Information bias (epidemiology)

In epidemiology, information bias refers to bias arising from measurement error.^[1] Information bias is also referred to as observational bias and misclassification. A Dictionary of Epidemiology, sponsored by the International Epidemiological Association, defines this as the following:

"1. A flaw in measuring exposure, covariate, or outcome variables that results in different quality (accuracy) of information between comparison groups. The occurrence of information biases may not be independent of the occurrence of selection biases.
2. Bias in an estimate arising from measurement errors."^[2]

Misclassification

Misclassification thus refers to measurement error. There are two types of misclassification in epidemiological research: non-differential misclassification and differential misclassification.

Nondifferential misclassification

Nondifferential misclassification is when all classes, groups, or categories of a variable (whether exposure, outcome, or covariate) have the same error rate or probability of being misclassified for all study subjects.^[2] It has traditionally been assumed that in the case of binary or dichotomous variables nondifferential misclassification would result in an 'underestimation' of the hypothesized relationship between exposure and outcome. However, this has more recently been challenged in that results of individual studies represent a single estimate and not the average of repeated measurements and thus can be farther (or nearer) from the null value (i.e. zero) than the true value.^[3]

Differential misclassification

Differential misclassification occurs when the error rate or probability of being misclassified differs across groups of study subjects.^[2] For example, the accuracy of blood pressure measurement may be lower for heavier than for lighter study subjects, or a study of elderly persons may find that reports from elderly persons with dementia are less reliable than those without dementia. The effect(s) of such misclassification can vary from an overestimation to an underestimation of the true value.^[4] Statisticians have developed methods to adjust for this type of bias, which may assist somewhat in compensating for this problem when known and when it is quantifiable.^[5]

References

^ Rothman, K.; Greenland, S.; Lash, T. (2008). Modern Epidemiology (Third ed.). Philadelphia: Lippincott Williams & Wilkins. p. 137. ISBN 978-0-7817-5564-1.
^ ^a ^b ^c Porta, M., ed. (2008). A Dictionary of Epidemiology (Fifth ed.). New York: Oxford University Press. p. 128. ISBN 978-0-19-531449-6.
^ Jurek, A. M.; Greenland, S.; Maldonado, G.; Church, T. R. (2004). "Proper interpretation of non-differential misclassification effects: Expectations vs observations". International Journal of Epidemiology. 34 (3): 680–687. doi:10.1093/ije/dyi060. PMID 15802377.
^ Copeland, K. T.; Checkoway, H.; McMichael, A. J.; Holbrook, R. H. (1977). "Bias due to misclassification in the estimation of relative risk". American Journal of Epidemiology. 105 (5): 488–495. PMID 871121.
^ Greenland, S. (1988). "Variance estimation for epidemiologic effect estimates under misclassification". Statistics in Medicine. 7 (7): 745–757. doi:10.1002/sim.4780070704. PMID 3043623.

v t e Biases
Cognitive biases	Acquiescence Ambiguity Affinity Anchoring Attentional Attribution Actor–observer Correspondence Authority Automation Availability Mean world Belief Blind spot Choice-supportive Commitment Confirmation Selective perception Compassion fade Congruence Cultural Declinism Distinction Dunning–Kruger Egocentric Curse of knowledge Emotional Extrinsic incentives Fading affect Framing Frequency Frog pond effect Halo effect Hindsight Horn effect Hostile attribution Impact Implicit In-group Illusion of transparency Mean world syndrome Mere-exposure effect Narrative Negativity Normalcy Omission Optimism Out-group homogeneity Outcome Overton window Precision Present Pro-innovation Proximity Response Restraint Self-serving Social comparison Social influence bias Spotlight Status quo Substitution Time-saving Trait ascription Turkey illusion von Restorff effect Zero-risk In animals
Statistical biases	Estimator Forecast Healthy user Information Psychological Lead time Length time Non-response Observer Omitted-variable Participation Recall Sampling Selection Self-selection Social desirability Spectrum Survivorship Systematic error Systemic Verification Wet
Other biases	Academic Basking in reflected glory Déformation professionnelle Funding FUTON Inductive Infrastructure Inherent In education Liking gap Media False balance Vietnam War Norway South Asia Sweden United States Arab–Israeli conflict Ukraine Net Political bias Publication Reporting White hat
Bias reduction	Cognitive bias mitigation Debiasing Heuristics in judgment and decision-making
Lists: General Memory

Misclassification

Nondifferential misclassification

Differential misclassification

References

Further reading