FO DVOA Ratings: Dallas 10th

[JD_KaPow]

A couple notes:

• If special teams weren't included, we'd be 7th, in front of Philly. The Eagles drop to about 17th if ST is excluded.
  
• The week 1 loss to SF is really dragging us down. By my rough estimate, we're 4th in DVOA over weeks 2 through 6, behind Denver, Baltimore and Green Bay.

http://www.footballoutsiders.com/dvoa-ratings/2014/week-6-dvoa-ratings

Yet Dallas, the team now being lauded as the best in the NFC by most NFL observers, is still only tenth in our DVOA ratings.What gives?

The issue is certainly not this Sunday's win. The DVOA system loved this weekend's performance, giving Dallas a single-game DVOA of 57.2%. The final score [was] closer than the performances of the two teams because the Cowboys had special teams gaffes and didn't recover any of the game's three fumbles. The Cowboys also have nice, high ratings for their Week 2 win over Tennessee and their Week 4 victory over New Orleans. However, their very close Week 3 and Week 5 wins (over St. Louis and Houston) are around 0.0%. Put that together with their very bad Week 1 loss to the 49ers, and the Cowboys just don't come out looking like [a] dominant team overall no matter how good they were in Seattle this week.

We still have Dallas as a very probable playoff team, of course. It's tough to blow a 5-1 record, although it is easier to screw up when one of your division rivals is also 5-1. Although the Eagles are slightly higher than the Cowboys in DAVE, the Cowboys have an easier remaining schedule, so we have the Cowboys making the playoffs in 78.3 percent of simulations and the Eagles in 72.4 percent of simulations.

 

[Idgit]

I can live with that. Especially since what's important when it finally comes down to the playoffs will be the recent trends and not the cumulative season total. You'd almost think this would be a combination of the cumulative score and a rolling three week snapshot, where the rankings were based on the recent performances rather than the cumulative total. That would give you a better idea of who was hotter at the moment.

 

[TheMarathonContinues]

Their system seems to be flawed because of late game scoring in games where the game was decided. Its whatever. I'll take a top 10 team. I think we're better than that but i'm not quick to call us the best team in the league yet.

 

[JD_KaPow]

Their system seems to be flawed because of late game scoring in games where the game was decided. Its whatever. I'll take a top 10 team. I think we're better than that but i'm not quick to call us the best team in the league yet.

They take garbage time into account, but they're careful not to be too aggressive about that: after all, big comebacks are getting more common.

Anyway, one of the things I like about this is that it doesn't overreact to a single game, no matter how impressive or awful that game is. Dallas is, after all, the same team that barely squeaked by Houston and St. Louis not all that long ago.

 

[JD_KaPow]

I can live with that. Especially since what's important when it finally comes down to the playoffs will be the recent trends and not the cumulative season total. You'd almost think this would be a combination of the cumulative score and a rolling three week snapshot, where the rankings were based on the recent performances rather than the cumulative total. That would give you a better idea of who was hotter at the moment.

Later in the year, I believe around week 9, they switch to a weighted DVOA, where earlier games in the season get progressively lower weights. Assuming Dallas keeps up the good play, their march up the ratings should accelerate when Week 1 starts getting devalued.

 

[AdamJT13]

DVOA is a yards-based statistic. Our defense is 29th in yards allowed per play (and 23rd in yards allowed per possession). Teams have moved the ball on us quite well. But our defense has been good at getting turnovers (fifth in turnovers per possession) and making stops when we need one (seventh in points allowed per possession).

So, the end result is that we're better on the scoreboard (and therefore in wins/losses) than our per-play stats would suggest, mainly because of those differences in the defense's numbers.

 

[JD_KaPow]

DVOA is a yards-based statistic. Our defense is 29th in yards allowed per play (and 23rd in yards allowed per possession). Teams have moved the ball on us quite well. But our defense has been good at getting turnovers (fifth in turnovers per possession) and making stops when we need one (seventh in points allowed per possession).

So, the end result is that we're better on the scoreboard (and therefore in wins/losses) than our per-play stats would suggest, mainly because of those differences in the defense's numbers.

Well, DVOA certainly takes the takeaways into account, although not the particular timing of them, as you say. It's probably true, though, that the big discrepancy we've had so far between yards per drive and points per drive isn't really sustainable. Something's got to give.

 

[Fletch]

It all comes down to our offense controlling the clock, which in return has benefited our defense, keeping them fresh for the second half of games.

 

[percyhoward]

Adam already said most of what I was going to say. Football Outsiders uses DVOA for its supposed predictive ability, which means yards get more weight than takeaways and, therefore, even points. Takeaways (and even simply good red zone defense, to a degree) are considered random occurrences. Also, you get penalized for having played weaker opponents, as if this says something about how you would have fared against stronger opponents. Maybe it does, but you can only find that out by facing them.

You'd rather lead the league in points per drive differential than DVOA. And DVOA should catch up to that stat over the course of the season anyway. Eventually.

 

[JD_KaPow]

Adam already said most of what I was going to say. Football Outsiders uses DVOA for its supposed predictive ability, which means yards get more weight than takeaways and, therefore, even points. Takeaways (and even simply good red zone defense, to a degree) are considered random occurrences. Also, you get penalized for having played weaker opponents, as if this says something about how you would have fared against stronger opponents. Maybe it does, but you can only find that out by facing them.

My criticism of DVOA is along these lines: it doesn't really know if it wants to be predictive or descriptive.

However, takeaways are not considered random occurrences. Offenses are penalized and defenses are credited for forced fumbles and INTs. Fumble recoveries are considered random (not 50-50--certain fumbles are more likely to be recovered by one team or the other--but not indicative of a particular skill). And fumbles and INTs really kill your DVOA, so it's not clear to me that they are undervalued relative to yards.

And I find your criticism of weighting based on strength of the opponents odd. Of course you have to take that into account when evaluating teams against each other. It obviously isn't perfect, but it's far better than treating a game against the Jaguars the same as a game against the Broncos.

Points per drive differential is a fine descriptive stat, but not so much of a predictive stat.

 

[Nation]

There is a lot about DVOA that I believe in (strength of schedule, fumble luck, record in 1-7 point games) and it still spits us out as a top ten team so I don't really have any problems with it. Considering who we have faced it probably is fair to call the defense league-average although I have hopes they will continue to improve. If we bully New York and Washington around like we should I think DVOA will show us love similar to the rest of the metrics out there.

 

[percyhoward]

My criticism of DVOA is along these lines: it doesn't really know if it wants to be predictive or descriptive.

However, takeaways are not considered random occurrences. Offenses are penalized and defenses are credited for forced fumbles and INTs. Fumble recoveries are considered random (not 50-50--certain fumbles are more likely to be recovered by one team or the other--but not indicative of a particular skill). And fumbles and INTs really kill your DVOA, so it's not clear to me that they are undervalued relative to yards.

And I find your criticism of weighting based on strength of the opponents odd. Of course you have to take that into account when evaluating teams against each other. It obviously isn't perfect, but it's far better than treating a game against the Jaguars the same as a game against the Broncos.

Points per drive differential is a fine descriptive stat, but not so much of a predictive stat.

Oh DVOA wants to be predictive. Merely accounting for turnovers isn't the same as weighting them in proportion to their effect on scoring. DVOA can't do that because of the randomness of turnovers.

And I'm not criticizing the idea of taking strength of opponent into consideration. On average, it's obviously a successful indicator of future success. Just pointing out that, when all is said and done, you'd rather have the edge in the stat with the higher win correlation, like points per drive. If the Cowboys drop in points per drive differential, DVOA was right. If they don't, DVOA will catch up by season's end.

 

[AdamJT13]

Well, DVOA certainly takes the takeaways into account, although not the particular timing of them, as you say.

It does take turnovers into account, but how those plays are weighted or scored would make a big difference. DVOA takes into account your performance on every single play, while in reality, it's usually only a handful of plays that really determine whether you win or lose.

It's probably true, though, that the big discrepancy we've had so far between yards per drive and points per drive isn't really sustainable. Something's got to give.

Those rankings probably will migrate closer together (hopefully in the right direction), but there have been teams that have sustained similar discrepancies for entire seasons. There usually is a team or two with a difference of about 10 positions in the rankings, and some teams have had bigger discrepancies. The 2010 Falcons, for example, had full-season rankings that are almost the same as ours right now -- 23rd in yards per possession, sixth in points per possession and sixth in turnovers per possession.

 

[JD_KaPow]

Those rankings probably will migrate closer together (hopefully in the right direction), but there have been teams that have sustained similar discrepancies for entire seasons. There usually is a team or two with a difference of about 10 positions in the rankings, and some teams have had bigger discrepancies. The 2010 Falcons, for example, had full-season rankings that are almost the same as ours right now -- 23rd in yards per possession, sixth in points per possession and sixth in turnovers per possession.

Sure. A 16 game season is really short, and things don't always average out over such a time period. But if you're trying to predict how a team will do going forward, unless you have a model that consistently accounts for such discrepancies, assuming that they will average out is the best bet.

There have been plenty of baseball teams that have sustained winning records over 162-game seasons despite giving up more runs than they scored. But in predicting their performance the next season, the run differential is generally more useful than the W-L record.

 

[JD_KaPow]

Oh DVOA wants to be predictive. Merely accounting for turnovers isn't the same as weighting them in proportion to their effect on scoring. DVOA can't do that because of the randomness of turnovers.

And I'm not criticizing the idea of taking strength of opponent into consideration. On average, it's obviously a successful indicator of future success. Just pointing out that, when all is said and done, you'd rather have the edge in the stat with the higher win correlation, like points per drive. If the Cowboys drop in points per drive differential, DVOA was right. If they don't, DVOA will catch up by season's end.

Well sure. When all is said and done, I'd prefer to have the edge in the stat with the highest possible win correlation: wins. If I want to understand if the points per drive differential (or the win rate) is sustainable going forward, I need to understand how much to regress the various components that go into it (i.e. build a predictive model).