## WP: Instant Offensive Improvement

I've poured over the data, and spent the last few days doing lots of advanced math, and I think I've figured out the Redskins' problem: They don't have any good players on offense.

There isn't much that can be done about their roster at this point in the season, but here are five things the 'Skins can do to improve their offense overnight.

### 12 Responses to “WP: Instant Offensive Improvement”

1. Anonymous says:

Variance per play is not the same as variance per game. If the objective is to increase game variance, then you have two competing strategies: increase variance per play and reduce the total number of plays per game [by running low variance plays]. I.e., the equiv. of the old Princeton basketball offense (hold the ball and hit threes). The problem in football is that you can't hold the ball to kill clock and then make a high variance play every possession...without significantly reducing your overall efficiency.

2. Jeffy B. says:

I read this first over at the Washington Post site and as a Redskins fan, I greatly enjoy reading your stuff about them specifically. I think this analysis is spot on with respect to what they need to do to try to win games. I find it frustrating to watch teams that are down by several scores in the second half punt the ball to the other team on fourth down, no matter how far to get a first. Once a team starts to get behind, they have to take chances to try to catch up, otherwise they are just playing to lose less badly. With the rare exception of the safety, a team can't score if they don't have the ball, so to give the ball intentionally to the other team is not something to be taken lightly.

I guess I do consider it to be a positive sign that they appear to be using the rest of the season to audition free agents. I would like to see them try out a ton of quarterbacks, whether from other pro leagues or UDFAs, as they don't seem to have other options at this point. In some ways, having a half a season that they don't have to worry about winning is too valuable of an opportunity to squander. Most teams never get the chance to try out talent and experimental schemes against NFL caliber opponents who are actually trying to win games.

3. Will says:

Brian, Anon. above makes an interesting point (echoed by one of the comments at WaPo) that the mechanics of football prevent you from running the "Princeton" type offense that is both trial reducing (fewer plays) and variance increasing. Might there be a way to measure the tradeoff in variance caused by each incomplete pass resulting in more plays/possessions? It seems like this is already inherently in WP, but might be hard to pull out.

4. Jeffy B. says:

One rough equivalent to the Princeton offense in terms of the total number of drives for each team might be to not kick field goals or perhaps not kick field goals or punts (ala Kevin Kelley http://sportsillustrated.cnn.com/2011/writers/scorecasting/09/15/kelley.pulaski/index.html ) The team trying this strategy would just try to get the first down until they get into the end zone. There might be fewer possessions for either team in the game, due to having longer drives, but you would be going for a bigger payout. I guess a team could also go for a two point conversion on every touchdown, but that wouldn't really lengthen the possessions by that much.

5. James says:

Extra points and two-point conversions are untimed, so attempting them makes no difference to the length of the game.

6. Odoacer says:

Brian,

Your poker analogy is incorrect. A short stacked player going all in is analogous to a player passing a lot late in the game when down. Both good and bad players do this as do both good and bad football teams.

A better analogy is the "Kill Phil" strategy which emerged a few years ago after being developed by poker mathametician David Sklansky. The strategy is useful for someone playing against a clearly superior player (the large number of successful poker pros named Phil is the genesis of the name). The strategy actually calls for playing far few hands than what standard theory would dicatate but betting aggressively, if not all in pre-flop, when a strong hand emerges. The most basic version is to go all in pre-flop whenever one has AA, KK, or AK but fold every other hand. By doing so, the outcome of the play will usually be 50/50 regardless of the skill of the other player. The key idea is not to increase variance per hand but rather to lower the the impact of skill (or R2 of skill on the regression). Which it seems to me is exactly what an inferior team should do.

What would the "Kill Phil" strategy look like in football? I'm not sure but perhaps running 5 straight times right up the middle, taking as much time off the clock between plays, followed by a play action pass for a 50yd bomb. The question remains as to how much play calling can impact conditional probabilities.

Regardless, while your thesis on increasing variance is mathamatically correct, an alternative model which lowers variance but also lowers the impact of skill can also be acceptable.

Greg

PS. The Princeton offense seeks to do both. A slow, methodical offense lessens the impact of athleticism/skill while an emphasis on 3pt shooting increases variance.

7. JD Mathewson says:

Since you can't really run a Princeton basketball offense in pro football, and simply because football isn't basketball, I'd say that, for the most part, same strategies that maximize variance per play in football will maximize variance per game.

I qualify that because you want to adjust your risk acceptance for the leverage of any given play at that point in the game. You want to increase the chances of big plays (and big turnovers) in situations where the reward is high, not low.

But that's always true, even with the Princeton offense.

8. Anonymous says:

What did you pour over the data?

9. bigmouth says:

Does this mean the Redskins should go back to Sexy Rexy? The man is the very definition of high variance, lol.

10. Brian Burke says:

Help me out! What's the proper spelling?

11. bigmouth says:

You "pore" over data -- no "u".

12. Brian Burke says:

Thank you!