Comments on: DNA and Guilt Redux http://xrlq.com/2008/05/28/dna-and-guilt-redux/ Politische Kommentare mit Snarkenremarken Mon, 14 May 2012 18:16:35 +0000 hourly 1 http://wordpress.org/?v=3.3.2 By: Xrlq http://xrlq.com/2008/05/28/dna-and-guilt-redux/comment-page-1/#comment-410803 Xrlq Mon, 02 Jun 2008 22:25:50 +0000 http://xrlq.com/?p=3412#comment-410803 James, while a positive result certainly <i>does</i> increase the chance that the suspect is guilty, the question is by how much, and whether a jury can be expected to assess it reasonably. Your reference to a "one in a million false positive rate," when the actual odds of a false match in this instance were 1 in 3, further illustrates my point. Both numbers, however, are tainted by the prosecutor's fallacy, which is why I'm arguing that even the 1 in 3 figure probably shouldn't be admitted, as it too implies a level of ex post facto certainty that simply isn't there. As to your suggestion for estimating the odds that the killer is in the database, I'm not sure how that is supposed to work unless you had a larger sample of killers from that time period. I do know that for a 1972 killer to have made it into a database that didn't even exist at the time. Robert Baker never made it into that database; who else didn't, either? If I had to bet on the odds of the original killer being in that database, I'd put those odds just south of 1 in 3, making a false match more likely than a true one. <blockquote>Note most such searches are run with more markers so false hits are less likely.</blockquote> True, but even then we still have to have at least a ballpark figure on the likelihood of the killer being in the database. Without having a decent idea of how likely <i>both</i> a true <i>and</i> a false match were, there is no way to assess which outcome is more likely to have occurred than the other, or by how much. It may not be an issue in the case of full match, assuming there wasn't anything unusually odd about the database making it nearly impossible for the killer to be in there. But when you start with odds that look like 1 in 1.1 million and are really only 1 in 3, it doesn't take a lot to overcome that. James, while a positive result certainly does increase the chance that the suspect is guilty, the question is by how much, and whether a jury can be expected to assess it reasonably. Your reference to a “one in a million false positive rate,” when the actual odds of a false match in this instance were 1 in 3, further illustrates my point. Both numbers, however, are tainted by the prosecutor’s fallacy, which is why I’m arguing that even the 1 in 3 figure probably shouldn’t be admitted, as it too implies a level of ex post facto certainty that simply isn’t there.

As to your suggestion for estimating the odds that the killer is in the database, I’m not sure how that is supposed to work unless you had a larger sample of killers from that time period. I do know that for a 1972 killer to have made it into a database that didn’t even exist at the time. Robert Baker never made it into that database; who else didn’t, either? If I had to bet on the odds of the original killer being in that database, I’d put those odds just south of 1 in 3, making a false match more likely than a true one.

Note most such searches are run with more markers so false hits are less likely.

True, but even then we still have to have at least a ballpark figure on the likelihood of the killer being in the database. Without having a decent idea of how likely both a true and a false match were, there is no way to assess which outcome is more likely to have occurred than the other, or by how much. It may not be an issue in the case of full match, assuming there wasn’t anything unusually odd about the database making it nearly impossible for the killer to be in there. But when you start with odds that look like 1 in 1.1 million and are really only 1 in 3, it doesn’t take a lot to overcome that.

]]> By: James B. Shearer http://xrlq.com/2008/05/28/dna-and-guilt-redux/comment-page-1/#comment-410784 James B. Shearer Mon, 02 Jun 2008 21:16:37 +0000 http://xrlq.com/?p=3412#comment-410784 3 5 Not sure why you think the test should not be admissable. Basically a positive result for the test greatly increases the chance that the suspect is guilty. If there was initially a reasonable chance (say 1% or more) that the suspect was guilty then a positive result of a test with a one in a million false positive rate makes guilt almost certain. If the initial chance was very low (say one in a million) then a positive result raises the chance to the 50-50 area. Either way you need some initial estimate of the suspects guilt to interpret the test. As for estimating the odds that the killer is in the database this can be done in the same way you can estimate the chances that a dead wife was murdered by her husband by looking at previous similar cases. This is not only cold case run against a database, you can use the other results to estimate the chance of a true hit. Note most such searches are run with more markers so false hits are less likely. 3 5

Not sure why you think the test should not be admissable. Basically a positive result for the test greatly increases the chance that the suspect is guilty. If there was initially a reasonable chance (say 1% or more) that the suspect was guilty then a positive result of a test with a one in a million false positive rate makes guilt almost certain. If the initial chance was very low (say one in a million) then a positive result raises the chance to the 50-50 area. Either way you need some initial estimate of the suspects guilt to interpret the test.

As for estimating the odds that the killer is in the database this can be done in the same way you can estimate the chances that a dead wife was murdered by her husband by looking at previous similar cases. This is not only cold case run against a database, you can use the other results to estimate the chance of a true hit. Note most such searches are run with more markers so false hits are less likely.

]]> By: Xrlq http://xrlq.com/2008/05/28/dna-and-guilt-redux/comment-page-1/#comment-409772 Xrlq Thu, 29 May 2008 15:38:35 +0000 http://xrlq.com/?p=3412#comment-409772 Initially, I did. Once I understood the prosecutor's fallacy problem, I concluded that a partial DNA match has no (measurable) probative value, and should therefore be excluded. The only exception would be if someone could find a credible way of assessing the odds that the killer is in the database. Initially, I did. Once I understood the prosecutor’s fallacy problem, I concluded that a partial DNA match has no (measurable) probative value, and should therefore be excluded. The only exception would be if someone could find a credible way of assessing the odds that the killer is in the database.

]]> By: nk http://xrlq.com/2008/05/28/dna-and-guilt-redux/comment-page-1/#comment-409768 nk Thu, 29 May 2008 15:16:21 +0000 http://xrlq.com/?p=3412#comment-409768 I misunderstood you, then. My argument at Patterico's was that the potential for prejudice outweighed the probative value at trial. I thought you thought that it should be admitted. I misunderstood you, then. My argument at Patterico’s was that the potential for prejudice outweighed the probative value at trial. I thought you thought that it should be admitted.

]]> By: Xrlq http://xrlq.com/2008/05/28/dna-and-guilt-redux/comment-page-1/#comment-409757 Xrlq Thu, 29 May 2008 14:14:35 +0000 http://xrlq.com/?p=3412#comment-409757 Not sure what our point of disagreement is. Since we can't tell how likely the killer was to be in the database, we have no way of knowing whether a "true" match to the killer himself is any more likely to occur than a "false" one to an unrelated individual who randomly matches 5 1/2 of his DNA indicators, let alone by how much. Not sure what our point of disagreement is. Since we can’t tell how likely the killer was to be in the database, we have no way of knowing whether a “true” match to the killer himself is any more likely to occur than a “false” one to an unrelated individual who randomly matches 5 1/2 of his DNA indicators, let alone by how much.

]]> By: nk http://xrlq.com/2008/05/28/dna-and-guilt-redux/comment-page-1/#comment-409654 nk Thu, 29 May 2008 04:04:36 +0000 http://xrlq.com/?p=3412#comment-409654 I think you were spot on with your lightning example. Otherwise, we have disagreed on this before, at Patterico's. Daubert rule notwithstanding, this should not be admitted as evidence at trial. Possibly for probable cause in a request for a warrant. An investigative tool only. I think you were spot on with your lightning example. Otherwise, we have disagreed on this before, at Patterico’s. Daubert rule notwithstanding, this should not be admitted as evidence at trial. Possibly for probable cause in a request for a warrant. An investigative tool only.

]]> By: Phelps http://xrlq.com/2008/05/28/dna-and-guilt-redux/comment-page-1/#comment-409598 Phelps Wed, 28 May 2008 22:45:13 +0000 http://xrlq.com/?p=3412#comment-409598 Too long to be published as written. Also, it includes math, which is too complicated for a journalist to understand, so they won't bother to read it. Too long to be published as written. Also, it includes math, which is too complicated for a journalist to understand, so they won’t bother to read it.

]]>