Could infant mortality analysis with time-invariant covariate effects be misleading? Evidence from urban Turkey

Murat G. Kirdar, Middle East Technical University
Mehmet Ali Eryurt, Hacettepe University
Ismet Koc, Hacettepe University

This study investigates whether the usual practice of conducting infant mortality analysis with time-invariant covariate effects could be misleading. For this purpose, it compares the results of an infant mortality analysis that does not allow for time-invariant covariate effects with the results of one that allows. The comparison of cumulative regressions with split regressions reveals that the effects of many covariates change substantially by the age of an infant. Considering different dynamics are associated with infant mortality in urban and rural areas and three-quarters of the population lives in urban areas the study is limited with the analysis of urban Turkey. The study uses data from Demographic and Health Surveys conducted in Turkey between 1993 and 2008. We pooled 4 sets of DHS data to provide a large sample size and increase statistical power. In these surveys each woman was asked for a history of all her births, including the month and year of each. Information on birth and death date used in the study is based on retrospective birth histories of women age 15-49. If the baby died before one month, date of death is recorded in a daily basis. In addition to the birth histories of the women, information on a wide range of of individual (on the mothers and children), household and community-level variables is collected in TDHSs. Event history analysis is applied to measure the impact of a rich set of variables on infant mortality risks. Mother’s age, mother’s education, birth order, preceding birth interval, prenatal care, place of delivery, gender of the child, family wealth, region and ethnicity included in the analysis as main independent variables. The results show cumulative analysis cannot uncover certain changing covariate effects over the age of the infant, which are uncovered by the infant mortality analysis with time-variant covariate effects.