While watching games, I’m simultaneously intrigued by and skeptical of stats that announce that player X does Y at Z effectiveness when his team wins, and not when his team loses. It’s a nice thought, that one player’s specific behavior might indicate, or even predict, his entire team’s outcome, but the choices of data are mostly arbitrary.

To make this exercise a little more robust and interesting, I decided to inject some statistical steroids into the traditional analysis. Below is a graph that shows the difference in five indicators (assist ratio, defensive rating, offensive rating, true shooting percentage, and usage percentage) between team wins and losses of all nine of the Houston Rockets regular rotational players (click for a full-sized interactive version).

First, the boring technical explanation. Offensive and defensive ratings are points per 100 possessions. Assist ratio is the number of assists and individual player has per 100 possessions. True shooting percentage is a field goal rate that considers three-point shooting percentage and free throw percentage. Usage rate is an indication of how much of a team’s posessions is “used” by a certain player (i.e., how many times a player shoots, goes to the line, etc.).

On the x-axis are players, excluding Asik due to lack of sample size. Leading each cluster of players is the Houston Rockets as a whole (the red bar). Also on the x-axis are the players’ performance in five statistical categories. The y-axis shows the players’ percentage change in those five aforementioned categories from games that the team wins to games that the team loses. Please note this is the percentage change in the statistic, not the nominal change. So if a player goes from 10 of something when the team wins to 9 of something when the team loses, his difference in this chart is -10%, not -1.

Finally, a warning about how to interpret these results. These data DO NOT show causality. In fact, the directionality of these data is completely unknown. Basically, we don’t know if the Houston Rockets start losing once Dwight Howard is featured more, or if Dwight Howard is featured more once the Houston Rockets start losing. Either scenario is equally plausible for each indicator.

Onto some quick hitting observations.

- Francisco Garcia appears to be the most offensively volatile but defensively consistent player. The difference in his offensive rating and TS%, especially his TS%, is the greatest on the team, while the difference in his defensive rating is the smallest.
- The point guards actually have more assists per possession in losses than in wins.
- The team’s “stars” and Casspi (?) see an uptick in usage during the team’s losses.

My opinion is that the final two bullet points indicate that there is a relationship between losing games and deferring to a few players. Regardless of what happens first, in games that the Houston Rockets lose, certain players pass more and certain players shoot more. Instead of trying to run an offense, players begin heroballing, thinking that it’s their responsibility to ignite a comeback. That’s usually a recipe that leads to bad results, barring extremely efficient performances from those privileged few. Perhaps sticking with the offense at hand, understanding that lumps will always be encountered, would be a more sustainable and effective solution.

View this discussion from the forum.