By IJay Palansky and Phil Curry
In sixteen NHL cities that aren’t Toronto, much of the talk over the past few weeks has been about the “momentum” each team carries into the playoffs. During lulls in the game when color commentators are desperate for something passably intelligent to say, they trot out the good ol’ clichés about how either their club’s strong close to the regular season could be the ticket to post-season glory, or how the team’s stumble to the finish line spells disaster for the second season.
Ordinarily our reactions to trite explanations like these range somewhere between Phil’s rolling his eyes and IJay’s wanting to rip off his own ears so he doesn’t have to listen any more (yes, he recognizes that muting the TV would be a less melodramatic solution). But when it comes to team momentum going into the playoffs, the Department of Hockey Analytics thought there might actually be a kernel of truth behind this particular platitude.
It makes sense that a team’s performance right before the playoffs might be more meaningful than the team’s performance earlier in the year for various reasons. A struggling team might have important injuries that carry over into the playoffs or a surging team might have gotten healthy. Maybe a team finally managed to sort out its defensive systems or line combos. Or maybe players just happened to be riding a natural performance peak – like a runner or swimmer who carefully manages her training so that her performance peaks right at the Olympics.
To test the theory, for every playoff series since 1998 we determined each team’s probability of winning as a function of the regular season point differential between the two teams and their relative “momentum” coming into the playoffs. We measured momentum by the points each team earned in (a) the last 5 games of the regular season, and (b) the last 10 games of the regular season.
The results are shown in the graph. Each dot represents a single team in a single first round series. The location of each dot reflects the probability that the team would win the series as determined by our model.
As expected the team with more regular season points was more likely to win each series. While that’s hardly newsworthy, we were able to quantify just how much it mattered. For each extra point a team had over its playoff opponent, it was, on average, 1% more likely to win the series. For example, all else equal, a team with 10 more points than its opponent stood about a 60% chance of moving to the next round.
There was also a fairly strong relationship between performance over the last ten games of the regular season and the likely outcome of the first round of the playoffs. In fact, a single point of difference over the last ten games meant a 2.2% increase in the likelihood of winning the series -- or more than twice as powerful a predictor of playoff success as a point difference over the entire regular season. So the probability that the team with the ten point regular season edge would win the series shrinks from 60% to 55.6% if its opponent earned just 2 more points than it did over the last ten games.
However, momentum over just the last five games of the season doesn’t seem to matter. Why does the last 10 games matter but not the last 5? That’s hard to say. Possibly because in the last 5 games teams that have locked up a playoff spot start resting their players, or maybe five games is just too small a sample size to be meaningful.
What does all this mean? First, since there seems to be no correlation between a team’s performance over the last 5 games and its playoff performance, it means teams should be less concerned about a little rust at the start of the playoffs and should freely rest their players as the season winds down.
Second, as far this year’s first round match-ups are concerned, in the Eastern Conference it means that the Rangers might be more of a favorite than their 2 point regular season lead over the Flyers indicates, due to their superior record over the last ten games (6-2-2 versus 4-3-3). In the West, it means that despite the Blues’ late season implosion, their series with the Blackhawks still promises to be a close one, since the Blackhawks had the second worst record of all playoff teams over the last ten games.
I liked this article, however I have one quarrel; I find the information interesting but I'm not entirely sure it points towards momentum. Most people would define momentum as the results of several games immediately before said game influencing the outcome of that game. However, in your data, the previous 5 games does not have an influence, instead the previous 10 does. That makes me think that perhaps there is something else going on here, not momentum but a different factor. For example, perhaps the teams that are best are simply more likely to win the previous ten games than a team that is inferior, therefore being more likely to win in the playoffs regardless of momentum. Perhaps teams that are losing more in the previous ten games are doing so because of injuries and will likely lose in the playoffs for the same reason.
I'm skeptical to attribute this data to momentum as opposed to any other possible factor due to the previous 5 games not being representative of future results; in my mind, it would be that sample that would convince me of momentum, instead of last ten games. I understand your point of resting a teams best players, but I still find myself skeptical to accept the conclusion.
Regardless, I thought the article was very interesting and well put together. Good read
Hi - I only just saw this comment today, so I apologize for the late reply.
The general point you are making is well taken. Better teams should also be more likely to have better records over the last 10 games. Thus, if you were simply to look at correlation coefficients between records in the last 10 games and probabilities of winning series, what you would be capturing, to a large degree, is that better teams win playoff series more often.
This is why multivariate regression techniques are so important. By running a regression, we were able to "control" for the underlying qualities of the teams in the series. By having a term that captures their relative strengths, the momentum term really is separate from the underlying quality of the team.
In the regression, we considered the relative records of the teams as well as their relative "momentum". That is, there are teams that were 10 points better, for example, than their opponent who had better records over the last 10 games (which you would expect), but also teams that were 10 points better but had worse records. What we are finding is that, among teams who were 10 points better than their opponents, those that closed out the season (relatively) strong won more often than those who finished (relatively) weak.
I hope this helps, and thanks for your comment. Again, I'm sorry it took me so long to notice it!
Very interesting topic, appreciate it for putting up. bfgfcbefceeg