Regression Analysis and Causal Inference: Cause for Concern? Perspectives on Sexual and Reproductive Health
2012 | Download
This Viewpoints article from Perspectives on Sexual and Reproductive Health provides a critical review of the uses and misuses of regression analysis in this and other fields, and the validity threats and issues associated with the claims, conclusions and recommendations that typically result. It calls for inferential humility in interpreting results from these methods, and provides recommendation for better practice and use.
Perspectives has long been known for its applied focus and emphasis on real-world problems and solutions, and has filled a critical niche in our field. Certainly, there is value in publishing good noncausal studies—for example, exploratory studies can yield hypotheses to be tested in future research, while descriptive and correlational studies of sexual health surveillance data can help identify the need for interventions among specific populations. But a real-world focus thrives on causal conclusions and implications; without addressing causality, this journal’s contributions to the promotion of sexual and reproductive health (like the contributions of similar journals in other applied fields) would be unnecessarily limited. However, the demonstration of meaningful causality with real-world implications is rarely straightforward; in fact, it is usually difficult and often quite messy. Statistical methods such as regression analysis can help guide us through the swamp. But methods do not in themselves provide valid inferences—they are just tools to be skillfully used in the quest for such inferences.