3

I am a statistician working as an applied analyst on a paper. As I went through a set of analyses, I decided upon a detailed and thorough data analysis method using a regression model. Traditionally, that regression model is presented alongside a set of assumptions that one should check in the model approach. I had a gut feeling those assumptions didn't matter in a broad sense. I performed some simulation studies to show that that was the case. The results are interesting and we will attempt to publish a paper about those as well.

However, turning to the present analysis, should I present without proof the results of the "robust" interpretation where assumption checking is eschewed. Does it make sense to await a reviewer to raise a flag about this issue? There is a lot of pressure to publish this quickly... which is not unusual or surprising. Would it make more sense to include a sentence saying, "The authors will show in later work that the assumptions of the modeling approach can be relaxed when (blah blah blah)"?

  • I think this depends largely on the nature of the work. Most applied journals these days rarely speak of statistical assumptions. In contrast, if this is a statistics paper then assumptions might be relevant. Does the paper tell a story even if assumptions are not discussed? – Behacad May 20 '14 at 17:53
  • @Behacad it is not destined for a statistics journal. Most likely a pain journal, but with a very public health oriented focus. The story that is told is a very compelling one. The statistical results and conclusions are consistent with that story. It's just that I was worried whether or not the statistics were "right" and it turns out that the assumptions don't matter, which is interesting on its own. I endeavor to be thorough in describing analyses, so I am fine with brushing aside the detail to be established at a later point. – AdamO May 20 '14 at 18:53
3

(I have turned my comments into an answer).

Whether or not to discuss statistical assumptions depends largely on the nature of your work. Many authors of statistical textbooks and manuals discuss assumptions in great depth. They also highlight the importance of reporting the results of assumptions testing in your manuscripts. In my experience, this is very rarely done for several reasons. For example, results are often the same regardless of assumptions (this is not always the case), if you are reporting statistically significant results and effect sizes then perhaps the need to discuss assumptions is less important, word limits often restrict the ability to discussion assumptions, and it is often very rare these days to find discussion of statistical assumptions. Note that these points relate to applied journals and papers. If the paper is statistical in nature, or you have non-significant results and need to highlight that your statistics were appropriate, then perhaps discussion of assumptions is warranted. I very rarely run across discussion of assumptions for statistics ranging from t-tests to survival analyses and latent growth modelling etc. In statistics papers though its often important to highlight what assumptions are needed and such.

I have published in the pre-eminent journals on pain and you certainly do not need to discuss assumptions in them most of the time. If you review recent papers you will see that these assumptions are very rarely discussed. A simple "we did regression" and "here are results of regression" will suffice (with more elaboration, of course!).

1

I wouldn't claim that you're going to do anything. If you think the assumptions are really problematic for the result and want to indicate that they can be relaxed, you could say something like "preliminary investigations of alternate models which do not assume such-and-so are consistent with these results (data not shown)".

If the reviewers insist, you can then decide whether it's worth it to put a minimal amount of that analysis in the paper.

0

There is one issue with publishing the proof in a subsequent: if the proof is too simple, you will probably have difficulties to publish it. In addition, if experts of your field find the proof is obvious (without reading your "later work"), your current paper may lose some credibility.

I would wait a bit (say 2 weeks) before publishing in order to have time to work seriously on the proof and check it is really an interesting work. Alternatively, I would send the paper to the journal after organizing it in such a way that I can add the proof without changing the structure (adding the proof in the appendix plus a quick comment in the main body - it is suitable for a non-math paper). Reorganizing the paper after submission is inappropriate since it does not respect the work the referee already did.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.