I'm performing binary logistic regression in SPSS; y is dichotomous variable; and both Xs are continuous variables.
I performed three models and I have troubles interpreting model with both predictors and with continuous by continuous interaction
In first model one predictor was introduced, and result was as hypothesized: negative and significant B.
In second model another predictor was introduced. Model was signifficant (Chi-Square= 6.39, p<0.05), but both predictors were insignifficant, although in the same direction as in first model.
I tried to solve this issue by introducing an interaction of those variables in the third model. Finally, third model showed the best fit (Chi-Square=12.77, p<0.01). But this time predictors showed opposite direction (became positive), and interaction became negative predictor. I'm having hard times interpreting meaning of interaction.
I have following questions regarding these results:
What does this interaction mean? How to interpret negative interaction of continuous variables in binary regression model?
Why predictors changed their direction when interaction was introduced?
Is it odd that this interaction have lowest B value, but also lowest p value?
Is there a way to calculate beta value for this interaction?