clock menu more-arrow no yes

Filed under:

2019-2020 NCAA Women’s Hockey Projected Records & Pairwise

Projecting the final Pairwise Rankings using KRACH-generated records

John Quackenbos, BC Athletics

KRACH-Projected Records & Pairwise Rankings:
NCAA “National Collegiate” Women’s Hockey

Welcome to the 2019-2020 NCAA Women’s Hockey Records & Pairwise Projector! Here’s a Q&A to help you understand what’s going on here. Fair warning — there’s some math involved.

What is this?

This calculator uses KRACH to project each team’s full-season record (through the end of the regular season). It then uses those records to exactly calculate what the Pairwise Rankings would be based on those results.

This allows you to compare teams based on how strong their remaining schedule is — a team may have several losses, but may have a soft schedule the rest of the way. Since they would be favorites in the rest of their games, they wouldn’t be projected to have many more losses.

But my favorite team is #1 in KRACH. Shouldn’t that mean they are projected to beat everyone the rest of the way?

No! The beauty of KRACH is that it gives you percentage odds of any team winning any game. We add those season-long fractional results into the team’s actual wins and losses to give them a final record. For example, if Team A is 20-10-5, and they have 70% odds to win their last game, then Team A is projected to have a 20.7—10.3—5 final record. That is, they are given credit for 0.7 wins and 0.3 losses from that unplayed game.

This is sort of like what ESPN does to give college football teams Projected Records, except this uses KRACH instead of FPI.

How about ties?

You only get credit for a tie if a completed game ends in a tie. Games that are being projected are only projected as a fraction of a win and a fraction of a loss.

Even if we could pull out some fractional number of ties from that projection, it wouldn’t make any difference, since a tie is just defined at 0.5 wins and 0.5 losses anyway.

How do you deal with “bad wins” in the RPI, since every unplayed game has fractional wins and losses?

A “bad win” is one that hurts the team’s RPI even though they won the game — so a win by a really good team against a really bad team. By definition, that applies only to games that a team wins.

What we did was we took each game and determined whether a “full win” against that opponent would lower the team’s RPI. If it did, then we took out the value of the full win times the number of wins they were given credit for in that game — so, if the “Game RPI” for a “bad win” was 0.600, and the team was projected to win it 95% of the time, we took out 0.570 “points” and 0.95 wins from the RPI calculation.

That’s... pretty complicated.

It is! Sorry. Just go with it; it’s right.

What about mid-season tournaments? You don’t know who is going to play in, say, the Beanpot finals before the semifinals happen.

Ah, but we can give odds of each potential finals matchup. We take the odds of Team A and Team C winning their semifinals, plus the odds of Team A and Team C losing their semifinals, and that gives us the odds of those teams facing each other in the finals. So, Team A might be projected to play Team C 70% of the time and Team D 30% of the time in the finals — we give Team A credit for playing 0.7 games against Team C and 0.3 games against Team D, which gives you 1.0 full game. That’s why if you look at some teams’ Head To Head comparisons, you’ll see a fractional number of Head To Head games. Beanpot teams that don’t play each other in the semifinals are a good example of this.

How about the “Quality Win Bonus”?

Teams are given credit for fractional QWBs, too. That includes — stay with me here — fractional fractional QWBs for teams that don’t know who they’re playing in the finals of a mid-season tournament.

Wow, you really thought of everything.

Hell yeah I did.

Here are a few notes and tips:

  • Like the other calculators here at BC Interruption, you have the option to change game results and see what would happen. Just change scores in the “Results” tab.
  • The “Grid” tab will show you the familiar Pairwise Comparison grid. However to make things a little easier, on the “PWR” tab, we’ve also included a matchup calculator that will allow you to easily see the comparisons between two selected teams. Just select two teams, and the comparisons will populate.
  • Score margin in games does not affect the rankings — only win/lose/tie.
  • To remove results, you need to double click on the cells with scores in them and delete everything in the cell for both team’s scores. The sheet reads non-blanks as a score: for example, two blanks is “no game,” but two [space]’s will be read as a 0-0 tie.
  • To add a game, you would add a line in the “Results” tab like the rest of the lines. Just make sure you spell each team name exactly the same as the rest.
  • Actual game results do not update automatically. I will be periodically updating throughout the week. The spreadsheet will then update with the new results if you refresh the page. You can see which dates’ games have been included at the top of the calculator.
  • Once you change results in the “Results” tab and switch over to the “PWR” tab, give it a few seconds. The results will update, but probably not instantaneously. It’s a pretty big spreadsheet with a lot of iterative calculations.

If you have any questions, comments, or bugs to report, I can be reached at grant dot salzano at gmail dot com.