Unmodelled BP v Modelled BP

[Kim Newman]

I would like to clarify if I should be reporting the Unmodelled (BP) age range or the Modelled (BP) age range.

If I was to calibrate dates using just the IntCal20 or SHCal20 curve the calibrated dates would output as Unmodelled BP (or likelihoods as per Format > Show > Output)

But I am calibrating dates from Indonesia which is within the ITCZ so am using a mixed curve -- Mix_Curve("Mixed","IntCal20","SHCal20",U(0,100)).  I'm looking to have an updated list of calibrated dates from a number of old excavations across the region.

The table output of a Mixed Curve provides both Unmodelled (BP) (likelihood) and Modelled (BP) (Posterior) age ranges.  My understanding in Bayesian modeling is that a prior and the likelihood are used to produce a Posterior which is essentially the modelled answer.  So my understanding is that if I was to write out a calibrated age I would use the Modelled (or Posterior) age range. 

I have had someone questions me on this given that a single curve result would be presented as Unmodelled (BP), so as I am new to this I was hoping someone could clarify if my understanding is correct.

Cheers

Kim

[Richard Staff]

Hi Kim,

The important thing (as described in Millard, 2014 - https://doi.org/10.2458/56.17455) is that, whatever data you do present, it is clear what has been done, such that any reader of your work will be able to correctly interpret and potentially reproduce your numbers.

In your case, I think that you are right, and you just want to present the "modelled" data, specifying either in the table caption, or the related text, what the modelled data relate to. I.e., allowing for the uncertainty of competing NH vs SH air mass influence on your samples, and explicitly stating which calibration curves (IntCal20 and SHCal20) and what mixing relationship (the full range between 100% NH and 100% SH) you have used. As well as any other assumptions (prior information) that have gone in to the model.

In order for your data to be reproducible though, it is important to provide the raw, uncalibrated 14C data too, although these can always be included in supplementary material/appendices if it becomes too "clunky" for the main text/tables.

In your case, I wouldn't have thought that the "unmodelled" data would be necessary/meaningful, so I wouldn't include it. So long as you include the other information (above), others would be able to generate it for themselves, if they had any reason to.

Hope that helps!

Richard 🙂

---

From: ox...@googlegroups.com <ox...@googlegroups.com> on behalf of Kim Newman <kim.n...@griffithuni.edu.au>
Sent: 30 June 2021 07:51
To: OxCal <ox...@googlegroups.com>
Subject: Unmodelled BP v Modelled BP

 

--
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/0769b184-10ec-4a8c-b941-a33d9fef3849n%40googlegroups.com.

[Kim Newman]

Thank you for answering my question Richard, it's reassuring that my understanding was of the use of Unmodelled/Modelled is correct.  I will also be presenting this data in a table so that the raw and calibrated values will be available as well as an explanation of the curve.

Cheers

Kim

[Christopher Ramsey]

Yes - it may seem odd that modelling is needed for a simple mixture of the curves. The reason that it is is that although we tend to specify the mixture using a normal distribution, the actual mixture must lie between 0-100% and you can specify the mixture in different forms (say uniform between 0-100%). The modelled version includes this effects whereas the unmodelled just propagates the errors using a standard error model.

Best wishes

Christopher

> On 5 Jul 2021, at 09:06, Kim Newman <kim.n...@griffithuni.edu.au> wrote:
>
> Thank you for answering my question Richard, it's reassuring that my understanding was of the use of Unmodelled/Modelled is correct. I will also be presenting this data in a table so that the raw and calibrated values will be available as well as an explanation of the curve.
>
> Cheers
> Kim
>
>
> On Friday, July 2, 2021 at 10:15:14 PM UTC+10 Richard wrote:
> Hi Kim,
>
> The important thing (as described in Millard, 2014 - https://doi.org/10.2458/56.17455) is that, whatever data you do present, it is clear what has been done, such that any reader of your work will be able to correctly interpret and potentially reproduce your numbers.
>
> In your case, I think that you are right, and you just want to present the "modelled" data, specifying either in the table caption, or the related text, what the modelled data relate to. I.e., allowing for the uncertainty of competing NH vs SH air mass influence on your samples, and explicitly stating which calibration curves (IntCal20 and SHCal20) and what mixing relationship (the full range between 100% NH and 100% SH) you have used. As well as any other assumptions (prior information) that have gone in to the model.
>
> In order for your data to be reproducible though, it is important to provide the raw, uncalibrated 14C data too, although these can always be included in supplementary material/appendices if it becomes too "clunky" for the main text/tables.
>
> In your case, I wouldn't have thought that the "unmodelled" data would be necessary/meaningful, so I wouldn't include it. So long as you include the other information (above), others would be able to generate it for themselves, if they had any reason to.
>
> Hope that helps!
>
> Richard 🙂
>

> From: ox...@googlegroups.com <ox...@googlegroups.com> on behalf of Kim Newman <kim.n...@griffithuni.edu.au>
> Sent: 30 June 2021 07:51
> To: OxCal <ox...@googlegroups.com>
> Subject: Unmodelled BP v Modelled BP
>
> I would like to clarify if I should be reporting the Unmodelled (BP) age range or the Modelled (BP) age range.
>
> If I was to calibrate dates using just the IntCal20 or SHCal20 curve the calibrated dates would output as Unmodelled BP (or likelihoods as per Format > Show > Output)
>
> But I am calibrating dates from Indonesia which is within the ITCZ so am using a mixed curve -- Mix_Curve("Mixed","IntCal20","SHCal20",U(0,100)). I'm looking to have an updated list of calibrated dates from a number of old excavations across the region.
>
> The table output of a Mixed Curve provides both Unmodelled (BP) (likelihood) and Modelled (BP) (Posterior) age ranges. My understanding in Bayesian modeling is that a prior and the likelihood are used to produce a Posterior which is essentially the modelled answer. So my understanding is that if I was to write out a calibrated age I would use the Modelled (or Posterior) age range.
>
> I have had someone questions me on this given that a single curve result would be presented as Unmodelled (BP), so as I am new to this I was hoping someone could clarify if my understanding is correct.
>
> Cheers
> Kim
>
>
> --
> 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/0769b184-10ec-4a8c-b941-a33d9fef3849n%40googlegroups.com.
>
> --
> 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/0e9564e2-41f8-4afb-a641-1b5e5030fc5cn%40googlegroups.com.

[Kim Newman]

Thanks for that explanation Christopher,  I had actually thought that was the case but wasn't confident in my understanding.

Appreciate you clarifying this

Cheers

Kim