Buffalo Bills: 7-5 – at Denver

[enthusiast]

no, it doesn't. have you even used it before?

you're criticizing me for going too in depth and then saying I'm not accounting for things?

 

[misterchainsaw]

Looking at Marrone's decisions using http://wp.advancedfootballanalytics.com/winprobcalc1.php

Play 1: 4th and 1 from own 46, 11:51 to play, 13-10 lead. Win expectency: 64%
Result of punt: 1st and 10 KC from own 8, win expectency: 66%
Result of league average punt (about 40 net yards): 1st and 10 KC from own 14, win expectancy: 65%
Another 50-50 decision with a slight edge to what they went with.

Play 2: 4th and 1 at BUF 41, 7:03 to play, 17-13 KC. Win probability 22%
Result of punt: 1st and 10 KC from own 3, win expectancy: 25%
Result of league average punt: 1st and 10 KC from own 21. Win probability 23%
Another 50-50 decision with a slight edge to the decision made.

Every single one of his decisions were "correct" by the numbers assuming league average and evenly matched teams. The fact the Bills have a good defense and a poor offense pushes the first two even further towards being "correct", and considering the FG route requires another offensive drive and the Chiefs are a ball control team built on moving the chains, I think he was right on the 3rd decision too.

 

[misterchainsaw]

no, it doesn't. have you even used it before?

you're criticizing me for going too in depth and then saying I'm not accounting for things?

It's taking into account FAR more than you are. Yes sometimes teams do or don't have their full allotment of timeouts, but the most often scenario is that team's do (or have one less than their full allotment) and the aggregate data will reflect that. It also takes into account the 2 minute stoppage.

 

[enthusiast]

It's actually incredibly abstract and discounts real scenarios in favor of aggregate scenarios. If you want to look at what to do on 3rd and X that may be viable but outwardly projecting a full game without regard to team specifics is useless


like come on you have to laugh at their option to input your own win percentage at start/spread, that's a clear effort to look like they can account for context worth a damn

 

[enthusiast]

Consider for a moment putting in the same data for 2007 Patriots vs 2008 Lions

you could swap the teams and literally have it output the same percentage

 

[Taro Tsujimoto]

With 2:32 left in the game, Buffalo Bills head coach Doug Marrone decided to go for it on fourth-and-10 from the 15 with his team trailing by four.

With three timeouts and the two-minute warning remaining, conventional wisdom would be to kick the field goal and hope your defense can force a punt.

In this case, ESPN's win probability model agrees that kicking the field goal was the way to go.

In the past 10 seasons, teams have attempted to convert a fourth down with between 8 and 12 yards to go in the 4th quarter or overtime 559 times and earned a first down on 214 of those plays (38.3 percent). Had the Bills converted the first down (putting them on the 5) and assuming a 10-second run-off, their expected win probability would be 60 percent.

Assuming an incomplete pass and a 5-second run-off (which is what happened), the Bills' expected win probability would drop to 16.7 percent.

So if we use 38.3 percent as the probability of converting the first down, the Bills' expected win probability by going for it was 32.5 percent.

After kicking a field goal, the Bills' expected win probability would be 38.1 percent.

http://espn.go.com/blog/buffalo-bil...g-for-it-on-4th-and-10-was-the-wrong-decision

Well, then.

 

[misterchainsaw]

It's actually incredibly abstract and discounts real scenarios in favor of aggregate scenarios. If you want to look at what to do on 3rd and X that may be viable but outwardly projecting a full game without regard to team specifics is useless


like come on you have to laugh at their option to input your own win percentage at start/spread, that's a clear effort to look like they can account for context worth a damn

You're being absurd, but fine, let's go more in depth. They have the percentage of at least one first down under the situation given as 67%. 67% loss, 33% chance of getting the ball back. Of the non-3 and outs, they'll average around 5 yards or so if they're being conservative (they averaged 2 a play running up the gut before they passed). So punt from their own 25, average 40 yard net put, Bills start around their own 35 with ~1:50 left. Win probability is 39% at this point (and this is bias towards the Bills as they wouldn't have time outs left).

.33 * .39 = .1285%. Not much different.

 

[enthusiast]

http://espn.go.com/blog/buffalo-bil...g-for-it-on-4th-and-10-was-the-wrong-decision

Well, then.

I don't understand why they're arbitrarily making it 4Q/OT attempts, that skews it unnecessarily

 

[misterchainsaw]

http://espn.go.com/blog/buffalo-bil...g-for-it-on-4th-and-10-was-the-wrong-decision

Well, then.

38.1% is WAY too high for win probability kicking the ball away with that little time left.

EDIT: It also doesn't take into account the possibility that the conversion would result in a TD. They literally assume all conversions to put the ball on the 5 yard line. That's horrible analysis.

 

[enthusiast]

You're being absurd, but fine, let's go more in depth. They have the percentage of at least one first down under the situation given as 67%. 67% loss, 33% chance of getting the ball back. Of the non-3 and outs, they'll average around 5 yards or so if they're being conservative (they averaged 2 a play running up the gut before they passed). So punt from their own 25, average 40 yard net put, Bills start around their own 35 with ~1:50 left. Win probability is 39% at this point (and this is bias towards the Bills as they wouldn't have time outs left).

.33 * .39 = .1285%. Not much different.

67% is far too high given today's performances and that's evident in the final performance statistics

 

[enthusiast]

38.1% is WAY too high for win probability kicking the ball away with that little time left.

he said, without any data

 

[misterchainsaw]

67% is far too high given today's performances and that's evident in the final performance statistics

Is it? How many times did KC go 3 and out today out of how many possessions?

 

[enthusiast]

they had 5 3 and outs and 1 turnover on less than three plays for a total of 6 out of 12 drives

this is not counting the end of half


so on a given drive 50%, not accounting for the predictable calls to kill clock at the end of a game

 

[misterchainsaw]

they had 5 3 and outs and 1 turnover on less than three plays for a total of 6 out of 12 drives

this is not counting the end of half


so on a given drive 50%, not accounting for the predictable calls to kill clock at the end of a game

That's quite a small sample size you got there. A true rate of 37% is not out of the realm of possibility.

And I count 5 out of 12 (which includes the fumble). Which is 41%. Not far away.

No team in the entire NFL has a 3 and out percentage worse than 30% for the season. http://www.sportingcharts.com/nfl/stats/three-and-out-percentage/2014/. You can probably bump that 5-10% because turnovers inside the first 3 plays aren't counted, so 37% in a conservative situation seems quite reasonable.

 

[26CornerBlitz]

@MatthewFairburn

Full postgame comments from #Bills CB Leodis McKelvin which are NSFW: http://www.syracuse.com/buffalo-bil...l?hootPostID=44b99b293c7cc63fe87aa449188b4b81

 

[Chad Thundercock]

http://espn.go.com/blog/buffalo-bil...g-for-it-on-4th-and-10-was-the-wrong-decision

Well, then.

The flaw I see there is assuming they exactly get 10 yards on 4th, thus the low 60% they get a TD from the 5.

 

[misterchainsaw]

The flaw I see there is assuming they exactly get 10 yards on 4th, thus the low 60% they get a TD from the 5.

Yeah, I edited my post above when I noticed that. That's just flat out awful analysis. Not surprising from amateur hour I mean ESPN.

 

[enthusiast]

That's quite a small sample size you got there. A true rate of 37% is not out of the realm of possibility.

And I count 5 out of 12 (which includes the fumble). Which is 41%. Not far away.

No team in the entire NFL has a 3 and out percentage worse than 30% for the season. http://www.sportingcharts.com/nfl/stats/three-and-out-percentage/2014/. You can probably bump that 5-10% because turnovers inside the first 3 plays aren't counted, so 37% in a conservative situation seems quite reasonable.

sample size? it's the only sample that's relevant re the likelihood of this team's 3-and-out performance against the chiefs. progression to seasonal mean doesn't mean it's suddenly going to click 5 drives in a row

 

[misterchainsaw]

sample size? it's the only sample that's relevant re the likelihood of this team's 3-and-out performance against the chiefs. progression to seasonal mean doesn't mean it's suddenly going to click 5 drives in a row

What the hell are you talking about, it doesn't have to click five times in a row.

Sample size is relevant because when you're talking about a sample size of 11 heading into the last drive (with 5 3 and outs), the odds that that does not represent the "true" 3 and out % is much higher than with a bigger sample. The fact that NO NFL defense maintains a 3 and out rate above 35% tells me that the "true" 3 and out rate statistically is probably less than what the Bills had maintained through the game.

To take this to the extreme example, I'm not going to claim KC's 3 and out rate following their opening possession was going to be 100% because the Bills stopped them 3 and out on the first possession.

EDIT: Statistically speaking, if we assume that 37% 3 and out rate (again, a HIGHER rate than any NFL defense has forced or any NFL offense has flunked into), they'd have a 38% chance of going 3 and out on 5 or more of their first 11 possessions. Well within reason. There's no statistical evidence that the 3 and out rate for a future possession would reach something absurd like 50%.

DOUBLE EDIT: I like how you eliminated the end of half drive when the Chiefs had already successfully gotten a first down on it, btw.

 

[Myllz]

I wonder how much longer Coughlin will have a job. 333 rushing yards against right now.

 

[missingmika]

McKelvin after the game.....i hope no one spray paints a ***** on my lawn again.

Well didnt say it, but definately thought it.

 

[enthusiast]

What the hell are you talking about, it doesn't have to click five times in a row.

Sample size is relevant because when you're talking about a sample size of 11 heading into the last drive (with 5 3 and outs), the odds that that does not represent the "true" 3 and out % is much higher than with a bigger sample. The fact that NO NFL defense maintains a 3 and out rate above 35% tells me that the "true" 3 and out rate statistically is probably less than what the Bills had maintained through the game.

To take this to the extreme example, I'm not going to claim KC's 3 and out rate following their opening possession was going to be 100% because the Bills stopped them 3 and out on the first possession.

EDIT: Statistically speaking, if we assume that 37% 3 and out rate (again, a HIGHER rate than any NFL defense has forced or any NFL offense has flunked into), they'd have a 38% chance of going 3 and out on 5 or more of their first 11 possessions. Well within reason. There's no statistical evidence that the 3 and out rate for a future possession would reach something absurd like 50%.

DOUBLE EDIT: I like how you eliminated the end of half drive when the Chiefs had already successfully gotten a first down on it, btw.

how do you think 6/13 turns to 6/18 without five positive drives in a row

I'm surprised that the bills only scored 13 points, statistically we should have expected them to put up 21? How can we account for this mystery seeing as that's their true scoring rate

incidentally I removed the drive out of perceived charity to KC, I forgot it was a conversion and not just a clock drain. if you want to factor that in against a prevent defense, be my guest

 

[Paradigm]

Hey everyone, we're famous

https://twitter.com/donchefbunny/status/531619661708677120/photo/1

(From the Simpsons/Futurama crossover episode)

 

[misterchainsaw]

how do you think 6/13 turns to 6/18 without five positive drives in a row

It doesn't have to. What are you talking about? Going 5/11 on 3 and outs is not near enough evidence to say the future rate of 3 and outs is going to be 45% when the best defenses in football don't do 35%. It's not like we're talking about a bad KC offense. They are actually quite competent when it comes to ball control and sustained drives, which is the very skill you need when you only need 10 yards to wrap up the game.

What would happen, if you could run possessions to infinity, is that the 3 and out possession % would approach the true % for future drives.

And again, they only had 5 3 and outs including the fumble. Not sure where you get 6 from.

I'm surprised that the bills only scored 13 points, statistically we should have expected them to put up 21? How can we account for this mystery seeing as that's their true scoring rate

The same concept that is the reason sample sizes are important: variance. .

A team that holds their opponents to 3 and out 33% of the time would be expected to hold opponents to 5/11 3 and outs 4.5 times through a season. That doesn't neccessarily mean that the 12th drive of those games is more likely than 33% to be a 3 and out. Eleven possessions is simply too small of a sample to make that claim. In this case, with KC being a good ball control offense, it seems likely that it wouldn't be an extraordinarily higher % than the best NFL defenses maintain over reasonable sample sizes.

incidentally I removed the drive out of perceived charity to KC, I forgot it was a conversion and not just a clock drain. if you want to factor that in against a prevent defense, be my guest

Prevent? The Bills would have loved to make KC punt from inside the 10 yard line with 40 seconds left. If they forced 3rd then 4th down, I suspect the Bills call timeouts.

 

[NotABadPeriod]

Whatever the math says, I'm not gonna fault Marrone for going for it on 4th down. If you're gonna err do it on the side of aggression instead of the conservative **** he does the other 99% of the time.