Quantitude S6E12 Suppression Unrepressed
Jan 7, 2025
Dive into the intriguing world of suppressor variables that magically enhance prediction without direct links to dependent variables. Discover humorous tales, from lunch with a titan to Sigmund Freud's theories. Explore defense mechanisms in academia and their quirky impact. Unravel how introducing new variables—like a Cookie Monster metaphor—can clarify complex relationships in statistical models. Plus, delightful anecdotes about street magicians and squirrel-proof bird feeders add charm to these enlightening discussions!
AI Snips
Chapters
Transcript
Episode notes
How Suppressors Boost Predictive Validity
- A suppressor variable increases the predictive validity of other predictors despite having no direct relation to the outcome.
- Classic case: Z correlates with X but not Y, and including Z boosts X's explained variance (e.g., X goes from 10% to 20%).
Cookie Monster Venn Diagram For Suppression
- Suppression works by removing irrelevant variance from a predictor so its relevant component better overlaps the outcome.
- Patrick uses a Cookie Monster Venn diagram: Z 'bites out' the non-overlapping part of X, increasing X's proportional overlap with Y.
Paul Horst's Pilot Ability Example
- Paul Horst's pilot ability example shows verbal ability as a suppressor that removed irrelevant variance from mechanical and spatial tests.
- Verbal test skills were unrelated to piloting but removed the 'verbal' noise in paper tests so reasoning scores better predicted pilot skill.