Cool data mining stuff:
Sent: Wednesday, June 22, 2005 8:13 PM
To: JDM Society List
Subject: [Jdm-society] Roles and decisions in conflict situations
It is often useful to predict the decisions that people will make when they are involved in conflict such as wars, takeover battles, and union-management disputes. In research on how best to make such predictions, Scott Armstrong and I obtained forecasts for eight real and diverse conflicts. We provided the participants in our research with between three and six plausible decisions for each conflict. By choosing at random from among the decisions, one would expect to be 28% accurate.
Here are our findings on the percentage of correct forecasts from four different methods:
Role-play simulations using novices 62%
Experts' structured analysis of analogies 56%
Experts' unaided judgments 32%
Game theorists' judgments 31%
(Chance 28%)
People involved in conflicts are often advised to "Stand in the other person's shoes", in other words, to think hard about the roles of the other parties. Are list members aware of any evidence that this approach improves the accuracy of predictions?
Suppose I asked experts to indicate, for each party in a conflict, which decision the party would prefer, why it would be preferred, how the party would try to achieve the preferred decision, and to assess the chances that they might achieve it. Having done that exercise, how accurate would you expect the experts' forecasts of the actual outcomes of the eight conflicts used in the above research to be on average? [____%]
If I asked novices (undergraduate students) to do this, how accurate would you expect their forecasts to be? [____%]
As well as answers to the previous questions, I'd appreciate any suggestions on how best to get people to "stand in the other person's shoes" so as to obtain the most accurate forecasts of their decisions.
Kesten Green
----------------------------------------------------
Dr Kesten C Green, Business and Economic Forecasting Unit, Monash University
www.conflictforecasting.comContact at home: P: +64-4-976-3243; M: +64-21-456-516
Jdm-society mailing list
Jdm-society@mail.sjdm.org
http://www.sjdm.org/mailman/listinfo/jdm-society
"Do not believe in anything simply because you have heard it. Do not believe simply because it has been handed down for many generations. Do not believe in anything simply because it is spoken and rumored by many. Do not believe in anything simply because it is written in Holy Scriptures. Do not believe in anything merely on the authority of Teachers, elders or wise men. Believe only after careful observation and analysis, when you find that it agrees with reason and is conducive to the good and benefit of one and all. Then accept it and live up to it." The Buddha on Belief, from the Kalama Sutta
AZMY data mining tools http://www.azmy.com/
http://www.datamininglab.com/toolcomp.html
http://www.xore.com/prodtable.html
http://www.cs.uvm.edu/~xwu/PPT/SEKE-02.ppt
glossary www.twocrows.com/glossary.htm
Do the observations come from a particular distribution? http://www.itl.nist.gov/div898/handbook/prc/section2/prc21.htm
NIST statistics handbook http://www.itl.nist.gov/div898/handbook/
stat course http://www.mnstate.edu/wasson/ed602.htm
stat package http://www.statistixl.com/
mining blog data http://www.pewinternet.org/ppt/BUZZ_BLOGS__BEYOND_Final05-16-05.pdf
regression tutorial
http://mtsu32.mtsu.edu:11308/regression/level3/multicorrel/useexcel.htm http://www.statsoft.com/http://www.hs.ttu.edu/hdfs3390/hothand.htm
http://portal.acm.org/citation.cfm?id=956849
Transformations are widely used in statistics to reduce data to standard forms. Some common methods of re-expressing data are as follows:
Centering -- The sample mean (column mean) is subtracted from the data values in order to obtain centered ``anomalies'' having zero mean. All information about mean location is lost.
Standardizing -- The data values are centered and then divided by their standard deviations to obtain ``normalized anomalies'' (meteorological notation) having zero mean and unit variance. All knowledge of location and scale is lost and so statistics based on standardized anomalies are unaffected by any shifts or rescaling of the original data. Standardizing makes the data dimensionless and so is useful for defining standard indices. Also note that correlation coefficients are unaffected by any linear transformations such as standardization.
Normalizing -- Normalizing transformations are non-linear transformations often used by statisticians to make data more normal (Gaussian). This can reduce bias caused by outliers, and can also transform data to satisfy normality assumptions that are assumed by many statistical techniques.
FLORIDA JUDGES THROW OUT DUI CASES FOR LACK OF SOURCE CODE FOR BREATH TESTERS http://tampatrib.com/floridametronews/MGBUBJ5QK9E.html
[The interesting thing is that if the same principle was applied to computerized voting, we'd have to rerun two presidential elections. Anyone for a class action suit?]
TAMPA TRIBUNE - Hundreds of cases involving breath-alcohol tests have been thrown out by Seminole County judges in the past five months because the test's manufacturer will not disclose how the machines work. All four of Seminole County's criminal judges have been using a standard that if a DUI defendant asks for a key piece of information about how the machine works - its software source code, for instance - and the state cannot provide it, the breath test is rejected, the Orlando Sentinel reported Wednesday.
Prosecutors have said they do not know how many drunken drivers have been acquitted as a result. But Gino Feliciani, the misdemeanor division chief in the Seminole County State Attorney's Office, said the conviction rate has dropped to 50 percent or less.
-----Original Message-----
From: jdm-society-bounces@mail.sjdm.org [mailto:jdm-society-bounces@mail.sjdm.org]
On Behalf Of Reifman, Alan
Sent: Wednesday, May 25, 2005 7:08 PM
To: spsp-discuss@stolaf.edu; jdm-society@mail.sjdm.org
Subject: [Jdm-society] new book "freakonomics"
Many of you have probably already heard of the new book "Freakonomics" (or perhaps have even already read it). It is written by University of Chicago economist Steven Levitt and journalist Stephen Dubner. I, like many Americans, first heard of Levitt in an August 3, 2003 New York Times magazine profile of him, written by Dubner. Here's a message I sent to the SPSP list in 2003 shortly after seeing the Levitt profile:
What has made Levitt stand out is his unconventional scholarly portfolio. He, by his own admission, has limited grasp of many traditional economic ideas, is not associated with any theories in economics, and does research on what seem to be unusual topics. In fact, Levitt recently characterized himself as being "adisciplinary," in his blog (see below).
The talent he has, though (in addition to an amazing work ethic -- he can sure crank out the papers), is in figuring out clever ways to design analyses of archival data to address interesting questions.
http://www.src.uchicago.edu/users/levit/recentpublications.htm
Do Sumo wrestlers throw matches? Do real estate agents go the extra mile to help their clients get the best possible deal in selling their homes? Might the legalization of abortion (starting in certain states in the late 1960s and culminating in the 1973 Roe v. Wade decision nationally) be a major factor in the early 1990s crime drop? Were contestants of certain demographic groups discriminated against on "The Weakest Link"?
In all of these cases (and others), Levitt and his collaborators came up with ingenious comparisons to test within the relevant datasets that would go a long way toward answering the questions. Some of the comparisons I could anticipate while reading the scenarios, but most I could not.
Beyond the empirical analyses, however, the storytelling is also spellbinding in places. Two examples, in particular, are the story of how one individual's strategic use of the Superman radio show helped eviscerate the Ku Klux Klan, and of how one of Levitt's colleagues ended up embedding himself in a Chicago crack-dealing gang as part of his research studies.
At about 200 pages, Freakonomics is a pretty quick read, one that I largely found exciting (there were a couple of studies that I thought were less compelling than others, however).
The aforementioned study linking abortion to later reductions in crime has been especially controversial. I've seen Levitt and his collaborator on that study, John Donohue, discuss this matter in various venues. The point they make, as I see it, is that the central concept of the study is not abortion, per se, but rather reducing the incidence of unwanted pregnancies and poorly cared for children, which can be accomplished by many non-abortion means, such as sex education, abstinence, contraception, parent education, adoption, etc.
The authors also maintain a blog related to the book (including reader comments), so there's opportunity for extended discussion on many of the issues raised in the book: http://www.freakonomics.com/blog.php . Levitt has even introduced new topics on the blog that were not in the book, including one that is near and dear to my heart -- the "Moneyball" approach to using statistics in sports decision-making (Levitt does not dispute that Oakland A's GM Billy Beane was able to compile play-off-caliber teams for many years with a much lower payroll than other teams; Levitt's contention is that the A's did NOT employ game-management strategies that were appreciably different from those of other teams).
One last thing: I want to state for the record that on May 11, 2001, more than two years before ever hearing of Levitt, I sent a message to the SPSP list suggesting the use of "The Weakest Link" to study discrimination:
As I've noted, however, the difference between Levitt and me is that he actually goes ahead and DOES these studies!
*********************************************************
Alan Reifman, Ph. D., Associate Professor
Dept of Human Dev't and Family Studies
College of Human Sciences
Texas Tech University
Lubbock, TX 79409-1162
(806) 742-3000
http://www.hs.ttu.edu/hdfs/Faculty/reifman.htm
Modeling the Internet and the Web: Probabilistic Methods and Algorithms by Pierre Baldi, Paolo Frasconi, Padhraic Smyth, Pierre Baldi