However, I was wondering whether the predictor*log interactions should be included when conducting the logistic regression analysis that tests my hypotheses. When the interactions of the continuous independent variables and their logs are included, the coefficients and significance (as observed in the SPSS output) is different compared to when only the predictors (and no interactions with logs) are included. See screenshots for the difference in regression output.
So basically I want to know whether I need to include the Box-Tidwell test interactions in my main model. Or do I need to first run a binary logistic regression to conduct the Box-Tidwell test, then see whether the assumption is met or not, then run a new binary logistic regression with only my predictors, and then use this output to report my results?
I hope my question is clear. If not, I'd be happy to clarify.
The current study provides a concise review concerning molecular biology of cancer initiation and progression. A summary of some important progresses made in the last two years to present from most recent ground breaking studies in cancer biology is discussed. This synthesis shows a novel understanding of molecular and cellular changes underlying c...