Boundaries for directly dated events

[Pavol Hnila]

Hello all,

I am a postdoc working on a chronological model of a tell settlement, where some settlement phases are represented by only three radiocarbon dates: one for the beginning (e.g. last rings of a construction beam), one for the use, and  one for the destruction (e.g. hoard of carbonized seeds burnt during the final conflagration). Construction and destruction were likely very short events, whereas the duration of the use can be expected to run several decades. I am not sure whether which of the following variants is methodologically preferable:

Variant A (Sequence with Boundaries defined by R_Date):
Sequence()
{
Boundary(R_Date(Construction));
R_Date(Use);
Boundary(R_Date(Destruction));
};

Variant B (Sequence with Start and End Boundaries calculated by OxCal)
Sequence()
{
Boundary("Start");
R_Date(Construction);
R_Date(Use);
R_Date(Destruction);
Boundary("End");
};

Variant C (Sequence with Start, Intermediate, and End Boundaries calculated by OxCal)
Sequence()
{
Boundary("Start Phase");
R_Date(Construction);
Boundary("Start Use");
R_Date(Use);
Boundary("End Use");
R_Date(Destruction);
Boundary("End Phase");
};

I know that intermediate boundaries make sense for the definition of groups with more dates, so that number of dates inside a group does not influence the overall span. Yet even grouping of single dates between boundaries seems to influence the results. Is is because they are treated as additional events, which then "squeeze" the modeled spans of original dates? So I included this option among the variants, hoping that someone experienced can elucidate on the meaningfulness or not of intermediate boundaries for single dates.

Or, maybe, there is some Variant D that I did not consider?

Best wishes,
Pavol

[Thomas S. Dye]

Aloha Pavol,

The Boundary() command estimates the age of something, typically a
context boundary, for which there is no directly associated dating
evidence.

In your case, it seems you have directly associated dating
evidence for Construction and Destruction. Perhaps this would
work?

Sequence()
{
R_Date(Construction);
R_Date(Use);
R_Date(Destruction);
};

Others on the list might have better ideas how to model your
situation.

All the best,
Tom

Pavol Hnila <phn...@gmail.com> writes:

> Hello all,
>
> I am a postdoc working on a chronological model of a tell
> settlement, where some
> settlement phases are represented by only three radiocarbon
> dates: one for the
> beginning (e.g. last rings of a construction beam), one for the
> use, and one
> for the destruction (e.g. hoard of carbonized seeds burnt during
> the final
> conflagration). Construction and destruction were likely very
> short events,
> whereas the duration of the use can be expected to run several
> decades. I am not
> sure whether which of the following variants is methodologically
> preferable:
>

> *Variant A (Sequence with Boundaries defined by R_Date):*
> Sequence()
> {
> Boundary(R_Date(Construction));
> R_Date(Use);
> Boundary(R_Date(Destruction));
> };
>
> *Variant B (Sequence with Start and End Boundaries calculated by
> OxCal)
> *Sequence()
> {

> Boundary("Start");
> R_Date(Construction);
> R_Date(Use);
> R_Date(Destruction);
> Boundary("End");

> };*
> *
> *Variant C (Sequence with Start, Intermediate, and End
> Boundaries**calculated by
> OxCal)*

> Sequence()
> {
> Boundary("Start Phase");
> R_Date(Construction);
> Boundary("Start Use");
> R_Date(Use);
> Boundary("End Use");
> R_Date(Destruction);
> Boundary("End Phase");

> };*
> *

> I know that intermediate boundaries make sense for the
> definition of groups with
> more dates, so that number of dates inside a group does not
> influence the
> overall span. Yet even grouping of single dates between
> boundaries seems to
> influence the results. Is is because they are treated as
> additional events,
> which then "squeeze" the modeled spans of original dates? So I
> included this
> option among the variants, hoping that someone experienced can
> elucidate on the
> meaningfulness or not of intermediate boundaries for single
> dates.
>
> Or, maybe, there is some Variant D that I did not consider?
>
> Best wishes,
> Pavol

> **


--
Thomas S. Dye
https://tsdye.online/tsdye

[MILLARD, ANDREW R.]

Model A should work. If you have multiple dates for the Use, Model A will have a uniform prior on the span of the phase, whereas Tom's suggestion will have a prior that is biased to longer lengths with more dates for Use.

My question would be whether you are really sure that these samples represent the construction and destruction events, and therefore are also boundaries in the model. Grain may be stored for a few years before charring. If you had unthreshed material then I would go with Model A, but here I think the confidence in the association may be too strong. Similarly with a date on a beam. Because repairs can take place this can be at or after the start of the phase. For these cases Model B is better. But if this is all constrained within a sequence with high chronological resolution the choice of Model A or Model B may not make much difference numerically.

Model C says that there are distinguishable periods of construction and destruction, and I don't think that is likely.

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

-----Original Message-----
From: ox...@googlegroups.com <ox...@googlegroups.com> On Behalf Of Thomas S. Dye
Sent: 17 March 2022 15:52
To: ox...@googlegroups.com
Subject: Re: Boundaries for directly dated events

[EXTERNAL EMAIL]

--
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/87lex8lgq2.fsf%40tsdye.online.

[Pavol Hnila]

Thank you, Andrew and Thomas, for your replies!

Andrew, it is correct - in most cases I cannot make such distinctions
and I was automatically using either Model B or its variant with a phase
that combined Construction+Reconstruction+Use (the site offers ample
evidence for reconstructions and some beams indeed have different
radiocarbon dates). Yet in one particular case - the one that made me to
think about Model A - I am quite confident about the attributions (see
Appendix). I am happy to hear that the model is methodologically correct
(when attributions are correct).

Since Model B also has a uniform prior on the span, will it tend to
equalize the distances between the three R_Dates and as a result the
Start and End can be skewed? I'm somehow afraid that the three R_Dates
will be treated as random samples from the period enclosed by the Start
and End boundaries and an unknown period will be added on both ends. If
this is the case, is there - except of adding more radiocarbon dates -
any workaround for constraining the algorithm to treat the construction
and destruction dates as being VERY CLOSE to the Start and End boundary?
UniformSpanPrior=FALSE seems to make little difference and it is not an
option for me, unless I can apply it locally for a part of a bigger model.

Model C (with intermediate boundaries) was indeed my attempt to separate
the "use period" and to force the algorithm towards distinguishing
better between the three events. But the intermediate boundaries seem to
shorten the probability ranges of all three events, not only
construction and destruction. Would you agree that Model C is the
preferable choice in cases where we have  different number of samples
defining a sequence inside a single settlement phase  - e.g. 2
construction samples, 4 use samples and 1 destruction sample?

I also expect that in bigger models the numeric differences between
individual variants will be very small, nonetheless I would like to
understand the effects of all of this choices.

Best wishes,
Pavol

Appendix (for those interested in the context of the samples):
The quoted beam was found in situ in the foundations, which originally
consisted of a special timber-rubble framework. The sampled beam had
traces of bark and its early date does not contradict the assumption
that it must have been part of the original construction. Similarly, the
grain. I called it a "hoard of seeds", but only to stress that it was
not a single seed but a small intentional accumulation. It was not a
storage or refuse context, but a deposition in an open-air courtyard
with a series of small shallow pits/cup-marks integrated into a walking
level. They likely served for offerings or other temple activities. It
is unlikely the seeds were lying around for longer, since in such a case
they would have been either removed by rodents/insects/birds or would
have disintegrated quite quickly. The most likely interpretation is that
they were deposited very shortly (days or few weeks at most) before the
conflagration that destroyed the entire site. For the purposes of the
model, given its resolution of five years, I treat the seed deposition
event as being contemporary with the destruction.

[Erik Marsh]

Pavol – 

 

Is there any workaround for constraining the algorithm to treat the construction and destruction dates as being VERY CLOSE to the Start and End boundary?

You can try using different types of boundaries. Uniform boundaries do assume the dates are evenly distributed between Start and End. But you have other options like Zero_Boundary, Sigma_Boundary, and Tau_Boundary – check the manual. Tau Boundaries are often used for burn events, when you have good reason to assume that most of the dates are from very close to the ending boundary (see this post, which has some published examples).

Hope this helps,

Erik

[Pavol Hnila]

Thank your Erik for your suggestion and the useful link! Though I could not figure out how to make it work for my sequence with three dates (except with intermediate boundaries), it might work when rearranging my events inside a bigger sequence. I need to think more about it.

Best wishes,
Pavol

--
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/e9910d5a-26c4-41c6-a59d-c4c0b311dc63n%40googlegroups.com.