Draft Prep: Correlates to Wins
Today's column examines NBA statistics through the lens of 'correlation,' to determine which stats were more or less likely to be found alongside other stats last season.
For example, we find that as field goal percentage (FG%) increased it became harder to accrue 3-pointers (3PT), assists (AST) and high free throw percentages (FT%). This makes sense intuitively, since PFs and Cs dominate the FG% category, but by using correlations we can objectively quantify these relationships.
Such knowledge empowers owners to more confidently draft a team as they see fit. That could mean balancing competing categories or piling up stats in some categories while de-emphasizing others. How strong is the correlation between assists and turnovers? What about the impact of high-volume 3-point shooting on a player's FG%? Going into your draft with a broad category-based strategy can help you avoid 'missing the forest for the trees,' which so often happens when player names and positions dominate your picks. Fantasy leagues are won with numbers, plain and simple, which is something all GMs should embrace.
Note: The population I'm using is the same I used for my end-of-season rankings back in June: "I tossed out players who didn’t appear in at least 30 games ... as well as players who didn’t average at least 20 minutes per game. After culling these two groups I was left with 216 players (enough for a 15-team league with 14-player rosters)."
You'll notice that the chart above includes three non-scoring categories: minutes, FG attempts and FT attempts. The categories with the most negative correlations were blocks, FT%, 3-pointers and FG%. Those with the strongest positive correlations were PTS, TO, STL and AST.
Be mindful of the pithy notions that correlation does not imply causation, and that the direction of the correlation can only be inferred. Turnovers and assists may show a strong positive correlation, but we can safely assume that it's the assists leading to more turnovers, not vice versa. Even more glaring is the minutes category -- obviously, increased minutes lead to increased points/assists/steals/etc., not the other way around!
Minutes-played has some of the most intriguing results. Naturally, it has a positive correlation with every statistical category under consideration. Points, FG attempts and FT attempts are particularly beholden to minutes, so you'll need to find high-volume scorers (like Kevin Durant, Carmelo Anthony or James Harden) in the early rounds. Steals and assists also show a meaningful positive correlation with playing time.
On the other hand, minutes-played has almost no significant correlation with blocks, FG% or FT% (unweighted averages). Even rebounds and 3-pointers show only a faint positive correlation with playing time. Late-round fantasy picks almost always play fewer minutes than early-round and middle-round picks, inferring that there might be copious shot-blocking in the late rounds, in addition to significant 3-pointers and boards. I say 'might be' because we're projecting future results on past events, with inherent risk.
Last season's examples of low-minute, high-block players include Jermaine O'Neal (0.9 blocks), Omer Asik (0.8), Samuel Dalembert (1.1), Timofey Mozgov (1.2) and Miles Plumlee (1.1), none of whom averaged more than 25 minutes per game. None of them are great fantasy options but they also contribute in rebounds and FG%, and they're representative of available shot-blockers at the bitter end of your drafts (or, more likely, off the waiver wire during the season).
Examples are just as easy to find if you're hunting late-round 3-pointers. Among players who averaged under 25 minutes per game, we find Marcus Morris (1.2 threes), Manu Ginobili (1.3), Mike Miller (1.3), Jordan Farmar (1.7), Greivis Vasquez (1.4), Tim Hardaway Jr. (1.6), Jared Dudley (1.1), Danny Green (1.9), Vince Carter (1.8), and many more. For a detailed look at position eligibility throughout drafts, check out my recent column, "Positioned to Win."
The three 'big man' categories (Blocks, Rebounds and Field Goal Percentage) show a strong positive correlation amongst themselves, with negligible or negative correlations to all other stats (the one exception is the mild positive between blocks/points). This is somewhat self-evident, but it underscores the fact that if you 'punt' blocks you may unintentionally hamper your ability to compete in rebounds or FG%. It also suggests that if you follow the ever-intriguing path of punting FT% to load up on big men, you'll need to figure out which other categories might pair well with BLK/REB/FG%. You can't win a typical league (roto or H2H) with three dominant categories.
I take an in-depth look at punting strategies in this year's NBA Draft Guide, so be sure to also check that out.
Since FT% is most strongly correlated with 3-pointers, and vice versa, you'd need to quickly decide whether you'll ditch both categories or attempt to stay competitive from beyond the arc (a tricky strategy, but it's not impossible). Three-pointers are unique in that they show a strong positive correlation with FT% but a strong negative correlation with FG% (in addition to REB and BLK). In this scenario, you could target guys with high 3-pointers and mediocre-to-poor FT%, like J.R. Smith, Trevor Ariza and Wilson Chandler.
This is a broad-strokes look at NBA stats and there are many exceptions to the rules, but those exceptions only increase the utility of correlations by highlighting unique players and statistical combinations. Other players who fit this unconventional mold include Russell Westbrook, Lance Stephenson, Draymond Green, Spencer Hawes and Joakim Noah.
Three categories display no glaringly negative correlations: PTS, STL and TO. The implications branch in two directions: you should be able to target those three categories without hampering your ability to compete elsewhere, and you can target other categories without weakening your ability to compete in PTS, STL and TO.
I'll now go through each major category to show it's 'strongest' (most positive) and 'weakest' (most negative) correlations. The correlations are only listed if they had a value (positive or negative) of at least 0.2. Correlations between 0.4--0.6 are in bold, and those above 0.6 are bold and italicized. I've deemed anything below 0.2 to be 'negligible.'
Points
Strongest: TO, AST, STL, 3PT, FT%, REB
Weakest: None, though it's high correlation with turnovers is of course a negative in 9-cat leagues.
Negligible: BLK, FG%
3-pointers
Strongest: FT%, AST, PTS, STL
Weakest: REB, BLK, FG%
Negligible: TO
Rebounds
Strongest: BLK, FG%, PTS
Weakest: 3PT, FT%, AST
Negligible: STL, TO
Assists
Strongest: TOs, STL, PTS, 3PT, FT%
Weakest: BLK, FG%, REB
Negligible: None
Steals
Strongest: AST, TO, PTS, 3PT
Weakest: None
Negligible: BLK, FG%, REB, FT%
Blocks
Strongest: REB, FG%
Weakest: 3PT, FT%, AST
Negligible: STL, TO, PTS
Field Goal Percentage
Strongest: REB, BLK
Weakest: 3PT, FT%, AST
Negligible: STL, TO, PTS
Free Throw Percentage
Strongest: 3PT, PTS, AST
Weakest: BLK, REB, FG%
Negligible: STL, TO
Turnovers
Strongest: AST, PTS, STL
Weakest: None
Negligible: BLK, FG%, FT%, 3PT, REB
And when you move from 9-cat to 8-cat, eliminating turnovers, here are the categories which gain or lose the most (aka those most or least correlated with TO):
Gain the most: AST, PTS, STL
Lose the most: None
Negligible: FT%, REB, 3PT, FG%, BLK
I've made two more charts which I will share with anyone who requests them. The first displays the r-squared values for the categories above. They are intimately related, but this chart (the 'coefficient of determination') describes how much variance in one category is attributable to another. The other chart shows Pearson's r values, but this time substitutes every player's z-scores in place of their actual stats. My standard disclaimer, which I'll repeat, is that I'm not a statistician. If you have any comments or questions, email me or send me a message on Twitter @Knaus_RW!
