Infant mortality time series are random walks with drift: are they cointegrated with socioeconomic variables?

Abstract

Previous time series analyses of infant mortality have failed to provide evidence to support their implicit assumption that infant mortality data used behaved as a stationary time series. The present study applies the augmented Dickey Fuller Test to infant mortality time series for Sweden (1800-1989), United Kingdom (1839-1989) and United States (1915-1989). The null hypothesis that each of these series is non-stationary is accepted at standard levels of significance. A conceptual framework of infant mortality which uses a combination of physical and social overhead capital as factors in a production function is developed to explain the finding of non-stationarity as derivative from the non-stationarity of a stock of health-enhancing capital. Estimation of econometric models of the socioeconomic determinants of infant mortality using differenced data with ARIMA estimation is inconclusive. Estimation of a bivariate cointegration model supports the hypothesis that infant survival and GNP/Capita are cointegrated for 19th century Sweden but not for 19th century UK. Bivariate analysis of 20th century Sweden, UK, and US data demonstrated no cointegration. This may be due to the onset of disequilibrium in the economic determination of infant mortality in the present era as technological advances and demographic shifts began to play a larger role. Supplementing the bivariate analysis with measures of unemployment, and crude birth rate in the 20th century permitted the detection of cointegration in US and UK. The multivariate results may suggest that improvements in 20th century UK GNP/capita have had greater impact on infant survival relative to US GNP/capita.