Background: This paper discusses appropriate strategies for multivariate data analysis in epidemiological studies.
Methods: In studies where determinants of disease are sought, it is suggested that the complex hierarchical inter-relationships between these determinants are best managed through the use of conceptual frameworks. Failure to take these aspects into consideration is common in the epidemiological literature and leads to underestimation of the effects of distal determinants.
Results: An example of this analytical approach, which is not based purely on statistical associations, is given for assessing determinants of mortality due to diarrhoea in children.
Conclusions: Conceptual frameworks provide guidance for the use of multivariate techniques and aid the interpretation of their results in the light of social and biological knowledge.