Forum Replies Created
>fit=mixed(log(A)~B+C+(1+B|id)+(1+B|item), method=”S”, data=dat2, progress=TRUE, cl=cl)
here i would like to get regression estimates and random effect variance on the original values, not on log-transformed
another problem I’m encountering is that when I try to use emmeans emmip or summary -functions it fails when the model includes the expand_re=TRUE specification:
>fit=mixed(log(A)~B+C+(1+B||id)+(1+B||item), method=”S”, data=dat2, progress=TRUE, expand_re=TRUE, cl=cl)
>Grid=ref_grid(fit, at=list(C=c(-2, -1, 0, 1, 2)))
>EMM=emmeans(Grid, specs=c(“C”, “B”))
>emmip(EMM, type=”response”, B~C, CIs=TRUE)
Error in eval(predvars, data, env) : object ‘re1.B1’ not found
One thing more:
Is there a way to back-transform the values returned by summary-function for a mixed object?
Many thanks I was now being able to plot figures:
>R=ref_grid(fit, type=”response”, at = list(a = c(-2, -1, 0, 1, 2)))
>A2=emmeans(R, specs=c(“a”,”b”, “c”))
>emmip(A2, b~a|c, CIs=T)
And also confidence intervals for a result table, e.g:
Here a=continuous predictor, b, and c categorical predictors.
Its a pity if no variance explained or equivalent value cannot be computed, that would be highly appreciated by researchers.
Of course we get sense of how strong the predictors are by looking at the figures.
Ok, so the emmeans seems to be the way to go as I have a rather complex models to report, and in some models I have factors with three levels (=multiple parameters?).
Thanks, I opened an issue also on sjPlot -GIT page.
I believe I need to few other values, at least confidence intervals and variance explained for the fixed and random parts of the model.
What would be best way to get these?
but at least the confint seem to take a lot of computation time at least if continuous predictors are involved.