01-07-2016 12:39 PM
I have an nlmixed model with a continous outcome and I would like to:
a) construct a profile likelihood confidence interval for my estimated parameter(s) and
b) check how "well-behaved" the profile likelihood is by graphing the (converged) model's log likelihood against a range of fixed values of the parameter.
I'm used to coding in Stata and can't manage this in SAS (beyond doing it manually by repeatedly fixing the value of the parameter in question and running the model until I identify the upper and lower limit values that give me a log likelihood that is 1.92 lower than in my actual model etc. etc.) I need to do this for many CIs/models so I'd be very grateful for any assistance. I have The Little SAS Book.
For b) I'd like to put the relevant parts of my model in a DO loop that iterates through different fixed values of the parameter (I think I've got that far) and for each run of the DO loop I want to extract into a new SAS dataset the value of the converged model's log likelihood (i.e. the final value that appears in the SAS "IterHistory" table) and its associated parameter value. Then I can do the graph.
For a) I think the pragmatic approach will probably be to identify the relevant values from the new SAS dataset. Or could the profile likelihood CIs be found directly without the need for complex coding? The only examples I have found so far for nlmixed/profile likelihood CIs have been for binomial outcomes and use coding with macros that I can't follow.
Very grateful for any assistance.