Solved
Contributor
Posts: 70

# Logistic Regression

Hello

1) How can we say new model is better?

2)  If SC or AIC criteria under the intercept column would have been increased in the 2nd model then what would be the conclusion?

3) If probability for LR and Score is >.05 and therefore insignificant but the probability of Wald is <.05 and thus significant. In such a case we reject or accept the null hypothesis?

Thanks

Accepted Solutions
Solution
‎11-23-2016 02:48 AM
Super User
Posts: 10,213

## Re: Logistic Regression

1) there are many model goodness-fit statistic Like :
Pearson chisquare , H-L test, AUC (the area under ROC curve)

2)That says covariates can explain more variance of data. That model would be better.

3)
"probability for LR and Score is >.05 " is test for the whole model.
If the whole model is insignificant, so the parameter significant doesn't mean anything.
i.e. model is not good.

Therefore, It should accept H0: beta=0 .

All Replies
Solution
‎11-23-2016 02:48 AM
Super User
Posts: 10,213

## Re: Logistic Regression

1) there are many model goodness-fit statistic Like :
Pearson chisquare , H-L test, AUC (the area under ROC curve)

2)That says covariates can explain more variance of data. That model would be better.

3)
"probability for LR and Score is >.05 " is test for the whole model.
If the whole model is insignificant, so the parameter significant doesn't mean anything.
i.e. model is not good.

Therefore, It should accept H0: beta=0 .

Contributor
Posts: 70

Thanks a lot.

Contributor
Posts: 70

## Re: Logistic Regression

When do we prefer Likelyhood ratio over Wald and Score?.

Any Example?.

Super User
Posts: 10,213

## Re: Logistic Regression

2) covariables can explain more variability of data (the difference between intercept and covariables is bigger).
the latter one is a better model.

"When do we prefer Likelyhood ratio over Wald and Score?."
I don't know .

Likelyhood ratio is usually for Contingency Table.Check PROC CATMOD which can also do Logistic Regression.
☑ This topic is solved.