07-22-2016 12:05 PM
I wrote the following code to fit a NHPP on my data. The problem is while it gives me the parameters and their relative errors, it does not provide a fitting error, like Anderson Darling. I want to see how far my fitted model is from the general data (e.g. mean square error). I want to know which model, power-law or log-linear, provides a better fit. I need your help, ASAP. Thank you!
proc reliability data=app1;
mcfplot time*day(-1) / fit=model noconf;
07-22-2016 03:34 PM
This is kind of a kludgy way around it, and I am not sure that it will work. If you use the parameter estimates to calculate predicted values at each time point, you could get the residuals, and from there calculate PRESS values as a way of comparing. Another would be to look at Kolmogorov-Smirnov tests comparing the observed and predicted values.