by Adam Wodon/Managing Editor (@chn-adam-wodon)
NPI is here. How’s it going?
This is the point in time where Pairwise got pretty stable. By that I mean, it was a decent enough indicator of what would happen at the end of the year. About 75 percent of teams in the top 14 at this point of the year, traditionally made the NCAAs.
It’s hard to tell whether this is because the math is so great, or because teams that play well early tend to be good teams that keep playing well. But it was probably some combination.
Whether this holds true for NPI remains to be seen. But at the moment, the top of the NPI and Pairwise (and KRACH, for that matter) don’t look a whole lot different from each other.
You can check this all out yourself. CHN has fully converted the site over to using NPI over Pairwise in all places where it’s relevant. That’s been fun. Check out our recent podcast, where we deep dive into the rationale and history of the NPI, and its weightings.
These are my early thoughts:
* It’s likely that when Dartmouth finally loses a game, some of those ECAC teams at the top will drop moreso than would be normal from just one loss. We’ll soon found out (or maybe not?).
* Overall, the NPI seems better than Pairwise/RPI, at least from a logic standpoint. It gets us closer to KRACH’s methodology, which I still prefer (with some tweaks there). We publish KRACH as our “CHN Power Ratings,” and stand behind that. It’s what powers the strength of schedule adjustments in our player CHIP rating, and it powers the Probability Matrix. Meanwhile, I have two math PhDs working on perfecting that algorithm (as much as it can be).
* Tim Danehy, the progenitor of NPI and long-time Pairwise guru, said on our recent podcast that he agrees KRACH is “pure” math, and that in a perfect world, that could be used. But he said there’s still too many quirks with sample sizes and insular schedules to make it good enough for use in college hockey. Honestly, I’d still prefer KRACH, and have touted it as a replacement for RPI for 20 years. But I get his point. Both algorithms rely upon an iterative process to get the numbers to line up. I believe KRACH with some tweaks (home/road splits, etc…) would be the ideal world.
* There are a couple interesting differences in the tweaks (which we’re apparently now calling “dials,” by the way) to the NPI as opposed to RPI, both of which I’m on board with. For one, in postseason conference tournament games, there is no home/road weightings. Home/road for the regular season is fine. But in conference tournaments, where a team truly did “earn” home ice, it was seen as a punishment. The higher seed gets potentially three more home games, but has an upside of only 0.8 wins for each. We’ve talked about this a bunch on our podcast since Colorado College coach Kris Mayotte brought it up a couple years ago, when his team barely missed the NCAAs after losing a best-of-3 4/5 series at home to Omaha.
* It’s worth remembering that things like the home/road weighting were never meant to be that precise in the first place. A 1.2/0.8 weight implies a 60 percent chance of the home team winning. That’s historically too much. Last season may be an anamoly or the start of a trend, but it was the lowest winning percentage for home teams we have on record, around 53 percent. It’s even worse than that if you just factor in conference games, where schedules are more balanced. Last season was around .500. Which also goes to show the need for home/road weighting — because that higher number in non-conference games implies that, more often, stronger (read: bigger schools) teams host weaker teams. The point of the home/road weighting most of all, thus, is not to get an exact correction for home-ice advantage, but rather to incentivize bigger schools to schedule more road games at smaller programs’ buildings. As a result, don’t expect that weighting to change. But it also points out that making it 1/1 in the conference playoffs is a good idea.
* There was another small tweak to home/road weightings. For 3×3 OT games, only regulation time will be weighting. The 3×3 portion will not be home/road weighted. That’s because 3×3 OT, from the data, is a crapshoot. That means if you lose 3-on-3 OT on the road, it goes down as basically a tie for NPI calculations.
* All of which brings us back to our favorite topic — home-ice Regionals. And I’ve written and spoke about this 62 ways ’til Sunday, so won’t go into it all again. Read here, here and here (and better solutions). But my point stands: These tinkerings only further demonstrate that the math is not good enough to be giving home-ice advantage to teams in the NCAA Tournament. The math is OK, it’s the best we have, and I’m not blaming the math. But these are philosophical decisions driving these tinkerings. Quality Win Bonus, the amounts of the home/road weightings, etc… They are not necessarily based upon sound science. And again, I have no problem with this. It’s not supposed to be 100 percent pure, and can’t be. But if I have to hear one more time … “well, if the math is good enough to pick the teams, why isn’t it good enough to seed the teams” … I swear I will lose it. People say this to me as if it’s a “gotcha” moment, and — sorry, no — I’ve been writing about that very thing for 30 friggin’ years. So again, in short — it’s the best we have for picking the teams, we need something, and it’s better than humans doing it. But it’s not good enough math to give teams the extra advantage of home ice. Which is definitely an extra advantage still in the NCAA Tournament, despite the shrinking regular-season numbers.
* That debate would be much more enjoyable if everyone just said “we get the math isn’t perfect, but we want home-ice Regionals for other reasons” — rather than trying to convince me the math is good enough for that purpose.