I’m new to statistical modelling, and I’ve managed to pick quite a complicated first problem.
Essentially, I have a bivariate linear mixed effects model, where I am trying to model an imaging parameter I (over patient age) against a neurological outcome N (over patient time since diagnosis), where there are shared covariates between I and N. I also have random effects for the patients, specifically for y-intercept and slope. The model converges, everything looks good, but what I am now trying to do is estimate how well the model will perform on future data.
I’ve been using cluster bootstrapping in R to calculate the optimism-corrected R2 values, as well as calibration curves. I think my main question is: Should I only do optimism-correction for marginal (fixed effects) R2 and calibration slope? Or, should I include optimism-correction for conditional (fixed+random effects) R2 and calibration slope? I’m asking because I’ve found that it is relatively straightforward to set up optimism-correction for marginal R2 and calibration slope, but I’ve not been able to sort it for conditional R2 and calibration slope.
Thank you, and apologies for my ignorance on this!
E
