The same SPSS models give DIFFERENT results

[Tim_SPSS]

This absolutely baffles my mind, so I hope you can help me out. I'm doing statistical analyses in SPSS with logistic regression models. I do crude analyses with a single dependent ("A") and a single independent ("B"). Moreover, I do adjusted analyses with a number of additional variables ("X"). I decided to put B in block 1, and then put all my X variables in block 2. This way, from what I know, the output will show both a crude model, and a model adjusted for all the X variables.

Now, here is the catch. When I create a model with JUST A and B, so no extra variables in block 2, the Exp(B) and 95%-CIs I find are DIFFERENT than what I find when I look at block 1 in a model with more variables in block 2. Why is this? As far as I am concerned, this should not be the case.

Thanks, Tim

 

[Stats fan]

Hi Tim,

I think it has to do with missing values. In the analysis with only B as predictor, SPSS omits cases that have missing values on A or B. In the analysis with B as predictor in Block 1 and X as additional predictors in bock 2, SPSS omits cases that have missing values on A, B, or X in all models.

 

[Tim_SPSS]

Hi Tim,

I think it has to do with missing values. In the analysis with only B as predictor, SPSS omits cases that have missing values on A or B. In the analysis with B as predictor in Block 1 and X as additional predictors in bock 2, SPSS omits cases that have missing values on A, B, or X in all models.

Hi Stats fan,

Thanks for your answer, that would make sense. Given I am writing a scientific article, any idea of what method would be preferable? The results I get from both methods do not differ drastically, so I am thinking there is no better option per se.

Thank you.

 

[spunky]

When I create a model with JUST A and B, so no extra variables in block 2, the Exp(B) and 95%-CIs I find are DIFFERENT than what I find when I look at block 1 in a model with more variables in block 2. Why is this? As far as I am concerned, this should not be the case.

So... do you end up with the exact same predictors in both models? Or does one model has more predictors than the other?

 

[Tim_SPSS]

Variables are the same in both models. Except for those in block 2, of course.

 

[spunky]

I'm just trying to understand if your question could be addressed by pointing out that if you are fitting two different models, there's no reason for any of the coefficients to remain the same, unless all the predictors are perfectly uncorrelated.