concept

The statistical approach in which a prior view is updated as data arrive. In episode 01 Marvin Barth explains it as the shape of his method while insisting that what he did is not proper Bayesian statistics. Start with episode 01, at about 26 minutes.

Across the episodes

01

Enters as the vocabulary of Barth's method, invoked and disclaimed in the same breath

Barth's plain-language contrast with classical hypothesis testing: start from a belief and revise it.

"In Bayesian statistics, and I'm bastardizing this a bit, but this is roughly the difference, is you start with a view and then you say, do the data as they come in, each new piece of data I get in, update my view in a different way or does it cause me to just stay where I was?"
Marvinasserted#
Marvin, The Grant Williams Podcast: Forecasts and Fortunes, Episode 1 (2 Nov 2026), 26:10

He defines the term Grant has already been using, so the method can be described in its own vocabulary.

"Prior, that's your view that you enter with, was that we did have polling errors, but I wasn't sure whether they would be big enough to overcome what seemed like a big democratic advantage."
Marvinasserted#
Marvin, The Grant Williams Podcast: Forecasts and Fortunes, Episode 1 (2 Nov 2026), 26:36

Barth borrows the logic without claiming the rigor.

"I use a Bayesian process, but I wouldn't call it Bayesian statistics if that makes any sense."
Marvinasserted#
Marvin, The Grant Williams Podcast: Forecasts and Fortunes, Episode 1 (2 Nov 2026), 27:18

The disclaimer he says he makes throughout the piece.

"This is not proper Bayesian statistics."
Marvinasserted#
Marvin, The Grant Williams Podcast: Forecasts and Fortunes, Episode 1 (2 Nov 2026), 35:17

Connections

Barth's process does not reach significance.