New estimates of ∆R (Marine20 vs Marine13)

[melt...@ntu.edu.sg]

Dear Christopher,

As you may recall from another thread, I compared U-Th dates and 14C dates on a number of mid-Holocene samples from a site to estimate ∆R, for each sample and also for the whole population.

With the introduction of Marine20, I've now recalculated everything.

A comparison of results is below:

Sample          ∆R (yr) ± 1sigma     ∆R (yr) ± 1sigma

                  for Marine13         for Marine20

Sample 1            -15 ± 27            -164 ± 44

Sample 2            173 ± 31             -18 ± 54

Sample 3            -39 ± 28            -212 ± 51

Sample 4              7 ± 35            -191 ± 56

Sample 5            -20 ± 28            -206 ± 51

Sample 6             23 ± 26            -118 ± 45

Sample 7             20 ± 27            -142 ± 44

Sample 8            -10 ± 26            -160 ± 44

Sample 9             31 ± 31            -107 ± 47

Sample 10            22 ± 27            -125 ± 46

Sample 11            -6 ± 31            -149 ± 48

Sample 12             4 ± 27            -153 ± 48

Overall Site         15 ± 9             -145 ± 14

The average difference is ~160 years, close to what's stated at

http://calib.org/marine/upnotice.html -- so that's nice to see.

But I notice the uncertainties in ∆R are systematically larger with Marine20.

I'll think a bit about the mathematical reasons why that might be, but before I go far down that road, can I double check that these larger uncertainties are real and expected, based on the new calibration curve?

Separately, I realize the Overall Site ∆R seems to be a single value of ∆R that fits ~95% of the individual sample ∆Rs.  If I instead wanted to estimate a larger uncertainty for the Overall Site ∆R that encompasses ~95% of all the variability (and uncertainty) at that site, is there a way to do that in OxCal?

Thanks, and best regards,

Aron

[Christopher Ramsey]

Dear Aron

Thanks for your message.

It is good to see that the new numbers agree with the values from the calib.org site.

The uncertainties on Marine20 are larger than those in Marine13, reflecting the uncertainties seen in different realisations of the curve - and I think more realistic.

On the uncertainty in the Delta_R, I think you need to estimate the variance in the site values and then convert that to the 1-sigma standard deviation to use in Delta_R. A conservative approach would be to calculate the standard deviation of the different estimates of Delta_R you have from your samples.

Others, with more experience of the Delta_R may be able to add to this.

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/feef887c-e1e4-48cf-aecf-9325d25d5092n%40googlegroups.com.

[melt...@ntu.edu.sg]

Dear Christopher,

Thanks for your prompt and helpful suggestions, and for all your hard work this week!  I'll follow through on estimating the variance and so forth.

Incidentally, I also tried out the updated http://calib.org/deltar/ website ... while their uncertainties in most cases are smaller than from OxCal, this was true both for Marine13 and Marine20.  I wonder if it means that http://calib.org/deltar/ isn't fully propagating all the errors through their calculations?

Meanwhile, the means of each ∆R (from individual samples) are similar between http://calib.org/deltar/ and OxCal, for Marine13 and Marine20 respectively -- so everything tracks one another and looks good.

Lastly, just like the uncertainties increased in OxCal in going from Marine13 to Marine20, the uncertainties roughly doubled in http://calib.org/deltar/ in going from Marine13 to Marine20 -- more support for your argument that the larger uncertainties in Marine20 are real.

Thanks again,

Aron

[Christopher Ramsey]

Aron

I think the standard way of calculating Delta_R as on http://calib.org/deltar/ is not to use the uncertainty in the curve. This is because that will be used when you later calibrate and it is then used twice.

Best wishes

Christopher

> To view this discussion on the web visit https://groups.google.com/d/msgid/oxcal/21bf7858-7fdc-4e01-a691-c544b9ccffe3n%40googlegroups.com.