Definitions

Confounding - a distortion of an association between an exposure and disease brought about by extraneous factors.

Interaction - the interdependent operation of two or more factors to produce an unanticipated effect (a statistical concept).

Effect modification - difference in the biologic effect of an exposure according to the presence or absence of a cofactor.

Parameter - the error-free, hypothetical (unknown), numerical quantity that decribes something about a population (e.g., RR, OR, µ, "sigma-squared").

Statistic - an error-prone, observed (calculated), statistical function of a sample (e.g., RR^, OR^, "xbar," s).

Estimator - a statistical function that calculates the probable location of a parameter (e.g., RR^ is the estimator of RR).

Random error - error that results from chance. Random error tends to equilibrate ("balance out") over the long run. The standard statistical techniques of estimation and hypothesis testing address random error in studies.

Systematic error - bias, error that results in a directional error of an estimate, usually because of flaws in a study's design. Systematic error tends to be directional (i.e., either overestimate or underestimate risk).

Selection bias - a study bias due to errors in sampling, so that the sample tends to be nonrepresentative of the population.

Information bias - a bias due to measurement error or misclassification.