Questions: Exposure Measurement Error and Exposure Assessment

Question 1

A case-control study finds no association between dietary fat intake (measured by self-report questionnaire) and heart disease. A validation study shows the questionnaire produces non-differential misclassification of fat intake. What is the most appropriate interpretation of the null result?

• A. Dietary fat is genuinely unassociated with heart disease in this population
• B. The null result may be a false negative caused by attenuation bias: non-differential misclassification moves the estimated odds ratio toward 1.0
• C. The null result is a false positive — non-differential error inflates associations toward the null
• D. Differential recall bias among cases is masking a true association

Answer: (B)

Non-differential misclassification mixes true exposure categories: high-fat eaters are sometimes classified as moderate, moderate as low. This averaging shrinks the apparent contrast between groups and biases relative risk estimates toward 1.0 (the null). A null finding in this context may reflect attenuation bias rather than a true absence of association — a classic false negative. The study cannot rule out a real effect.

Question 2

A researcher suspects that women diagnosed with breast cancer recalled their past hormone use more thoroughly than healthy controls. If this differential recall bias is present, the estimated odds ratio will most likely be:

• A. Biased toward the null, making the association appear weaker than it is
• B. Biased away from the null, making the association appear stronger than it is
• C. Unaffected, because subjective recall errors cancel out across large samples
• D. Biased toward the null, but only if hormone use is a rare exposure

Answer: (B)

Cases who are more thorough reporters of exposure will appear to have higher exposure prevalence than controls, inflating the apparent association. Differential misclassification biases the OR away from the null here. This is the canonical recall bias scenario in case-control studies of cancer. Critically, differential error can bias in either direction depending on which group over- or under-reports — its direction is not predictable without knowing the error structure.

Question 3

Non-differential misclassification of a binary exposure typically produces an observed relative risk closer to 1.0 than the true relative risk.

• T. True
• F. False

Answer: True

True. Non-differential misclassification mixes exposure categories symmetrically across disease groups, diluting the true contrast. The observed association is attenuated — biased toward the null — because misclassified individuals are essentially counted in the wrong group, reducing the apparent difference between truly exposed and unexposed. This is sometimes called the 'dilution' effect.

Question 4

Differential misclassification usually biases effect estimates toward the null, so it makes associations appear weaker than they truly are.

• T. True
• F. False

Answer: False

False. This is the key distinction between the two error types. Non-differential misclassification predictably biases toward the null. Differential misclassification — where the error pattern differs between cases and controls — can bias estimates in either direction: toward the null, away from the null, or even reverse the direction of an association. Its unpredictability is precisely what makes it more dangerous: a researcher cannot use the usual 'conservative' interpretation.

Question 5

Why is differential misclassification considered more dangerous than non-differential misclassification in epidemiologic studies? What makes the direction of bias unpredictable?

Model answer: Non-differential misclassification has a predictable effect: it dilutes true contrasts and biases estimates toward the null. Researchers can treat null findings with skepticism and know the direction of the bias. Differential misclassification occurs when the error pattern differs between cases and controls — for example, if cases recall past exposures more thoroughly than controls (recall bias). In this situation, the bias can move the estimate toward or away from the null, or even create an apparent association where none exists. Because the direction depends on which group over- or under-reports and by how much, it cannot be predicted from first principles without validation data. This means differential misclassification can produce false positives, false negatives, or distorted effect sizes in ways that are not transparent to the reader.

The practical upshot is that non-differential error makes studies conservative (underestimates real effects), while differential error can make studies misleading in either direction. Exposure validation is the only way to quantify and correct for the actual error structure rather than guessing at its direction.