CecileProust-Lima
commented
2 months ago Dear Ross, sorry for the delay.
ExternVar can be used when your outcome is ordinal. It's just that you should use the method "conditional" and not the "twoStageJoint". Here is an example :
library(lcmm) paquid$age65 <- paquid$age - 65
estimation of the first model (ordinal mixed model)
mthresholds <- lcmm(HIER ~ age65male, random=~ 1, subject='ID', data=paquid, link='thresholds') mthresholds2 <- lcmm(HIER ~ age65male, random=~ 1, subject='ID', mixture=~ age65,ng=2,data=paquid, link='thresholds',B=random(thresholds))
estimation of an external model (linear mixed model in this case)
ext <- externVar(mthresholds2,fixed= IST ~ age65, random=~age65, data=paquid,method="conditional",M = 100) summary(ext)
I hope this helps, best Cécile
rossneville
VivianePhilipps
Dear Prof. Proust-Lima
I am very much enjoying using the lcmm package and I have found it very user-friendly (largely due to the worked out vignettes on the lcmm webpage).
I have successfully used lcmm to estimate a latent class model. Given that I’m estimating latent classes from a 4-level ordinal outcome, I employed the thresholds link function.
The lcmm model works well, producing two distinct classes (of children screen time across ages 3, 5, 7, and 9).
I was hoping to use the externVar command to estimate the degree to which class membership (1. high initial screen time steadily declining and 2. high stable screen time) predicts an external outcome at age 9 (reading and literacy).
As far as I can tell, the externVar command does not work for latent class models estimated with ordinal outcome variables (i.e., where link=“thresholds”). Is this correct / still the case?
If I am mistaken and there is a vignette where external is used in the context of an ordinal outcome, can you please point me in the right direction .
If I am right and there is no way of estimating an external outcome using externVar where the lcmm model has the thresholds link function, can you share your thoughts on how such an analysis could be accomplished.
I can extract the predicted class membership for each observation in the dataset and use that as a categorical predictor in a typical regression model. That works fine. However, this does not take into consideration that class membership is measured with error…hence why you came up with externVar.
I would most appreciate any guidance / support / additional resources that you can provide.
Thank you for your consideration of my (long winded) question.
Regards Ross