

Logit intercepts for MARKOVS 

Message/Author 

Johnny Wu posted on Thursday, February 22, 2007  10:38 am



Hello, If I want c#2 to be my stayer class, do I do this: %c#2% c2#1 ON c1#1@20; OR this: %c#2% c2#1 ON c1#1@20; (I thought that more negative logit thresholds = higher probability, such that a 20 logit threshold transition from class 1 time 1 to class 1 time 2 would indicate that the probability of transitioning from class 1 time 1 to class 2 time 2 is ONE)? thank you 

Johnny Wu posted on Thursday, February 22, 2007  8:21 pm



Sorry I meant "a 20 logit threshold transition from class 1 time 1 to class 1 time 2 would indicate that the probability of going from CLASS 1 TIME 1 to CLASS 1 TIME 2 is ONE)? 


The c2# on c1# language refers to logistic regression terms described in chapter 13 of the UG. These are in logit scale, not threshold scale, so signs should not be reversed. +20 is a high logit and therefore a high probability. The threshold parameterization is only for observed categorical variables in order to cover ordinal variables (thresholds are ordered). 

Back to top 

