Slicing and Dicing Pace Numbers
Last time, I kicked off
this little study with an overall look at controlling pace and its role in determining winners. Most “overall” statistics though, aren’t really analysis-friendly. From basic metrics like points per game to advanced measurements like offensive efficiency, we often find at least one filter that allows us to dive deeper.
The simplest example is the home/road split. From following the sport, we have general ideas of a team’s ability. By dividing their season into home/road, a dichotomy may emerge that offers us more information about a team. Historically, the Denver Nuggets are Pepsi Center masters, while they just tread water at lower altitudes. Some splits may be cases of confirmation bias, while many others are certainly conclusive.
When I really think about it, basic overall statistics are becoming antiquated in analysis, across all sports. It’s tough finding true utility in all-encompassing numbers. By applying even one modifier, we glean more insight (while being careful to observe significance, randomness, and sample size). These modifiers, like lineup data or shot locations, are quickly growing in complexity.
The data I used for my initial study was extensive enough that I can apply splits and not diminish the sample, which lets us look a little deeper. Hopefully, we can learn a more detailed story about the efficacy of controlling pace. Remember, the blanket approach yielded a winning percentage of 50.7% for teams that controlled the pace of the game, and 38.2% of games over the past 16 seasons qualified under “pace controlled”. Let’s slice the data.
Turtles v. Cheetahs
We’ll first examine teams by describing their usual pace. I added one condition on the derived pace controller: whether they were slower or faster than the average pace of that NBA season. Table 1 here lists the records of slower teams.
Recall that margin is the percentage difference in pace required to qualify a game as “pace controlled,” a calculation I detailed in part one.
By splitting the data, we find that slower than average teams won 54.3% of their games when they controlled pace (i.e. limited the number of possessions)1. Since this is about to devolve into a messy paragraph of numbers, here’s a summary table of the split.
The data claims there hasn’t been much merit to purposely playing faster. I’d speculate that this is because fast teams typically score often in transition, hence raising their pace. “Running” is an act that all teams can pull off; according to Synergy Sports, the league scores 1.12 points per possession in transition, considerably higher than the average PPP. If a fast team pushed the tempo, a slow team can aptly adjust, to a degree.
Also, a fast team would probably lack the ability to defend competently for the entirety of the shot clock. Upon facing a team with league average pace and elite offensive efficiency, their incompatible modus operandi would hurt them in the half court.
We also learn a second fact: slowing down is more frequent than speeding up, occurring in about 48 additional games per season. Every season but one had more games controlled by slow teams than fast teams. So not only are slow teams ostensibly more effective when controlling game pace; they actually do it more often too.
Lastly, if I set the cutoff point for a “slow” team as the average minus one standard deviation, the winning percentage increases to 55.5%. The uptick makes logical sense, as only extreme teams are in the sample, and teams in the middle 68% area are less likely to use pace control as a strategy. Doing the same for fast teams (mean plus one standard deviation) their win rate ups to 47.7%. The sample size percentages drop to 13% and 11%, respectively.
Davids v. Goliaths
Basketball stats pioneer Dean Oliver once suggested that slowing down a game could increase chances of winning for underdogs. His reasoning was that an opponent would theoretically have fewer possessions to impose their superiority, thus increasing variability in result. It certainly makes sense intuitively, and Dr. Oliver even provided mathematical verification. Can I empirically test it against history?
The main hurdle is establishing the underdog and favorite team in a game. I figured the practical way was to examine Vegas point spreads (nothing more delicious than clashing two databases). They’re a reasonable measure for the difference in skill between two teams, and home court advantages are built in (myth or not – it’s more about perception). Best of all, they are frozen at the point in time, so any information about the team after such a game isn’t accounted.
Covers.com documents point spreads going back to the 2005-06 season, so we’ll be working with a smaller sample here – 8 seasons (up to this year’s All-Star Break). For the non-degenerates, a team with a plus sign, such as +7.5, is an underdog relative to their opponent. In simple terms, the number represents the skill difference. Therefore, a higher number means the two teams are farther apart in perceived skill.
The small numbers, on the other hand, means the two teams are relatively close, so I discarded any point spreads below 3.5. Games with spreads from -3 to +3 may be manufactured from public bias and other awesome gambling technicalities. Essentially, they create cloudy error bars such that we can’t distinguish the favorite with absolute precision.
When a team was getting more than three points (their Vegas point spread closed at +3.5 or higher), I designated them as an underdog. First, let’s look at the outright win rates of those teams without any further filters.
In all, any team tagged as a 3.5 point underdog, or worse, won 26% of their games.
With that in context, now let’s compare it to a specific subset: games where the underdog controlled the pace, and slowed it down by limiting possessions (the same categorization as Table 1 above).
Aggregating the red bars produces a 30.8% winning percentage. The observation confirms Dr. Oliver’s hypothesis: slow pacing underdogs indeed improve their winning percentage by about five percentage points.
Note the red vertical bars; they are higher than their blue parent data set. The significant finding here is that in every year but one, the win rate is better* when reducing the number of possessions. The consistency in each season demonstrates that this isn’t randomness – it’s extremely unlikely to happen seven seasons in a row. The effect of slowing down tempo as an underdog is positive, and compliant with Dean Oliver’s theory.
*I didn’t control for a potential bias here. The distribution of point spreads is not the same every season (i.e. a season may have more 7 point underdogs and less 3 point underdogs, etc.) The difference in win percentage won’t be uniform for this reason. Not to worry though, as the distribution does not vary greatly season to season.
I’ve been diving deeper and deeper into my original data set, so you’ve probably noticed that my sample size has dramatically shrunk to 6.6% of all games, as indicated by the color-corresponding lines. It’s possible that I’ve compromised my sample by allowing randomness to invade it. But 6.6% is still 581 games, a substantial amount. To be safe, I expanded the sample and found the same conclusive results2.
For all games in which underdogs controlled the pace, speeding up actually occurred more often than slowing down, counter to Dr. Oliver’s advice. Deservedly, their winning percentage was nearly nine percentage points lower.
It’s nice when intuitive ideas pan out empirically, isn’t it?
What We Learned
Let’s resurface from that tide of pace data we just submerged in. What can we do with this information? The initial instinct is that any team preparing for a game can better their chances by slowing down the pace. There are a few points to consider though.
First, this would only prove successful should the underdog team’s natural pace be slow, and their opponent plays at a speed closer to league average. Remember that by my definition, two teams that both run 85 possessions per game would be excluded from this study. We’re looking for pace mismatches and whether an underdog capitalized on it.
Second, to expand on Dr. Oliver’s theory, there is eventually a point when a team’s chances can’t be salvaged by any strategy. If Vegas lists you as a 15 point underdog, your probability of winning the game is as likely as a Steve Novak lay-up. So let’s consult a final chart, where we list the outright win rates by point spread:
As the spread passes +5.5 points, the slow pace idea fails to keep up (and our sample diminishes). In other words, the discrepancy in team skill overtakes any edge gained from implementing such a strategy. But observe the underdogs of fewer than 6 points: the virtues of limiting possessions are quite evident. Obviously, the smaller sample sizes cause some variation in nailing down the true advantage, but for such games, we see their winning percentage raised six percentage points, from 36% to 42%.
I started this study with a simple question, based on the verbal spouting of TV commentators. Does controlling pace influence the game’s result? The answer is, as it commonly is, sometimes yes, sometimes no. In certain contexts, it’s an agreeable tactic. That 6% I found above is a clear edge for moderate underdogs, significant enough to be employed, as long as it doesn’t weaken the other gears in the team’s machine3.
To rephrase, pace control is a key tactic for some teams and a non-factor for others. It could be a product or accident of both teams, or the deliberate enforcement of one. Identifying those teams, or specific games, could comprise a part three of this study, providing examples to demonstrate these ideas.
It’d also provide some interesting betting possibilities. A four point underdog typically has money line odds of around +150, for an implied probability of 40%. Based on the table above, if they in fact win 43% long-term, then a market inefficiency exists. In order to exploit it, one would have to predict when an underdog team applies a slow pace approach. Those could be avenues to revisit at the end of the season.
We should remember that despite these findings, every team is fitted for a certain style and pace, and they all have their limitations. The best teams are flexible; using an array of lineup combinations, they can adjust, and mitigate the advantages of their opponents’ game plans.
Footnotes and outtakes
1. The binary nature of wins and losses does cause evil randomness in the results. A team that controlled pace might’ve lost on a last second shot. This would go both ways, but in any case, a close game with such a result meant the pace control wasn’t very effective anyway. For completeness, I calculated net rating of the controlling team (instead of wins/losses), and averaged them for each season. The correlation coefficient between that and winning percentage was a very strong 0.861. I’ve used the latter for this study, as it’s far more interpretable.
2. 3.5 points as a threshold for weeding out “similar strength teams” can be moved to increase sample size. I tested at +1.5 points (thereby upping the sample to 754 games) and found that overall underdogs won 30.1% of games and ones with slow pace won 35.3%. Like the +3.5 case, the difference in win rate is approximately five percentage points. Remember, it’s not about the win rates; the idea is comparing them relative to one another.
3. Keep in mind that no matter how I split the data, I shouldn’t expect major increases in win percentage. The four factors are the stalwarts of success, so while pace may affect the equation in certain spots, a dramatic bump would defy convention. I confirmed this by computing a “Pace Comfort Level”, which is exactly what it sounds like*, so feel free to skip the following calculation.
*Take the team that is being forced to play outside their typical pace. Calculate the percentage difference between that and the game pace. A higher number represents a controlling team pacing the game farther away from its opponent’s usual pace, or comfort level.
On games with a defined team controlling pace, I ran a linear regression on offensive rating with the four factors as the independent variables, yielding an anticipated 0.96 R-squared. Then, I added “Pace Comfort Level” to the model. While the estimated coefficient was positive and significant, the increase in R-squared was trivial.
* If you’ve braved this entire piece, the final nugget I discovered was that home teams controlled pace more often – over 60% – but it didn’t equate to winning more often.
* Thanks to Ian for his thoughts as I constructed this study.