Power and effect size, before the experiment rather than after
Lesson 5 of 6 · Statistics you can defend
1 · Learn the move · Flipped interaction
Power analysis answers one question: given the smallest effect worth caring about and the noise you expect, how many biological replicates buy a real chance of seeing it? Run before the experiment, it sizes the work. Run after, on the effect you happened to observe, it is arithmetic theater - observed power restates the p-value and adds nothing. The flipped interaction fits because the inputs live in your head, not the model's: smallest effect of scientific interest, variance from a pilot or the literature, alpha, design. Let the model interview you, one question at a time, then draft the calculation as a script you run. The number comes from the script and goes in your log with its inputs, where a reviewer can check all three.
Interview me to set up a power analysis for [experiment]. One question at a time, wait for each answer: smallest effect size of scientific interest and its units, expected variability and its source (pilot, literature, guess - label which), alpha and sidedness, design and any pairing. Then draft a script that computes required n, with every input as a named variable and a comment citing its source. Do not output a bare n from your own arithmetic - the script I run is the record. If any input is a guess, have the script print the n across a range for that input.
2 · Your turn. You write the prompt
You are sizing a mouse cohort for a metabolic study before the animal protocol goes in, and the committee form asks for a power justification. You have a pilot with 6 animals and a hunch that the interesting effect is around a 15 percent change. Let the model interview you, then produce the script and the sentence for the form.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.