Do public health surveys provide representative data? Comparison of three different sampling approaches in the adult population of Croatia.

We investigated the sample representativeness in three different types of population-based public health surveys in Croatia. Responses from the household sampling based Croatian Adult Health Survey (CAHS), health insurance register based Croatian Health Survey (CHS) and a telephone survey (TPS) were analysed and compared to gender, age and education composition of the Croatian adult population, based on the 2001 Census. The raw (unweighted) survey data were used and analysed with Spearman's rank test and distance analysis. The results indicated that TPS had the most similar gender composition compared to the Census data. TPS also had the most similar age composition in men, while CHS had the most similar age composition in women. Finally, CAHS had the most similar education composition to the Census data. Three population subgroups were substantially under-sampled in all three surveys--men, younger people, and elderly from the lowest educational classes. For these sub-groups, advanced sampling methods should be employed in order to obtain more precise estimates from public health surveys.