Let's start today's column with some quick trivia. What do Kelenna Azubuike, Bostjan Nachbar and Martell Webster have in common--besides all being swingmen?
Give up? The answer is that all three were amongst the NBA's top 20 preseason scorers last year, averaging 16.6, 18.2 and 17.7 points per game, respectively. All three outscored both LeBron James (14.1 ppg) and Kobe Bryant (13.2). So too did Anthony Roberson (14.2 ppg), and for his trouble Roberson did not even make the Denver Nuggets and did not play in the NBA last season.
These are examples I've cherry-picked, and points per game certainly isn't anywhere close to a complete measure of player performance. Still, numbers like those indicate why there is inherent distrust of whatever happens in the preseason, good or bad. This is completely understandable. After all, the exhibition season is more about experimenting and tuning up than it is about winning. Teams play combinations of players who may never see the floor again while facing opposing lineups of widely-varying strength.
Nonetheless, past studies have found some meaning in preseason performance at the team level. Roland Beech of 82games.com looked at the issue two years ago and found a surprisingly decent .40 correlation between preseason record and regular-season record, in comparison with a .57 correlation with record the previous season. (Correlation measures the strength of a relationship between two variables, with 1 indicating a perfect relationship and 0 indicating none at all.) Noteworthy amongst Beech's conclusions that preseason performance was almost entirely useless in predicting the performance of teams that had won at least 50 games the previous season, but became very useful for teams that had won fewer than 30 games.
Last year, Devin Black took a slightly different perspective on the issue for RealGM and found similar results.
Intuitively, this makes sense. Good teams are much more likely to rest their players and use the preseason as an opportunity to slowly build into the regular campaign, knowing they have a long season ahead of them. Teams coming off of poor seasons are developing young players who will continue to play during the regular season or integrating newcomers into the lineup, making preseason a more meaningful indicator.
If the preseason can hold some predictive value for teams, is the same true for individuals? The only study I've seen to consider the issue was done last year by the SuperSonicSoul blog, but focused exclusively on rookies and looked at just a handful of statistical categories.
In studying the preseason, I was limited to looking at last year's numbers, available at DougStats.com. While this isn't an ideal sample size, we'll have to make do. The first surprise when I plugged the numbers into my spreadsheet was how normal they looked, at least when I ranked players by Wins Above Replacement Player. While Gerald Wallace led the NBA in preseason WARP and Ronnie Brewer and LaMarcus Aldridge joined him in the top 10, the rest of the group featured a who's who of All-Stars like Dwight Howard, Kevin Garnett, Chris Paul, Steve Nash, Allen Iverson and Dirk Nowitzki. Even with nothing on the line, the best talent tends to rise to the top.
Like Beech did for teams, I wanted to compare the predictive power of the preseason to numbers from the previous regular season, so I took the group of slightly more than 200 players who played at least 250 minutes in both the 2006-07 and 2007-08 regular seasons and at least 100 minutes in the 2007 preseason. Again, preseason performance proved surprisingly robust. The correlation of 2006-07 and 2007-08 per-minute Win% was .788; by comparison, preseason Win% had a .583 correlation with 2007-08 performance.
The more interesting question might be this: Does preseason performance offer us any additional information? After all, if what we're seeing is simply that better players play better no matter the circumstance, we already can identify these players based on their previous regular-season performance. To look at the matter this way, I created a regression seeking to predict 2007-08 regular-season performance from 2006-07 performance and 2007 preseason performance. Combining the two improves correlation slightly over 2006-07 performance alone, up to an r of .814. Looking at the coefficients on the two variables indicates that preseason Win% is about a fourth as useful as a predictor as Win% the previous season.
Looking at the players whose preseason performance differed the most from how they had played the previous season reveals a mixed bag. The group of strong preseason performers included breakout players Brewer and Louis Williams, while sophomores LaMarcus Aldridge and Jordan Farmar did not see their per-minute performance improve dramatically but did add many more minutes. However, another second-year player, Hilton Armstrong, actually took a step backwards after his strong October and improvement wasn't in the cards for then-Sonics Luke Ridnour and Chris Wilcox.
Struggling in the preseason proved meaningless for David West, who was actually amongst the 10 least valuable players in exhibition play before posting an All-Star campaign. Still, a poor month of October was prophetic for the rest of the 10 players who dropped off the most from the 2006-07 season, a group which includes Drew Gooden, Donyell Marshall and Darko Milicic. They would ultimately see their collective per-minute winning percentage drop from .530 to .452.
Another interesting group is those players who had no track records, the rookies. The correlation between preseason performance and regular-season performance was stronger for rookies (.618) than for veteran players (.583). If we had predicted how rookies would do from their preseason efforts, we would have come pretty close on several players; Kevin Durant's winning percentage was virtually unchanged. Al Horford couldn't keep up his All-Star-caliber play in October, but that still provided an early indication he would challenge for Rookie of the Year. Even the two major exceptions amongst the rookies make sense--Juan Carlos Navarro and Luis Scola played much worse in exhibition games, which would make sense as they adjusted to the NBA game.
Looking through the numbers, it looks like the biggest factor that works against preseason stats is not the uneven motivation and level of competition but instead the brevity of the schedule. Break down the numbers by category and the effort stats (rebounds, blocks, etc.) carry over closely into the regular season, while shooting percentages--more variable even season to season--have a dramatically lower correlation.
Speaking of fluky shooting, let's go back to our original trio of players. Judging their overall performance, they played well, though they were not amongst the league's best preseason performers. Nachbar's play was buoyed by a bizarre number of trips to the free-throw line--eight attempts a night. Webster (58.4 percent from the field and 50.0 percent on threes) and Azubuike (51.8 percent) are more stereotypical examples of players who just had a hot streak at a noteworthy time.
While it's best to be cautious of such players, don't throw out preseason stats altogether. The evidence indicates there is some predictive value to how players play even in seemingly meaningless games.
Kevin Pelton is an author of Basketball Prospectus.
You can contact Kevin by clicking here or click here to see Kevin's other articles.