The effect of misclassification on the estimation of association: a review

Area of Science:

• Epidemiology
• Psychiatric Research
• Biostatistics

Background:

• Misclassification of categorical variables is a pervasive issue across scientific disciplines.
• It critically impacts the accuracy of association measures, particularly in psychiatric research.
• Commonly used association metrics like risk ratio and odds ratio are susceptible to bias.

Purpose of the Study:

• To review the consequences of misclassification in statistical analysis.
• To demonstrate the magnitude of bias introduced by misclassification using a numerical example.
• To highlight the impact of misclassification on confidence intervals.

Main Methods:

• Review of statistical literature on misclassification.
• Numerical example illustrating bias in risk ratio estimation when misclassification occurs in the exposure variable.
• Discussion of non-differential misclassification scenarios.

Main Results:

• Misclassification introduces significant bias in risk ratio and odds ratio estimates, especially when probabilities are not negligible.
• Non-differential misclassification (solely in one binary variable, independent of the other) biases estimates towards the null value.
• Ignoring misclassification typically results in confidence intervals that are artificially narrow, and uncertainty can broaden them.

Conclusions:

• Misclassification poses a substantial threat to the validity of research findings, necessitating careful consideration.
• While avoiding misclassification by design is optimal, statistical methods can mitigate bias when it cannot be prevented.
• Accurate measurement and appropriate statistical handling of misclassification are crucial for reliable research outcomes.