Friday, January 7, 2011

Coaches of Bad Teams Should Take More Risks to Increase Win Rate

This season there’s quite a bit of turnover in the coaching ranks, which is nothing new for the NFL. And as usual, many of the positions left vacated are from teams lacking talent and performing quite poorly. Four of the biggest coaching vacancies – Carolina, Denver, Cleveland and San Fransisco – come from teams with a combined 17 wins this year and one total playoff appearance in the past five years. Another – Oakland – is coming off an 8 win season and seems to have some talent, but is one of the worst managed teams in the NFL and a perennial bottom-feeder.

Some of these positions will be filled by coaches with thin or questionable resumes, and are likely to start their new jobs a hair away from the hot seat. My belief is that these coaches have a way to over perform expectations. But to do so requires accepting more risk than almost any ever seem willing to take.

Earlier today, Patrick forwarded Jim and I a terrific article which gives a blue-print for what the underdog should do to win a game. For those too lazy to click through, the idea is this:
- An underdog has an expectation that is worse than the favorite.
- The underdog should work to maximize variance to give themselves a better shot at winning. Increasing variance, even if it slightly hurts your average expectation, increases your top-end expectations and therefore improves your chance of winning.
- Subsequently, the favorite should do everything they can to reduce variance, as the more likely both teams are to perform at their mean expectation; the more likely the favorite is to win.

The article expresses this with examples based in basketball, but the same principles apply to football. Let’s say Team X is a 3 point favorite over Team Y. Team X is expected to score 20 points, Y to score 17. Team X has the best chance of winning the game by forcing the game to play exactly to expectations so that they win by three points. Team Y maximizes its odds of winning by taking risks and trying high variance plays to increase its maximum score-band. This way, they have a better chance of winning the game. The draw-back is that they also are more likely to lose by significantly more than 3 points.

This article helped bring into focus a post that’s been bouncing in my head for some time. The basic premise is simple: A coach of a team likely to lose should take chances on plays with poor expectations, but that can give their team a big edge when they work. That’s not to say they should always take chances. But it’s to say that they should be taking these chances far more often than any coach has ever been known to do. If I were to accept a position as the Carolina Panthers’ head coach for ’11, here’s what I’d do to maximize my chances of winning.

Blitz the quarterback almost all the time, and coach my DBs to jump routes.
Unfortunately I don’t have a lot of data about how often this works in the NFL. But defensively, this will make two outcomes far more regular than what is currently seen. First, my team will force more negative plays for the offense. They will sack the quarterback more often than expected. They will intercept the ball more often. Etc. This turns the tide in my favor, ending drives earlier with fewer scoring and even potentially my team scoring on those plays.

Second, my team will allow more big plays for the offense. If my DB jumps a route, misses and has no safety help over the top, that’s going to be a big gain if not a touchdown. While this is obviously a negative outcome, there are some mitigating factors to them. One, if my team is truly a bad team, my opponent has a pretty decent chance of scoring anyway. Two, my defense is getting off the field quickly, and isn’t likely to be worn down in the fourth quarter.

Kick onside a lot.
“A lot” here I would define as somewhere around 20% of the time. This year, teams scored on average 4.1 times per game. Add in the kick at the beginning of either the first or second half and you come out to around 5 kick-offs per team per game (obviously worse teams will have fewer kickoffs). I think one on-sides kick per game when the opponent isn’t expecting it would be a good number.

For the five teams I listed above, average starting field position allowed was somewhere between the 29 and 34 yard line (Carolina’s was at the 33.2 line). This includes punts, but is a pretty decent proxy for general field position. A failed onside kick would result in my opponent starting somewhere around my own 40-45 yard line. I’m giving them a free 25-30 yards when I fail.

However, when I succeed, I’m also giving myself an extra possession. NFL teams this year averaged somewhere around 2 points per drive. My having a worse offense would hurt this number of course. However, that’s mitigated by the fact that, with the average starting position somewhere between the 25 and 30, my getting the ball starting somewhere past my 40 yard line will increase that value. I also deprive my opponent of the 2 points his drive was valued at. This onside kick could be a net 4 point swing in expectation, a huge advantage in the NFL.

And onside kicks aren’t exactly a very low probability of success. No, I didn’t ask Sean Payton for his opinion there. Instead I went to the data, looking at the 2010 success of the onside kick. I eliminated all fourth quarter onside kicks though, believing that most of those would be expected onside kicks (apologies to Tom Coughlin). There were only 14 onside kicks not in the 4th quarter this season. Kicking teams recovered six of those, a 43% success rate.

So here you have a question of how often you’re willing to give up 25 yards of possession in order to potentially receive a full extra possession yourself. Obviously the more you kick onside, the more it’s expected and the more you’ll fail. But, while I don’t see data for it, I have a hard time imagining 25 yards of field position is worth close to the 4 point swing I would create by getting the extra possession. Even if it’s worth as much as 2 points, I could afford to succeed only 33% of the time and have an equal expectation.

An additional benefit gained is that teams would be forced to react. They would likely be using more of a hands team to guard against onside kicks, which leads to their starting with worse field position when I kick it deep. And if they don’t put out their hands units, my success rate likely moves far closer to the 43%.

Consider never kicking a FG unless it’s 4th and 10+ between my opponent’s 10 and 30 yard line. When kicking, run fakes regularly (25% or more).
I tried finding good data on success of fakes but couldn’t find much. My all plays spreadsheet had only three plays with a fake punt or FG clearly marked. Two of those were successful. But I remember the Ravens ran a fake punt against the Dolphins, which wasn’t on my sheet as a marked fake, so who knows how many I’m missing. Fakes, when unexpected, tend to be fairly successful. Obviously the more they’re expected, the less they’ll work.

When outside my 30, I’d choose to punt and work on having an excellent punting game that can pin my opponents inside their 10 yard line, or to simply go for it on 4th down (see below for more). This replaces a FG attempt of 47+ yards, which across the NFL were less than 58% to be made. When inside the 10, I’d go for the TD as even if I fail, I’m pinning my opponent deep in his territory, making it harder for him to score. 4th and 3-10 yards is somewhere in the range of 25% to convert. This is giving up about a point of expected value, but fits the theme of higher variance even with lower expected value, and is mitigated by your opponent starting inside their ten yard line where if I kicked the FG, the subsequent kick-off puts them around the 30. Twenty yards of field position helps claw back that point of value.

Go for it on 4th and 3 or less every time I’m not in my opponent’s FG range.
Again looking at all 2010 data, the success rate of 3rd down or 4th down and 3 or less to go was 44%. I also looked at the bottom five offenses (AZ, Car, Chi, Sea, StL) in DVOA and found their success rate was 39%. This fits with the idea of taking negative expected value plays which result in significantly positive advantages when they work. When I’m outside my FG range, I extend my drive and increase my chances to score more points in the game. When I’m in my FG range, I improve my chances of scoring a TD for seven points. And the deeper I go into my opponent’s territory, the less likely they are able to drive the field and score if my fourth down attempt fails.

Go for two point conversions every time.
Looking at data on 3rd or 4th and 2 or less from 2010, offenses had a 48% success rate (42% for the bottom five DVOA teams). This again goes with the idea of taking plays with negative expected value, but improves chances of winning significantly if and when they work. Those bottom five DVOA offenses averaged two TDs and a FG per game. First, since I’m not kicking FGs much anymore, I’m more likely to score three TDs instead of two and a FG. Second, if I convert more than one of my two point conversions, I’m requiring my opponent to either score four times, or also convert more than one two point conversion with me to match my three scores.

The total combination of these five things reduces my overall expected point total per week. But it also increases my max expected point total per week, and should increase the number of games I should win against superior opponents. If Carolina would have an expected 5-11 record next season – not unreasonable given the lack of talent – it’s more likely that two things happen:
1) I experience more blow-outs in games I lose, and
2) I win more than five games.
Would I be willing to accept a few 42-7 losses in order to lead my team to an 8-8 or 9-7 record?

This actually comes to the one huge draw-back, and likely reason we don’t see this happen pretty much ever. It’s not assured that I will over perform my expected win total by three or four games by employing these strategies. And coaches with poor winning records who suffer a higher than expected number of 42-7 type blow-outs do not last long in this league. As a coach, I’m more likely to survive longer by simply playing vanilla, expected football and going 5-11 my first year, than I am to play high risk football and steal a few wins to bring my team to the middle of the pack.

It’s easy to armchair quarterback these decisions when I’m never realistically going to get a shot to run an NFL team for a year. But to be one of those 32 lucky individuals getting that shot, I can understand why they would want to maximize their chances of being invited back the following year, even if it means costing themselves a higher win rate. Still, it’s fun to ponder the ideal, and to think about the shot I’d love to take.

*Note: All data for this article were from Football Outsiders.


  1. While you know I am sympathetic with your general thrust, you get into a lot of trouble when you think in absolutes. Always do this, always do that, almost always do the other. It is very easy to see many game situations where this black/white thinking would get your team into a ton of trouble.

    You tie the game as the clock tick to 0:00. Do you go for 2?

  2. I think I write more in absolutes than that I think in them. Possibly a fault but I just skip the word "almost" in there. Obviously there are situations where I wouldn't take the higher risk play.

  3. You do write in absolutes a bit, but what I'm getting at isn't the difference between "almost always" and "always". There are situations where ALL teams should ALWAYS be aggressive, and there are situations where NO team should EVER be aggressive, and I know you aren't talking about those spots.

    But even with the spots in the middle, and all or none approach is dangerous. If you always blitz then they will always throw screen passes. If you on-side kick more than about once every 3 games then you are going to see significant diminishing returns.

    I absolutely agree that finding ways to steal possessions is the trick for worse teams to beat better teams, and I agree that the best way to achieve this is with a high risk/reward style, even one with an overall negative expectation. Where I disagree is how often a team needs to do this to be successful.

    I think there are offenses and defenses that can be installed that naturally lend themselves to this style, and I also think that these tend to be the preferred systems for fans. Everyone hates the WCO. Everyone hates bend-don't-break. Only a few teams can really pull off either successfully, and those are teams that already have superior talent.

    The others should all install Gregg Williams - style defenses (or something similar). The others should all install Peyton-style no-huddles. (I have no clue why this isn't more popular).

    I don't think inferior teams need to install high risk PLAYS to be more successful, they need to install high risk SYSTEMS instead.


About This Blog

Twitter: oblong_spheroid

  © Blogger templates The Professional Template by 2008

Back to TOP