Tis the season to look at polls. If you are interested in this election (and who isn’t?) you might find yourself spending much of your day looking at the latest polling data. And what do these polls show? From Nate Silver’s site – fivethirtyeight.com – we see that last summer it looked like Barack Obama was predicted to win. And then around the Republican convention, John McCain took the lead in the polls. More recently, the polls once again predict an Obama victory.
In evaluating the quality of these predictions we don’t look at just one poll or one snapshot. To get an accurate sense of what’s going to happen, we consider a number of snapshots. And hopefully, as the amount of information we gather accumulates, our forecast improves.
Polling NBA Rookies Before the Preseason
Our analysis of NBA rookies follows this basic philosophy. As we consider a variety of snapshots we hope to zero in on an accurate forecast.
The first snapshot we will consider is what the players did in college. As noted in September – and Table One – the college numbers indicated that Michael Beasley and Kevin Love will be productive NBA players (and Greg Oden can be included in this list). Meanwhile, the college numbers of Russell Westbrook, Brook Lopez, Jerryd Bayless, D.J. Augstin, Joe Alexander, O.J. Mayo, Eric Gordon, Robin Lopez, and Anthony Randolph suggested that these players may not be prolific producers of wins in the NBA (at least not during their rookie seasons).
Of course, this is just a suggestion. A few months ago we saw a second set of numbers in the NBA’s summer league. And these numbers… okay, it doesn’t look like summer league numbers mean much (at least that was my argument last summer).
The Preseason Numbers
In the past few weeks, though, new numbers have come in. We have now seen these rookies in the preseason. Now preseason basketball is not the same as the regular season. Teams don’t play their starters much of the game and many of the players on the court are never going to appear when the games count. Nevertheless, there does appear to be a link between preseason performance and what see in the regular season.
In 2006-07 and 2007-08 there were 34 rookies who played at least 100 minutes in the preseason and 500 minutes in the subsequent regular season. For these 34 rookies we see a 0.81 correlation between Win Score per minute in the preseason and the regular season. In other words, 66% of the variation in what a rookie does in the regular season is explained by what he did the preseason (below I explain the difference between correlation and explanatory power)1.
If we add in what we learned in college last year, our explanatory power improves. When we consider both Win Score per minute in college and in the preseason, we are now able to explain 75% of the variation in regular season performance (by the way, this doesn’t mean that college performance only explains 9% of NBA performance). In sum, it looks like we now know something.
Table Two reveals what we have learned. Both the college and preseason numbers indicate that Greg Oden and Kevin Love will be above average NBA players as rookies. In other words, an average NBA player posts a WP48 [Wins Produced per 48 minutes] of 0.100 and both Oden and Love are expected to eclipse this mark.
Now most rookies are below average. In fact, an average rookie only posts a 0.042 WP48. If we consider the mark of an average rookie, then it looks like Mario Chalmers, Ryan Anderson, Mareese Speights, Michael Beasley, and O.J. Mayo – will be above average freshmen.
For a few other rookies, though, we now have two snapshots that should give their teams (and fans) some concern. The model – which looks at both college and preseason performance — projects that Derrick Rose, Brook Lopez, Eric Gordon, Jerryd Bayless, Russell Westbrook, D.J. Augustin, and Joe Alexander will be below average for rookies. And a below average rookie is really not someone who will help a team win many games.
A Few More Thoughts
Other notes on these results:
- The model only projects Greg Oden to be just a bit above average. Oden, though, was hurt in college so this projection is probably too low. I would guess that Oden’s rookie numbers will come closer to what we saw in the preseason. In other words, I expect Oden to be the most productive rookie in 2008-09.
- Beasley looked great in college, but his preseason numbers are quite bad. So the second round of numbers on Beasley should leave Miami fans unhappy.
- In contrast, Memphis fans should become more optimistic. O.J. Mayo didn’t play well in college. But in the preseason he was close to average for an NBA player. So the projection on Mayo has now been upgraded from below average as a rookie to above average for a first-year player (but still below average for an NBA player).
As always, we need to close this discussion by noting that although college and preseason numbers tell us quite a bit, we are still left with just a projection. For some of these players the projection for the rookie season will be incorrect. Furthermore, these projections are not telling us anything about performance beyond the rookie season (although there is a link between rookie performance and what a player does the rest of his career). So it’s possible that a number of rookies with bad numbers today will develop into productive NBA players in the future. And of course, it’s also possible that won’t happen.
1- The correlation coefficient -or “r” – tells us the strength and direction of a relationship. But if we want to know how much of the variation in a dependent variable (what we are trying to explain) is explained by our model, we turn to r2. For a model with only one explanatory variable – such as a model linking preseason performance to regular season performance – r2 is simply the correlation coefficient squared. So a correlation coefficient of 0.81 between preseason and regular season performance tells us that we are explaining 66% (0.81 * 0.81) of the variation in regular season performance (with just preseason numbers).
It’s important to remember that r2 is really the ratio of the explained sum of squares (or how much variation our model explains) to the total sum of squares (how much variation there is to explain). So of the variation that exists in regular season performance, preseason performance explains 66%. If we consider both preseason and college performance, we can then explain 75% of this variation. And again, this doesn’t mean that college performance only explains 9% of NBA performance.
The WoW Journal Comments Policy
Our research on the NBA was summarized HERE.
Wins Produced, Win Score, and PAWSmin are also discussed in the following posts:
Finally, A Guide to Evaluating Models contains useful hints on how to interpret and evaluate statistical models.