Monday, September 8, 2025

Parametric bootsrap of a mixed model

 Lately I have been tackling a problem about estimating precision and trueness of dental measurements. I know that a mixed effects model is a best way to reveal these estimates, so I go that way. 

I built some models with lm4::lmer and nlme::lme. Since lmer doesn't enable heterogenity in variances, I mainly focused on lme. I hadn't thought other packages like glmmTMB or similar although they might have best options. Since lme couldn't calculate 95% CIs, I wrote a simple script doing first nonparametric bootstrap, then parametric bootstrap. 

 I extracted variance components from the main by simulation, and calculated CIs for precision and trueness. Mixed models did a big job. 

But, after all those serious work for simulations, I discovered that parameters package have this ability builtin its functions like simulate_model and model_parameters. 

model_paramaters  didn't work smoothly but simulate_model did the same job what my code of 100+ lines and many hours of work did immediately. I was surprised by the magical ease those functions supplied. 

 

No comments:

Post a Comment