خريطة عمل بطاريات الاختبار

This is the usual condition:

the researcher knows only one type of "test" and applies it to any data.

These are the conditions that should be considered:

(i) the kind of response variable (including assumptions about the data-generative process leading to observed values, based on logic, experience, and analogies), and (ii) the experimental design.

Too frequently: the research was designed to fit the only test the research was familiar with. The design interacted with the system being tested to give the result that was expected. This is further degraded by minimal sample sizes that are fantastic if all of our assumptions hold true and we do not want enough data to figure out by how much we are wrong.

Based on logic and experience (time and funding), develop an experimental design sufficiently robust and internally consistent that you can reliably answer the question being asked. Additionally have model systems where you can ask the question in other ways to see how methodology may influence the outcome.

One step better is to use logic and experience to set up a simulation that then is used to generate an "expected model" which is then compared to real data. However, getting the simulation might be a research project in its own right. The push against this approach is that research and publication are evaluated based on being novel. If everything is "novel" then it is harder to use experience to set up the experimental design.