Should I report significant correlations of variables, that are not part of my hypotheses?

[mcfilu]

Hi,

I correlated all my study variables, and some of the demographic variables, with each other, to see if there were any significant associations. I found significant correlations between some study variables with demographic variables. For example, let's say I analyzed which type of candy participants like eating the most; and the amount of candy XY eaten correlated with participant's educational status. This would seem like a "spurious" association, as in there would be no obvious explanation why participant's education should be associated to how much of candy XY they eat. My questions:

1) Is it common to to this sort of preliminary correlational analyses to explore associations between variables?

2) Should I report significant correlations, even if they are not part of my study questions/hypotheses?

3) If yes, should I mention these significant correlations as well in my discussion? Or can I simply report them in my results part, and them not mention them anymore in the discussion?

Thank you in advance !

Edited November 25, 2019 by mcfilu

[SciOfInnovation]

Yes! These correlations probably need to be included. Not reporting data that doesn't fit your hypothesis is a failure of rigorous research methodology and an ethics issue.

Depending on the significance of these correlations your study may require significant redesign or at least analysis of possible study shortcomings. You should probably do some analysis of variance between different demographics to determine if simple random sampling is even appropriate, or if you need some form of sample stratification to achieve a reasonable statistical confidence interval. (Since you did not mention your data collection method, I am assuming you use random sampling and could not measure every member of the population.) Also, make sure to plot the data to visually see if it looks like a straight line relationship between the demographic and the variable. Correlation only captures linear relationships, and not more complex relationships like the bell curve of human IQ.

1) This sort of preliminary correlational analysis is very common and a good way to begin validating data quality.

2) Yes, even if there is truly no significance this is part of your study design and data collection method. It shows what your data sample looks like.

3) This belongs in the discussion/conclusion, and possibly not results depending on the level of detail you want to get into.

Ask your supervisor what they think, they can probably give you better guidance on how to handle this difficult question. You didn't give a lot of details about your research, so this answer could be way off base.

[taacc]

On top of the suggestions made above my comment, I would clarify that including findings that were not in your original hypotheses is always encouraged, as long as you're clear about what they are. It only becomes a problem to include them if you change your hypotheses to fit to these significant findings you weren't expecting. I would just say that while you did not hypothesize that these correlations would be significant ahead of time (apriori), you found them after doing preliminary correlational analyses/you decided to do them posthoc, and then you can go on to explain what they might mean to your overall results and discussion (to the appropriate extent of their relevance).