GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2009-07 > 1247739666


From: Alan R <>
Subject: Re: [DNA] For Age Estimates beware of tossing STRs away
Date: Thu, 16 Jul 2009 10:21:06 +0000 (GMT)
References: <007c01ca0574$c6d62a70$6400a8c0@Ken1><458251.74284.qm@web86608.mail.ird.yahoo.com><009501ca05c5$bf0cbe20$6400a8c0@Ken1>
In-Reply-To: <009501ca05c5$bf0cbe20$6400a8c0@Ken1>


Ken-that's what I understood regarding confidence intervals too.  As for my other question, can all the fast and at least some of the medium STRs be removed as a sort of compromise between Tim's method and the need to not have huge confidence intervals? Would that work? Does that make sense?

Alan


Alan

1 standard deviation or 67 percent confidence intervals sort of speak for themselves; 1/3 the time the difference of true and measured value of whatever will be greater than 1 standard deviation.

2 standard deviation or 95 percent confidence intervals also speak for themselves; 5 percent of the time the difference of true and measured value of whatever will be greter than 2 standard deviations.

It's a psychological or subjective thing what kind of certainty or uncertainty you want to live with.  Of course these are just the statistical confidence intervals due to the random nature of mutations.  There will be additional uncertainties or confidence interval due to the mutation rates not being precisely measured, nature's rules for the mutational behavior of the STRs being different than what is in the model, uncertainties for the finite sampling, etc.  Some of these latter uncertainties should probably be called unknown biases in that whatever uncertainties they contain get transmitted into systematic pushing of all age estimates up or down by some amounts.

This thread: