Health Impact Evaluation of Aspirational Districts Program in India: Evidence from National Family Health Survey

The Aspirational District Program (ADP) is a unique initiative of Government of India launched in 2018 that aims to reduce inter-district multidimensional inequality. ADP aims to bring the most backward districts to catch up with the rest of the other districts in the country. The program is comprehensive in its scope as it targets the improvement of several key development indicators spanning health and nutrition, education, agriculture and water resources, financial inclusion and skill development and basic infrastructure indicators. Aspirational districts (ADs) are eligible for enhanced funding and priority allocation of various initiatives undertaken by the central and state governments. Our research estimates the causal impact of ADP on the targeted health and nutrition indicators using a combination of propensity score matching and difference-in-differences (PSM-DID). We use the fourth and fifth rounds of National Family Health Survey data collected in 2015-16 and 2019-21 respectively which serve as the pre and post-treatment data for our analysis. Moreover, we take advantage of the transparent mechanism outlined for the identification of aspirational districts under ADP , which we use for propensity score matching for our PSM-DID. We find a 1% increase in home deliveries conducted in the presence of skilled birth attendants in aspirational districts as a result of ADP . Other than this, we do not find evidence of positive or negative impact of ADP on any other health and nutrition indicators. Future research efforts should be made toward impact evaluation of ADP on all the targeted indicators, which will provide a comprehensive evaluation of ADP .

Health inequalities can be both a cause and a consequence of inequitable economic growth and development, and one cannot be isolated from the other 1–4. Health inequality has been documented across countries as well as within a country across socioeconomic and disadvantaged groups as well as sub-national spatial units. In India and other low and middle-income countries (LMICs), while there have been several targeted interventions through social welfare and poverty alleviation schemes, the primary focus of such schemes has been to address deprivations, and therefore any implications of the scheme for addressing inequality are inadvertent. In 2018, the Government of India launched the ‘Transformation of Aspirational Districts’ or Aspirational Districts Program (ADP henceforth)5 with the objective of reducing regional multidimensional inequalities. ADP identified the most backward districts in the country and developed a comprehensive framework to prioritise development in these districts so that they can catch up with the other districts in the country. While health inequality is not the only targeted dimension, the health and education sectors equally contribute to 60% of the total outcomes. Other dimensions include agriculture and water resources, skill development, financial inclusion and basic infrastructure. We evaluate the impact of ADP on targeted health indicators in aspirational districts (ADs) before and after the policy intervention using the fourth and fifth rounds of the National Family and Health Survey (NFHS) data by employing a combination of propensity score matching and a difference-in-difference analysis. Our research makes an important contribution to the literature by assessing the impact of a policy exclusively designed to promote inclusive development by reducing multidimensional regional inequality. In the context of developing countries, particularly India, although there has been an increase in efforts toward research on health inequity in the past two decades6, 7, research studies have predominantly focused on documenting health disparities, which has been possible due to the availability of high-quality unit-level data. However, research evaluating health policies and programs at the sub-national level is lacking, which can contribute to policy making surrounding health inequity6, 7. Our research evaluates the ADP in India from the lens of health inequity and discusses the findings, which we believe would be of interest to researchers, practitioners and policy makers. To the best of our knowledge, our research is the first to evaluate ADP. In an earlier policy paper by Green and Kapoor8, although authors attempt to assess the ADP by accounting for changes in targeted indicators, the analysis is not causal, which the authors acknowledge. We fill this gap by estimating the causal impact of ADP on targeted health indicators. Our paper is . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint NOTE: This preprint reports new research that has not been certified by peer review and should not be used to guide clinical practice. structured as below. Section provides a brief description of the ADP followed by data and the methodology in section . Results of our analysis are presented in section followed by a discussion of the findings in section and conclusion in section respectively.

Regional inequalities in India are multidimensional in nature and existed for a long time. Researchers have found evidence of widening economic inequality during the last three decades post-liberalization 9–11. While economists have traditionally focused on income inequalities, increasing efforts have been made to document the evidence on health inequalities6, 7, 12, 13. Experts have acknowledged that reproductive health, child health and nutrition are some of the greatest challenges for India14. Strong evidence of district-level inequality in maternal mortality rates 15 and under-five child and infant mortality rates16–19 exists. Similar evidence of high inequality across districts exists for maternal health care utilization that includes indicators like antenatal and post natal care, institutional birth or delivery in the presence of skilled birth attendants 14, 20–26. Child nutrition also exhibits stark inequality across districts manifested in indicators of stunted, wasted or underweight chil- dren12, 21, 27–30. With regard to child immunization, while significant achievements have been made, full immunization is a distant thing, with several regions performing dismally low on vaccine preventable diseases21, 25, 27–29, 31, 32. In a recent work, Subramanian et al13 developed district-level maps of several child health and nutrition indicators using visual tools that capture district-level heterogeneity for different health and nutrition indicators. Aspirational Districts Program (ADP) The ‘Transformation of Aspirational Districts’ or Aspirational Districts Program (ADP) was launched by the Government of India in January 2018 with the objective of achieving inclusive development through a reduction in multidimensional inequalities by focusing on districts that have been lagging in several development indicators. Although, ADP is comprehensive and not limited to health indicators, health and nutrition is a key sector for the identification of backward districts as well as targeted program outcomes. The targeted outcomes are 49 performance indicators that include 81 data points, of which 13 indicators including 31 data points are from the health domain. The remaining indicators measure performance related to education, agriculture and water resources, financial inclusion and skill development and basic infrastructure.5 Aspirational districts (ADs) are eligible for enhanced funding and priority allocation of various initiatives undertaken by the central and state governments. Three key mechanisms underlying ADP that are expected to drive ADP toward its goal are convergence (of central and state schemes), collaboration (of central, state-level officers and district collectors) and competition among districts driven by a mass movement. A robust monitoring and evaluation framework is a key feature of ADP, which has been a major shortcoming of past initiatives33. ADs have been identified through a transparent mechanism based on the ranking of a composite index developed from 11 core measurable indicators summarized in the table 1. Health and nutrition indicators have been assigned one of the highest weightages (contributing to 30%) in the composite index that includes four variables - antenatal care, institutional deliveries, stunting and wasting of children below 5 years each having equal weights. The other sectors used for construction of the composite index include infrastructure with 30% weightage, deprivation measured by landless household dependent on manual labor contributing one-fourth to the index, and the remaining 15% is contributed by education. As it can be seen in table1, ADP guidelines also mention the data source used for identifying the ADs along with the indicators (measured in percentages unless explicitly stated)5. The data for health variables are taken from the fourth round of NFHS, deprivation from the Socio Economic Caste Census (SECC) data, education from Unified District Information System on Education (UDISE) and infrastructure variables from the respective ministries. Using the above mechanism, the composite index for every district is constructed and districts are ranked based on this composite index to arrive at the 117 most backward districts such that each state has at least 1 district as part of the ADP. West Bengal declined to be part of the ADP, hence after excluding 5 districts of West Bengal, there are 112 ADs8. A key feature of ADP is continuous monitoring and evaluation of targeted indicators. The performance metrics contribute to the overall and sector-specific ranking of a district, which is publicly available on the website Champions of Change dashboard of ADP (http://championsofchange.gov.in/site/coc-home/). This is expected to foster healthy competition among the ADs, which drives their monthly rankings. Several ADP targeted indicators are collected every few years that inhibits continuous monitoring. Therefore, such indicators are measured through survey or self-reported values of districts (non-validated or survey validated). While, the above approach is indispensable for continuous monitoring, yet relying solely upon them for assessing 2/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint Table 1. Variables used to identify ADs (Source: Aspirational Districts Primer5 Indicator Source Sector Weight Landless household dependent on Manual Labour SECC D7 Deprivation 25% Ante Natal Care NFHS-IV Health 7.5% Institutional Deliveries NFHS-IV Health 7.5% Stunting of children below 5 years NFHS-IV Health 7.5% Wasting of children below 5 years NFHS-IV Health 7.5% Elementary Drop-out Rate U-DISE 2015-16 Education 7.5% Adverse pupil teacher ratio U-DISE 2015-16 Education 7.5% Unelectrified household Ministry data Infrastructure 7.5% Household without individual toilet Ministry data Infrastructure 7.5% Un-connected PMGSY village Ministry data Infrastructure 7.5% Rural Household without access to water Ministry data Infrastructure 7.5% the impact of ADP could lead to a biased evaluation. Self reported indicators by districts are often not validated, which could be upward biased due to the competition faced by the districts. Moreover, causal impact evaluation of ADP, require measurement of outcomes for non-ADs too. Therefore, a nationally representative health outcome data is required for estimating the casual impact of ADP. We use district-level National Family Health Survey (NFHS) data to estimate the impact of ADP on targeted health and nutrition outcomes. NFHS data have been a trusted and reliable source of health data for decades among researchers, practitioners and policymakers, which has shaped important health policy decisions for the country. NFHS data collection is not motivated to reflect any desirable outcomes of ADP but rather centered around the objective of tracking the health status of the population in the country. Therefore, using NFHS data for the health impact evaluation of ADP is unlikely to suffer from desirability bias unlike the self-reported district performance indicators. Second, there are no large methodological differences in execution of NFHS data collection that make inter-district comparison unmeaningful. Finally, NFHS is nationally representative, and we are able to obtain estimates of health indicators for all districts irrespective of whether AD is present. Therefore, access to the fifth round of NFHS data facilitated the health impact evaluation of ADP. Data and Methods We estimate the causal impact of ADP on targeted health and nutrition outcomes by employing fourth and fifth rounds of district-level NFHS data. We take advantage of the timing of the launch of ADP in 2018, which was between the NFHS-4 and NFHS-5 data collection. The ability to use the fourth and the fifth rounds of NFHS data as the pre and post-intervention data respectively allows us to compare the changes in ADP-targeted health and nutrition outcomes5. We use a combination of propensity score matching (PSM) and difference-in-differences (DID) to estimate the causal impact of ADP on targeted health and nutrition outcomes. A growing body of literature in health economics and policy has employed the combination of PSM and DID (PSM-DID) to estimate the causal impact of a policy intervention using observational data34–38. A reasonable choice of casual inference model for the impact evaluation of ADP could be a DID model given the availability of NFHS-4 and NFHS-5 data that could serve as pre- and post-treatment data. DID estimates are unbiased if the pre-treatment trend of the outcome variable between the control and the treatment group are parallel. To test for the parallel trend assumption more than one period of pre-treatment data is needed. However, we are unable to test for pre-treatment parallel trends due to data limitations, as NFHS data are not available for districts but only for the states prior to NFHS-439. PSM has been widely used for causal impact evaluation in observational studies40–43. In PSM, the probability of treatment assignment depends on observed covariates, which are used to estimate the propensity score. The estimated PS is matched between the treatment and the control units to create a valid counterfactual group for the treatment units. PSM enables the design and analysis of observational studies to mimic certain aspects of randomized controlled trials. Essentially, the PS acts as a balancing score by ensuring that the distribution of observed baseline covariates is similar between the treated and untreated units given their PS values 43. It allows us to estimate the Average Treatment Effect in Treated (ATT) 44 for an intervention. Using PSM before DID enables comparison between the groups, and strengthens the plausibility of the parallel trends assumption, thus reducing the selection bias and dependence upon the unobservables and observables consistent over 3/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint time or having similar trends between the treatment and the counterfactual34, 38. In line with Heckman, Ichimura, and Todd (1997)45, PSM-DID method allows for estimation of temporally invariant unobserved outcome differences between individuals in the treatment and control groups, akin to fixed effects in panel data analysis. Therefore, we use the PSM-DID method for impact evaluation of the ADP on potentially targeted health and nutrition outcomes that are measured in NFHS. PSM-DID We would like to estimate the average treatment effect on the treated (ATT) on selected outcome variables, which would provide us with the casual estimate of the impact of ADP. Following Cheng et al.(2015)46, the ATT is given by AT T = E(Y P i,post −Y P i,pre|X P i ,ZP i ,Di = 1) − E(Y NP i,post −Y NP i,pre|X P i ,ZP i ,Di = 1) (1) where Y P i and Y NP i denotes the treated and the non-treated outcomes for district i with an additional subscript pre and post indicating time period for pre or post intervention. The outcomes are conditional upon a set of observed and unobserved attributes of the district denoted by Xi and Zi respectively. Di is an indicator variable with value 1 for treatment else 0. While, E(Y NP i,post −Y NP i,pre|X P i ,ZP i ,Di = 1) is unobserved, a fundamental assumption in the matching literature is that the treatment assignment can be assumed to be random if the treatment and the control groups are matched on the observed covariates, i.e. X P = X NP = X, which implies that E(Y NP i,post −Y NP i,pre|Xi,ZP i ,Di = 1) = E(Y NP i,post −Y NP i,pre|Xi,ZP i ,Di = 0). Therefore, the ATT can be rewritten as AT T = E(Y P i,post −Y P i,pre|X P i ,ZP i ,Di = 1) − E(Y NP i,post −Y NP i,pre|X P i ,ZP i ,Di = 0) (2) In the above equation, we can also get rid of the unobservables, if we believe thatZi are time invariant or have the same time trend between the treated and the untreated. Moreover, given the assumption of conditional Independence, the potential outcomes are independent of the treatment status when matched on the covariates. Therefore, the ATT can be estimated conditional on the propensity scores instead of conditional upon observed covariates. PS or the probability of treatment assignment is estimated using a logit regression as in equation 3 where Di is the binary variable with value 1 if the ith district is an AD else 0. X is the vector of observed covariates. P(Di = 1) = eXΓ 1 + eXΓ (3) In the context of ADP, we have information regarding the treatment assignment criteria. Therefore, we know the set of Xs or the observed covariates that were employed for treatment assignment as described in table 1. We use the same for matching on covariates in order to find a close counterfactual for every treated district. Subsequently, the ATT conditional upon PS can instead be written as AT T = E(Y P i,post −Y P i,pre|P(Di = 1,Xi),Di = 1) − E(Y NP i,post −Y NP i,pre|P(Di = 1,Xi),Di = 0) (4) We use full matching method for matching the PS to find a counterfactual for every treated unit. In this method, each treated unit is matched with one or more untreated units based on their PS by minimizing the matched sample’s total absolute within-subclass distances by the number of sub-classes selected and the units assigned to each subclass.43, 47. Full matching

has been demonstrated to be an effective matching technique for reducing bias due to observed confounding variables42. Evaluation Outcomes The targeted health and nutrition outcomes of ADP are listed in the ADP guidelines5. However, we restrict our analysis those indicators only for which we have a corresponding variable measured in NFHS as described in table 2. Out of 13 health indicators, 3 indicators related to anaemia among pregnant women, tuberculosis and health infrastructure are not available in NFHS, and thus could not be utilized in our analysis. Table 2 below lists the 11 health indicators (measured in percentages unless specified) along with the corresponding NFHS variable. Table 3 summarizes the mean of health data points for AD and non-ADs pre and post intervention (i.e. for NFHS-4 and NFHS-5) along with p-values stating the statistical significance for the difference of means between AD and non-AD. 4/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint Table 2: List of Health indicators (data points) targeted by ADP 1a Pregnant Women (PW) receiving four or more antenatal care check-ups against total ANC registration Mothers who had at least 4 antenatal care visits 1b ANC registered within the first trimester against total ANC registration Mothers who had an antenatal checkup in the first trimester 1c PW registered for ANC against estimated pregnancies Registered pregnancies for which the mother received a Mother and Child Protection card 2 PW taking Supplementary Nutrition under the ICDS programme regularly Mothers who consumed iron folic acid for 100 days or more when they were pregnant 4a Sex Ratio at birth Sex ratio at birth for children born in last 5 years 4b Institutional deliveries out of total estimated deliveries Institutional births 5 Home deliveries attended by a SBA (Skilled Birth Attendance) trained health worker out of total estimated deliveries Home births that were conducted by skilled health personnel 6a Newborns breastfed within one hour of birth Children under age 3 years breastfed within one hour of birth 6b Low birth weight babies (Less than 2500 gms) No variable 6c Proportion of live babies weighed at birth No variable 7 Underweight children under 5 years Children under 5 years who are underweight (weight for age) 8a Stunted children under 5 years Children under 5 years who are stunted (height for age) 8b Children with Diarrhoea treated with ORS Children with diarrhoea in the 2 weeks preceding the survey who received ORS 8c Children with Diarrhoea treated with Zinc Children with diarrhoea in the 2 weeks preceding the survey who received zinc 8d Children with ARI (Acute Respiratory Infection) in the last 2 weeks taken to a health facility Children with fever or symptoms of ARI in the 2 weeks preceding the survey taken to a health facility or health provider 9a Severe Acute Malnutrition (SAM) Children under 5 years who are severely wasted (weight for height) 9b Moderate Acute Malnutrition (MAM) Children under 5 years who are wasted (weight for height) 10a Breastfeeding children receiving adequate diet (6-23 months) Breastfeeding children age 6-23 months receiving an adequate diet 10b Non-Breastfeeding children receiving adequate diet (6-23 months) Non-breastfeeding children age 6-23 months receiving an adequate diet 11 Children fully immunized (9-11 months)(BCG + DPT3 + OPV3 + Measles1) Children age 12-23 months fully vaccinated based on information from (i) either vaccination card or mother’s recall (ii) vaccination card only No. ADP Health Indicator (Data points) NFHS variable 5/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint Table 3: Summary Statistics of ADP targeted health indicators pre and post intervention 1a Mothers who had at least 4 antenatal care visits 60.59 53.00 0.00 52.78 40.17 0.00 1b Mothers who had an antenatal checkup in the first trimester 71.72 67.53 0.00 61.12 52.68 0.00 1c Registered pregnancies for which the mother received a Mother and Child Protection (MCP) card 96.34 95.55 0.10 89.29 87.80 0.18 2 Mothers who consumed iron folic acid for 100 days or more when they were pregnant 45.87 40.40 0.01 31.31 23.83 0.00 4a Sex ratio at birth for children born in the last five years 947.50 943.38 0.74 932.27 942.90 0.32 4b Institutional births 89.64 82.30 0.00 80.10 68.79 0.00 5 Home births that were conducted by skilled health personnel 2.76 5.17 0.00 3.71 5.77 0.00 6a Children under age 3 years breastfed within one hour of birth 45.20 40.91 0.03 44.96 45.36 0.82 7 Children under 5 years who are underweight (weight for age) 28.44 35.92 0.00 31.53 40.93 0.00 8a Children under 5 years who are stunted (height for age) 32.41 39.42 0.00 34.90 42.91 0.00 8b Children with diarrhoea in the 2 weeks preceding the survey who received oral rehydration salts (ORS) 17.77 24.34 0.05 53.42 53.27 0.95 8c Children with diarrhoea in the 2 weeks preceding the survey who received zinc 9.13 12.52 0.07 22.19 20.57 0.30 8d Children with fever or symptoms of ARI in the 2 weeks preceding the survey taken to a health facility or health provide 43.60 48.35 0.15 72.04 69.47 0.10 9a Children under 5 years who are severely wasted (weight for height) 7.36 8.64 0.00 7.43 9.00 0.00 9b Children under 5 years who are wasted (weight for height) 18.17 20.68 0.00 20.12 23.90 0.00 10a Breastfeeding children age 6-23 months receiving an adequate diet 11.77 11.14 0.41 9.72 8.80 0.20 10bNon-breastfeeding children age 6-23 months receiving an adequate diet 1.18 1.12 0.89 14.89 14.89 1.00 11 Children age 12-23 months fully vaccinated based on information from either vaccination card or mother’s recall 76.74 76.80 0.96 62.61 59.53 0.10 11 Children age 12-23 months fully vaccinated based on information from vaccination card only 81.85 81.78 0.96 62.61 59.53 0.10 NFHS- 5 NFHS- 4 No.Indicators Mean (Non- AD) Mean (AD) p-val. Mean (Non- AD) Mean (AD) p-val.

PSM We begin with the results of PSM. We closely follow Table1 for choosing covariates for PS estimation for treatment assignment. We use indicators for all variables in table1 except ‘Unconnected PMGSY village’ under the infrastructure sector due to its 6/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint unavailability. We tried as much as possible to use the data from the sources listed in Table1, but for a few variables, the listed data source was not available. Therefore, instead, we used the variable from NFHS-4. We substituted ‘Household without individual toilet’ with the inverse of ‘Percent of Population living in households that use an improved sanitation facility’ from NFHS-4 and ‘Rural Household Without Access to Water’ with the inverse of ‘Percent of Population living in households with an improved drinking water source’ from NFHS-4. All the above indicators are used as independent variables in the logit regression for the estimation of PS. Subsequently, the estimated PS are matched using full matching in order to create a valid counterfactual for the treated districts. We have a total of 601 districts from district-level NFHS data, of which 109 are ADs. Figure 1 plots the density of observed covariates across ADs and non-ADs before and after matching. We also compared the summary statistics of covariates between ADs and non-ADs before and after matching as described in table 4 and table 5 respectively. We are left with 91 ADs and 355 non-ADs after full matching based on estimated propensity scores, which satisfies the assumption of common support. The assumption of common support is a crucial assumption for making meaningful comparisons across control and treatment units. A scatterplot of estimated propensity scores for comparison between ADs and non-ADs is provided in the appendix. Figure 1. Density plots across AD and non-AD for co-variates before and after matching PSM-DID The estimates of ATT from PSM-DID analysis for the targeted health and nutrition outcomes (as mentioned in table 2 are presented in table 6. We find an increase of 1% in home births conducted by skilled birth attendants (SBA) in ADs due to ADP and an increase of more than 9% in percentage of children taken to a health facility who had fever or ARI symptoms in ADs relative to non-ADs as a result of ADP. The results also suggest weak evidence of an increase in the percentage of stunted children in ADs. 7/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint Table 4. Summary of Covariate Balance before matching ‘ Mean St. Mean Var. eCDF Variable AD Non-AD Diff. Ratio Mean Max Landless household dependent on Manual Labour 30.2397 24.5434 0.3830 1.1236 0.1101 0.1911 Full Antenatal Care 14.6356 21.6954 -0.5230 0.6458 0.1273 0.2088 Institutional Births 68.8885 80.3258 -0.6842 0.9943 0.2040 0.3498 Stunted Children(<5 years) 42.8942 34.8777 0.9746 0.7249 0.2090 0.3858 Wasted Children(<5 years) 23.8192 20.1223 0.4390 1.2840 0.1105 0.1886 Elementary Dropout rates 7.4513 4.6893 0.4976 1.5501 0.1776 0.2982 Adverse Pupil Teacher Ratio 28.9527 22.8524 0.4356 1.6738 0.1402 0.2458 Unelectrified household 23.1067 12.8397 0.7634 1.1093 0.2287 0.3641 Household without individual toilet 67.9798 49.2710 0.9761 0.7684 0.2416 0.4272 Household without access to water 15.3510 10.9193 0.3178 1.3277 0.1068 0.2083 Table 5. Summary of Covariate Balance after matching Mean St. Mean Var. eCDF Variable AD Non-AD Diff. Ratio Mean Max Landless household dependent on Manual Labour 29.8023 31.6006 -0.1209 1.1142 0.0535 0.1414 Full Antenatal Care 15.1505 18.3066 -0.2338 0.6962 0.0562 0.1413 Institutional Births 70.7187 69.7060 0.0606 0.8176 0.0290 0.1114 Stunted Children(<5 years) 42.3758 41.2926 0.1317 0.8757 0.0401 0.1169 Wasted Children(<5 years) 23.7538 22.2937 0.1734 0.9099 0.0438 0.1050 Elementary Dropout rates 7.0248 7.1477 -0.0221 0.9426 0.0263 0.1157 Adverse Pupil Teacher Ratio 28.2272 26.9172 0.0935 1.0944 0.0350 0.0923 Unelectrified household 22.0343 18.6160 0.2542 1.1700 0.0652 0.1937 Household without individual toilet 67.0868 63.9678 0.1627 0.9310 0.0503 0.1752 Household without access to water 15.1637 19.0958 -0.2820 0.5022 0.0572 0.1346 8/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint We conduct several robustness checks for the validity of our results. In the first set of robustness checks, we report results from two matching methods other than full matching that include nearest neighbour matching (NNM) and genetic matching (GM). In the second set of robustness checks, we include additional variables for PS estimation, and subsequently valid counterfactuals are created using all three matching methods followed by DID estimation. Table 6. Results of PSM-DID No. Indicators Est. St. err. t-stat. p-val. 1a Mothers who had at least 4 antenatal care visits 0.58 1.57 0.37 0.71 1b Mothers who had an antenatal checkup in the first trimester -0.09 1.45 -0.06 0.95 1c Registered pregnancies for which the mother received a Mother and Child Protection (MCP) card -0.11 0.69 -0.16 0.88 2 Mothers who consumed iron folic acid for 100 days or more when they were pregnant 0.32 1.93 0.17 0.87 4a Sex ratio at birth for children born in the last five years -7.14 13.88 -0.51 0.61 4b Institutional births -0.37 1.04 -0.36 0.72 5 Home births that were conducted by skilled health personnel 1.00 0.43 2.33 0.02 6a Children under age 3 years breastfed within one hour of birth -2.46 1.56 -1.57 0.12 7 Children under 5 years who are underweight (weight for age) 0.99 1.04 0.95 0.34 8a Children under 5 years who are stunted (height for age) 1.45 0.88 1.65 0.10 8b Children with diarrhoea in the 2 weeks preceding the survey who received oral rehydration salts (ORS) -1.71 3.84 -0.45 0.66 8c Children with diarrhoea in the 2 weeks preceding the survey who received zinc 0.70 2.39 0.29 0.77 8d Children with fever or symptoms of ARI in the 2 weeks preceding the survey taken to a health facility or health provide 9.80 4.33 2.27 0.02 9a Children under 5 years who are severely wasted (weight for height) 0.41 0.59 0.70 0.48 9b Children under 5 years who are wasted (weight for height) 0.47 0.99 0.48 0.63 10a Breastfeeding children age 6-23 months receiving an adequate diet 0.84 0.77 1.09 0.28 10b Non-breastfeeding children age 6-23 months receiving an adequate diet -0.02 0.58 -0.04 0.97 11 Children age 12-23 months fully vaccinated based on information from either vaccination card or mother’s recall 1.48 1.60 0.93 0.35 11 Children age 12-23 months fully vaccinated based on information from vaccina- tion card only -0.48 1.55 -0.31 0.75 Robustness Checks Alternative matching methods We use two matching methods in addition to the full matching method used earlier, which includes Nearest Neighbor Matching (NNM) and genetic matching (GM). NNM pairs every treated unit with the closest eligible control unit, most commonly, the matching is based upon propensity scores. Typically, treated units with the highest propensity scores are paired first. On the other hand, GM is nonparametric matching, which is a generalization of PS and Mahalanobis distance matching that may include the propensity scores or the observed covariates or both48, 49. We have 96 treated and another 96 control districts after matching based on the NNM and GM method. The density plots and covariate summary statistics for both methods before and after matching are included in the appendix. Table 7 reports the PSM-DID results using matching methods NNM and GM. We report the results of only those indicators for which we observe a statistically significant impact. The complete set of results for all indicators is reported in the appendix. For both NNM and GM, we observe an increase in the percentage of home births conducted by SBA in the treatment district due to ADP, which is consistent with our previous results from the full matching method. Additionally, we also find a reduction in the percentage of institutional births using NNM, although no such impact is observed for GM or the full matching method. 9/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint Table 7. Results from alternative PSM-DID No. Indicator Est. St.err. t-stat. p-val. Nearest Neighbor Matching 4b Institutional births -1.77 0.87 -2.04 0.04 5 Home births that were conducted by skilled health personnel 1.23 0.30 4.13 0.00 Genetic Matching 5 Home births that were conducted by skilled health personnel 1.09 0.37 2.98 0.00 Use of additional covariates for estimation of propensity scores The covariates used for the estimation of PS were guided by the ADP operational guidelines5. However, there are few variables used for the identification of AD that have multiple indicators in NFHS-4, which measure different intensities of the same variable. For example, the NFHS-4 has three variables for antenatal care, of which the first one is the percentage of mothers who had full antenatal care, the second measures one antenatal checkup in the first trimester and the third measures atleast four ante- natal care visits. For our earlier analysis, we used full antenatal care as the covariate for the antenatal care variable as described in table 1. Similarly, we also have two indicators for wasted children, and the indicator we used earlier is percent of children wasted. However, another indicator measures the percentage of children severely wasted. As a robustness check, we included the above three variables for the re-estimation of PS in addition to the original variables that we used as independent variables. Accordingly, new counterfactuals are created using the re-estimated PS corresponding to the three matching methods used earlier. We have 91 treated and 379 control districts for the full matching method based on re-estimated PS, and another 96 treated and control district pairs after matching using the NNM and GM methods. The density plots and covariate summary statistics for all of the above matching methods before and after matching have been included in the appendix. Table 8 reports the PSM-DID results using full matching, NNM and GM based on re-estimated PS. We report the results of only those indi- cators for which we observe a statistically significant impact. The complete results for all indicators are reported in the appendix. We find strong evidence of an increase in the percentage of home births conducted by SBA. The findings are robust and do not depend upon the choice of method of matching that we use. This is in alignment with our main analysis as well as robustness checks that we have run. Weak evidence also exists towards an increase in percent of children fully immunized as a result of ADP. In addition, weak evidence towards undesirable and adverse impact of ADP also emerges in the form of a reduction in the percent of registered pregnancies, a reduction in the percentage of institutional births, a decrease in the percentage of non-breast feeding children receiving adequate diet and an increase in percentage of stunted children in the ADs. However, other than the indicator for home births conducted by SBA, no other findings are robust. Table 8. Results from different PSM-DID method with additional variables for estimating the propensity scores No. Indicator Est. St.err. t-stat. p-val. Full Matching 1c Registered pregnancies for which the mother received a Mother and Child Protection (MCP) card -0.89 0.53 -1.70 0.09 5 Home births that were conducted by skilled health personnel 1.15 0.36 3.15 0.00 8a Children under 5 years who are stunted (height for age 1.31 0.75 1.74 0.08 10b Non-breastfeeding children age 6-23 months receiving an adequate diet -1.09 0.67 -1.62 0.10 Nearest Neighbor Matching 1c Registered pregnancies for which the mother received a Mother and Child Protection (MCP) card -0.82 0.45 -1.81 0.07 5 Home births that were conducted by skilled health personnel 1.46 0.36 4.07 0.00 11 Children age 12-23 months fully vaccinated based on information from either vaccination card or mother’s recall 3.42 1.85 1.85 0.06 Genetic Matching 4b Institutional births -1.68 0.93 -1.81 0.07 5 Home births that were conducted by skilled health personnel 1.21 0.40 3.06 0.00 11 Children age 12-23 months fully vaccinated based on information from either vaccination card or mother’s recall 2.89 1.76 1.65 0.10 10/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint

Welfare schemes and policies in LMICs are mostly targeted toward addressing severe multidimensional deprivations. Inequality reduction is likely to be an inadvertent but desirable impact of welfare schemes. With this as the stage, ADP in India is a unique of its kind experiment given that it aims to reduce inter-district multidimensional inequalities by identifying the most backward districts in the country, and making additional efforts through ADP so that the ADs catch up with the rest of the country. We evaluate ADP from the lens of the health sector. The evaluation metric consists of a composite index, of which health and nutrition indicators carry 30% weightage spanning 31 health data points (over 13 performance indicators) out of a total of 81 data points (over 49 performance indicators). Out of 13 targeted health and nutrition indicators measured using 31 data points, we run an impact evaluation of ADP for 18 health and nutrition data points across 10 indicators. Prior to our work, assessment of ADP was been limited to documenting the changes or existing potential for improvement using distance-to-frontier analysis8, 50. However, none of the above research attempts to estimate the causal impact of ADP. Therefore, we believe that our study is one of the first to estimate the causal impact of ADP on targeted health and nutrition outcomes. We take advantage of the transparent AD identification method and the availability of district-level nationally representative health survey data before and after treatment to estimate the measurable impact of ADP on health and nutrition outcomes. We find strong evidence of an approximately 1% increase in percent of home births conducted by SBA, which is robust across several specifications. We do not find robust evidence of any positive or negative impact of ADP on any other health and nutrition indicator. The COVID-19 pandemic has tested the resilience of health systems globally including those that have been considered to be highly robust51–54. The COVID-19 pandemic has been characterized by major disruptions in routine health services in both high-income countries and LMICs53, 55, 56, which has had a profound negative impact on the utilization of Reproductive, Maternal, Newborn, Child Health (RMNCH) services53, 57–59. This has contributed to an increase in maternal mortality rates during and after the pandemic 60, 61. In India, even prior to the pandemic multiple studies have indicated low utilization of RMNCH services in several regions21, 24, 26, 62, 63, although there has been a significant increase in the utilization of RMNCH services in the last decade primarily driven by the cash incentive scheme (Janani Suraksha Yojana or the safe motherhood scheme) 64–67. In addition, several other factors driving the uptake of RMNCH services include community-level health infrastructure20, education24, 68, exposure to mass media 24, 68 and health insurance coverage69. Despite this increase in the uptake of RMNCH services, substantial inequalities still exist24, 68. As part of RMNCH services in India, home births conducted in the presence of SBAs were low between 1992 and 2006 with pronounced rural-urban and poor-rich differences, with rural and poor women being less likely to utilise SBA services 62. India witnessed a substantial increase in the percentage of home births conducted by SBA between 2006 and 2016 with the figures increasing from 41% in 2006 to 81% in 2016 70, which further increased to 90% in 2019-2169. The cash incentive scheme for safe motherhood (Janani Suraksha Yojana) started in 2005 has been instrumental in bringing this change 24, 64–68, 70. Despite this impressive development, stark inequalities still exist in home births conducted by SBA across states and within states across districts. The latest round of NFHS reveals that the highest and the lowest average SBA percentages across the states are 100% and 57% respectively, which are even higher across districts ranging between 32% and 100%69. In the past, efforts have been made through India’s Reproductive, Maternal, Newborn, Child and Adolescent Health (RMNCH+A) strategy to improve child and maternal health outcomes across poor-performing districts of the country33. Although, RMNCH+A has limited success in achieving its objectives, lim- itations and learnings from the initiative have served as useful guidance for future endeavours that led to the building of the ADP. We find a significant increase in the percentage of home births conducted in the presence of SBA of approximately 1% in ADs as a result of the ADP. While institutional deliveries are encouraged, if the mother chooses to deliver at home, the emphasis is that home deliveries should be attended by an SBA5. During the pandemic, there was a reduction in the percentage of institutional deliveries71–73, and pregnant women often chose to deliver at home instead of a health facility due to the fear of transmission of COVID infection63. While it is possible that the increase in SBA could be driven by a reduction in the percentage of institutional deliveries due to the pandemic, we do not find evidence that could substantiate differences in institutional births between ADs and non-ADs. In a pre-COVID study including 145 hotspot districts with low utilization of RMNCH services, a 10% increase in exposure to mass media in a particular district or a 10% increase in four or more antenatal care (ANC) in a particular district was likely to increase SBA by 2.5%24. Therefore, a 1% increase in SBA in the ADs as a

of the ADP is not a small change given that part of the evaluation period coincided with the covid-19 pandemic. Amidst such glaring inequalities, a 1% increase in SBA during home deliveries in the AD is impressive for the ADP. We do not find a robust impact of ADP on any other indicator other than a positive and statistically significant impact on home births conducted by SBA. Three key mechanisms that are expected to drive the ADP towards its goal are conver- gence of central and state schemes, collaboration of central, state level officers and district collectors and competition among districts driven by a mass movement. The targeted outcomes of ADP are comprehensive, and are not limited to health and 11/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint nutrition indicators but include education, agriculture and water resources, financial inclusion and skill development and infrastructure outcomes too. ADP explicitly mentions that the initial focus of ADs should be on achieving low-hanging fruits, which when coupled with competition among ADs could be driving our results of no impact on other health and nutrition indicators5. If improvements in targeted health and nutrition outcomes are less visible relative to other targeted sectors over smaller periods of time, competition among ADs might channel efforts and resources toward sectors in which changes are rapidly visible. Therefore, although our findings do not find evidence of significant improvement in health and nutrition in- dicators, there could be improvement in other non-health indicators as a result of ADP, which is beyond the scope of our research. We also point out a few limitations of our findings. Our research utilizes the fourth and fifth rounds of NFHS data. However, an important caveat remains unstated and is related to the execution of data collection for the fifth round of NFHS data. NFHS-5 data collection was disrupted due to the pandemic. While we do not doubt the integrity of the data collection process, NFHS data collection was executed in two phases - the first was pre-pandemic and the second phase was during the pandemic. It is possible that our results could be influenced by the two phases of data collection. There is no debate that adverse impacts of the pandemic have been felt unequally globally across countries and within countries, which has further exacerbated existing health inequalities24, 74, 75. In the context of ADs, by design most backward districts in the country, predominantly rural regions were identified as ADs. As evident from Table 3, there are significant disparities in different development indicators between ADs and non-ADs. In light of the evidence of the heterogeneous impact of COVID-19 across different socioeconomic groups and regions76–79, it cannot be ruled out that the rural areas in India were hit harder due to their already stressed health systems even pre-pandemic80–82. Therefore, it is possible that findings of a limited positive impact of ADP on health and nutrition outcomes could be masking the gains reversed by the pandemic. ADP could have averted several adverse outcomes due to the pandemic. Theerefore, it is comforting to see that there is no evidence of widening health inequalities between ADs and non-ADs. However, taking this limitation into account, we suggest future research directions for impact evaluation of ADP in India. First, ADP warrants a comprehensive evaluation across all targeted indicators as there might be several interrelated push and pull factors across sectors that need not co-move in the same direction in the short run. Second, further research would be beneficial to circumvent confounding related to COVID in our results, which might be possible with the post-pandemic sixth round of NFHS data that has already been planned for execution between 2023-24.

The Aspirational District Program (ADP) is a unique initiative of the Government of India launched in 2018 that aims to reduce inter-district multi-dimensional inequality. ADP aims to bring the most backward districts to catch up with the rest of the other districts in the country. The program is comprehensive in its scope, as it targets the improvement of several key development indicators for spanning health and nutrition, education, agriculture and water resources, financial inclusion and skill development and basic infrastructure. We evaluate the impact of ADP on health and nutrition indicators and find evidence of a significant increase in the percent of home births conducted by skilled birth attendants in aspirational districts as a result of ADP. Our findings are robust to different specifications. Future research efforts should be made toward impact evaluation of all the targeted indicators, which will provide a comprehensive evaluation of ADP.

1. Thomas, K. & Sudhakar, K. Health-care inequalities: an indian perspective. The Lancet 356, S35 (2000). 2. O’Donnell, O., Van Doorslaer, E. & Van Ourti, T. Health and inequality. In Handbook of income distribution, vol. 2, 1419–1533 (Elsevier, 2015). 3. Clark, R. World health inequality: Convergence, divergence, and development. Soc. science & medicine 72, 617–624 (2011). 4. Grimm, M. Does inequality in health impede economic growth? Oxf. economic papers 63, 448–474 (2011). 5. Aayog, N. Aspirational districts (2018). 6. Bhan, N., Rao, K. D. & Kachwaha, S. Health inequalities research in india: a review of trends and themes in the literature since the 1990s. Int. journal for equity health 15, 1–8 (2016). 7. Bhan, N. & Rao, K. D. Health inequities research in india: A review of themes, patterns and opportunities. BMJ Glob. Heal. 1, A34–A35 (2016). 8. Porter, M. E. & Stern, S. An assessment of aspirational districts programme. Inst. for Compet. (2020). 12/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint 9. Suryanarayana, M., Agrawal, A. & Prabhu, K. S. Inequality-adjusted human development index: States in india. Indian J. Hum. Dev. 10, 157–175 (2016). 10. Lolayekar, A. & Mukhopadhyay, P. Growth convergence and regional inequality in india (1981–2012). J. Quant. Econ. 15, 307–328 (2017). 11. Singhal, A., Sahu, S., Chattopadhyay, S., Mukherjee, A. & Bhanja, S. N. Using night time lights to find regional inequality in india and its relationship with economic development. PloS one 15, e0241907 (2020). 12. Joe, W., Mishra, U. S. & Navaneetham, K. Health inequality in india: evidence from nfhs 3. Econ. Polit. Wkly. 41–47 (2008). 13. Subramanian, S., Sarwal, R., Joe, W. & Kim, R. Geo-visualising diet, anthroprometric and clinical indicators for children in india: Enabling district prioritisation for interventions (2020). 14. Paul, V . K.et al. Reproductive health, and child health and nutrition in india: meeting the challenge. The Lancet 377, 332–349 (2011). 15. Goli, S., Puri, P., Salve, P. S., Pallikadavath, S. & James, K. Estimates and correlates of district-level maternal mortality ratio in india. PLOS Glob. Public Heal. 2, e0000441 (2022). 16. Kumar, C., Singh, P. K. & Rai, R. K. Under-five mortality in high focus states in india: a district level geospatial analysis. Plos one 7, e37515 (2012). 17. Sankar, M. et al. State of newborn health in india. J. perinatology 36, S3–S8 (2016). 18. Bora, J. K. Factors explaining regional variation in under-five mortality in india: an evidence from nfhs-4. Heal. & Place 64, 102363 (2020). 19. Puranik, A., Binu, V ., Biju, S. & Subba, S. H. Spatio-temporal assessment of infant mortality rate in india. Indian journal public health 62, 32–38 (2018). 20. Singh, P. K., Kumar, C., Rai, R. K. & Singh, L. Factors associated with maternal healthcare services utilization in nine high focus states in india: a multilevel analysis based on 14 385 communities in 292 districts. Heal. policy planning 29, 542–559 (2014). 21. Awasthi, A., Pandey, C., Chauhan, R. K. & Singh, U. Disparity in maternal, newborn and child health services in high focus states in india: a district-level cross-sectional analysis. BMJ open 6, e009885 (2016). 22. Ali, B., Debnath, P. & Anwar, T. Inequalities in utilisation of maternal health services in urban india: Evidences from national family health survey-4. Clin. Epidemiol. Glob. Heal. 10, 100672 (2021). 23. Singh, N., Patel, R. & Chauhan, S. Geospatial analysis of utilization of maternal health care services in india. GeoJournal 1–20 (2021). 24. Mishra, P. S., Sinha, D., Kumar, P. & Srivastava, S. Spatial inequalities in skilled birth attendance in india: a spatial-regional model approach. BMC Public Heal. 22, 1–13 (2022). 25. Kumar, G. & Reshmi, R. Availability of public health facilities and utilization of maternal and child health services in districts of india. Clin. Epidemiol. Glob. Heal. 15, 101070 (2022). 26. Usman, M., Reddy, U. S., Siddiqui, L. A. & Banerjee, A. Exploration of spatial clustering in maternal health continuum of care across districts of india: A geospatial analysis of demographic and health survey data. Plos one 17, e0279117 (2022). 27. Kumar, C., Singh, P. K., Singh, L. & Rai, R. K. Socioeconomic disparities in coverage of full immunisation among children of adolescent mothers in india, 1990–2006: a repeated cross-sectional analysis. BMJ open 6, e009768 (2016). 28. Panda, B. K., Kumar, G. & Mishra, S. Understanding the full-immunization gap in districts of india: a geospatial approach. Clin. Epidemiol. Glob. Heal. 8, 536–543 (2020). 29. Puri, P., Khan, J., Shil, A. & Ali, M. A cross-sectional study on selected child health outcomes in india: Quantifying the spatial variations and identification of the parental risk factors. Sci. reports 10, 6645 (2020). 30. Biswas, M. Identifying geographical heterogeneity in associations between under-five child nutritional status and its correlates across indian districts. Spatial Demogr. 10, 143–187 (2022). 31. Khan, J., Shil, A. & Prakash, R. Exploring the spatial heterogeneity in different doses of vaccination coverage in india. PloS one 13, e0207209 (2018). 32. Singh, S. K. & Vishwakarma, D. Spatial heterogeneity in the coverage of full immunization among children in india: Exploring the contribution of immunization card. Child. Youth Serv. Rev. 121, 105701 (2021). 13/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint 33. Taneja, G. et al. India’s rmnch+ a strategy: approach, learnings and limitations. BMJ global health 4, e001162 (2019). 34. Wang, H., Yip, W., Zhang, L. & Hsiao, W. C. The impact of rural mutual health care on health status: evaluation of a social experiment in rural china. Heal. Econ. 18, S65–S82 (2009). 35. Chen, Y . & Jin, G. Z. Does health insurance coverage lead to better health and educational outcomes? evidence from rural china. J. health economics 31, 1–14 (2012). 36. Su, B., Thierry, G. H., Chen, Q. & Zhao, Q. The new cooperative medical scheme and self-employment in rural china. Sustainability 9, 304 (2017). 37. Dong, W. et al. Effects of china’s urban basic health insurance on preventive care service utilization and health behaviors: Evidence from the china health and nutrition survey. PLoS one 13, e0209890 (2018). 38. Zhou, Q., He, Q., Eggleston, K. & Liu, G. G. Urban-rural health insurance integration in china: impact on health care utilization, financial risk protection, and health status. Appl. Econ. 54, 2491–2509 (2022). 39. Delhi, P. Conducting the nfhs/annual health surveys (2019). 40. Austin, P. C. A critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Stat. medicine 27, 2037–2049 (2008). 41. Caliendo, M. & Kopeinig, S. Some practical guidance for the implementation of propensity score matching. J. economic surveys 22, 31–72 (2008). 42. Stuart, E. A. & Green, K. M. Using full matching to estimate causal effects in nonexperimental studies: examining the relationship between adolescent marijuana use and adult outcomes. Dev. psychology 44, 395 (2008). 43. Austin, P. C. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivar. behavioral research 46, 399–424 (2011). 44. Imbens, G. W. Nonparametric estimation of average treatment effects under exogeneity: A review. Rev. Econ. statistics 86, 4–29 (2004). 45. Heckman, J. J., Ichimura, H. & Todd, P. E. Matching as an econometric evaluation estimator: Evidence from evaluating a job training programme. The review economic studies 64, 605–654 (1997). 46. Cheng, L., Liu, H., Zhang, Y ., Shen, K. & Zeng, Y . The impact of health insurance on health outcomes and spending of the elderly: evidence from china’s new cooperative medical scheme.Heal. economics 24, 672–691 (2015). 47. Greifer, N. Matching methods (2018). 48. Sekhon, J. S. & Grieve, R. A new non-parametric matching method for bias adjustment with applications to economic evaluations. In iHEA 2007 6th World Congress: Explorations in Health Economics Paper(2008). 49. Diamond, A. & Sekhon, J. S. Genetic matching for estimating causal effects: A general multivariate matching method for achieving balance in observational studies. Rev. Econ. Stat. 95, 932–945 (2013). 50. Sharma, S. & Sharma, V .Efficiency Assessment of Maternal Health Services in the Aspirational Districts of EAG States in India: A Data Envelopment Analysis Approach (Institute of Economic Growth, University Enclave, University of Delhi, 2020). 51. Saulnier, D. D. et al. A health systems resilience research agenda: moving from concept to practice. BMJ Glob. Heal. 6, e006779 (2021). 52. Haldane, V .et al. Health systems resilience in managing the covid-19 pandemic: lessons from 28 countries. Nat. Medicine 27, 964–980 (2021). 53. Arsenault, C. et al. Covid-19 and resilience of healthcare systems in ten countries. Nat. Medicine 28, 1314–1324 (2022). 54. Paschoalotto, M. A. C., Lazzari, E. A., Rocha, R., Massuda, A. & Castro, M. C. Health systems resilience: is it time to revisit resilience after covid-19? Soc. Sci. & Medicine 115716 (2023). 55. Pujolar, G., Oliver-Anglès, A., Vargas, I. & Vázquez, M.-L. Changes in access to health services during the covid-19 pandemic: a scoping review. Int. J. Environ. Res. Public Heal. 19, 1749 (2022). 56. Fejfar, D. et al. The impact of covid-19 and national pandemic responses on health service utilisation in seven low-and middle-income countries. Glob. Heal. Action 16, 2178604 (2023). 57. Shapira, G. et al. Disruptions in maternal and child health service utilization during covid-19: analysis from eight sub-saharan african countries. Heal. policy planning 36, 1140–1151 (2021). 14/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint 58. Amouzou, A. et al. Health service utilisation during the covid-19 pandemic in sub-saharan africa in 2020: a multicountry empirical assessment with a focus on maternal, newborn and child health services. BMJ global health 7, e008069 (2022). 59. Aranda, Z. et al. Disruptions in maternal health service use during the covid-19 pandemic in 2020: experiences from 37 health facilities in low-income and middle-income countries. BMJ Glob. Heal. 7, e007247 (2022). 60. Chmielewska, B. et al. Effects of the covid-19 pandemic on maternal and perinatal outcomes: a systematic review and meta-analysis. The Lancet Glob. Heal. 9, e759–e772 (2021). 61. Khalil, A. et al. A call to action: the global failure to effectively tackle maternal mortality rates. The Lancet Glob. Heal. 11, e1165–e1167 (2023). 62. Pathak, P. K., Singh, A. & Subramanian, S. Economic inequalities in maternal health care: prenatal care and skilled birth attendance in india, 1992–2006. PloS one 5, e13593 (2010). 63. Ghosh, A. & Ghosh, R. Maternal health care in india: a reflection of 10 years of national health mission on the indian maternal health scenario. Sex. & Reproductive Healthc. 25, 100530 (2020). 64. Lim, S. S. et al. India’s janani suraksha yojana, a conditional cash transfer programme to increase births in health facilities: an impact evaluation. The Lancet 375, 2009–2023 (2010). 65. Powell-Jackson, T., Mazumdar, S. & Mills, A. Financial incentives in health: new evidence from india’s janani suraksha yojana. J. health economics 43, 154–169 (2015). 66. Sengupta, N. & Sinha, A. Is india’s safe motherhood scheme leading to better child health care practices? Glob. Soc. Welf. 5, 49–58 (2018). 67. Rahman, M. M. & Pallikadavath, S. How much do conditional cash transfers increase the utilization of maternal and child health care services? new evidence from janani suraksha yojana in india. Econ. & Hum. Biol. 31, 164–183 (2018). 68. Ali, B. & Chauhan, S. Inequalities in the utilisation of maternal health care in rural india: Evidences from national family health survey iii & iv. BMC public health 20, 1–13 (2020). 69. Prasad, R. D., Ghosh, K. & Shri, N. Does health insurance coverage improve maternal healthcare services utilization in india? evidence from national family health survey-5, 2019–21. J. Public Heal. 1–13 (2023). 70. Ali, B., Dhillon, P. & Mohanty, S. K. Inequalities in the utilization of maternal health care in the pre-and post-national health mission periods in india. J. Biosoc. Sci. 52, 198–212 (2020). 71. Kumari, V ., Mehta, K. & Choudhary, R. Covid-19 outbreak and decreased hospitalisation of pregnant women in labour. The Lancet Glob. Heal. 8, e1116–e1117 (2020). 72. Nair, M., Group, M. W. et al. Reproductive health crisis during waves one and two of the covid-19 pandemic in india: incidence and deaths from severe maternal complications in more than 202,000 hospital births. EClinicalMedicine 39 (2021). 73. Goyal, M. et al. The effect of the covid-19 pandemic on maternal health due to delay in seeking health care: experience from a tertiary center. Int. J. Gynecol. & Obstet. 152, 231–235 (2021). 74. Bambra, C., Lynch, J. & Smith, K. E. The unequal pandemic: COVID-19 and health inequalities (Policy Press, 2021). 75. McGowan, V . J. & Bambra, C. Covid-19 mortality and deprivation: pandemic, syndemic, and endemic health inequalities. The Lancet Public Heal. 7, e966–e975 (2022). 76. Basu, P. & Mazumder, R. Regional disparity of covid-19 infections: an investigation using state-level indian data. Indian Econ. Rev. 56, 215–232 (2021). 77. Avula, R. et al. Disruptions, restorations and adaptations to health and nutrition service delivery in multiple states across india over the course of the covid-19 pandemic in 2020: an observational study. PLoS One 17, e0269674 (2022). 78. Chakrabarti, A., Bair, E. F. & Thirumurthy, H. Routine child immunizations in india during the covid-19 pandemic. SSM-Population Heal. 22, 101383 (2023). 79. Chakraborty, T. & Mukherjee, A. Economic geography of contagion: a study of covid-19 outbreak in india. J. Popul. Econ. 36, 779–811 (2023). 80. Sarkar, A. & Chouhan, P. Covid-19: District level vulnerability assessment in india. Clin. epidemiology global health 9, 204–215 (2021). 81. Middya, A. I. & Roy, S. Geographically varying relationships of covid-19 mortality with different factors in india. Sci. reports 11, 7890 (2021). 15/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint 82. Panneer, S., Bhat, L., Rice, L. & Kantamaneni, K. The second wave of covid-19 and beyond: Rural healthcare. Econ. Polit. Wkly. 57 (2022).

We are thankful to J V Meenakshi, Ram Singh, Sangeeta Bansal and participants at the Annual Economics Conference 2022 organized at IISER Bhopal for their useful comments and feedback on an earlier version of the paper. Author contributions statement SKA conceived the project, SM prepared the data and conducted the analysis under supervision of SKA, SM wrote the first draft of the paper, SKA prepared the current draft. All authors reviewed the manuscript. (Sandip K. Agarwal (SKA); Shubham Mishra (SM)) Additional information Competing interests The authors declare that they have no competing interest. 16/16 . CC-BY-NC-ND 4.0 International licenseIt is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprintthis version posted September 6, 2023. ; https://doi.org/10.1101/2023.07.27.23293263doi: medRxiv preprint