On this page
Identify the missing model inputs
A test starts with a null reference model and calculates a statistic’s tail area under it. It has not assigned prior probabilities to the null and alternatives. It has not specified a distribution of possible effects under an alternative.
Rewrite the sentence without reversing it
For a fictional upper normal test with p = 0.02, a suitable description is that the null model assigns probability 0.02 to a statistic at least as large as the observed one. The statement depends on the specified model and planned upper direction.
List the separate posterior ingredients
| Ingredient | Question |
|---|---|
| Prior | What probability is assigned before this data? |
| Null likelihood | How is the observed data modeled under the null? |
| Alternative likelihood | How is it modeled under a specified alternative? |
| Updating | How are those pieces combined? |
Keep the reported effect visible
A probability interpretation alone does not describe the estimated effect’s magnitude or usefulness. Report the estimate, uncertainty and design next to the p-value. Do not replace the mistaken 2% claim with an equally unsupported 98% probability of profit.