Poll-wise: Can polls predict college hockey’s national tournament field better than the PairWise?

Parker Mackay (39 - Minnesota Duluth) (2019 Omar Phillips)
Minnesota Duluth was ranked in the PairWise and USCHO.com Division I Men’s Poll last season on its way to a second straight national title (photo: Omar Phillips).

Some time during every college hockey season – usually around January – many fans switch their focus from national polls to the PairWise Rankings, hockey’s ranking algorithm.

The logic for the switch is pretty straightforward: the PairWise determines the field for the national tournament, and the polls do not.

Why bother with polls at all?

The reason is hidden in plain sight. Strictly speaking, the PairWise does not determine the entire national tournament field: teams that win their conference tournaments automatically qualify, regardless of their rankings in the PWR.

Hockey fans know this, of course, but they rarely discuss it during the regular season, probably because it is so difficult to predict. However, this detail makes a difference: polls are slightly better than the PairWise at identifying tournament-crashing, “Cinderella” teams.

Most of the time, the teams that win their conference tournaments are also highly ranked in the PairWise, meaning that the 16-team tournament field strongly resembles the top 16 teams in the year-end PWR.

However, over the 10 seasons from 2008-09 to 2017-18, 15 schools with relatively low PairWise rankings earned berths in the Division I men’s tournament by winning their conference tournaments. That equates to 1.5 teams per year, or almost 10 percent of the field.

Typically, the PairWise had all but left those teams for dead. Not one of them was in the PWR’s top 25 during the first week of March. In fact, their average ranking was in the bottom half of teams nationwide.

Polls identified at least some of these teams. Sure, sometimes tournament-crashers blindside everyone. Robert Morris, for example, finished the year in the middle of its league standings and was ranked 49th of 59 teams in the PWR just weeks before going to the tournament in 2014.

One-third of the aforementioned Cinderella teams, though, received votes in the USCHO.com poll in the first week of March. One team was in the poll outright (Air Force, No. 19, 2012), two were the top team in the Others Receiving Votes category (RIT and Denver, 2010 and 2014), and four of the five were in the top 25. The average poll position for that third of Cinderellas was 22.5, just a couple of positions outside of the formal top 20.

Of course, the PWR’s inability to predict automatic-bid teams may be unsurprising – it wasn’t built to do that. The algorithm weighs every game’s result equally, regardless if it happened in October or March, meaning that it is blind to trends.

Polls, on the other hand, are more flexible and can identify teams that are playing well right now, for better or for worse – that flexibility can be a liability when selecting an entire tournament field, which is a big reason why the PairWise was created.

With this in mind, we’ll ask a complementary question: If polls are better at predicting the hot teams that win tournaments and get automatic bids, is the PairWise better at predicting the whole tournament field, which mostly includes at-large bids based on regular-season success?

Before we answer that question, let’s address the obvious. Even casual fans know that the PWR and the polls converge as the season progresses. In the first week of March, for example, 15 of the top 16 teams in the PairWise are also in the top 16 spots in the USCHO.com poll. This level of convergence is likely because pollsters mimic the PWR, even if not consciously.

However, it is also true that human pollsters and PWR computers may simply build similar rankings independently of one another, especially as more games are played. As evidence, consider those same sixteen teams atop the PWR in March: about ten of them are typically in the top 16 of the preseason poll, released over five months earlier. About three-quarters of them (about 12) are in the preseason poll’s top 20. The convergence of the polls and the PWR that we see during the season, then, is probably a combination of mimicry and simply being right.

Regardless, we’ll return to the implications of this convergence for the casual fan; for now, we’ll ask whether the PWR predicts the tournament field better than polls do.

It doesn’t.

Over the decade in question, the top sixteen spots in the PWR during the first week of March included an average of 13.4 of the 16-team tournament field. As mentioned, 1.5 teams per year earn strictly automatic bids by winning their tournaments, leaving about one team (1.1) that falls from the PWR during the conference-tournament season each year.

The polls show similar results. The top 16 spots in the USCHO.com poll include an average of 13.2 tournament teams, compared to the PWR’s 13.4 – two statistically indistinguishable numbers. The polls and PWR also behave similarly when we expand our selections to 20 teams. During the first week in March, the top 20 teams in the poll include an average of 14.2 tournament teams, compared to 14.1 in the PWR’s top 20 – another trivial difference, statistically.

The middle of the season tells a similar story. In the first week of January, the top 16 spots in the poll and the PWR include 11.2 and 10.8 tournament teams, respectively. The top 20 spots of the poll and PWR included 12.7 and 12.5 tournament-bound teams. Again, these differences are not statistically significant, but that is the point – the PairWise and the USCHO.com poll predict the national tournament field with indistinguishable accuracy, both in January and in March.

Of course, this is not an argument against the PairWise.

For example, as the number of teams playing in the conference tournaments dwindles, and as the number of plausible tournament scenarios becomes manageable, the PWR is a uniquely useful tool for predicting the tournament field. Also, this is not an argument about changing the selection process for the national tournament.

That’s part of the mystique of college hockey – just ask fans of Minnesota Duluth, whose 2018 championship run was catalyzed by a scintilla of a PairWise point. (Or, for that matter, ask fans of Minnesota, which was that same scintilla of a point away from the tournament.)

However, fans’ shift from the polls to the PairWise each January seems unfounded. If fans want to accurately predict the tournament field, even late in the regular season, the PWR is no better than the polls. In fact, polls often resemble the PairWise, and might even be a little better at picking the at-large bids.

This January, instead of mass-migrating to the PairWise, some fans might consider staying Poll-Wise, too.

P. A. Jensen (@PrideOnIceCream) has written for USCHO.com before, including a previous feature on preseason polls.