Can we talk about that whole Bayesian inference thing for dating pottery?

I was at a dig last summer in New Mexico and this PhD student kept telling me to use Bayesian analysis on the pottery layers instead of just standard radiocarbon dating. I figured she was just some academic who never got her hands dirty, so I basically ignored her. Fast forward 8 months later, I'm trying to sort out a site where the layers were all mixed up from an old flood. Standard dating gave me a mess of dates that made zero sense. I finally gave in and ran the data through a basic Bayesian model this guy online showed me. It actually cleaned everything up and showed a clear sequence of occupation from 1100 to 1300 AD. Now I feel like an idiot for wasting all that time. Has anyone else had a similar experience where some "book smart" advice ended up being the real deal?

3 comments

3 Comments

anthony_jenkins612d ago

The whole 95% chance between 1070 and 1120 AD is basically what I got from my model, so you're right that it's not a single number. But I still think the Bayesian output is more useful than just a range. When I ran it on my flood damaged site, it flagged three charcoal samples as clear outliers that the standard radiocarbon dates had me scratching my head over. Removing those outliers gave me a much tighter timeline of occupation that matched the stratigraphy I could actually see in the ground. So while the model doesn't hand you a fixed date like 1150 AD exactly, it does give you a practical way to throw out bad data and see the real pattern. That's way more useful than just knowing your error bars got smaller.

dylan_thompson2d ago

The 1100 to 1300 AD thing is close but not exactly right. Bayesian methods don't give you a magic number like that, they just narrow down the probability ranges for each layer. So you probably got something like 95% chance the oldest occupation was between 1070 and 1120 AD or something similar. I made that same mistake at first, thinking the model spat out fixed dates. It really just makes your error bars smaller and shows which dates are outliers, which is why it fixed your flood messed up site.

jenny_sullivan972d ago

Totally agree, it's way more about narrowing the odds than giving you a neat little answer.