Module 09 Summary

What this module established

Every slope in a multiple regression is conditional on the other predictors. An unadjusted training coefficient of 888 became 82 once experience entered — and the true value was 50.

Carry forward

• An interaction coefficient is a DIFFERENCE of slopes. The social channel's slope was 3.099 − 2.136 = 0.963, not −2.136.
• R-squared cannot detect the wrong shape. The curved model had the higher R-squared; only its residuals, swinging +4.85, −10.24, +5.39, gave it away.
• One row in 41 changed a slope from 0.522 to 2.295, with a Cook's distance of 38.0 against a next-largest of 0.09.
• Twenty pure-noise predictors reached R-squared 0.26 on n = 60, while adjusted R-squared went NEGATIVE at −0.116.

Before moving on

Move on when your residual inspection is clean, your influential points are investigated, and no predictor entered because of its p-value.

Mark complete