GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2006-02 > 1140890391


From: Doug McDonald <>
Subject: Re: [DNA] SNP Test Necessary: WEBB-BIRNBACH R1b SuperX HaplotypeStudy Report
Date: Sat, 25 Feb 2006 11:59:51 -0600
References: <200602250337.k1P3br9C012337@mail.rootsweb.com> <43FFD6A6.8050601@sbcglobal.net> <44007833.50408@scs.uiuc.edu> <44008504.8070207@sbcglobal.net>
In-Reply-To: <44008504.8070207@sbcglobal.net>


Alfred A. Aburto Jr. wrote:
> > Doug McDonald wrote:
>
>> Alfred A. Aburto Jr. wrote:
>>
>>> They are not accurate, because they vary with sample size, and also
>>> the results vary with samples of the same size (look at the web page
>>> results for DYS389I where there is a factor of 2 variation and the
>>> sample sizes are comparable). What one would like to see in terms of
>>> accuracy are results like, for example, 2.16E-03 +/-0.04 from
>>> scientist A, and then another totally independent result like
>>> 2.167E-03 +/-0.008 from scientist B and so on ... then in that case
>>> you would have accurate and repeatable results. The results at the
>>> YHRD web site don't show that behavior.
>>
>>
>>
>> yes they do
>>
>> They are quite consistent, it's just that the individual studies
>> have large standard errors.
>>
>
> What worries me is that the fundamnetals don't look right. What I mean
> is that, in the case I'm thinking about for DYS389I, in the YHRD
> results we have Dupuy et al. 2004 with 1766 samples giving a mutation
> rate of 2.26E-03, and then Gusmao et al. 2005 with 1793 samples giviing
> a mutation rate of 1.11E-03. A factor of 2 difference! The first thing,
> before publishing, a scientist must ask is why do I differ from this
> other result by this factor of 2? The sample sizes are both "large" at
> 1766 and 1793 meioses and they are of comparable size, yet there is a
> factor of 2 difference.

No ... the sample sizes that matter are 4 and 2 respectively ...
the number of mutations, not the number of samples. That's your mental
miscalibration.



>What is it that is causing that? It seems that
> the fundamentals are at odds (or something is wrong with them that we
> don't understand).

The "expected error" is not dependent on (the square root of )
the total number of samples, it is dependent on (the square root of)
the number of mutations.

The expected errors are .0011 and .0008 respectively. Thus these results
in fact agree as expected.

Doug McDonald


This thread: