5
$\begingroup$

I'm trying to submit a chapter of my thesis to a journal. We used nested ANOVAs where we nested the group comparisons within year to account for potential bias due to different observers between the two years. There were stark contrasts between year although that is not of interest. My adviser says to leave all significant findings regarding year, out of the paper, because it's not relevant to our overall "story". I agree that it is not relevant, but I wanted to see what the consensus was on this. I know often that analyses that are run are left out of papers, but part of the results of a simple ANOVA model? Do I owe it to the readers to be clear about evidence of observer bias or can I just leave it out because it isn't important to the message of the paper.

Improve this question
$\endgroup$

2 Answers 2

Reset to default
3
$\begingroup$

I would like to see the entire model included somewhere in the paper, probably in the methods section or in a supplement/appendix. If it the model is huge, a table is a nice way to present this. You could even organize the table to distinguish between "interesting" variables (e.g., drug dose) and boring covariates that are likely to be trivially true (e.g., rater). I saw this done recently and thought it was very convenient, but I'm having a hard time tracking the paper down.

However, once you've "disclosed" your model, there's no need to give equal time to each of its components when discussing the results. You could ignore the covariates completely, or you could start the results section with something like

After controlling for observer bias (see complete model in Appendix 1), we found that....

If the bias is particularly surprising or you have done something to control it (e.g., provide raters with a standardized rubric), then it might be worth a slightly longer mention.

Improve this answer
$\endgroup$
4
$\begingroup$

If the bulk of your paper is taken up with discussion of the ANOVA results it would seem odd to omit to mention controlling for observer bias; & odd to want to, as your having taken it into account forestalls potential criticism for not having done so. If you're giving effect estimates and standard errors you need to specify the model they're embedded in if you want anyone else to be able to use them. If, on the other hand, you merely mention in passing some conclusions drawn from the ANOVA, it doesn't seem necessary to go into too much detail about it.

Improve this answer
$\endgroup$
2
  • $\begingroup$ Thanks for the answer. To be clear, I mention the model in the methods, my adviser just wants to take out what she views as superfluous material in the results section. $\endgroup$
    – Mina
    Commented Feb 4, 2014 at 5:14
  • 1
    $\begingroup$ Well, what is the superfluous material? I think @Matt's answer covers the possibilities nicely. A results table is both conventional & handy, & there would be little sense here in not giving it fully if you give it - without necessarily expending a single word on discussion of an uninteresting effect. $\endgroup$
    – Scortchi ♦
    Commented Feb 4, 2014 at 9:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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