jean997
Does this indicate a lack of causal relationship?
I have a question regarding the results I obtained using your methodology. The p-value for the CAUSE model exceeds 0.05. Does this indicate a lack of causal relationship?
model1 model2 delta_elpd se_delta_elpd z
1 null sharing 0.2030073 0.1568134 1.2945782 2 null causal -0.9606224 1.2914358 -0.7438406 3 sharing causal -1.1636296 1.1884105 -0.9791479 p-value testing that causal model is a better fit: 0.16 Posterior medians and 95 % credible intervals: model gamma eta q [1,] "Sharing" NA "0.04 (-0.52, 0.49)" "0.05 (0, 0.27)" [2,] "Causal" "0.04 (0, 0.08)" "-0.02 (-0.56, 0.49)" "0.04 (0, 0.24)"
I think this case is a little borderline personally. I think it is ok to make inference based on the 95% credible interval for gamma, though in the paper we use the model comparison p-value. The model comparison p-value is more conservative. Basically this is telling you that when you fit the model with a causal effect, most of the posterior is positive. The causal model also fits better than the non-causal (sharing only model). However, the causal model doesn't fit overwhelmingly better than the sharing model. So I think this is weak evidence of a causal effect.
