Will Roberts

No call at ALL on that hook on JVR?? #LGRW

Always happy to answer questions or have a thoughtful discussion around other improvements you'd like to see. I hope this at least explains why fumbles hurt a player's score so badly, because sooner or later, fumbles will come back to bite them!

Part of what makes the sport of football unique is the shape of the ball and what it does when it hits the ground. Since luck is the largest contributor to whether or not a team recovers, I felt controlling for all fumbles was more consistent and unbiased to outcomes.

I wanted to allow players to at least have a chance that a fumble doesn't derail their entire season in the model. That may still happen if a TE puts the ball on the ground, but it would be even worse if only counting lost fumbles.

If the model was built only on fumbles lost, a player would certainly avoid a penalty if their team recovers, but then would get hit TWICE as hard if they lose it, since fumbles lost occur on roughly 1/2 of all fumbles. And 1/(1-0.5) = 2

It is essentially a 50/50 coin flip as to whether or not the fumble is lost. This is why I've decided to build the model as penalizing ANY fumble, because the ball is literally out of the player's hands at that point. There's no "good" fumble.

I've heard the argument that only lost fumbles should penalize a player, but when looking at the numbers, it is really just luck as to whether a player's team recovers or not, as evidence by the Fumbles Lost to Fumbles ratios for each position

TEs fumbling only 20 times in 934 recorded games in '25 is why TEs are hurt so bad. And why QBs aren't, although I will be adjusting that for next season. QBs had only been based on rushing TDs, and I'm going to base it on total TD instead to increase the penalty.

One thing I don't think most people realize is just how rare fumbles actually are in football, relatively speaking. It's why there's always so much excitement when one happens. Here's a look at the total number of fumbles in CFB by position over the last few years

Thus, if something doesn't happen frequently, like fumbles, it has a higher weight since you are dividing by a smaller number. That's why completions, rushing yards, etc. have smaller weights. There's a lot of them.

The way Production Score is built is that it relates each stat, in this case fumbles, to its relative frequency to touchdowns. This gives roughly how many "points" each stat would be worth if everything a player did counted on the scoreboard.

With football season just about over, one of the biggest sources of feedback I've gotten in this first year is surrounding fumbles. I've heard it is too punishing for a player, especially if their team doesn't lose the ball or points aren't scored off of the turnover.

U.S. WorldTour team EF Pro Cycling to sell naming rights as part of plan to win Tour de France and Tour de France Femmes.

The Athletic speaks to team CEO Jonathan Vaughters about his plans to take on the sport's biggest teams.

🔗 www.nytimes.com/athletic/700...

Andrew Copp’s December to Remember

prashanthiyer.substack.com/p/andrew-cop...

Great write up, Prashanth!

Thanks for the feedback! Definitely something I've struggled with too. Turnovers, and especially scores, result in massive values because they just don't happen that often compared to TDs for offensive players. I'll see if I can find a way to tone that down in the off-season. Thanks for the help!

Glad you like it! I created the model these cards use. How do you think it does in evaluating player performance? Always open to feedback if you have any!

Always happy to take feedback and answer any questions on the model, but hope these changes help make things a little clearer when using it as a tool to evaluate player performance.

The expected result is that PAR values will fall generally speaking, but the hope is that it will give a better impression of how that player is performing compared to a starting caliber player at their position.

Rather than using the true median, this new version of PAR will compare a player's Production Score to the median of all STARTING players at their position.

If a player received significant playing time, regardless of how they performed, they would be near the top of the league by PAR making it hard to tell how they're doing compared to other players.

Production Above Replacement (PAR) had been measured against the true median of ALL players in the dataset. While this helps to keep 50% of players above a PAR of 0 and 50% below, it doesn't give us much insight as to how a player is performing.

After careful consideration and review of the Production Score model, I've decided to go ahead and make one rather significant update.

The Red Wings Centennial logo at center ice is ssssssooooo much cooler than Chicago's #LGRW

Top 5 WRs by PAR so far this season (min. 3 games)

1) Mario Craver - Texas A&M +10.4
2) Cade Harris - Air Force +9.5
3) Chris Brazzell II - Tennessee +9.3
4) Jaden Barnes - Appalachian State +8.9
5) Omar Cooper Jr. - Indiana +8.9

Good stat analysis is always welcome.

Oh. So that was a perfectly rational thought. Don't have tons of those.

And then there's the stat about 0 multi goal comebacks for Dallas, and I suddenly don't feel that bad. #LGRW

Just heard that Colorado has 2 SHG this series and genuinely wondered if the Red Wings had 2 all season. #LGRW

Migration season is under way, and I'm super excited the orioles finally found our feeder! #birds