Below is this week’s article for the Toronto Star, and at the bottom is a link to the pertinent regression results. I think the article more or less speaks for itself, but I did want to add a couple of additional thoughts.
First, even though the data set included all Finals and Semi-finals since 1980, it’s still a relatively small sample. That said, the GFDiff relationship was significant at the 1% level and the GADiff was significant at the 5%, so the numbers appear solid.
Second, at some point I’ll add in data for all playoff series and/or for all teams and we’ll see how that affects things. And/or run separate regressions for the early playoff rounds or the regular season. If I had to guess, I’d guess that if anything the GFDiff coefficient would increase if we added regular season and early playoff rounds because by-and-large teams reaching the final 4 are going to be stronger defensively and strong offenses are more likely to thrive against weaker defensive teams. Of course, the reverse is going to be true as well. However, because teams with strong defenses but not-so-strong offenses almost by definition play a lower-event style, they’re going to be more susceptible to short-term randomness — think the NJ Devils at the end of 2012-13 when they just couldn’t buy a goal even though they played very good defense. (Another way to think about this is that more offense generally means more opportunities, thereby reducing game-to-game variance).
Third, it would be interesting to run the numbers for just the post-salary cap era. The problem with doing that is sample size. But in light of the fact that teams generally pay much more for offense than defense, it occurs to me that putting together the type of team that can offensively dominate is likely more difficult, and almost certainly more difficult to sustain over multiple years, than it was previously. Defense is going to be easier to repeat (e.g., it’s not subject to the high variance that shooting percentage is), and I expect it’s also easier to maintain under the salary cap. From that perspective, then, I wouldn’t be surprised if a really good defensive team might not have a slightly lower probability of winning the Cup in any single year, but would be more likely to maintain its performance, and so over a period of several years in the aggregate it would have a better overall probability of winning the Cup than a team that has huge offensive numbers for a year or two and then either (a) regresses to the mean and/or just has an off year, or (b) is forced to re-tool for cap reasons.
With that, here’s the article:
“Defense wins championships.”
This might well be the most widely used truism in sports. Hockey, football, basketball, soccer: “Defense wins championships.” It’s so ingrained that it’s probably the mantra of checkers players too.
Implicit in this seemingly universal belief is that in the last rounds of the playoffs, when competition stiffens, offense may fizzle or be neutralized, but stout defense will hold the line and carry the day.
Everyone seems to believe. But is it true?
Obviously there’s something to it. Teams that treat their nets like storage lockers for their opponents’ hockey pucks aren’t likely to get too far in the playoffs.
Then again, as reflected in the table, it’s not as if the team with the best defense wins the Cup every year. Far from it.
Since 1980, only five teams that finished first overall in goals against in the regular season won the Cup. There were exactly the same number of Cup winners who finished first in goals for. So between offense and defense being the ticket to playoff glory, that’s a push.
But surely even if they weren’t first overall, virtually all Cup champs have been strong defensively, right? Not so. It turns out that since 1980, out of 33 Cup winners seven were tenth or worse in regular season goals against. Sure, four of those teams happened to have players named Gretzky, Messier, Kurri, Coffey, Lemieux, Jagr, and Francis. But hey, those still count.
On the other hand, only four teams that were tenth or worse in regular season goals for won the Cup, suggesting that having a strong offense might well be more of the recipe for success than stifling defense.
That’s all fine and dandy, but it’s hardly scientific. Which means it’s time to put the fancy in fancy-stats.
A regression analysis is a powerful statistical tool for estimating the relationship between variables; for example, how team defense affects the probability of winning a playoff series. I ran a species of regression called a “probit” to estimate the impact of team defense and team offense on the chances of winning the Cup.
Team defense was measured using regular season goals against. Team offense was measured using regular season goals for. For the two teams playing in each Finals and Semi-finals since 1980, I calculated the difference between them in goals for and goals against to determine each team’s defensive and offensive strength relative to the other.
For example, in 2012, the last full NHL season, the Kings played the Devils in the Finals. The Kings scored a paltry 194 goals that year, but allowed only 179. The Devils scored 228 and allowed 209. So the Kings’ “goals for differential” was -34 (their 194 goals minus the Devils’ 228). In other words, the Devils’ offense was 34 goals better than the Kings’ that year. The Kings’ “goals against differential” was -30 (179 minus 209), meaning the Kings allowed 30 fewer goals than the Devils.
Using teams’ goal differentials, the probit regression estimated the contribution of team defense and team offense to the outcomes of each of the 99 Finals and Semi-finals series since 1980.
The results: Team offense actually had a slightly greater effect on the likelihood a team would win. Each additional goal for above the opposition increased a team’s probability of winning the series by 0.46 per cent. For every goal allowed fewer than the opposition, a team’s probability of winning increased by 0.37 per cent.
Those numbers might not sound particularly big, but the differences in goals for and goals against between teams in the Finals are often 30 or more, and even reached as high as 125 in one instance (goals for differential between the 1983 Oilers and Islanders).
What this tells us is that when it comes down to crunch-time, it doesn’t matter much whether a team is an offensive powerhouse or constructs the stingiest of defensive shells. Score lots. Allow few. Both are almost equally good. What matters, at least as far as team offense versus team defense is concerned, is plain old goal differential.
|Year||Stanley CupChampion||Regular SeasonGoals Against||GA Rank||Regular SeasonGoals For||GF Rank|
* Note: Due to the fact that since 1980 NHL seasons have been 48 games, 80 games, 82 games, and 84 games, all goal totals have been normalized to an 82 game season to allow an “apples-to-apples” comparison.