RE: OMEGA HAS A NONZERO BLOCK

From:"Kowalski, Ken" Subject:RE: [NMusers] OMEGA HAS A NONZERO BLOCK Date: Mon, 7 Oct 2002 14:57:44 -0400 Come on Lew, you would never question a domain expert's prior? I find that hard to believe. Moreover, as your message suggests we often wear different hats and sometimes assume the role of a domain expert anyway. If I was assuming the domain expert role and was going to specify a prior I would certainly want to provide the information to support my prior to give credence to its use...am I the only one that feels this way? In the absence of any 'credible' prior information, fixing the correlation to 1 is the best we can do on the basis of the MOF and the current data. I'm assuming that fixing the correlation to 0.5 is inconsistent with the current data in hand (i.e., a considerably higher MOF). Nevertheless, if there is credible extra information that the correlation is better represented by being fixed at 0.5, then that's fine. Serge's simulation results though seem to suggest that the correlation being estimated on the boundary value of 1 is more likely to occur when the true correlation is indeed high...presumably higher than 0.5 (Serge correct me if I'm overstating your results). Ken