Players don't get any hotter than Kevin Durant was when his Oklahoma City Thunder hosted the Golden State Warriors on Jan. 17.
After missing on his third shot attempt of the game, Durant made his next eight in a row. He scored 15 points in the first quarter and had 29 by halftime en route to a career-high 54. Durant saved his greatest heroics for the final period, when he stymied a Warriors comeback attempt by making three triples in a little more than a minute, following it up with a shorter jumper on the Thunder's next possession.
By that point, Durant looked like he couldn't miss. But he did. In fact, Durant didn't score from the field again, misfiring on his last three shot attempts and scoring his lone two points of the final four and a half minutes at the free throw line.
I couldn't help but think about Durant's performance against Golden State on Friday at the MIT Sloan Sports Analytics Conference in Boston, when John Ezekowitz and Carolyn Stein were presenting their paper (co-authored with Andrew Bocskocsky) on revisiting the so-called "hot hand" in basketball.
The hot hand has been a hot topic among academics since Thomas Gilovich, Robert Vallone and Amos Tversky authored a paper disputing its existence in 1985 by showing that players on the Philadelphia 76ers were no more likely to make their next shot after a make than a miss. With additional support from studying streaks at the free throw line and in practice situations, they concluded that past shooting success did not predict more in the future.
Data insights, Durant's example
The example of Durant, however, shows why a well-designed study using play-by-play data might not detect the existence of the hot hand. You see, Durant didn't necessarily go cold down the stretch against the Warriors. Instead, Golden State altered its defense by denying him the basketball and using double-teams, leaving Durant the option of forcing more difficult shots or passing to teammates.
The results Durant posted were influenced by the perception of everyone involved: his teammates and coach, who were more likely to defer and try to get him the ball; the defense, which was more likely to deny him easy opportunities; and Durant himself, who was willing to expand his shot selection to take advantage of his hot streak.
More sophisticated data has allowed today's researchers to account for these factors in assessing players' streaks. In a presentation at the 2009 Sloan Conference, John Huizinga and Sandy Weil showed that players attempt more difficult shots after made baskets than after misses. This year's paper expanded on that effort by using SportVU camera-tracking data to quantify the difficulty of every shot from the 15 arenas equipped with SportVU cameras last season based on location on the floor, the shooter's ability and the location of nearby defenders.
The results Bocskocsky, Ezekowitz and Stein found were consistent with Durant's experience. The hotter a player has been over his previous four shots -- accounting for their difficulty -- the more likely he is to take his team's next shot. On average, an extra make over his past four attempts translates into the next shot coming from six inches farther away from the basket, and with defenders in closer pursuit.
After accounting for those factors, the researchers determined that players are in fact more likely to make a shot of the same difficulty after a hot sequence than a cold one. They typically add a little more than half a percentage point to their shooting percentage for each additional made shot in their past four. So an average player coming off a 4-for-4 stretch, compared to going 0-for-4 on shots of the same difficulty, would see his field-goal percentage go from 45.1 percent to 47.3 percent if he didn't change his shot selection.
If that's not quite the same as being "on fire" in NBA Jam, as Ezekowitz joked, it's still enough of an effect to potentially explain the feeling players at every level occasionally experience of being "in the zone."
From a practical standpoint, the existence of the hot hand matters in how players and coaches react to hot streaks. The evidence still shows they overestimate the size of the effect, as Durant did against the Warriors. Players must be cautious not to force things too much, while coaches should remind them that an open shot for a colder teammate is still generally better than a heavily contested one during a good stretch. Oklahoma City could get away with Durant struggling down the stretch because his 3s built a comfortable lead. Not all players and teams who overestimate the hot hand are so fortunate.
Elsewhere at the Sloan Conference
• Along with the hot-hand paper, all three other basketball research papers at the conference (including the winning paper on the elements of rebounding by Rajiv Maheswaran and his team at Second Spectrum) utilized SportVU data, which Stats Inc. has provided for academic work.
With cameras now in all 30 arenas, every team in the league has access to tracking data from every game. But there are wide differences in how effectively they've converted the raw information into usable conclusions, with teams that installed the cameras early (such as the Toronto Raptors, as Grantland's Zach Lowe highlighted last year) having a major first-mover advantage.
The NBA also has made select SportVU stats available on its website and even begun to incorporate them into box scores, a huge step forward for the public. Still, most of what has been learned from SportVU remains proprietary to teams, who view those conclusions as a competitive advantage. We've seen only a small amount of what SportVU potentially has to offer.
• As exciting as SportVU's potential is, the sample size on numbers derived from it remains tiny. Until we have data for every game from multiple seasons, it's tough to determine how much of the differences among players in categories such as opponent shooting percentage within five feet of the basket is noise as opposed to skill.
There's also the challenge of interpreting new statistics. Former Orlando Magic coach Stan Van Gundy joked about learning that Paul George has run nearly 150 miles this season ("Of what possible use is this information?" he asked rhetorically), but the same is true of performance-based stats such as secondary (or "hockey") assists. Are they as valuable as actual assists? Or valuable at all? We can compare players to each other in these skills, but quantifying their importance is much more difficult.
• One thought-provoking note from the rebounding study: The top players at converting defensive rebound opportunities are much better than their peers on the other end. Of the top 10 in converting offensive rebounds, only Gerald Wallace has ever made an All-Star appearance, but eight of the top 10 on defense have been All-Stars. That group includes last year's three leading vote-getters for MVP (LeBron James, Durant and Carmelo Anthony) and former MVP Kevin Garnett.
One possibility is that defensive rebounding is an indicator of other athletic ability that helps stars dominate. There is some evidence for this -- at certain positions. Point guards with high defensive rebound rates in my college database tend to outperform both their WARP projections and their spots in the draft. Four of the top 10 point guards in defensive rebounding developed into All-Stars, while none of the bottom 24 have become All-Stars. But at other positions, defensive rebounding doesn't look like an indicator of success.
The cynical explanation is that because multiple players often have a chance at grabbing defensive boards, role players might be deferring to their star teammates, allowing the bigger names to pad their stats.
• For the first time, all 30 NBA teams were represented at the Sloan Conference. As they promised last year, when they were the only team not in attendance, the Los Angeles Lakers sent three people (including panelist Jeanie Buss) to Sloan.