Which Atlanta Dream Players Are Most Instrumental to Team Wins?

This is a question that can be approached from several angles. There are several standard arguments:

1) The player that provides the nebulous quality of "leadership".
2) The player who has the best statistics.
3) The player that has the best performances in breakout games.
4) The player with the best "clutch" performance.

Most of the "best player" talk centers on #2 - the best stats. #1, #3 and #4 are to some extent, unquantifiable. What is "leadership"? How do you define "clutch" in basketball? What kinds of games are breakout games? Generally, the idea is that you create some sort of metric that determines the excellence of the player. Some of these metrics even have names like "Win Shares" or "Boxscore Value" - metrics which attempt to answer the question by claiming, "this player was worth X number of wins".

I decided to take a new approach (*) - instead of counting points, or rebounds, or whatever, I came up with the following definition:

"A player is instrumental to a team's wins if
a) there is a strong relationship between the player having a good game and the team winning, and
b) there is a strong relationship between the player having a poor game and the team losing."

After all, who cares about what kinds of stats you have if your team doesn't win any games? Or if all of your big games come during team losses. I had to come up with two measures: team success and player success, and determine if there is a relationship between the two.

The measure for team success is easy: wins and losses. A team earns one point for a win, and zero for a loss. We assume that it makes no difference if you win by two or win by twenty, and likewise for losses.

For player success, I used the Boxscores metric created by David Sparks. (If you want to know about the specifics, you can do a Google search.) Boxscores is a "linear metric", meaning that it gives you points for doing good things and takes away points for doing bad things. It is cumulative, and is not a "per minute" metric - it rewards players who play a lot.

For each game, a Boxscore value was calculated for a player. This would be problematic for four players - Nikki Teasley, Tamera Young, Ivory Latta and Armintie Price. Teasley was waived, and Latta was signed to replace her. Young was traded to the Sky for Price. I decided to give players who only played part-time with the Dream values of "zero" for games in which they were not with the team. Even for these part-time players, we can still measure the impact they had - did the team win when they started showing up, or when they left?

Finally, I used the statistical measure of correlation between an individual player's Boxscore values and team wins. (Trust me, you don't want me to explain correlation.) The strength of correlation is that it always yields a number between -1 and 1. A correlation = 1 implies a strict one-to-one relationship in one direction: A goes up when B goes up and vice-versa. A correlation = -1 implies a strict one-to-one relationshp in the opposite direction: A goes up when B goes down, and vice-versa. A correlation = 0 implies that A and B have no relationship.

So out of the thirteen Dream players, here are the players whose Boxscore values during a game had the highest correlation with whether or not the team won that game:

Holdsclaw 0.448
Lyttle 0.420
Lehning 0.404
Teasley 0.247
McCoughtry 0.214
Snow 0.137
Miller 0.119
Lacy 0.102
Latta 0.057
Castro Marques 0.005
De Souza -0.046
Price -0.121
Young -0.309

I don't know how strong the correlation has to be before it's significant, but I'm going to assume that any correlation > 0.3 is at least a medium corrlelation.

At the top is Chamique Holdsclaw. Whenever Chamique had a good game, it was likely that the Dream would win - and it was likely that when she had a bad game (or didn't play) that the team would lose. During the last 11 games of the season - when Holdsclaw's knees finally caught up with her - the Dream went 4-7, losing four of its last five and getting swept in the playoffs by the Shock.

Up next is Sancho Lyttle. By any standard of "breakout", Lyttle had a breakout year, named to the All-Star Game for the very first time. It was also true that the Dream were most likely to win when Lyttle had a good game.

The suprising third person is Shalee Lehning. This surprised me. Lehning is the only other player whose performances correlate with team victories. Think of the playoffs between the Dream and the Shock. Holdsclaw played in the first game but was clearly playing through the injury and contributed nothing. Lehning was out for the season with a severe shoulder separation. Lyttle played, but had two substandard games. Our correlation metric suggested bad news for the Dream, and sure enough, the Dream played close in the first game (but lost) and horribly in the second.

For Lehning detractors - and there are a lot of them - we need to remember that correlation only indicates how two variables change in sync - but there might not be causation. Both variables, for example, could be related to some other hidden variable. It could be that Lehning's strong correlation with team wins could be that Marynell Meadors kept her in when the team was winning which allowed her to build her Boxscore value during the wins.

For some players - Jennifer Lacy and Ivory Latta - their game performance correlate with team wins at a level no better than random chance.

What's very suprising to me are the low correlations presented by Iziane Castro Marques and Erika de Souza - de Souza actually has a marginally negative correlation. It seems that whether or not those two players were hot didn't have much to do with whether the team won or not. Then again, you could simply make the reverse argument that was made for Lehning - that when the team was clearly winning, the players were pulled from games which adversely affected their ability to correlate wins with good Boxscore values.

Finally, the cases of Armintie Price and Tamera Young. For each player, good performances were correlated with the team losing, not winning. Most likely, in each case, the player was kept in for garbage time when the team was losing. Price's only two good games this year with the Dream were both losses - the first was a 91-84 loss in Seattle where she scored eight points and had three rebounds. The second was in the final game of the year, the playoff loss at Gwinnett Arena, which was her best game of the year by Boxscore value - she had 10 points, six rebounds and three assists in a game where the Dream had nine players and were desperately trying to find someone with a good game.

As for Young, I suspect that her good performances are strictly related to garbage-time situations. Clearly, Young wasn't really necessary for the team to win. The Young/Price trade made some sense - but what does all of the above imply about Shalee Lehning - or Erika de Souza or Iziane Castro Marques?


(*) - At least I think this is a new approach, as I haven't read about anyone else using it.