Preseason is, finally, drawing to a close. So I think a fair question is what, if anything, we might have learned from it. An 82games study a while ago indicated a 40% correlation between preseason win percentage and regular season. (Yikes, for us Celtics fans, though they didn't do a point differential, which should be a slightly better predictor and would have the Celtics at -3.57 point per game projected at 31 wins rather than 11 wins. Thanks Knicks bench!). The current batch of advanced stats can already do better than that, however.
For most players who have been in the league a few years, especially on the same team, adding a couple of exhibition games of data shouldn't change our assessment of their abilities much at all.
There are some exceptions where there is a significant discontinuity and preseason gives us significant new information, like Derrick Rose coming off of injury or, to a lesser extent, a player on a new team with a new role. But, I think rookies' performance is probably the most informative and interesting preseason data since this is the first time we have seen them play against NBA competition.
With that in mind I used data from RealGM on rookies for the last three years in both preseason and regular season to create the Tableau visualization below. I kept things pretty simple looking at simple linear box score metrics Alt Wins and Win Score (new). Also Rebounds, Assists, field goal percentage and turnovers. The highlights and my inferences:
Once you get to 50 minutes most measures are starting to give you statistically valid information, with this set of data 70 to 80 minutes in the preseason seemed to be the 'sweet spot' where the correlations were high(ish) without dropping too many players.
Win Score correlated better than Alt Win score as far as correlation between pre-season and regular season, with a Goodness of Fit score of .39 at my favored minute filters. I suspect the higher correlation is mostly because of the heavier weight on rebounds since I didn't apply a position adjustment.
Looking at the stats scaled to per 48 minutes, rebounds and blocks are the most predictive stats, followed by assists, But all are somewhat positionally influenced (as you can see on that tab). For rebounds the R2 drops from .75 to .47 when only looking at Bigs. For assists it dropped from .78 for all positions to .68 only including Guards and .12 for Point Guards.
Field goal percentage is much more variable given the amount of minutes/shots the rookies got and the variable nature of shooting. Even free throws had little predictive power.
Steals, turn overs and personal fouls were all in middle ground. In bad news for Kelly Olynyk fans the fouls committed R2 was .58, indicating some shorter outings for the big guy.
All the correlations look pretty similar to what I have seen others report from college to pros, with exception of shooting because of the small preseason sample. The interesting questions will be what happens when the two data sets contradict each other.
The next step will be to apply this information to this year's batch of rookies once the preseason is done.
On the visualization minutes on both preseason and regular season are adjustable, as are positions. Tab over to the various stats.