In the previous sections, we learned about different study designs and the types of effect measures that are used to describe associations between exposure and outcome. When an association is detected, we must question whether we can derive causal inferences from that association. In other words, does the observed association reflect the “true” association? Can we conclude that the exposure has an independent and actual effect on the risk of an outcome?
Issues of bias and confounding can affect the validity of study results and our ability to derive causal inferences. In this section, we provide a definition of bias and discuss the types of bias that we encounter in epidemiological study designs. We will discuss confounding in the next section.
Definition of bias
Bias is any systematic error in the design or conduct of a study that results in the deviation of the observed measure of association from the true measure of association.1 Bias affects the validity of our results and our ability to draw conclusions from a study. Although bias cannot be completely eliminated, researchers should aim to minimize biases and to identify/account for biases when interpreting results.1 There are two main categories of bias that we commonly encounter in epidemiological studies — selection bias and information bias. Note that while bias can occur in all study designs, they’re most common in observational study designs. We’ll focus on biases encountered in case-control and cohort studies throughout this section.