 |
 |
All, Looking for a little help with a mixture model problem. I have some dose versus time data that for some events look like a sigmoidal growth model and for other events look asymptotic with no inflection point. I have attempted to do something like Gammerman and Smith's 1996 model of the Elston and Grizzle ramus data. I choose my mixture component indicating variable for each model from a bernouilli distribution and then add one to either the zero or one from the bernouilli. This mixture component indicating variable then chooses which model to use for the mean of my normal distribution. When I run this model, I receive undefined real result error messages. After troubleshooting, it looks like I get the error when the mixture component indicating variable =0. Has anyone else run across this? Note that I am deliberately not forcing reconciliation between two means, but rather, choosing one or the other. Thanks, John Neal %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%