borderline significantp = 0.09
We then used backwards elimination regression modeling and eliminated all non-significant (statistically significant if p ≤ 0 .05 from the Wald statistic) interactions from t he initial model (the interaction with home smoke rule was borderline significant in the final model p = 0.09 but was retained because it was significant in some iterations of the model) and then one-by-one eliminated each of the non-significant main effects whose removal did not cause the beta coefficients of the other variables in the model to change.