3 ms·

Given the surprisingly poor perfection of dna "matching" and the tendency for public and law-court rhetoric to seriously misrepresent the odds of a mistake... a

Given the surprisingly poor perfection of dna "matching" and the tendency for public and law-court rhetoric to seriously misrepresent the odds of a mistake... at this point anyone who would voluntarily submit dna - or anything other than fight very hard against handing it over - has grotesquely poor judgement.

asdff 4y ago
Can you site a source documenting this poor perfection of DNA matching? I dug into this after reading your comment and I found the opposite was true, that a false positive as an odds on the rate of 1 in 1 billion. When you couple it with the fact you also need more than just DNA evidence to convict (which was done with the golden state killer, DNA was not the only source of evidence), the odds grow even smaller of 1) you having no alibi or potentially being at the place of the crime at the time and more evidence, and 2) you being that 1 in 1 billion match to the real killer. https://www.parliament.uk/documents/post/postpn258.pdf https://www.parliament.uk/documents/post/postpn258.pdf
- yetanotherloser 4y ago
  pop article and still too forgiving, but this is not a bad start: https://freakonomics.com/2008/08/are-the-fbis-probabilities-about-dna-matches-crazy/ https://freakonomics.com/2008/08/are-the-fbis-probabilities-... (it's still excessively charitable about the odds because the genetic factors are not really perfectly independent and even a tiny bit of interdependence wreaks havoc with large probability calculations) The link you cite with "1 in billions" odds is exactly the sort of misrepresentation I'm talking about. Combine smaller-than-pretended probabilities with big database trawling and it becomes a bit like the "birthday problem": it provides really good odds of finding someone to pin the crime on, but very little confidence that it was really them.
  - asdff 4y ago
    The birthday problem is different because there are 365 possibilities and every individual can only have 1 out of 365 states. For DNA there are over 300 million variants in the human genome and you can have different combinations of these variants. As far as matching 9 of 13 loci, thats not surprising to me nor concerning. If you left a sample at a crime scene it would match all 13 loci. Today the FBI uses at least 20 loci. On top of that, merely being a dna match does not mean you are going to jail. You need corroborating evidence just like how finger prints are used. The odds of a false positive that 1) is a dna match and 2) was in the right place at the right time and lacks and alibi and has a potential motive for the crime must be a trillion to one.
    - yetanotherloser 4y ago
      Birthday problem analogy comes up when trawling databases, not in single matches. Even in single matches you need to be careful not to apply the maths of combining independent probabilities on things that aren't really independent. As soon as someone starts quoting one-in-billions it is reasonable to fear that they might be trying to pull the "prosecutor's fallacy". Was the article I linked helpful?

asdff 4y ago

Can you site a source documenting this poor perfection of DNA matching? I dug into this after reading your comment and I found the opposite was true, that a false positive as an odds on the rate of 1 in 1 billion. When you couple it with the fact you also need more than just DNA evidence to convict (which was done with the golden state killer, DNA was not the only source of evidence), the odds grow even smaller of 1) you having no alibi or potentially being at the place of the crime at the time and more evidence, and 2) you being that 1 in 1 billion match to the real killer. https://www.parliament.uk/documents/post/postpn258.pdf https://www.parliament.uk/documents/post/postpn258.pdf

yetanotherloser 4y ago
pop article and still too forgiving, but this is not a bad start: https://freakonomics.com/2008/08/are-the-fbis-probabilities-about-dna-matches-crazy/ https://freakonomics.com/2008/08/are-the-fbis-probabilities-... (it's still excessively charitable about the odds because the genetic factors are not really perfectly independent and even a tiny bit of interdependence wreaks havoc with large probability calculations) The link you cite with "1 in billions" odds is exactly the sort of misrepresentation I'm talking about. Combine smaller-than-pretended probabilities with big database trawling and it becomes a bit like the "birthday problem": it provides really good odds of finding someone to pin the crime on, but very little confidence that it was really them.
- asdff 4y ago
  The birthday problem is different because there are 365 possibilities and every individual can only have 1 out of 365 states. For DNA there are over 300 million variants in the human genome and you can have different combinations of these variants. As far as matching 9 of 13 loci, thats not surprising to me nor concerning. If you left a sample at a crime scene it would match all 13 loci. Today the FBI uses at least 20 loci. On top of that, merely being a dna match does not mean you are going to jail. You need corroborating evidence just like how finger prints are used. The odds of a false positive that 1) is a dna match and 2) was in the right place at the right time and lacks and alibi and has a potential motive for the crime must be a trillion to one.
  - yetanotherloser 4y ago
    Birthday problem analogy comes up when trawling databases, not in single matches. Even in single matches you need to be careful not to apply the maths of combining independent probabilities on things that aren't really independent. As soon as someone starts quoting one-in-billions it is reasonable to fear that they might be trying to pull the "prosecutor's fallacy". Was the article I linked helpful?