Author: Chris Carson
Date: 16:48:12 05/31/02
Go up one level in this thread
On May 31, 2002 at 17:11:43, Dann Corbit wrote: >On May 31, 2002 at 07:16:45, Chris Carson wrote: > >>Dann, >> >>If you believe that ratings inflation exists, then do the stats/research and >>present a proof. > >There is no question that the ratings have moved. >Since the data pool constantly changes, there is no point in trying to make a >research experiment because it is not repeatable! >In other words, you could do a simulation, but it is impossible to show exactly >what is going on. Not a simulation, but actual hypothesis testing. Null: No ratings drift, H1: Ratings drift. This is a simple experimental design, you can use every list published by FIDE as input data. This is really a simple, I do this type of experiment everyday (this would be a simple test with the data already supplied by FIDE). :) > >>It may exist and there is plenty of data to look at. If you >>can show this, it would prove your point. You should be able to establish >>averages for today, 5 years back, 10 years back, 20 years back. Show N >>population, break out #GM's, #IM's, ... Show the std deviation. If there is >>inflation, you can show it to be significant and establish a 95% confidence >>level. You can show the rate of inflaciton over time, factors that influence >>the inflation (based on data not opinion) and make recommendations for >>correcting this. It would be impressive. The records are there, you may have >>to dig them out, but I "know" the data is there. Spend some time, and then >>please present your findings, I would really like to see what you find and what >>you have to say after you discover it. > >Trying to show exactly what is happening is like herding cats. Get the darn >thing to hold still, and maybe you could formally prove something. Not at all. Simple design. Simple two tailed analysis (one tailed if only checking for inflation). I have research assistants that can set this up and do the analysis, it is routine, but I am skeptical and see no practical use in testing this, if you think there is significant drift (use 95% confidence level), then test it. The hardest part is getting the FIDE lists, but they are all there if you look for them, the stats are easy, Excell or SPSS will do. > >Mostly, it is not possible to do better than educated guesses. I disagree here. > >Drift up or down does not matter anyway. As long as enough games are pumped >through the system, the distance between players should remain relatively >accurate and that is all that matters. We agree, there is most likely no significant difference.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.