[note: this probably belongs in qbtheory] In response to dlgood (below) I believe it can be done with regression analysis and a suitably large sample (possibly requiring multiple tournaments). Each observation is a team's score in a match The dependent variable = Team Total Points in Match The independent variables are Dummy variables [1,0] to indicate whether a particular player was on the team, we would thus need 1 dummy variable for each player. Dummy variable [1,0] for each opposition team Dummy variable [1,0] for pack author Some additional requirements team compositions must change across the data set. I.e. we cannot have collinearity among members (e.g. Jim Dendy and Albert Whited must appear separately from each other some of the time). This will only happen if a team rotates members within or between tournaments. Opponents must be seen multiple times (though not necessarily by the same team) Authors must be read multiple times (though different rooms can take this into account) >From this (the coefficient's on the dummy variables for players) we can see each player's individual contribution to the team's score (in total points (combining tossups and bonuses), while controlling for opponent and pack. Comments? Anyone want to try it out? By:dlgood99 Date:10/11/01 9:30 am I think the key way to going about this is not to look at what stats we have at our disposal and make a formula, but rather to determine what the criteria for good players and find statistics that can demonstrate them. Here is what we are looking for: (averaged on a per 20 tossup basis) Positives: Points contributed on tossups Power Tossups Points contributed on bonuses Negatives: Interrupts Factors to control for: Quality of opposition (strength of schedule) Quality of teammates (shadow effect) Difficulty of question sets played on Tournament format Subject area Things to discount: Pickoffs, bouncebacks, or whatever you call them I would suggest that getting an accurate statistical ranking of players that controls for each of these factors within a reasonable margin of error is probably too difficult to do. At a minimum, it would require scorsheets from each game, an analysis of the average per packet score. In my opinion, the requirements are too complex to do this through statistical analysis. If somebody can come up with a model for this, I'd be interested in seeing it.
This archive was generated by hypermail 2.4.0: Sat 12 Feb 2022 12:30:45 AM EST EST