NFL Coaches Are Getting 4th Down Completely Wrong

Plenty of people that rely on only numbers are getting this wrong, also. Is the Defense you are facing great on short-yardage? Have you failed or succeeded on 4th and 1/4th and 3 already? How good or great is your own Defense if you fail? More than just numbers involved in it.

If the defense you are facing is great in short yardage, that would be included in the numbers. Previous successes or failures should not materially change your approach as each instance is, for all intents and purposes, a mutually exclusive event. I don’t think failing or succeeding is indicative of the next try failing or succeeding, but I could test it (again, with numbers). How good is your own defense? Again, quantifiable. Many fail to recognize that math is simply our best objective model of the world and underpins everything, whether we like it or not.

"Previous successes or failures should not materially change your approach as each instance is," This is patently not true.

Imagine how bad a coach would be if he refused to try just because they failed to get it on an earlier drive. Yes, context matters (which is quantifiable), but emotion should not play a part in decision making.

Imagine how bad a coach would be if his team failed on, say, six consecutive 4th and 2's against a Defense. Then went for it a seventh time on 4th and 2, and gave the Harbaugh face when they got stopped, again. It's not emotion. It's situational awareness.

@EaglesXsandOs Situational awareness and the ability to see and feel the momentum/mentality on the field and understand that psychology affects what happens on every snap doesn't show up in a spreadsheet. I don't know why its difficult for the math people to understand that.

@EaglesXsandOs The reason the math/data guys and football guys bash heads so often as I see it, is because data has become such a big part and "the truth", that some act like the game is a simulation and card counting.

I’m a football guy who happens to appreciate how math can help. Go check out my hundreds of posts breaking down plays on film. Context certainly matters, but “football guys” tend to overstate the importance of momentum as well, which appears to be a very real and significant phenomenon, but really doesn’t hold up under scrutiny. We want our human assessment of the psychology of the game to matter and be correct, but that fails just as often, if not more, than going by the numbers.

Every decision can be framed as: Win probability if I get it (40%) WP if I fail (30%) WP if I punt (33%) In this scenario, you can see that going for it IS worth it, but some people don’t want to risk that 3% that’s guaranteed. As mentioned in the video, game context certainly shifts all these probabilities, but I don’t think coaches should become psychologists and look at every decision through the lens of the emotional/momentum impact. That ends up being more about job security than actually making the optimal decision. This is a fundamental challenge to a lot of peoples beliefs and change is always uncomfortable, but over time, I think we will continue to see a shift towards 4th down aggressiveness as coaches like Ben Johnson pave the way for the next generation of coaches to take emotion out of their decision making.

I'll always laugh that we all knew this intuitively playing Madden against your HS friends back in the day.

Would love to see a team follow this for a season. My first thought was blackjack; play the book move for the best odds… It really makes sense to go for it when you think about it like that. Panthers utilized it a lot last season and it certainly gave them a slight edge.

A coach will need job security to really do this (like Ben Johnson), as the failures are magnified. Punting all the time on 4th and short is also a failure but much less noticeable. Many coaches would rather have death by a thousand cuts than one big failure.

Yeah I doubt any coach would last long doing it especially if they have an unlucky season. Always fun to imagine and discuss though! Love this kind of content.

On the topic of momentum: People say the 4th down chart ignores momentum. I tested whether failing gives opponents an extra boost beyond field position. The answer, not really.

Ben Johnson

This is boss

Dope videos.

I call it The Dan Campbell effect. Hus against the spread record is the best in NFL history and it's cuz of his aggressive fourth down decisions and it feels like his edge has started to slip away the last couple years because more and more are now copying him

Cool!

Amazing video

Staley lost his job as HC of the LA Chargers b/c he stupidly lost games failing on 4th down, gifting other teams turnovers.

and now Ben Johnson has been out-classing the league for years now and is on the forefront of the 4th down evolution

I'm not sure your points for down and yard line are right. What is your point value for 4th and 1 on your 20 vs. other team 1st and 10 on your 20?

Go: −0.81 Punt: −1.70 Opponent 1st and 10 on your 20: -4.52 Going for it at 70%: 0.70 × (+0.75 by converting ) + 0.30 × (−4.52 by failing) = 0.53 − 1.34 = −0.81 Going is about 0.9 points better. Break-even = (Punt − Fail) / (Convert − Fail) = (−1.70 − (−4.52)) ÷ (0.75 − (−4.52)) = 2.82 ÷ 5.27 = 53% So going for it is very clearly worth it under normal circumstances The expected points model in NFLverse is based on 17 real plays from that scenario, so its -1.06 in that scenario is totally unreliable. Thats why I built my own model that is more robust and not reliant on small samples.

Thanks but if nobody goes for it on the 20 how can u validate-.81?

My model uses more data. From 2018-25, in the first 3 quarters, teams converted 3rd & 1 between their own 15-25 around 73% of the time (273 plays). Between their own 11-30, they converted 72% of the time (577 plays). Teams actually went for it on 4th & 1 only 16 times between their own 15 and 25, and they converted 81% of those (13 of 16). Between their own 11 and 30, they went for it 52 times and converted 75% (39 of 52). So no matter how conservative your model is, it’s still well above breakeven.

I think 3rd and 1 is a lot different from 4th and 1 pysholgically

Sure, then use 4th down conversion rates which are even higher (nearly 30% above breakeven)

Few realize Claude can one shot these

How could you get a first down inside your own 10🤔

Silly mistake by me, but technically, the other team could turn it over on your half-yard line, that would make 4th and 1 from your 9 and a half yard line possible 😎

Relevant that not all 4th and 1s are created equally @StatsbyLopez has a great paper that touches on this 4th and .1 and 4th and 1.9 are very different plays (both counted as 4th and 1) The former is an auto QB sneak, while the latter you might even throw a pass

Great piece Eric!

But aren’t scoring rates higher for the opposing team if they get the ball ten yards from the end zone?

It’s all factored in. Every 4th down decision has a breakeven rate. The estimated success % (from historical data) determines if it’s worth it or not

But how is it factored in? In aggregate?

BS The generic conversion rate for 4th % 2 is ~ 57%; 4th & 3 ~ 45% But generic teams aren’t playing. A good D is more likely to get a stop. Matchups also affect % It’s never just that grid That grid also undervalues turnovers, which are outcome determinative. Punts aren’t

Yeah but the historical data doesn’t support the notion that coaches are specifically not going for it against good defenses and vice versa. Also, one of the main underlying points here is that coaches systemically overrate the emotional/momentum impact of failure and that leads them to making decisions that throw away points because punting feels safe.

our model using stochastic game theory has v(1,10,21)=-.16, v opponent having ball on our 20 -4.37 and assuming 40 yard punt value to us of -1.19 so we need p*(-.16) +(1-p)*(-4.37)>-1.19 so we need over 70% chance to go for it

To me, the punt is overrated. A 40-yard punt from our 20 gives them the ball at their own 40, so the punt should be worth about −1.9 to us. Real punts from the 18–22 average about −1.7 (1,466 punts, 2018–25). Using −1.19 makes punting look about half a point better than it is. Converting looks underrated. 1st & 10 at our own 21 is worth about +0.75 in real play-by-play data (1,062 plays), not −0.16. I’d like to know which seasons your values are fit to, and what your model says 1st & 10 at the opponent's 40 is worth. If it’s older data, it would explain it because offenses are much better now.

Do not believe the 50/50 math of going on 4th and 1. That is flawed. Take an extreme case. Best NFL D against my high school O. Still 50/50? Hardly, and the NFL is some version of this in every game. I think the chart would be better at 45/40/35% make, then see.

Not sure what you’re referring to. It’s not 50/50 and I never suggested that. Within the NFL, matchup matters but only a little

Not a you saying thing...just what the models assume. It's just not possible that it right.

The models don’t assume that at all

But they do, ask any AI. Google was my first one and this one is from Grok. So a 1st and goal from the 1 that is a goal line stand and misses 4 times in a row would be a statistical impossibility....but it happens often.

The scenario you described of 4 consecutive plays from 1 yard line and getting exactly 0 yards on each try has happened 23 times total in the last 25 years. Thats less than 1 per season. And it’s not statistically impossible to not convert. Of drives that had 1st and goal at 1 and never lost yards and tried on 4th down if necessary, 99.1% ended in a TD.

well....I trust in your analysis but man....crazy stuff.

As a data guy, this is preposterous. There are so many variables that should dictate the decision. Eg. How many conversions in the empirical data came b/w playoff contenders v bottom dwellers? Timing also matters - if I’m down 14 and 1:30 on the clock, converting on 4th & 1 vs prevent defense is much easier. Is Mike Alstot in my backfield? Is Larry Allen on the line? Is Mark butt fumble Sanchez taking the snap?

The video mentions that there are game context factors that shift the probabilities, including time and score.