What is the difference between differential and non-differential misclassification?

Non-differential misclassification means measurement errors are unrelated to the other variable and typically (for a binary exposure) bias toward the null; differential misclassification means the error depends on the other variable and can bias the estimate in either direction.

Is recall bias a type of information bias?

Yes. Recall bias is a differential information bias in which subjects with the outcome remember or report past exposures differently from those without it, distorting the measured association.

How does information bias differ from selection bias?

Information bias comes from how variables are measured or classified, whereas selection bias comes from who is included in or retained for the analysis.

Information Bias

Information bias is a systematic error that arises from inaccurate measurement or classification of exposure, outcome, or covariates. When subjects are placed into the wrong category — exposed counted as unexposed, or diseased as non-diseased — the resulting misclassification can distort the estimated association. Its effect depends critically on whether the errors are unrelated to the other variable (non-differential) or related to it (differential).

Definition

Information bias is a distortion of an exposure-outcome association caused by error in the measurement or classification of exposure, outcome, or other study variables, such that subjects are systematically assigned to incorrect categories.

Scope

The entry covers the sources of measurement error, the key distinction between non-differential and differential misclassification and their typical directions of effect, and common named forms such as recall and interviewer bias. It is a methodological reference and provides no clinical guidance.

Core questions

How accurately were exposure and outcome measured or classified?
Is the misclassification non-differential or differential with respect to the other variable?
What direction would the misclassification be expected to push the estimate?
Could the measurement process itself depend on knowledge of exposure or disease status?

Key concepts

Misclassification
Non-differential misclassification
Differential misclassification
Recall bias
Interviewer / observer bias
Sensitivity and specificity of measurement
Regression dilution

Mechanisms

Information bias originates in imperfect measurement instruments, fallible recall, or inconsistent classification. Its consequences hinge on a key distinction. Non-differential misclassification, where errors are unrelated to the other variable, typically (for a binary exposure with two categories) biases the estimate toward the null, blurring a real effect. Differential misclassification, where the error in one variable depends on the value of the other — for example, cases recalling past exposures more thoroughly than controls (recall bias), or interviewers probing exposed subjects more closely (interviewer bias) — can bias the estimate in either direction and is harder to anticipate. Because information bias arises within data collection, it is conceptually distinct from confounding (a common cause) and selection bias (driven by who is included). Measurement error in a confounder can also leave residual confounding even after adjustment.

Clinical relevance

Information bias is one reason a reported association may be too strong, too weak, or pointed the wrong way, so scrutinising how exposures and outcomes were measured is part of appraising evidence. The concept describes how study findings can be distorted; it is not advice for any individual's diagnosis or treatment.

Epidemiology

Information bias is a concern in every design but is especially salient where exposure is reported after the outcome is known — for instance recall bias in case-control studies — and where outcome ascertainment may differ by exposure status. Awareness of misclassification motivates validation sub-studies and blinded or standardised measurement.

History

Recall, interviewer, and observer biases were catalogued through twentieth-century epidemiology as recurring threats in observational studies. The formal treatment of misclassification in terms of sensitivity and specificity, and the result that non-differential error usually biases toward the null, became standard parts of methodologic texts, while glossaries such as Delgado-Rodríguez and Llorca (2004) organised the many named information biases.

Debates

Does non-differential misclassification always bias toward the null?: The toward-the-null result holds under common conditions (notably independent, non-differential error in a dichotomous exposure), but exceptions arise with more than two exposure categories or with dependent errors, so the 'always conservative' intuition can mislead.

Key figures

Kenneth Rothman
Sander Greenland
Miquel Delgado-Rodríguez

Seminal works

delgado-rodriguez-2004
grimes-schulz-2002-bias

Frequently asked questions

What is the difference between differential and non-differential misclassification?: Non-differential misclassification means measurement errors are unrelated to the other variable and typically (for a binary exposure) bias toward the null; differential misclassification means the error depends on the other variable and can bias the estimate in either direction.
Is recall bias a type of information bias?: Yes. Recall bias is a differential information bias in which subjects with the outcome remember or report past exposures differently from those without it, distorting the measured association.
How does information bias differ from selection bias?: Information bias comes from how variables are measured or classified, whereas selection bias comes from who is included in or retained for the analysis.