With the rise of analytics and the global popularity of the NBA and basketball overall, NBA General Managers seem to have more resources at their disposal than ever before. Presumably this fact, paired with decades of history to learn from, would lead to teams being better at picking the best available player when it’s their turn to make a selection in the draft. We would hope that teams would have become less likely to select busts early in the draft or even just pick a decent player when a better player was available. However, GMs don’t appear to have gotten more competent at drafting over time — in fact, they seem to have gotten worse.
If NBA GMs were absolutely perfect at drafting, we might expect the best player would always go first, the second best player would go second, and so on. Likely there would be a bit of deviation from this order as GMs draft players who fit holes in their rosters, especially later in the draft when better teams are drafting. (A team with good starting and backup point guards might draft a wing who they expect to be slightly worse than an available point guard, for instance.) But in general, we will use GM’s proficiency for drafting the best available player as a proxy for their overall drafting competence.
We want some objective (or as objective as possible) measure to determine which players had the best careers. There are a lot of metrics that attempt to measure a player’s total positive impact — Player Efficiency Rating, Wins Produced, Win Shares, and many others. For our purposes, we’ll use win shares, as it gives us an intuitive and mostly reliable idea of a player’s total career contributions on both offence and defense. We’ll look at the total win shares over a player’s career, as this signifies both a player’s ability to perform on the court and his ability to stay on the court.
To measure how well GMs performed at picking the best available player (or at least the available player who would end up with the most career win shares), we can use a tool called “Spearman’s rank correlation coefficient”, which will allow us to compare the rank of the career win shares of a player among the other players in his draft class to the position he was selected. If teams drafted perfectly according to this metric, the player in a draft class who would go on to have the most win shares would be picked first, the player who would go on to have the second most win shares would be picked second, and so on. A perfect draft would get a score of one, and a draft done completely random would get a score around zero. This ensures that GMs aren’t penalized in draft years without top talent; as long as they draft the one of the best players available, the metric will report that they did well, regardless of whether the best player available was LeBron James or Hedo Türkoğlu. Years where GMs did a better job of picking the top talent available would tend to have scores closer to one, whereas years where teams weren’t reliably picking the best available players will have scores that are lower.
If teams have improved at drafting (and developing) players, we would expect to see our Spearman’s rank correlation coefficients get bigger over time. We’ll look at NBA drafts between 1979 (the year the NBA added the three point line) to 2015 (so we have several years of data for draftees), and we’ll focus specifically on the first round of draft picks. These years will give us an idea of drafting in the modern NBA era, and limiting ourselves to the first round can cut out a lot of noise from later draft picks, whose career win shares may often be a consequence of whether they are given an opportunity to play almost as much as their actual ability. Having one or two diamonds in the rough appear out of the second round can really screw up how well this metric thinks teams did at drafting in a way that seems unfair to me — second round picks seem to act more like dice rolls on prospects who are particularly unlikely to achieve their best-case scenario career.
Graphing how well teams did at drafting over time, we find something surprising:
Teams appear to have gotten worse at drafting, particularly if you compare the current millennium with the previous one. The linear correlation coefficient (R^2) between Spearman’s coefficient and the year of the draft suggests a quite strong correlation (above 60%) between the draft year and how well teams drafted. Something has happened to make teams worse at drafting — or at least look like they’re worse at drafting — but what?
Maybe Drafting Just Got Harder?
As basketball has grown both domestically and internationally, NBA general managers have had to make their decisions among a larger number of prospects and among prospects who have been playing against a wider variety of competition. Maybe the marginal advantage gained by additional resources being devoted to scouting potential NBA players is swamped by the growing number and diversity of those potential NBA players. The best player in major division college basketball might not be as good as that awesome young player in the EuroLeague, but making a straightforward comparison is particularly hard when they are playing against different kinds of competition with different rules.
So maybe we should GMs some slack when we compare them to their forebears. The task of determining who is the best player might just have gotten much more difficult.
Maybe Teams Drafting Early Are Swinging for the Fences
Another possible component contributing to the apparent decline in drafting competence might be that teams have begun drafting for upside more than drafting for expected value. A “safe” early pick who ends up being good but not great won’t make a bad team a top contender, and likely won’t even carry a really bad team to the playoffs (see: Phoenix, Minnesota, many others). But a “risky” early pick who ends up being a top ten player can completely turn around your franchise’s fortunes.
Perhaps teams now approach the draft with more of a high-risk, high-reward attitude than they did in the past. If a good player won’t turn your franchise around, why take the safe bet on a merely good player when an unlikely but potential superstar is available? This is especially true when missing on a potential superstar will just result in you getting more chances at early picks in future drafts, when a good player might put you in the purgatory of being unable to go far in the playoffs but being too good to get early draft picks. Finding that early draft picks are less consistently good makes GMs appear to have gotten worse at drafting, but GMs might actually be adopting a different strategy than just picking the available player who is most likely to be good.
Do Total Win Shares No Longer Reflect What Teams Value Most?
Rudy Gobert has averaged about 1.3 times as many win shares per season as Andre Drummond, but I would bet that teams value him (as a basketball player, of course) much more than 1.3 times as much as they value Andre Drummond. Win Shares are based on a combination of defensive rating, which is a Wikipedia-tier statistic — fine for getting a decent idea of what’s going on, but inadequate when you really need the detailed truth –, the points a player generates per possession (through his own shots and through assists), and his number of possessions.
Perhaps over time teams have come to more greatly value precisely the things players can do to help their teams win without accumulating win shares. Things like spacing the court, setting good screens, throwing passes that lead to an assist on the next pass, and stifling pick and rolls help teams win but aren’t perfectly captured by win shares. However, things like shooting and assisting in ways that produce efficient offense and making opponents generally score less efficiently when you are on the floor are captured by win shares, and teams definitely value those things. Maybe a change in values can explain part of the change over time we found, but I have a hard time imagining that this offers a complete explanation, since win shares still capture a lot of the things that teams consider valuable, even as the game has changed and teams have embraced more advanced analytics.
I don’t think that GMs getting less competent is a likely explanation of their apparent decline in their ability to draft. However, the fact that they look like they’ve gotten worse at their jobs according to at least one intuitive way of measuring how well they are doing brings up interesting questions about the incentives driving teams’ decisions and how these have changed over time.