Nylon Calculus: Dispelling playoff myths
The regular season is more than halfway complete and teams are now turning for home. The horse race is sorting itself out, as the frontrunners for the title emerge. We can begin to anticipate which teams will make the playoffs and even fantasize about possible matchups.
Still, there’s only so much a team’s regular season record can tell us about playoff prospects. Just ask the Raptors. Ask Chris Paul.
When confronted with such an uncertain future, we tend to seek comfort in the well-worn myths of the postseason. And what better caretakers for the NBA’s institutional memory could we ask for than the players, coaches, and broadcasters who have experienced past championship victories and defeats most viscerally? These guys KNOW what works in the playoffs and they have all the clichés to prove it.
But which of our most-trusted playoff axioms can hold up under the scrutiny of a systematic analysis? Which of these truisms can actually predict playoff outcomes? Will defense really win you championships? Do elite teams actually have a switch to flip?
Let’s find out.
A simple model of playoff outcomes
There are a variety of different approaches available for modeling playoff outcomes. I opted to use ordinal logistic regression to estimate the probability that a team would win a given postseason series (in 4, 5, 6, or 7 games) or lose it (in 4, 5, 6, or 7 games). I examined the period from 2003 to 2017 during which the playoffs consisted of four rounds of seven-game series (i.e., the no. of observations = 15 postseasons x 15 series per postseason x 2 teams per series = 450). For explanatory variables, I compared each team’s regular season performance with that of their playoff opponent. I evaluated the predictive power of various regular-season team stats based on model fit using cross validation (training models with 90 percent of the data and using the other 10 percent for testing with the random 9:1 partitioning repeated 1,000 times).
My base model used a team’s regular-season advantage (or disadvantage) in the Simple Rating System (SRS) to assess the likelihood of the eight playoff series outcomes (W4, W5, W6, W7, L7, L6, L5, L4). The Simple Rating System reflects a team’s average margin of victory adjusted for its strength of schedule. It’s a useful measure of team quality and strongly predictive of playoff success.
So, for example, in the first round of the 2016 Western Conference Playoffs, the 73-9 Golden State Warriors had a regular-season SRS of plus-10.4 compared to the plus-0.3 SRS compiled by their opponent, the 41-41 Houston Rockets. That’s more-or-less a plus-10 point-per-game advantage in SRS for the Warriors, which — as is evident from the chart above — the base model suggests is equivalent to a 97 percent series win probability (and a 59 percent chance at a sweep).
The Simple Rating System is pretty closely tied to a team’s record and — by extension — its playoff seeding. You can separate the effect of home court advantage by including it as an additional term in the playoff odds model. For two playoff opponents with equivalent regular-season SRS, home court advantage tilts the series win probability to a 56-44 advantage.
Along the same lines, I evaluated a variety of regular-season team stats available from Basketball-Reference as possible predictors of playoff outcomes: offensive rating, defensive rating, pace, 3-point attempt rate, and the “four factors” (on both sides of the ball) — eFG%, rebounding rate, free-throw rate, and turnover rate. In each case, the models included SRS plus one additional team stat. I compared each new model’s fit to that of the base model to test whether any of the team stats was predictive of playoff success. Finally, I contrived some relationships between these statistical models and a few playoff myths of note to make my post more interesting.
Testing the myths
We’ll test five different myths. The first one is a Charles-Barkley Special, but it runs deeper than Chuck. There’s a whole series of related varieties: “Live by the 3, die by the 3”, “3-point shooting offenses have more variance”, “I don’t like jump-shooting teams”…you’ve heard it before.
So, does a team’s propensity for jump shooting — or, more specifically, 3-point shooting — confer a risk for an early playoff exit?
To answer this question, I built a playoff odds model which included terms for SRS and 3-point attempt rate. Again, I used the difference in regular-season SRS between a team and their playoff opponent to predict outcomes. But, here, I also tested the effect of the difference in regular-season 3-point attempt rate between each team and its playoff opponent. This sort of model allows us to “control for” team quality while evaluating the predictive power of regular-season 3-point attempt rate.
The verdict? I found that, given two evenly-matched playoff opponents (i.e., two teams with no difference in regular-season SRS), the team who shot 3-pointers at a higher rate during the regular season was more likely to LOSE the series. But, the effect was small. For example, a difference in regular-season 3-point attempt rate of plus-8 percentage points translated to a 48 percent series win probability (note: a plus-8% difference in 3PAr is at the high end of the actual differences observed for evenly-matched playoff opponents over the past 15 years). This magnitude of a shift in playoff odds was less than what we estimated earlier for home-court advantage. Moreover, this model which included regular-season 3-point attempt rate did NOT fit the data any better than the base model (with SRS only) during cross validation — suggesting that 3-point attempt rate offers limited power in predicting playoff outcomes.
In other words: jump-shooting teams really CAN win championships! Hooray.
Bear Bryant was talking about Alabama football when he coined this famous championship axiom, but it gets applied to the NBA all the time, too. I re-used the same model framework from above to assess the recent validity of Bear’s claim — this time using SRS and defensive rating.
I found that given two evenly-matched playoff opponents (i.e., two teams with no difference in regular-season SRS), the team with the better (lower) regular-season defensive rating was more likely to WIN the series. Once again, though, the effect was small. For example, a difference in regular-season defensive rating of minus-5 points per 100 possessions translated to a 52 percent series win probability (Note: a 5-point difference in DRtg is at the high end of the actual differences observed for evenly-matched playoff opponents over the past 15 years).
It’s worth mentioning that — if there is no difference in SRS — when one team has an advantage in defensive rating, the opponent would be expected to have a more-or-less equivalent and counterbalancing advantage in offensive rating. Indeed, when I modeled offensive rating in the place of defensive rating, I found the complementary result: a plus-5 point per 100 possession advantage in regular-season offensive rating yields a 48 percent series win probability.
This magnitude of a shift in playoff odds was similar to the effect we observed for 3-point attempt rate. And these models — which included regular-season defensive or offensive ratings — did NOT fit the data any better than the base model (with SRS only) during cross validation, suggesting a lack of predictive power.
The model suggests that if two playoff opponents are evenly-matched, the more defensive-minded team holds NO particular advantage in the series. In other words: defenses certainly win championships, but offenses do too.
I’m not sure if this could really be considered a cliché. I feel like it’s a pretty commonly held belief, though — that the seven-seconds-or-less Suns offense was a gimmick,and the team was never a legitimate championship threat.
To clarify, I’m not evaluating whether pace changes in the playoffs — teams definitely DO play slower in the postseason (Note: I observed an average decrease in pace of 2.7 possessions per game across 225 playoff series). What I wanted to evaluate was whether teams who played at a faster pace during the regular season were at an advantage or a disadvantage in the playoffs.
I found that given two evenly-matched playoff opponents (i.e., two teams with no difference in regular-season SRS), the team with the faster regular-season pace was more likely to LOSE the series. Like the other stats, though, the effect of pace was very small. For example, a difference in regular-season pace of plus-3.5 possessions per game translated to a 49 percent series win probability (Note: a plus-3.5 difference in pace is at the high end of the actual differences observed for evenly-matched playoff opponents over the past 15 years). Based on cross validation, regular season pace was NOT predictive of playoff success.
In other words: fast-paced teams aren’t doomed.
There’s not a stat that I know about which is a definitive measure of a team’s physicality. But, one reasonable surrogate for physicality might be defensive free-throw rate. A team that fouls a lot might also be described as physical, right?
I found that given two evenly-matched playoff opponents (i.e., two teams with no difference in regular-season SRS), the team who allowed a higher regular-season free-throw rate (i.e., the more “physical” team) was more likely to LOSE the series. For example, a difference in regular-season free-throw rate of plus-4 percentage points translated to a 46 percent series win probability (Note: a plus-4 percentage point difference in free-throw rate is at the high end of the actual differences observed for evenly-matched playoff opponents over the past 15 years). Regular-season free-throw rate was not predictive of playoff success based on cross validation.
In other words: physicality is overrated.
And, finally, we come to a myth that I won’t try to undermine.
If we take a team’s turnover tendencies as an indication of their attention to detail, we can test the flip-switching myth using a playoff odds model of regular-season SRS plus regular-season turnover rate.
I found that given two evenly-matched playoff opponents (i.e., two teams with no difference in regular-season SRS), the team who had a higher regular-season turnover rate was more likely to WIN the series.
For example, a difference in regular-season turnover rate of plus-3 TOV per 100 plays translated to a 65 percent series win probability (Note: a plus-3 difference in turnover rate is at the high end of the actual differences observed for evenly-matched playoff opponents over the past 15 years). Of all the stats evaluated, turnover rate had the biggest impact on playoff odds — and it was the only stat which improved the fit of the base model (SRS-only) in cross validation, albeit by a small margin (1 percent improvement in r2).
The data appear to suggest that sloppy regular-season teams leave themselves room for improvement in the playoffs. In general, teams cut back on turnovers in the playoffs (Note: I observed an average decrease in turnover rate of 0.8 TOV per 100 plays across 225 playoff series), but teams with larger regular-season turnover rates tended to make bigger reductions during the playoffs.
In other words: Some teams CAN flip the switch in the playoffs.
I wanted to look for more evidence of switch-flipping — so I checked to see if teams’ peak performances over short stretches of the regular season could help predict playoff success (in addition to SRS averaged over the whole season). That is, should we expect that a team like the 2018 Cavs — who have had some high points and some low points this season — will have more or less playoff success than a more consistent team with the same average SRS? To find out, I tested the effect of peak net ratings (ORtg-DRTg) from the regular season on playoff odds. I looked at rolling averages of a team’s net ratings for stretches of games of varying length. Ultimately, a team’s performance over a period of 20 to 25 games added the most information when paired with full-season SRS.
I found that given two evenly-matched playoff opponents (i.e., two teams with no difference in regular-season SRS), the team with the better 20-game stretch (i.e., a higher peak of average net ratings during any 20 games of the season) was more likely to WIN the series. Total SRS and peak 20-game net ratings are highly correlated (r = 0.89), so the effect sizes are not really meaningful; but, it’s interesting that SRS + peak net rating is marginally more predictive of playoff odds than SRS by itself. Basically, a team with a higher ceiling and lower floor in the regular season will tend to beat a higher floor/lower ceiling opponent in the playoffs.
In other words: Flip. That. Switch.
Predicting 2018 outcomes
For fun, let’s try to apply our model to some highly anticipated 2018 playoff matchups. Based on current regular-season stats, if the Warriors face the Rockets in the Western Conference Finals, they would have an SRS advantage of plus-1.16 and a higher turnover rate by 0.8 TOV per 100 plays. The model predicts the most likely outcome is Golden State in six, with an overall series win probability of 64 percent for the returning champs. Sounds reasonable.
But, bettors beware. The models described above have some serious blind spots. They don’t account for regular-season injuries. They don’t incorporate assumptions about team quality from previous years (based on team or player stats). And they probably don’t pay enough attention to star power, which is especially important in the playoffs.
You can see the limitations of this approach when you unleash the model on a potential Cavs-Raptors playoff matchup. Cleveland has an overwhelming minus-7.37 point deficit in SRS (with a plus-1.0 TOV per 100 plays higher turnover rate), which translates to a measly 9 percent series win probability and 34 percent chance of a sweep for Toronto. But, what about the LeBron effect?
You can see in the table above, LeBron’s teams have tended to outperform playoff expectations – winning more convincingly and losing more competitively than their regular-season SRS stats would suggest. We can formalize this trend by adding a LeBron term to the base model. Doing so, I found that given two evenly-matched playoff opponents (i.e., two teams with no difference in regular-season SRS), the team with LeBron was more likely to WIN the series (76% percent series win probability).
Next: Nylon Calculus -- How does missing a rookie season affect development?
Still, even after accounting for the power of playoff-LeBron, the model considers the Cavs underdogs against the Raps (22 percent series win probability). Will this really be the year when LeBron loses his stranglehold on the Eastern Conference, then? Well, I’ll keep believing one last myth — that LeBron makes the Finals every damn year — until I see some evidence to the contrary.