Dear Allstat,
1. Which is the best test to assess the goodhess of fit
for the above?
2. When i tried to fit a logistic regression in Stata with 9 predictors and
650 patients, the Pearson godness of fit test gave a p-value of 0.89. When i
added an extra term which was levels of creatinine ( a continuous variable),
the pearson gave a relatively significant value of 0.07. Both of these
p-values relate to the reduced models where we have purely significant
factors in the equation. Is this due to interaction between creatinine and
the other variables?
3. Does the Hosmer-Lemeshow test not work well with many subjects? For when
i tried in spss 10 ( it provides the previous test for the goodness of fit),
the model without creatinine was significant before it was reduced ( when it
became non-significant) and the model with Creatinine was not significant
both before being reduced and afterwards. The initial models here giving
reverse goodness of fit p-values compared to the pearson p-values in the
stata version.
Thank you fr your time.
Best wishes
Ian
|