Reading Binary Logistic Regression results

#1
Hi!

Need some help on Binary Logistic Regression results interpretation.

1) On Block 0: Begining Block i have overall percentage of 76%, but after adding the independent variables in Block 1 i still have 76%. Does it mean the independent variables do not predict the dependent variable?

2) In Cox&Snell r2 I just have 0.027 and in Nagelkerk r2 i have 0,040, are these values too low? Does this mean the model is not adequate?

3) Can someone explain me what i should have on Wald test?

4) In the "Variables in the equation" table i have variables with sig.=0.000 but B=0.009 This is too low, right?

My results seem too poor, even when i use Backward stepwise model.
Can someone explain me how to read the Logistic regression results? And what values are ok, and what are not?
Thank you so much.

Marisa
 
#4
You might want to know that a lot of statisticians dislike, sometimes intensely, Stepwise because it is not robust between samples and generates questionable results sometimes when there is high multicolinearity - if you leave one of these variables in and another out which occurs at times.

1) On Block 0: Begining Block i have overall percentage of 76%, but after adding the independent variables in Block 1 i still have 76%. Does it mean the independent variables do not predict the dependent variable?
I assume you are using what SPSS calls hiearchial regression rather than stepwise here. With linear regression there is a F change test, which if signficant means you added predictive validity to a model in that bloc. I don't know if this occurs in the logistic regression test. I am not sure what you mean by "overall percentage".


2) In Cox&Snell r2 I just have 0.027 and in Nagelkerk r2 i have 0,040, are these values too low? Does this mean the model is not adequate?
There is no statistical answer to this question. You have to decide if these numbers are substantively meaningful to you - that is are they high enough to matter in the context of what you are studying. This is made more difficult by the fact that there is no clear metric for them, unlike linear regression they do not stand for amount of explained variance (or really anything which is intuitively obvious IMHO). The best way to answer this question is to look at the literature in this area and consider what results they found. Then ask how your values compared.

3) Can someone explain me what i should have on Wald test?
There are a lot of different Wald test. If you mean the Wald Chi Square for individual predictors than the easiest thing to look at is the associated p value. If it is less than .05 you can reasonably conclude this variable added predictive value to the model. If you are talking about the overall model Wald value, then this means (if you are below a .05 p value) that the model overal had predictive value.

4) In the "Variables in the equation" table i have variables with sig.=0.000 but B=0.009 This is too low, right?
I am not sure if you mean the sig (which is the p value) or slope (B). Whether it is "too low" is a substantive judgement. Do you feel it has enough impact on the dependent variable to matter? If Y moves .009 units does this matter?