I have had feedback from a reviewer that I need to restate my hypotheses as null hypotheses. My supervisor disagrees with this suggestion. Having done a bit of digging, I found Szucs & Ioannidis (2017) (https://www.frontiersin.org/articles/10.3389/fnhum.2017.00390/full) agree that in most cases testing the null is very limiting, because (amongst other concerns they list) it does not take account of the probable outcome of the research given prior research results and theoretical framing of the issue. I prefer to state the directional alternative hypothesis since these statements are very precise about what we are looking for and what we expect to find; the findings therefore respond to a very specific question. I understand there's a long history of convention etc. behind stating the null, though - what is the current thinking on this?
if you wish to draw inferences using statistical tests then you should express a null hypothesis since the text express the degree to which this is unlikely. A useful book for you might be the one by sidney seigel on statistical tests for the behavioural sciences
There are many criticisms of NHST, but I don't think it is going away anytime soon. I think you will just have to do what the reviewer says. I often do what the reviewer says as well as provide my alternative. In your case, do the traditional null hypothesis phrasing, then state specifically what you expect the alternative might be. It sounds like it won't be too hard to make everyone happy.
I also feel like Christopher, you need to follow reviewer's suggestion. In that case you include both in your research paper. Also, your alternative one should be specific, clear in objective and precise.
Many thanks for the responses. I understand that reviewers have a lot of power and I in fact did change the draft to state the null while also keeping in our alternative directional hypotheses, but my supervisor wanted me to change it back because she said it is actually poor practice. I am therefore more interested in what is considered best practice these days (or are we in the middle of a shift?) rather than how to manage reviewers per se. My supervisor's comments are: "Statistically, we ARE testing the null hypothesis already. Most statistical tests begin by assuming the null hypothesis is true, and then test the alternative against the null. .05 is set as an error rate to test how likely would it be to get the data you got in your study IF the null hypothesis was true. However, if your introduction leads you to expect one more than the other, you write just one hypothesis at the end of the introduction, and word it as a directional hypothesis. The hypothesis is expressed as the alternative hypothesis, and the statistical test is showing whether we can reject the null in favour of the alternative. If you swap it around and are trying to reject the alternative hypothesis , it changes the alpha rate to 95%! Statistically, we nearly always test the alternative hypothesis against the null hypothesis even if we expect the null to be retained." I would therefore be interested in hearing what the current thinking is regarding stating the null. Cheers!
Bryan R Lawson Christopher A Varnon & Shirshendu Roy
- Badges
- Science topic
- Similar topics
- Cognitive Science
- Thinking