I was inspired over the holidays to learn some new R package. This somehow snowballed into forecasting the results for all the NACS finals and creating FiveThirtyEight style tables to summarize them `¯\_(ツ)_/¯`

. A snapshot of the main table is below. Check out the full results here!

You can also view predictions for all players in each state/province.

These predictions are based on simulations using everyone’s matchplay ratings. For the most part, I think the ratings do a pretty good job. There are definitely some quirks — for example, it gives four-time Texas champ, Colin MacAlpine, only a 6% chance of winning of winning Texas. Having lower odds makes some sense, as Colin doesn’t have a bye and his seed gives him a relatively difficult path to finals. But I also think some of his potential opponents are massively overrated (*cough* philgrimaldi *cough*). Of course we’ll see how it plays out!

I’m definitely wary of sharing tables that say someone has a “<1% chance” of winning — please know that this is not necessarily my opinion or your skill, just the results of a simulation with a naive/imperfect model! I very much hope some of these 1%'ers prove the odds wrong.

As always, thanks to Matchplay(@haugstrup) and IFPA(@Shep) for providing APIs that make this kind of thing possible.

Enjoy! And do let me know if you have any questions, feedback, or suggestions for improvement! This was just a fun side project, so there very well may be errors. Let me know and I’ll get them fixed ASAP. Also, I will probably rerun the simulations next week to account for any changes in the matchplay ratings.

Edit: Arkansas, Louisiana, & Virginia have now been added. Nova Scotia and North Dakota do not have an even 16 players, which my code cannot handle, so they are not included.

UPDATE (1/13): There was a major bug in my code, and the previous results were showing the odds for a best-of-1 series instead of a best-of-7. Tables have now been updated. For the most part the relative ordering of players will be similar. However, the %'s are now a lot more opinionated, especially in states where there is only one or two heavy hitters. Apologies to anyone who ran out to Vegas with those previous numbers. Thanks to @Pinball_Yeti for pointing this out!

FINAL UPDATE (1/19): I reran simulations with a fresh pull of ratings from matchplay. Results may have shifted if ratings changed. I also corrected any states that I was informed had changed, which may have also affected outcomes. Lastly, I have added a new feature that compares each player to all others in the field, so you can see the predicted outcome of all possible matchups.