Wiki143:Articles for deletion/Accuracy Paradox

This page is an archive of the proposed deletion of the article below. Further comments should be made on the article's talk page, if it exists; or after the end of this archived section. The result of the debate was DELETE. jni 29 June 2005 07:13 (UTC)

Accuracy Paradox

At best, this is original research. No relevant Google hits for this phrase. Eric119 00:03, 22 Jun 2005 (UTC)

• Delete nonsense. Sounds like someone has misunderstood an A-level statistics lesson. The 90%/10% mean that, given you tested positive, there's a 90%/10% chance that you do/don't have it. It does not say outright how often the test reports positive/negative, and it does not say that there is a 10% chance of testing positive when you're in the clear. See bayes theorem, and you'll see the article is missing a piece of information. -Splash 01:47, Jun 22, 2005 (UTC)
• Delete. Original research. --BaronLarf 01:52, Jun 22, 2005 (UTC)
• Delete. Utter nonsense. What User:Splash said. -Hmib 03:33, 22 Jun 2005 (UTC)
• Delete. Original research? Possible neologism (zero relevant hits on Google), and what Splash said. Crush with giant killer robots. --  B.d.mills  (T, C) 05:23, 22 Jun 2005 (UTC)
• Delete, elementary statistical inference poorly explained; I've never heard it called a paradox. Peter Grey 05:49, 22 Jun 2005 (UTC)
• The Mayo Clinic called it a paradox, and they seem to have had some trouble figuring it out.... "When symptomatic patients with abnormal electrocardiographic stress test responses first came to be referred for coronary angiography, the proportion with disease-the predictive accuracy of an abnormal test response-was noted to be very high. But when the identical testing procedure was later extended to asymptomatic subjects, the proportion with disease was surprisingly low. This puzzling paradox was eventually resolved through Bayes' theorem...". [1]. Kappa 06:35, 22 Jun 2005 (UTC)
• The article is nothing but elementary statistics. Of course, nobody says that a test with 10% false negative rate AND 10% false positive rate is any useful for testing for a disease which, at average, appears in one out of 1000 persons tested. Delete. - Mike Rosoft 08:25, 22 Jun 2005 (UTC)
  • So this web page [2] is wasting space discussing a concept that everyone understands intuitively? Kappa 08:45, 22 Jun 2005 (UTC)
• Delete. An explanation of the (more or less) counter-intuitive consequences of conditional probability could be useful, though. Leibniz 10:48, 22 Jun 2005 (UTC)

• Merge and redirect: Notice that this article has both capital letters. It's an article on one particular paradox. Whether other people waste space or not is not our concern. Med student tutorial pages are also not our concern. The question is whether this particular phrasing is a famous enough term that a person needs an encyclopedia to explain it and will search only by this phrase. I don't see much popularity for this phrase. Therefore, a mention of it in paradox and in conditional probability would be sufficient. Have the redirect point to the latter. Geogre 11:40, 22 Jun 2005 (UTC)
  • Well, it's not even a paradox - it's just a misunderstood point of (conditional) statistics.-Splash 12:30, Jun 22, 2005 (UTC)
    • Paradox has the definition: "an apparently true statement or group of statements that seems to lead to a contradiction or to a situation that defies intuition". I believe this fits that definition. Kappa 12:52, 22 Jun 2005 (UTC)
    • Hmmm, I'm not sure which part of this pseudo-example is paradoxical. First, we should ignore the example in the article since it is incomplete and mathematically wrong. The Smurfs example on your link is (one of) the correct formulation(s) - note it has an additional piece of information. But even in that example, if you improve the true-positive rate from 99% to 99.9% (you improve the test), the chances that the smurf is ill, given that they test positive also goes up (to 20.1%) which is hardly counter-intuituve. If you increase the false-positive rate to say 3% i.e. you make the test worse (with true +ve still 99%), then the chances that the smurf is ill, given that they test positive, fall to 14% — again hardly surprising. If it were a paradox, something would go 'the wrong way'. It is perhaps surprising that a 99% accurate test only picks up 20% of actual cases, but there is nothing paradoxical about it. Perhaps an example on the conditional probability page is in order, but certainly not the malformed one of this article, and certainly not under a 'paradox' heading.-Splash 13:37, Jun 22, 2005 (UTC)
      • The counterintuitive part is the test has a very low rate of false positives, but a large proportion of the positives are false. Kappa 14:03, 22 Jun 2005 (UTC)
      • Hmmm, perhaps. But the counterintuition only results from a cursory examination of the facts. Seeing as the OED more or less concurs with the def you give above, I suppose this could be loosely termed a paradox but, given the almost complete absence of the terminology in actual usage (including on Google), and the (mathematically correct) page that Cyan found below, I still think this page should be removed.-Splash 16:28, Jun 22, 2005 (UTC)
• Delete. As far as I can see the article is actually wrong, but I have never heard of the Accuracy Paradox and I would suggest this is a new coinage and not suitable for a heading in an encylopedia. If there is anything in it (I think not) it should appear elsewhere, eg conditional probability.
• Delete, insufficiently notable error. At most, an explanation correcting the misunderstanding could be merged to conditional probability. This is not a paradox nor a notable meme, even if one confused source once used the word. Barno 13:51, 22 Jun 2005 (UTC)
• Hmmm this error is mentioned under "false positive" "False positives can produce serious and counterintuitive problems when the condition being searched for is rare" but since it's non-notable maybe that should be taken out. Kappa 14:03, 22 Jun 2005 (UTC)
• Delete, this is already adequately explained in the correct place, Bayesian inference#False positives in a medical test. -- Cyan 15:49, 22 Jun 2005 (UTC)

• I'm glad you found that! Maybe the conditional probability, Bayes theorem and Bayesian inference bunch need some merging, but certainly not with the help of this article.-Splash 16:28, Jun 22, 2005 (UTC)

• Delete, had this been an actual paradox I may have said otherwise. Phoenix2 23:19, 22 Jun 2005 (UTC)
• Delete. Speaking as one in howevermany of course. hydnjo talk 00:43, 23 Jun 2005 (UTC)

The above discussion is preserved as an archive of the debate. Please do not modify it. Subsequent comments should be placed on a related article talk page, if one exists; in an undeletion request, if it does not; or below this section.