Lisa, you mentioned that the study variable in question was not among those initially tested with chi-square. I suggest that you repeat the former test with this particular study variable now included. Every variable has to account for itself in research.

Can you explain your rationale more? Thanks.

Hi Lisa,

If I'm understanding your question correctly, you're referring to a model with a single predictor (which is not significant) to a model with several predictors, in which the one variable is now significant?

Lisa, you stated that the study variable that now became significant was not included in the chi-square analysis. If that is the case, you can now repeat the ch-square test including the particular variable.

As Jason asked, if the variable was included in the original analysis, there should be nothing to worry about the validity of your analysis.

@Jason - Yes, in that initially looking at any differences in the outcome between groups (study vs control). This was not significant. However, once running the full model which included controls, plus another study variable, the study variable was significant. This is a different variable than what was tested with t-test and chi-square.

Thanks for your input.

Greetings! Can anyone pls address the following queries?

I have a dichotomous dependent variable and 50 predictors (ordinal or binary nominal).

1. If I first conduct Chi-square tests and then only put the significant predictors in the binary logistic regression, is that okay? If I do this, my Nagelkerke R square becomes low but if I put all the 50 predictors in the binary logistic regression model, the Nagelkerke R square looks pretty good, around 0.555. The results in the two cases don't vary too much.

2. If I input all the variables in the model, for some ordinal categorical variables, two confounding scenarios occur:

>in one case, the overall variable is shown significant but none of it's category is significant.

>on the other hand, 1 or 2 categories become significant but the overall variable is insignificant.

What should I do?

3. Since all my independent variables are categorical (ordinal or binary nominal), how can I perform collinearity test in SPSS? Is that test necessary if I follow the steps in 1?

4. If I perform linear regression to get the tolerances and VIFs, then if I see none of the tolerances are near to 0 or 1, and none of the VIFs is greater than 5, and then if I perform binary logistic regression with all the variables, is that an accepted method?

Regards.