Methodological approaches to conducting pooled cross-sectional time seriesanalysis: The example of the association between all-cause mortality and per capita alcohol consumption for men in 15 European states

Aim: To compare different statistical models in order to estimate the assoc iation of alcohol consumption and total mortality when time series data ste m from different regions. Data and Methods: Data on per capita consumption in 15 European countries were combined with standardized mortality rates co vering different periods between 1950 and 1995. An indicator of region-spec ific drinking patterns was measured without reference to a concrete time po int, thus generating a hierarchical data structure. Two groups of models we re compared: pooled cross-sectional time series models with different error structures and hierarchical linear models (random coefficient models). Res ults. If historical time is not controlled for in cross-sectional models, t his might result in estimating a negative association between alcohol consu mption and total mortality. Hierarchical linear models or cross-sectional m odels controlling for historical time, however, resulted in the expected po sitive association. Only hierarchical linear models were able to adequately estimate the moderating effect of drinking patterns on the association bet ween alcohol consumption and total mortality. Conclusion: For pooled cross- sectional time series data, control for the potential impact of historical time is of utmost importance. Hierarchical linear models constitute a super ior alternative to analyze such complex data sets, especially as time-indep endent characteristics of regions can be implemented in the model. Copyrigh t (C) 2001 S. Karger AG, Basel.