For the statistically oriented, base concepts like “clutch” and “momentum” generally aren’t a thing. Instead, a different sort of conventional wisdom rules the day: Margins of victory mean more than wins; results are largely random walks of possessions; series are largely random walks of games; and playoff games are like regular games between stronger teams. Despite encouraging overt oversimplification, these ideas remain compelling because metrics built with them in mind have been pretty good at making predictions in many contexts across several sports and in the NBA’s regular season.
But the NBA playoffs are a different beast entirely. There, many of the traditional ways of thinking — like the philosophy that the playoffs are where winners win and where champions win championships — turn out to be true.
A few years ago, I came up with a heuristic for picking NBA champions that I call the “Five by Five” method: Of teams that are 1) within five games of the best win-loss record in the NBA and 2) have won a championship within the previous five years, pick the one that won a championship most recently. If no team at the top has a ’chip, pick the team with the best record.1 This correctly picks the eventual NBA champion in 21 of 32 seasons (66 percent) since 1984 (including three of four since I came up with it).
The Golden State Warriors, of course, entered these playoffs with a better résumé than the typical favorite. Defending champions? Check. Best record in NBA history? Check. No teams within five wins in the regular season? Check. Yet, playoff models that rely heavily on margin of victory and game-by-game simulations — like ESPN’s Basketball Power Index and FiveThirtyEight’s Elo projections — gave the Warriors less than a 45 percent chance of repeating.
The Cavs beating the Warriors in this year’s NBA Finals would be an upset no matter how you look at it. FiveThirtyEight’s projections give the Warriors a 69 percent chance of winning the series. But if we factor in the conventional wisdom — not something we say here very often — the Warriors look even stronger.
In particular, I believe there are three clichés about the playoffs that the conventional wisdom gets right and the unconventional wisdom gets wrong — and they’re all good news for Warriors fans:
1. The playoffs are a whole different ballgame
The obvious problem with using models built predominantly on regular-season data to predict postseason games is that the playoffs have demonstrably different dynamics. For one thing, both teams try very hard to win each game. When it’s win or go home, there’s no more jockeying for better playoff (or lottery) position, no more using throw-away games to experiment with new lineups, the best players tend to get more minutes per game and no one’s resting on the second night of (non-existent) back-to-backs.
For the most part, this has one inescapable effect: The better team wins more often. For example, modeling playoffs separately suggests that a team that is projected to win a regular home game against a particular opponent 66 percent of the time under our Elo model is likely to win an equivalent game 69 percent of the time in the playoffs. Over a seven-game series, those small differences add up: The odds of that team with home-court advantage winning a seven-game playoff series against the same opponent rise from 57 percent to 63 percent.
But that’s just the beginning. Better teams win individual games more often, but they also win series more often than their game-winning rate suggests. In other words, there is skill to winning playoff series, and better teams are better at this skill. We can see this by tweaking how we make our projections. Normally, to project a series, we first model how much of a favorite a team is in each game (either sequentially or all at once) and then calculate its odds of winning four out of seven. But an alternative, less granular method is to take the same inputs (like team strength), ignore individual games and just model the odds of each team winning each series. The fact that this method produces very different — not to mention more accurate — results tells us something about NBA series dynamics. Here are the projected series win rates for teams with home-court advantage using three different Elo-based methods:
The model that projects series results directly tends to add 8 to 10 percentage points to the better team’s chances (though of course this diminishes a bit as the gap between teams gets larger). Coming into the Finals, the Warriors are 64 Elo points ahead of the Cavs, making that difference 9 percentage points — a fairly significant boost to Golden State’s outlook.
2. Winning is everything
Or, more shockingly for stats-types: Margin of victory is useless.
Virtually any model worth its salt long ago abandoned uncouth metrics like wins and losses in favor of more nuanced metrics like margin of victory. But such proxies are only valuable so long as they’re a more reliable predictor of winning than winning itself. In contexts like baseball (where results are fairly close to a random walk) or football (where sample sizes are always small), looking predominantly at differentials in runs or points may be a sound analytical strategy. But in many sports — and especially basketball — winning is a demonstrably discrete and valuable skill independent from scoring and allowing points. This is why, as I’ve written before, a good model for predicting outcomes should account for both a team’s winning percentage AND its margin of victory.
How much you weight each one depends on the amount of data you have and what you’re trying to predict.
In the NBA playoffs, the skill of winning is especially important because — and this is seriously one of my favorite things in all of sports — being good at “winning” manifests both in playoff games and playoff series. “Winning” teams not only win playoff games more than their margin of victory implies, but they win playoff series even more often than that.
Thus, seemingly unlikely things like the Warriors’ coming back from 3-1 to beat the Thunder in seven games happen significantly more often for teams with a demonstrated knack for winning.
Overall, here’s how the chances of winning a seven-game series change when you adjust the relative weight of win rates and margin of victory to account for playoff dynamics:
Note how stratified the regular season coloring is: For any given level of margin of victory, having 10 more wins only affects the odds of winning a (theoretical) series by 5 to 10 percentage points. In the playoffs, on the other hand, the ranges are staggering. If team A is a net of 2 points per game better than team B, its chances of winning the series may vary by as much as 30 percentage points depending on the teams’ relative win-loss records.
3. Champs gonna champ
Championship experience matters. A lot. And recent champs win way more than similarly situated non-champs.
Winning an NBA championship is really hard. And surprise: Teams that pull it off and then keep playing at a high level are likely to do well in the playoffs.
Indeed, having demonstrated an ability to win a championship continues to be a strong indicator of a team’s ability to compete for another for several years. Below is a chart that shows all teams since 1984 that finished within five games of the best record in the regular season arranged by their winning percentage and how many other teams finished within five games of the best record as well (our proxy for competition). The differences between those that had won a title within the previous five years (our proxy for championship experience) and those that had not are withering:
Nine of the 10 teams that had the best record free and clear (no teams within five games) won the championship (including all five with championship experience). That leaves 22 seasons in which no such team existed. In 12 of those 22, the title was won by a team with championship experience (out of 29 such teams “competing”, or 41 percent); in seven, the title was won by a team with no championship experience (out of 50 such teams competing, or 14 percent). Only three seasons had champions that finished more than five games back (of 273 such playoff teams in those years, or ~1 percent).
Of the 16 teams that had the best record in the league, at least one challenger within five games and no championship experience, only two won the title — and both times it was the San Antonio Spurs.
All clichés considered
To show how much these clichés potentially affect real predictions, I’ve created a series of models to predict the outcome of the NBA Finals series between the Warriors and the Cavs using different methods and sets of variables discussed above (for championship experience, I introduced a “has more recent championship experience within past five years” variable2). Here are the results, with methods ordered from least to most predictive:3
|PROJECTION BY||USING||CHANCE OF WINNING|
|Game||Elo difference — regular-season dynamics||69.0%|
|Game||Elo difference — playoff dynamics||74.0|
|Series||Elo difference and win difference||90.2|
|Series||Elo difference, win difference and championship experience||94.2|
|Series||Championship experience and win difference||95.0|
Obviously, this doesn’t account for particulars like the Cavs turning it on or that a twice-injured Steph Curry has looked shaky at times. And as of this writing, betting lines seem to agree with the standard models, generally installing the Warriors as a slightly better than 2-1 favorite. But if history is any kind of guide this year, the unconventional wisdom may be significantly understating the Warriors’ chances.
Check out our latest NBA predictions.