Set bootstrap confidence intervals in MGA for two tailed test and bonferroni correction

[ElNicolosi]

Hello,

I was wondering if any of you could help me with an issue to which I cannot find any answers online and/or in books.

I need to carry out posthoc multiple comparisons between three groups after an OTG. However, I cannot use the permutation test as my sample sizes are quite different (respectively 45-108-99). So I am opting for PLS-MGA non-parametric test. Yet, this test is a one-tailed test and I actually have no directional hypothesis. So I need to carry out a two-tailed test at 5% confidence. This mean 2.5% right and 2.5% left.

My question is how should I set up the p value in the bootstrap table of the MGA two get the corresponding two-tailed test? Since, with three groups I have a total of three comparisons to make (A-B, A-C, BC) I thought to apply the Bonferroni correction (0.05/3=0.016). However, provided that the MGA is a one-tailed test should I first divide 0.05/2 and then by the number of comparisons? So 0.05/2= 0.025/3=0.0083?

So, am I doing correctly when I set in the bootstrap table for the MGA selecting the option "two-tailed bootstrap" and "p-value" 0.0083?

Any advices would be much appreciated.

Thank you in advance for taking the time to read this!
Ele

[jmbecker]

The Bonferroni correction should be the same regardless of whether you take 5% devided by 3 and then devide by 2 (for both ends) or devide 5% by 2 (for both ends) and then devide by 3. In any case it should be 0.008333333 for the lower end and 99.99166667 for the upper end.

The PLS-MGA output is not affected by the "two-tailed bootstrap" and "p-value" settings. They only matter for confidence intervals and highlighting of results.