+ Reply to Thread
Results 1 to 3 of 3

Thread: Fisher's exact test - relationship to OR CI, RR CI

  1. #1
    Points: 89, Level: 1
    Level completed: 78%, Points required for next Level: 11

    Posts
    2
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Fisher's exact test - relationship to OR CI, RR CI




    Hello,

    I'm very amateur in statistics, and I'd be thankful for helping me with these questions. We're studying a gene, whose alleles (A, B, C) can carry a distinct modification. Each of our specimen can be classified as one of the alleles, and either carries the modification, or not. We would like to know, whether some alleles are more/less likely to carry the modification than the others. To determine this, we are comparing each possible pair of alleles (A and B, A and C, B and C) with Fisher's exact test.

    I was told that along with Fisher's, I should also determine OR and RR, and to discard the results, where (95%) CI includes number 1, even though the p is significant.
    Q1. should I discard the results where OR CI contains 1, or where RR CI contains 1, or in both cases?
    Q2. is calculating OR and RR in this case necessary/somehow useful? I'm only interested to know whether likelihood of modification differs between alleles (which I'll learn from p), and which one is more likely to be modified (that I can see from input data)...

    In reality, we are working with more than 3 alleles (its something around 10), and I'd like to do some correction for multiple comparisons.
    Q3.I was performing Bonferroni correction, but recently I read, that it may not be suitable for everyone-with-everyone comparisons within a group, what's exactly our case. Is that true?
    Q4. Also I read that when dealing with high number of comparisons, Bonferroni becomes too strict. Is it possible to overcome it? (e.g. if I do 100 tests, is it possible to Bonferroni-correct my p-values/significance level with numer 20 only, not 100?)

    Thank you for replying any of these!

  2. #2
    Pirate
    Points: 15,159, Level: 79
    Level completed: 62%, Points required for next Level: 191
    victorxstc's Avatar
    Posts
    875
    Thanks
    229
    Thanked 332 Times in 297 Posts

    Re: Fisher's exact test - relationship to OR CI, RR CI

    1. Why discarding them? It indicates that they are non-significant but you can keep them too. Sometimes non-significant is as valuable.
    2. Calculating OR (and RR) is almost always good and in many cases, necessitated by international report guidelines. At least reporting it has no harm, but improves your results.
    3. No Bonferroni can be suitable for such a comparison (all the possible pairwise comparisons), if the number of pairwise comparisons is not too high.
    4. Yes there are modifications to Bonferroni method which are better when the number of pairwise comparisons gets too high. Bonferroni-Holm correction is one of them and a very good method too.
    "victor is the reviewer from hell" -Jake
    "victor is a machine! a publication machine!" -Vinux

  3. #3
    Points: 89, Level: 1
    Level completed: 78%, Points required for next Level: 11

    Posts
    2
    Thanks
    0
    Thanked 0 Times in 0 Posts

    Re: Fisher's exact test - relationship to OR CI, RR CI


    Thank you!

    I'd also like to know 2 more things if possible
    to Q1. - I think I did not express myself properly - if the Fisher's p comes out as significant, but then either OR CI or RR CI contains 1, does it mean that consequently, result of Fisher's is considered non-significant?
    Q3. when performing all possible pairwise comparisons - whats the best way to correct for multiple comparisons then?

+ Reply to Thread

           




Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts






Advertise on Talk Stats