Florence : A 2021 NWHL Bubble Prediction Model

Meet Florence.

The fourth iteration of my prediction model is yet again almost identical to her predecessors Clementine and Wesley (both of which have previously been used to predict the outcome of AIHL seasons) but with one major difference. Your girl finally dedicated two weeks of her life to transferring everything over from Excel to R.

While I would love to upgrade Florence even more currently available statistics make this difficult and thus she is based on the same concepts that Clementine was run on until such point that we have a big enough sample size of the new tracking data that I feel comfortable upgrading her. For a full rundown of how Florence (and thus Clementine) works please see my post from a little over a year and a half ago.

The basics, however, are as follows:

Offensive and Defensive ratings are produced based on how much stronger or weaker a team is perceived to be from the average expected goals for or against.
Matchups are charted using a Poisson distribution model to predict the likelihood of four separate game outcomes (W, L, OTW, OTL) and those numbers are stored to be referred back to later.
A Monte Carlo Simulations is then run to predict the outcome of the Group Stage, The Seeded Stage and then the Playoffs with all results being stored and the completion of each ‘season sim’.
We then chuck the entire thing into hyperdrive and run that baby 50,000 times or for long enough that I can cook an entire pavlova and wait for it to cool.

At the completion of this we have a a % chance of ever available outcome from Boston loosing every game and Connecticut winning it the cup in OT, to the slightly more likely (A Boston vs Toronto Final that is sure to tear a continent apart).

I will be trying to run an update to the model every evening with new results coming out in the morning ET time so I recommend following me on Twitter if that’s your jam.

And now, with my sincerest apologies to everyone’s teams, here is Florences 2021 NWHL Bubble Predictions.

Screen Shot 2021-01-23 at 7.23.04 am.png

Even just based on the above overview we can already see a couple of things clearly, one being that to the surprise of no one the Boston Pride look to be separating themselves from the top of the pack yet again. Closest to them are the Rivs who have jumped into a fairly secure looking second seed after the release of the final rosters, followed by the Whitecaps and Buffalo. Last-minute adjustments caused Toronto to lose some ground from my initial predictions largely due to the fact that I still had Shiann Darkangelo on their roster by accident despite her trade to the Riveters. Event Connecticut at the bottom of the ladder is still in with a shot with a 24 % chance of making the semifinals out of the group stage.

To look a little deeper at each teams chances we can inspect the point spreads across those 50,000 simulations to see just how far their luck can stretch. The points spread can give up a little more information than a simple average can, instead, showing up where the vast majority of simulations where clustered.

For the Boston Pride we can see that a long tail on that point spread helps drag the average down, with the largest number of occurrences clustered around that 11-12 point range.

Similarly, when looking at the median of each team’s points range we can see how close things really are between those middle four teams with only slight difference in their curves.

Despite all of the above I also feel inclined to note that with the short season, a hot streak can mean anything, and as shown in the Rank Occurrences above there’s a simulation for everything, up to an including Boston exiting the group round in 6th place (sure there’s a 0.01% chance of it happening, but it could still happen).

And with that, I leave you with the final Group seed rankings chances, now go forth with all the statistical knowledge in the world and remember that at the end of the day the most correct piece of analysis is still the following:

‘Just win baby’ - Al Davis (1963)