WEBVTT

NOTE Diagnostics: can the posterior be trusted?: captions match the on-screen steps.

1
00:00:00.200 --> 00:00:06.130
Step 1 of 9. Diagnostics: each sampler runs as four chains, seeds 90125 to 90128.

2
00:00:06.130 --> 00:00:12.482
Step 2 of 9. R-hat, bulk and tail ESS and MCSE for every coefficient, checked against R's posterior package.

3
00:00:12.482 --> 00:00:20.738
Step 3 of 9. Trace plots: four chains from dispersed starts overlap after warm-up.

4
00:00:20.738 --> 00:00:24.864
Step 4 of 9. Rank plots: near-flat histograms mean the chains mix well.

5
00:00:24.864 --> 00:00:34.153
Step 5 of 9. HMC: one leapfrog trajectory of the original HMC.fn, slowed down.

6
00:00:34.153 --> 00:00:42.465
Step 6 of 9. The Hamiltonian stays nearly constant along the path, so the proposal is accepted.

7
00:00:42.465 --> 00:00:55.930
Step 7 of 9. Auto-run: the chain fills in the posterior, replaying R's draws.

8
00:00:55.930 --> 00:01:04.417
Step 8 of 9. Posterior predictive check: replicated trials against the observed counts at each dose.

9
00:01:04.417 --> 00:01:12.880
Step 9 of 9. Test statistics with their Monte Carlo error: no sign of misfit.