WEBVTT

NOTE As submitted (2023) vs corrected: captions match the on-screen steps.

1
00:00:00.200 --> 00:00:07.727
Step 1 of 8. Corrected: all four methods agree on the dose slope β₁.

2
00:00:07.727 --> 00:00:13.826
Step 2 of 8. Flip the toggle to As submitted (2023): the original code, bugs and all.

3
00:00:13.826 --> 00:00:21.011
Step 3 of 8. Dose became a 7-level factor, and each method was fed a different likelihood.

4
00:00:21.011 --> 00:00:27.828
Step 4 of 8. The four posteriors for β₁ now disagree, for reasons unrelated to the algorithms.

5
00:00:27.828 --> 00:00:36.644
Step 5 of 8. β₂ never entered the Metropolis likelihood, so its chain wanders freely.

6
00:00:36.644 --> 00:00:43.357
Step 6 of 8. Back to Corrected: the same functions, given the data the question describes.

7
00:00:43.357 --> 00:00:52.110
Step 7 of 8. Why and how it was fixed: decision record DR-001, under Methods.

8
00:00:52.110 --> 00:01:05.560
Step 8 of 8. The decision in one line, then context, options, what happened and what I'd change.