It’s Not Accuracy They Want…

Jason Kuznicki

Jason Kuznicki is a research fellow at the Cato Institute and contributor of Cato Unbound. He's on twitter as JasonKuznicki. His interests include political theory and history.

Related Post Roulette

4 Responses

  1. Boonton says:

    I hate to be a nudge but after tutoring statistics off and on for years the ‘coin’ analogy here is a bit irksome.

    Let’s say out of 100 cars 10 are carrying drugs. Now imagine a dog who barks at 5 cars carrying drugs and not at any of the others. We’d say casually this dog is ‘always right’, is 100%. But he isn’t.

    Likewise consider a dog that barks at 20 cars catching the 10 that have drugs but also barking at 10 other ‘false positives’. The article says this dog is no better than flipping a coin since he is wrong 50% of the time. But he isn’t. He catches 100% of the drug smuggling cars. A coin flip would generate 45 false positives, 5 true positives catching only 50% of the smugglers.

    I’m not saying a dog with so many false positives is acceptable but we should keep both sides of the coin in mind. The 2nd type of dog may be more desirable when the stakes are much higher, say a bomb sniffing dog where you want zero people with bombs to get through.

    In terms of probable cause, the 2nd type of dog has lots of false positives BUT that doesn’t mean his barks are no better than coin flips. If he barks at you the odds are still greater that you’re carrying drugs.Report

    • Boegiboe in reply to Boonton says:

      So, the piece of information we lack is: What percent of traffic stops did dogs trigger off of? Your examples may be in the right ballpark, but we can’t assume that. Drug-sniffing dogs don’t sit in every traffic police vehicle. They may be summoned when the traffic cop suspects something is fishy. So, drug dogs may be triggering on more than 50% of all stops. The article quotes Justice Souter writing that some dogs in testing do as poorly as identifying 60% of clean cars has containing drugs. (This would be even more useful information if we knew how many true positives these hyper-triggering dogs detected.)

      Anyway, you’re absolutely right that the dogs may be doing much better than coin flips. But your last sentence is incorrect: You don’t know that the vast majority of all sniffs result in negatives from the dogs, and this assumption is necessary for that statement to be true.Report

  2. DensityDuck says:

    And in point of fact we have “human-hunting dogs”–scent trackers.Report