Who this is for: This is the advanced, more technical companion to Residual gaps, explained. It is written for HR, works councils and anyone who wants to go a level deeper into the statistics behind a residual gap - not required reading, but useful when you want to understand why a gap is trustworthy rather than just read the number. Comfortable with the basics is enough; no statistics background is assumed.
Description: This deck takes the model summary panel PayGap produces for every run and explains each line in turn. It covers why the model is fit on log(pay) and how that turns pay drivers into percentage steps; the two fit metrics (R² and adjusted R², and what a widening gap between them signals); the two significance checks (the p-value and the F-statistic, and why they are read together); the two comparison scores (AIC and BIC, meaningful only when comparing runs); and the smearing factor that corrects the log-to-euro back-conversion. It closes with a one-question-per-metric recap and a five-step sequence for validating a model before acting on any gap.
How this helps you: Read the model summary as a set of health checks before acting on a residual gap. The material tells you when a gap is reliable enough to act on (high R² with adjusted R² close behind, p-value under 0.05, a high F-statistic, a smearing factor near 1.0, and at least five employments per factor) and when the model needs more tuning before any number can be trusted. The validation steps run in order - each one gates the next - so you know there is no point reading fit before you have confirmed there is enough data.