Interpreting logistic regression

[noetsi]

I got findings I have not seen before with this many cases and I wanted a second opinion I don't know how to move the attachments around for which I apologize. As you can see in the testing Global Null Hypothesis none of the models are statistically significant. None of the parameters are either (most are not close). My guess is this might be influenced by only have 83 of the least common cases and 7 predictors (there are 533 total cases). I don't know how to interpret the Hosmer and Lemeshow test results at all, I have never seen something like this. The DV has two levels, 1 is satisfied/highly satisfied and 0 is not satisfied/highly dissatisfied.

One thing I did which may not be correct is for the 8 levels of the categorical variable (geographic areas that make up a state) I created seven dummies and put the 7 of them in the model (leaving out a reference level - the 8th level of the categorical variables). Perhaps I should just put one of the predictors in instead given the small case size?

 

[noetsi]

I went with a theory that the issue was too many predictors. But even using one of the dummies they were not significant. The SE looks huge to me relative to the effect size.

 

[hlsmith]

Not sure why the H-L test has perfect fit?

Side note, how many cases per region, provide a table. Data shouldnt be too sparse, but table will help. Could it be that region just doesnt predict satisfaction?

 

[noetsi]

I was asking, in part, why the HL test was a perfect fit. I don't understand that. It could be that area does not predict anything, but the descriptives don't really support that. The average level is 3.29 and individual areas range from 3.11 to 3.53 which seems a good deal on a 4 point scale where almost no one is below 2.

There are plenty of cases in each area. The problem may be that relatively few were dissatisfied. The number of those range from a low of 5 in one area to a high of 16 in another.