We used data from two rounds of the nationally representative, longitudinal Indian Human Development Surveys (IHDS). The surveys were funded by the National Institutes of Child Health and Human Development (NICHD) and were produced jointly by the National Council of Applied Economic Research (NCAER), New Delhi, and the University of Maryland. IHDS-I (2004–2005) was administered to 41 554 households—27 010 rural and 13 126 urban households. The rural sample was drawn using stratified random sampling of defined units. In urban areas, a stratified sample of towns and cities within states was selected by probability proportional to population size [25, 26]. Eighty-three percent of the households interviewed in 2004–2005, were re-interviewed in IHDS-II (2011–2012), and an additional replacement sample of 2134 households was added. Appendix 3 presents an analysis of households that were not interviewed in IHDS-II and were lost to follow-up. The IHDS-II survey was administered to 42,152 households in 33 states and union territories, 384 districts (of 612 districts), 1503 villages, and 971 urban areas.
Both rounds of the IHDS survey include a household interview with information about household asset ownership, and an interview with ever-married women of reproductive ages (15–49 years) with information about birth history, reproductive health, and antenatal and delivery care for the most recent birth. IHDS-I recorded 11 670 births, and IHDS-II recorded 13 881 births (demographic characteristics presented in Appendix 1). The implementation of the ASHA program started in April 2005. IHDS-I covers births from 2000 to 2005, and IHDS-II covers a 6-year period (2005–2011) after the implementation of the ASHA program.
Measurement of program exposure and outcomes
We estimated the effect of the ASHA program on four outcomes: (1) whether the respondent received at least one ANC visit, (2) had 4 or more ANC visits, (3) delivered in a health facility, and (4) had a skilled attendant present at the time of birth. Following the 2008 WHO recommendations, we considered medical doctors, nurses, and auxiliary nurse midwives (ANMs) as skilled birth attendants [27].
IHDS-II survey gathered information on the type of health care provider seen by women at any point during their last pregnancy. A woman was considered exposed to an ASHA if she reported that an ASHA assisted her in response to at least one of the following questions: “Where did you get a pregnancy card made?”; “Did you get help from anyone for making a pregnancy card/registration?”; “Who visited you when you were pregnant?”; “Who facilitated or motivated you to go to a health facility for delivery?”; and “Who arranged the transportation to take you to the health facility for delivery?”. We explored three specifications for defining the exposure: (S1) Cluster-level “intensity” of ASHA exposure was calculated as number of women who reported exposure to an ASHA in a cluster during the last birth divided by the total number of eligible women who had a birth in 6 years preceding the survey in the cluster. This measure takes a value between 0 and 1 and captures direct individual exposure to an ASHA, as well as indirect effect that may result from the presence of an ASHA within the community. (S2) All women in clusters in which at least one woman reported seeing an ASHA received a value of “exposed” (i.e., 1). (S3) Women received a value of either 0 or 1 depending on their individual reported exposure.
Characteristics of the women who report receipt of ASHA services
To understand whether the ASHA program is reaching its target population, we calculated the uptake of the program nationally, disaggregated by demographic characteristics, based on self-reported ASHA exposure (S3) using IHDS-II data. To understand whether ASHAs are differentially used by individuals belonging to a certain demographic in areas where the ASHA program is active, we restricted the analysis to only those clusters where ASHA services were reported. We used logistic regression to investigate the association between a range of individual and household characteristics and receipt of ASHA services in clusters where an ASHA has been reported. We estimated this model at the national level and separately for rural areas and high-focus states.
The logistic models account for sociodemographic characteristics of the women—including maternal age (15–19, 20–24, 25–29, 30–34, 35–39, 40–49 years), maternal education (1–5 years, secondary 6–11 years, 12 years or more of education), maternal caste (upper/forward caste, scheduled caste (SC), scheduled tribe (ST), other backward caste (OBC)), religion (Hindu, Muslim, other religions), parity, and household wealth index. OBC, SC, and ST are official Government of India caste classifications for minority groups of historically disadvantaged people. Continued disparities by caste exist in education, income, and social networks. We used polychoric principal component analysis (PCA) to estimate household wealth using information about household asset ownership (bicycle, sewing machine, generator, mixer, motorcycle, television, air cooler, clock, fan, chair/table, cot, telephone, mobile phone) and household characteristics (type of cooking place, type of toilet, availability of electricity, type of chulha (hearth), water source, wall type, roof type, floor type) from the IHDS household surveys. The polychoric procedure, unlike the standard PCA, retains ordinal variables without breaking them into dummy variables [28]. The first component of the polychoric PCA was used to create wealth quintiles, explaining 27% of the variance for the 2005 and 30% for the 2012 data.
Association between exposure to ASHAs and maternal service utilization outcomes
We assessed the effect of exposure to ASHAs, measured as cluster-level exposure intensity (S1), on the outcomes, using multivariate difference-in-differences models fitted using ordinary least squares regression [29]. The difference-in-differences approach assesses the effect of the ASHA program by controlling for baseline differences between exposed and unexposed populations, and for temporal differences that may have resulted from underlying changes over time [30]. We used fixed effects at the cluster level to control for baseline differences between clusters and any unmeasured time-invariant factors that may have resulted in selective uptake or targeting of the ASHA program. The models controlled for individual and household characteristics: maternal age, maternal education, household wealth index, birth order, maternal caste, and religion. All regression models were adjusted for survey design features using cluster-level sample weights, and standard errors were corrected for correlation across individuals in the same cluster using robust standard errors.
We tested the sensitivity of our findings to various model specifications. We estimated the models separately for rural areas and for high-focus states (results not presented). About 30% of the clusters had 3 or fewer eligible women. Since ASHA exposure is defined as the cluster average, we tested the robustness of the results to cluster sizes greater than two and greater than three. Additionally, we also used the three different specifications for exposure to the ASHA program for further robustness checks. Around 2005, a national cash assistance program called Janani Surakshna Yojana was launched to incentivize women living below the poverty line to deliver in health facilities or with a skilled attendant [31]. The ASHAs play a vital role in rolling out this program. We could not control for cash assistance in the above models for skilled birth attendance (SBA) and facility birth, as data on cash transfer were only available for women who delivered in a health facility. To disentangle the effect of the cash transfer on SBA and facility births from the effect of the ASHA program alone, we tested a multinomial logistic model with SBA and facility birth coded as three-level categorical outcomes. Facility births were coded as birth at home, birth in a health facility without any financial incentive, and birth in a health facility with a financial incentive, and SBA was categorized as no SBA, SBA without financial incentive, and SBA with financial incentive.