My casual jeremiad writ large.

[interstice]

http://www.latimes.com/news/local/la-me-dna20-2008jul20,0,1506170,full.story

I remember using this very example in my stats class last summer. Basically, even though in one sample a certain (combination of) gene(s) may have 1-in-a-billion chance of showing up at random (i.e. a false positive match), that sample may not be representative. For example, the gene may be more common in the Amish thus increasing the false positive rate. One 14-year-old overachieving Chinese student expressed concern about allowing too much freedom for the defense in a legal argument. She also didn't like my reductionist equation of intelligence to reasoning speed (I used this example for a brief and casual lecture on generalized linear models I think); she may have been right there, actually.

The genetics is a little more interesting than just that. You'd like your claims to be "conservative"; that is, the actual chance to be even smaller than you claim. However, in this case it always seems possible that there will be some isolated group of people who may have certain common genes. So do we actually model population flows and take the argument one level higher, making arguments about distributions of genes themselves being unlikely? Should we allow this reasoning in court, having seen the travesty of explanation that even basic DNA testing provokes? But then maybe it's been so hard because people have had an intuitive understanding and suspicion of exactly what's going wrong now!

If not, do we throw everything out, or instead of trying to "do it right" fall back to our flawed and incomprehensible zeitgeist to make our decision for us?

Or is there even a deeper distinction between an unlikely individual within a population (legally admissible argument) and an unlikely population (inadmissible)?

Comments

[random-walker.livejournal.com]

OK, it wasn't this "very example" of course, but it was an example of the same concept as this very example. Same thing.

[suicide-sam-e.livejournal.com]

I think that DNA "evidence" is, ideally, treated as all evidence. (Although in this case, as you said, it is being over-analyzed to death.) It's just part of a case built to establish guilt. Drawing parallels between the presence of DNA at some suspected crime and the similar DNA profile of a suspected criminal.

Like that BS line from ballistics technicians about how the rifling on a bullet slug is "unique as a fingerprint". It is not. But if 1. a suspect is found to have a gun, and 2. the patterns on a bullet fired from this gun match similar patterns on the same type of bullet, then an argument can be made that it was the gun used.

As DNA is less likely to be "falsely" similar (except in atypical mono-type populations), I think it is much stronger evidence than the markings from a weapon made in the same factory as thousands of almost-identical weapons.

Now that I complain about its ambiguous nature... I wonder how often a defense brings in its own contrary ballistics experts? Or is a ballistics lab a cost-prohibitive endeavor only large governments could afford to use to back their claims?

[suicide-sam-e.livejournal.com]

And for a historical perspective, remember that before "DNA" they were convicting people using simple blood typing. (Which is still used as a general paternity test.)

[random-walker.livejournal.com]

The problem here is that, as opposed to blood typing, DNA is incredibly strong evidence with a few possible bugs. That is, apart from atypical populations (which are really kind of a Devil's Advocate argument) and other extremely rare and poorly-documented exceptions it is more-or-less infallible (except for fraud and other deliberate human manipulation). It's so much stronger than blood typing, that to give it proportionally more weight would be to disregard almost everything else.

[suicide-sam-e.livejournal.com]

But you have to compensate for the fallibility of a "jury of peers". Fuck, is "peers" defined loosely in the judicial system. I would wager that to the average American, DNA testing is only slightly above modern witchcraft. (Along with most other dedicated sciences.) Unless joe six-pack and the old lady who's a total shut-in and couldn't get out of jury duty can "look at DNA" themselves, they're probably not going to accept it as a sound method (except possibly on a whim, i.e. "That scientist fellow had nice hair. His testimony is fine with me!").

Most people don't even know what "DNA" is an acronym for. I don't think I can spell it. And I'm not 100% what part of DNA "DNA" is—is it the helix ladders the individual genes form the rungs of? That's my best guess.

So, from a prosecuting point of view wherein one is trying to make a convincing agrument, it is negligent to rely heavily on DNA evidence to win a conviction.

(Even with a well-spoken DNA expert explaining the subject to a jury, that is not going to guarantee the jury being able to understand DNA or the relevance of DNA. Fortunately, I believe these few perpetually dim bulbs would be compensated for in the jury's numbers: If they're in the deliberation room, all, "What was that DNA the guy went on about for half an hour?" and the other jurors clearly think they are a dolt, the few jurors incapable of comprehending DNA testing will, since it is human nature, just go along with the majority of the jury's view on the subject.)

[random-walker.livejournal.com]

Are juries even allowed to refute expert testimony*? Don't they just have to take the 1-in-however-many chance at face value? Even though it may sound cynical, I think I'd rather have jurors focus on the probability and statistics, than the actual science. The science is interesting but not really relevant to the case as the somewhat counterintuitive issues of "false positives". The idea that even an educated juror will discover a fundamental flaw in PCR during the course of a trial is just laughable.

I mean "allowed" in a practical sense, the same way that most jurors believe that jury nullification is not allowed.

[random-walker.livejournal.com]

I also think that paternity testing should go to DNA methods. Or at the very least, if you get a match with blood typing it should be double checked at the DNA level. A negative on blood typing would of course be sufficient.

[suicide-sam-e.livejournal.com]

I believe they do (or did) this on the "Montel Williams Show". To find out if a given man is indeed "her baby's daddy," they perform a blood test, and if that is proven they go to the next level, DNA. The blood typing is more of a "screening" for the more expensive DNA test. (Although I think that DNA testing has gotten much more economical of late...)

Of course, some of these featured men decline going to the level of DNA confirmation. If the child's blood type matches theirs, the kid looks reasonably like them, and they (presumably) can recall having sex with the mother, those three are convincing enough for them. Whether the reason is because they are generally "good" men, or because they want to be able to fall back on "that was never proven at the DNA level!" argument re: paternity, I do not know.

[random-walker.livejournal.com]

I think it's both. Once you are yourself convinced that you're "guilty", it becomes altruistic to "confess" before the airtight evidence is in. Optimal strategy.

[suicide-sam-e.livejournal.com]

There is a margin of error on blood type tests, though, is there not? I recall it being much larger than a well-performed DNA test margin of error. (Although I do not know if, like the DNA test, the majority of the error originates from tests which are not well-performed in the first place.)

Still, it's only like 5% or something.. I think.

[random-walker.livejournal.com]

I'm assuming that the tests themselves are basically infallible (which is pretty much the case if they are conducted properly, the probability being so very small), meaning the margin of error is basically 0. The fact is that e.g. A+ blood is just not that rare meaning you can be confused with other people.

As you say, DNA tests of that level are getting very cheap so it's a no-brainer to do one except for your own personal strategy (as you mentioned), or privacy issues.