Analysis by Comet

Since I’m not using this site hardly at all anymore, to keep up with our work, check out our new Bluesky page “@shownspace.com” and/or subscribe to the Shown Space Substack.

The short of it is that we presented research on player contribution at the MIT Sloan Sports Analytics Conference in March and we received 2nd place! So we decided to take the manuscript and build a website so that the metrics developed can be more accessible.

For those who are interested in keeping up with more UFA focused frisbee analytics, my friend and I started a new website (shownspace.com) that is still largely a work in progress. Short post on my Substack (analysisbycomet.substack.com/p/its-been-a-w…) with more details.

@watchUFAtv ...reliably contribute to success/failure on the field, it's possible that this bias will correct itself.

@watchUFAtv ...bias is something that we'd want perpetually predicting future outcomes. Of course currently, the model I'm using in simulations is a Bayesian logit model only taking into account home field advantage and team strength. So, as I'm able to include more attributes that...

One fun part of building models and simulations is seeing where they fall short. Not because it's fun to be wrong but because finding concrete biases can be enlightening. In evaluating the @watchUFAtv 2024 season simulation I built, it's interesting that... #30DayChartChallenge

Day 21 of #30DayChartChallenge is "green energy". Expansion teams can be considered "green", so here they're outlined. Since the pandemic, @watchUFAtv expansion teams are either terrible or really good with no in between.

@logclarke @watchUFAtv Yes, x axis is season from beginning to end

@Comet_Miller @watchUFAtv Is the x axis season week?

Back to a few final plots showing the simulated season, mostly to finish out the #30DayChartChallenge uncertainty section. The white lines (which end up looking like a funnel) represent a team's path through the simulated 2024 @watchUFAtv season.

@watchUFAtv For example, though unlikely, the simulation shows it's possible Montreal gets back to being about an average (1500) team. After week 1, it certainly seems as though they'll outperform expectation. Simulation didn't know about the Tobe Decraene impact 😅

For day 28 of #30DayChartChallenge we have two trend plots showing the week 1 @watchUFAtv Super Series game. Two slightly different ways to view how the game progressed. The fourth quarter pull away really put the game out of reach.

The previous ridges plot are individual opponent Elos, which is why there is much more variability. In the true average Elo rating plot, the distributions are much tighter showing avg opponent Elo for each simulated season. Conclusions still the same, just two different views

In a late night post, I realized I made a mistake in the title of the plot. The previous plot wasn't actually *average* opponent Elo rating, but *sorted* by the average opponent Elo rating. This plot shows the actual average opponent difficulty over the 1000 sims.

@larrypmac_tor @itsdanielcohen @watchUFAtv Aka the "regression to the mean" phenomenon tends to, on average, miss the extreme results that often happen in a single actually played-out season. We got a random, unpredictable world we live in (but *on average* pretty predictable)😅

@larrypmac_tor @itsdanielcohen @watchUFAtv Ha, they did! 0.9 wins per season (on average) over the sims. Lower bound of the 95% interval is 0 wins, upper bound is 3 wins. I will say, as is customary with lots of models (especially top-down ones based on team rather than player), the extremes are tough to get accurate.

For perhaps a more analytical approach, (but not to take away from the writing and analysis of @itsdanielcohen) here is another way to view strength of schedule. I simulated the upcoming @watchUFAtv season 1,000 times... More details below in 🧵 #30DayChartChallenge

@samth @watchUFAtv @itsdanielcohen I regret that the title of the plot makes it more confusing than it should be 😣 the distribution is made up of opponent Elos in all 1000 season sims. So the plot is showing 1000 times 12 regular season games. Title should say *sorted* by avg elo. Costly late night mistake 🚩

@Comet_Miller @watchUFAtv @itsdanielcohen This is all the distributions of the average ELO ratings of each opponent, not the distribution of the average ELO over all opponents, right?

@itsdanielcohen @watchUFAtv There's a lot more to say that can't be explained in a thread, but the last comment I'll make for now is that this is player agnostic, so no, this does not take into account the players NY lost or the player your favorite team added. It's solely based on estimated team strength

@itsdanielcohen @watchUFAtv On the right extreme, the Hustle, Royal, Breeze, Phoenix, Shred, Glory, Rush, and Wind Chill all get the opportunity play the star-laden (but weakened?) dynasty this year. Between those two extremes, it's tough to parse exactly, but trends are noticeable if you know the schedule