The analysis proceeded in three steps using a purposeful selection method (Hosmer & Lemeshow, 2000). 1) Each variable was examined using a univariate test. Variables showing p<.20 based on the individual tests were retained and entered simultaneously in a saturated model. 2) Using the log-likelihood ratio test, variables with p>.05 were removed sequentially based on their significance. The effect of removal of each variable was scrutinized by examining the change in the coefficients in order to identify potential confounders. 3) Once the variables for the final model were identified, tests of potential moderating effects of RA and PA with sex were examined.
