I am performing a mediation analysis with mixed models in the presence of a (potential) treatment * mediator interaction. I would like to test if this interaction is significant. Is there anything wrong with using the same logic in this code to preform this test? My model.m and model.y that I provide to the mediate function are both lmer objects, and test.TMint says that it is not implemented for these models quite yet.
So my question is, is there anything theoretically wrong with this:
mod.m <- lmer(M ~ X + (1|id), data = data)
mod.y <- lmer(Y ~ M * X + (1|id), data = data)
https://github.com/kosukeimai/mediation/blob/a9a5c7b747b1d96d9a6df7b7cfe3d7719a1b0780/R/medtests.R#L78-L84
Hello,
I am performing a mediation analysis with mixed models in the presence of a (potential) treatment * mediator interaction. I would like to test if this interaction is significant. Is there anything wrong with using the same logic in this code to preform this test? My model.m and model.y that I provide to the mediate function are both lmer objects, and test.TMint says that it is not implemented for these models quite yet.
So my question is, is there anything theoretically wrong with this:
mod.m <- lmer(M ~ X + (1|id), data = data) mod.y <- lmer(Y ~ M * X + (1|id), data = data)
med.results <- mediate(mod.m, mod.y, mediator = "M", treat = "X" )
acme_diff = med.results$d1.sims - med.results$d0.sims
quantile(acme_diff, probs = c(0.025, 0.975))