Hello and happy new year,

I was wondering.... is it possible to add a hypothesis that you hadn't previously planned for a study?

The long and short of it is... I'm analysing the results of an experiment (developmental psychology), and although the predictions have not been met, I have a hunch that I might see some interesting effects if I analyze a different aspect of the data. Is it above board to state this as an exploratory hypothesis and do the relevant analyses, or should everything be pre-planned?

Anyone come across this kind of thing before?

Tudor

Think I read an article somewhere that adding in a hypothesis post-hoc is bad practice, but I may have remembered wrongly. I am sure your supervisors would have better answers to your question.

Hi Chaotic

Thanks for your response. I've just been having a google and come across some really interesting (recent) papers on this kind of thing. There are lots of variations of it... and at worse it is fishing/data dredging/p-hacking (i.e., basically hunting through your data to see how you can get significant results), and is completely wrong yet widely done. On the other hand, if the author is transparent about what they are doing and why (i.e., states that it is a posthoc hypothesis for x reason, as opposed to presenting it as a a priori hypothesis) it seems it is acceptable (I found a few examples of this kind of transparent reporting in my search).

For anyone interested:
https://link.springer.com/content/pdf/10.1007%2Fs10869-016-9456-7.pdf
https://en.wikipedia.org/wiki/Testing_hypotheses_suggested_by_the_data

Tudor

Ps. I like to get opinions, chat to people, and come to my own conclusions before approaching my supervisors :)

I think its fine to say that your have unexpected and interesting results if the data were interpreted in a certain way, and speculate on the implications. But to actually insert new hypotheses after the research was done seems too much like torturing data to yield the results you want. After all, didn't someone once said that if you torture data sufficiently, they can confess to anything?

But as I said, I am no expert on the matter, and have yet to start on the PhD journey. This is based on readings from a while ago, and I might not have recalled it properly.

It depends entirely on the specifics of what you are doing. If I were to conduct additional analyses posthoc and make it clear this is what I was doing and why, then there is no problem. There are some (only a few I have been able to find) good examples of it in APA journals.

Unfortunately, the p-hacking/data dredging/torturing literature seems to suggest that many researchers are doing this and NOT stating it - but rather, they are presenting them as a priori hypotheses. And that is where there is an issue.

Back to my own research... I don't know if I can be bothered to do further analyses! I'll see if the sups think it's worth it.

You can tell the story of your research exactly as you want.
I often deliberately try NOT to come up with a hypothesis to avoid confirmational bias and a subsequent closed mind if results dont pan out the way I want.
There is absolutely nothing wrong with running an experiment for the hell of it and then trying to see what story emerges. If you then decide to tell people that you were looking for that result all along then that is up to you. Nobody has to see the scaffolding behind your thought process and it appears to me that researchers go out of their way to obfuscate how and why they did something.
In my opinion, the reason for trying something, or the hypothesis, is completely irrelevant. My philosophy is that I couldn't care less what you thought might happen. I only care about what you discovered.

I get your point but if you are doing statistical tests then you need a pre-specified hypothesis - it is part of the methodology (Null Hypothesis Signficance Testing - NHST). If you don't have one then the method isn't being applied correctly - as the idea is to reject the null hypothesis (so that you can entertain your alternative hypothesis).

Of course if you are using a different methodology, a qualitative one for instance, then a hypothesis might not be necessary (or may even be completely against the principles of that particularly methodology).

I think it is fine to discuss what seems to be the emerging pattern from your data, and to propose a hypothesis on the basis of your findings - that would seem to me to be an important aspect of making a contribution to knowledge. I think too often people get hung up on wanting to 'prove' something and to get the expected results, rather than see what actually comes out of a study. I think, like you said, the problems arise when people try to retrofit a hypothesis as if the results were expected all along.

Yeh, I've come to the conclusion that I will do it if it as "posthoc analyses" is going to add to the study in a meaningful way (it may be just as well to say in the discussion that I noticed something anecdotally - and save myself lots of time in coding and analysing the data just to make a point).
Cheers Chickpea