04-15-2018 02:41 AM
Hello, I want to ask some questions regarding Tweedie Distribution in SAS:
1. Can I assess goodness of fit (using Kolmogorov-Smirnov or Anderson Darling) for Tweedie?
2. I use TECHNIQUE=CONGRA, NMSIMP, and QUANEW for Tweedie (PROC HPGENSELECT is used), but the dispersion parameter has significant difference (CONGRA > 44, NMSIMP > 1500, QUANEW > 16000) while the power parameter is around 1.6 for all technique used, is there anyone who knows why it is like this?
Thank you in advance!
04-16-2018 01:44 PM
1. You can assess goodness of fit by using the AIC, AICC, and BIC criteria. The regression procedures use information criteria because it is generally applicable to regression models, whereas the K_S test is a univariate test. You can also look at the scaled Pearson chi-square, which is close to 1 in models that fit well.
2. Please post your log. I suspect that some of the optimizations are not converging.