A QUANTILE REGRESSION APPROACH FOR CHILD MORTALITY ANALYSIS Prafulla Kumar Swain*, Vishal Deo 1 and Gunjan Kumar P. G. Department of Statistics, Utkal University, Vanivihar, Bhubaneswar - 751 004, India. 1 Department of Statistics, Ramjas College, University of Delhi, New Delhi - 110 007, India. E-mail : prafulla86@gmail.com Abstract : In this paper, we have explored a quantile regression approach to study the factors affecting the child mortality in India. The annual health survey data has been used for application and the results of quantile regression have been compared with those of a linear regression (LR) model. Factors, such as safe delivery, private delivery, mothers’ post natal check within 48 hours, breast feeding within 1 hour, full immunizations, fathers literacy rate., etc are found to be significantly associated with child mortality (P value < 0.05). The results have demonstrated that using quantile regression leads to better interpretation and more specific inference about the predictors of child mortality. Hence, we suggest that the quantile regression could be used as an alternative to LR in mortality analysis. Key words : Child mortality, Quantile regression, Linear regression, Annual health survey. 1. Introduction In recent years, quantile regression has been widely used to assess the distribution of a response variable given a set of explanatory variables. Koenker and Bassett (1978) argued on the omnipresent Gaussian assumption for the model error and formalized asymptotic properties of the least absolute deviations estimator of the quantile regression model for independent observations. Indeed, the emerging quantile regression has many advantages over linear regression model, viz., in quantile regression, we can assess how the centre of a conditional distribution varies with changes in subject characteristics, and one can examine how any percentile of the conditional distribution is affected by the changes in subject characteristics. Although, quantile regression came to light as an econometric regression model, now it is being used in various fields, especially, medicine [Austin and Schull (2003)], education and policy [Haile and Nguyen (2008)], natural resource management [Cade et al . (2005)], forestry [Zhang et al. (2005)], in computing stand density index [Ducey and Knapp (2010)] and in evaluating the spread rate of forest diseases [Evans and Finkral (2010)]. Pourhoseingholi et al. (2008) used *Author for correspondence Received May 18, 2017 Revised August 26, 2017 Accepted September 11, 2017 Int. J. Agricult. Stat. Sci. Vol. 13, No. 2, pp. 457-463, 2017 ISSN : 0973-1903 ORIGINAL ARTICLE a quantile regression approach to study the quality of life in breast cancer patients. Fenske et al . (2013) undertook a comprehensive analysis of the determinants of child stunting in India using additive quantile regression approach. In this work, our objective is to determine the predictors associated with child mortality in India based on annual health survey data using a quantile regression. While a number of studies have looked at the correlates of infant mortality, most of them have exclusively concerned themselves with estimating the mean effect of variables such as mother’s education, child’s sex, urbanization level and birth order etc. on infant mortality. Such estimates miss an important point for policy makers: exogenous variables and policy interventions may affect infant mortality differentially at different points in the conditional distribution. India has the world’s highest percentage (21%) of under-five deaths, estimated at 1726000 in 2009 [UNICEF (2009)]. Decentralized district-based health planning is essential in India because of the large inter- district variations. In the absence of vital data at the district level, the State level estimates are being used for formulating district level plans as well as setting the