5% expected outliers in a set of dates?

[Erik Marsh]

Hi all –

For some reason, I have the idea that in a set of 100 dates, all things being equal, we can expect 5 of them to outliers. Is this right? Or am I mixing this up with the 5% chance that an individual outlier date needs to be shifted (as in Bronk Ramsey 2009:5)?

thanks!

Erik

[Christopher Ramsey]

Erik

Yes - this possibly conflates a number of different conventions:

1. We usually use the 95% range - which is likely to be exceeded in ~ 5% of cases.
2. For this reason the agreement index thresholds were set to fail at a similar point - so you would expect ~ 5% of these to report a warning.
3. When using formal outlier analysis, in the absence of any other information we often choose 5% as an outlier probability.

This is ultimately rather circular and there could be other things at play. However from experience it seems about right. In general dates on known age material fit a normal distribution fairly well at 1 and 2 sigma but less well at 3 and by 4 or 5 sigma there are certainly more outliers than you would expect. So partly this level masks the fact that our uncertainties are slightly longer-tailed than Normal. All this depends on what you are dating and how.

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/f64b7408-1519-4e55-839b-e811e4c4af25n%40googlegroups.com.

[Erik Marsh]

thanks Christopher! Very helpful.

[Erik Marsh]

In case it's useful to others, I came across a citation that backs this up:

"we should expect one in 20 radiocarbon ages to fall outside of the 95.4% probability range and can only hope that it is not so far outside that range as to make our interpretations importantly wrong." (Hamilton and Krus 2018:199)

Hamilton, W.D., Krus, A.M., 2018. The Myths and Realities of Bayesian Chronological Modeling Revealed. American Antiquity 83, 187–203. https://doi.org/10.1017/aaq.2017.57

[Ray Kidd]

Hello,

Being old and probably past it, gives me the nerve to stick my posterior over the parapet, so to speak.

Out of 20 Radiocarbon ages you would expect one to fall outside the 95.4% Probability range.   That is true.  But that's it.  No more.  You cannot point to the one that falls outside the range and say 'That is the one, the rest are fine'.  Any one of the remaining 19 Radiocarbon Ages still has the individual probability of being outside the 95.4% range.  That is what is meant.by a 95.4% probability range, it is applied to a distribution.  Now I can rest and await the flack!

Best wishes

Ray

-----Original Message-----
From: Ray Kidd <rayfo...@aol.com>
To: erik....@gmail.com <erik....@gmail.com>
Sent: Wed, Feb 2, 2022 8:05 pm
Subject: Re: 5% expected outliers in a set of dates?

Hello,

Being old and probably past it, gives me the nerve to stick my posterior over the parapet, so to speak.

Out of 20 Radiocarbon ages you would expect one to fall outside the 95.4% Probability range.   That is true.  But that's it.  No more.  You cannot point to the one that falls outside the range and say 'That is the one, the rest are fine'.  Any one of the remaining 19 Radiocarbon Ages still has the individual probability of being outside the 95.4% range.  That is what is meant.by a 95.4% probability range, it is applied to a distribution.  Now I can rest and await the flack!

Best wishes

Ray

--
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/ce9ed259-4bc2-4578-ac7f-0a983eafdbaen%40googlegroups.com

.

[Ray Kidd]

[Robert Lewis]

Yes. This is often underappreciated.

In fact, if you have 11 c14 dates,  chances are better than 50% that at least one falls outside the 2 sigma range.

--

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/1739586053.2361997.1643835825522%40mail.yahoo.com.

[Ray Kidd]

I think, philosophically, once you have something you can look  at,, it is no longer a probability.

You can also consider that the probability is two tailed, so 2.5% each side at 5%

regards

Ray

-----Original Message-----
From: 'Robert Lewis' via OxCal <ox...@googlegroups.com>
To: ox...@googlegroups.com
Sent: Mon, Feb 7, 2022 12:39 pm
Subject: Re: 5% expected outliers in a set of dates?

To view this discussion on the web visit https://groups.google.com/d/msgid/oxcal/CAPM3Oif50tvnRwPeNO5TF1qBwBgAQM4P%2BTRGvDeY0D48R-U-WQ%40mail.gmail.com.