Week 14 Discussion

[steelman]

^^...that's not how percentages or probabilities work...

If it worked like that, teams would be doing it.

Going for two is higher variance at the average 2 point/PAT success rates. Variance is irrelevant in the long run. EV is points on the board. Higher EV = more points. More points = good.

NFL teams don’t do it because they are extraordinarily risk averse. They will willingly sacrifice points for the benefit of less public scrutiny. We know this because this is what happens. See: tonight.

Like I said, Mike Tomlin played this game a couple years ago. What the Steelers found out is that it didn't work out and Tomlin abandoned it. On top of the league average statistics, each team has their own statistics, and it would be stupid to ignore those stats in favour of league wide stats, which don't matter one bit as far as coaches are concerned

[PR0v3]

And if you don't get it you are putting yourself in a greater hole.

NFL PAT success rate = 94.4%. 100 PATs = 94.4 points.

NFL 2-point conversion rate is generally in the 45-60% range. 100 2 point conversion attempts = 90-120 points.

Going for two scores more points in the long run, it’s really not up for debate. It’s better to go for two when down 8.

That's the logic gamblers use to get themselves in trouble.

Yes, over the long term it balances out, but in the short term, it's far less successful to go for 2 than it is to kick the PAT. The Steelers experimented with this a few years ago. The only way the odds work out is if teams go for it every time, no exceptions. Otherwise you're trying to time the odds rather than just playing the long term statistics. In the end, it cost the Steelers too many times and Tomlin abandoned the plan.

Long term odds and immediate odds don't always align. Flip a coin 10 times. If the first 9 times it lands on heads, what are the odds it lands on heads again? It's 50/50 obviously, but long term, what are the odds heads comes up 10 times in a row? It would be far, far less than 50/50. That's why if you want to take advantage of the long term odds, you need to go all in. Go for 2 every time or you're not playing the odds properly.

You are more likely to fail to score any points on a single attempt going for two vs. PAT, but you are also more likely to score two points going for two than kicking a PAT. If you convert two point conversions at a rate greater than half that of a PAT, which most teams do, the only time it is correct to kick a PAT is in specific situations.

[PR0v3]

^^...that's not how percentages or probabilities work...

If it worked like that, teams would be doing it.

Going for two is higher variance at the average 2 point/PAT success rates. Variance is irrelevant in the long run. EV is points on the board. Higher EV = more points. More points = good.

NFL teams don’t do it because they are extraordinarily risk averse. They will willingly sacrifice points for the benefit of less public scrutiny. We know this because this is what happens. See: tonight.

Like I said, Mike Tomlin played this game a couple years ago. What the Steelers found out is that it didn't work out and Tomlin abandoned it. On top of the league average statistics, each team has their own statistics, and it would be stupid to ignore those stats in favour of league wide stats, which don't matter one bit as far as coaches are concerned

It didn’t work because they didn’t convert at a rate > half that of a PAT. They probably could have though if they stuck with it given that historical figures show that the NFL does on average, and the Steelers offense was above average.

He was unwilling to see it through though for the long run. The NFL isn’t a long run business though, if things don’t work out immediately you will be axed. It’s no wonder the status quo isn’t often challenged, despite pretty clear evidence it should be.

[honcho55]

Poking my head back in to this thread to comment on the PAT vs 2pt conversion.

PROv3’s posts are all spot on. That is all.

[Servo]

- Eagles inside the 5, Boston Scott TD since Sanders "cramped up". FML

- Eagles inside the 5, Dallas Goedert just doesn't even block and Sanders doesn't score on his lone attempt. FML

- OT, Sanders breaks a tackle, goes outside....and instead of running left around his blocker clean into the endzone...he runs LITERALLY into the dude's back...and I lose by 3 points.

Jesus F***ing Christ, talk about a real kick in the dick. Miles Sanders = blind.

[Nanananananana]

- Eagles inside the 5, Boston Scott TD since Sanders "cramped up". FML

- Eagles inside the 5, Dallas Goedert just doesn't even block and Sanders doesn't score on his lone attempt. FML

- OT, Sanders breaks a tackle, goes outside....and instead of running left around his blocker clean into the endzone...he runs LITERALLY into the dude's back...and I lose by 3 points.

Jesus F***ing Christ, talk about a real kick in the dick. Miles Sanders = blind.

Boston Scott>Saquon

[Phaded]

He was unwilling to see it through though for the long run. The NFL isn’t a long run business though, if things don’t work out immediately you will be axed. It’s no wonder the status quo isn’t often challenged, despite pretty clear evidence it should be.

You have several flaws to your argument and are ignoring many key football fundamentals.

Your success rates for 2pt conversions are based on league wide averages off of an incredibly small sample size (more than half the teams in the league had 0.1 or 0.0 attempts per game in 2018). Each team will have varying success within them. Most teams have a set play or two they go to in these clutch situations - with film study, teams will pick up on what these plays are and the success rate would likely go down.

You keep talking about "long run" success like it is relevant, it's not. The season end conversion rate is meaningless if you were not able to succeed when it matters most. You have to look at it during each individual game which is short-term. Teams only get a handful of touchdowns each game and going for the two-point conversion and failing could literally cost you the game. Each conversion attempt is an isolated incident in itself.

A lot of NFL games are tightly contested. Picture a low-scoring game that going into the 4th quarter is 13-13. Team A jumps out to a lead with a field goal that made it 16-13, now if Team B had opted to go for the 2 point conversion early in that game and failed, it would have been 13-12, now 16-12 and as opposed to a field goal to tie it - they would need a touchdown to take the lead.

You don't have some revolutionary approach - if going for a two point conversion every attempt was the right move, someone would be doing it.

The Saints went for two point conversions twice on Sunday - failed both times, and ended up losing by two as a result of those decisions.

[jenkins.math]

Poking my head back in to this thread to comment on the PAT vs 2pt conversion.

PROv3’s posts are all spot on. That is all.

Yes Prov3's analysis is correct. But as I've learned both here and in daily life, most people are mathematically clueless and do not understand anything beyond basic arithmetic. People will argue the math but only because they don't understand or grasp it, so since they don't understand it, it must be incorrect. It's much easier to stick with the status quo than to data mine, think logically, and base a decision based on said logic.

[Phaded]

It's not so much that the math is incorrect as much as it is the data to come up with the math.
When your data is wrong, your math is going to be wrong.

[PR0v3]

It's not so much that the math is incorrect as much as it is the data to come up with the math.
When your data is wrong, your math is going to be wrong.

The data is based on 2-point conversion rates over the history of the NFL. By this point, the sample size should be large enough to be statistically significant.

[PR0v3]

He was unwilling to see it through though for the long run. The NFL isn’t a long run business though, if things don’t work out immediately you will be axed. It’s no wonder the status quo isn’t often challenged, despite pretty clear evidence it should be.

You have several flaws to your argument and are ignoring many key football fundamentals.

Your success rates for 2pt conversions are based on league wide averages off of an incredibly small sample size (more than half the teams in the league had 0.1 or 0.0 attempts per game in 2018). Each team will have varying success within them. Most teams have a set play or two they go to in these clutch situations - with film study, teams will pick up on what these plays are and the success rate would likely go down.

You keep talking about "long run" success like it is relevant, it's not. The season end conversion rate is meaningless if you were not able to succeed when it matters most. You have to look at it during each individual game which is short-term. Teams only get a handful of touchdowns each game and going for the two-point conversion and failing could literally cost you the game. Each conversion attempt is an isolated incident in itself.

A lot of NFL games are tightly contested. Picture a low-scoring game that going into the 4th quarter is 13-13. Team A jumps out to a lead with a field goal that made it 16-13, now if Team B had opted to go for the 2 point conversion early in that game and failed, it would have been 13-12, now 16-12 and as opposed to a field goal to tie it - they would need a touchdown to take the lead.

You don't have some revolutionary approach - if going for a two point conversion every attempt was the right move, someone would be doing it.

The Saints went for two point conversions twice on Sunday - failed both times, and ended up losing by two as a result of those decisions.

Based on the data, going for two is better in the short and long run. In the short term, variance is slightly higher, but variance isn't a good enough reason to take a suboptimal approach unless in specific circumstances (many late game scenarios, like when you only need one point to win the game). I keep talking about the long run because variance is irrelevant over a large enough sample size.

Last night, when the Eagles were down 8, they gave up any chance of winning the game in regulation by kicking the PAT (unless they went for two after the second TD, which would be suboptimal strategy, for reasons I can explain), and kicking the PAT only saved their chances of losing in regulation by about 15%. When kicking two PATs after two TDs when down 14, the expected point total is 13.9. It's a losing play, in the long run this strategy will lose games. They beat the odds last night, however.

The obvious counter to your example is if Team B had opted to go for two and converted, now they would need a FG to win the game outright and not even have to mess with overtime. A TD would also force the other team to then have to score a TD, since they'd then be up by 4 (if they convert the PAT in a late game scenario). Your example is just another demo of the hyperconservative mindset that NFL coaches have.

[Phaded]

It's not so much that the math is incorrect as much as it is the data to come up with the math.
When your data is wrong, your math is going to be wrong.

The data is based on 2-point conversion rates over the history of the NFL. By this point, the sample size should be large enough to be statistically significant.

Which would be great if the average 2-point conversion rate of all NFL teams over the history of the NFL was truly representative of the probability of success for any given team against any other given team at any point in time. However, there are far more other variables to consider to have a true mathematical understanding of the probability and thus why I feel that the argument and sample size being utilized is deeply flawed.

I understand and appreciate what you are trying to do - but it's not as simple as you have illustrated. If it was, NFL teams would be doing it more often.

[lukkynumber13]

Oh golly, me dumber because this thread... AGAIN

[PR0v3]

It's not so much that the math is incorrect as much as it is the data to come up with the math.
When your data is wrong, your math is going to be wrong.

The data is based on 2-point conversion rates over the history of the NFL. By this point, the sample size should be large enough to be statistically significant.

Which would be great if the average 2-point conversion rate of all NFL teams over the history of the NFL was truly representative of the probability of success for any given team against any other given team at any point in time. However, there are far more other variables to consider to have a true mathematical understanding of the probability and thus why I feel that the argument and sample size being utilized is deeply flawed.

I understand and appreciate what you are trying to do - but it's not as simple as you have illustrated. If it was, NFL teams would be doing it more often.

True. Obviously we don't really know the exact probability of a successful 2-point conversion attempt by the Eagles in the 4th quarter on 12/9/19 against the Giants versus a PAT. It's more complex than a coin flip. In general though, the book says, "down 8 towards the end of the game, go for two." The commentators, which is how this all started, were not wrong.

Don't really agree with the last point though. It took the NBA what, 30+ years to recognize the value of the 3-point shot? Most NFL coaches aren't exactly known for their optimal strategic decision making. The game has hardly evolved at all on that front. For instance, there is also growing evidence that teams go for it on 4th down way too infrequently, and that you should only really punt when you are in an extremely poor situation.

[Phaded]

Don't really agree with the last point though. It took the NBA what, 30+ years to recognize the value of the 3-point shot? Most NFL coaches aren't exactly known for their optimal strategic decision making. The game has hardly evolved at all on that front. For instance, there is also growing evidence that teams go for it on 4th down way too infrequently, and that you should only really punt when you are in an extremely poor situation.

I don't really agree with the comparison to the value of a 3-point shot in the NBA. You have many more opportunities on a game-to-game basis to make up for a missed 3-point shot. In the NFL, you have what - anywhere from 1 to 5 chances usually to make up for a missed two-point conversion? Plus in the NBA, there is even the possibility you get the ball back immediately if the three pointer is missed.

It's interesting that you mention the 4th down thing - I think I remember hearing that Harbaugh hired some data analyst or something like that to assist in the decisions regarding 4th down conversion attempts and other probability factors.