Page 2 of 2
Re: Sinclair(2009) JMCB
Posted: Mon Apr 14, 2014 6:26 am
by AhmedSahlool
Hi,
Do you mean that I apply the PMETHOD=GENETIC in the individual models and then use the generated initials values in the bivariate model, or I apply it directly in the the bivariate model.
Thank you.
Re: Sinclair(2009) JMCB
Posted: Mon Apr 14, 2014 11:52 am
by TomDoan
AhmedSahlool wrote:Hi,
Do you mean that I apply the PMETHOD=GENETIC in the individual models and then use the generated initials values in the bivariate model, or I apply it directly in the the bivariate model.
Thank you.
No. If the individual models would give a good set of guess values for the bivariate model, that would be nice. Unfortunately, they don't because the correlations in the joint model are so high. You need to use PMETHOD=GENETIC on the bivariate model. GENETIC uses a randomization to set the initial values for the population and then does a systematic attempt to improve upon those. Given how ill-behaved the likelihood is, you probably will need to increase the population scale-up. See page UG-117 for a description.