iiinova — where your digital self keeps evolving

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

With all of that in mind, I would consider SPM an offense-leaning estimate of player impact. If you're interested in the full SPM ratings, check out the article I wrote for SI. It has updated prior-informed RAPM and Win Probability Added rankings for 2026 as well 🙂 www.si.com/wnba/wnba-aw...

si.com

WNBA Awards by the Numbers: Why A’ja Wilson Is the Advanced Stats MVP

The numbers don’t lie.

000

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

(Limitations cont'd) - Even with padding, stats in a 44-game season can fluctuate and aren't always "stable" - The scale of the SPM ratings aren't comparable to NBA metrics. I find that my WNBA RAPM coefficients are more compressed than what I see elsewhere.

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

(Limitations cont'd) - Defensive contributions are underexplored. There are more than 2x as many offensive features as there are for defense. SPM captures defense only to the extent it shows up in those stats, so defenders whose impact doesn't manifest through those stats will be underrated.

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

There are several limitations to the SPM model, as I have built it. - Prior-Informed RAPM is a noisy, heavily regularized target that leaves signal on the table. Any biases in RAPM carry through to SPM. - It was designed to be descriptive of how impactful a player's production was, not predictive.

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

As Dan Rosenbaum noted in 2004 when he introduced SPM, the SPM ratings tend to be "cleaner" than RAPM. Face validation was encouraging, too. At least seven of the top 10 players should be All-WNBA selections. It wasn't perfect, though. See limitations below for more on that.

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

More results: The validation-selected model had a validation R2 of 0.43 and a test R2 of 0.44 SPM also achieved an adjacent season mean Spearman correlation of 0.68, higher than both prior-informed RAPM* (0.58) and ordinary 1YR RAPM (0.34) (*Stability for prior-informed RAPM may be inflated.)

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

A Generalized Additive Model performed best in validation. Linear regression worked well, too, and might be the simpler choice here. I constrained each GAM term to be monotonic to help avoid oddly shaped relationships that didn't seem to have a plausible basketball explanation.

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

The model’s inputs are one season’s worth of 14 individual statistics, all of which are shown below. Each of the features are: - Converted to a per 100 rate - Padded toward the league average - Measured relative to a season's specific league rate (this helps with era shifts)

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

Why? My goal was to build a descriptive single-season SPM model. Prior-informed RAPM is more stable and predictive than ordinary 1YR RAPM. Compared with multi-year RAPM, prior-informed 1YR RAPM accepts more noise in exchange for staying closer to what happened in that specific season.

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

Excuse a brief aside on the target variable: SPM traditionally uses multi-year RAPM, since it's divorced from counting stats and it provides a relatively more stable estimate of lineup-based impact. For this SPM model, though, I used prior-informed, 1-Year RAPM as the target variable.

100

Dan Falkenheim @thefalkon.bsky.social · 29/09/2026

What's the value of a WNBA player's individual production? Enter Statistical Plus-Minus. It estimates impact by weighing different contributions from a player's stat line. SPM is more stable than RAPM, and it often produces more sensible rankings. Ranks + methodology below 👇

A data visualization showing the top 10 leaders in WNBA Statistical Plus-Minus. The order is as follows: 1. A'ja Wilson (+3.22), 2. Veronica Burton (+2.29), 3. Kelsey Mitchell (+2.28), 4. Caitlin Clark (+2.18), 5. Paige Bueckers (+2.08), 6. Aliyah Boston (+2.05), 7. Breanna Stewart (+2.05), 8. Jackie Young (+2.05), 9. Olivia Miles (+1.83), 10. Kayla McBride (+1.77)

The gap between A'ja Wilson and second place (0.93) is almost double the gap between second place and 10th place (0.52)

210

Dan Falkenheim @thefalkon.bsky.social · 22/09/2026

With the playoffs fast approaching, let’s check in on the eighth-seeded Liberty. Where do their players shoot more or less often than the league average? For the first time in Jonquel Jones's career, fewer than half of her shots (47.0%) are coming in the paint.

A data visualization showing where each player on the Liberty shoots more (red zones) or less (blue zones) often compared to league average. Fewer than 50% Jonquel Jones's shots have come at the rim, while players like Pauline Astier and Rebekah Gardner bring rim pressure.

030

Dan Falkenheim @thefalkon.bsky.social · 17/09/2026

Animating a team's defensive shape as a convex hull using SkillCorner's Spanish Liga ACB open tracking data. kudos to @skillcorner.bsky.social for making this public. it's a great opportunity for people to get hands-on experience with tracking data!

020

Dan Falkenheim @thefalkon.bsky.social · 17/09/2026

Where do Phoenix Mercury players shoot more or less often vs. league average? Alyssa Thomas and Natasha Mack stand out for their shot frequency in the paint. Kahleah Copper's shot profile is relatively even compared to the rest of the league Chart excludes players on new teams.

010

Dan Falkenheim @thefalkon.bsky.social · 15/09/2026

Where do Los Angeles Sparks players shoot more or less often vs. league average? The Sparks guards don't rack up a ton of attempts near the rim, but Dearica Hamby and Rae Burrell help make up for that.

A visualization showing where Sparks players take shots more (red zones) or less (blue zones) often compared to league average. Dearica Hamby and Rae Burrell take a high rate of shots near the rim. On the whole, the Sparks are relatively balanced and avoid long twos. Monique Akoa Makani's chart includes her shots from her time with the Mercury.

040

Dan Falkenheim @thefalkon.bsky.social · 03/09/2026

It's fair to ask why I focused on one-year RAPM when multi-year RAPM is more predictive. I wanted to build a descriptive one-year BPM/SPM model. So, I also wanted a more stable one-year target. For forward-looking uses, time decay or multi-year RAPM is likely the better route.

121

Dan Falkenheim @thefalkon.bsky.social · 03/09/2026

Research here was made possible by the previous work on stabilizing RAPM with priors by Fearnhead and Taylor (2010) and Jeremias Engelmann. Engelmann's writeup on incorporating priors back in January formed the foundation for what I did here.

110

Dan Falkenheim @thefalkon.bsky.social · 03/09/2026

tl;dr: Single season prior-informed RAPM... - Only uses previous three-year RAPM as a prior (no box score metrics) - Caps the prior penalty share at 1/3 (weak prior) - Improves predictive performance and maintains sensitivity to current season

110

Dan Falkenheim @thefalkon.bsky.social · 03/09/2026

A second issue: RAPM wants to lean hard on the prior, which pulls the estimates toward three-year RAPM The prior's penalty share can become a hyperparameter that balances predictiveness with responsiveness Setting it to 1/3 was a sweet spot (Spearman = 0.95 w/ one-year RAPM)

A visualization plotting the prior share of total regularization on the x axis and mean test RMSE on the y axis. The plot reaches it's first "elbow" when the prior share is set to 20%, then steadily drops and reaches a local minimum around 75%.

A visualization plotting the prior share of total regularization on the x-axis and mean spearman correlation on the y-axis. As the historical prior share crosses 50%, the resulting estimates show more correlation with three-year RAPM. Before 50%, the resulting estimates show more correlation with one-year RAPM.

110

Dan Falkenheim @thefalkon.bsky.social · 03/09/2026

I wanted to see if there was a way to both improve one-year RAPM's predictiveness *and* retain its responsiveness To do that, I used each player’s previous three-year RAPM as a weak prior for the one-year model. Doing so provides better predictiveness than ordinary two-year RAPM

The same line graph as the post above, with the prior-informed result overlaid on top. Note: Single-season Prior-Informed RAPM is not strictly a "one year horizon" model as it contains prior information from the previous three seasons; however, it does only use a single season in the training window.

110

Dan Falkenheim @thefalkon.bsky.social · 03/09/2026

Taking a step back: RAPM generally improves with at least three seasons in the training window. That makes sense, as the model can work with a larger sample. A drawback of multi-year RAPM, though, is that more stability comes with less sensitivity to current year performance.

A line graph showing how ordinary RAPM performs across different time horizons. The line graph drops sharply (improves) at three years, plateaus, and then begins falling again around eight years. A separate configuration is selected through validation for each time horizon (seasons included in the training window).

120

Dan Falkenheim @thefalkon.bsky.social · 03/09/2026

One-year WNBA RAPM is noisy. Can encoding prior information, without using box score inputs, help? Yes. Prior-Informed RAPM closes about half of the predictive gap with three-year RAPM, while estimates stay rooted in current season play. Ranks and details below 👇

A visualization showing the top 10 and bottom 10 WNBA players by prior-informed RAPM in 2026. Vernoica Burton, A'ja Wilson, Jackie Young, Natasha Howard and Aliyah Boston rank as the five most impactful players by prior-informed RAPM. Prior-Informed RAPM uses a three-year RAPM prior and trains on a one-season horizon.

162

Dan Falkenheim @thefalkon.bsky.social · 02/09/2026

Where do Seattle Storm players shoot more or less often compared to league average? Their guards like to get to the paint and near the rim, and their bigs don't shy away from above the break threes.

030

Dan Falkenheim @thefalkon.bsky.social · 01/09/2026

Where do Connecticut Sun players shoot more or less often vs. league average? No surprise: Their bigs live in the paint, and their guards aren't afraid to shoot the midrange shot.

A visualization that shows each Connecticut Sun player's shot frequency heatmap. Red zones indicate where a player shoots more often compared to league average; blue zones indicate where a player shoots less often than league average. In general, the Sun's frontcourt players take a high rate of shots in the paint.

030

Dan Falkenheim @thefalkon.bsky.social · 23/08/2026

on first glance, the Fire allow transition and second chance opportunities at a relatively high clip (both of those contexts are efficient for the offense), which I think is at least a small part of what's driving their defensive rating down. would like to dig in more, though!

020

Dan Falkenheim @thefalkon.bsky.social · 22/08/2026

Paige Bueckers loves to attack the midrange, particularly on her right side. (The Dallas Wings' empty-side actions might play a part in this, too.) She currently leads the league in right-side midrange field goal attempts, with 25 more attempts than Courtney Williams in that zone.

A relative shot frequency heatmap that shows where Paige Bueckers shoots more or less frequently compared to other guards in the WNBA. In general, Bueckers shoots more frequently in the right mid-range and at the top of the key. She shoots less frequently near the rim and from the left wing beyond the arc.

070

Dan Falkenheim @thefalkon.bsky.social · 21/08/2026

(On the Fire's league-best scramble efficiency since July 1: Part of it is due to a small sample size. They rank 14th in scramble frequency. They also rank second in scramble three-point attempts and fifth in scramble 3PT FG%, which is helping to raise those numbers.)

121

Dan Falkenheim @thefalkon.bsky.social · 21/08/2026

For reference, here are the typical tables showing how each WNBA team has performed across different play contexts for the full season.

111

Dan Falkenheim @thefalkon.bsky.social · 21/08/2026

How has every WNBA team stacked up efficiency-wise across different play contexts since July 1? Quietly, the Portland Fire have had one of the league's better half-court offenses. The New York Liberty's half-court defense has also cratered without Leonie Fiebich and Satou Sabally.

110

Dan Falkenheim @thefalkon.bsky.social · 19/08/2026

Sports Illustrated is hiring (!) for a women's sports (particularly women's hoops)-focused Staff Writer. I'm biased, but this is a really excellent opportunity to drive our coverage in the magazine, on the website and across social platforms. Come work with us 🙂 www.comeet.com/jobs/minutem...

comeet.com

Job opportunity: Staff Writer, Sports Illustrated at Minute Media

Sports Illustrated (SI) is the trusted voice for a changing sports landscape. With 70 years of excellence in sports journalism under its belt, SI is the essential destination for fans looking for expe...

084

Reposted by Dan Falkenheim

Richard Cohen @richardcohen.bsky.social · 15/08/2026

By the way, for anyone interested in the full numbers behind what I mentioned on the podcast yesterday - These are the records of the eight playoff teams against the bottom seven in the league: 17-3 13-2 13-3 14-2 14-3 13-5 14-1 10-4 Top group vs bottom group has been a bloodbath. #WNBA

2308

Dan Falkenheim @thefalkon.bsky.social · 03/08/2026

They are! They've been quietly solid all year.

010

Dan Falkenheim @thefalkon.bsky.social · 03/08/2026

Why track kills? Defensive rating can feel abstract, while kills turn sustained defensive execution into a concrete, attainable goal. Kill rate is closely tied to stop rate and defensive rating. It is a descriptive companion to broader measures of defensive quality.

050

Dan Falkenheim @thefalkon.bsky.social · 03/08/2026

A “kill” occurs when the defense prevents the offense from scoring on three straight possessions. College programs have tracked kills for well over a decade, so let's do the same for the WNBA. Here are the league's leaders in kills per 100 defensive possessions:

A table showing how each team ranks by stop and kill rate. The Valkyries generate more kills than any team in the league.

3112

Dan Falkenheim @thefalkon.bsky.social · 02/08/2026

I've been thinking a lot about your stories and your writing this week. The past few days have been disheartening, to say the least.

010

Dan Falkenheim @thefalkon.bsky.social · 02/08/2026

L.A. short-circuited a potential rebuild by trading for Kelsey Plum ahead of the 2025 season, and now it's back to where it began. For @sportsillustrated.bsky.social, I chronicled how the Sparks reached this point and what the future looks like from here: www.si.com/wnba/with-ke...

si.com

With Kelsey Plum Traded to Mercury, Sparks Have to Start Over Yet Again

Los Angeles short-circuited a potential rebuild by trading for Plum ahead of the 2025 season, and now it's back to where it began.

021

Dan Falkenheim @thefalkon.bsky.social · 01/08/2026

If you're looking for local Minnesota LGBTQ+ organizations to support ahead of tomorrow's Lynx game, consider: RECLAIM: www.reclaim.care TIGERRS: tigerrs.org Queermunity: queermunitymn.com Feel free to add more!

040

Reposted by Dan Falkenheim

Julia Poe @juliapoe.bsky.social · 29/07/2026

DiJonai Carrington says she will be a game day decision for the foreseeable future. Could make her debut for the Sky as early as tomorrow’s game against Connecticut.

0477

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

The assist credit heuristic was inspired by Dean Oliver's writing from the late 80s 😂 For that rabbit hole and more, I wrote up my results and what goes into the numbers for Sports Illustrated: www.si.com/wnba/most-cl...

si.com

The Most Clutch Players in the WNBA This Season

By the numbers, these are the players who leave their mark when the pressure is highest.

020

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

My WPA includes misses, makes, free throws, assists, rebounds, turnovers, steals, blocks, fouls drawn and committed. Credit for assists and rebounds can be ambiguous. My stab at it is detailed below.

An image explaining the process behind assigning rebound and assist WPA. Rebound WPA: The shift in win probability associated only with securing possession, given that the shot has already been missed. Assist WPA: The shift in win probability associated with making the shot is divided between the assister and the shooter. The assister receives WPA credit proportional to the historical make rate for the shot location (34% on most threes, for example). The assister’s credit is capped at 50%. The shooter receives the rest.

120

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

The hand-wavy, imprecise explanation is that a semi-markov model estimates the home team's win probability by averaging a team's odds of winning across future game states. tried to go more in-depth with the simplified sheet below. (forgive any errors/typos!)

A sheet showing some of the underlying math behind a simplified version of the WNBA semi-markov win probability model. It is not intended to be an exact replication and omits certain features of the model for space.

110

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

After testing XGBoost, GAMs, Stern’s Brownian motion model, and other methods, I landed on a semi-markov model. On out of sample games, the no-odds variant (used for WPA) is roughly on par with Inpredictable. The odds-aware variant slightly outperforms ESPN Analytics.

A plot showing how the no-odds calibrated semi-markov model performs in out of sample games vs. Inpredictable. In 2025, the models are about equal in game-equal brier LogLoss. In 2026, Inpredictable holds a slight edge.

110

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

Win Probability Added (WPA) quantifies that impact by looking at how much a player swung their team’s estimated chances of winning. It’s really important, then, to have a good underlying win probability model to ensure those estimates reflect real uncertainty!

110

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

What is clutch, anyway? The W defines it as the final five minutes when the game is within five points. That’s not a bad definition, but it doesn’t distinguish the impact of a layup to pull the game within three with 4:59 left vs. a game-winning shot.

110

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

Who has been the WNBA’s most clutch player this year? To answer that, I built and tested a win probability model, then measured how much each player shifts their team’s odds of winning late in close games. Sabrina Ionescu comes out on top. Here’s what goes into the numbers 👇

A table showing the top 10 WNBA players by Win Probability Added (WPA) per 5 clutch minutes. (Minimum 25 clutch minutes played.) Clutch time uses the WNBA's definition and occurs during the final five minutes of a game when the score is within five points. WPA includes shot attempts, free throws, turnovers, assists, rebounds, steals, blocks, fouls drawn and fouls committed.

231

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

The Cyberiad, and Stanislaw Lem in general, is such a treat. Not only intellectually rich but also presages so much. (And Michael Kandel is a wonderful translator.)

000

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

"Our ancestors created it for the simple reason that anything else would have been too easy for them; in their megalomania they thought to make the very sand beneath their feet intelligent. Quite pointless, for there is absolutely no way to improve on perfection.“ — Stanislaw Lem, The Cyberiad

110

Dan Falkenheim @thefalkon.bsky.social · 28/07/2026

yes! the definition used was "a play that occurs soon after an offensive rebound, before the offense and the defense reset" so, basically putbacks and quick kickout attempts after an offensive rebound. typically happens < 5 seconds of offensive rebound

010