Causality and Comparative Studies

Virtually any statistical analysis of practical importance is potentially susceptible to various forms of bias. Therefore, it might be supposed that biased comparisons would be of central interest to statisticians and other research methodologists. In fact, the field of statistics has traditionally treated bias more as a minor annoyance than a topic worthy of serious academic consideration. In scientific articles, discussion of possible bias is often relegated to a few paragraphs on the “limitations” of the study. These afterthoughts often have a perfunctory quality; they seem intended more to deflect criticism than to invite a serious evaluation of the study’s validity.

Why has the subject of bias received so little attention by statisticians? Perhaps the main reason is that any meaningful discussion of bias in comparative studies cannot avoid dealing with the core concept of causality. Unlike ordinary statistical concepts, bias pertains explicitly to the causal nature of things. The existence of bias cannot be determined by mere calculation; it depends as well on the interpretation of the data in light of some causal theory. Thus, bias has become an “inconvenient truth” for the almost strictly mathematical discipline that statistics has become.


As we suggested in Chapter 1, the field of statistics has long regarded causality as lying outside the bounds of its scientific mandate. Statistics deals with empirical ...

Get Bias and Causation: Models and Judgment for Valid Comparisons now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.