A question of sample size.

#1
Hi All,
I think I'm going to commit the cardinal internet forum sin and make my first post a possibly boneheaded question for people who have better things to do.... But here we go anyways:)

I have a CS background with a very minimal amount of stats and have a question related to internet gaming statistics. My question is:

How can I determine what a large enough sample size would be to determine the average win rate in a game for a single player over thousands of games without actually playing thousands of games. Is 100 games enough? This is a 15 vs 15 game. Assuming that players are of different skill levels on completely randomized teams and the outcome of the game is only affected by the skills of the individual players.

ex) player A plays 100 games, wins 55%. this is +/-X% of what that player would see if they were to play 10000 games.

What sort of statistics problem does this translate into and how can I represent it in notation and solve?

If this is too basic perhaps someone can point me toward a good resource for figuring this out myself?

Thanks!
Sean
 

rogojel

TS Contributor
#2
hi,
that would be a one sample p test. The sample size with your notation would be n= ((2/X)^2)*P*(1-P) where P is you original estimate of the win rate. The constant 2 is related to the confidence level, with 2 that would be the usual 95% confidence.

Alternatively you could use any sample size calculator available on the net to get the sample size.

regards
rogojel