I have some general questions about bootstrapping in NLME. 1) is the bootstrapping in NLME parametric or non-parametric? 2) if non-parametric, are the parameter estimate boundaries enforced during bootstrap? 3) if you have a well behaving model, the parameter CI’s from the bootstrap seem very small. does the bootstrap reinitialize for every run, or do the parameter estimates carry from one run to the next. If that is the case the entire parameter space is not really being resampled/reestimated, and the bootstrap may be stuck in a local minimum? 4) I would not have to bootstrap if I could get the standard errors to run, in the current version that seems impossible.
Dear Wolowitch 1) is the bootstrapping in NLME parametric or non-parametric? non parametric in the sense the individuals are resampled with replacement and then the fit is performed again if non-parametric, are the parameter estimate boundaries enforced during bootstrap? It should. Note that QRPEM in this version does not enforce boundaries but FOCE ELS should if you have a well behaving model, the parameter CI’s from the bootstrap seem very small. does the bootstrap reinitialize for every run, or do the parameter estimates carry from one run to the next. If that is the case the entire parameter space is not really being resampled/reestimated, and the bootstrap may be stuck in a local minimum? Here is what should be done. you first run your model without bootstrap. You copy the model to the workflow and accept all fixed and random. Then you shift to bootstrap. The bootstrap will start every run with the final estimates (the same) you got from your initial run. This is to prevent drifting. I would not have to bootstrap if I could get the standard errors to run, in the current version that seems impossible. You are talking I guess about the complex model you showed me a while ago. I believe that this model is not fully identifiable and had too many parameters. This is not the right example to demonstrate the bootstrap feature. I did not succeed to improve your model but we can talk next week about that if you have time. Best Regards; Serge