In this tumultuous Week 15, we saw the winless Colts beat the Titans, the seemingly unstoppable Packers lose to a floundering Chiefs team, and his holiness, Tim Tebow, lose for the first time in 7 games. Kansas City jumped out to a 6-0 lead with two early field goals, but just as importantly, kept the ball out of Aaron Rodgers' hands by sustaining long drives. The Chiefs' first three drives totaled 38 plays consuming over 19 minutes of game clock. Newly anointed interim head coach Romeo Crennel was also faced with a few early 4th-down decisions. Using our Markov model, we can see how these drives developed.
The Chiefs received the opening kickoff and proceeded to drive 79 yards on 14 plays, not letting the Packers offense take the field until six minutes had already run off the clock. Here is a look at how the probabilities developed throughout the drive:
Kyle Orton and company converted on two third downs (plays 3 and 9) thanks to short completions and runs after the catch, but after two incomplete pass attempts on 2nd-and-goal and 3rd-and-goal from the 1-yardline, the Chiefs decided to take the field goal.
As Advanced NFL Stats followers like yourselves will know, it almost always makes more sense to go for it on the goal line. Going for it nets almost twice as many expected points as a field goal (4.52 to 2.40) given that the estimated conversion rate is about 68%. Even if the Chiefs don't convert, they still net +0.53 expected points by pinning the Packers so deep. Now this should be qualified: this takes a league-average offense as the baseline and the Packers certainly are not a league-average offense (although they looked it on Sunday).
After a Packers missed field goal and another KC 9-play, 4+ minute drive that ended in a field goal, the Chiefs offense took the field again, this time at their own 15. What followed was an eerily similar 15-play, 9-minute drive:
This time, after Jackie Battle and Thomas Jones were stuffed on back-to-back plays, the Chiefs decided to go for it on 4th-and-1 from the 3-yardline. Similarly, going for it was the correct decision. Going for it yields 4.10 expected points while taking the field goal results in 2.38. Since it was slightly later in the game, we can take Win Probability into account as well. Going for it gives the Chiefs an 81% chance of winning the ball game versus a 79% chance if they take the field goal. Again, this should be qualified for the same reason - the Packers are not a league-average team, so those probabilities are probably lower. Given that the Chiefs were significant underdogs, however, it makes even more sense to be taking risks.
Jackie Battle was once again stuffed, but backed up against their own end zone, the Packers were forced into a quick three-and-out. This would have been a huge win for the Chiefs had Tim Masthay not unleashed a 71-yard punt to even the field position battle. No one else would score before halftime and the Chiefs would go on to strip the Packers of their undefeated season.
Given that they were severe underdogs, I don't fully understand Romeo Crennel's decision to take the field goal on the first drive. In fact, I would argue that it would have made sense for the Chiefs to go for it on 4th-and-6 from the 14 on their second drive, instead of taking the second field goal. The Field Goal attempt nets on average 0.05 expected points over going for it in that situation - essentially negligible - but the reward for a conversion is much higher. After all,
underdogs need aggressive and risky strategies to win. Props to Crennel for taking the chance on the third drive, though.
Keith Goldner is the creator of Drive-By Football, and Chief Analyst at numberFire.com - The leading fantasy sports analytics platform. Follow him on twitter @drivebyfootball or check out numberFire on Facebook.
We're the chiefs really underdogs in this game? They completely dominated the line of scrimmage. Without any other info, just watching this game showed they were somewhat close in talent, if not the chiefs as abetter team.
Anon-
The Packers are down 2 starting linemen and and their best receiver in the last week.
Is there a decent way to quantify the impact a #1 receiver like Greg Jennings has on the productivity of the other receiving options? Even though he doesn't always put up big numbers himself he probably draws the best coverage options, increasing the efficiency of other players. On Sunday it certainly looked like something (his absence ?) made it harder for Rodgers to find open options.
That has to be a troll post if I've ever seen one.
Anon #2 : yes, you're right. And that's why it is conceivable that the Chiefs thought they had a good shot of winning that game, and thus didn't need to increase variance,
We want the Lions 98 yard game winning TD drive!
Per request, here is the Lions game-winning drive. Remember that the model isn't as well fitting here given that time/score are not variables. So the Lions were never going to punt the ball or kick a field goal. That being said, the 48-yard completion to Megatron was by far the biggest play in the drive:
http://i.imgur.com/6PyRa.png
According to the model, only a 9% chance of a TD starting from their 2 - but again this needs to be qualified since they were never going to punt or kick a field goal.
And as to the first comment, you cannot possibly consider KC as anything but an underdog. GB was favored by 14 despite the injuries and KC was floundering. If you truly believed that before the game then hats off and you could be making millions in vegas.
KG, when would Crennel know in real world terms when he moved from being an underdog to a favorite? If BB says it's 83/17 for Green Bay before the game...but then starts his game probability real-time chart at 50/50...and shows KC at 86/14 to win with 5:44 left in the second quarter when the Chiefs are down at the 3...are the Chiefs still a dog, or are they now the favorite.
Should Crennel be thinking, "I'm 86% to win right now with my 6-0 lead, I need to be conservative." Not likely, regardless of what the chart says.
Should he be thinking "We were almost two touchdown underdogs before the game started, and I'm only up 6-0...I need to be aggressive and risky." They're not 2-TD dogs any more.
Or, should he be thinking, "Why does that Burke guy start the game at 50/50. That doesn't help me at all right here."
Vegas has in-game betting that adjusts on the fly. Is there any way for BB's charts to better approximate the true percentages when one team is at least 60/40 or better to win before kickoff (which describes 13 of the first 15 games this week by the way)?
Would keep you from having to say stuff like "Now this should be qualified," and "Again, this should be qualified for the same reason - the Packers are not a league-average team, so those probabilities are probably lower."
Would help the discussion I think to have a better sense of what the true numbers are when a pre-game underdog has taken an early lead...
Jeff that's an interesting point. It would be interesting to adjust the in-game probabilities based on our pre-existing evaluation of the teams. I would argue that the Packers should still have been favored for almost the entirety of the game (up until the last Battle TD, probably). I'm sure those in-game Vegas lines reflected that too.
I bet Gb -148 to win the game at halftime. That's not that far away from 50/50.
Thanks anon, that's very helpful. Would guess no-juice line would be suggesting about a 57-58% win probability for GB at the half according to the market, at the time BB's chart was showing 70% for KC.
Agree with you KG about KC being the "true" underdog most of the game. Gets confusing with the different percentages floating about though. Tougher to visualize what coaches should do based on that moment's challenges.
Has there been a discussion in the past about why BB uses 50/50 at the start of every game, when so few NFL games have probabilities exactly at true pick-em? It's probably something that's come up before.
Quick note...when you hear old school oddsmakers complaining about the new corporate types and their "algorithms," it's often tied to these "in-game" betting options. Old school guys want to post a number based on their feel for the game and how the bettors are likely to invest...while the new age math guys prefer their in-game algorithms. Maybe the anonymous poster just above can add some context to what he thinks of in-game lines in his experience...
If algorithms are being used to create in-game lines, it is most likely because of the time-sensitivity of the information since it is nearly impossible for a human to accurately update lines after every play without some computational aide.
As for pre-game lines, I use an algorithm based on point differential (similar to the pro-football-reference srs ratings) to create my own lines, and they are usually very close to Vegas lines except for games involving teams with major injuries, coaching changes or something else that cannot be easily measured objectively.
It is a good way to see how much certain injuries are affecting the lines and take advantage of over-adjustments. For example, the Matt Schaub injury was initially valued at around 8 points by Vegas linesmakers, which was a clear over-adjustment in my mind, and sure enough the Texans easily covered the spread in Yates' first few games before clinching the division.
So, even if algorithms are being used for in-game betting, surely there must also be a human involved to account for events that would be extremely difficult to measure with an algorithm, such as injuries and weather changes. I'd love to hear someone else's thoughts on the subject as well.
Can you do an analysis of how many points Kubiak cost Houston with his unwillingness to use all four downs?
Kicking field goals from the 5 yard line down 21 points -- in the second half? At home against an awful defensive team? With the top seed on the line?
And it wasn’t just the downright stupid choices on 4th down, but on 2nd and 3rd down also. Kubiak’s unwillingness to go for anything, meant passes on 2nd and 3rd even when in clear 4-down territory, and it meant relying on the arm and quick decisions of a third string rookie QB rather than making use of an elite running unit.
Moreover, Carolina knew it. Knowing they could ignore the run, the 31st ranked Panther defense was able to stuff his team, and essentially end any chance of a long playoff run
Sfree -
Down 7-0, 4th and 3 from CAR 41. Punt - Poor decision (+0.98 net EP going for it).
Down 7-0, 4th and 2 from CAR 31. FG - Poor decision (+0.70 net EP going for it).
Down 21-0, 4th and 5 from CAR 7. FG - Poor decision (0 net EP difference but since they were down 21, +0.03 net WP difference going for it.)
Down 21-3, 4th and 8 from CAR 16. FG - Okay decision, not great (EP says FG is more efficient option, but since they were down 18, going for it nets +0.02 WP).
"We're the chiefs really underdogs in this game? They completely dominated the line of scrimmage. Without any other info, just watching this game showed they were somewhat close in talent, if not the chiefs as abetter team."
Is your argument really that the chiefs weren't the underdog because they won the game? Are you being serious or is this a troll post.
No. The argument is that as the game went on, the Chiefs coaching staff may have realized they weren't big underdogs anymore based on the play on the field and score. Jeff Fogle makes my argument more sensicial.
There was a 3rd chance to go for it on 4th and short with 8 minutes left in the game and chiefs up 9-7 and they kicked a 20 yard FG....that one blows the other 2 decisions away. Not only did they announcers agree with "taking the points" as they always do, they also lauded the Chiefs for making it a 2-score game. Seriously.
Thanks Keith for the Kubiak summary. I guess it's not so bad by NFL stupidity standards, but that only partially covers it.
These NFL coaches -- and announcers -- are really maddening.
Has anyone at Advancednflstats ever done a season summary of how many points particular coaching staffs cost a team with these "decisions"?
Maybe you ought to incorporate mgmt foolishness as a variable in the WP estimates. There is variablity among teams, no? NE, NO and Atl go for it at least some times.
Yeah that would be interesting. The typical management foolishness is factored into the WP since it's based on actual results - since most coaches act the same way, the win probability estimates will reflect that. That being said, there is definitely variability among coaches.
First of all, Brian's said before he starts the game charts at .5 GWP because then we can accurately capture how well each team has performed. If he started the game at 86/14 and the Packers immediately jumped out to a 21 points lead then the Packers could only earn 0.14 GWP, which doesn't accurately reflect the game.
Furthermore, adjusting the in-game probabilities by team strength causes other problems. I think it's safe to say that the Saints have a better chance of coming back from behind than the Ravens due to the teams' respective offense and defense. However, the Ravens defense makes it less likely opposing team's score so coming from behind is less likely.
Furthermore, a good offensive team could have an effective running game (49ers), or a passing game (Packers). That makes a difference too. That requires a much greater understanding of teams in context-dependent situations, and is not nearly as simple as starting the game at 86-14. What's the GWP for a team with a 7-0 lead with 6:00 left in the 1st quarter, when they were 85% favorites compared to 45% underdogs?
Thanks James. Was there ever an article on this topic that somebody could link to?
Agree that using the pre-game estimates would influence the measurements of accomplishment. Would suggest calling them "accomplishment" graphs or something similar then. I don't think what Brian is showing matches the definition of "win probability graph" if a team that enters with an 86-14 win probability starts off at 50/50.
My understanding is that he's showing what the probability for a generic Team A would be against a generic Team B given that moment's circumstance (with those generic teams starting out at dead even in terms of abilities given it's 50/50 before kickoff)...rather than what those particular teams are actually dealing with. Have I written a fair description? If you click on the Green Bay/Kansas City box in real time with the "Chiefs" threatening in the second quarter, the probability you see isn't actually for the "Chiefs" winning that game (even though it's labeled that way), it's for a generic team winning in that situation against an equal opponent.
In-game betting markets count in quality of opponent, and also reflect the issues you described James in terms of comeback potential for certain styles. If the line isn't right the moment it goes up, bettors with knowledge of those things bet in a way which moves the line.
In your comment, you said, "That requires a much greater understanding of teams in context-dependent situations, and is not nearly as simple as starting the game at 86-14."
Isn't starting the game at 50/50 much worse in terms of context-dependent situations (simpler and worse)for expressing what the actual percentages are for those teams in that game?
And, in terms of serving the reader, Why don't the "probabilities" as described each week in the NY Times match the "probabilities" at kickoff by the same author in his graphs? Is there a way to clarify things for readers?
Has anyone at Advancednflstats ever done a season summary of how many points particular coaching staffs cost a team with these "decisions"?
The ZEUS guys did this a few years ago with their Critical Call Index. See http://pigskinrevolution.com/CCI.html