Summary
The episode works through five questions on information bias and data quality. The hosts calculate how nondifferential misclassification weakens an odds ratio and show that poor specificity can matter more than poor sensitivity when an exposure is uncommon. They then show how differential recall can produce an association where none exists, and they debate whether a study can describe its estimate as conservative. Later segments work through surveillance bias, regression dilution with repeated measurements, digit preference near clinical thresholds, and the racially patterned error of pulse oximeters.