I’ve completed the first cut NBA projections for 2013-2014, and was surprised to see second-year Blazers point guard Damian Lillard rank 4th in overall RotoValue for both an 8 category and 9 category format. Last year’s rookie of the year was a good player, but not even in the top 20 overall last season. So why is he so high in 2014?
Well, a second-year player has just one year of data I use in projections, and in that rookie year he was an iron man, playing all 82 games, logging almost 39 minutes per contest. So his projections also have him playing 82 games, and again averaging nearly 39 minutes. Indeed his projected per-game averages hardly move from his rookie year: 19.1 ppg, 3.5 rpg, 6.4 apg. Still, while those are good stats to be sure, they’re not world-beating. So why so high a ranking?
It turns out playing time matters a lot. Compare last year’s rookie of the year to his predecessor, Kyrie Irving. Irving’s per-game stats project much better, but also he’s only played 51[ref]Out of a possible 66 in the strike-shortened year[/ref] and 59 games in his 2 NBA seasons, so he projects to just 59 games this year. When you look at projections on a per-game basis, Irving ranks in the top 10 in both formats, well ahead of Lillard. So the model does think Irving should be more productive when he plays; it’s just less confident that Irving will play more.
The site actually has two projection data sets, one derived from the other. The first one is “Raw Projection”, the output of a model with mean reversion and attempting to incorporate player aging based on historical data. That model only projects players who have played at least one year in the NBA, and it ignores what team a player is on. The Projection data makes two adjustments: first, it uses preseason per-minute data (and an arbitrary season of 70 games) to project rookies; and second, it “normalizes” playing time on a team. Basically I add up all the projected minutes on a team, and compare that to how many minutes a team should play (I assume 48.4 minutes for each of the 82 games, to allow for about 6-7 OT periods per team per year). If the team totals are too high, the model looks to remove minutes from players it considers weaker on a per-minute production basis until the projected total is at the target. Conversely, if they’re too low, the model adds minutes to stronger players (who aren’t already at very high levels). So when a team loads up on veterans who normally would project to a lot of time, the projection will adjust playing time down for some of them; conversely a rebuilding team needs to have someone play minutes, so those whose per-minute stats are better will be projected to play even more.
I’m not crazy about using the very small sample size of preseason data for rookies, but I don’t have college data on the site, and trying to convert college stats into some equivalent pro context is itself a major undertaking. And the model worked pretty well with Lillard last year, as it pegged him to start and play a lot, with projected averages of 17.1 points, 2.7 rebounds, and 6.4 assists per game, compared with actual regular-season averages of 19.1, 3.5, and 6.5.
I remembered being surprised at Lillard’s projection last year, but I also said, “Damian Lillard, the projected starter in Portland … has posted impressive numbers in the preseason, especially shooting. He’s certainly a player worth watching, and he may wind up as the year’s most valuable rookie.”
No rookie this year so far looks as good as Lillard did last year, but the best so far is Sixers’ Michael Carter-Williams. He was just the 11th pick overall, but he projects to start for a team that will give him a lot of time. So he too might be a nice sleeper pick for this year.