Summary
The episode works through four questions that students often find difficult in questionnaire design. It uses a hypothetical comparison of two surveys to show that non-response bias depends on both the share of people who did not respond and the gap between responders and non-responders, and it discusses the 2011 National Household Survey as a Canadian case. It then explains how random measurement error attenuates associations, works a multi-step example that combines the Spearman-Brown formula with attenuation, and contrasts random error with the systematic under-reporting of weight found in the Canadian Community Health Survey. The hosts examine how response options and scale ranges shape answers, show how an undeclared don't-know code can produce a plausible but wrong mean, and close with a debate on whether means are appropriate for Likert data that ends in a shared rule for reporting single items and multi-item scales.