Stating your design before asking for a test
Lesson 2 of 6 · Statistics you can defend
1 · Learn the move · Context + audience
A test question without a design statement is a request for a guess. The model cannot see your bench, so every design fact you leave out gets replaced by the most common case in its training data - unpaired, independent, normal, generous n. The fix is mechanical: write the design down before the question. Experimental unit. Factors and levels. Biological n and technical n, and how they nest. Paired or unpaired, and why. Outcome type and how it is measured. Known structure: plates, batches, days. Then ask for candidate tests with what each assumes, not for the answer. A design statement takes four minutes and converts the model from a guesser into a consultant who has actually read the protocol.
Here is my design, stated before any test question. Experimental unit: [unit]. Factors and levels: [factors]. Replication: [biological n, technical n, nesting]. Pairing: [paired or unpaired, and why]. Outcome: [type, how measured]. Known structure: [plates, batches, days]. Given only this design, list candidate analyses with what each assumes and what would rule each out. Mark anything you still need as [NEEDS: ...]. Do not pick one yet.
2 · Your turn. You write the prompt
Your qPCR experiment crosses two genotypes with two treatments, three biological replicates each, every sample run in technical triplicate, spread across two run days. You have been asking "what test do I use for qPCR data" and getting generic answers about t-tests. Write the design statement first, then the question.
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.