Contending with Bias
Bias presents a conundrum. On the one hand, the presence of bias is viewed as a flaw that invalidates the results of a comparative study. On the other hand, nearly every study is affected to some extent by bias, and most scientific progress has been based on such imperfect research. This chapter discusses the ways that bias has traditionally been addressed and suggests the need for a broader perspective.
We have divided the sources of bias into five general categories:
- Sampling of the study subjects
- Assignment to exposure states
- Adherence to nominal exposure states
- Exposure ascertainment
- Outcome measurement
Assignment bias is widely considered to be most critical and has received by far the greatest attention. Internal validity is often accepted as the sine qua non of causal inference. Conversely, other forms of bias are often viewed as much less serious concerns. While recognizing the possibility of these other forms of bias, investigators tend to downplay their importance. As a result, many methodologists have come to adopt a very skeptical attitude toward conclusions reached in the absence of randomized experimentation but to be somewhat more cavalier about other threats to validity.
The causal modeling perspective suggests that traditional attitudes may be shortsighted. Bias is an almost unavoidable fact of research life that must be taken into account in evaluating a study’s methodology. Rather than an attempt to engineer perfection, the research ...