If overcrowding harms health care quality, the impacts of encouraging more people to use services are not obvious. Impacts will depend on whether marginal entrants benefit and whether they benefit enough to offset the congestion externalities imposed on inframarginal users. We develop a general-equilibrium model that formalizes these ideas. We examine them empirically by studying JSY, a program in India that paid women to give birth in medical facilities. We find evidence that JSY increased perinatal mortality in areas with low health-system capacity, was particularly harmful in more-complex births, reduced the quality of facilities' postnatal care, and generated harmful spillovers onto other services.

IT is increasingly common for governments to incentivize individuals to use health care.1 The rationale is clear; barriers, such as a lack of information or resources, might prevent individuals from seeking beneficial care. However, health systems in many low- and middle-income countries (LMICs) that use demand-side incentives are often poorly resourced and provide poor-quality care.2 Given this, incentivizing demand could backfire. The additional demand may lead to congestion externality if, by causing fixed resources to be spread more thinly, it results in the quality of care falling for everyone, including those who would have used the care without any subsidy. Moreover, when health care quality is poor, the marginal group who takes up care due to the incentive may not benefit. Finally, to manage an increase in demand for the incentivized care, providers may pull effort and resources away from other types of care, which could risk a wider deterioration in quality.

This paper examines these issues in the context of the world's largest conditional cash transfer program: India's Janani Suraksha Yojana (JSY). Introduced in 2005, JSY pays substantial cash incentives—about 28 times the average rural daily wage for casual labor—to women who give birth in a health facility.3 The scheme does not provide increased funds to facilities to cover the cost of additional deliveries. JSY is widely credited with causing a rapid decline in home births, which fell from 80% in 2005 to 40% in 2011 in the states targeted by the program.4 However, rigorous research has found that it had no overall impact on birth outcomes (Powell-Jackson et al., 2015). This is puzzling as a set of simple medical procedures can mitigate the vast majority of mortality risks associated with childbirth for both mother and infant (Lawn et al., 2005). The pervasive poor quality and low capacity of much of India's government health system might explain why bringing births into medical facilities had no overall effect on health. At less than 1% of GDP, India's public spending on health is a third (relative to national income) of other major emerging economies, and the system has a huge shortage of qualified health professionals (Rao et al., 2011) compounded by high rates of absenteeism (Chaudhury et al., 2006).5

This study explores the effects of incentivizing the demand for health care in this highly supply-constrained system. We begin by creating a flexible general-equilibrium model of birth outcomes and families' decisions about whether to deliver at home or in a facility. Crucially, we allow for births to be heterogeneous in their complexity and in how much they benefit (in expectation) from taking place in a facility. We allow births to sort into facilities endogenously, whether partially or fully or not at all, on the basis of these benefits. Key to our model is a congestion, or overcrowding, externality that causes the expected riskiness of all births in medical facilities, and especially the most-complex facility births, to increase when the overall use of facilities rises relative to their capacity.

In our model, the introduction of a subsidy has two effects. First, it results in private benefits (or harms) to the marginal births that take place in a different location as a result of the subsidy. Second, it results in social costs to the births that would have already occurred in facilities, the inframarginal facility births, which now become riskier because of increased congestion. The degree to which births select into facility delivery based on their private gains plays an important role in shaping both effects. For example, in the extreme case of perfect selection on gains (a classic Roy model), the marginal births will be harmed by being incentivized into facilities, as these births were making a privately optimal choice to deliver at home before the subsidy. Moreover, in this case the inframarginal facility births that will be exposed to the additional congestion are the most-complex births, which are precisely the births that are most harmed by congestion. In this extreme case, then, a subsidy unambiguously worsens health outcomes.

On the other hand, our model highlights that there are circumstances in which a subsidy could improve health outcomes. The more complex the marginal births are that are brought into facilities due to the subsidy, the more they will benefit from the move. At the same time, the fewer complex births that occur in facilities to start with, the smaller the aggregate size of the congestion externality. Weak initial sorting on gains but a marginal utility of receiving the subsidy that is increasing in birth complexity could give rise to such conditions.6 While it is theoretically ambiguous, and ultimately an empirical question, how the impact of the subsidy varies with respect to preexisting capacity, we show that there are several channels through which higher preexisting capacity increases the chance that a subsidy will be useful.

We examine the impacts of JSY empirically, focusing on two dimensions of heterogeneity that our model suggests will be key: preexisting system capacity and birth complexity. We use geographic disparities in the preexisting capacity of district health systems—disparities caused by the central government's allocation of funds and variation in states' political priorities (Kumar et al., 2011)—to assess how the effects of stimulating demand depend on supply constraints. Specifically, exploiting exogenous variation in JSY's rollout, we examine whether the preexisting capacity (specifically the numbers of doctors, nurses, and beds relative to the size of the population served) affected its impacts.

We find that JSY led to a large average increase, of 7.86 percentage points, in the probability that births were delivered in health facilities. In aggregate, this doubled the number of institutional deliveries per day for which each government-run secondary care facility was responsible. The increase in deliveries per facility was particularly stark in districts with below-median capacity in the secondary care system. Here caseload increased by a factor of 2.5, from 1.92 deliveries per facility per day to 4.80. We show evidence that while there was some degree of sorting on gains—more-complex births were more likely to occur in facilities than less-complex births both before and after JSY—JSY did not disproportionately draw in complex births. More- and less-complex births were equally likely to be marginal to the subsidy, suggesting that JSY was not particularly effective at targeting the deliveries that would have privately gained the most from moving to a facility.

We show that in these districts with below-median capacity in the secondary health care system, the average risk of perinatal mortality increased as a result of JSY by 0.90 percentage points, or by 24.3% relative to the rate before JSY. Although JSY caused the same increase in the institutional delivery rate in areas with above-median preexisting capacity, it had no impact on perinatal mortality, neither positive nor negative, in these areas. We show further that this increase in perinatal mortality risk was entirely concentrated on complex births. This pattern of heterogeneity can, in our model, only be rationalized by the presence of a congestion externality. Moreover, we find that JSY reduced the probability that infants received any checkup between 2 and 10 days after a facility birth, which could indicate that facilities reacted to congestion by discharging women earlier, or by reducing checkups for babies still in inpatient care.

Our empirical analysis goes further by examining whether the above responses are short-lived and whether the effects of the subsidy spill over into other areas of care. We present more tentative evidence suggesting that harmful effects of JSY in low-capacity areas may have persisted up to 5–10 years after the start of JSY: over this period, low-capacity areas with higher compliance with JSY had higher mortality than low-capacity areas with lower compliance. Finally, we also find that stimulating demand for institutional delivery had adverse spillover effects on the proportion of children who were up to date with all vaccinations.

Our findings have clear implications for policy. They show that the capacity of a health system to deliver a rapidly increasing amount of incentivized care without reducing the quality of services should be a first-order consideration when deciding whether to adopt demand-side incentives. These considerations are likely to be particularly crucial in determining whether to incentivize the take-up of invasive, time-consuming, and potentially risky procedures such as institutional delivery.

Our paper contributes to several literatures. First, it contributes to the study of demand-side incentives in health, which have proved especially popular policies in LMICs (Kremer and Glennerster, 2011; Dupas and Miguel, 2017). We bring an explicit focus on congestion and offer a flexible general-equilibrium model of the health effects of incentives in the presence of congestion externalities. Demand incentives are generally effective at increasing uptake of health care (Lagarde et al., 2009; Giedion and Díaz, 2010), and therefore, unless there is spare or rising capacity in the system, they will increase congestion. The mixed evidence on the impact of such incentives on health is consistent with the idea that congestion externalities might be key to shaping their effects. For less invasive health interventions, such as the health checkups, vaccinations, or nutrition advice that early conditional cash transfers in Latin America typically incentivized, health benefits have sometimes, but not always, been found (see the review by Lagarde et al., 2009). However, when demand for more invasive procedures has been targeted with no change to the supply side, studies, including previous studies of the effect of JSY, have generally not found health benefits (Giedion and Díaz, 2010; Das and Hammer, 2014; Powell-Jackson et al., 2015; Okeke et al., 2020). Conversely, where the supply side has been improved simultaneously, health benefits have often followed (Miller and Vera-Hernández, 2013; Gruber et al., 2014; Cesur et al., 2017). Our paper contributes by directly examining the role of supply in explaining heterogeneity in how demand-inducing policies affect health. We do so within a single health system and focus on invasive procedures whose riskiness depends directly on quality. Finally, by exploring heterogeneity by case complexity, we provide evidence that congestion externalities, rather than only the private impacts of health care on the marginal cases, are important in limiting the effectiveness of demand-inducing policies in low-capacity areas.

Second, this paper adds to the evidence on the importance of health care quality. In many health systems in LMICs, including in rural India, inputs such as trained health professionals and equipment are often not effectively translated into evidence-based care (Chaudhury et al., 2006; Das et al., 2008, 2016, 2018; Mohanan et al., 2015). Nevertheless, we show that even if existing health professionals and equipment are not used as effectively as they might be, their availability can still be an important predictor of how increased use of health care translates into health outcomes.7 Interestingly, we find that capacity in secondary-care health facilities is a very important predictor of the impacts of increasing institutional deliveries on health.

Third, our paper contributes to the literature on congestion externalities in health systems. Overcrowding has been the focus of much attention within the medical literature (Hoot and Aronsky, 2008), and a growing number of papers have also considered these issues from an economic perspective (Freedman, 2016; Hoe, 2022; Marks and Choi, 2019). We focus on India's public health system, which, similarly to public health systems in many LMICs, operates with far fewer resources than the systems in the United States or the United Kingdom, where the empirical economic research on congestion externalities has focused to date. We might expect congestion externalities to play a more prominent role in such resource-constrained settings. Differently from much of the empirical work, which has focused on day-to-day fluctuations in demand (Freedman, 2016; Hoe, 2022; Marks and Choi, 2019), we focus on a context where a high level of congestion is a constant feature of the health system and thus we explicitly model how patients react endogenously to congestion in choosing whether or not to seek health care.

The final literature we contribute to considers the spillover effects from policies that target one aspect of health care onto another aspect, in line with Holmstrom and Milgrom (1991). Previous work has found that when health providers are incentivized in reimbursement contracts to carry out more of a certain procedure, this often comes at the expense of nonincentivized tasks (e.g., Dumont et al., 2008). Our paper is the first to assess whether incentivizing the demand for specific procedures leads to a deterioration in the use or quality of other procedures. We find that it does. Whether such a reallocation is optimal depends on the relative productivity of the different procedures. In this case, however, the reallocation of resources prompted by JSY provided no aggregate health benefits but caused a reduction in childhood vaccinations.

This paper proceeds as follows. Section II sets up our conceptual framework. Section III provides a basic overview of JSY, while section IV discusses our data, choice of sample, and construction of measures. Section V presents our methods and results for assessing the impact of JSY and discusses the different mechanisms. Section VI concludes.

We begin by creating a theoretical model to clarify the different mechanisms through which subsidizing facility delivery might affect birth outcomes. We develop a general equilibrium model of both health and facility choice that allows births of heterogeneous complexities to select into facility delivery endogenously. Crucially, the model allows for congestion externalities whereby an increase in the overall use of facilities increases the risk associated with facility delivery for all births.

Given the model's various mechanisms, it does not provide an unambiguous prediction of whether or not subsidizing demand will improve health outcomes, although it nests a classic Roy model that does give unambiguous predictions. Instead, our goal is to characterize the features of the underlying environment that will shape a subsidy's impact.

A. Setup

Determinants of health.

Births are indexed by i and may result in either a good health outcome (Hi=1) or a bad health outcome (Hi=0). We allow births to be heterogeneous in their ex-ante risk, in how much they gain or lose from facility delivery, and in the degree to which they suffer from congestion. We summarize this heterogeneity by allowing births to differ along one underlying dimension: their “complexity,” denoted by ηi. Without loss of generality, we assume that ηi is uniformly distributed across the interval [0, 1], with 0 being the least-complex birth and 1 being the most complex.

Women may give birth either at home (Fi=0) or in a health facility (Fi=1). We let the ex-ante probability of a good health outcome for home birth i be denoted πh(ηi) while the ex-ante probability of a good health outcome from a facility birth is πf(F/X,ηi), where X is capacity, FiFidi is overall usage, and F/X is congestion. We make the following assumptions over how complexity affects the private costs/benefits from facility birth:

  • A1. More-complex births are always more risky than less-complex births that occur in the same location: dπh(ηi)dηi<0 and dπf(F/X,ηi)dηi<0,F/X.

  • A2. More-complex births have more to gain (or less to lose) from facility delivery: πf(F/X,ηi)ηi-dπh(ηi)dηi>0,F/X.

Our model nests a model with no congestion externality. In that case, we have assumptions A1 and A2 and a “no externality” condition (i.e., dπf(F/X,ηi)d(F/X)=0ηi). If a congestion externality is present, then we further assume the following:

  • E1. Increased congestion increases the risk of facility births: dπf(F/X,ηi)d(F/X)<0,ηi.

  • E2. More-complex births are at least as harmed by congestion as less-complex births: d2πf(F/X,ηi)d(F/X)dηi0.

  • E3. The harmful effects of congestion on the health of facility births are increasing in current congestion: d2πf(F/X,ηi)d(F/X)2<0.

Preferences.

In the absence of any subsidy, let utility over place of birth be driven by a consideration for health and a consideration for nonhealth costs and benefits, with the weight given to health being λ[0,1]. Then let utility be derived from a subsidy s at a marginal rate of θ(ηi), with θ(ηi)>0,ηi, which we consider as the marginal utility of consumption and which is allowed to depend on the complexity of the birth. The possibility of this dependence can capture, for example, the idea that more-complex births might be disproportionately from poorer families,who may derive higher utility from the subsidy than richer families. Overall, then, we denote utility as
u(Hi,Fi,ηi,νi)=λHi+(1-λ)Fiνi+Fiθ(ηi)s,
(1)
where νi is an idiosyncratic preference for facility delivery. Given this preference is likely to be formed of many different factors (e.g., tastes, family tastes, social norms, or distance to a facility), the central limit theorem suggests νi will be well approximated by a normal distribution. We assume this distribution is independent from birth complexity: νi|ηiN(ν¯,σν2). Thus, expected utility for a birth of complexity ηi and idiosyncratic preference νi, occuring in location Fi
U˜(Fi,ηi,νi)=λ[Fiπf(F/X,ηi)+(1-Fi)πh(ηi)]+(1-λ)Fiνi+Fiθ(ηi)s.
(2)

Although not formally modeled, we note that a lack of responsiveness to ex-ante risk (i.e., a low λ) could be driven by uncertainty over birth complexity rather than families not valuing health outcomes per se.

B. Social Planner's Problem

We first consider the allocation of births to facilities that a social planner trying to maximize overall health would choose.8 We assume that the social planner can perfectly observe each birth's complexity, and thus each birth's ex-ante risk conditional on place of delivery. Conditional on allocating a fraction F of births to facilities, the social planner will always assign the fraction F with the highest complexity. In other words, the social planner will perfectly sort births based on their ex-ante gains from facility delivery. The solid line in figure 1 plots the private benefits gained by the marginal births as the planner allocates more and more births to facilities. This curve slopes downwards for two reasons: (1) as the proportion of births occurring in health facilities increases, the complexity of the marginal birth falls and with it the idiosyncratic gain for the marginal birth falls; (2) as the total number of births in facilities increases, crowding increases, facility quality decreases, and the gain for the marginal births decreases.
Figure 1.
Social Planner's Problem

Figure plots the private benefit to the marginal birth and the social marginal cost to all births already in facilities as more births enter under the condition that births enter in the order of their complexity, from most complex first to least complex last. Points B and C are discussed in section IID.

Figure 1.
Social Planner's Problem

Figure plots the private benefit to the marginal birth and the social marginal cost to all births already in facilities as more births enter under the condition that births enter in the order of their complexity, from most complex first to least complex last. Points B and C are discussed in section IID.

Close modal

In choosing how many births to allocate to facilities, the planner will trade off gains to the marginal births with the marginal increase in the congestion externality (the “social marginal cost,” or the dashed line in figure 1), which is experienced by all births already taking place in facilities. The social planner chooses F to equate the social marginal cost from moving an additional birth to the facility with the private marginal benefit to the birth that is moved. This social optimum is marked as point A in figure 1.

C. Decentralized Location Choice and Sorting

We next explore the decentralized equilibrium where families choose where to deliver based on their own preferences and ex-ante risk, without considering any externalities. We assume that families have rational expectations, that is, that their beliefs about ex-ante risk associated with home and facility births and how this is impacted by complexity and congestion are correct, and that they are accurately able to predict how the introduction of the subsidy will affect overall facility usage. Given our setup, the probability that a birth of complexity ηi will occur in a facility given congestion F/X and subsidy s, which we define as F¯(ηi,F/X,s), is
F¯(ηi,F/X,s)=Prλπf(F/X,ηi)+(1-λ)νi+θ(ηi)s>λπh(ηi)=Φλ1-λ[πf(F/X,ηi)-πh(ηi)]+11-λθ(ηi)s+ν¯σν2.
(3)
In equilibrium, the proportion of births occurring in facilities is the fixed point, F:
F=01Φλ1-λ[πf(F/X,ηi)-πh(ηi)]+11-λθ(ηi)s+ν¯σν2dηi.
(4)

In appendix section B.2, we show that dFds>0, that is, a subsidy will unambiguously increase the proportion of facility births. The degree to which births of complexity ηi are marginal to the subsidy is described by dF¯(ηi,F/X,s)ds [equation (B.8) in appendix B].

D. Health Impacts of the Subsidy

Let H¯(ηi,X,s) be the proportion of good health outcomes for births of complexity ηi when capacity is X and the subsidy is s:
H¯(ηi,X,s)=F¯(ηi,F/X,s)πf(F/X,ηi)-πh(ηi)+πh(ηi).
(5)
Differentiating with respect to s, we see that the marginal impact of a subsidy on the proportion of good health outcomes for births of complexity ηi is composed of two effects:
dH¯(ηi,X,s)ds=dF¯(ηi,F/X,s)dsπf(F/X,ηi)-πh(ηi)(1)PrivateCost/BenefittoMarginalBirth(s)+F¯(ηi,F/X,s)dπf(F/X,ηi)d(F/X)1XdFds(2)Crowdingexternalityoncurrentfacilitybirths.
(6)

Appendix B.1 gives the full expressions for the total differentials.

In other words, effect (1) is the expected private health cost or benefit to marginal births of complexity ηi. It is the difference in ex-ante risk between facility and home births for births of complexity ηi at the current levels of crowding (F/X) weighted by the extent to which births of this complexity are marginal to the subsidy. Effect (2) is the congestion externality. It is the reduction in the probability of a good health outcome from facility births of complexity ηi that occurs because of the overall increase in facility births due to the subsidy. This is scaled by the proportion of births of complexity ηi that will be subject to this externality (i.e., the proportion currently taking place in facilities).

To find the overall effect, we average across complexities:
H(X,s)=01H¯(ηi,X,s)dηi,dH(X,s)ds=01dH¯(ηi,X,s)dsdηi.
(7)

Overall, a positive subsidy will be optimal whenever there is a marginal gain in health from moving from a zero subsidy to a marginally positive subsidy, that is, whenever dH(X,s)ds|s=0>0. This will depend on the sign and magnitude of the private marginal benefits accrued, and, if these are positive, then whether they are large enough to offset the congestion externality imposed on births currently taking place in facilities. We now discuss each effect in turn.

Private benefits/costs to the marginal births.

The sign and size of the marginal private costs or benefits will depend on which births are marginal to the subsidy. The more complex the set of marginal births is, the more the marginal births will gain from being moved into facilities.

In our framework, the complexity of the marginal births depends both on (1) the weight that families give to ex-ante health risk in the decision about where to deliver, λ, and (2) on how the marginal utility from the subsidy (θ(ηi)) varies across the complexity distribution. Figure 2 illustrates how varying both of these factors changes, across the complexity distribution, the prior probability that births will occur in facilities (plot a), the degree to which births are marginal to the subsidy (plot b), and the overall marginal benefits/harms from the subsidy (plot c). We construct example scenarios using parametric functional forms for πh(ηi)πf(F/X,ηi) and θ¯(ηi) (see appendix section B.3.3 for details). Each example scenario is calibrated to match pre-JSY mortality rates (3.7%), facility delivery rates (20%), and to have the same constant aggregate responsiveness of facility usage to the subsidy.
Figure 2.
The Same Increase in Institutional Delivery Rates Due to a Subsidy Will Translate into Higher Health Gains if There Is Little Initial Sorting on Health Gains but the Marginal Utility (MU) of Consumption is Increasing in Birth Complexity

Figure plots, by complexity: (a) The prior probability of facility birth [equation (3)]; (b) marginality to the subsidy [equation (B.8)]; and (c) the private marginal benefit from the subsidy [equation (6)]. Details of the parametric functional forms used for πh(ηi)πf(F/X,ηi) and θ¯(ηi), parameter choices, and calibration are in appendix B.3.3. In the “high sorting” scenarios, λ=0.9 while in the “low sorting” scenarios, λ=0.2. In the “const. MU” scenarios, dθ¯(ηi)/dηi=0 while in the “incr. MU” scenario, dθ¯(ηi)/dηi>0.

Figure 2.
The Same Increase in Institutional Delivery Rates Due to a Subsidy Will Translate into Higher Health Gains if There Is Little Initial Sorting on Health Gains but the Marginal Utility (MU) of Consumption is Increasing in Birth Complexity

Figure plots, by complexity: (a) The prior probability of facility birth [equation (3)]; (b) marginality to the subsidy [equation (B.8)]; and (c) the private marginal benefit from the subsidy [equation (6)]. Details of the parametric functional forms used for πh(ηi)πf(F/X,ηi) and θ¯(ηi), parameter choices, and calibration are in appendix B.3.3. In the “high sorting” scenarios, λ=0.9 while in the “low sorting” scenarios, λ=0.2. In the “const. MU” scenarios, dθ¯(ηi)/dηi=0 while in the “incr. MU” scenario, dθ¯(ηi)/dηi>0.

Close modal

As shown by the “high sorting” (blue dashed lines) example in figure 2, a high λ, indicating a high degree of selection on gains, implies that the most-complex births already largely occur in facilities even without the subsidy (plot a), which means that they are not marginal to the subsidy (plot b). In this scenario, the subsidy will primarily draw in the next most-complex births out of the births currently taking place at home (plot b).

As λ approaches 1, the model approaches a pure Roy model with complete selection on gains. In this case, without the subsidy, births select into facilities in the order of their complexity (from the most complex to the least complex) until the marginal private health gain to the last birth is zero (marked by point B in figure 1). Even without the congestion externality too many births occur in facilities relative to the social optimum (point A in figure 1). Subsidizing demand in this case unambiguously worsens health. Under perfect selection on gains, the marginal births to an infinitesimally small subsidy are unaffected by whether or not they occur in a facility. A discrete increase in the subsidy (point C in figure 1) brings in births that are harmed by the fact that they are now taking place in a facility. This is on top of any costs from the overcrowding externality.

If births respond only very weakly to ex-ante health risk in selecting into facilities (i.e., if λ is low), then there is a pool of highly complex births that currently take place at home and could gain a lot from being moved into facilities. If the marginal utility of consumption θ(ηi) is independent of ηi (as in the solid line in figure 2), the subsidy will always draw in the births uniformly from across the complexity distribution. However, if the marginal utility of consumption θ(ηi) is increasing in birth complexity (as demonstrated by the dotted line in figure 2), then more-complex births will be more likely to be marginal. Since more-complex births have more to gain from facility delivery, this would result in larger private gains from the subsidy. Such a scenario might arise if complex births are disproportionately from poorer families, who might, for example, have poorer access to prenatal care and a higher marginal utility of consumption.

Congestion externality.

In the presence of a congestion externality (i.e., if E1–E3 are true), then an increase in aggregate facility usage due to the subsidy will decrease the probability that facility births of all complexities result in a good outcome. As shown in figure 3c, in our framework it will always be the most-complex births that will be most harmed by increased congestion. This is for two reasons. First, under sorting on gains, more-complex births are more likely to occur in facilities to begin with and therefore to be exposed to the congestion (figure 3a). Second, by assumption E2 more-complex facility births are more harmed by congestion than are less-complex facility births (figure 3b). More generally, the more births that occurred in facilities to begin with and the more elastic total facility usage is to the subsidy, the greater the size of the externality.
Figure 3.
Under Sorting on Gains, the Costs of the Congestion Externality Will Be Concentrated on Complex Births

Figure plots, by complexity: (a) The prior probability of facility birth [equation (3)]; (b) the marginal reduction in facility quality due to the subsidy [equation (B.4)]; and (c) the social marginal cost from the externality [equation (6)]. Details of the parametric functional forms used for πh(ηi)πf(F/X,ηi) and θ¯(ηi), parameter choices, and calibration are in appendix B.3.3.

Figure 3.
Under Sorting on Gains, the Costs of the Congestion Externality Will Be Concentrated on Complex Births

Figure plots, by complexity: (a) The prior probability of facility birth [equation (3)]; (b) the marginal reduction in facility quality due to the subsidy [equation (B.4)]; and (c) the social marginal cost from the externality [equation (6)]. Details of the parametric functional forms used for πh(ηi)πf(F/X,ηi) and θ¯(ηi), parameter choices, and calibration are in appendix B.3.3.

Close modal

Impact of preexisting capacity.

When there is no sorting on gains, the initial equilibrium is independent of capacity, and higher preexisting capacity will unambiguously mean a subsidy is more beneficial (or less harmful) because higher capacity (1) increases the health gain for marginal births; (2) means the identical absolute increase in facility usage translates into a smaller increase in congestion F/X; and (3) means the marginal increase in congestion will be less harmful because higher-capacity areas are starting from a lower initial level of congestion (assumption E3). When there is sorting on gains, including in the extreme case of the Roy model (λ=1), how capacity affects the impact of a marginal subsidy will be theoretically ambiguous, as we document in appendix B.3. Thus, this is ultimately an empirical question. Intuitively, the same three factors as listed above go in the direction of a subsidy being more useful, or less harmful, in higher-capacity areas. However, higher-capacity areas will have more births already occurring in facilities, and thus more births will suffer from the increase in congestion.

E. Conclusions from the Model

Our model illustrates that the impact of a demand subsidy for facility delivery is theoretically ambiguous, rendering it a key question for empirical research. It highlights that the impact of a subsidy will be composed of two key effects: first, the benefits/costs to the marginal births; and, second, social costs to all facility births due to the congestion externality.

The framework illustrates that the ways in which health impacts from the subsidy vary across the distribution of birth complexity are informative about whether or not a congestion externality is at play, or whether impacts can be explained by the aggregation of private gains/losses to the marginal births alone. Conditional on being marginal to the subsidy, private benefits are likely to be concentrated on more-complex births and any private costs concentrated on less-complex births. Conversely, any social costs due to a congestion externality will always be concentrated on complex births. This insight is key to how we interpret our findings in section VE.

The model shows that a positive subsidy is optimal if the marginal births benefit from being moved into facilities and the magnitude of this benefit outweighs the externality cost borne by births already occurring in facilities. This is more likely to be the case if facilities are less crowded (more capacity, less usage) to begin with, and if it is more-complex births (who have more to gain from facility delivery) who are marginal to the subsidy.

JSY was launched in April 2005 with the aim of reducing maternal and infant mortality through increasing rates of institutional delivery (Ministry of Health and Family Welfare Government of India, 2005).

Many life-saving procedures can only be carried out in health facilities or by a trained birth attendant (Lawn et al., 2005), which creates many reasons why facility delivery could improve birth outcomes, especially for more-complex births. However, evidence of poor quality in the form of few health professionals compounded by absenteeism, a lack of physical resources, and the presence of dangerous pathogens could all reduce the potential gains from facility delivery or even make it more risky than home birth (Chaudhury et al., 2006; Das et al., 2008; Mohanan et al., 2015; Das et al., 2016). We note, though, that under a congestion externality it could be the case that all births gain privately from facility delivery yet the aggregate impact of increasing facility delivery on birth outcomes is negative.

We focus on JSY's effects on rural households in states that the scheme designated as “Low Performing States” based on their low prior institutional delivery rates. In rural areas of these states, JSY is universal and cash incentives are more generous than in either “High Performing States” or in urban areas. Specifically, JSY provides a financial incentive of Rs.1400, or USD 32, to all pregnant women who give birth in a government health facility.9 JSY also provides an incentive of Rs. 600, or USD 14, to community health workers for every pregnant woman they bring to a facility. JSY did not provide new funding to help facilities expand obstetric services (Ministry of Health and Family Welfare Government of India, 2005).

We focus our analysis on the households primarily targeted by JSY: rural households in the nine states designated by JSY as “Low Performing States.”10 Together, the rural population of these states comprised 33.9% of India's population in the 2001 census and 48.7% of India's deaths of infants under seven days of age between 1990 and 2001.11

A. Perinatal Mortality and Place of Birth

Data on perinatal mortality, defined as stillbirth after 22 weeks of pregnancy or death within 7 days of birth, come from the 2007–2008 DLHS-3 pregnancy roster for ever-married women, which recorded all pregnancies since January 1, 2004. We focus on perinatal mortality given its particular sensitivity to the quality of care during the birth (World Health Organization, 2006). As our sample, we use all births (i.e., live births and stillbirths but excluding miscarriages and abortions) that occurred between January 1, 2004 and December 31, 2007 and within nine quarters of JSY's rollout in the district as our sample. We thus drop the 36 districts where JSY was not rolled out until after quarter 4 of 2007, which leaves us 256 districts. Table 1 provides sample descriptives. Place of birth is available for each respondent's most recent birth, and thus effects here should be interpreted as averages for this subpopulation.

Table 1.

Sample Descriptives

MeanStandard DeviationNo.
Birth outcomes    
Perinatal Mortality 0.0398 0.195 104,057 
Stillbirth 0.0153 0.123 104,057 
Died within 7 days | Live birth 0.0248 0.156 102,461 
Place of birth    
Home 0.723 0.447 81,242 
Government Hospital/CHC 0.124 0.330 81,242 
PHC 0.0598 0.237 81,242 
Private/NGO Hospital/Clinic 0.0926 0.290 81,242 
Maternal characteristics    
Age 24.34 5.432 104,057 
Husband's age 30.53 11.20 104,057 
Age at marriage 16.32 3.413 104,057 
Years of education 7.184 2.057 104,057 
Household characteristics    
Scheduled caste 0.198 0.399 104,057 
Scheduled tribe 0.164 0.370 104,057 
Other backward class 0.446 0.497 104,057 
Below poverty line 0.333 0.471 104,057 
Hindu 0.860 0.347 104,057 
Muslim 0.122 0.327 104,057 
MeanStandard DeviationNo.
Birth outcomes    
Perinatal Mortality 0.0398 0.195 104,057 
Stillbirth 0.0153 0.123 104,057 
Died within 7 days | Live birth 0.0248 0.156 102,461 
Place of birth    
Home 0.723 0.447 81,242 
Government Hospital/CHC 0.124 0.330 81,242 
PHC 0.0598 0.237 81,242 
Private/NGO Hospital/Clinic 0.0926 0.290 81,242 
Maternal characteristics    
Age 24.34 5.432 104,057 
Husband's age 30.53 11.20 104,057 
Age at marriage 16.32 3.413 104,057 
Years of education 7.184 2.057 104,057 
Household characteristics    
Scheduled caste 0.198 0.399 104,057 
Scheduled tribe 0.164 0.370 104,057 
Other backward class 0.446 0.497 104,057 
Below poverty line 0.333 0.471 104,057 
Hindu 0.860 0.347 104,057 
Muslim 0.122 0.327 104,057 

B. Capacity

We construct measures of preexisting capacity in each district using three inputs that are easy to measure and over which national guidelines exist: (i) beds, (ii) doctors/medical officers, and (iii) nurses/midwives. We distinguish between inputs in primary-care and secondary-care facilities.12 We use the facility survey of the DLHS-2, which, in 2003, surveyed government health facilities in 182 of the 256 districts we focus on in this paper. These data cover all secondary-care health facilities and a random sample of primary-care health centers (PHCs). We estimate the number of beds, doctors, and nurses/midwives per 10,000 of the rural population in each district's primary- and secondary-care facilities using population figures from the 2001 census. Further details and density plots of these measures are given in figure A1.

We use exploratory factor analysis to assess the dimensionality of these six measures and to create summary indices. We first run a single factor analysis on all six measures, which estimates that there are two orthogonal factors with eigenvalues greater than 1 (the Kaiser criterion for retaining a factor) which implies that these measures can be well summarized by two underlying factors: one primarily reflecting secondary capacity and the other primary capacity (factor loadings in columns 1 and 2 of table A1). We next run two separate factor analyses for the primary and secondary capacity measures to create summary indices for each without imposing orthogonality. These are the indices that we use in our analysis, and columns 3 and 4 show the loadings. In our main analysis, we use an indicator of whether the district had above- or below-median preexisting capacity in the primary and/or secondary system based on these factor measures. We use the continuous measures in robustness analysis.

C. JSY Rollout

We estimate JSY's rollout from the DLHS-3. Each respondent was asked whether they had received a payment through JSY or another state-specific scheme for their last birth. Once JSY was rolled out in a district, all births in government institutions should have received payments. We define JSY as being active in a district from the first quarter (after its official launch) in which 25% of births that occurred in government institutions were reported to have received a JSY payment in both that quarter and the following year.13 There is strong variation in the timing of implementation within and between states (figure A2). In robustness analysis, we use an alternative, fractional measure of JSY intensity—the proportion of eligible births that received the payment (see figure A3).

As we hope to use the rollout of JSY to identify causal effects, we now examine how the rollout relates to longer-run trends in our outcomes of interest. A correlation between the rollout and already-existing trends could suggest that parallel-trends assumptions required for our differences-in-differences approach to be valid might not be plausible. We use the DLHS-2 to calculate district-level changes in rates of institutional delivery, 7-day infant mortality, and vaccinations between 1990 and 2001.14 We then run district-level regressions of JSY's start date on these changes interacted with the prior capacity of the district (table A2). Columns 1, 4, and 7 show that the start date of JSY is uncorrelated with long-run changes in these variables, confirming the finding of Powell-Jackson et al. (2015) that the rollout appears uncorrelated with prior trends. The other columns of table A2 show that the start date of JSY is correlated with capacity in the primary- and secondary-care sectors, but not with the interaction of preexisting capacity with the long-run changes. Figures A4, A5, and A6 in appendix A provide a more detailed analysis and find no evidence that districts where JSY rolled out at different points were on different trends before the program.

A. Empirical Strategy

We seek to assess the causal impact of JSY on place of birth and perinatal mortality and how these impacts vary by the preexisting capacity of the district health system. We exploit the rollout of JSY across districts between its formal launch in April 2005 and the end of 2007. Impacts may vary with the time JSY had been operational (“event time”) if, for example, it took time for all households to hear about the scheme or if facilities adapted gradually to the new demand. For each outcome, we thus begin our analysis with an event study to estimate the effect of JSY at each event-time period k. Specifically, we estimate the following linear probability model:
Yibdt=α+k=-99βk1{Kibdt=k}+θb+θd+θt+νibdt,
(8)
where Yibdt denotes our outcome of interest (perinatal mortality or place of birth) for birth i, of birth order b, in district d, in quarter t. Kibdt denotes the event time (in quarters) since JSY became active in district d, with Kibdt=0 in the quarter of initiation. θb, θd, and θt are, respectively, birth-order, district, and quarter-of-birth fixed effects. νibdt is the error term, which we allow to be arbitrarily correlated among births within a district over time by clustering standard errors at the district level (Bertrand et al., 2004).

We do not have a large “always untreated” sample, as JSY had rolled out to almost all districts by the end of our sample period. Therefore, we must restrict two lags to identify the model (Borusyak et al., 2021); we thus set β-9, the first event period, and β-1, the period immediately before JSY's rollout, to zero. The specification allows us to test for differences in nonlinear pretrends—by testing the null hypothesis H0:β-8=β-2=0—but not for differences in the linear component of pretrends (Borusyak et al., 2021). This test complements our earlier analysis, in section IVC, which suggested that the rollout was unrelated to longer-run trends in the outcomes of interest.

Our identifying assumption is that, conditional on district, quarter-of-birth, and birth order, birth-specific shocks are mean independent of JSY's rollout, that is, E[νibdt|1{Kibdt=k},θb,θd,θt,k={-8,...,-2,0,...,9}]=0. This is a multiperiod parallel-trends assumption and rules out that the rollout of JSY was related to birth-specific shocks but does not rule out that the rollout was related to birth-specific gains from JSY. Under this assumption, and the assumption of no systematic heterogeneity in treatment effects across cohorts (Sun and Abraham, 2021), βk is the average causal effect of JSY k periods after it began.15

To examine dynamic treatment effects by preexisting capacity, we repeat this analysis but interact JSY's rollout with capacity indicators:
Yibdt=α+k=-99βk1{Kibdt=k}+k=-99γkCd1{Kibdt=k}+θb+θd+θt+νibdt,
(9)

where Cd is a vector containing measures of the capacity of district d prior to the rollout and where we again impose effects at k=-9 and -1 to be zero. The identifying assumption here is E[νibdt|1{Kibdt=k},Cd1{Kibdt=k},θb,θd,θt,k={-8,...,-2,0,...,9}]=0. This rules out that the rollout within districts with the same preexisting capacity level was related to birth-specific risks once conditioning on quarter of birth and birth order. In our main analysis, Cd is simply a binary indicator for district d having above-median preexisting capacity; in this case, βk represents the effect of JSY in low-capacity districts k periods after it began and βk+γk represents JSY's effect in high-capacity districts.

Where we do not reject the null of no differential nonlinear pretrends, we next place more structure on our analysis to increase precision while maintaining the ability to pick up treatment effects that vary over time. Thus, using the identical sample, we estimate a model that allows the dynamic treatment effect associated with JSY to comprise both a level shift and a trend break:
Yibdt=α+βshiftJSYdt+βbreakJSYdt×Kibdt+θb+θd+θt+νibdt,
(10)

where JSYdt is an indicator for whether JSY had begun in district d and quarter t, that is, JSYdt=1{Kibdt0}. We estimate the identical model allowing for an interaction between treatment terms and preexisting capacity to examine heterogeneity.

Finally, we summarize the dynamic treatment effects—estimated using specifications equations (8) and (9)—into a single average effect. To do so, we follow Borusyak et al. (2021); and estimate a weighted average effect, β^average, and corresponding standard error se(β^average)^, using weights proportional to sample size at each event time:
β^average=k=09wkβ^k,se(β^average)^=k=09wkse(β^k^),wk=i1{Kibdt=k}k'=09i1{Kibdt=k'}.
(11)

When treatment effects vary with time, this approach is preferable to a two-way fixed effects regression, which recovers a weighted average of treatment effects that places negative weight on longer-run effects (Borusyak et al., 2021). To increase precision, we estimate dynamic effects under the restriction that the coefficients on the lags of JSY's rollout, β-8 through β-2, are zero after first testing this restriction.

B. Impacts on Institutional Delivery

Our event studies indicate that JSY led to a very substantial increase in the rate of institutional delivery and that the impact of JSY increased with the time it was operational in a district (see figure 4a). After 2 years of operation, the analysis suggests that JSY had increased the probability that a birth took place in a medical facility by around 20 percentage points. Dynamic effects look similar across areas with different preexisting capacities, and we cannot reject that effects prior to JSY's rollout are zero for all levels of preexisting capacity (figure A7).
Figure 4.
Dynamic Effects of JSY

Figures plot dynamic effects of JSY on (a) Institutional delivery (all districts), (b) perinatal mortality in low secondary-care capacity districts, and (c) perinatal mortality in high secondary-care capacity districts. Each figure plots (1) the event-time effects [equations (8) and (9)]. We plot coefficients βk (in graphs a and b), k=-9,...,9 and βk+γk (in graph c); (2) dynamic effects allowing for a level shift and a trend break as estimated from equation (10); (3) average effects of JSY as defined in equation (11).

Figure 4.
Dynamic Effects of JSY

Figures plot dynamic effects of JSY on (a) Institutional delivery (all districts), (b) perinatal mortality in low secondary-care capacity districts, and (c) perinatal mortality in high secondary-care capacity districts. Each figure plots (1) the event-time effects [equations (8) and (9)]. We plot coefficients βk (in graphs a and b), k=-9,...,9 and βk+γk (in graph c); (2) dynamic effects allowing for a level shift and a trend break as estimated from equation (10); (3) average effects of JSY as defined in equation (11).

Close modal

Next, we estimate the average of these dynamic effects, weighted according to the event-time distribution of our sample [equation (11)]. Overall, JSY increased the probability of institutional delivery by 7.86 percentage points (p<0.001), or 35% of the mean before the policy's launch, across all sample districts (see table 2, panel A, column 1) and 7.62 percentage points across districts with capacity data available (column 2). These effect sizes are in line with those found by Powell-Jackson et al. (2015). The increase appears similar in districts with more and with less prior capacity (columns 3–5).

Table 2.

Effect of JSY on Institutional Delivery and Perinatal Mortality by Pre-Existing Capacity

Panel A: Institutional delivery
(1)(2)(3)(4)(5)
JSY 0.0786*** 0.0762*** 0.0769*** 0.0830*** 0.0748*** 
 (0.0095) (0.0114) (0.0170) (0.0127) (0.0181) 
JSY × High secondary cap.   0.0018  0.0183 
   (0.0170)  (0.0208) 
JSY × High primary cap.    −0.0143 0.0152 
    (0.0143) (0.0239) 
JSY × High sec. × High prim.     −0.0433 
     (0.0313) 
Observations 81242 59901 59901 59901 59901 
Number of districts 256 182 182 182 182 
Mean Prior to 2005Q2 0.2240 0.2260 0.2260 0.2260 0.2260 
Panel B: Perinatal mortality 
 (1) (2) (3) (4) (5) 
JSY 0.0015 0.0032 0.0090** 0.0048 0.0111** 
 (0.0027) (0.0033) (0.0039) (0.0038) (0.0044) 
JSY × High secondary cap.   −0.0082**  −0.0117** 
   (0.0035)  (0.0052) 
JSY × High primary cap.    −0.0043 −0.0040 
    (0.0034) (0.0041) 
JSY × High sec. × High prim.     0.0061 
     (0.0063) 
Observations 104057 76804 76804 76804 76804 
Number of districts 256 182 182 182 182 
Mean Prior to 2005Q2 0.0347 0.0370 0.0370 0.0370 0.0370 
Panel A: Institutional delivery
(1)(2)(3)(4)(5)
JSY 0.0786*** 0.0762*** 0.0769*** 0.0830*** 0.0748*** 
 (0.0095) (0.0114) (0.0170) (0.0127) (0.0181) 
JSY × High secondary cap.   0.0018  0.0183 
   (0.0170)  (0.0208) 
JSY × High primary cap.    −0.0143 0.0152 
    (0.0143) (0.0239) 
JSY × High sec. × High prim.     −0.0433 
     (0.0313) 
Observations 81242 59901 59901 59901 59901 
Number of districts 256 182 182 182 182 
Mean Prior to 2005Q2 0.2240 0.2260 0.2260 0.2260 0.2260 
Panel B: Perinatal mortality 
 (1) (2) (3) (4) (5) 
JSY 0.0015 0.0032 0.0090** 0.0048 0.0111** 
 (0.0027) (0.0033) (0.0039) (0.0038) (0.0044) 
JSY × High secondary cap.   −0.0082**  −0.0117** 
   (0.0035)  (0.0052) 
JSY × High primary cap.    −0.0043 −0.0040 
    (0.0034) (0.0041) 
JSY × High sec. × High prim.     0.0061 
     (0.0063) 
Observations 104057 76804 76804 76804 76804 
Number of districts 256 182 182 182 182 
Mean Prior to 2005Q2 0.0347 0.0370 0.0370 0.0370 0.0370 

Standard errors (clustered by district) in parentheses. Effects are weighted (weights proportional to sample size, defined in (11)) averages of dynamic effects (estimated using specifications (8) and (9)). All estimates control for quarter-of-birth, birth-order, and district fixed effects. The first column includes data for all sample districts. The remaining columns include data only for districts with capacity data available. High secondary cap. (High primary sec.) is an indicator equalling 1 if district has above-median secondary- (primary-) care capacity, as defined in section IVB. *p < 0.1, **p < 0.05, and ***p < 0.01.

As we see in table A3 in appendix A, breaking this down into different types of institutional deliveries, we see that this increase was comprised of an increase of 6.30 percentage points (p<0.001) in the probability of births occurring in government-run secondary-care facilities, an increase of 3.14 percentage points (p<0.001) for government-run primary-care facilities, and a decrease of 1.58 percentage points (p=0.003) for private facilities (in most states, payments were not given for births in private facilities).

The magnitudes of these increases are large relative to the prior rate of institutional delivery and the capacity of the government health system. A back-of-the-envelope calculation suggests that an increase in the proportion of births occurring in government facilities (secondary and primary) of 9.44 percentage points (table A3) translates into an increase in the average number of births per day occurring in each government-run secondary-care facility, or in a government-run primary-care facility under the supervision of that secondary facility, from 1.73 before JSY to 3.59 afterwards. Even though the increase in institutional delivery was the same across districts with more and with less capacity in the secondary-care system, the difference in the number of facilities means the increase in caseload was more pronounced in districts with low preexisting secondary-care capacity (from 1.92 to 4.80) than with higher capacity (1.54 to 2.38).16

C. Impacts on Perinatal Mortality

We now move on to assess impacts on perinatal mortality. In line with Powell-Jackson et al. (2015), we find that the average effect, across districts with high and low preexisting capacity, of JSY is close to zero and not statistically significant; this holds both in terms of the dynamic effects (figure A8a) and when averaged over event time (table 2, panel B, columns 1 and 2).

However, the results are markedly different in districts with below-median preexisting capacity in the secondary health care system from districts with above-median capacity. In the former, low-capacity districts, figure 4b shows that JSY increased the risk of perinatal mortality, with effects growing somewhat in magnitude over time. The average impact across event times (table 2, panel B, column 3) is an increase in the risk of perinatal mortality of 0.90 percentage points (p=0.020). In contrast, in districts with above-median secondary capacity, JSY had no impact at any event time or averaged across event times (p=0.811), with the difference in the average effect between these above-median capacity districts and the districts with low capacity being statistically significant (p=0.022). These magnitudes are substantial compared to the rate of perinatal mortality of 3.70% in the 15 months before JSY's launch.

Although point estimates for the effect of JSY of perinatal mortality in districts with low primary-care capacity are positive (an increase in mortality risk), these are not statistically significant (table 2, panel B, columns 4 and 5). Neither is the difference in effects between districts with high and low primary-care capacity.

D. Robustness

In appendix A, we explore the robustness of our results. We show results using a static two-way-fixed-effects specification, that is, Yibdt=α+β1{Kibdt0}+θb+θd+θt+νibdt. Effect sizes for perinatal mortality (table A4) are nearly identical to the weighted average of the dynamic effects, which is unsurprising as the dynamic effects do not vary substantially with time (figure 4). For institutional delivery, the static specification produces smaller, but still highly statistically significant, estimates (table A5), which is expected given that, as shown in figure 4, the dynamic effects increase with time (Borusyak et al., 2021). We further show that results are robust to using a probit model (tables A6 and A7) and to using a fractional indicator of JSY's rollout (table A8 and A9), to using continuous measures of preexisting capacity (tables A10 and A11), and to allowing for differential time trends by preexisting capacity (tables A12 and A13). In table A14, we show that we find no evidence of heterogeneity in the impact of JSY on perinatal mortality across districts that adopted JSY earlier and later.

Given that health system capacity was not randomly assigned, heterogeneity by capacity could, theoretically, be capturing heterogeneity across distinct but correlated characteristics of districts (see table A15). We mostly rely on our conceptual framework to motivate why system capacity is a natural dimension over which heterogeneity would arise. However, we also use LASSO penalized regression to analyze which district characteristics predict heterogeneity in the impact of JSY on perinatal mortality. We allow for heterogeneity to be predicted by district-level measures of poverty, men's education, women's education, distance from different health facilities, and our two measures of system capacity. Figure A9 shows that at all levels of penalization, the standardized coefficients on both measures of capacity are greater than all alternative predictors. Moreover, LASSO sets coefficients on all other predictors to zero well before setting either of the capacity coefficients to zero. This reassures us that the heterogeneity we have recovered is not simply standing in for heterogeneity across other dimensions.

E. Mechanisms

We have shown that JSY increased the perinatal mortality risk in areas with low preexisting capacity in the secondary-care system. Our theoretical model demonstrates that two mechanisms could have led to this. Even in the absence of any congestion externality, JSY could harm health if the marginal births moved by the subsidy were riskier when taking place in a facility rather than at home. As discussed in section III, if facilities are poor quality, this is not implausible. However, our model also suggests a second mechanism: that the overall increase in facility delivery rates increased congestion in facilities, leading to a deterioration in the quality of medical care for all facility births, including those that would have taken place in facilities even in the absence of the subsidy.

In this section, we explore the mechanisms at play. We begin by looking at how the impacts of JSY differed by birth complexity as an attempt to shed light on the role of a congestion externality in driving the negative effects of JSY. To this end, we create a proxy for birth complexity by using variables containing information that would have been known to families at the time they were choosing where to deliver: complications experienced during pregnancy, the number of previous pregnancies, whether the woman had previously had a miscarriage, whether she had previously had a stillbirth, and whether she had previously given birth to a baby who subsequently died. We use these variables, and their interactions, to predict perinatal mortality in the full sample for which this information is available (the sample of last births). To do so, we used penalized LASSO regression to avoid overfitting. We then predict a risk index using the chosen predictors. See table A17 in appendix A for details. We divide births into those with a risk index above and below the median.

In column 1 of table 3, we estimate the impact of JSY on institutional delivery using methods identical to those we employed earlier but now interacting the secondary health system capacity with our proxy for birth complexity in addition to JSY's rollout. The first important thing to notice here is that even in the absence of JSY, in both low-capacity (line e) and high-capacity areas (line f), more-complex births were more likely to occur in facilities than less-complex births, suggesting that births select into health facilities at least partially on this basis of their expected complexity. While this suggests that, prior to JSY, births sorted into facilities somewhat based on gains, lines (a)–(d) of column 1 show that JSY drew in more-complex and less-complex births to the same degree, and this was true in both low- and high-capacity areas.17 We argued in section II that for JSY to stand the best chance of being effective at improving health outcomes, it would have to disproportionately induce the most-complex births into health facilities. We see here that this does not seem to have been the case.

Table 3.

Impacts on Institutional Delivery, Perinatal Mortality, and Post-Birth Checkups by Birth Complexity

Institutional DeliveryPerinatal MortalityCheckup day 1 | FacilityCheckup days 2–10 | Facility
(1)(2)(2)(3)
(a) JSY × Low sec. × Low risk 0.0782*** 0.0041 −0.0042 −0.0666*** 
 (0.0178) (0.0045) (0.0250) (0.0243) 
(b) JSY × Low sec. × High risk 0.0755*** 0.0159*** −0.0373 −0.0749*** 
 (0.0175) (0.0053) (0.0257) (0.0260) 
Difference (b−a) −0.0027 0.0118*** −0.0331* −0.0083 
 (0.0099) (0.0043) (0.0175) (0.0203) 
(c) JSY × High sec. × Low risk 0.0766*** 0.0004 0.0231 −0.0000 
 (0.0132) (0.0035) (0.0234) (0.0261) 
(d) JSY × High sec. × High risk 0.0802*** 0.0084* 0.0334 −0.0329 
 (0.0154) (0.0044) (0.0261) (0.0262) 
Difference (d−c) 0.0036 0.0079* 0.0102 −0.0328 
 (0.0135) (0.0041) (0.0228) (0.0231) 
(e) Low sec. × High risk 0.0526*** 0.0176*** 0.0530*** 0.0444** 
 (0.0063) (0.0031) (0.0184) (0.0173) 
(f) High sec. × High risk 0.0701*** 0.0160*** 0.0192 0.0621*** 
 (0.0124) (0.0038) (0.0218) (0.0226) 
Observations 59,880 59,888 16,498 15,858 
Number of districts 182 182 182 182 
Mean Prior to 2005Q2 0.2260 0.0370 0.6972 0.4225 
Institutional DeliveryPerinatal MortalityCheckup day 1 | FacilityCheckup days 2–10 | Facility
(1)(2)(2)(3)
(a) JSY × Low sec. × Low risk 0.0782*** 0.0041 −0.0042 −0.0666*** 
 (0.0178) (0.0045) (0.0250) (0.0243) 
(b) JSY × Low sec. × High risk 0.0755*** 0.0159*** −0.0373 −0.0749*** 
 (0.0175) (0.0053) (0.0257) (0.0260) 
Difference (b−a) −0.0027 0.0118*** −0.0331* −0.0083 
 (0.0099) (0.0043) (0.0175) (0.0203) 
(c) JSY × High sec. × Low risk 0.0766*** 0.0004 0.0231 −0.0000 
 (0.0132) (0.0035) (0.0234) (0.0261) 
(d) JSY × High sec. × High risk 0.0802*** 0.0084* 0.0334 −0.0329 
 (0.0154) (0.0044) (0.0261) (0.0262) 
Difference (d−c) 0.0036 0.0079* 0.0102 −0.0328 
 (0.0135) (0.0041) (0.0228) (0.0231) 
(e) Low sec. × High risk 0.0526*** 0.0176*** 0.0530*** 0.0444** 
 (0.0063) (0.0031) (0.0184) (0.0173) 
(f) High sec. × High risk 0.0701*** 0.0160*** 0.0192 0.0621*** 
 (0.0124) (0.0038) (0.0218) (0.0226) 
Observations 59,880 59,888 16,498 15,858 
Number of districts 182 182 182 182 
Mean Prior to 2005Q2 0.2260 0.0370 0.6972 0.4225 

Standard errors (clustered by district) in parentheses. Effects are weighted (weights proportional to sample size, defined in equation (11)) averages of dynamic effects (estimated using specification (9)). All estimates control for quarter-of-birth, birth-order, and district fixed effects. High sec. is an indicator equalling 1 if district has above-median secondary care capacity, as defined in section IVB. More-complex (less-complex) biths are defined as those with above (below) median values for the LASSO-predicted index of birth complexity (detailed in table A17 in appendix A). *p < 0.1, **p < 0.05, and ***p < 0.01.

Next, we turn to how JSY's impacts on perinatal mortality differed by births' complexity in column 2. In rows (a) and (b) we see that practically all the increase in perinatal mortality in low-capacity areas was driven by more-complex births. We estimate that the perinatal mortality risk of these complex births in low-capacity areas increased by 1.59 percentage points due to JSY, while less-complex births saw no change. The difference between the two is highly statistically significant. Interestingly, we even see tentative evidence that in higher-capacity areas too, more-complex births may have suffered due to JSY, although to a lesser extent than in low-capacity areas. The fact that the harms of JSY appear to be concentrated on more-complex births despite JSY having moved more-complex and less-complex births into facilities at the same rate suggests that a congestion externality is key to explaining JSY's perverse effects.

To see this, first imagine the special case of our model with no congestion externality. In this case, all of the aggregate harms due to the subsidy are made up of private harms to the marginal births. In our model, conditional on moving the same proportion of less-complex and more-complex births into facilities (which we see empirically), the more-complex births would privately gain more or be harmed less. Therefore, the fact that harms appear concentrated on complex births is inconsistent with the version of our model without a congestion externality.

However, our model predicts that under sorting on gains (which we see empirically), the overall cost of the externality will be disproportionately borne by complex births. This is the case for two reasons. First, a disproportionate number of complex births will already be occurring in facilities and thus will be subject to the externality. Second, we anticipate (as set out in assumption E2) that congestion will harm more-complex facility births the most. Thus, the fact that we find clear evidence that the harms associated with JSY are concentrated on complex births is evidence that a congestion externality is crucial to understanding our results.

Besides this indirect evidence on the importance of congestion externalities, the results in columns 3 and 4 provide more direct evidence on congestion. Here we look at whether respondents reported that their baby received a checkup after their facility birth. Note that we condition on births occurring in facilities here as it would be surprising if moving births into facilities did not result in some increase in the proportion reporting a check from a health professional;18 instead, we want to focus on the quality of care provided by facilities. In column 3, we look at whether or not respondents reported that their baby received a checkup within the first 24 hours. Overall, we do not see evidence that JSY changed the likelihood of this, which, given that even a very cursory checkup from a health worker would be counted here, is perhaps not surprising. In column 4 we look at whether babies received a checkup between days 2 and 10. We take this as an indicator of how long women remain in facilities after birth, and the intensity of the continued care they receive (both in facilities and in the community). Here we find that in low-capacity areas (but not in high-capacity areas) JSY reduced the probability of such checkups by around seven percentage points (lines a and b). We interpret this as evidence that in low-capacity areas, facilities reacted to congestion by either discharging women and babies earlier or by providing less continued care. Furthermore, the fact that this reduction was equal for more-complex (line b) and less-complex births (line a) suggests that facilities did not prioritize maintaining more-intensive care for more-complex births in reacting to this congestion.

F. Spillovers onto Vaccination Rates

We employ the same empirical strategy to examine spillover impacts of JSY on another service that the local health system performs: childhood vaccinations. Our outcome here is the proportion of vaccinations a child has received of those that a child of the same age should have received.19 Given that polio vaccinations are largely administered by a parallel system, the “Pulse Polio Initiative,” we focus on the six nonpolio childhood vaccinations that are delivered through the public health system.

In Table 4, we find that, averaged across event time, JSY reduced the proportion of the recommended vaccinations that children aged 9 months or older had received by 2.05 percentage points (p=0.007). Breaking this down by the six nonpolio recommended vaccinations, we find that JSY significantly reduced the probability that children received four of these recommended vaccinations. Interestingly, the negative effect of JSY on vaccination rates is smaller in vaccines given closer to birth than in those given later. This is consistent with JSY reallocating resources and reducing vaccination rates, but this is offset for vaccines given at an early age because JSY also led to an increase in institutional deliveries, which could have increased parents' attachment to health care providers right after birth.

Reductions in vaccination rates do not vary with preexisting capacity (figure A10, table A18).20 In appendix A, we show these findings are robust to using a static specification (table A19), to using a probit model (A.20), and to using a fractional indicator of JSY intensity (A.21). In appendix C, we argue that these results are not driven by differential selection caused by JSY's impacts on mortality.

G. Tentative Evidence on Medium-Run Effects

We have provided evidence that in the short run JSY caused an increase in perinatal mortality in areas with low preexisting rates of capacity in the secondary-care system. A crucial question is whether this was a short-term effect, in which case the impacts could have been due to transition difficulties, or whether JSY's negative impacts persisted. Here, we briefly assess whether there is any evidence of persistence. To do this, we draw on the birth record of the fourth National Family Health Survey (2014–2015). The NFHS-4 records detailed information, including JSY uptake and place of delivery, of all births since 2010 in addition to mortality data for a longer history of births.

In the short term, we used the staggered rollout of JSY across districts to estimate its effects. However, by 2010 JSY had been operational for upwards of 4 years in most districts, making this approach less plausible. Instead, we explore how variation in JSY compliance between districts correlates with variation in both institutional delivery rates and perinatal mortality rates over the medium run. Specifically, we create an indicator for whether or not 90% or more of births that should have been eligible for a JSY payment (i.e., births that occurred in a government facility) between 2010 and 2015 reported receiving one. We use the identically constructed measures of pre-JSY health system capacity.

Beginning with rates of institutional delivery in table 5, we find that, after controlling for quarter-of-birth fixed effects, districts where compliance was greater than 90% had institutional delivery rates of 4.5 percentage points greater than in districts with lower compliance (column 1). Note that because we have no historical data from this data set on institutional delivery, we cannot control for district fixed effects.

Table 4.

Effect of JSY on Vaccination Rates

(1)(2)(3)(4)(5)(6)(7)
MeanBCGDPT-1DPT-2DPT-3MeaslesVitamin A
JSY −0.0205*** −0.00818 −0.0134* −0.0114 −0.0311*** −0.0322*** −0.0225** 
 (0.00757) (0.00741) (0.00805) (0.00993) (0.0109) (0.00909) (0.00920) 
Observations 79298 96284 95886 95175 93894 79298 79298 
Age of Administration (Months)  1.5 2.5 3.5 
Mean Prior to 2005Q2 0.639 0.786 0.717 0.614 0.510 0.648 0.557 
(1)(2)(3)(4)(5)(6)(7)
MeanBCGDPT-1DPT-2DPT-3MeaslesVitamin A
JSY −0.0205*** −0.00818 −0.0134* −0.0114 −0.0311*** −0.0322*** −0.0225** 
 (0.00757) (0.00741) (0.00805) (0.00993) (0.0109) (0.00909) (0.00920) 
Observations 79298 96284 95886 95175 93894 79298 79298 
Age of Administration (Months)  1.5 2.5 3.5 
Mean Prior to 2005Q2 0.639 0.786 0.717 0.614 0.510 0.648 0.557 

Standard errors (clustered by district) in parentheses. Effects are weighted [weights proportional to sample size, defined in equation (11)] averages of dynamic effects [estimated using specification (8)]. All estimates control for quarter-of-birth, birth-order, and district fixed effects. Sample for each regression includes all children who are older than the recommended age of administration for that vaccination (displayed here as “months at admin.” Sample for the mean coverage rate (column 1) includes only children older than 9 months who should have received all vaccinations. *p < 0.1, **p < 0.05, and ***p < 0.01.

Table 5.

Relationship between JSY Intensity, Institutional Delivery, and Perinatal Mortality after 2010

(1)(2)(3)(4)(5)
Inst. Del.Mort.Mort.Mort.Mort.
High Intensity 0.0452** 0.00710*** 0.00743* 0.00732** 0.00726* 
 (0.0218) (0.00249) (0.00438) (0.00368) (0.00385) 
High sec. cap. × High Intensity   −0.00462 −0.00907* −0.00662 
   (0.00577) (0.00533) (0.00556) 
Historical perinatal mort. rate    0.515***  
    (0.0894)  
Observations 24,277 38,589 27,815 27,815 110,805 
Quarter of Birth FEs × × × × × 
State FEs    ×  
District FEs     × 
Sample: Post-2010 births × × × × × 
Sample: Pre-2005 births     × 
(1)(2)(3)(4)(5)
Inst. Del.Mort.Mort.Mort.Mort.
High Intensity 0.0452** 0.00710*** 0.00743* 0.00732** 0.00726* 
 (0.0218) (0.00249) (0.00438) (0.00368) (0.00385) 
High sec. cap. × High Intensity   −0.00462 −0.00907* −0.00662 
   (0.00577) (0.00533) (0.00556) 
Historical perinatal mort. rate    0.515***  
    (0.0894)  
Observations 24,277 38,589 27,815 27,815 110,805 
Quarter of Birth FEs × × × × × 
State FEs    ×  
District FEs     × 
Sample: Post-2010 births × × × × × 
Sample: Pre-2005 births     × 

Standard errors (clustered by district) in parentheses. High secondary cap. is an indicator equalling 1 if district has above-median secondary-care capacity, as defined in section IVB. Pre-2005 sample comprises births from 2000 up until the launch of JSY in 2005, quarter 2. Historical perinatal mortality rates calculated using this pre-2005 sample. *p < 0.1, **p < 0.05, and ***p < 0.01.

We then proceed to analyze the medium-run effects on perinatal mortality. Column 2 shows that districts with 90% or higher JSY compliance had perinatal mortality rates that were 0.71 percentage points higher than districts with lower compliance (controlling for time trends). Columns 3–5 show that, consistent with our short-term results, this result is driven by low-capacity areas, although whether the difference is statistically significant depends on the precise specification. Note that these results are subject to some important limitations. In columns 3 and 4, where we only use births from 2010 onwards, the sample size is much smaller than in table 3 and we cannot include district fixed effects. In column 5, we append the data with births between 2000 and 2005, which enables us to include district-level fixed effects, although this assumes that the unobserved district-level components remain constant over that long period, which is a strong assumption.

While the extent to which different districts complied with JSY over the medium run may not be exogenous for many reasons, these associations are certainly consistent with JSY continuing to have perverse effects on perinatal mortality up to 10 years after it was first introduced. The associations are perhaps most convincing at suggesting that it is unlikely that the short-term harmful impacts of JSY in low-capacity areas turned beneficial over the medium run. In many ways, this seems unsurprising given that since JSY's introduction in 2005, India's public spending on health care has remained unchanged at less than 1% of GDP, far less than other rapidly growing economies. Low and steady levels of public expenditure since JSY's introduction perhaps gave little scope for a rapid improvement in the capacity of the supply side over this period.

Conditional cash transfer programs are frequently used to increase the uptake of health services, but evidence on their impact on health is mixed (Lagarde et al., 2009). In this paper, we bring to the fore the role of congestion externalities in explaining the detrimental effects of JSY on perinatal mortality in areas of India that have a low-capacity health care system. The overall impact of JSY on health is composed of both its effect on inframarginal facility births (those that would have taken place in a facility regardless of JSY) and its effect on marginal births (those births induced to take place in the facility because of JSY). Under a congestion externality, the effect on the inframarginal births will always be detrimental due to increased congestion. Hence, for JSY to improve outcomes, any positive effect on marginal births must outweigh the congestion externality.

The theoretical model highlights interesting interactions between the size of the congestion externality and what drives selection into institutional delivery. Under full selection on gains, the inframarginal facility births are the riskiest and, consequently, they are the most harmed by the additional congestion caused by JSY. Moreover, under full selection on gains, the births that are marginal to JSY are harmed from switching from home to facility delivery. Hence, in this case, JSY will have detrimental effects on health. On the contrary, if the initial selection into health facilities is unrelated to health gains, then there could be a large pool of potentially marginal births that could gain a lot from facility delivery. Furthermore, if more complex births are more responsive to the subsidy, this would give JSY a better shot at attracting the births that would particularly gain more from facility delivery.

Unfortunately, our empirical analysis reveals that JSY was not successful at attracting disproportionately complex births; more- and less-complex births were similarly attracted Moreover, we find evidence that in low-capacity districts, JSY generated additional and harmful congestion. In these districts, JSY decreased the likelihood that babies born in facilities received a postnatal checkup after the first day, which suggests that they were discharged sooner or simply that checkups were overlooked. We show that the detrimental effects of JSY were concentrated on complex births, a result that in our model can only be explained by the presence of a congestion externality. Overall, JSY caused perinatal mortality to increase in low-capacity districts, and more tentative evidence suggests that these harmful effects may have persisted for up to 10 years.

Our general equilibrium framework points to alternative approaches on how to use public funds to improve health outcomes when congestion externalities are a concern. First, policies that help identify complex cases and direct those cases toward health facilities could be particularly useful. Our framework demonstrates that, for a given degree of capacity and usage, increasing the complexity of the cases that select into formal health care will improve overall health. While there are often important barriers to patients discerning the usefulness of a given health care intervention and this is a fundamental barrier to individuals seeking out the optimal amount of care (Arrow, 1963), good-quality prenatal care could help direct more complex births towards facilities.

Second, our framework highlights the potential of policies that effectively increase the capacity of health care systems and the quality of care they provide, including both policies that increase the resources available and those that ensure those resources are used efficiently (Dupas and Miguel, 2017; Kremer and Glennerster, 2011; Björkman and Svensson, 2009). Our framework suggests that such efforts are likely to improve outcomes for existing users of health care and to attract more individuals to take up care. Also, our empirical results highlight the stakes of such efforts to improve quality, especially as efforts intensify to ensure universal access to health care. We found that in areas with few of the most basic health care inputs—beds, nurses, and doctors—increased use of health care harmed overall health.

1

For example, many conditional cash transfer programs condition payments on the take-up of health care (Lagarde et al., 2009).

2

See, for example, Das et al. (2016, 2008) and Mohanan et al. (2015).

3

The JSY payment is Rs. 1400 in the states on which we focus (see section III). The average daily wage for casual labor (across rural areas) in 2004/5 was Rs. 48.89 (NSS, 61st round).

4

Authors' calculations using District Level Household and Facility Survey-3 and National Family Health Survey-4.

6

For instance, poorer women may have higher average birth complexity if they have less access to prenatal care and might also have a higher marginal utility of consumption.

7

While it might be the case that the presence of health professionals and beds is correlated with other drivers of care quality, we consider it likely that these physical and human resources are causally important determinants of quality in our setting.

8

This is equivalent to a social planner maximizing welfare without knowledge of idiosyncratic preferences.

9

Exchange rate: 44 Rs./USD (April 2005). By contrast, in High Performing States only pregnant women with a Below Poverty Line card are eligible and transfers are lower (Rs. 700). Urban births also receive lower transfers (Rs. 1000).

10

These are Uttar Pradesh, Chhattisgarh, Bihar, Madhya Pradesh, Rajasthan, Assam, Orissa, Jharkhand, and Uttarakhand. Jammu and Kashmir was also designated a “Low Performing State” but we estimate that JSY only began in a single district before 2008 and so we drop it from all analysis.

11

Calculated from DLHS-2 using sample weights.

12

Secondary-care facilities are hospitals, community health centers, and first referral units.

13

The latter condition prevents erroneous reports of respondents receiving JSY leading us to mistakenly infer a too-early start date where there are few births recorded in a district-quarter cell.

14

The birth roster in the DLHS-2 does not include stillbirths, so we check for parallel pretrends only for 7-day mortality. We use “short” differences for vaccination rates (between 2001Q1 and 2002Q4) as DLHS-2 only contains vaccination information for recent births.

15

Table A14 in appendix A formally tests for evidence of cohort heterogeneity and finds no evidence of it.

16

Taking India's 2005 crude birth rate of 24.2 per thousand people (data.worldbank.org) we estimate the average number of births per day in each district using population figures from the 2001 census. We estimate the average number taking place each day in government health facilities both before JSY's rollout (using the pre-JSY observed rate in the district) and after (using this baseline rate plus 0.0944). We divide these figures by the number of secondary-care facilities in the district.

17

In itself, this result is not informative about the degree of selection on gains and/or the degree to which the marginal utility of consumption is increasing in birth complexity. As we can see from figure 2, the fact that the subsidy has the same average impact on usage above and below the median could be the result of different combinations of selection on gains and marginal utility of consumption.

18

Table A16 in appendix A shows that we do indeed see the overall rate of postnatal care increasing with JSY, although this increase is concentrated in high-capacity areas.

19

These data are available for each respondent's two most recent surviving children.

20

One reason why heterogeneity in effects might not mirror that found for perinatal mortality is that in areas with lower secondary-care capacity roughly one-half of the increase in institutional delivery came through primary-care facilities (table A3). This increase in the use of primary-care facilities for deliveries might have mitigated some of the detrimental effects of JSY on vaccination rates if giving birth there meant local health professionals could follow up more easily with the parents on the child's vaccinations.

Arrow
,
Kenneth J.
, “
Uncertainty and the Welfare Economics of Medical Care
,”
American Econmic Review
53
:
5
(
1963
),
941
973
.
Bertrand
,
Marianne
,
Esther
Duflo
, and
Sendhil
Mullainathan
, “
How Much Should We Trust Difference-in-Differences Estimates?
Quarterly Journal of Economics
119
:
1
(
2004
),
249
275
.
Björkman
,
Martina
, and
Jakob
Svensson
, “
Power to the People: Evidence from a Randomized Field Experiment on Community-Based Monitoring in Uganda
,”
Quarterly Journal of Economics
124
:
2
(
2009
),
735
769
.
Borusyak
,
K.
,
X.
Jaravel
, and
J.
Spiess
, “
Revisiting Event Study Designs: Robust and Efficient Estimation
,” (
2021
). arXiv:2108.12419.
Cesur
,
Resul
,
Pınar Mine
Güneş
,
Erdal
Tekin
, and
Aydogan
Ulker
, “
The Value of Socialized Medicine: The Impact of Universal Primary Healthcare Provision on Mortality Rates in Turkey,
Journal of Public Economics
150
(
2017
),
75
93
.
Chaudhury
,
Nazmul
,
Jeffrey
Hammer
,
Michael
Kremer
,
Karthik
Muralidharan
, and
F. Halsey
Rogers
, “
Missing in Action: Teacher and Health Worker Absence in Developing Countries
,”
Journal of Economic Perspectives
20
:
1
(
2006
),
91
116
.
Das
,
Jishnu
, and
Jeffrey
Hammer
, “
Are Institutional Births Institutionalizing Deaths?
World Bank Future Development Blog
(
2014
),
1
8
.
Das
,
Jishnu
,
Jeffrey
Hammer
, and
Kenneth
Leonard
, “
The Quality of Medical Advice in Low-Income Countries
,”
Journal of Economic Perspectives
22
:
2
(
2008
),
93
114
.
Das
,
Jishnu
,
Alaka
Holla
,
Aakash
Mohpal
, and
Karthik
Muralidharan
, “
Quality and Accountability in Health Care Delivery: Audit-Study Evidence from Primary Care in India
,”
American Economic Review
106
:
12
(
2016
),
3765
3799
.
Das
,
Jishnu
,
Liana
Woskie
,
Ruma
Rajbhandari
,
Kamran
Abbasi
, and
Ashish
Jha
, “
Rethinking Assumptions about Delivery of Healthcare: Implications for Universal Health Coverage,
BMJ
361
(
2018
), k1716.
Dumont
,
Etienne
,
Bernard
Fortin
,
Nicolas
Jacquemet
, and
Bruce
Shearer
, “
Physicians' Multitasking and Incentives: Empirical Evidence from a Natural Experiment
,”
Journal of Health Economics
27
:
6
(
2008
),
1436
1450
.
Dupas
,
P.
, and
E.
Miguel
, “
Impacts and Determinants of Health Levels in Low-Income Countries
” (pp.
3
93
), in
Abhijit V.
Banerjee
and
Esther
Duflo
(eds.)
,
Handbook of Economic Field Experiments
, Vol.
2
(
Amsterdam
:
Elsevier
,
2017
).
Freedman
,
Seth
, “
Capacity and Utilization in Health Care: The Effect of Empty Beds on Neonatal Intensive Care Admission
,”
American Economic Journal: Economic Policy
8
:
2
(
2016
),
154
185
.
Giedion
,
Ursula
, and
Beatriz Yadira
Díaz
, “
A Review of the Evidence
” (pp.
155
177
), in
Maria-Luisa
Escobar
,
Charles C.
Griffin
, and
R. Paul
Shaw
(eds.)
,
The Impact of Health Insurance in Low- and Middle-Income Countries
(
Washington, DC
:
Brookings Institution Press
,
2010
).
Gruber
,
Jonathan
,
Nathaniel
Hendren
, and
Robert M.
Townsend
, “
The Great Equalizer: Health Care Access and Infant Mortality in Thailand
,”
American Economic Journal: Applied Economics
6
:
1
(
2014
),
91
107
.
Hoe
,
Thomas P.
, “
Does Hospital Crowding Matter? Evidence from Trauma and Orthopedics in England
,”
American Economic Journal: Economic Policy
14
:
2
(
2022
),
231
262
.
Holmstrom
,
B.
, and
P.
Milgrom
, “
Multitask Principal-Agent Analyses: Incentive Contracts, Asset Ownership, and Job Design
,”
Journal of Law, Economics, and Organization
7
:
Special
(
1991
),
24
52
.
Hoot
,
Nathan R.
, and
Dominik
Aronsky
, “
Systematic Review of Emergency Department Crowding: Causes, Effects, and Solutions
,”
Annals of Emergency Medicine
52
:
2
(
2008
),
126
136
.
Kremer
,
Michael
, and
Rachel
Glennerster
, “
Improving Health in Developing Countries
” (pp.
201
315
), in
Handbook of Health Economics
, Vol.
2
(
Amsterdam
:
Elsevier B.V.
,
2011
).
Kumar
,
A. K. Shiva
,
Lincoln C.
Chen
,
Mita
Choudhury
,
Shiban
Ganju
,
Vijay
Mahajan
,
Amarjeet
Sinha
, and
Abhijit
Sen
, “
Financing Health Care for All: Challenges and Opportunities
,”
The Lancet
377
:
9766
(
2011
),
668
679
.
Lagarde
,
M.
,
A.
Haines
, and
N.
Palmer
, “
The Impact of Conditional Cash Transfers on Health Outcomes and Use of Health Services in Low and Middle Income Countries (Review)
,”
Cochrane Database of Systematic Reviews
2009
:
4
(
2009
), CD008137.
Lawn
,
Joy E.
,
Simon
Cousens
, and
Jelka
Zupan
, “
4 Million Neonatal Deaths: When? Where? Why?
The Lancet
365
:
9462
(
2005
),
891
900
.
Marks
,
Mindy
, and
Moonkyung Kate
Choi
, “
Baby Boomlets and Baby Health: Hospital Crowdedness, Hospital Spending, and Infant Health
,”
American Journal of Health Economics
5
:
3
(
2019
),
376
406
.
Miller
,
Grant
,
Diana
Pinto
, and
Marcos
Vera-Hernández
, “
Risk Protection, Service Use, and Health Outcomes under Colombia's Health Insurance Program for the Poor
,”
American Economic Journal: Applied Economics
5
:
4
(
2013
),
61
91
.
Ministry of Health and Family Welfare Government of India
, “
Janani Suraksha Yojana Guidelines for Implementation
” (
2005
).
Mohanan
,
Manoj
,
Marcos
Vera-Hernández
,
Veena
Das
,
Soledad
Giardili
,
Jeremy D.
Goldhaber-Fiebert
,
Tracy L.
Rabin
,
Sunil S.
Raj
,
Jeremy I.
Schwartz
, and
Aparna
Seth
, “
The Know-Do Gap in Quality of Health Care for Childhood Diarrhea and Pneumonia in Rural India
,”
JAMA Pediatrics
169
:
4
(
2015
),
349
357
.
Okeke
,
Edward N.
,
Zachary
Wagner
, and
Isa S.
Abubakar
, “
Maternal Cash Transfers Led to Increases in Facility Deliveries and Improved Quality of Delivery Care in Nigeria
,”
Health Affairs
39
:
6
(
2020
),
1051
1059
.
Powell-Jackson
,
Timothy
,
Sumit
Mazumdar
, and
Anne
Mills
, “
Financial Incentives in Health: New Evidence from India's Janani Suraksha Yojana
,”
Journal of Health Economics
43
(
September
2015
),
154
169
.
Rao
,
Mohan
,
Krishna D.
Rao
,
Ak Shiva
Kumar
,
Mirai
Chatterjee
, and
Thiagarajan
Sundararaman
, “
Human Resources for Health in India
,”
The Lancet
377
:
9765
(
2011
),
587
598
.
Sun
,
Liyang
, and
Sarah
Abraham
, “
Estimating Dynamic Treatment Effects in Event Studies with Heterogeneous Treatment Effects
,”
Journal of Econometrics
225
:
2
(
December
2021
),
175
199
.
World Health Organization
, “
Neonatal and Perinatal Mortality: Country, Regional and Global Estimates
,”
technical report
(
2006
).

Author notes

We thank Orazio Attanasio, Britta Augsburg, James Banks, Dita Eckardt, Casper Hansen, Sonya Krutikova, Peter Levell, Kate Smith, and workshop and seminar participants at the University of Essex and the Institute for Fiscal Studies for helpful comments and feedback. We would like to thank the Centre for Public Policy at the Institute for Fiscal Studies for generous financial support. In addition, this project has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (grant agreement No. 695300-HKADeC-ERC-2015-AdG).

A supplemental appendix is available online at https://doi.org/10.1162/rest_a_01206.

This is an open-access article distributed under the terms of the Creative Commons Attribution 4.0 International License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. For a full description of the license, please visit https://creativecommons.org/licenses/by/4.0/legalcode

Supplementary data