I would use the number of unique comparisons. So if you had 3 variables you would have 3 unique combinations of comparisons (eg 4 = 6). Just depends on how bad a type I error would be to your question.
Hello all,
When constructing correlation matrices, a number of people claim that one should apply the Bonferroni adjustment if the number of variables is high. I did some background research on that, but couldn't find a source which actually states as per when (i.e. how many variables make up 'many') one should use that. I am also confused if this counts only for listwise correlations or also for pairwise?
I would use the number of unique comparisons. So if you had 3 variables you would have 3 unique combinations of comparisons (eg 4 = 6). Just depends on how bad a type I error would be to your question.
Allright, thank you for responding to my question and your opinion. Still, I am confused as per when (is there a cut-off line/threshold) one better applied Bonferroni adjustments?
I am also happy about a lit reference...I just want ot understand that...
Thanks!!
Well after looking at my #2 post, you could probably down grade the correction. Since 1 is only getting compared twice, same with 2 and 3 (see below). So you could multiply the p-values by 2 or equivalently divide alpha by 2. I have also seen in papers that you should correct based on the total number of comparisons, that is typically what I do, which is much more conservative.
Example:
1,2
1,3
2,3
As for a reference, all you have to due is type "Bonferroni Correction" into Google Scholar and you get many papers on the topic.
Thank you, I got your point now and will do some more lit research on it.
Tweet |