GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2003-06 > 1055821323


From: "John F. Chandler" <>
Subject: Re: [DNA] Systemic Errors
Date: Mon, 16 Jun 2003 23:14 EST
In-Reply-To: bwhitta1@bellsouth.net message <007801c33455$07290860$7ec9d6d1@bwhitta1> of Mon, 16 Jun 2003 16:14:45 -0600


Bob wrote:
> John Chandler. Would you give it one more shot at defining "Systemic Errors".
> I have been reading every message for 4-5 months and I haven't gotten it yet.
> Put it in very simple terms for me [much like Ann does]. I have a hunch I know
> but I don't want to confuse the issue if I am wrong. Thanks for your help and
> contributions to this board.

Systematic errors are any errors that you can't eliminate by either of
two expedients: (A) making more careful measurements or (B) averaging
together the results of multiple tests. The opposite of "systematic"
error is "random" error. Consider what happens when a marker drops out
or "fails". That is either a lab error or a personal mutation that made
the marker itself disappear. If the marker is in a gene, then a
mutation would probably have been fatal, so you would suspect a (random)
lab error and definitely rerun the test and hope you get a proper
measurement the second time. Similarly, there are the "clerical
mutations" that we talk about from time to time. These are basically
random and can be kept to a minimum by simply following strict
procedures for collecting and reporting results. In terms of Y STR
testing, an obvious systematic error is to use a wrong calibration
for converting fragment length into a repeat count -- we've seen a lot
of that. For something more abstract, like ethnicity, calibration again
comes into the picture by way of the reference samples. If the samples
are not representative of the intended target populations, then the
results for everybody will be wrong, though not necessarily by the same
amount. Furthermore, it you test multiple members of a family in order
to take an average (that's expedient "B"), you can't expect the errors
to tend to cancel out, as they would for random errors.

John Chandler

This thread: