As a Warriors fan, my team has been incredibly successful over the past decade. With the departure of Kevin Durant last year, another season-ending injury for Klay Thompson, and a depleted bench comprised of undeveloped young players, the two remaining Warriors' stars Stephen Curry and Draymond Green have lead the team to a 21-20 record good for 9 seed in the Western Conference (as of March 17th). After a 26-point blowout loss to the defending champion Lakers, Curry noted to the media that opposing teams remembered the dominance his team had and wanted their revenge.
Steph: "Teams still want to beat us and beat us bad. They still have a lot of memories from the last five, six years." pic.twitter.com/81JALMG8bt
— Warriors on NBCS (@NBCSWarriors) March 1, 2021
Recently, I've noticed some players on opposing teams greatly outpace their normal levels of production against the Warriors:
- Versus Jazz on March 14th: Rudy Gobert had 24 PTS and 28REB on 80% FG, from his season averages of 14.5 PTS, 13.5 REB on 64.7% FG
- Versus Lakers on March 16th: Talen Horton-Tucker had 18 PTS, 10 AST, and 4 REB, from his season averages of 7.7 PTS, 2.2 AST, and 2.6 REB
- Versus Lakers on March 16th: Montrezl Harrell had 27 PTS on 78.6% FG, from his season averages of 14.5 PTS on 63% FG
I'll be the first to admit that there are very reasonable explanations for this. Against the Jazz, between Kevon Looney's injuries diminishing his athleticism and James Wiseman's inexperience undermining his defensive capabilities, the Warriors did not have a reliable large paint presence. As good of a defender Draymond Green is, he was often drawn out to the perimeter to guard the plethora of three-point shooters on the Jazz roster. This lack of paint defense allowed Harrell to exploit many of the same weaknesses Gobert did two days prior.
On another note, it seems every single day one of my friends, a fan of another team, is complaining about an opposing player having an abnormally good game. We viscerally feel that the opposing player performs exceptionally well against only the team we're rooting for. Logically speaking, this is likely due to a selection bias. We watch our team by far the most and therefore notice opposing players performing well only against our teams regularly. Yet, the concept still piqued my interest: are certain teams more likely to allow opposing players to have exceptionally good games? If we look at the games played from the start of this season to the All-Star Break, will we see certain teams have allowed more exceptionally good games or outlier games as I will refer to them moving forward?
We first must define in more statistical terms what an outlier game is. I decided that looking for a player's top 10% of games in terms of performance would be ideal. Over the course of 36 or so games that the average team played in the first half of the season, that would average out to around three games for a player that didn't miss any time. This year, COVID has thrown a massive wrench into all that. With the shortened offseason and packed NBA schedule, players are getting injured at a greater rate and are being forced to sit out games as per the contact tracing protocol. Very few have actually played all their team's games, meaning our average is closer to two outlier games per player.
The natural follow-up is what statistic do we specifically look for that 10% in. I decided to use two of them for parallel analysis: points and Game Score. Since I'm looking at this from the perspective of a fan thinking a player is performing better than normal, more often than not that fan is thinking about scoring. But scoring is just one aspect of the game; John Hollinger's Game Score attempts to boil down a player's full stat line (points, offensive and defensive rebounds, assists, steals, blocks, shooting splits, fouls, and turnovers) into a quick one-number metric. This is particularly appealing since Game Score would would be a more holistic measure of statistical performance.
After web scraping game logs from every player from every game before the All Star Break, we can identify the top 10% of games for each player by both points and by Game Score. Then we tally them up grouping by the opponents. Because each team played a varying amount of games (depending on COVID postponements), we should also divide by the number of games so that we aren't penalizing teams for giving up more outlier games by virtue of playing more. Below is a table of the outliers allowed per game by both methods:
At first glance, it appears that generally our two metrics from which we determined outlier games are in agreement. For the teams that do have disparities, I've thought about similarities in defensive scheme and talent and have found no compelling explanation for the observed differences. (If you have a theory, please let me know below.) Nevertheless, this does actually show that some teams allow outlier games at almost twice the rate of some others. Overall, the better defensive teams in the league have much lower rates, with the notable exception of the Blazers who have the third-worst defense in the league.
We should accordingly control for teams' defensive ability for both our outlier metrics. I've done this graphically:
Again we would expect the results to be similar between the two graphs since our initial numbers were also similar. The light gray band surrounding the two trend lines represents the 99% confidence interval of a polynomial regression between Outliers per Game and Defensive Rating. Teams outside the bands are those who either give up significantly more (if above) or significantly less (if below) outliers even controlling for their defensive rating. Three teams are below the confidence intervals on both graphs: the Blazers, Thunder, and Clippers. The two teams to be above both intervals are the Warriors and Mavericks.
Long story short, Curry's intuition of opposing players overperforming against his team is corroborated by the data. To my Warriors (and Mavericks) fans out there, that feeling you have of some "random" player going off on a nightly basis, it's completely justified.
Comments
Post a Comment