Defense Wins Championships? No, Offense And Balance Do | Stat Geek Idol

This is a Sweet 16 submission in our inaugural Stat Geek Idol contest. It was conceived of and written by Adrian Atkinson of Tobacco Road Blues and HoopSpeakU.

‘Defense wins championships.’

How many times has Billy Broadcaster uttered that cliche or Jimmy Journalist penned it? And they’re getting it straight from the top, you know. Head coaches (both at the collegiate and professional levels) have been espousing this conventional wisdom for generations.

From a basketball perspective, it is easy to see why coaches preach this mantra: it is simply harder to get maximum effort on the defensive end of the court than on the offensive end (so the ‘championships’ carrot is dangled to entice solid two-way exertion).

While it may be effective as a motivational instrument in a coach’s toolbox, is there any evidence that defense does, in fact, win championships?

In reality, study after study has revealed that it takes a combination of very good (to great) offense and very good (to great) defense to win college basketball titles. “Defense wins championships” is not technically incorrect—it’s just incomplete. One could just as easily state “offense wins championships” or, most accurately (but least pithily), “a combination of offense AND defense wins championships.”

The analysis in this paper will help to shed additional light on the relative roles of offense and defense in winning NCAA Tournament games. Let’s start with a brief description of the methodology.

Splitting Team Into Tiers Based On Their Efficiency

All NCAA Tournament teams from 2003 to 2012 (i.e., the KenPom era) have been categorized into a 5 x 5 matrix of (offense, defense) pairings. Based on its adjusted offensive and defensive efficiencies, each team is placed into a tier ranging from 1 (elite) to 5 (poor). Specifically:

Tier 1: teams ranked 1-10 in adjusted efficiency
Tier 2: teams ranked 11-25 in adjusted efficiency
Tier 3: teams ranked 26-50 in adjusted efficiency
Tier 4: teams ranked 51-100 in adjusted efficiency
Tier 5: teams ranked 101 or worse in adjusted efficiency

For example, a team ranked in the top ten in both adjusted offensive and defensive efficiency (like Ohio State, Michigan State, and Kentucky this season) would be labeled as a (1,1). A team like 2012 Missouri (top ten offense, sub-100 defense) would fall into the (1,5) bin. Historically, most 15 and 16 seeds are categorized as (5,5), (5,4), or (4,5).

Table 1 provides a 25 x 25 matrix summarizing the NCAA Tournament win-loss record for all possible match-ups between tiers. For example, (1,1) teams have a record of 73-12 in NCAAT games since 2003. The (1,1) row of Table 1 shows the win-loss breakdown for this type of team against all other (offense, defense) combinations.

(click to enlarge)

Borrowing from the economics concept of strong and weak stochastic dominance, the cells in Table 1 have been color-coded to represent their level of dominance. The main diagonal consists of match-ups between identically-tiered teams (e.g., a (2,2) vs. a (2,2)). Moving up and to the right from the main diagonal, dominant teams begin to emerge (and, moving down and the left, dominated teams are revealed in a mirror image).

A strongly dominant team is a member of better offensive and defensive tiers than its opponent (e.g., a (2,2) strongly dominates a (3,3) or a (5,4)). A weakly dominant team belongs to one better tier than its opponent with the other tier being equal (e.g., a (1,2) weakly dominates a (1,4) or a (1,5)). In match-ups containing no dominance, one team is in a superior offensive tier with the other belonging to a superior defensive tier.

From 2003 through the opening weekend of the 2012 NCAA Tournament, 43% of all NCAAT games have involved a strongly dominant team. Those teams have won 91% of those contests (240-24). An additional 29% of all NCAAT games have included a weakly dominant team, with those squads winning 68% of the time (120-57). Four percent of all match-ups paired identically-tiered teams.

What Dominance Patterns Imply About Offense Vs. Defense

Getting back to the article’s thesis, what can Table 1 tell us about the relative importance of offense versus defense?

In match-ups involving a weakly dominant team, the team in the better offensive tier (when defenses are in the same tier) wins 67.3% of the time. The team in the better defensive tier (when offenses are in the same tier) wins 68.4% of such games.

Not much evidence here of offense or defense being more crucial than the other, but things get a little more interesting when examining the games in the “no dominance” category. In these games (that involve one team with a superior offense versus another with a superior defense), we can sum the offensive tier and defensive tier to assess which team is stronger overall. For example, when a (1,3) matches up against a (4,2), the (1,3) has the stronger adjusted efficiency profile despite lacking either strong or weak dominance.

In 48 of the 142 “no dominance” games, the superior offensive team also has the stronger overall profile. Those teams have won 70.8% of those contests. In 44 games, the superior defensive team has been the stronger overall team, winning 70.5% such match-ups. In the remaining 50 “no dominance” games, the teams have been equally matched from an overall efficiency standpoint—for example, a (1,2) vs. a (2,1) or a (2,5) vs. a (4,3). In these match-ups, the stronger offensive team has won 60% of the time.

In an admittedly small sample, (1,2) teams have won seven of eight match-ups against (2,1) teams. In these typically high-profile games (usually Elite 8 or Final Four showdowns), elite offenses with very good defenses have clearly gotten the better of elite defenses with very good offenses. Score a point, perhaps, for the “offense wins championships” arguments.

Actual vs. Expected Wins Based On Offense/Defense Combos

Table 2 examines the average profile for each of the 25 (offense, defense) combinations. Expected wins are calculated using seed expectations based on 2003-2011 data. For example, the average No. 1 seed has won 3.33 NCAAT games, and the average No. 6 seed has won 0.94 NCAAT games. Table 2 lists the average seed for teams within each (offense, defense) combination.

(click to enlarge)

Using the distribution of seeds and their associated seed-expected wins, expected wins for each combination can be calculated. Using this method, teams in the (1,1) category are expected to win 3.05 NCAAT games on average. Teams in the (3,3) category are expected to win 0.86 on average.

Looking at the difference between actual NCAAT wins and expected NCAAT wins is another way to assess the relative merits of strong offenses versus strong defenses. As seen in Table 2, teams in the (1,1) and (1,2) categories are clearly the big winners in terms of exceeding seed expectations.

Pairing an elite offense with an elite/very good defense is a time-proven recipe for March success. However, pairing an elite offense with a merely good (or worse) —Tier 3 or below)—defense leads to underperformance in March. Likewise, pairing an elite defense with a non-elite offense also tends to led to falling short of expectations in the NCAA Tournament.

Overall, there are 208 teams in this sample who were in a stronger offensive tier than defensive tier. Those teams won 1.04 NCAAT games on average versus an expected average of 1.11—or about 6% worse than expected. The 194 teams who were in a better defensive tier than offensive tier won an average of 0.88 games as compared to an expected average of 1.03—or about 15% worse than expected.

This provides more evidence that offensive efficiency is more important than defensive efficiency to a team’s ultimate NCAA Tournament success.

The Case for Balance

With both offensive-dominant and defensive-dominant teams falling below their seed-expected win totals, some category of teams must be exceeding theirs.

Indeed, teams that fall into the “perfectly-balanced” class (e.g., (1,1), (2,2), (3,3)…) have exceeded their win expectation by 18% (an average of 1.03 actual NCAAT wins versus an expected average of 0.88). Even when throwing out the thoroughly dominant (1,1) teams (whose seed-expected wins are probably underestimated since all No. 1 seeds are considered—not just the (oftentimes) cream of the crop (1,1) No. 1 seeds), the remaining balanced teams exceed their seed-expected wins by 11%.

Looking at Table 2, one can see that the main diagonal (i.e., perfectly balanced) teams have fared well in exceeding/meeting expectations. As one moves farther and farther from the main diagonal (i.e, becoming more and more unbalanced), teams underperform their seed expectations more dramatically.

Balance Rules, But If You Must Choose, Choose Offense

So what are the take-home messages from this analysis?

1.) You still must be very good on both ends to win big in March.

2.) But if you can’t be elite on both ends, it is much, much better to be elite offensively (and very good defensively) rather than elite defensively (and very good offensively).

3.) In games involving teams of similar overall strength, the one with the stronger offense will win more frequently in March.

4.) Offensive/defensive balance is important; it is hard to win big in March if one unit is significantly worse than the other (right, Duke and Missouri?).