From: Daniel Toebbens gmx de> Date: 19 aug 2007 Subject: Re: PanAtlantic and Dojo24 [was: Call for data] -------- Original-Nachricht -------- Datum: Sat, 18 Aug 2007 19:33:10 -0400 Von: Larry Kaufman comcast net> An: shogi topica com Betreff: Re: PanAtlantic and Dojo24 [was: Call for data] =20 > There is an obvious flaw in simply using a linear=20 > equation to make these comparisons. Neither rating system=20 > allows for negative ratings. Therefore, any player with a=20 > Pan-Atlantic rating less than your intercept value=20 > (1190 overall, 1215 for Europe) would automatically=20 > appear underrated, even if his Dojo rating is just one=20 >(or zero).=20 Of course. But that just means that a player needs a strenght of about 1200= PanAtlantic points at least, before he can expect to get any points signif= icantly different from zero in the Dojo. For weaker players the Dojo simply= cannot be used as a gauge, since the typical entrance strength is too high= . > Since all of the American players in your=20 > sample are over 1500, this strongly biases the result of > the comparison to make Europe appear underrated. Not really. The linear average of the American sample is of course reliable= only for the range, in which the majority of data lie; that is 1200 - 1800= in the Dojo24 rating. For this region, all the European data points are b= elow or in the lower fringes of the American ones.=20=20 > I would suggest redoing the calculation with the=20 > intercept value assigned. It's probably not fair to use=20 > zero, as a beginner in Europe is weaker than a typical=20 > beginner in Dojo, but maybe an intercept of 400 > might be reasonable (this used to be the minimum, and now=20 > might be considered the rating of a novice but not a=20 > beginner). That's what I had expected initially, too. But the data points I have simpl= y do not support this. There are too many players with relatively high PanA= tlantic rating, who nonetheless have very low rating in Dojo24. And the slo= pe of the data does not allow for a linear relation with an intercept as lo= w as 400. This would not fit the data at all. I'll have to see about gettin= g some graphics set up somewhere.=20 Its possible to force the curve to go down at low ratings, using an allomet= ric function=20 PanAtlantic =3D a + b*Dojo24^c=20=20 and fixing 'a' to a low value. Using an exponent of 0.4 gives reasonably go= od results over the whole data range. The resulting curves are PanAtlantic =3D 400 + 81.5*Dojo24^0.4 for all data,=20 PanAtlantic =3D 400 + 76.2*Dojo24^0.4 for European data,=20 PanAtlantic =3D 400 + 86.4*Dojo24^0.4 for American data,=20 However, the American sample is dominated by the lists I got from Larry an= d George. There might be some selective effect in this, which distorts the = data. The fact, that they are all within a comparable narrow rating range i= ndicates this. I really would like some data from e.g. the Ohio players.= =20 > I should also tell you to remove the datapoint that=20 > shows a Dojo rating of 2350, as I happen to know that this=20 > handle is often used by visiting pros and so is grossly=20 > inflated.=20 This of course just increases the difference. This data point was one of th= e two outliers in the American data set, for which the PanAtlantic rating w= as notably low. The other one was from Junki Saito, who only played 6 games= in one PanAtlantic tournament (and does not appear any more in the newest = rating list). Without those, the American linear average is PanAtlantic = =3D 1140 + 0.59*Dojo24.=20=20=20 --=20 Psssst! Schon vom neuen GMX MultiMessenger geh=F6rt? Der kanns mit allen: http://www.gmx.net/de/go/multimessenger --^---------------------------------------------------------------- This email was sent to: = bobd science uva nl EASY UNSUBSCRIBE click here: http://topica.com/u/?a2i6Ys.= a7fLI9.= Ym9iZEBz Or send an email to: shogi-unsubscribe topica com For Topica's complete suite of email marketing solutions visit: http://www.topica.com/?p=3DTEXFOOTER --^----------------------------------------------------------------