From akriman@darwin.helios.nd.edu Fri Feb 7 10:43:36 2003 Received: from mxu2.u.washington.edu (mxu2.u.washington.edu [140.142.33.7]) by lists.u.washington.edu (8.12.1+UW01.12/8.12.1+UW02.12) with ESMTP id h17IhZCK086124 for ; Fri, 7 Feb 2003 10:43:35 -0800 Received: from mailspool.helios.nd.edu (mailspool.helios.nd.edu [129.74.250.7]) by mxu2.u.washington.edu (8.12.1+UW01.12/8.12.1+UW02.12) with ESMTP id h17IhXQY027394 for ; Fri, 7 Feb 2003 10:43:33 -0800 Received: from darwin.helios.nd.edu (darwin.helios.nd.edu [129.74.250.114]) by mailspool.helios.nd.edu (8.9.2/8.9.2) with ESMTP id NAA09377 for ; Fri, 7 Feb 2003 13:43:32 -0500 (EST) Received: (from akriman@localhost) by darwin.helios.nd.edu (8.12.6/8.12.6/ND-cluster) id h17IhWSJ017576 for classics@u.washington.edu; Fri, 7 Feb 2003 13:43:32 -0500 (EST) Date: Fri, 7 Feb 2003 13:43:32 -0500 (EST) From: Alfred M Kriman Message-Id: <200302071843.h17IhWSJ017576@darwin.helios.nd.edu> To: classics@u.washington.edu Subject: Re: If the Delphic Oracle had a website... X-Uwash-Spam: Gauge=III, Probability=3%, Report="QUOTED_EMAIL_TEXT, SPAM_PHRASE_03_05" > J.F. Gannon wrote > I came in 100% Aquinas, 90% Augustine, 83% Aristotle. This result is at > least consistent. But I rather suspect that there is something inconsistent > about being 100% Aquinas and 96% Kant, not that you are inconsistent, but > that the test is defective. I found some of the formulations ambiguous, but > was reluctant to keep choosing the last option in all of those instances. Much of the strangeness in the results has to do with the normalization. I suspect that most people's answer sets result in a poor match to any of the philosophers. The raw value of the correlation of any test-taker's views with any named philosopher (or rather his or her internal test representation) is probably typically no more than 20%. It is very easy to be in 20% agreement with both of two philosophers who are in substantial disagreement. The Aquinas/Kant numbers which JFG observed are inconsistent were earned by Elizabeth Vandiver: > And I, too, come out as 100% Aquinas, and 96% Kant. And here I always > thought I *disagreed* with Kant. Strange. ... Suppose that "in reality" -- i.e., according to EV's answers to the inadequately specified questions -- she is in 12.5% agreement with the selector's internal representation of Aquinas, and in 12% agreement with the representation of Kant. This is possible even if EV is largely in disagreement with Kant, and likewise Kant with Aquinas. EV might feel that her agreement with Aquinas should be higher, but disagreements she considers not essential nevertheless have their weighting. What is essential for one philosopher is not essential for all, so it is tricky to establish appropriate weights, even assuming the philosopher's position can be and has been understood. Also, as JFG noted, the none-of-the-above catch-all is tempting. This not only lowers raw correlations over-all, it also leads to a bullet-vote effect: a question that is unusually clearly formulated, even though it is not so important, may influence the result more than a poorly-worded question for which people choose the catch-all. Since the output numbers are normalized up to a maximum of "100%" (presumably linearly), EV's would (under assumption) be scaled up by a factor of eight and she would be graded "100%" Aquinas and "96%" Kant. Meaningless fun, with stress on the meaninglessness (in the first pass I was "100%" Sartre). Diana Wright observed earlier: > Retake the test & change one little answer. That flipped me > from Sartre to Augustine. This too is an artifact of the small overlaps. Changing one answer might change a couple of philosophy description vectors by a small amount each, but when the agreement in all cases is in the cellar, the small difference among small numbers is magnified. We don't have mountains of information here, just mole-hills under the magnifying glass. DW simply moved a grain from one to the other. There is a whole science of test-writing, and these "selectors" are constructed by amateurs. AMK .