-
Set the Gate
Write the one question your model must answer. If you’re listing pairwise comparisons, you need ANOVA or planned contrasts… not whack-a-mole.
-
Define “Meaningful” Before “Significant”
Pick the smallest effect that would change a choice. If your margin doesn’t steer behavior, you’re measuring trivia.
-
Visual Proof of Life
Plot distributions and intervals before testing. If the picture and the p-value disagree, fix the model, not the plot.
-
Alpha With a Seatbelt
State alpha and the multiple-comparisons plan. If you test more, protect more. Curiosity needs guardrails.
-
Size + Bounds
Always pair effect sizes with confidence intervals. Magnitude without uncertainty is marketing, not inference.
-
Assumptions Are Design Choices
Independence, measurement, sampling… decide them up front. Assumptions unchecked become conclusions undone.
-
Save the Recipe
Version the data, seed randomness, save code or output. Reproducible beats impressive.
-
End With a Verb
Close every report with action: implement, pause, collect more. Numbers without verbs don’t change the world.