Week 10 - Assumptions + Diagnostics
Is the Model Telling the Truth?
For four weeks we have been glancing at residual plots and saying “nothing obviously wrong.” This week each of those glances becomes a named assumption, with a diagnostic, a consequence when it fails, and something to do about it.
We work two examples on purpose. One is a model you already know (sleep, stress, and depression in the ASAP study) that passes with only a small caveat. The other is the Spotify model from Week 7, with an R² of .62, that breaks every rule until a single transformation fixes it. Both outcomes are normal. Most of the time, diagnostics take five minutes and one sentence in your write-up.
Regression assumptions are claims about the residuals, not about your raw variables. A skewed predictor is fine. A skewed set of leftovers is worth a look.
Prepare
Slides
Data
Lecture: ASAP Study (.csv) · Popular Spotify Songs (.csv) · Study, Tutoring & Grades (.csv)
In class: Same data from activity last week <DASS survey data (.csv, 20 MB)
For Next Time
📋Lab 9 - Groups, Slopes & Interactions
In-Class Check at the start of class.
Back to course schedule ⏎