Chapter 5 Code 5.19: Effect of marriage on divorce A->M->D

[Jaskanwal]

In code 5.19, it feels like the effect of marriage on divorce should be captured as follows:
mu = pm.Deterministic("mu", a + bA * age_shared + bM * marriage)
i.e., the variable marriage_shared should be replaced by marriage.

I am trying to understand how A->M->D relationship is captured when running 5.21 for the values of A specified in 5.20. In the current formulation, the variable marriage_shared is controlling the A->M-D relationship and my understanding is this is set equal to data["MedianAgeMarriage_std"].value. I believe this can have an effect on counterfactual plots (Figure 5.6 left) due to A->M->D dependence (if any).

[jaredsnyder]

I noticed the same issue, I tried to fix it by updating the code to look like this:

# With PyMC3 we have to simulate in each model separately

 

# Simulate the marriage rates at each age first

age_shared.set_value(A_seq)

 

with m5_3_A:

    m5_3_M_marriage = pm.sample_posterior_predictive(m5_3_A_trace, samples=1)

    marriage_shared.set_value(m5_3_M_marriage['marriage'].flatten())

    m5_3_M_marriage = pm.sample_posterior_predictive(m5_3_A_trace)

It didn't seem to effect the plots but I could see that being the result of D being independent of M conditioned on A (or as a result of a mistake I made in attempting a fix)