RE: OMEGA HAS A NONZERO BLOCK

From:"Serge Guzy" Subject:RE: [NMusers] OMEGA HAS A NONZERO BLOCK Date:Mon, 7 Oct 2002 09:30:00 -0700 I do agree with Ken that there is no science in the decision of fixing the correlation to let say 0.5. I will add that fixing a correlation to zero or 1 is more scientific as it is related to the notion of independence or linear dependence(often in the log domain)between parameters. Based on our individual background, we make the decision of what seems to be more or less scientific and Ken has a good point about fixing Ka to a unique value when we know that Ka should be different for each individual. If you think in terms of simulation, fixing Ka or forcing 100% correlation between 2 parameters when the true population dispersion is very very small(Ka) or the true correlation between the two parameters is very high(~100%correlation) is the same. Of course I am interested to know to what extent my assumptions will affect my prediction power. Did somebody try to bootstrap the observed data as well as the intraindividual noise to see if the "illness" was still present with correlation ~1. In my simulations, when I would get a correlation near to 1 when in fact the true correlation was zero, the Bootstrap approach would give me correlation ranging from zero to 1 while a true correlation near to 1 would stay near to 1 for all bootstrap samples.