Six Sigma Basics: Correlation & Regression

A: 50 N·m is the hard boundary. A slope that small across the full span hints the x-axis is scaled in a smaller unit, a classic vendor paste error when lbf·in data gets relabeled as SI. Physics doesn't force zero intercept once bias exists, R² thresholds aren't mandated, and heteroscedasticity needs context before killing a startup.

A: Cause-and-effect is the boundary. ISO 13053 frames improvement around CTQs and verified relationships, not just numerical association. There’s no universal r threshold, scope exclusions like that don’t exist, and p-values alone don’t show causality.

A: n=8 is the number that bites. At that size, leverage is high and slope stability is poor, making it unsuitable for decisions tied to safety. R² and range help but don't fix sensitivity to single points.

A: Three points is the trigger engineers watch. A short run trending the wrong way suggests process shift not yet breaching limits. Specs alone lag behavior, while knee‑jerk shutdowns or cherry-picking data miss the signal.