Can I "discard" p-values from my linear mixed model?

Question

I have a linear mixed effects model as follows in R:

model.int = lmer(value ~ group * category * time + (1|patient), data=modeldata)

anova(model.int) then gives me the following list of p-values:

1. group
2. category
3. time
4. group * category
5. group * time
6. category * time
7. group * category * time

As I use the mixed model on 5 sets of data, this results in altogether 5*7 = 35 p-values. Now we know that a higher number of p-values results in a higher percentage of type I errors, which is why we should correct for multiple comparisons.

But, among the 7 p-values, the only ones concerning my hypothesis are the two marked bold. The other p-values do not interest me. For example, I already know that time and group have a significant effect on value by study design, that is specifically the reason why I put it as a fixed effect in the linear mixed model!

My question is, can I just report the 2 p-values concerning my hypothesis in my paper and discard the rest? And if yes, do I correct for multiple comparisons only among the 2*5 = 10 p-values concerning my hypothesis or for all 35?

Answer 1

My question is, can I just report the 2 p-values concerning my hypothesis in my paper and discard the rest? And if yes, do I correct for multiple comparisons only among the 2*5 = 10 p-values concerning my hypothesis or for all 35?

Be careful. It could give the impression that you didn't include the other terms in your models, and doing that is a big mistake. I get that there might be a lot of output which is extraneous to your point, so perhaps include only the terms of interest in the body of your paper, but be sure to include a footnote referring to an appendix where you show the full output.