Community Effects on Infant Mortality in Rural India: A Multilevel Approach to Prioritize the Program Inputs

Damodar Sahu, Institute for Research in Medical Statistics (ICMR)
Arvind Pandey, Indian Council of Medical Research
Sada Nand Dwivedi, All India Institute of Medical Sciences

Background: In spite of substantial improvement in the general health condition of people in India over time, infant mortality rate continues to be high in certain parts of the country particularly in the rural area where two-third of the population lives. This is a matter of great concern. High infant and child mortality was found to be associated mainly with the low socio-economic development (poverty, illiteracy, education, sanitation, potable water, lack of access to utilization of health care services at the community level). The program managers and policy makers have been at the same time harping on the issue of identifying vulnerable locations to prioritize the program to meet demand-supply gap in-term of basic facility. They looked forward for the research in the area. A number of studies has examined determinates of infant and child mortality in India and other countries. These studies used mostly individual level information. Recent literature has however emphasized the importance of community (or cluster) level variables. A population that have superior infrastructures such as approach to all weather road, health centre, education facilities etc. is likely to have better health condition of people and low level of infant and child mortality. Such factors as above at the community level definitely influence individual perception and decision to access services. Under this premise it is very important to know the effectiveness of community level indicators and their linkages with individual level indicators to guide program managers to achieve the convergence. Most of the studies reported so far especially from developing countries have considered community level variables in data analysis but these variables have been disaggregated at individual level that would result into distortion of traditional methods of data analysis regarding their important assumption of independence of records. This consideration does not take into account the hierarchical structure present in the data where children are nested in a community. In fact, all children in a community have common community level variables. Therefore, traditional methods of data analysis may obviously result into under-estimation of standard errors of estimates showing significant association of a covariate with dependent variables, which may not be true in real sense. In contrary, comparatively a new method “Multilevel Analysis” considers the hierarchical structure of data in analysis providing reliable results. Study design: We propose to use two--level model to examine the linkages of various socio-economic and programme factors that may affect infant survival. Accordingly we shall be incorporating variables at two levels- level-I: individual level variables of children and parents; and level-II: community level variables; from data collected under National Family Health Survey, 1998-99, India. The main focus of the analysis will be on assessing the effectiveness of the community level variables like health services and infrastructures available at the community (or cluster) level on infant mortality, by controlling the effects of other individual level factors in rural India. Hypothesis: There is a strong positive relationship between survival chance of infants and the availability of health services in the community. (Whether there was a health centre facility i.e. primary health centre or community health centre/rural hospital, or government hospital/dispensary or private hospital/clinic within 2km of the sample cluster, weather all-weather road within the cluster, whether opportunity of higher education (middle school or above) within 2 km of the sample, whether health or family welfare film show organized at community in the past one year) in the community (or cluster). Data and Methodology: We propose to use the data from the India’s National Family Health Survey, 1998-99 (NFHS-2) where complete birth history information were collected from all eligible ever-married women aged 15-49 surveyed including sex, twin birth, date of birth, survival status for each live birth, and age at death for children who had died. A child-based file is created to include selected child-specific information from birth history, and mother and household characteristics. The present study will be based on most recent births born three years preceding the survey because the desired data on maternal care in NFHS-2 were collected only for these births. Two statistical models are estimated using maximum likelihood estimators. The outcome variables for the two equations are survival status of child before attending his/her first birthday (‘1’ for surviving and ‘0’ for not surviving). The independent variables are (1) demographic variables (breastfeeding duration, child is twin, sex of the child, birth order, preceding birth interval, and mother’s age at the time of birth); (2) socio-economic variables (religion-caste/tribe, mother’s education, father’s education, mother’s work status, mass media exposure (watching TV at least once in a week) and land holding); (3) environmental variables (house type, electricity, sanitation, cooking fuel, potable water, separate kitchen room, and crowding); (4) maternal care (number of antenatal check-up visits, number TT injection received, and place of delivery); and (5) community level variables (whether there was a health centre facility i.e. primary health centre or community health centre/rural hospital, or government hospital/dispensary or private hospital/clinic within 2km of the sample cluster, whether all-weather road within the cluster, whether opportunity of higher education (middle school or above) within 2 km of the sample, whether health or family welfare film show organized at community in the last one year). Expected findings: As in the earlier studies, the findings of the present study may have impact of some of the individual level variables such as demographic (sex, preceding birth interval, mother’s age at the time of birth etc.), socioeconomic (mother’s literacy, religion-cast/tribe), environmental (sanitation facility, safe drinking water), and maternal care variables (received TT, institutional delivery). In addition, there may be strong negative association between the community level variables - availability of health centre (within 2km) in the community, presence of good transport including all-weather road within the cluster etc. and the infant mortality in the study population.