need help choosing the right model

WeeG

TS Contributor
#1
following another thread I opened, I need your advice to choose the correct logistic model. I had 8 IV's, I ran 8 univariate models (used the method of choosing the right model suggested but Hosmer and Lemeshow). Then I took all significant IV's and put them in 1 model. None were significant ! I started reducing the most none significant ones, until I got a model with 1 IV which is significant and another one with P.Value of 0.07. This is model #1. Then I took off the second IV and stayed with a univariate model...this is model #2.
My problem is, that some tests I did after modelling, give unclear choice, I'll be more specific:

Model #1:
beta1=0.735 with a SE of 0.35 and P.Value of 0.036
beta 2=0.871 with a SE of 0.486 and P.Value of 0.073
Hosmer and Lemeshow statistic=11.82 with P.V=0.159
Chi Sqaure statistic=60.69 with P.V=0.3108
log likelihood=-35.63 AIC=77.262 BIC=83.495
Area under the ROC curve = 0.739

Model #2:
beta=0.83 with a SE of 0.344 and P.Value of 0.016
Hosmer and Lemeshow statistic=5.59 with P.V=0.693
Chi Sqaure statistic=30.29 with P.V=0.694
log likelihood=-37.37 AIC=78.75 BIC=82.90
Area under the ROC curve = 0.6771

According to Hosmer and Lemeshow, the second model is better (right ? I am not sure). According to the BIC, also model 2 is better. According to AIC and the area under the ROC curve, the first model is better.

What do you think ?

(there is no BIG correlation between the 2 IV's)