I've been playing around with the Advanced NFL Stats Win Probability calculator investigating the "go for two first" idea when down 15 in the fourth quarter.

All probabilities are with 6:00 left in the fourth quarter, 1st and 10. I chose this because that was the time left in the Missouri-Oklahoma game which inspired me to look at this. The data is for NFL games though.

If you kick the extra point, let's assume you make the extra point 100% to cut the deficit to 8 and kick deep, and give the opposition the ball at their own 30. The probabilities are given from the point of view of the trailing team.

A) Win Probability when down 8 and opponent has ball on own 30: 8%

B) Win Probability when down 7 and opponent has ball on own 30: 10%

C) Win Probability when down 9 and opponent has ball on on your 45: 5%

D) Win Probability when down 9 and you have the ball on your own 45: 13%

A models the game state after kicking the extra point.

B models the game state after a successful two point conversion and kicking deep.

C models the game state after missing the two point conversion and onside kicking and your opponent recovers.

D models the game state after missing the two point conversion and onside kicking and you recover.

**********

If you give two point conversions a 50% chance of being successful and onside kicks a 20% chance of being successful, I get a win probability of 8.3% overall if you go for two, which is only 0.3% better than kicking...close to breakeven and could be made so with slight tweaking of variables.

Some other notes/problems with the analysis:

The data are from all NFL games since 2000 (all after introduction of two point conversion).

This is NFL data and might not fully translate to college...maybe the two minute warning and different kickoff field position, etc., have an effect.

There is no option to select how many timeouts left either side has, the data is just across all situations that have happened.

There are some weird things in the data where holding everything else constant and giving a team WORSE field position INCREASES their win probability. Maybe this is a sample size issue, but that is what has happened in reality so far. Maybe teams with worse field position have changed their playcalling in such a way as to increase their win probability in certain situations.

I didn't take into account any probabilities of big returns.

Possible reasons this analysis doesn't show a bigger effect:
The leading team derives some offsetting advantage on defense knowing they are two scores up.

The effect is too small to measure with an analysis this crude.

Teams that are behind make even bigger mistakes in the remainder of the game that dwarf the benefit of going for two down 15, but with perfect decision making they would reap the full benefits.

**********

Anyway, I'm fully open to the above being torn to shreds, it was just something I was looking at.

I made assumptions that were pretty favorable to the trailing team (50% two point conversion rate, 20% onside kick recovery rate) and showed a tiny benefit going for two. Change the inputs and I'm sure kicking can come out slightly ahead. In any case, it's close to breakeven if you believe the above.

By way of comparison, Lovie Smith's failure to challenge the Jay Cutler touchdown that was ruled a fumble in Week 7 of the NFL cost the Bears 19% of Win Probability, a massive mistake.

I completely understand the theory of why you want to go for two first when down 15 late, and truth be told I would probably do it myself (assuming my team was at least decent at converting). But there seems to be a gap between theory and empirical data...could be many reasons for this, but it's just food for thought.

Chiefs just punted from the Bills 33. Touchback. Lol.

Chilly challenged a catch that was so obvious that only an idiot would challenge.

yeah, see post 76 of the late game thread and the next 30 posts, every variety of lol chilly used

chilly deserves his own thread

Tied 10-10 in overtime Bills at the KC 40 w/ less than two minutes left, instead of running three plays, running down the clock and setting up a fg to win and and if u miss tie. Buffalo runs pass plays and on the third down fitzpatrick gets called for intentional grounding taking them out of fg range, stopping the clock and leaving a minute plus for KC to drive down the field for the winning score.

chilly against belichick might be the most one-sided coaching matchup that there could possibly be

This isn't surprising IMO. Using 50% made it expected-points neutral. That early in the game, win probability is pretty close to linear in points. That 0.3% reflects the slight curvature.

Something to keep in mind is that for a lot of this kind of stuff, including some of the most glaring examples ITT, the difference in probability actually isn't that great. I wrote the OP after a team scored a touchdown with 30 seconds left and kicked an extra point to go up 6. It was obviously stupid, but kicking instead of going for 2 probably dropped them from 99% to 98.99% to win the game. The standard thing people, including me, like to bring up is being down 14 and scoring a TD. Going for it is better even if you aren't quite a flip to convert, and it's easy to show that. However, it doesn't affect your win probability much if you kick because the scenario where it makes a difference isn't very likely.

A lot of this discussion is like which lines to take in poker. Betting out in a certain spot when a check raise would clearly be better will have a minuscule impact on your overall win rate, but those leaks add up. One of the things I want to do is try to figure out how much the worst coaches cost their teams. My guess is something like a quarter to half a game per season, but I could see it being significantly smaller or larger than that.


Tough to say. Kickers are certainly better in the NFL but everything is different.

I said in a different thread that I don't like his model much. He basically lumps together situations and looks at how often the team won in that spot. The problem is that a huge sample size would be required for this. How often is a team down 10 away with the ball between their own 30 and 50 with between 5 and 6 minutes left in the second quarter? I don't know exactly how he lumps them together, but I think it could be improved a lot to reduce sample-size problems.

Are you trying to figure out the gap between the best and the worst, or are you just trying to figure out how much Coach X screws up? It's probably impossible to figure out since you can't just add up the WP of all of their "mistakes," but I'd be shocked if, for example, Chilly isn't consistently costing the Vikings at least one win per year. That's 0.0625 WP per game, which isn't much at all if he botches a handful of decisions every Sunday (which he does).

um rex ryans fake punt from their own 20 i think it was like 4th and 20 and the punter tried running

Yeah, I'm not sure how to combine them.

It's the same for players or the offense/defense. If you take a team like LSU that is very good defensively and weak on offense, and just add up the changes to win probability for their two units in a single game then their offense might cost them 200% of a game and their D gets them 280% (numbers courtesy of my ass).

rex gross mayne tho

They have that stat in baseball, it's called Win Probability Added, and you can get a value more than one in a single game. In fact, the all time record for a single game is held by someone who lost, he came in as a defensive substitute and hit three go ahead or game tying homeruns, yet they lost.

http://www.insidethebook.com/ee/inde...s_art_shamsky/

I don't mind the Bills being aggressive here. They were moving the ball through the air and really struggling on the ground. They needed more yards for a better FG attempt. Too often I've seen a team "settle" for a game tying or winning 50+ yarder and it not work out. What really killed them was the 20 yards punt.

LOL Mike Shanahan.

I agree with Jared's big post. I just wanted to point out that when you're down 15 late in the game, neither kicking nor going for two can be shown to be a big mistake by the empirical data. There might be something to the "defense gets an advantage when up two scores" theory, or the benefit of going for two could be real but miniscule.

Definitely agree there are some problems with the data, but it's the best I'm aware of and certainly covers broad strokes. There's a discussion on one of the Advanced NFL Stats pages about why integrating timeouts is problematic.

I think that the the O is better off down 7 or 9. A lot of DCs will play prevent or close to it because oh we're up two scores.

Mike Tomlin sucks and fails to challenge an obvious TD and then proceeds to kick a 19 yard FG 2 plays later. COY Mike Tomlin.

omg jaredL isnt gone forever

Also, lol at some of the stories in this thread. Im not a huge american sports fan, but its pretty amazing how stat based sports can be still screwed at such a high level like the nfl...

This can't be emphasized enough. It really is getting borderline mind-boggling that Chilly is the head coach of one of the 32 best football teams on the planet. I mean the ****ing ball didn't come within 2 feet of the ground. WTF was he challenging? It's amazing how many bad (in-game, at least) head coaches there are in the NFL. I'm beginning to think there are about 3-4 coaches in the entire league who make net-positive in-game decisions. Everyone else just hopes to **** up less than the other guy.

Chan Gailey hasn't won a game in a long time. He's rusty. lol.

I just realized that I misread your post and thought it was all with 6 to go in the first, not fourth. Disregard the part about win probability being linear, that's not true with that little bit of time left. The obvious example is going from 8 to 9 up. That's huge with 6 left but not with 51 minutes left.

This is a good example of it being unlikely to matter. For it to be helpful to go for it on the first one, OU would need to score another TD, get an onside kick and then score. That's pretty unlikely and it's actually worse than that because sometimes they would wait to go for it, score with a bit of time left, get an onside kick and score. It's not surprising that the win probability would be close in the two situations. I'm guessing it's actually less than the 0.3% that you got.

The thing about going for two down 9 isn't that it seriously increases your chances. It's that almost everyone that is paid to tell us about the game, and probably a majority of those paid millions to make these decisions, think that it's awful when it is at the very worst just as good as kicking the PAT.

I think we can agree that the problem is more with the lack of logic used to explain the choice. After going through this exercise, I wouldn't have a problem with a coach saying "I've crunched the numbers, it's a close decision, and I decided to kick/go for two based on situational factors X, Y, and Z."

X, Y, and Z could be any number of things specific to your team, the other team, and the situation, such as:

Your best available kicker has a much lower than average rate of converting extra points.

Your team is really good/bad at two point conversions (maybe you have a Boise State style trick play that you feel has a high probability of working one time).

There is a major injury on the other team.

The other team's offense is really bad and you think giving them good field position with a failed onside kick is much less of a problem than with an average offensive team.

You think there's something exploitable on their onside kickoff returns.

However, we're probably always going to hear something like:
"You always wait to go for two until you absolutely have to."
"Why?"
"You just...do."

Shocked he doesn't.

This is a good post, and makes me laugh thinking about a coach in a postgame interview explaining to his fans that even though his decision failed, it was worth a quarter of a win more in equity. He'd be a 2+2 hero, but on the hot seat.