Bayesian Log-Linear Regression Models

This feature requires SPSS® Statistics Standard Edition or the Advanced Statistics option.

The design for testing the independence of two factors requires two categorical variables for the construction of a contingency table, and makes Bayesian inference on the row-column association. You can estimate the Bayes factors by assuming different models, and characterize the desired posterior distribution by simulating the simultaneous credible interval for the interaction terms.

1. From the menus choose:

   Analyze > Bayesian Statistics > Log-Linear Models

2. Select a single, non-scale, row variable from the Variables list. You must select at least one non-scale variable.
3. Select a single, non-scale, column variable from the Variables list. You must select at least one non-scale variable.
4. Select the desired Bayesian Analysis:
   • Characterize Posterior Distribution: When selected, the Bayesian inference is made from a perspective that is approached by characterizing posterior distributions. You can investigate the marginal posterior distribution of the parameter(s) of interest by integrating out the other nuisance parameters, and further construct Bayesian confidence intervals to draw direct inference. This is the default setting.
   • Estimate Bayes Factor: When selected, estimating Bayes factors (one of the notable methodologies in Bayesian inference) constitutes a natural ratio to compare the marginal likelihoods between a null and an alternative hypothesis.

     Table 1. Commonly used thresholds to define significance of evidence

     Bayes Factor	Evidence Category	Bayes Factor	Evidence Category	Bayes Factor	Evidence Category
     >100	Extreme Evidence for H0	1-3	Anecdotal Evidence for H0	1/30-1/10	Strong Evidence for H1
     30-100	Very Strong Evidence for H0	1	No Evidence	1/100-1/30	Very Strong Evidence for H1
     10-30	Strong Evidence for H0	1/3-1	Anecdotal Evidence for H1	1/100	Extreme Evidence for H1
     3-10	Moderate Evidence for H0	1/10-1/3	Moderate Evidence for H1

     H0: Null Hypothesis

     H1: Alternative Hypothesis

     ¹

     ²

   • Use Both Methods: When selected, both the Characterize Posterior Distribution and Estimate Bayes Factor inference methods as used.