I’ve conducted an ordinal logistic regression analysis with 3 independent variables. I then want to ask if adding an interaction to an independent variable is warranted. What is the best approach for determining whether to retain the interaction term or not? Do you base this decision on the p-value of the parameter estimate of the interaction term, p-value of the interaction term in the Wald Test, or p-value of the interaction term in the Likelihood Ratio Test? For my data, the p-value of the parameter estimate of the interaction term and in the Wald Test are < 0.05, but the p-value on the Likelihood Ratio Test is > 0.05.
