Multiplicity correction is typically not required unless you have hundreds of measurements, where p-values of 0.01-0.05 would indeed be likely to pop up by chance. If your reviewers/bosses still ask you to do this, put all your 12 p-values into p.adjust(.., method="BH") function in R, this will do Benjamini-Hochberg "false discovery rate" corection.

Thank you Dr Mikhail

Either Bonferroni or Benjamini correction, it does not differ for me.

You mean that I will multiply by 12 (number of genotypes in the 4 SNPs). I think this is too great number for the 4 SNPs.

Can I make chi square for each SNP (2 columns X 3 raws) and this will produce only one p value. This p value will be multiplied by number of SNPs?

Why I correct p value? although the SNPs are unrelated and the genetic distance is very large and each SNP on different chromosome?

Hi Iman

Do you have many confounders and covariates?  I'm not sure you need to correct for multiple testing in a case-control study considering only 4 SNPs and no interaction variables.  Bonferroni is also very strict so if you maintain significance for certain SNPs then great, but it might exclude a lot of relevant ones because of it's stringency. Consider false discovery rates (FDRs) instead.

Regards,

Nathaniel

Thank you Dr Nathaniel

I want to correct for multiple SNPs

Both Bonferroni or false discovery rates (FDRs), does not differ for me.

I want any type of correction for multiple SNPs

Dear Iman,

I appreciate your acceptance that when you do 4 analysis (4 SNPs), the probability for a false positive increases by 4 times and hence you consider 0.0125 (0.05/4) as p value threshold or you multiply your obtained p values by 4. I guess you are bit unclear with the number of p values. Ideally for each SNP you'll get one p value for genotypic association (3x2 table if co-dominant model is assumed) and one p value for allelic association (assuming additive model). So 4 for genotypes and 4 for alleles. You need not add the numbers from genotypes and alleles. So it is 4 p values for genotypic association analysis and 4 for allelic association analysis. Ideally you should adjust for multiple testing but still you need not ignore a significance that disappeared after multiple testing. You report the significance in the paper and discuss the possible mechanisms. However, it is good to  mention that significance did not withstand multiple testing.

Thank you too much Dr Veera

Your answer is very clear and logic

I want to know if all SNPs are far from each other (e.g. on different chromosome). Why should I correct p value? I study each SNP alone, and its effect alone. 

What I understand about correction for multiple comparisons, that each p uncorrected value is comparing each SNP against other 3 SNPs. Then I correct p values by multiplying by the number of SNPs.

Here I study each SNP effect alone (not versus the other 3 SNPs). Why should I correct for multiple comparisons?

@Iman there is no comparison between SNPs. You are just trying to find out if any of the SNPs is associated with disease. More the SNPs I have to do the analysis more I get false positives. That is why you have a threshold in range of 10^-8 to 10^-12 in genome wide association studies, where you screen millions of SNPs for association. All the best.

Dr Veera

In Bonferroni for ANOVA, we compare variables. Why when using it in SNPs, we do not compare SNPs? 

Dear Iman,

The comparison in ANOVA is between the categories of a variable and not between variables. Similarly the comparison is between the genotypes of a SNP and not between the SNPs. Please read the attached articles for more clarifications.

http://bib.oxfordjournals.org/content/3/2/146.abstract

http://www.nature.com/nprot/journal/v6/n2/full/nprot.2010.182.html

Thank you Prof. Dr. Veera

Dear all,

I am zalina, a phd student 

I am from Malaysia

I need your help

I am having problems in responding to comments about my manuscript especially regarding

"multiple testing should definitely be applied. 3 polymorphisms in different models including several diplotypes and haplotypes are analysed. The correction factor might be small, but so is the difference between the observed p-values and the significance level."

In my manuscript, I only studied one gene.

Correction for multiple testing was not performed since only one gene was tested and I followed previously published article by Belfer et al. 2013

1. Belfer I, Segall SK, Lariviere WR, et al. Pain modality- and sex-specific effects of COMT genetic functional variants. Pain. 2013;154(8):1368–76

 Please help me and guide me to answer the editor comment

 zalina

Would you please send me the table you are talking about