ANOVA with Holm Correction

[Emily Janio]

Hi all,

I’ve been asked to run an ANOVA with a Holm correction. I need to compare Stigma (continuous outcome) across five different survey versions (categorical independent variable).

I understand that in this case, I can employ the -oneway- command:
oneway Stigma Survey
And then use an option like -bonferroni- to do a correction.
But it seems like in Stata, in order to get the Holm correction, I need to employ the -anova- command:
anova Stigma Survey
And then I need to use the -test- command with the -mtest(holm)- option. Here is where I get a bit confused. It seems like I need to make a matrix to refer to when I use -test-. I just don’t understand how to make this matrix. I would like to compare Stigma across the five survey versions. How can I do this with a Holm correction?

[Joseph Coveney]

I just don’t understand how to make this matrix. I would like to compare Stigma across the five survey versions. How can I do this with a Holm correction?

You can do the following in order to see how to construct the contrast matrix.

anova Stigma Survey

testparm i.Survey // Optional:  in order to see which individual tests make up the omnibus test
test , showorder // Optional:  in order to see the design matrix
matrix input T = (0 1 0 0 0 0 \ 0 0 1 0 0 0 \ 0 0 0 1 0 0 \ 0 0 0 0 1 0)
test , test(T) mtest(noadjust) // Optional:  in order to confirm that your contrast matrix is what you want

test , test(T) mtest(holm) // <- This

// An alternative syntax that doesn't require the explicit matrix (it uses the general syntax for -test-):
test 2.rep78 3.rep78 4.rep78 5.rep78, mtest(holm)

[Joseph Coveney]

I wrote

test 2.rep78 3.rep78 4.rep78 5.rep78, mtest(holm)

Make that

test 2.Survey 3.Survey 4.Survey 5.Survey, mtest(holm)

(or however your category numbering is) for your dataset.

I had confirmed that alternative syntax using the auto dataset that ships with Stata. And unintentionally neglected the mutatis mutandis when copying-and-pasting the command line into the post.