This analysis was done very similarly to my research on big men that I discussed last week. The same basic guidelines were used for selecting players: they needed to play at least 600 minutes in the NBA (and were in the 2010-2015 draft classes). This still isn’t all that much of a sample size, but at least indicates real playing time, enough to take something away from. Correlating college stats to NBA advanced numbers just won’t mean all that much if those advanced numbers (which are notoriously suspect to swings in small samples) are based on 100 garbage time minutes.
‘Wings’ are just as broad a category as ‘big men’, if not more so. Some of the players in this grouping play minutes at point guard, others at power forward. A handful are dynamic scoring stars, many others are limited role players. But that’s kind of the point: The draft prospects in this draft are a correspondingly diverse set of players with different skills, strengths, and roles, which means they deserve to be compared to a group that reflects that variety. Breaking down the NBA players and prospects into more specific categories could be interesting, but players change roles so quickly even in their rookie NBA seasons (think Donovan Mitchell, projected as a defense-first prospect, who turned in one of the best scoring seasons ever by a rookie), that this seems like an idea that sounds better than it would be in practice.
Again, I only look at the stats from the prospects’ last two seasons in college, and then average them together equally. I believe that upperclassmen get overly penalized for their age, and as “age” is a separate category in my regression analysis, I don’t need to drag the stats of the juniors and seniors down further. I also don’t use per-40 stats, mostly because if players are going to make it in the NBA, they should be able to play big minutes in college. Sure, some coaches don’t play their best players as much as they might, but maybe there is some limitation to the player that they know about, hence the lack of minutes. I take this into account when looking at prospects, acknowledging that their numbers are low because their minutes were reduced, but when looking at the overall model, I prefer straight per game stats.
Here are the simplified results of the regression analysis. ‘Significant’ means that the statistic explained some of the variability in the NBA advanced statistics to at least a small extent. ‘Model-usable’ indicates that while not useful by itself, the inclusion of that stat strengthened the overall model of correlation. ‘Insignificant’ suggests that the college stat had no real bearing on the players’ advanced metrics in the NBA.
Wing Stats Significance
Compared to the big men, there was somewhat more variation for the wings in which college statistics were significant for NBA success. A mere three stats – age, turnovers per game, and assists per game—were noteworthy or even usable for all the advanced metrics, and only one, blocks, was irrelevant across the board. That meant evaluating the rest of the stats, and measuring their importance, was complicated somewhat. In the end, the Win Shares model was by far the weakest of the three, pushing rebounds and steals towards greater status, and sending points and three point makes in the opposite direction. Therefore, if I had to tier the importance of the statistics in my model, it would be something like:
Tier 1: Assists, Age, Turnovers
Tier 2: Rebounds, Steals
Tier 3: True Shooting
Tier 4: Points, Three pointers
Tier 5: Blocks
Coefficients of Wing Stats Model
|PPG||Negative (but not in model)||Negative (but not in model)||Positive|
|RPG||Positive||Positive||Positive (but not in model|
|SPG||Positive||Positive||Positive (but not in model|
|BPG||Positive (but not in model||Negative (but not in model)||Negative (but not in model)|
|3PT||Negative (but not in model)||Positive (but not in model)||Negative|
|TS||Positive (but not in model||Positive||Positive|
The coefficients in this analysis are much easier to dissect. The six most important stats all have the same coefficient sign for each model, and they all make sense: Turnovers and age are negative (better to be lower); assists, rebounds, steals, and true shooting are positive (better to have higher numbers). So a young prospect with low turnovers and high rebound, steal, and assist numbers would rate incredibly well by my model, while an older prospect with high scoring numbers (even efficiently) but little in the way of peripheral stats would not fare so well. Why are various scoring and shooting statistics relatively irrelevant for wing player success compared to more “do-it-all” type stats?
I think the easiest explanation is that while it’s difficult for poor shooters to become great ones, it’s very common for them to at least make their way to average, or slightly above. Look at two of the key role players in the Western Conference Finals: PJ Tucker took four threes total in three years of college, yet this past season shot 37.1% from outside on 3.8 shots per game, and Trevor Ariza has taken six to seven threes per game over the last four seasons even though he shot just 23.7% on three per game at UCLA. A more recent example—Jaylen Brown shot just under 30% from deep two years ago at Berkeley, and took a mere three of them per game. This past season, he worked his way up to 39.5% on 4.4 attempts per game. None of these guys has ever been or will ever be Steph Curry, or even Klay Thompson or Kyle Korver. But they all went from being relatively putrid outside shooters (or non-shooters) in college to solidly average or above in the NBA.
What’s much harder to improve on is intrinsic feel for the game. Some players are great at reading a basketball court and knowing what to do on it. They aren’t necessarily skilled, or athletic—they’re just smart, and back up those smarts with enough other tools to be productive basketball players. The difficult part about evaluating basketball IQ and feel is that there is no real measurement for it statistically. But I do think there are some statistics that, more than others, show that players know what to do and where to be on the court: Assists, rebounds, steals, and turnovers prominent among them.
Assists can be the result of ball dominance of sheer court vision, true. But for most of the wing players in college, they don’t have the ball enough to really rack up assists that way. Instead, they generate assists through the flow of the offense—by reading the court. Rebounds and steals are helped by athleticism and size, as well as gambling for them, which doesn’t always help a team win. However, plenty of big, athletic players aren’t great rebounders, and many quick, long-limbed ones don’t get a lot of steals. Why? Because they don’t have the ability to sense where other players are on the court as well as others, or instinctively know where a pass or rebound is going. These are skills that translate across all levels of basketball, and are hence useful in evaluating prospects for their jump to the NBA.
The NBA has also shifted in a way that favors wing players that do other things besides score or create for others in isolation. Those skills are still incredibly valuable, of course, but most teams right now are looking for versatile, smart players who can do many things on the basketball court. What better way to show whether prospects can be that type of player than through the “basic” stats that best indicate productivity outside of scoring. A non-threat wing who is a great rebounder or passer can still get played off the court in the playoffs, absolutely. Yet they have a better base to build their career in the modern NBA than a fantastic isolation scorer who can’t do much else positively when he doesn’t have the ball, or his shot isn’t falling. In a couple years, the NBA might have pivoted yet again, in a direction we can’t foresee. Right now, versatility (especially for wing players), is key.