I want to study which factors have an effect on the hormone PTH. PTH is used as a dichotome variable and is our outcome variable, the dependent variable. We're using generalized linear mixed models in SPSS. If we study the effect of one factor on PTH, we obtain a different p-value than when we use multiple factors at a time. Is there a correction for interactions we can use?
1
-
1$\begingroup$ How come a hormone is a binary variable? What did you measure / observe in your experiment? $\endgroup$– dipetkovCommented Sep 20, 2023 at 16:53
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
It seems to me you are asking about two different things. It's normal for a predictor's p-value to change when other variables are included as predictors, as now the coefficient of the first predictor is adjusted for these additional predictors. The best thing to do if the results change significantly is to find out why this is. E.g. it may be that the relationship between the original predictor and dependent is actually spurious and driven by the other predictor(s), other predictors may be suppressors, or the like.
P-value correction usually refers to correcting your critical p-value as lower when conducting several tests within a single dataset. I think it's not typical to correct p-values in the context of a single multiple regression (i.e. to consider each predictor as one test) so you don't necessarily need to do any adjustments when considering your multiple regression output, whether you have interactions or not.
However, as it seems you have factorial predictors, it's customary to adjust p-values when conducting post-hoc comparisons between factorial levels (Tukey's correction is a common one, and it's available in SPSS).
