Model vs. test


TS Contributor
we just had an interesting discussion about the chi-squared test, that could be generalized. I think the core of the discussion was whether one should prefer a simple statistical test or a statistical model - e.g. a chi-squared test vs. a logistic regression.

Given that in practice one is most often also interested in questions that the model can answer but the test can not, I would suggest that except for rare, clear-cut cases, building a model is preferable to simply running a test.

What do you think?


Less is more. Stay pure. Stay poor.
Agreed. It took me a little while before I got on board with this earlier in my career, but now that I know logistic reg well, I try to stay in that boat most of the time.

Benefits: can easily add options to address sparse data, get odds ratios, get bootstraps or other CIs, get accuracy metrics, and most importantly control for other variables and run contrasts or estimates.