Goodness of Fit test for xtgee

[Dapeng Cheng]

Hi Stata experts,

I am using -xtgee- to run models. I have a base model that only includes control variables and the second model includes control and independent variables. Now I want to compare the goodness of fit for these two models. I tried to use AIC or BIC with the command -estat ic- but it is not supported. I know GEE is based on quasilikelihood theory rather than maximum likelihood theory.

I read some papers that used GEE in my field. I found that they reported -2 log-likelihood value and delta (a sign of triangle) fit from the base model (χ2 ). I searched -2 log-likelihood and some people say it is deviance. My questions are:

1. Is -2 log-likelihood value the same with deviance reported in GEE output?
2. What command can I use to compare the goodness of fit for two gee models?

[Marcos Almeida]

GEE models use quasilikelihood instead of ml, hence we can’t have AIC. But there is an user-written program named qic which provides its equivalent for GEE. That said, hopefully it applies for xtgee as well.

[Dung Le]

Hi Stata experts,

I am using -xtgee- to run models. I have a base model that only includes control variables and the second model includes control and independent variables. Now I want to compare the goodness of fit for these two models. I tried to use AIC or BIC with the command -estat ic- but it is not supported. I know GEE is based on quasilikelihood theory rather than maximum likelihood theory.

I read some papers that used GEE in my field. I found that they reported -2 log-likelihood value and delta (a sign of triangle) fit from the base model (χ2 ). I searched -2 log-likelihood and some people say it is deviance. My questions are:

1. Is -2 log-likelihood value the same with deviance reported in GEE output?
2. What command can I use to compare the goodness of fit for two gee models?

Another way to test goodness-of-fit is to compute root mean square error and mean absolute prediction error.