There’s an old sports cliche that says “good teams win the close games.” But do they really? If so, how often?
In college hockey, exactly how indicative are the tight games of a team’s moxie?
A sabermetric analysis of one-run games in baseball inspired this statistical foray. Stickball’s story is that on average, any given team’s winning percentage in one-run games will be closer to .500 than its overall success rate. Would the same be true of hockey, which has far less scoring? Only the numbers would tell.
First, a disclaimer. USCHO is not the Elias Sports Bureau, and the venerable ESB doesn’t track college hockey anyway. Thus, the forthcoming figures were painstakingly researched over hours of checking and double-checking. Despite every exacting effort, perfect accuracy can not be guaranteed, and there simply wasn’t enough time or manpower to get a broader sample for analysis. Mock this investigation if you will … but just recall that there’s another old saying that dictates not to knock it if you haven’t tried it.
In that regard, many thanks to a close but unnamed friend for his professional statistical advice, and also to my brother Jeff Sullivan, operator of Lookout Landing, one of the most entertaining and insightful baseball sites out there. Even if it is about the Mariners. (Actually, the team’s blundering haplessness is directly proportional to the hilarity of the content.)
On to the method. First, I collected every Division I team’s records from the past two full seasons (2006-07 and ’07-08). I then categorized each game played as either a “close game” (henceforth referred to as CG) or a “decisive game” (DG). For our purposes, I defined a CG as any game that was decided by one goal, one goal plus an empty-netter, or one ending in a tie. I then scanned the results for differences between the categories.
[Editor’s note: data tables for this article have been posted to Scribd.com and are available for viewing and/or download. Chart 1 and Chart 2.]
While winning percentages in the DGs ran the gamut from four percent (Merrimack, ’06-07) to 95 percent (Michigan, ’07-08), the range in the close games was a much tighter 50-point spread: 24 percent (Western Michigan, ’07-08) to 74 percent (Denver, also last year).
Right away, the CG cluster draws attention (see Chart 1). We’ll find out soon enough what it’s worth. For you statheads, the correlation coefficient for the data set is about .29, and the nature of the relationship is what data analysts call “statistically significant,” meaning that it’s too strong to have occurred by chance.
First to the teams on the ice. Last year’s national champions, Boston College, finished 25-11-8. However, the Eagles only won half of their CGs (7-7-8), going 18-4 in its decisive games. Second-place Notre Dame won 54 percent of its tight contests, compared to 70 percent in the DGs. From the 2007 title game, Michigan State had very little difference (.625 CG, .682 DG), but runner-up BC had a large disparity of over 24 percentage points: .558 to .800.
Note that not only are the CG/DG differences considerable, but that these were also teams that played — and won — at least a half-dozen tight, pressure-packed playoff games on the road to the NCAA finals.
Apart from the title contenders, we look at the extremes. Rensselaer went 10-18-8 overall in ’06-07, and the Engineers only won a single DG out of 15 — under seven percent — but took 26 of 42 CG points for a 55 percentage-point difference. Last year, Air Force (21-12-6) took 16 of 42 points in CGs, but 32 of 36 in DGs. There are plenty of other such examples; these are just the teams at the far corners of the map.
So there is clearly a difference between nail-biters and blowouts when it comes to wins and losses. But what is the difference? Well, this data’s regression formula is: CG win pct = .42 + .15 x DG win pct: when the DG winning percentage changes, the corresponding rate in CG only changes 15 percent as rapidly. But, the close-game rate is spotted 42 points, giving it a big boost. (This formula addresses the “line of best fit,” if you can recall that term from stats class: it’s the line on the chart that falls closest to the most points. If a team had a perfectly average relationship between its winning percentages, its dot would fall dead on the line.)
Looking to the first chart, you can see how the CG points cluster around .500, and how gradually the averages increase when compared to the DGs. The relationship between the variables is simply not very strong.
To simplify the chart a bit, I chopped the teams into six different groups, arranged by ascending DG percentages (see Chart 2). Plotting the six points in the second chart, CG versus DG, it’s clear as day how the results are related: bad teams aren’t as bad in CGs as they are in DGs, and good teams aren’t as good in CGs as in DGs.
So what does this mean? Well, it stands to reason that good teams are good because they score more goals than their opponents, and chances are that they do it in a balanced fashion. It’s highly unusual for a team to come away with a winning record and a negative goal differential (goals scored minus goals allowed) … this would imply that the team is losing by blowouts, but only winning by hairs-breadths.
In other words, legitimately good teams have solid DG percentages, but being good has only the loosest bearing on the outcome of a close contest.
One benefit of this study is to analyze past teams in this manner, and maybe find an elusive explanation for surprising finishes. For example, Northeastern finished 16-18-3 last year despite being 11-5-3 as of January 18. Some may have called the 5-13-3 finish a collapse, but in reality it may have just been a convenient example of regression. The Huskies were 8-2-3 in their first 13 close games, coinciding with — and feeding — their quick ascent, but the Huntington Hounds were only 5-5-0 in CG contests during the slide. The 13-7-3 CG record is still very strong given the 3-11 DG result, but it’s easy to see how NU was running at an early pace that it was highly unlikely to sustain.
But we can also apply our observations to this year’s pack (see the sidebar). The top-ranked Irish are 6-1-2 in tight games, 12-2 otherwise, so they look perfectly legit so far. Highly-ranked Boston University is only 4-3-1 in the CGs, but 10-2 otherwise. Are the Terriers as good as their ranking? That’s not much of a stretch. What about lowly Michigan Tech, at 5-17-2? MTU’s CG record is 5-7-2, leaving the Huskies at 0-10 in the more decisive games. Not good news for Tech fans.
So in the end, there are a few impressions to be taken from this study. First, there is a decided difference between how teams perform in close games and in the rest. Second, it is clear that tight games are a bad team’s best friend … and a stellar squad’s worst enemy.
Obviously, being a good team will aid in winning games of all shapes and forms. Talent isn’t meaningless just because the score is close. But one reason for the moderating effect of CG percentages that shouldn’t be overlooked is that many nail-biters are such because the teams are evenly matched. It wouldn’t come as much of a surprise if either team won a pick-em matchup, so unpredictable bounces are understandably magnified.
There is much, much more work to be done in the interests of quantifying this game. Hockey is so much more fluid than baseball, basketball or football, that breaking down individual moments becomes an inevitably arbitrary act.
However, the information gleaned from this project has painted a picture that should add a little more numerical clarity to our general understanding of how the game is played. It may not resonate at the same frequency as Moneyball, but if it stimulates a new way of looking at the game we all love so much, it will have made its mark in an equally indelible way.
