BookmarkSubscribe
Choose Language Hide Translation Bar
amm9010
Community Trekker

Should I use Bonferroni correction when making two linear contrasts with Fit Model?

I'm conducting a series of univariate REML models and have two preplanned orthogonal contrasts. I am wondering if I need to correct the acceptable alpha to guard against Type I error, or if this correction is taken into consideration by the JMP program when I add a second contrast?

0 Kudos
2 ACCEPTED SOLUTIONS

Accepted Solutions
jiancao
Staff

Re: Should I use Bonferroni correction when making two linear contrasts with Fit Model?

JMP doesn't automatically make the adjustment when you use Custom Test from the Estimates menu to add additional contrasts. You may apply the Bonferroni correction. (As a note, when there are a large number of comparisons, the Bonferroni correction can be too conservative.)

0 Kudos
dr_winfried_koc
Community Trekker

Re: Should I use Bonferroni correction when making two linear contrasts with Fit Model?

A slightly less conservative option is the Bonferroni-Holm procedure:  the smaller p value needs to be tested on the alpha/2 level first, the other p value can then be tested on your planned  alpha level if the first test was statistically significant.

A further option for dealing with the multiple testing issue is the false discovery rate approach (FDR-approach) for which a nice and easy usable JMP Add-In is available.

0 Kudos
4 REPLIES 4
jiancao
Staff

Re: Should I use Bonferroni correction when making two linear contrasts with Fit Model?

JMP doesn't automatically make the adjustment when you use Custom Test from the Estimates menu to add additional contrasts. You may apply the Bonferroni correction. (As a note, when there are a large number of comparisons, the Bonferroni correction can be too conservative.)

0 Kudos
dr_winfried_koc
Community Trekker

Re: Should I use Bonferroni correction when making two linear contrasts with Fit Model?

A slightly less conservative option is the Bonferroni-Holm procedure:  the smaller p value needs to be tested on the alpha/2 level first, the other p value can then be tested on your planned  alpha level if the first test was statistically significant.

A further option for dealing with the multiple testing issue is the false discovery rate approach (FDR-approach) for which a nice and easy usable JMP Add-In is available.

0 Kudos
jiancao
Staff

Re: Should I use Bonferroni correction when making two linear contrasts with Fit Model?

Thanks for mentioning the FDR Pvalue add-in. It has been incorporated into JMP known as the Response Screening under Modeling.

0 Kudos
amm9010
Community Trekker

Re: Should I use Bonferroni correction when making two linear contrasts with Fit Model?

Thanks so much for the suggestions, that all definitely helps!

0 Kudos