State trends in infant mortality, 1968-83.

This paper presents an analysis of state trends in infant mortality rates (IMRs) for 1968-83. In order to take into account the large random error component associated with state IMRs, weighted least squares estimates are used to fit log-linear models to these trends. Using simulated data, these estimates are shown to be nearly unbiased and to provide valid significance tests. However, the power to detect changes in trend is rather limited, especially in small states. Using these methods, separate analyses of White IMRs in 49 states and non-White IMRs in 30 states were completed. Nine states are identified which had infant mortality trends less favorable than the national experience and 1981-83 rates more than 5 per cent above the national average.