Shrink%

Questions and techniques relevant to the modeling process.
Post Reply
jkuti
Posts: 9
Joined: Tue Aug 26, 2014 1:55 pm

Shrink%

Post by jkuti » Thu Jul 21, 2016 11:34 am

After convergence of a Pmetrics run, the html file gives population parameter estimates including mean, SD, median, CV and a term called "shrink%". To be honest, I have never paid much attention to it because it is never listed in the old NPAG out files, but sometimes it is a small number and sometimes it is zero. What does it refer to and how should it be interpreted?

Thanks much...

JK

User avatar
mneely
Site Admin
Posts: 209
Joined: Tue Nov 08, 2011 11:08 am
Contact:

Re: Shrink%

Post by mneely » Mon Jul 25, 2016 6:28 am

It's a new statistic for us that we are exploring. It is calculated as the % of total variance in each model parameter probability distribution that is due to variance in the posterior parameter probability distribution, i.e., mean(var(EBD)) / popVar , where EBD is the empirical Bayesian distribution or posterior distribution.

The idea is that as data are less informative about individuals, e.g. very sparse, non-optimal sampling, then posterior variance increases because we know less about each subject.

As I said, it's new, and we are still exploring the ramifications and whether it truly corresponds to the concept of shrinkage in the parametric world, see Savic, R. M. & Karlsson, M. O. Importance of Shrinkage in Empirical Bayes Estimates for Diagnostics: Problems and Solutions. The AAPS Journal 11, 558–569 (2009).

Post Reply