Informing Resource-Poor Populations and the Delivery of Entitled Health and Social Services in Rural India:  A Cluster Randomized Controlled Trial FREE

The delivery of health and social services is often inadequate in developing countries such as India in part because of limited resources allocated by governments and donor agencies.^1- 2 However, increased spending by itself may not be sufficient to improve outcomes. Efficiently targeting resources to impoverished communities and getting public workers to perform their duties have remained significant challenges for many public services worldwide.^1,3- 5

Inadequate service delivery may result from weak mechanisms of accountability in developing countries.^1,6- 8 For example, 25% to 35% of publicly funded health workers and school teachers are absent on any given day but are rarely fired for repeated absences.^9- 11 A number of developing countries, including India, are decentralizing control over local public services to local communities,^12- 13 which is intended to increase service providers' accountability to the local community.⁷ However, anecdotal reports and surveys indicate that communities and members of local government are uninformed of what services they are entitled to and what state-mandated controls they have over these services.^4,8 Local oversight committees rarely meet and committee members are also frequently unaware of service delivery issues. Widespread poverty and social divisions such as caste also make individual or collective action to improve service delivery difficult in these settings.¹⁴

Uttar Pradesh ranks 23 out of 32 states in India in terms of the proportion of people living below the poverty line.¹⁵ The poverty line in India is defined as the amount of money needed to purchase one's caloric needs and does not include money required for clothing or shelter.¹⁵ Nearly 35% of India lives on less than $1 a day and represents a region with the greatest percentage of impoverished people next to sub-Saharan Africa.¹⁶ The National Family Health Survey¹⁷ found that only 64% of rural women in Uttar Pradesh receive antenatal care. Seventy-six percent of births are unassisted by a trained health care provider, and only 10% of rural women report receiving any postnatal care. Eighty percent of children younger than 2 years have not received their full immunizations. Although there is a paucity of detailed data at the district level on such services, other surveys in Uttar Pradesh find similar results.^18- 19 Even though education is free, most individuals receive less than a high school education, and less than 60% of the population is literate.^17,20 In Uttar Pradesh, 26% of teachers are absent on any given day.²¹

Interventions that increase the accountability of public service sectors to the community may improve the quantity and quality of services delivered. One way to enhance the accountability may be to provide information that empowers stakeholders. For example, a recent newspaper campaign in Uganda that publicized diversion of primary school funds led to a reduction in embezzlement from 80% to 20%.²² However, informational campaigns that involve random assignment to intervention and control groups and rigorous evaluation of outcomes are lacking.

We therefore sought to conduct a community-based randomized controlled trial to determine the impact of information delivery on receipt of health and social services. We hypothesized that if individuals have detailed information about public services they are legally entitled to, they may be empowered to demand these services from public workers and less likely to tolerate petty corruption from government workers.

We conducted this cluster randomized controlled trial from May 2004 to May 2005 in Uttar Pradesh, a state in northern India with a population of 170 million.²⁰ Of the 70 districts in this state, we focused on 21 central, central-eastern, and southern districts in which we have previously conducted surveys. Districts consist of approximately 14 blocks and each block consists of about 65 village clusters. From a comprehensive list of blocks and village clusters, we used a random-number generator to randomly select 1 block within each district and then randomly select 5 village clusters within each block. We then randomly assigned districts to intervention and control arms. By randomly selecting only 5 village clusters of about 1000 in each district, we spread the selection of 105 village clusters over 21 districts to minimize any potential for contamination between intervention and control villages. Although the districts were adjacent to one another, no 2 blocks were adjacent to each other and the village clusters were far apart. Travel between them would be difficult. A village cluster (also referred to as a gram panchayat) is the smallest unit of government and consists of approximately 1 to 3 adjacent villages that elect a local head and council members. There are on average 956 village clusters per district. Each village cluster included in the study had on average 409 households and an average population of 2343.²⁰

The districts were all drawn from a single state, and availability of and access to health and education services are largely determined by policy at the state level. The population is ethnically the same and similar in culture. The population speaks a single language, Hindi.

The Figure illustrates the flow of participants through the trial. Five village clusters from each of 21 eligible districts were randomly chosen for the intervention or to serve as control. We sampled 10 households per village cluster. We identified 5 low-caste (also referred to as scheduled caste) households from each village cluster by first selecting an arbitrary starting point in the main low-caste neighborhood and then selecting every fifth household from this starting point. We used a similar method to identify 5 mid- to high-caste households from each village cluster. Villages are not designed in any systematic way and consist of numerous crisscrossing alleys and paths. In the intervention arm, 252 low-caste and 296 mid- to high-caste households were systematically selected for the baseline survey. Twelve intervention households moved before the final survey. In the control arm, 245 low-caste and 252 mid- to high-caste households were systematically selected for the baseline survey. Eight control households moved before the final survey.

Our cluster-randomized trial sample size calculations were based on a 5% significance level and 80% power. The sample size and power calculations are driven by the number of village clusters, rather than the number of households per village cluster. For proportional outcomes, to detect a 0.2 increase over a control proportion of 0.5 with 10 households per cluster and a conservative coefficient of variation of 0.5, we estimated needing 94 total clusters (47 per arm). Increasing the number of households above 10 does not significantly decrease the number of village clusters required. For school fees, to detect a 10-rupee decline from a control of 35 rupees with 10 children per cluster, standard deviation of 15 rupees, and a coefficient of variation of 0.5, we estimated needing 82 total clusters. Our actual sample size included 105 total clusters.²³

Households were invited to participate in a baseline survey if they had at least 1 child going to public primary school in the village. Children are not required to go to school, although they are guaranteed a right to basic education. A recent survey of 69 districts in Uttar Pradesh in 2005 found that 93% of all children aged 6 through 14 years are enrolled in school. Sixty-three percent of all children are enrolled in government schools.²⁴ In some villages, none of the mid- to high-caste households had children attending a public school. In this case, we included 5 mid- to high-caste families even if their children did not attend a public school. In 3 intervention village clusters, we could not locate any low-caste households and selected mid- to high-caste households in their place. Five intervention clusters and 1 control village cluster had fewer than 5 low-caste households. A total of 21 additional mid- to high-caste households in intervention villages and 2 extra mid- to high-caste households in control villages were selected to maintain the overall sample size. This method resulted in a slightly uneven distribution of low-caste vs mid- to high-caste households. Of 548 households in intervention village clusters, 252 (46%) were low caste. Of 497 households in control village clusters, 245 (49%) were low caste.

We used the Indian census definition of a household, “a group of persons who commonly live together and would take their meals from a common kitchen,” and the state government's definition for caste.^25- 28 We asked each household's caste and used a comprehensive list provided by the government to identify which caste category, low or mid- to high caste, the household belonged to. Low caste refers to “untouchables” (also called scheduled castes), who are at the bottom of the caste and socioeconomic hierarchy.²⁹ The criterion for listing a person in this category is “extreme social, education and economic backwardness arising out of the traditional practice of untouchability.”²⁶ Midcaste refers to “other backward classes,” which are defined as certain castes of historically low socioeconomic status. “High caste” refers to all others. We chose half our sample from scheduled castes because they depend disproportionately on government services compared with the higher castes, who are more often able to afford private physicians and schools. Because overcharging school fees is widespread in villages, we chose households that had school-going children to assess any impact on this outcome. We focused on districts in northern India where we had previously conducted survey work because it gave us the insight and familiarity needed to perform an informational campaign. The surveys were conducted by in-person interviews by a team of trained research assistants with experience in administering rural household surveys in Uttar Pradesh. Surveys were conducted in the local language of Hindi and were pilot tested in villages outside the sample village clusters before use. At baseline, we did not tell the households that any informational meetings would be done later, nor did they know that they would be reinterviewed at 1 year.

Both parents from each household were asked several questions about access to health and social services. Health services questions included whether a nurse midwife had come to the village in the past 4 weeks; whether there was a pregnant woman in the household within the past 12 months and, if so, whether she had received a prenatal examination, tetanus shots, and prenatal supplements (iron/folic acid tablets); and whether there was an infant younger than 1 year in the household and, if so, whether he or she had received any vaccinations. Social services questions included how many children went to primary school in the village for the previous academic year and how much in school fees they were charged, whether a village council meeting had occurred in the past 6 months, and whether development work was performed in the village.

An information campaign was conducted in each intervention village cluster in June 2004. The information campaign was conducted in 2 rounds in each village cluster, separated by a period of 2 weeks. Each round consisted of 2 to 3 meetings, as well as distribution of posters and leaflets. Residents were informed in advance about the dates and locations of meetings, and separate meetings were held in low- and mid- to high-caste neighborhoods. Each meeting lasted about an hour and consisted of a 15-minute audiotaped presentation that was played twice, opportunities to ask questions, and distribution of leaflets. People were notified that the information was collected from the government and distributed in the public interest by our research team and a nongovernmental organization based in Uttar Pradesh, Sahbhagi Shikshan Kendra.³⁰

To ensure that the information campaigns were uniform, research assistants read a scripted introduction and were allowed to answer questions only to which the answers were already written on the leaflets. Any other questions or issues were not answered. The presentation, leaflets, and posters focused on the following information. Health services information included the specific days and hours a nurse midwife is available in the village; the obligation of the nurse midwife to provide free prenatal and postnatal care, including tetanus vaccines and prenatal supplements for mothers and health care and vaccinations for infants; health centers available for more specialized care; and where to complain about quality or quantity of health services. Social services information included how much school fees are for low- and mid- to high-caste children, sources and oversight of education funds, obligations of oversight committees, requirements for semiannual village governance meetings, organization and funding of village government and development work, right to obtain copies of village records, and where to complain about education or village governance problems.

The information in the presentation and leaflets was obtained from the Uttar Pradesh health, education, and village governance departments. According to headcounts of everyone attending our informational meetings, 14% of the residents of the entire village cluster attended in round 1 and 11% attended in round 2. If there was some overlap in attendance, we estimate that 14% to 25% of the residents of each intervention village cluster attended a presentation. We reached our target level of attendance, which was 250 people in each round (about 11% of the average village population). No information campaign was held in the control villages.

Baseline survey participants were reinterviewed 12 months later by research assistants who had no knowledge of the intervention. To maintain this blinding, intervention group subjects were not asked whether they attended an informational meeting.

Two years after our intervention, we conducted focus group meetings among 20 randomly selected intervention villages. We conducted 3 focus group meetings in each village, one from each of mid- to high caste, low caste, and low-caste women. Each meeting consisted of about 20 adult volunteer participants who were asked whether they remembered the information campaign, whether they had discussed the information with others afterward, whether any of the villagers took up service delivery issues with the service providers, why more significant changes did not occur in the village, and what could be done to improve our information campaign.

Excess school fees were defined as the school fees paid by students minus the legal amount they can be charged (US $1 = 45 rupees). The unit of analysis for this outcome was individual children. The unit of analysis for other outcomes (eg, visits by nurse midwife, development work in village) was households. For each outcome, we compared intervention and control groups, adjusting standard errors for clustering at the village level. We used the regress and cluster commands from Stata 9.2 statistical software (StataCorp, College Station, Texas) for these analyses. P < .05 was set as the threshold for significance.

For 5 of 8 outcomes, comparing within-household changes from baseline to follow-up was not possible, because households that reported those outcomes at baseline were often not reporting on the same outcomes at 1 year. For example, a household reporting on prenatal outcomes at baseline would no longer have a pregnant woman to report prenatal outcomes on at 1 year. For these, we additionally conducted a multivariate regression comparing intervention to control at 1 year, using a random-effects model in which random effects are at the village cluster level and standard errors are clustered at the village cluster level. The regression adjusts for total population of the village cluster, district size, household caste, and highest education attained in the household. For the 3 remaining outcomes of visits by nurse midwife, village council meetings, and development work in village, we conducted a within-household difference-in-differences analysis, using a random-effects model at the village cluster level and clustering for standard errors at the village cluster level.

Focus groups were analyzed by proportion of respondents to questions. Quotations representing dominant themes were recorded.

Our informational campaign and surveys were approved by the village governance department of Uttar Pradesh and the institutional review board of Johns Hopkins University. Before conducting the surveys, we first approached the locally elected village head with the state permission and obtained permission from him. A research assistant then obtained verbal informed consent from both parents before administering the household survey. If the father of the household was not available, then another male adult was asked to substitute.

Intervention and control districts were similar in their socioeconomic indicators, and the mean characteristics are similar to the averages for the state (Table 1).^{20,24,31- 32} The districts ranged in size from 180 000 to 555 000 rural households and from 1.2 million to 3.7 million rural population.²⁰ Only 16% of their population lives in urban areas. One third of their population lives below the poverty line, which is an income of $90 per capita per year.²⁰ The infant mortality rate is 89 per 1000 live births.³¹ Only 46% of children are immunized. Although most children are enrolled in school, the literacy rate is only 56%.²⁴

The village clusters were characterized by substantial numbers of low-caste residents, a low literacy rate, less than universal access to electricity, and few primary schools (Table 1). Treatment and control village clusters did not differ in their demographic and developmental characteristics.

Our baseline survey indicated no significant differences between treatment and control households with respect to the health, education, and governance outcomes (Table 2). However, it did indicate inadequate delivery of relevant services for both groups. Only about two thirds of households had been visited by a nurse midwife, about half of pregnant women and infants had received specific health services, and children were charged excess school fees of about 30 rupees per year. Moreover, only about one third of households reported that a village council meeting had occurred or that development work in the village was performed.

After 1 year, intervention villagers noted increased delivery of several services compared with control villagers, and most outcomes either worsened or remained unchanged in the control villages (Table 2). Changes among low-caste and mid- to high-caste households were generally similar (Table 3 and Table 4). However, the differences between intervention and control low-caste households were not statistically significant for tetanus vaccinations and receipt of prenatal supplements (Table 3).

In a multivariate random-effects regression comparing intervention villagers with control villagers at 1 year (Table 5), in absolute percentage points villagers in the intervention group reported increased prenatal examinations (30%; P < .001), tetanus vaccinations (27%; P < .001), prenatal supplements (24%; P = .003), infant vaccinations (25%; P = .004), and decreased excess school fees (8 rupees; P < .001). When differences were calculated in change from baseline to final assessment between intervention and control groups for the 3 outcomes in which this was possible, village council meetings occurred 21% more often in intervention than control (P = .01) groups. There were no significant differences between intervention and control groups in visits by nurse midwives or in development work (Table 2).

Sixty-two percent (694/1118) of participants in 60 focus groups remembered the information campaign. Of those who remembered, at least 25% stated that they had taken up service delivery issues with individual service providers, although only 5% had brought it up in the village council meeting; 43% reported that service delivery improved in health and education services after the information campaign. However, less than 8% reported that there was any change in the functioning of the village council meetings.

Forty percent of focus group participants (278 members) stated that they had discussed the information with others. Of the 276 members who said they had not discussed the information with others, 58% believed it was futile, 20% said they did not understand the information well enough, and 9% said they were scared. When we asked why more significant change did not occur in village improvement activities, a prominent theme was that the elected village head tended to be unapproachable. Typical responses from mid to high castes were that the “village head's work is done for his own men.” Low castes responded similarly, also citing caste as a reason.

When asked how we could improve the information campaign, many reported that information campaigns should occur every few months and that the village head, as well as the village secretary, should participate.

We found that providing a structured informational program to villagers in north Indian districts about entitled services enhanced the delivery of health and social services among low- and mid- to high-caste households. Improvements occurred in prenatal services, infant vaccinations, excess school fees, and occurrence of village council meetings. Although nurse midwife visits did not increase, 20% to 25% more households in intervention villages reported that women were receiving prenatal services and that their infants were being immunized, suggesting that the nurse midwife was performing more duties when a visit did occur.

Although most outcomes either worsened or remained unchanged in control villages, almost all outcomes improved in intervention villages, suggesting that in an environment in which some services are generally worsening, at a minimum, the intervention prevented services from worsening.

Prenatal outcomes worsened in control villages, possibly a consequence of fewer nurse midwife visits, which may fluctuate from year to year, depending on local factors. The nonavailability of supplies or supervision, which can be affected by changes in the state government, may reduce her incentives to visit villages. The state government did change that year, with a new chief minister and political party for the state. Although most policies remained the same, the degree of implementation could change as a consequence of the new leadership. Intervention villages reported having 5% more pregnant women at 1 year compared with control, which could be explained by a few more new marriages or slightly increased fertility that year.

Development work in the villages increased in both arms and may have been influenced by new elections for village leadership that were due 3 months after our final survey. It is common for spending activity for development work in villages to increase just before local elections. It is unlikely that a shift in resources within the village clusters explains the increase in development work and decrease in prenatal services. Funds for village development work come directly to the village cluster accounts and are managed by the elected village head. Nurse midwives do not report to the village head and receive their supplies separately. Village council meetings worsened in the control groups and are supposed to be a forum for evaluating complaints and progress in the village cluster. However, with village elections upcoming, providing such a forum might have been perceived by the village head as hindering his chances for reelection and may therefore not have been called.

Teachers may get away with overcharging school fees because parents do not know the actual fee schedule and may be illiterate. The actual fees are between 0 and 12 rupees per child per year, depending on the child's grade and caste.³³ At 1 year, intervention households reported being overcharged 10 rupees less, whereas there was no appreciable decrease in charges for control households.

Focus groups suggest that the improvements resulted from increased communication among villagers and between villagers and service providers in intervention villages. However, there was little dialogue with the village leadership because of a sense of futility, as well as fear.

Previous work on this topic has described widespread deficiencies in service delivery.³⁴ However, few randomized trials of interventions to improve public service delivery have been conducted.³⁵ Our findings highlight the importance of empowering communities to facilitate individual and collective action,^1,34 which is especially important because a number of developing countries, including India, have decentralized control over public services to local communities.^12- 13

Empowerment often requires significant resources to involve community members in gaining control over issues that concern them.¹⁴ Although we informed villages about their rights to services, as well as how to complain, we did not advocate that they demand those services or that they complain. Significant improvements in basic services we observed suggest that villagers are capable of some degree of self-empowerment once they are given the correct information. However, significant barriers to collective action were apparent and clearly warrant further studies.

The total cost of the intervention was only $4000, which amounts to about $0.22 per household in a village cluster. If the government or local organizations could disseminate this information on radio or newspapers, the costs could be lower.

Strengths of our study include a large sample size in a state that shares many of the same service delivery issues as other Indian states,³⁴ the use of a practical intervention that can easily be adapted to other regions, and a rigorous cluster randomized controlled trial design. However, our intervention was not successful in increasing visits by nurse midwives or development work, indicating that further work is needed to determine how to improve these outcomes. In addition, low-caste households appeared to benefit less than mid- to high-caste households in 2 areas, tetanus vaccinations and prenatal supplements, suggesting that low-caste households have greater barriers to individual or collective action or that they are inadequately informed of the importance of these services.

Several limitations must be considered in interpreting our results. First, the data collected were based on self-report because there are no reliable written records on nurse midwife visits, prenatal services, vaccinations, or excess school fees. Although there are registers of village governance meetings, these are frequently falsified. Second, it is possible that intervention villagers were more likely to pay attention to service delivery than control villagers (recall bias). Third, our data do not include any morbidity or mortality outcomes. However, prenatal care has a documented association with morbidity and mortality outcomes.^36- 37 Fourth, we focused on villages in only 1, albeit large, geographic region. The content and dissemination of this intervention should be tailored to the circumstances of other Indian states and other developing countries. Future work may involve refining the nature and manner of information campaigns, accounting for the literacy levels of various communities, enhancing community oversight of services, targeting lower socioeconomic households better, and evaluating for sustainability. Sustainability outcomes may depend on targeting interventions in primary education and governance, along with health.³⁸

Seventy percent of India's population lives in the villages and would benefit from improved basic services.²⁰ Interventions that educate resource-poor populations about entitled services may improve the delivery of services. Such interventions are promising and low-cost means to improve the health and welfare of individuals in developing countries.