From: Rafal Smigrodzki (rms2g@virginia.edu)
Date: Wed Nov 27 2002 - 16:03:29 MST
Charles Hixson wrote:
> Where do you get your estimate of the accuracy of the testing? (I
> realize that you stipulated that the testing be well done, but please
> be aware that this isn't usually the case. Only a small number of
> sites are typically checked, e.g.)
### There is a core of 13 STR sites, but an arbitrarily large number of
sites can be checked. The only caveat is if the defendant has a monozygotic
twin. With approximately 30 million polymorphisms present on average in any
person's genome, the accuracy of DNA testing itself can be almost
arbitrarily high. A much more important issue is the possibility of fraud or
negligence in testing.
------
> How accurate do you estimate that the test would need to be for the
> official murder to be defensible? Is 50% false positives good enough?
> 10%? 1%? I don't believe that you'll get anywhere near 0.0001% this
> year, or next.
### I can do 1/1000 000 now, in my own lab if I wanted to.
I feel comfortable with anything better than 1/1000. I am willing to risk a
1/1000 chance of myself being wrongfully convicted and killed, if in return
I know that 99.9 of murderers caught with their DNA are terminated.
-------
And probably not in the next couple, either. (After
> that, who knows? With the current rate of change, it's possible.)
> OK. How accurate to you believe the current tests to be? Why do you
> believe this? At one point, a few years ago, I heard that they
> typically checked only 10 sites.
### Even 10 sites can give you high level of confidence. If each site has,
say 5 alleles at 20% each, you have 5e10 squared combinations (100 million
or so), with equal frequency in the population.
------
This is good usually enough if the
> test is to confirm what they have other good evidence to believe,
> which was the use at that time. Do you have any information that as
> they have begun to use the database for original suspect selection
> that they've improved the testing in the necessary way for
> statistical reasoning? (I don't, so I have been presuming that the
> original techniques have been maintained.)
### This I don't know.
Rafal
This archive was generated by hypermail 2.1.5 : Wed Jan 15 2003 - 17:58:27 MST