partially adjusteda	0.78 (0.75, 0.80)	0.75 (0.71, 0.78)	−0.03 (−0.08, 0.02)	−0.07 (−0.18, 0.04)
Fully adjustede	0.78 (0.75, 0.80)	0.75 (0.71, 0.78)	−0.03 (−0.07, 0.01)	−0.07 (−0.17, 0.03)
Physical abuse
Partially adjusteda	0.27 (0.24, 0.30)	0.26 (0.23, 0.29)	−0.01 (−0.05, 0.03)	−0.03 (−0.12, 0.06)
Fully adjustede	0.27 (0.24, 0.30)	0.26 (0.23, 0.28)	−0.01 (−0.05, 0.03)	−0.03 (−0.12, 0.06)
Psychological abuse
Partially adjusteda	0.32 (0.29, 0.35)	0.28 (0.25, 0.31)	−0.04 (−0.08, 0.01)	−0.08 (−0.18, 0.01)
Fully adjustede	0.32 (0.29, 0.35)	0.28 (0.25, 0.31)	−0.04 (−0.08, 0.01)	−0.08 (−0.18, 0.01)
Controlling behaviour
Partially adjusteda	0.72 (0.69, 0.75)	0.67 (0.64, 0.71)	−0.05 (−0.10, 0.00)	−0.10 (−0.21, 0.00)
Fully adjustede	0.72 (0.69, 0.75)	0.67 (0.64, 0.71)	−0.05 (−0.10, 0.00)	−0.11 (−0.21, 0.00)

^aAdjusted for stratification variable (block).

^bAdjusted for baseline mental distress score, age, household wealth, marital status, and block.

^cAdjusted for baseline work amount, age, household wealth, number of girls in household, and block.

^dAdjusted for baseline age, educational attainment, household wealth, marital status, age of marriage, and block.

^eAdjusted for baseline IPV exposure, educational attainment, and block.

We found modest reductions in the secondary outcomes IPV and work demands, but virtually no change in agency (Table 3). In fully adjusted models, we found a 3 percentage point decrease (95% CI: −7, 1) in exposure to IPV among women living in treatment compared to control hamlets, which was driven primarily by a decrease in partner controlling behaviour (decrease = 5 percentage points, 95% CI: −10, 0) and psychological abuse (decrease = 4 percentage points, 95% CI: −8, 1). We found virtually no reduction in women’s total work amount (adjusted mean difference = −0.12 h, 95% CI: −0.53, 0.30), although we found a slight reduction in the amount of time spent caring for children, the elderly, and the disabled (adjusted mean difference = −0.16 h, 95% CI: −0.37, 0.03). We did not find any meaningful changes in either women’s overall agency (adjusted mean difference = 0.03, 95% CI: −0.05, 0.10), nor its constituent parts.

5. Discussion

There is a dearth of research on structural factors affecting mental health, and our study evaluated the effect of one potential structural factor, access to affordable daycare. We found that access to daycare led to modest reductions in women’s mental distress. Affordable daycare may be one structural factor that can improve population mental health.

To our knowledge, this is the first randomized trial to evaluate the effect of access to affordable daycare on women’s mental health. Quasi-experimental studies have found mixed results (Ángeles et al., 2011; Baker et al., 2008; Rosero and Oosterbeek, 2011), and thus our study adds important information on this topic. These differences in results may be due to two reasons. First, although quasi-experimental study designs can provide strong evidence, they are nevertheless still susceptible to confounding by unmeasured factors, which may lead to biased study results. Our experimental study randomized participants to treatment assignment – which, in expectation, balances measured and unmeasured confounders between treatment groups – and thus can provide stronger evidence and reduced risk of confounding.

Second, these quasi-experimental studies were conducted in very different settings than our study (i.e., Quebec, Mexico, Ecuador), and access to affordable daycare might affect mechanisms linking daycare with mental health differently in different contexts. For example, employment is one hypothesized variable linking daycare with mental health (either positively by increasing women’s agency, or negatively by increasing women’s work burden). In Quebec, access to subsidized daycare increased women’s labour force participation (Baker et al., 2008), while in our study access to daycare led to virtually no change in paid work amount. In our study context, there are limited economic opportunities, which may cap daycare’s effect on paid employment. However, in Quebec daycare has the potential to lead to sizeable increases in women’s labour force participation due to far more economic opportunities. The heterogeneous effects of daycare on employment may help explain why the Quebec study found that access to daycare increased mothers’ depressive symptoms (potentially through an increase in overall work burden), while our study found that access to daycare led to a reduction in mental distress (potentially by changing work demands and reducing exposure to IPV). Thus, the mixed results in the literature may reflect true heterogeneous effects.

We found that access to affordable daycare corresponded to an 11% (95% CI: −23.1%, 1.1%) decrease relative to the mean number of distress symptoms, or a reduction of 0.08 (95% CI: −0.18, 0.01) standard deviation units. These effect estimates are of similar magnitude to other social interventions in LMICs that investigate mental health outcomes. For instance, a randomized controlled trial of a multifaceted livelihood intervention among the very poor conducted in 6 countries (including India) found a 0.10 standard deviation improvement in mental health score (Banerjee et al., 2015).

We found some evidence that daycare reduced IPV, but did not substantially affect work demands or women’s agency. IPV is consistently associated with poor mental health, and non-physical aspects of IPV, such as psychological abuse and controlling behaviour, have emerged as important predictors of poor mental health (Lagdon et al., 2014). Thus, the reductions we observed in psychological abuse (4 percentage point reduction, 95% CI: −8, 1) and controlling behaviour (5 percentage point reduction, 95% CI: −10, 0) might have contributed to the decrease in mental distress observed in our study. In addition, we found slight reductions in the amount of time women reported caring for children, the elderly, or disabled (−10 min, 95% CI: −22, 2). These shifts in work demand patterns could contribute to reductions in mental distress; prior work in this study population indicated that high amounts of care work are associated with greater mental distress (Richardson et al., 2017). Taken together, our study provides some indication that many mechanisms working together might lead to reductions in mental distress. However, it should be noted that our secondary outcomes are interrelated with mental distress, and our study was not able to distinguish if the changes in secondary outcomes led to reductions in mental distress or if reductions in mental distress led to changes in secondary outcomes.

We did not find compelling evidence that daycare increased women’s agency, although all agency effect estimates were positive, indicating that women randomized to daycare had very slightly higher agency scores in all agency domains. Increasing women’s agency is a transformational process that may take many years to come to fruition. Thus, the relatively short follow-up time in our study (approximately 1 year post intervention) might not have been a long enough period to affect substantial change.

The daycare model evaluated in our study was developed by our partner organization, Seva Mandir, and thus is not directly comparable to the anganwadis and crèches sponsored by the Indian government. However, our results do demonstrate that consistent access to affordable daycare might lead to reductions in mental distress among Indian mothers, which could inform current policy debates. Recent legislative efforts are scaling up access to daycare; in 2017, the Indian government enacted the Maternity Benefit Act, which requires all employers with more than 50 employees to provide daycare to children between the ages of 6 months and 6 years. Scale-up of these services may have unintended, positive consequences for mothers’ mental health throughout India.

Our study has a number of strengths, including detailed measures of women’s work burden and agency, a relatively large sample size, random allocation to daycare, high participation rates, and low loss to follow-up. However, our study has some caveats. First, we measured mental distress with the GHQ-12. Although the GHQ-12 has strong psychometric properties among Indian adults (Patel et al., 2008), it was initially developed for a European population. Qualitative research indicates that many Indian women experience symptoms of distress as physical complaints such as body aches, gynecological symptoms, weakness, and tiredness (Pereira et al., 2007), which are not captured with the GHQ-12. Thus, the GHQ-12 may miss symptoms of distress in this population, which may partially explain the relatively low average number of distress symptoms in this population. Second, utilization of daycare was moderate (less than half of women who had access to daycare used it), and therefore higher enrollment might correspond to more pronounced treatment effects.

Our study has additional limitations. First, the daycare program evaluated in this study was developed by an NGO that has been working in these communities for many years, and thus the model of daycare evaluated in this study may not be reproducible in other settings. Second, the effect of daycare on women’s mental health is likely contextual, and thus our results may not be generalizable to other contexts. Third, our time use survey did not capture other aspects of time use that may be relevant, such as working at a more leisurely pace or performing one task at a time, both of which are associated with better well-being among women (Floro and Pichetpongsa, 2010). Our study was not able to detect these work patterns, and thus the modest changes in work patterns observed in this study might belie greater shifts in women’s work demands. Fourth, our effect estimates had some degree of uncertainty. Therefore, although our study indicates daycare might be beneficial to women’s mental health in this context, these results should be interpreted cautiously.

In conclusion, our study found that access to affordable daycare led to modest reductions in women’s mental distress. These results offer evidence that expanded access to daycare might lead to improvements in population mental health. Future research in other contexts, as well as replication studies in India, would add to and potentially confirm results.

Appendix 1. Development of wealth index

We measured household wealth with 23 asset-based indicators that are commonly used to measure wealth in LMICs (Filmer and Pritchett, 1999). These indicators included housing characteristics (i.e., type of toilet facility, material of exterior wall, type of roofing, home electrification, source of drinking water), the number of durables owned (i.e., number of cell phones, watches/clocks, electric stoves, wood stoves, fans, televisions, bikes, motorcycles, wells, grain storage cans, pressure cookers, chairs/stools, beds, silver jewelry, gold jewelry, wedding ornaments), home ownership, and whether the household had a savings account. We created a summary wealth score using a polychoric principle component analysis (PCA), which is a type of PCA that can appropriately model ordinal variables (Kolenikov and Angeles, 2004). We used the Stata user-written command -poly-choricpca-, which uses the factor loadings from a polychoric PCA to estimate a wealth score for each individual. We used a one component PCA that explained 26% of the variance.

Appendix 2.

Percent answering affirmatively to women’s agency items at follow-up

Decision making in the home and control over income	Home decisions	Who usually makes the following decisions:	n	Respondent only	Jointly with other family members	Respondent not involved
		Decisions about health care for yourself?	2996	4%	61%	35%
		Decisions about how many children to have and when?	2994	2%	89%	9%
		Decisions about whether to use contraception?	2967	3%	87%	10%
		Decisions about the education of your children, including where they go to school and until which grade?	2991	4%	78%	18%
		Decisions about visits to your family or friends?	2994	4%	75%	21%
		Decisions about whether you can work?	2996	23%	55%	22%
		Decisions about where you can work?	2994	19%	56%	25%
	Control over income	Decisions about making major household purchases?	2996	1%	70%	29%
		Who decides how your husband’s earnings will be used?	2994	2%	82%	16%
Attitudes and perceptions		Do you agree or disagree with each statement:	n	Agree	Disagree
		Husband should help with chores if wife is working	3040	95%	5%
		A married woman should be able to work outside the home if she wants to	3031	93%	7%
		A wife has a right to express her opinion even if she disagrees with what her husband is saying	3033	93%	7%
Freedom of movement		Are you usually permitted to go to the following places on your own, only if someone accompanies you, or not at all?	n	Alone	Not alone	Not at all
		To the market to buy things?	3041	91%	8%	1%
		To a health center or doctor within the village?	3041	89%	11%	0%
		To the community center or other meeting place within the village?	3041	88%	12%	1%
		To homes of friends in the village?	3041	90%	9%	1%
		To a shrine/mosque/temple/church within the village?	3041	90%	10%	0%
Participation in the community	Family-related issues	Do you feel comfortable speaking up in public to:	n	No, not at all comfortable	Yes, but with a great deal of difficulty	Yes, but with a little difficulty	Yes, fairly comfortable	Yes, very comfortable
		Protest a man beating his wife?	3039	17%	9%	14%	41%	19%
		Protest a man divorcing or abandoning his wife?	3039	18%	9%	16%	39%	18%
	Outside home-related issues	Help decide on infrastructure (like small wells, roads, water supplies) to be built in your community?	3039	21%	10%	16%	35%	17%
		Ensure proper payment of wages for public works or other similar programs?	3038	12%	8%	15%	40%	26%
		Protest the misbehaviour of authorities or elected officials?	3038	22%	11%	20%	35%	13%
		Do you feel comfortable attending rural meetings unaccompanied?	3039	35%	8%	11%	30%	17%

Appendix 3.

Summary statistics for derived women’s agency scores at follow-up

	Mean	Standard deviation	Minimum	Maximum
Overall agency score	−0.12	0.73	−2.83	1.66
Household Decision-Making	−0.02	0.46	−1.57	1.85
Freedom of Movement	−0.16	0.65	−3.05	1.13
Participation in the Community	−0.02	0.94	−2.14	1.58
Attitudes and Perceptions	−0.06	0.27	−1.09	0.41