88 Comments

All the advanced stats for the women are broken. Rebounds are valued too highly. Look at any of the advanced stats it goes C, F, F, C, C, F, F, F, C, F then you start seeing some of the guards.

Expand full comment
author

Certainly PER and WS are not adjusted for position, but Estimated RAPTOR is explicitly set up so that the average G, F and C zero out at the league level. So that wouldn't be what's happening with that component of the composite, at least.

Expand full comment

You are also right about rebounds. Scoring on your own miss on a 1st offensive rebound gets downright silly: credit of 2 pts for FG + 2 offensive rebounds.

Expand full comment

Agree! and as I have pointed out to Neil many times, the WS stuff is ridiculous as it is based on theoretical and not actual wins. Indiana's 20 wins are higher than their theoretical by about 4 wins. What this means is the WS of Indiana players are low by about 25%.

Expand full comment

Dude last minute puts the obvious RoTY at #1 so his ranking don't look like a joke when CC is crowned. This guy either has a shitty model or he sold out to ESPN to create false drama.

Expand full comment
author

You can see the progression in the data every day of the season, smh

https://github.com/Neil-Paine-1/WNBA-stats/blob/master/Consensus%20Wins%202024.csv

Expand full comment

Neil, your ESPN article got downright silly when you tried to explain Rickea Jackson in 6th place. She was clearly the 3rd best rookie and some may argue the 2nd best, as she is not a one-trick pony. What you have shown is the 'advanced' metrics, as applied to individual players, is just a theoretical exercise that has no place in mainstream publications.

Expand full comment
Sep 18Liked by Neil Paine

A'ja has set the record for most regular season total points, total rebounds, ppg, rpg, and win shares. And she's a close second for best PER. That's a pretty good case for the greatest regular WNBA season ever. I wonder if eRaptor would agree

Expand full comment
author

Her 9.0 wins added are 4th-most ever in a season (behind 2000 Sheryl Swoopes, 2023 Breanna Stewart, and 2002 Tamika Catchings). Interestingly, when we pro-rate things to 40 team games, though, she drops to 11th:

1. Sheryl Swoopes, 2000: 12.1 W per 40 gm

2. Tamika Catchings, 2002: 11.9

3. Yolanda Griffith, 2000: 10.4

4. Sheryl Swoopes, 1999: 10.3

5. Yolanda Griffith, 2001: 9.9

6. Yolanda Griffith, 1999: 9.8

7. Cynthia Cooper, 1997: 9.7

8. Sheryl Swoopes, 2002: 9.6

9. Breanna Stewart, 2023: 9.6

10. Yolanda Griffith, 2004: 9.5

11. A'ja Wilson, 2024: 9.2

This is likely a residue of her comparatively less impressive on-court impact stats. The Aces were "only" +6.5 per 100 when she was on the court, and actually 1.8 pts/100 worse with her than without her. We don't have those numbers for the older seasons on the list above (there's a separate formula that deals with those seasons that have missing data), but Stewart in 2023 is an interesting comparison: The Liberty were +13.0 per 100 with her on the court, 11.5 better than when she wasn't playing. That helps explain why her season ranks slightly ahead of Wilson's 2024 even though Wilson had better individual numbers.

Expand full comment

9 of the top 10 are from 1997->2004. If these are the 'older' seasons, this comparison does not make sense as you are comparing results of 2 different calculations. It would be better to compare results using the 'separate formula' for all players being compared.

Expand full comment
author

The version that includes plus/minus only comes in over the past few seasons, when that data is first available, so that's not what's happening here. It may well be that it was easier to dominate relative to the league in earlier eras of the WNBA, or that Swoopes, Cooper, Griffith and Catchings were just that good relative to the competition. Everything here is relative to the league in that season.

Expand full comment
Sep 5Β·edited Sep 5Liked by Neil Paine

I really appreciate you putting this together. I at least briefly check probably 4-5 days out of the week.

Apologies if you've already answered this, but is there a way to use or apply a date filter? I think it's obvious why a lot of people are looking at these stats. So adjusting a timeframe to remove tough Liberty losses and/or an early 1-8 record would be....neat.

From what I understand you use basketball-reference.com, and to get a date filter for those stats requires a subscription to stathead. Notwithstanding the ease (or lack thereof) with which said filter may be applied, totally understand if this is a no.

Thanks again!

Expand full comment
author

Sorry I missed this earlier... No filter option on Datawrapper charts right now, sadly, but the raw data itself is available archived every day on GitHub:

https://github.com/Neil-Paine-1/WNBA-stats/blob/master/Consensus%20Wins%202024.csv

Expand full comment

These stats are complete trash. You subtract wins based on point differential πŸ˜‚. That tells me you underweight clutch time which insane.. then you use that to lower a players value on the team. So pretty much other players on the team bring down your ranking lmao. Fucking moron

Expand full comment

Kev, please give us your own equation for a composite stat. I can't wait to pick it apart and make you look like an idiot.

Expand full comment

A good start would be to allocate WS based on actual team wins! Not how many they should have won. At 35 game mark, actual wins, compared to WS wins, are off significantly (28%) for 5 of the 12 teams. At 28 game mark, this problem affected 2 teams. It now affects 5. In the course of 7 games, what was bad, is now much worse.

Expand full comment
Sep 3Liked by Neil Paine

Would you be able to do these ratings with real wins/loses and compare to the pythagream wins/loses to see what changes?

Expand full comment

Would you just divide the final number by the pythagoream win and multiply it by the real win or is it more complicated than that?

Expand full comment
author

It's a little more complicated, but not too bad -- you would take the difference between the team's actual and Pythagorean wins, and distribute that across all the players on the team according to their % of team minutes. So the Fever have won ~4 more games than their Pythag expected, and Caitlin Clark has played 17.7% of their minutes, so she would receive a bump of 0.7 wins. In a vacuum, this would boost her to the periphery of the Top 10 in each list overall, though you'd need to do the same thing for every player to know where she'd land for sure.

Expand full comment
Sep 3Liked by Neil Paine

I guess that would hurt the aces and storm the most. So that percentage isn't in the original calculation so it can't just be the multiplied and divided it seems.

Expand full comment
author

Using the data I post on GitHub (https://github.com/Neil-Paine-1/WNBA-stats), you can derive the % of minutes just by aggregating the player minutes at the team level. I don't have the actual/pythag wins in there, but those can be pulled from this BB-Ref table: https://www.basketball-reference.com/wnba/years/2024.html#all_advanced-team

Expand full comment

It also helps the mercury a bit, with how few games the wnba plays vs the nba and mlb it seems the Pythag isn't as statistically significant as the former sports

Expand full comment
author

I would say the Pythag is MORE significant in a sport with fewer games. One of the whole points of it is to find teams that had good/bad luck in small samples of games by having close wins/losses.

Expand full comment

Also is it a straight increase using minutes played or does it incorporate their value during their minutes played because just a straight add of time played seems a little reductive to the ratings

Expand full comment
Sep 2Liked by Neil Paine

Hey, I'm curious about using turnovers as a stat in ratings - the NBA players with the most turnovers ever is Lebron James. Him and other top scoring players are often the players with the most turnovers every season, because they simply have the ball more - in 2023, Trae Young and Luka Doncic had the most turnovers per game in the NBA.

Caitlin has the WNBA rookie record for assists - wouldn't it make sense she also has the most turnovers?

Are turnovers really a useful metric to factor into how a player is performing?

Expand full comment
author

Yes! Turnovers are consistently one of the strongest negative predictors of a player's effect on team offensive efficiency, even after controlling for assists and other elements of the player's role.

https://web.archive.org/web/20071022232808/https://sonicscentral.com/apbrmetrics/viewtopic.php?t=327

https://www.basketball-reference.com/about/bpm2.html

https://fivethirtyeight.com/features/the-hidden-value-of-the-nba-steal/

https://fivethirtyeight.com/features/weve-been-waiting-for-this-andrew-wiggins/

Not only are they highly negative on offense, but they also predict a negative impact on _defense_ because they often start high-value possessions going the other way:

https://fivethirtyeight.com/features/how-our-raptor-metric-works/#:~:text=Fastbreak%20turnovers%20committed,0.11%2Dpoint%20deduction.

Expand full comment

You can't fairly account for TO/A when as seen with CC's first 15 games her team was basically letting the ball bounce off their face on beautiful transition passes that SHOULD have led to assists. That's not even counting the 2-3 1 foot bunnies that were completed passes then missed. Those turnovers that get lumped on her not only hurt her but take away from her other stat (which is undervalued in the advanced metrics to begin with)

Expand full comment
Sep 19Liked by Neil Paine

OK, it's a bit surprising & I would hate to say "your method/equation is wrong", at first I assumed ESPN is trying to make controversial clickbait to get people talking, because the odds were like -3000 in Vegas for Clark ROTY late August/early September I believe - but if you've used the exat same equation in the past then it's fair

At this point, is Clark in the lead for ROTY by your calculations? it would be interesting to see the updated list after the regular season is over

Expand full comment

Yea but the models fail with generational talent.. example one lebron.. example two Luka.. you also have to factor in actual TO.. a non negligible portion of Clark’s turnovers are caused by her teammates fumbling a pass. You also need to factor in ACTUAL production compared to turnovers.. Clark has the most points scored and assisted on per game in the entire league.. she skews the chart because of her presence. You also don’t factor in usage rate enough. How is Clark the best in the league points+assists combined but her turnovers are still counted as the same as other players who are less impactful.

Expand full comment
Sep 1Liked by Neil Paine

Hey Neil. I love your ESPN articles. It's great to see someone use legit advanced statistics rather than just spouting off biased opinions. Don't let the haters get to you. They just don't understand.

Expand full comment

Cool to see Seattle’s Nneka Ogwumike at No. 4. The team gets a lot of production from its star Jewell Loyd and Skylar Diggins-Smith seems to be involved in every play, so it is surprising to see Ogwumike as one of the best in the WNBA.

Expand full comment
author

Nneka Ogwumike is awesome, she's long been a darling of the metrics (https://fivethirtyeight.com/features/the-rise-and-rise-of-nneka-ogwumike/) in addition to a former MVP and great all-around player. She's 2nd on the Storm in points per 100 possessions (https://www.basketball-reference.com/wnba/teams/SEA/2024.html#per_poss::26) despite not needing the ball a huge amount outside the regular flow of the offense.

Expand full comment

Neil, PER is a made up calculation that in many cases does not pass the eyeball test. Most extreme example: Bill Russell, considered one of the best players of all time and undoubtedly the best winner, is not in the top 50 players of all time per PER. Also, guard play is undervalued in PER, as evidenced by the fact none of the top 15 in history are guards. Is James Harden better than Magic? Is Kobe only the 40th best player of all-time. I don't think so.

Expand full comment
author
Aug 30Β·edited Aug 30Author

I don't disagree! I include PER as one of the composite stats because it is one of the most commonly used metrics, and it actually correlates much better with perception-type things like All-Star voting than other advanced stats. So I still think it has value as a way of quantifying perceived value (and for that reason, I'm actually really surprised Clark isn't higher in it). But there are far better metrics to measure actual value, especially when we want to include defense.

Expand full comment

How can she be, when it is clearly wrong! Other teams (except Phoenix) player totals are close to actual. A stat with anomalies of 30% are kind of meaningless. There has to be something unaccounted for or accounted for wrong. Also, vs Boston and Mitchell, her PER of 2 is silly, notwithstanding turnovers.

Expand full comment
author

My rules of thumb are that PER rewards volume scorers and doesn't care about defense or efficiency. Win Shares rewards efficiency and at least tries to incorporate defense, but undervalues usage rate (scoring volume) a lot. And then stats like BPM (not available for WNBA) and RAPTOR do a better job of balancing between those factors while also more properly accounting for defense and passing value.

Expand full comment

There is a problem with WS: Indiana has 15 wins, but the total for the individual players is 10.8! There are 4.2 wins unaccounted for. I do not know what other stats this affects, but to the extent it does, it is a problem.

Expand full comment
author

That's because Indiana has 11 Pythagorean Wins, or the wins we would predict based on their point differential -- a better predictor of true team quality (in single-season samples) than W-L record itself: https://www.basketball-reference.com/wnba/years/2024.html#advanced-team::5

Expand full comment

Pythagorean Wins!!! You might want to tell your ESPN readers your rankings are based on Pythagorean Wins and not actual wins. That of course, would go over like a lead balloon.

Also, raptor has the same limitation. Indiana players are missing 3 wins after last night. So, while the nuances between these various measures may be interesting to stat guys like yourself, when they end-up telling the same (wrong) story, what's the point.

Expand full comment
author

How this all is calculated is completely out in the open, and I even provide the data for people to download every day: https://github.com/Neil-Paine-1/WNBA-stats

It still amazes me that people think I set out to tear down Caitlin Clark, that I am being secretive, or have fudged numbers somehow... We decided to rank off of a composite of advanced metrics from before the season even began, and I've updated those stats every day on Substack ever since. These metrics have been in use forever in the NBA and they work, to the point that every team uses a RAPTOR-like stat (although there is tracking data in the NBA, so they work better there). If you want to read opinions based on the eye test or scouting, I'm not your guy. My job is to present the stats and explain them.

Expand full comment

I am not suggesting anything except basketball 'advanced' stats for individual players generate so many nonsensical results as to be meaningless. If being used forever means being published forever, then yes, but really used like in baseball, no. As for explaining them, you write like you are comparing real info such as true shooting, rebounds, assists,....as opposed to theoretically, this team is 30% worse than their record indicates, therefore the players are 30% worse than they appear, as reflected in their individual 'advanced' stats.

Expand full comment

And the sky is literally falling and so is your creditablity

Expand full comment

Aren’t you also the one who predicted that Iowa wouldn’t make the Final 4 this year based on your metrics?

Expand full comment
author

Yes, they faced a difficult path in the bracket and had just a 32.9% Final Four probability in ESPN's BPI model before the tournament. I wrote about how she and Iowa overcame every challenge they faced here: https://neilpaine.substack.com/p/caitlin-clark-and-iowa-met-every?utm_source=publication-search

Expand full comment

So Neil you're the egghead (whink) who writes that Angel Reese is better than Catlin Clark for ROY using data analytics. So where does winning factor in? Last 10 games, Angel is 3-7, Catlin is 7-3. Looks like the algorithm is missing something.

Expand full comment
author

Each other component metric (RAPTOR wins, Estimated Wins Added from PER) is scaled to equal the team's total Win Shares -- this ensures that no single metric gets more weight than the others, but it also has the side effect of making sure each one is tied to the team's overall performance. So that's how winning factors into the composite ratings.

Expand full comment

Is there any testing on how ERaptor and WARP correlate to win% or RAPM?

Expand full comment
author

For the NBA, Estimated RAPTOR has an 0.89 correlation with full-blown RAPTOR in the tracking era and 0.93 in the pre-tracking era. WNBA-wise, I haven't really tested against anything because I just decided to port it over mid-season.

Expand full comment

Hello! I love your substack! Is there any way to see the elo from the begining of the season? I would like to see the evolution of the elos along the 2024 and to do somekinf of visualitations

Expand full comment

Placing the sky over the fever is wild because fever leads them 2-1. Fever has also played more games. Whatever β€œalgorithm” this is doesn’t make any sense.

Expand full comment
author

The Sky have a better PPG differential (-1.5 vs -4.2) even after adjusting for strength of schedule: https://www.basketball-reference.com/wnba/years/2024.html

Expand full comment