Author: Martin Schubert
Date: 08:57:51 05/27/02
Go up one level in this thread
>My point is that there is a meaning for the probability that we believe that X >is better than Y. > I'm interested in one statistical model where you can calculate probabilities for the unknown paramteter. A statistical model which makes sense. Can you explain one? >If you need to make a guess which program is better, then there is a meaning to >the question what is the probability that you believe that you are right. > That's something you can't catch with statistics. The fact that you don't know the rating doesn't mean it varies. You can only talk about probabilities for things which happend by chance. Let's take a look at the easiest statistical test. How good is a medicine? A medicine helps with probablity p. You don't know p. The idea of statistics is now to get an estimation for p without further knowledge. Statistics can provide that. So what would you do? What kind of distribution would you assume for the p like you do it for the rating of Fritz? >Let assume for the discussion that the rating of Fritz is distributed normally >with average of 2720 and standard deviation of 50. >Let assume the same is for tiger except average 2700. > >These assumptions are before seeing games because we need to assume apriory >distribution and it is logical to assume that Fritz is probably better based on >previous expereince when Fritz was the program that was leading the ssdf list >more often than Tiger. > >let use the variable X for the rating of Fritz and the varaible Y for the rating >of Tiger. > >X and Y are constants but for us they are not constants because we do not know >them. > They are constants or they are not. If you keep them constant in your model you can't give probabilities for them. >We ask the question what is the probability that X>Y. We can't. > >Before having games we can say that it is slightly more than 1/2 based on our >assumptions. > >After having games we can have better knowledge about X and Y and >we may change our opinion. > >Uri Martin
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.