Going beyond with Bayesian updating.

The paper is a think-piece, not an experiment. It asks: what if we treat each extinction session like a new clue?

The author shows how Bayes’ rule can update the odds that a behavior is truly dying out. You start with a guess, then let the data talk.

No new data were collected. Instead, the paper gives the math recipe. Plug in response counts and the formula spits out the chance the behavior will stay gone.

The curve you draw is no longer eye-balled; it is a living probability line that moves with every zero response.

Gilroy et al. (2017) made the idea real. They wrote free software that uses the same Bayesian engine to pick the best delay-discounting model and give a solid ED50.

DeLeon et al. (2005) share the spirit. They show that fixing the observation window for every kid can fool you; tailor the window and your prediction sharpens—just like Bayesian updating tailors the extinction curve to the data you actually see.

Carey et al. (2014) sound a warning that links back here. They prove that sampling only the first few trials warps mastery curves; Guerin (1992) says the fix is to keep every data point and let Bayes weigh them instead of your eyes.

You no longer have to guess when extinction is ‘stable.’ Next time you run an extinction probe, drop the session totals into a simple Bayes spreadsheet. Watch the probability of ‘true zero’ rise or fall in real time. You will know, with numbers, whether to hold the line or change the plan.