I'm trying to come up with a stat which can be used to properly distinguish consistent players from inconsistent players. I'm not referring to "consistent effort", but rather those players who you can count on to be "consistently productive" game to game. But I'm having a tough time precisely defining it. Here are a couple ways: (1) The more consistent the player is, the less variation there is in his stats, game to game (2) The more consistent the player is, the less chance there is that he'll have a "poor game" (say, for instance, 50% of his usual production level) What does consistency mean to you? What I'd ultimately like to do so is rate player consistency, and then examine how it impacts winning games.
Consider the following examples of game performances over the course of 80 games -- A, B, C, and D. How would you order them, from least consistent season to most consistent season?
I would put: (most consistent) D, A, C, B (least consistent) And for me consistency is keeping your stats for all the season, maybe with a +/- 2 margin
That order makes sense. The difficulty is in deciding between D and A as more consistent. For A, the player plays at the exact same level for half the season, than plays at a much higher level for the second half of the season. If you use standard deviation (the typical of way of measuring variation in a sample) A and B have the same "consistency". Instead, the metric I'm considering is average difference from the averages over the last n games. So, using n=3 for instance, I get A (0.52), D (0.66), C (1.76), B (13.3). But it isn't even that simple. Problem is players who play less minutes will naturally end up having less "consistent" numbers, game to game, simply due to random variation. Actually, you could extend that to say players who are "less involved" in games, on average, will end up being less consistent. So, you have to somehow take into account the average involvement of a player per game. The method I'm looking into is simulating a player's season many times, using the exact same minutes he got each game. I randomly generate his box score stats each game based on the minutes played, his per-minute stats for the whole season, and his 2fg%, 3fg%, and ft%. Then, I take the average "consistency" of all these simulated seasons, and compare that to how consistent his season actually was. Seems somewhat complicated, but it looks like it could yield some interesting results. Is Stromile Swift really as inconsistent as people make him out to be, or is the apparent "inconsistency" simply due to his lack of minutes? Hopefully, I'll have a good answer to that question shortly.
