Dennis Roberts wrote <<< The other inherent problem here is that IF in fact these 600 are all topnotch ... there will not be THAT much variation amongst them on a variety of factors ... winning records ... etc. This relative lack of differentiation amongst the players will make it hard for ANY method to
uncover differential facets or dimensions of their behavior ... >>> I'm not sure of this. A couple points: First, even if this were the top 600 players, that would cover quite a range of skill. Second, the relative equality of skill may make it easier to uncover other factors, not harder. If an amateur plays a grandmaster, style is irrelevant. Any grandmaster will beat any mediocre player so often (nearly 100% of the time) that other factors cannot affect the outcome. OTOH, if two players are equal, then ONLY other factors can affect the outcome. Indeed, it might be useful to limit the dataset to the top 100 players. Not only would this increase N per matchup, but it would control for the factor that isn't interesting here: Skill. The problem would be that style may have different effects for people of different skill, but I don't see any way to deal with this, simply because data on the styles of lower-rated players is unlikely to exist. Peter . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
