variable k values vs. optimal k value

[Jason Padgett]

Hello,

I am wondering if someone could help me understand the advantages and disadvantages of using variable k values over an optimal k value.

Isn't it possible to determine an optimal k value that accounts for the variable sedimentation rates within a core?

Can one determine the optimal k value by running multiple models each with different k values and comparing a set of indices, e.g., agreement index, mean confidence interval, the number of agreement indices <60%, and mean normalized age deviation?

Are there other methods to determine the optimal k value?

Cheers and thanks,

-J

[Christopher Ramsey]

Jason

Essentially what you suggest is what variable k does - but the output is then weighted to those values which are most probable rather than picking one value. The main reason for not using it would be if you have other ways to estimate k from the sedimentation rate (varve counts etc). The other (older) approach which is just to make k as high as you can with the Amodel > 60% will probably give higher precision and gives the most optimistic estimate. A possible advantage of this is that the inputs to the model are more explicit. However, the variable k method is more robust - and will generally have higher uncertainties.

Best wishes

Christopher

> --
> You received this message because you are subscribed to the Google Groups "OxCal" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to oxcal+un...@googlegroups.com.
> To view this discussion on the web visit https://groups.google.com/d/msgid/oxcal/9b6573b4-0311-4bd4-bf1c-c152ffa23b98n%40googlegroups.com.

[MILLARD, ANDREW R.]

If you want an optimal value, you could always run a variable k model and take the value with the maximum posterior density. But choosing an optimal value when the true value is unknown seems to me to be deliberately ignoring an aspect of uncertainty that we know is there. It’s equivalent to calibrating a radiocarbon date and just using the mode of the posterior distribution. For that reason I’d always go for a variable-k model over a fixed-k model unless there are strong reasons to think that we can estimate k a priori.

 

Best wishes

Andrew

--

Dr. Andrew Millard

Associate Professor of Archaeology,

Durham University, UK

Email: A.R.M...@durham.ac.uk 

Personal page: https://www.dur.ac.uk/directory/profile/?id=160

Scottish Soldiers Project: https://www.dur.ac.uk/scottishsoldiers

Dunbar 1650 MOOC: https://www.futurelearn.com/courses/battle-of-dunbar-1650

 

 

From: ox...@googlegroups.com <ox...@googlegroups.com> On Behalf Of Jason Padgett
Sent: 06 June 2022 21:13
To: OxCal <ox...@googlegroups.com>
Subject: variable k values vs. optimal k value

 

[EXTERNAL EMAIL]

--

[Jason Padgett]

Christopher and Andrew,

Right on, your answers are very helpful! Thank you very much!

Cheers,

-J