The Combined Role of Subsidy and Discussion Intervention in the Demand for a Stigmatized Product

Abstract

This paper studies the joint role of subsidization and group discussion intervention in increasing the demand for sanitary pads—a product that is widely available but demand for which may be curtailed due to the psychological cost associated with menstrual stigmatization. The study deploys a field experiment in Nepal to randomly allocate discount coupons of various values so that participants face exogenous variation in the effective price of sanitary pads. In addition, a randomly selected group of women in the sample participate in menstrual-health-related group discussion intervention. The findings suggest that an increase in subsidy level increases the probability of adoption across both groups of women—those receiving only a subsidy and those participating in the discussion intervention coupled with a subsidy. Also, women participating in the discussion intervention have a higher adoption rate. The effects of group discussion intervention are concentrated among women with high psychological cost, whose purchase decisions are more likely to be affected by societal stigma. The results suggest that combining a subsidy with group discussion could provide a cost-effective strategy to increase the adoption of health technology, the demand for which is constrained by social norms.

1. Introduction

Many areas of health are governed by traditional beliefs and taboos, which can affect the demand and utilization of health products and services due to societal stigma.¹ Unlike cases where lack of information causes individuals to underestimate the marginal benefit of health products (Luby et al. 2004, 2005; Cairncross et al. 2005; Jalan and Somanathan 2008; Madajewicz et al. 2007; Dupas 2011), stigmatization can limit the adoption of even well-known health products by giving rise to psychological cost.² Consequently, subsidies that have proved successful in increasing take-up of health products (Ashraf, Berry, and Shapiro 2010; Meredith et al. 2013; Dupas 2014b) may not be as effective in the case of products associated with taboo issues. Pairing subsidies with an additional intervention designed to reduce the underlying psychological cost may be necessary. Public-health research indicates that providing a platform for group discussion and information sharing can be effective in reducing stigma-related psychological costs (Brown and Bradley 2002; Creel et al. 2011; Johnston-Robledo and Chrisler 2013; Henderson et al. 2017; Bobel et al. 2020).

In this paper, we report results from a randomized controlled trial conducted to evaluate the relative effectiveness of subsidy-only intervention against a combined intervention that pairs a subsidy with group discussion (subsidy-plus-discussion intervention) in increasing the demand for disposable sanitary pads. In many developing countries, as in our study setting of Nepal, menstruation is highly stigmatized due to deeply rooted sociocultural norms and religious beliefs. Despite sanitary pads being fairly well known and commonly available, they are not universally adopted. Rags, cotton, or wool are used instead (Budhathoki et al. 2018), which if improperly used can be detrimental to one’s immediate health and long-term well-being.³  Sommer et al. (2015) note that a combination of high cost, budget constraints, and menstrual stigma could explain poor menstrual-health management and the low adoption rate of sanitary pads. While menstrual-health products themselves are not taboo, societal stigma around menstruation creates product stigma (Klintner 2021).

Stigma can inhibit demand for menstrual products regardless of a woman’s knowledge about the product’s benefits. For those unfamiliar with the product, stigma creates a social environment that discourages them from gathering relevant information by discussing the matter with peers. Learning from peers is an important step in the adoption of new health products (Oster and Thornton 2012; Meredith et al. 2013), a channel that is curtailed for products associated with a stigma. Even for those with adequate information, menstrual stigma may dissuade them from taking the public action of purchasing the product from a local shop due to the shame or embarrassment connected to “unmentionables” or “controversial products” (Wilson and West 1981). Stigma leads to concern about social image or status, which is exacerbated when the action is public (Jensen 2006; Bursztyn and Jensen 2017). In our baseline survey, 32 percent of non-regular users of sanitary pads cite being uncomfortable during purchase as a reason for non-frequent use, while 23 percent cite price as the primary obstacle.⁴ Thus, in addition to budget constraints, the under use of sanitary pads can to some extent be attributed to the psychological costs created by stigma. Designing effective policies aimed at increasing the adoption of sanitary pads therefore requires an understanding of the relative importance of price and psychological costs.

We hypothesize that a reduction in psychological costs created by the stigma, achieved through participation in group discussion intervention, will make women more responsive to subsidies. To test this, we create two treatment groups—the first group receives a subsidy of a random value (“subsidy-only treatment”) and the second group receives a subsidy as well as an invitation to participate in a group discussion intervention session (“subsidy-plus-discussion treatment”). The discount coupon can be redeemed at local pharmacies and take one of five values (10 percent, 25 percent, 50 percent, 75 percent, or 90 percent).⁵ The random variation in price allows us to trace the demand curve. In the group discussion session, to which half of the study participants are invited, a health professional provides information about women’s health, including but not limited to menstrual hygiene and related social beliefs, as well as discussion of the stigma surrounding menstruation. Using information on coupon redemption, we estimate the demand curves and the price elasticities of demand for sanitary pads across the subsidy-only and subsidy-plus-discussion intervention groups.

Two features of our group discussion intervention need to be emphasized. First, it is not intended as an advertisement campaign for the product. Indeed, given that sanitary pads are widely known (85 percent of women in our sample already know about sanitary pads), the advertisement component of this treatment, if any, is likely to be small. The primary objective of the discussion intervention is to help reduce participants’ psychological costs of taking an action they consider to be stigmatized. Studies have argued that lack of discussion regarding menstruation creates a culture of shame (Johnston-Robledo and Chrisler 2013; McHugh 2020). Therefore, normalizing conversations around menstruation can be effective in lowering the internal guilt or perceived risk of social ostracization (Klintner 2021; Molina et al. 2021). As such, we focus on group discussion intervention in contrast to the door-to-door campaign done by Ashraf, Jack, and Kamenica (2013) and Meredith et al. (2013). The intervention creates an opportunity for women to openly discuss menstrual-health issues, including the norm of stereotyping and the labeling of menstruating women as “impure” with an associated stigma. Second, we do not expect the discussion treatment to immediately reduce deeply rooted menstrual stigma in the society, which is very difficult to do (Link and Phelan 2001; Gronholm et al. 2017). While the discussion intervention could have improved the adoption of menstrual-health products through a reduction in societal stigma, this is not necessary. As long as it is able to reduce the psychological costs arising due to menstrual stigma, women in the subsidy-plus-discussion treatment arm can use (redeem) sanitary pads in greater numbers even when the society’s menstrual stigma remains unchanged.⁶

One estimation challenge is that we only have 67 percent compliance in our discussion treatment. While the assignment of households to various treatment arms was done beforehand,⁷ discount coupons were distributed to participants at the end of the discussion session. Therefore, non-compliers did not receive their assigned discount coupons.⁸ As we argue in detail in a later section, this does not impact the estimation of the price elasticities of demand separately for the subsidy-only and subsidy-plus-discussion groups (assuming a linear demand curve) because the allocation of the discount is still random within the two groups and compliance is unrelated to discount levels. The estimates of the difference between the elasticities of the two groups could be biased due to self-selection. To account for the issue of non-compliance in our main results, we predict the propensity to comply (using the subsidy-plus-discussion treatment sample) and correct for selection by including the probability of compliance as a control variable. In other methods, we adopt Heckman’s two-step approach (Heckman 1976) and conduct multiple imputations on missing redemption values among non-compliers using compliers with a similar compliance probability and discount level. The main results are robust to these alternative empirical strategies; detailed discussion is provided in the Results section and supplementary online appendix S1. Despite our attempt to rectify the potential bias due to non-compliance, we note this to be an important caveat to the results.

The findings suggest that the use of disposable sanitary pads increases as the after-discount price falls across both the subsidy-only and the subsidy-plus-discussion intervention groups. The effects within the subsidy-only group are larger when moving from the 10 percent to 25 percent and 50 percent to 75 percent discount levels, where the adoption rate increases by more than 25 percentage points. Women participating in discussion interventions have a higher adoption rate at the respective subsidy levels in comparison to the subsidy-only group and the adoption rate within this group is largest when moving from the 10 percent to 25 percent and 25 percent to 50 percent discount levels. However, even within the subsidy-plus-discussion intervention group, the adoption rate is only 70 percent at the highest discount level (90 percent discount). This finding is in sharp contrast to those products unaffected by societal stigma and not solely designed for females (e.g., bednets, water purification technology, rubber shoes to prevent hookworm infection), for which take-up is much higher at low market prices (Ashraf, Berry, and Shapiro 2010; Meredith et al. 2013; Dupas 2014b).

If the impact of participating in discussion intervention resulted in the reduction of psychological costs associated with stigma, we should find a larger impact on women with higher (indirectly measured) psychological costs. We check this supposition by interacting the subsidy level and discussion treatment indicator with proxies of baseline psychological costs. The results show that women in the subsidy-plus-discussion treatment group with high psychological costs were more affected by the combined intervention—the demand curve for the women with high psychological costs is shifted more to the right. Overall, our results indicate that while budget constraints are an important determinant of demand for sanitary products, pairing subsidies with discussion about overcoming psychological costs imposed by social norms could be even more effective in boosting the adoption rate of these hygiene products.

One possible confounding factor affecting the interpretation of our result is the difference in how we distribute discount coupons to the two treatment groups. If distribution of coupons in the subsidy-plus-discussion treatment arm allows participants to compare the level of discounts each receives, it may cause resentment among those with less of a discount. Such comparison would suppress demand at higher price levels, making it harder to determine the impact of the discussion intervention at the lower discount levels. However, we find an adoption rate of 19 percent even at the lowest discount (10 percent) for participants in the subsidy-plus-discussion group, so this channel does not seem to be important. Another confounding factor is that the group setting may exaggerate the effectiveness of the discussion treatment due to the presence of peer effects. If the peer effects are comprised only of social spillover related to menstruation, we consider this to be a part of the mechanism through which discussion intervention should affect psychological costs. However, the group setting may have allowed peers to share information unrelated to menstrual health (e.g., the location of pharmacies where coupons can be redeemed). If this is the case, the effects of discussion intervention may be overstated. We argue that this is unlikely to be a major issue, as necessary information regarding the logistics of redeeming the coupon was clearly provided to all participants by the research team. Priming may also explain some of the differences in redemption rates between the subsidy-plus-treatment and subsidy-only groups (Haaland, Roth, and Wohlfart 2020). Actively inducing participants to discuss menstruation may influence them to redeem coupons by bringing the issue to the forefront of their cognition. While this is certainly possible, priming alone would not account for our findings that women in the subsidy-plus-discussion treatment group with high psychological costs are more responsive to price subsidies.

Our paper contributes to the literature studying the adoption of health products by highlighting the importance of pairing group discussion intervention with price subsidies to increase the demand for those health products whose use is curtailed by stigma. Past studies have found that welfare-enhancing technologies have suboptimal adoption in many settings, which may be attributed to an individual’s valuation of health or the attributes of the technology itself (Kremer and Miguel 2007; Foster and Rosenzweig 2010). A few studies evaluate the combined effects of subsidies and information interventions, with mixed findings. Ashraf, Jack, and Kamenica (2013) find that providing information substantially increases the effect of price subsidies on demand for an unfamiliar water purification technology. In contrast, Meredith et al. (2013) find that information or an increase in knowledge did not affect the purchasing decision for shoes, which prevent soil-transmitted helminths in Kenya, Guatemala, India, and Uganda. However, these results do not speak to situations where societal stigma influences the adoption of health technology and thus imposes psychological costs. Given that social norms and perceptions are considered to be an important driver of economic behavior (Akerlof and Kranton 2000; Bisin and Verdier 2011; Moffitt 1983; Polinsky and Shavell 2000), they should be incorporated in our understanding of the demand for health products.

Our finding, that discussion intervention combined with subsidization of products can increase demand more effectively than only providing a subsidy, has implications for devising cost-effective strategies to increase the usage of health products in situations where societal stigma reduces demand. Some previous academic studies directly distribute menstrual hygiene products and assess the outcome (Oster and Thornton 2012; Benshaul-Tolonen et al. 2021), but do not consider the role of psychological costs weighing down on the demand for such products. On the policy front, price reductions of menstrual products have been implemented. In India, a scheme called Yojana Ranjit, aimed at providing subsidized sanitary napkins to both girls in rural areas and those attending district council schools, was launched in March 2018 (Joshi 2018). Likewise, the Indian Tax Council announced a complete reduction of the tax rate on sanitary napkins from 12 percent to 0 percent in order to increase usage (Singh 2018). The Kenyan government repealed the sales tax on sanitary pads and tampons in 2004, which was followed by budgeting in 2011 that allocated $3 million each year towards sanitary pads for schoolgirls in relatively poor communities (Geertz, Lakshmi Iyer, and Francesca 2016). Recently, the Nepalese government allocated a budget to distribute free sanitary pads in government schools (Free Sanitary Pads for Public Schools 2019). Our results indicate that it may be more cost effective to provide relatively lower levels of a subsidy coupled with a program of group discussion, rather than providing high levels of subsidy without complementary discussion programs.

Finally, interventions usually provide menstrual products to school-age girls (Oster and Thornton 2011) and such interventions may have a limited effect as girls do not have much control over household expenditure. Instead, by focusing on adult females who have comparatively more decision-making power (although perhaps not total control over their finances), we find large effects with program interventions. Given the evidence of intergenerational transmission of knowledge, greater adoption of products by mothers is likely to have spillover effects on the sanitary-pad usage of their daughters and may help to remove a significant impediment to schooling caused by poor menstrual hygiene (Dolan et al. 2014; Montgomery et al. 2016; Benshaul-Tolonen et al. 2021).

The rest of the paper is organized as follows. The study begins with the experimental design, followed by the estimation strategy. Next, the discussion of the results are presented after which concluding remarks are provided. Supplementary online appendix S1 provides additional robustness exercises and supplementary online appendix S2 presents a simple model that illustrates how the presence of psychological cost affects women’s response to subsidization and the role of discussion intervention.

2. Experimental Design

Our study was conducted in five villages of the Bidur municipality of Nuwakot district, located 61 kilometers northwest of Nepal’s capital, Kathmandu.⁹ We began by administering a baseline questionnaire to all eligible households. In addition to a household’s demographic and socioeconomic characteristics, one female aged 15–50 from each household was asked questions about their menstrual health.¹⁰ The questionnaire also included several questions related to the societal stigma or norms associated with menstruation. We collected a total of 707 observations in the sample, although a few have missing information on some key variables and were therefore not used.

2.1. Subsidy Treatment

Our subsidy treatment closely resembles those implemented by Ashraf, Berry, and Shapiro (2010), Cohen and Dupas (2010), and Kremer and Miguel (2007). All households were randomly assigned a discount coupon for either 10, 25, 50, 75, or 90 percent for a specific brand of sanitary pad.¹¹ The price of sanitary pads is Rs. 70 ($0.62) for a pack containing six pads. Women normally use 3–4 packs per cycle, which means the total expenditure at the full price would be Rs. 210–280 ($1.85–$2.47). The majority of women in our sample belonged to a family with a monthly income of less than Rs. 25,000 (see the summary statistics of the baseline characteristics in table 1), so the expenditure on sanitary pads would represent over 3 percent of the household’s monthly income. The discount coupon could be used to purchase up to five packs of sanitary pads, enough for one cycle.

The coupon had to be redeemed within 45 days at one of two local pharmacies. These pharmacies were conveniently located at the main marketplace, allowing easy access from all villages in the study. The average distance from the five experiment villages to the marketplace where the pharmacies were located is approximately two kilometers. The pharmacies also held an adequate stock of sanitary pads and had qualified staff to keep detailed logs of the number of pads that were sold. The pharmacies were reimbursed for the expected number of redemptions at the list price of the product in advance, with a stipulation that the sales would be recorded accurately and the final tally returned to the researchers. To ensure that these pharmacies reported accurately, they were required to collect the discount coupons and redeemers’ signatures, in addition to maintaining a logbook. Research assistants corroborated the registration in the logbook by using discount coupons; in this way the discount coupons also acted as receipts for the transactions. This gives us confidence that the pharmacies did not have an incentive to misrepresent the sales.

An important point to highlight is that participants did not know about the subsidy treatment and their assigned level of subsidy in advance. For the subsidy-plus-discussion treatment group, coupons were distributed at the end of the discussion session according to the pre-assigned discount levels. Thus the incentive to attend the event did not vary by assigned discount rate. While the discussion intervention was taking place, individuals in the subsidy-only group within the community received their assigned discount coupons.

2.2. Subsidy-plus-Discussion Treatment

We paired the subsidy treatment with a discussion treatment in order to address the psychological costs due to menstrual stigma. While subsidy-only treatment has proved successful in many cases, it may be less effective for stigmatized products like sanitary pads. Furthermore, in general, the size of subsidies necessary to induce meaningful adoption are found to be very large (Dupas 2014a), which could be partly due to non-pecuniary factors inhibiting demand. If it is possible to identify and influence these factors, a smaller amount of subsidization might be sufficient to achieve the targeted adoption rate.

Using computer-based randomization, participants were allocated into either a subsidy-only or subsidy-plus-discussion group after stratifying on whether they had received menstruation-related awareness through NGOs in the post-earthquake era (pre-treatment).¹² We invited the households assigned to receive the discussion treatment to attend the session. The households were given a physical invitation card, which they had to bring to the program so that we could track attendance. The discussion intervention was organized at a local school closest to the study site and lasted for three hours.¹³ The intervention was administered by four female health workers from Kathmandu and a nurse from the Nuwakot district hospital.

The design of the group discussion intervention was motivated by the notion that while information about sanitary pads is widespread, menstrual stigma might be affecting their widespread use. Borrowing from the public-health literature, we focused on “normalizing” conversations regarding menstruation by fostering an environment to promote discussions.¹⁴ Open discussion about menstruation is considered inappropriate in many societal settings (Lee and Sasser-Coen 1996; Houppert 1999). A group intervention, as opposed to disseminating information individually, is more likely to achieve normalization. Interpersonal communications that take place between participants in a group setting are also likely to generate peer support, which is important in stripping away any shame associated with a stigmatized action or product (Tomori et al. 2014; Parikh et al. 2018; Aghaei, Mohraz, and Shamshirband 2020).

To stimulate discussion, the presenters highlighted how menstrual stigma lowered female well-being. This included (but was not limited to) (a) the social tradition of monthly isolation during menstruation, usually in the shed (Chaupadi); (b) stereotypical beliefs that menstruating females are “unclean” or “impure,” which aids the persistence of conventional traditions such as restricting menstruating women from entering the kitchen or places of worship and from touching male members of the household (temporarily considered “untouchable”); and (c) the role of stigma in suppressing conversations around menstruation. It was also discussed that proper menstrual-health management is a human right. To provide evidence of changes occurring at the legislative level, the participants were informed that the act of Chaupadi (the tradition of monthly isolation) was criminalized in 2017. Throughout the session, the participants interacted by sharing their personal experiences regarding barriers to proper menstrual-health management and issues that arise during their menstrual cycle.¹⁵

The discussion session also provided information about general female health and hygiene, including but not limited to menstrual hygiene. This included discussions regarding women’s health concerns related to pregnancy, uterine prolapse, and breast cancer. We focused on providing general information about female health rather than explaining the feature of the product itself, as sanitary pads were already very well known among our sample. Therefore, we do not believe that our discussion intervention provided any meaningful advertisement cues about the product itself.

We do not have any evidence on whether the stigma surrounding menstruation was actually diminished by our intervention, but given what we know about the evolution of social norms, this is highly unlikely. Gronholm et al. (2017) note that societal stigma is a complex phenomenon and that short-term interventions tend to have only short-lived, if any, impact. Furthermore, given that men are usually responsible for policing adherence to social norms, our intervention, which targeted women, would have a limited impact on the overall stigma at best. However, the intervention likely reduced the psychological costs produced by the stigma, which may make women more likely to adopt sanitary pads.

2.3. Baseline Characteristics and Randomization Check

The summary statistics for the variables from the baseline survey are presented in table 1. Almost 80 percent of the women in our sample are Hindus. The average age in the sample is 30 years and the average education level is the seventh grade. Seventy percent of the sample have a household income below Rs. 25,000. As with many communities in Nepal, traditional norms inform ideas around menstruation among many women in our study location. Menstruating women are colloquially referred to as becoming (temporarily) “untouchable,” as reported by over one-third of the respondents. In most cases, menstruating women in the community are prohibited from touching other household members and entering the kitchen or religious spaces; about 60 percent of our respondents were not permitted to enter their kitchen during menstruation. Some communities believe that menstruating women can cause bad luck, resulting in family illness or harvest failure. As such, a form of Chaupadi practice includes forcing them to sleep outside the house (in a shed, usually with livestock), which exposes them to dangers including death due to snake bites and asphyxiation.¹⁶ In our sample, over 56 percent of the women reported being kept in the shed during their period. Knowledge of menstruation was also not universal; while 74 percent of respondents reported menstruation as a physiological process, only 53 percent of individuals correctly identified the uterus as its source. Regarding use of sanitary pads, while over 75 percent of women reported ever using a sanitary pad, only 41 percent reported using them on a regular basis and 56 percent used them during their last menstruation.¹⁷ Clearly, there is still room for improvements in the usage of sanitary pads in the study locations.

There are in total 10 treatment arms in this study, with 5 different discount levels across the subsidy-only and subsidy-plus-discussion treatment groups. For all estimations, the omitted category in the regressions is comprised of those women who were assigned a 10 percent discount and were not invited to the discussion sessions.¹⁸ We check for balance across these treatment arms by regressing various baseline covariates on all interactions between the discussion treatment and discount-coupon-level indicators.¹⁹ The results are reported in tables 2 (for main demographic characteristics) and 3 (for menstrual-health-related variables). In general, we find that the covariates are balanced across all dimensions of randomization. Among estimates of 126 (14 variables × 9 coefficients) regressors, 6 estimates are statistically significant (3 at 10 percent and 3 below 10 percent levels), which is likely by pure chance. Furthermore, we present at the bottom of each table the F-statistics pertaining to the null hypothesis under the restrictions that the coefficients on the interaction terms between discount coupon indicators and discussion intervention treatment group are jointly equal to zero. Based on the F-statistic for each respective variable, we are unable to reject the null hypothesis. This provides further evidence regarding the proper implementation of treatment assignments.²⁰

2.4. Compliance

Participation in the discussion session was voluntary and therefore not universal; 67 percent of women who were assigned to the discussion treatment group attended the event. Table 4 reports the number of individuals in different randomization bins, with those in the subsidy-plus-discussion treatment group further divided into compliers and non-compliers. Non-compliance may pose selection problems and affect the interpretation of the treatment effect (Ye et al. 2014). For our study, an additional issue is that the non-compliers did not receive their assigned discount coupon at all since coupons were distributed at the end of the discussion session (to reiterate, for those in the subsidy-only group, coupons were distributed door to door while the discussion intervention program was ongoing).²¹ Hence, we are unable to observe whether the non-compliers would have redeemed their coupons. It is problematic if unobserved factors contributing to non-compliance also systematically affect coupon redemption. We discuss the potential problem of non-compliance and discuss our solutions in the next section, with additional robustness tests reported in supplementary online appendix S1.

3. Estimation

With randomization, we can estimate the price effects and the impact of attending the discussion session on the adoption rate of sanitary pads by estimating the following equation:

The dependent variable, Y_i, is an indicator that takes a value of 1 if an individual i redeems her coupon, and 0 otherwise. The explanatory variables include categories of discount D_ij (following the first summation in equation (1)), which takes a value in {1, 2, 3, 4, 5} for the corresponding discount percents in {10, 25, 50, 75, 90}. Coefficients on γ_j show the effect of effective price on the probability of redeeming the coupon among women in the subsidy-only group, where the omitted category includes individuals receiving a 10 percent discount. For example, γ₂ evaluates the effect of receiving a 25 percent discount coupon on the redemption rate compared to individuals receiving a 10 percent discount. In the second summation in equation (1), the discussion intervention status T_i interacts with discount indicators. The parameter β_j (j = 1, 2, …, 5) evaluates the difference in the redemption rate for those women who participate in the group discussion intervention compared to the subsidy-only group within a specific discount rate. For example, β₁ estimates the probability of coupon redemption among women in the subsidy-plus-discussion group who received a 10 percent discount compared to individuals in the subsidy-only group with a 10 percent discount. The sum of γ_j and β_j gives the redemption rate for women in the subsidy-plus-discussion group with the subsidy level j. The variable S_i indicates whether an individual is exposed to a campaign regarding menstrual health and hygiene following the 2015 earthquake (stratification variable).

The variable X_i is a vector of controls. Our parsimonious model specification only includes an indicator of whether an individual has ever used a sanitary pad as well as caste dummies. In the additional estimation, we also control for individual and household covariates including father’s education level, mother’s education level, age and age squared, indicators of income level, and relationship status with respect to the household head. We also include controls related to beliefs and practices about menstruation such as (a) whether an individual is confined to a shed during the time of menstruation (tradition of Chaupadi); (b) whether she is deemed untouchable; (c) the perceived cause of menstruation (hormones); and (d) the source of menstruation (uterus).

The estimation uses a linear probability model. Given the setting of multiple hypotheses in the model specification, we report q-values that adjust for the false discovery rate (false positive) (Storey 2002; Storey, Taylor, and Siegmund 2004). Additionally, we report the F-statistic from the joint hypothesis testing of whether coefficients on the interaction terms (β_j, j ∈ {1, …, 5}) are jointly equal to zero.

3.1. Elasticity Calculation

To calculate the elasticity of the demand curve, we assume linearity in each segment of the demand curve between two adjacent effective price points created by the coupons. The four segments are defined by discount-percent pairs (10, 25), (25, 50), (50, 75), and (75, 90). Thus, for each of the two demand curves pertaining to the subsidy-plus-discussion and subsidy-only groups, we get four elasticity estimates. For each segment, elasticity is defined as the ratio of proportionate change in the redemption rate and proportionate change in price. The proportions are calculated from the baseline of the lower price (and the respective quantity demanded) of the segment. For example, for a segment defined by the discount percent (10, 25),

where P_i(D) denotes the price level associated with discount D. This information can be directly derived from the coefficients in equation (1). For example, consider the segment between (10, 25):

3.2. Demand for Information: Compliers versus Non-compliers

One concern about non-compliance among the subsidy-plus-discussion group is that there could be self-selection in attending the discussion session.²²  Table 5 compares the baseline characteristics between compliers and non-compliers. Although compliers and non-compliers are not statistically different across the majority of baseline variables, we find that compliers are more likely to have used sanitary pads in the past. This is because compliers are more likely to fall into the higher income bracket (Rs. 25,000–39,000) compared to non-compliers.²³ The two groups may differ in other unobservables that determine redemption behavior. Given the possibility of systematic differences between compliers and non-compliers, we consider whether and how non-compliance affects our estimates.

To conceptualize the effect of non-random compliance in our elasticity estimates, consider a linear demand for sanitary pads (a simplified form of equation (1)):

where H_i denotes a high discount (H_i ∈ {0, 1} for simplicity), A_i ∈ {0, 1} denotes the discussion intervention status, and e_i are unobservables affecting demand. The variable β₀ is represented as the population mean of Y_i for the subsidy-only group (A_i = 0) and low discount (H_i = 0). The randomization of discussion intervention and discount treatments allows us to assign treatment effects such that β₁ is defined as the average treatment effect on the treated group receiving a high discount on Y_i, β₂ is the effect of discussion intervention coupled with a low discount, and β₃ represents the interaction effect of discussion intervention and a high discount.

To calculate the elasticity of demand for the subsidy-plus-discussion group, denoted by ϕ₁, we can use the formula similar to the Elasticity Calculation section:

The second equality utilizes the fact that |$\mathbb {E}[e_i|H_i = 0, A_i = 1] = \mathbb {E}[e_i|H_i = 1, A_i = 1] = 0$| due to the random assignment to the subsidy treatment arms.²⁴

A similar calculation to equation (2) shows that for the subsidy-only group, elasticity ϕ₀ is given as

From the equality established in equation (2), it is clear that the impact of discussion intervention on elasticity is due to changes in slope (β₃) and the location of the demand curve (β₂).

The effect of non-compliance is such that we are compelled to estimate ϕ₁ using a sample of individuals who chose to comply. Let C_i = 1 for compliers and C_i = 0 for non-compliers. In our selected sample due to partial non-compliance, our elasticity estimates for the treatment group, |${\hat{\phi }_1}$|⁠, is given by

The key issue is that if |$\mathbb {E}[e_i|C_i=1]\ne \mathbb {E}[e_i]$|⁠, it creates a bias in the elasticity estimates. This arises if compliance is systematically selected or related to potential outcome; i.e., |${{Pr}}(C_i=1|Y_i(1))\ne {{Pr}}(C_i=0|Y_i(1))$| and |${{Pr}}(C_i=1|Y_i(0))\ne {{Pr}}(C_i=0|Y_i(0))$|⁠, where Y_i(1) and Y_i(0) are the potential outcomes for compliers and non-compliers, respectively.

To correct for the effect of non-compliance in our sample, consider a latent model for compliance :

where |$C_{i}^{*}$| is the latent utility of compliance and W_i is a vector of individual characteristics, which may include variables that affect demand.

Then the estimated elasticity using the selected sample of compliers given in equation (3) can be written as

Given that e_i and u_i are correlated, then |${\mathbb {E}}[e_i|u_i \gt -W_i\gamma ] \ne {\mathbb {E}}[e_i]$|⁠. In other words, if unobserved factors affecting compliance also affect the demand for sanitary pads, then the elasticity estimates of the subsidy-plus-discussion group will be biased.

Note that we can estimate equation (4) using our subsidy-plus-discussion group to get consistent estimates of γ. Since households were randomly assigned to the treatment arms, we expect similar compliance behavior even in the subsidy-only group had they been invited to the discussion intervention. Thus the estimates |$\hat{\gamma }$| derived from the subsidy-plus-discussion sample can be used to predict the compliance probabilities for the whole sample regardless of the treatment status. Using these predicted compliance probabilities, it is possible to correct for the potential bias created by non-compliance.

To predict compliance probability, we closely follow the procedure to estimate the propensity score as highlighted in Imbens and Rubin (2015).²⁵ We use a logit model with the final specification including covariates pertaining to the following factors: (a) variables affecting the cost of attending the group discussion campaign (log of distance to the closest college, and marital status); (b) knowledge regarding women’s health (education, whether she received any health-related awareness following the earthquake, whether she reported the uterus as the source of menstruation); (c) socioeconomic status (caste, income); (d) past usage of sanitary pads (ever used sanitary pads, repeatedly used); (e) stigma endured regarding menstruation (kept in a shed during menstruation, considered temporarily untouchable during menstruation); and (f) benefits from attending the health campaign (the number of females in a household).

In figure S3.2(a) in the supplementary online appendix, we plot the empirical cumulative distribution function (ECDF) of compliance probability for both the subsidy-plus-discussion and subsidy-only groups. The ECDFs for these two groups coincide, suggesting that individuals in the subsidy-only group would have responded similarly had they been invited to the discussion intervention program.

3.3. Estimation Sample and Robustness Checks

For our main results, we estimate the specification given in equation (1) by excluding non-compliers from the sample, while accounting for the compliance probability as an additional control in an auxiliary model specification. In addition, we implement Heckman’s two-step approach to account for selection (Heckman 1976), where the first step predicts the compliance probability using a probit model and the second step controls for the inverse Mills ratio. Additionally, we adapt a multiple imputation approach by imputing the values on coupon redemption among non-compliers by using a random sampling from compliers with similar compliance probabilities within each discount category. These auxiliary exercises are discussed in the following section and details are provided in supplementary online appendix S1.

4. Results

In figure 1, we plot the redemption rate against prices corresponding to each discount level (excluding non-compliers). Figure 1(a) shows the demand for sanitary pads for the pooled sample; the 95 percent confidence intervals are shown by the dotted lines. Figure 1(b) plots the demand curves for the subsidy-only and subsidy-plus-discussion treatment groups. The horizontal bars indicate the 95 percent confidence intervals of the difference in redemption rates between the treatment groups at each price level.

Figures 1(a) and 1(b) show that adoption of sanitary pads increases as the effective price falls, but different patterns emerge across the two types of treatment groups. Overall, in figure 1(a), there is a sharp increase in redemption from 7 percent to 35 percent when the discount rate increases from 10 percent to 25 percent; however, the responsiveness to further reductions in price is more muted. Even at a 90 percent discount rate, redemption is only 26 percentage points higher than redemption at the 25 percent discount rate. The implication is that very high subsidies may not induce much of an additional adoption rate than moderate ones. In fact, as seen in figure 1(b), a moderate subsidy of 50 percent when combined with discussion led to a larger redemption than a 90 percent subsidy without discussion. There is a rightward shift in the demand curve due to participation in group discussion, although the difference is statistically significant only at the 10 percent and 50 percent discount rates. We did not find evidence that the horizontal difference between the two demand curves varies by the price level. We note that, since this graph is constructed from the sample that excludes non-compliers, the difference between the two demand curves may be overstated. As discussed below, we do find evidence on the relative effectiveness of combined intervention after accounting for other variables and correcting for non-compliance.

Table 6 presents the findings after estimating equation (1). The results in column (1) are based on the parsimonious specification that includes specific discount rate indicators, their interactions with the discussion treatment status, the stratification variable, caste dummies, and an indicator representing whether an individual has ever used sanitary pads. The 10 percent discount coupon is the omitted category. Additional controls are introduced in subsequent columns but the coefficients do not change much across these specifications, which is expected in an RCT study.

We report both heteroskedasticity-robust standard errors in parentheses and the q-stats in square brackets. The q-values are obtained after correcting for the false discovery rate—in each specification, we are conducting nine hypotheses tests (four indicators for discount levels and five interaction terms) and some of them could turn out to be statistically significant just by chance. We also report the F-statistic on the joint test of significance of the five interaction terms to assess whether the effect of the discussion intervention is jointly different from zero.

The regression results confirm the downward-sloping demand curve, demonstrating responsiveness to price subsidies. In the subsidy-only group, increasing the discount rate from 10 percent to 25 percent leads to 29 percentage points additional redemption. Likewise, moving from the 10 percent to the 50 percent discount rate increases redemption by 32 percentage points, which is only 3 percentage points higher; the additional 25 percentage point discount does not induce much of an additional adoption rate. It is only when moving to a 75 percent discount rate that redemption increased substantially to 47 percentage points. A further 15 percentage points discount (to 90 percent) increases the redemption rate only by an additional 4 percentage points. In summary, the magnitudes of subsidy treatment are larger when moving from 10 percent to 25 percent and 50 percent to 75 percent discount levels.

Looking at the coefficients on the interaction terms, we see that the effect of discussion intervention treatment induce more redemption across all discount levels. At the 10 percent discount level, there are 14 percentage points additional redemption than the subsidy-only group. At the 25 percent discount level, there are 9 percentage points additional redemption. The strongest effect is found at the 50 percent discount level, in which the discussion treatment induces an additional 22 percentage points redemption. Finally, at both the 75 percent and 90 percent discount levels, we observe an additional 15 percentage points redemption. Although the coefficients on these interaction terms have different magnitudes, we cannot statistically reject the null that they are equal to one another. Since each treatment bin contains limited observations, our study lacks the power to formally detect a heterogeneous effect of the discussion intervention at different price levels.²⁶

Our main results are robust to inclusion of additional covariates. Column (2) in table 6 includes household and personal characteristics such as the father’s level of schooling, mother’s level of schooling, income, relationship status (dummies), age and age squared, as well as baseline variables pertaining to the household’s attitude and one’s knowledge regarding menstruation. Additionally, column (3) controls for access to basic amenities such as the household’s distance to the nearest college and market distance (where pharmacies are located), respectively. Finally, column (4) adds the area fixed effects. The estimates are similar across columns (2)–(4) in table 6.

Columns (5)–(7) in table 6 accounts for non-compliance using various techniques. Column (5) controls for the compliance probability in the model specification.²⁷ Column (6) controls for the neighborhood-specific turnout (tol) for the discussion session to account for the rate of compliance across neighborhoods, and column (7) controls for the inverse Mills ratio (IMR).²⁸ The estimates presented in columns (5)–(7) are not affected by additional controls capturing compliance. This adds to our confidence that the effects of group discussion intervention are not severely driven by self-selection in participating in the discussion intervention program.

The results that adoption rates are relatively low even for those individuals in the subsidy-plus-discussion treatment group and receiving the highest discount rate of 90 percent are in contrast to studies focusing on bednets (Dupas 2014b; Cohen and Dupas 2010; Tarozzi et al. 2014), water purification solutions (Ashraf, Berry, and Shapiro 2010), and rubber shoes to prevent hookworm infection (Meredith et al. 2013)—for which the adoption rate is much higher at lower prices. As mentioned in Kremer and Miguel (2007), who find fairly low levels of the adoption of deworming medicine, women may have low private valuation for sanitary pads. Unlike the health products in the aforementioned studies, menstrual-health products are exclusively for women, whose adoption is significantly affected not only by social stigma associated with menstruation and psychological cost involved during the purchase of menstrual-health products, but also by their bargaining and decision-making power. These components can further explain relatively lower adoption rates even at the lowest price. In fact, table 7 shows that among non-regular users of sanitary pads, 32 percent report discomfort while purchasing sanitary pads as a major hurdle, while 24 percent report financial constraint as a deterrent. Consistent with the findings from other studies (Cohen and Dupas 2010; Ashraf, Jack, and Kamenica 2013; Dupas 2014b,a; Berry, Fischer, and Guiteras 2020), the results demonstrate that the quantity of the product demanded drops rapidly as price increases.

One possibility is that the observed difference between the two treatment groups may have resulted from the two different ways we distributed the discount coupons. To the extent that this resulted from knowledge spillovers or peer support, as is commonly discussed in the literature on peer effects, we argue that this is part of our intended mechanism. Group discussion intervention was precisely chosen in favor of a door-to-door distribution program to normalize discussion and to promote interpersonal communication regarding menstruation in a setting where discussion of menstrual topics is regarded as inappropriate due to societal stigma.

An unintended consequence of group distribution could be that it promotes a comparison of coupon levels across peers, which may influence behavior for non-price and non-menstrual-health reasons (resentment, jealousy, etc.). While we expect that even the lowest level of discount should trigger a non-negative response to demand, those who were assigned to receive the lowest discount (thus facing the highest price) compared to their peers may refrain from redeeming (i.e., due to resentment) even when the marginal benefit of redemption is positive. Such behavior would suppress redemption at the lowest discount level. However, we find that the effect of discussion intervention treatment is in fact economically significant at the lowest discount level (10 percent). Thus we do not expect this response to be prevalent in our sample. Moreover, we cannot rule out communication among members within the control group. In fact, our study setting involves tight-knit communities where information about the discount levels could be easily shared (although admittedly less immediately than among the treatment group). Additionally, communication between subsidy-only and subsidy-plus-discussion group is also possible. If anything, spillover of information from the subsidy-plus-discussion treatment group to the subsidy-only group will likely bias the estimates of discussion intervention downwards.

The results presented so far only include compliers in the subsidy-plus-discussion treatment group. To evaluate whether the exclusion of non-compliers drives the main results, we first conduct estimations using the multiple imputation method based on the propensity of compliance.²⁹ As described in greater detail in supplementary online appendix S1.2, this method indicates the division of the subsidy-plus-discussion sample into four blocks based on the quartiles of the compliance probability, such that the within-block compliance probability is similar between compliers and non-compliers. The underlying assumption governing this method is that non-compliers would have responded similarly to compliers facing the same level of discount within each block had they complied with the discussion intervention. We impute redemption status among non-compliers using random sampling (with replacement) of the compliers’ redemption values at the respective discount levels within each block based on the propensity score (of compliance) and estimate the model specification as given by table 6 (column (6)) using the partially imputed sample. This process is iterated 1,000 times to obtain the distribution of estimates at each discount level. The distribution of estimates using the multiple imputation method is presented in figure S1.2 in the supplementary online appendix. The blue and green vertical lines represent the mean value of the estimates from the multiple imputation method and main estimates from table 6 (column (6)) at the respective discount levels. The magnitudes of the mean values and the main estimates are close to one another, with the main estimates well bounded within the 5th and 95th percentiles of the distribution of estimates obtained from partially imputed samples.

Next we divide the sample into 10 equal bins of the same width based on the compliance probability, and plot the number of units across these bins, as shown in figure S1.3. The difference in frequency between the subsidy-only and subsidy-plus-discussion groups is largest at the bin with a compliance probability of 0.6–0.7. Using random sampling (with replacement), we sample the subsidy-only units to match the bin-specific number of observations in the subsidy-plus-discussion group and reestimate the model specification given by table 6 (column (7)) in the trimmed sample. This process is replicated 1,000 times. The results from such an approach using the trimmed sample (of the subsidy-only group) are shown in figure S1.4. The magnitudes of mean estimates from this approach are similar to the estimates obtained from the multiple imputation method in figure S1.2. These alternative exercises further add to the evidence that the main results are not severely driven by the selection of compliers in the subsidy-plus-discussion sample.

In the next exercise, we specifically focus on women in the lowest family income bracket (Rs. 0–24,999). This serves two specific purposes. First, it allows the focus of interventions among women facing relatively tight budget constraint to be shifted. Second, this exercise alleviates differences in baseline usage of sanitary pads between subsidy-only and subsidy-plus-discussion samples arising due to the difference in income across the compliers’ versus non-compliers’ sample.³⁰

The results for women in the lowest income bracket are presented in table S3.5. The effects of subsidies on take-up for women in the subsidy-only group are similar to the findings reported in table 6. However, the effects of discussion intervention are more pronounced at the lowest and highest levels of subsidy (10 percent and 90 percent discount) across the reported specifications. Column (1) indicates that the group discussion intervention increased take-up rate by 25 and 21 percentage points among women with 10 percent and 90 percent discount coupon, respectively. For discount levels of 25 percent and 50 percent, the effects of discussion intervention are around 12 percentage points. The F-statistics presented at the bottom of the table suggest that the effects of the group discussion intervention are statistically different from zero at the conventional levels of significance.

4.1. Elasticity Estimates

The elasticity estimates for both treatment groups are calculated as described in the Elasticity Estimation section, based on the point estimates given in table 6, column (6). The estimates are presented in table 8.³¹ The demand is inelastic for both groups for prices below Rs. 52.5. This result is consistent with low elasticity of health technologies at relatively low prices (Berry, Fischer, and Guiteras 2020). The table shows that the elasticity estimates are similar across subsidy-only and subsidy-plus-discussion groups for the lowest segment of the demand curve (e.g., between discount rates 75 percent and 90 percent). It is interesting to note that the demand is highly elastic at the highest price segment; the elasticity estimates for subsidy-only and subsidy-plus-discussion groups are −4.56 and −2.57, respectively.

4.2. Heterogenous Effects by Psychological Cost

As a way to demonstrate the role of the psychological costs due to menstrual stigma on redemption behavior, we examine the heterogeneous effects of interventions by levels of psychological costs. Due to a lack of a direct measure, we infer psychological costs by selected baseline characteristics and construct a dummy indicating a high-psychological-cost measure. We consider four plausible proxies of high psychological costs: housed in a shed during menstruation, viewed as untouchable, lacking recent experience with the product, and having a high index value for stigmatization. In the third case, those who did not self-purchase sanitary pads (by themselves) in the past cycle are classified into the high cost group, while in the fourth case we use an index of available measures of stigma to define high versus low stigma groups.

Consider the impact of participating in the discussion session among women in high- versus low-psychological-cost groups.³² The agent decides whether to purchase sanitary pads by comparing the marginal benefit of sanitary pads to two distinct types of marginal costs: the utility loss from reduction in other consumption and the psychological costs associated with purchasing sanitary pads. The subsidy-only treatment works by lowering the utility loss from reducing other consumption, whereas the discussion intervention treatment additionally lowers the psychological costs.

These two marginal costs work differently for the low- and high-psychological-cost groups. While both types of costs are important for the high-psychological-cost subgroup, the utility cost is much more relevant than psychological cost for those with low psychological cost. Thus we expect to find larger impacts of discussion intervention treatment among women bearing high psychological costs. Likewise, the demand curve for the high-psychological-cost subgroup should be to the left of the low cost subgroup in the absence of discussion treatment intervention. Following the discussion treatment, we would expect to see a rightward shift in the demand with effects concentrated among those with higher psychological cost.

We test these predictions by estimating an econometric model that includes interaction the between discount levels, the discussion intervention treatment indicator (T_i), and high stigma (high psychological cost) indicator (S_i). The discount levels are collapsed into high (High = {75, 90} percent), mid (Mid = 50 percent), and low categories (Low = {10, 25} percent) to ensure that we have enough observations in each cell. Altogether, we have 12 (3 discount levels × 2 treatment groups × 2 psychological-cost groups) cells. We run the following regression:

The omitted category is composed of women with a low psychological cost in the subsidy-only group who receive a low discount coupon. The estimate of δ₁ (δ₂) shows the effect of receiving a mid-level (high-level) discount coupon for this group of women (compared to the omitted category). Similarly, the κ’s depict the additional effect of discussion intervention on the low-psychological-cost women at each discount level. Next, the β’s estimate the difference in the redemption behavior between women with low and high psychological costs in the subsidy-only group at each discount level. Finally, the (κ + γ)’s estimate the redemption behavior of high-psychological-cost women in the subsidy-plus-discussion group compared to high-psychological-cost women in the subsidy-only group at the corresponding discount levels. The estimates of the γ’s show whether the subsidy-plus-discussion treatment effects for high-psychological-cost women are different from those for low-psychological-cost women. The predicted redemption for each subgroup can be obtained from the designated coefficients.³³

It should be noted that the proxies we use for unobserved psychological costs are correlated with demographic characteristics that may also impact redemption behavior. Table S3.7 in the supplementary online appendix shows the correlates of high psychological cost. We find that each proxy of high psychological cost varies in their association with baseline characteristics. Being housed in a shed (column 1) is strongly related to caste, family income, and age, but not education. Individuals in Newar and Janjati households, girls or women in high-income households, and relatively younger individuals are less likely to be housed in a shed during menstruation. Being considered untouchable (column 2) is mostly a caste-specific practice and not highly correlated with education, income, or age. Finally, not purchasing sanitary pads in the previous cycle is determined by schooling and family income, but not by caste membership.

The results from estimating equation (5) are presented in table 9. The dependent variable in each column is the redemption indicator. The column headings indicate the proxy used to categorize women into the high-psychological-cost group. The odd columns present results with basic controls and the even columns control for demographic variables. The redemption rate for women in the low-psychological-cost group with a low discount coupon is around 20 percent across all columns. The point estimates of the coefficients vary somewhat between different proxies for psychological cost, but are broadly consistent. The redemption rates for high-psychological-cost women within the subsidy-plus-discussion treatment group compared to high-psychological-cost women with the subsidy-only treatment group are positive across all subsidy levels and are represented by the sum of the coefficient on the double interaction term (discount level × discussion) and the coefficient on the triple interaction term at the respective subsidy level. The joint test rejects the null in most cases—the coefficients pertaining to the high-psychological-cost group with discussion intervention are similar to those of the high-psychological-cost group receiving only the subsidy treatment at the conventional levels of significance, as shown in table S3.7.

For ease of exposition, we use the estimates in table 9 and construct predicted redemption values to plot the demand curves for four groups of women formed by combining the levels of psychological cost and discussion intervention treatment status. Results are shown in figure 3. Panel A plots the demand curve when being housed in a shed is used as the proxy for high psychological cost (corresponding to estimates in column (2) of table 9). Panel B shows the demand curves when using the past purchase history as the proxy (corresponding to estimates in column 6). Also, panel C shows the results when an index measure constructed from available stigma-related variables is used to categorize the high stigma group (corresponding to estimates in column 8). The results vary slightly across the three proxies, most likely due to differences in the accuracy of classification.

We find different demand curves for women with low and high psychological costs in the subsidy-only group, indicating that the level of psychological cost is an important influence in their responsiveness to price changes. Panel B shows that women in the subsidy-only group who do not have a recent history of purchasing sanitary pads (high-psychological-cost group) are least responsive to additional discounts. Their expected redemption increases from 0.2 to 0.43 when moving from the lowest to the highest discount levels. On the other hand, for the same price change, women in the subsidy-only treatment group who have a purchase history (low-psychological-cost group) increase their redemption from 0.2 to 0.6. These findings are consistent with the hypothesis that women in the low-psychological-cost subgroup are more responsive to discount subsidies.

Women in both the low- and high-psychological-cost groups respond to the group discussion intervention treatment by shifting their demand to the right, with a larger shift for high-psychological-cost women. Also, for both low- and high-psychological-cost groups participating in the discussion treatment, the redemption rate at the mid-level discount exceeds what is observed with the high-level discount in the subsidy-only case. Participating in the discussion intervention but receiving the lowest discount is also not as effective. The expected redemption value for low-psychological-cost women receiving a low discount subsidy and joining the discussion intervention session is 28 percent, as shown in panel B. The responsiveness to the discussion treatment among women with high psychological costs at the low discount level is higher, which increases when combined with higher subsidy levels. For instance, the expected redemption rate for high-psychological-cost women with the discussion treatment is 59 and 70 percent at the mid and high discount levels, respectively. The results using an index measure to represent the high-psychological-cost category, presented in figure 3(c), are broadly consistent with the findings presented in earlier subfigures.

What might explain the greater responsiveness from high-psychological-cost women to the discussion treatment such that the demand is to the right when compared to the low-psychological-cost subgroup with the same treatment? One possible explanation is that some learning occurs for women with high psychological costs. In other words, they are relatively new to using the product, so the discount coupled with discussion intervention treatment may increase their willingness to experiment with the product. To test this possibility, we reestimate the demand curves as shown in figure 3(b) by restricting the sample to those who reported having previously known about sanitary pads in the baseline. The results from this auxiliary exercise show similar patterns to figure 3(b), providing suggestive evidence that an increase in responsiveness among the high-psychological-cost subgroup undergoing the discussion treatment is not being driven due to the novelty of the product.³⁴

The important finding from our heterogeneous analysis is that the subsidy-plus-discussion treatment intervention is consistently more effective for the subsample of individuals who are likely to have higher psychological cost. This finding provides suggestive evidence that discussion intervention treatment used in the study increases redemption through a reduction in the psychological cost associated with menstruation. However, we caution that the analysis uses imperfect proxies to identify women with high versus low psychological cost. Future studies can benefit from directly eliciting the underlying psychological costs.

5. Conclusion

In this study, we use a randomized controlled experiment to evaluate the relative effectiveness of the subsidy-only intervention against an intervention that combines subsidization with group discussion in inducing adoption of sanitary pads, a product whose demand is likely curtailed due to stigma surrounding menstruation. The study is based in rural Nepalese villages, where regular use of sanitary pads is limited. We find that the demand for sanitary pads is downward sloping for both the subsidy-only and subsidy-plus-discussion groups. Moreover, women invited to participate in the group discussion intervention have a higher overall demand for the product. These effects are also observed among women belonging to the lowest family income bracket (Rs. 0–24,999 per month), who arguably face tighter budget constraints. We also find that the impact of discussion treatment is concentrated among women who are likely exposed to higher levels of stigma in the baseline and thus have higher psychological costs.

Our results complement the findings of Ashraf, Berry, and Shapiro (2010), who find that information coupled with subsidies can increase the demand for an unfamiliar health product (a water purification product in their case). By extending the existing literature, we focus on the adoption rate of menstrual-health products, which are already well known among the participants in the study setting. Thus the results of this study inform the design of policies aimed at increasing the adoption of health products whose demand may be partly affected by stigma associated with the relevant health issue by emphasizing that group discussion should be provided concurrently with subsidies. The group discussion helps women overcome any barriers imposed by social stigma and make them more responsive to price subsidies. From a cost-benefit perspective, the findings of this study highlight that lower levels of subsidization combined with discussion intervention can help reduce the cost of providing health subsidies, assuming that the additional cost of delivering the program is lower than the realized savings.

We recognize several gaps in our study that could be filled in future studies. First, only women were invited to attend the discussion campaign. As the societal stigma clearly governs the treatment of girls and women during menstruation, it is necessary that the discussion or information programs target men as well. Second, this study focuses on the short-term effects of the program; however, discussion intervention may have long-term impacts through learning (Dupas 2014b, 2011). Finally, this study did not attempt to measure reduction in the societal stigma itself. The question of whether group discussion intervention programs can influence societal stigma about health warrants a rigorous investigation. These are some extensions that we look forward to incorporating in future studies.

Data Availability Statement

Due to the sensitive nature of the research supporting data is not available.

Footnotes

Author Biography

Vinish Shrestha (corresponding author) is an associate professor at Towson University, Towson, MD; his email address is [email protected]. Rashesh Shrestha is an economist at the Economic Research Institute for ASEAN and East Asia (ERIA), Jakarta; his email address is [email protected]. The views expressed in this paper are those of the authors. The research for this article was financed by the College of Business and Economics, Towson University. The authors would like to express sincere gratitude to Dipasha Bista, Barsa Budhathoki, Ashma Gautam, Rita Ghatani, Shanti Ghimere, Malati Gurung, Situ Kapali, Aastha Shrestha Neupane, Ranju Pandey, Sandhya Pathak, Sari Khatiwada, Sarita Regmi, Pratibha Shrestha, Sapana Shrestha, Shreeju Shrestha, and Prativa Thapa for helping us throughout the experiment phase as research assistants. The authors would like to thank Raju Amatya and Pranish Shrestha for providing logistical guidance, without which it would not have been possible to conduct this study. Also, the authors are thankful to Anja Benshaul-Tolonen, Eric Edmonds, Seth Gitter, Paul Glewwe, Swetha Peteru, Cleo Shrestha, Rakshya Gorkhali, Shreeju Shrestha, participants at the Mid-West International Development Economics Conference, DIAL Conference on Development Economics, and anonymous referees for valuable comments. A supplementary online appendix is available with this article at The World Bank Economic Review website.

References