Let's get this year started on a good note. Both in Green Bay, and here on NN. I present an argument for why immense regression in 2012 should not be expected. Or, rather put, why our progress in 2011 made sense.
Our DVOA was on a steady climb, from '07 through '09, but we hit a road-bump named Mike Singletary in '10, and rather that continuing our upward trend into positive DVOA, we wound up at -11.2%.
- Year : Team DVOA
- 2007 : -33.4%
- 2008 : -14.1%
- 2009 : -1.2%
- 2010 : -11.2%
- 2011 : 18.6%
-33.4% to 18.6% is a 52% change in four years, for a 13% change each year.
Last year, the top eight teams in DVOA, ranging from 27.0% to 13.9%, all had at least 10 wins. In 2010, the top seven teams (from 44.6% to 16.3%) all had at least ten wins. In 2009, seven of the top eight (29.1% to 16.5%) had at least ten wins. In 2008, seven of the top nine (31.8% to 13.1%) had ten wins. In 2007, the top seven teams (52.9% to 18.2%) had at least ten wins.
While DVOA, by Football Outsider's own admission, does not have much of a correlation to winning (nor is it really supposed to), in the past five years, at least, teams who score high do tend to win games. So how does FO justify its low win projection for the Niners? They expect a huge drop in team DVOA.
How huge? 25.5% ,to end the season at -6.9%. To be clear, FO is predicting us to fall to our 2008 and 2010 levels.
San Francisco's upward trend in DVOA is not, however, the only such area where that trend showed.
- Year : Point Differential (Rk.)
- 2007 : -145 (29th)
- 2008 : -42 (23rd)
- 2009 : +49 (13th)
- 2010 : -41 (20th)
- 2011 : +151 (4th)
Again, from -145 to +151, with four opportunities to improve, means an improvement of 74 every year. The 49ers were improving from year-to-year, until Singletary in 2010.
Point differential is well known to have a large correlation to both wins and future wins through what is called the Pythagorean Expected Record.
Pythagorean expectation is a formula created by famed sabermetrician, Bill James, for use in baseball. The coefficient for the formula is 2 in that sport, but people have found that a slightly higher coefficient for football works best, and the official one used by Pro Football Reference is 2.35.
When we take PS^2.35 and divide it by PS^2.35 + PA^2.35 (where PS = points scored; PA = points against), we come up with a team's expected record. This accounts for luck and also for a team's record in close games.
In other words, not only does this give us an accurate depiction of a team's real value, but it has predictive qualities.
- Year : Expected Wins
- 2007 : 3.7
- 2008 : 6.9
- 2009 : 9.5
- 2010 : 6.8
- 2011 : 12.3
The Pythagorean expectation loves low-scoring affairs that are a bit closer, to high-scoring affairs that might be further apart. For example, it regards a 17 - 10 victory slightly higher than it does a 38 - 28 victory, even though the deficit in the high-scoring game is 10 - and it especially prefers a 17 - 10 victory over a 45 - 38 one.
That might seem strange, but it makes sense when you think about it in terms of a game, or even a season. If a defense can hold teams to a low amount of points, then a seven point lead can be quite advantageous. But a seven point lead with a defense that has already given up 38 is not worth much at all. What's to stop those last seven points from tying it up? Certainly not your defense, which already gave up 38.
Because of this, Harbaugh's philosophy fits very well with the Pythagorean formula. All those Akers field goals made me grit my teeth just as much as you, and playing to hold a small lead near the end of the game is risky - but what it did do is give the 49ers the best expected record in the entire NFL in 2011.
Yes, even better than the Green Bay Packers, who finished with 15 wins, but were only expected to win 11.9; and better than the Saints, who led the league in point differential at 208, but were expected to win 12.1 games.
What Expected Wins does really well is it, as mentioned above, accounts for luck. Luck, such as having a really good record in close games (contests decided by one to seven points).
For example, the 49ers were 6 - 2 in close games last year, and if you "normalize" that record, it gives them two less wins/two more losses, moving their record to 11 - 5. Similarly, the Patriots were expected to win 11.6, but actually won 13 on the strength of a 4 - 2 record in close games (RiCG). If we "normalize" their RiCG to 50%, they lose one more game, and go 12 - 4, much like what their Expected Wins predicted.
I went ahead and "normalized" every team's record in 2011, and the results showed the same thing: in over 90% of cases, their "normalized" record was closer to their Expected Record than their actual record.
In other words, Pythagorean expectation is a very accurate tool for predicting a team's record, and the only reason it is off at all is because a team's RiCG is above or below 50%.
For the 49ers, this was actually off just a bit: 12.3 expected wins, 13 regular wins, and 11 normalized - but not by a lot, and I think it's because the Pythagorean system really values the way we played football last year. For a large majority of cases, it works really well, and even for us it's not like it was off by a significant margin (0.7 to 1.3; or 0.6 difference).
Another such working example, Oakland was expected to win 6.3 games, but actually went 8 - 8. However, they were an amazing 7 - 2 in close games. If we assume such a mark is unsustainable, and we "normalize" their RiCG to 50%, they go 4.5 - 4.5 in close games, and their "normalized" record becomes 5.5 - 10.5. 5.5 is much closer to their 6.3 expected wins than their 8 actual wins.
And, just to give you one more solid example, but in table form, we can look at the Colts for the last decade. The team consistently outperformed their Expected Wins because they consistently had an RiCG over .500. Thanks, of course, to Peyton Manning.
- Year : RiCG, (Actual Record), Normalized Record, [Pythagorean Expected Record]
- 2010 : 5 - 4, (10 - 6), 9.5 - 6.5, [9.1 - 6.9]
- 2009 : 7 - 0, (14 - 2), 10.5 - 5.5, [10.8 - 5.2]
- 2008 : 8 - 1, (12 - 4), 8.5 - 7.5, [10.2 - 5.8]
- 2007 : 5 - 3, (13 - 3), 12 - 4, [12.5 - 3.5]
- 2006 : 8 - 3, (12 - 4), 9.5 - 6.5, [9.6 - 6.4]
- 2005 : 3 - 0, (14 - 2), 12.5 - 3.5, [12.7 - 3.3]
- 2004 : 3 - 2, (12 - 4), 11.5 - 4.5, [11.5 - 4.5]
- 2003 : 7 - 3, (12 - 4), 10 - 6, [10.6 - 5.4]
- 2002 : 6 - 1, (10 - 6), 7.5 - 8.5, [9.0 - 7.0]
Notice in every year, except 2002, the Colts' Normalized Record brought them closer to what the Pythagorean formula would suggest their record should be.
Again, just to be clear, this means the Pythagorean formula has great predictive value, but it cannot tell us a team's RiCG.
If a team does well in close games, they outperform their Expected Record; if they don't, they underperform. But if we assume that a team's RiCG is bound to gravitate towards .500 from year-to-year (which is not necessarily a good assumption), then the Pythagorean formula becomes spot on (unless you are Peyton Manning, and you happen to have a knack for winning close games - and let's hope the 49ers' defense has similar clutch abilities).
Regardless, if you can predict a team's points for and against, you can estimate their record. And if you can temper that with any potential variance in their RiCG, then you can give them a range of records. This is what I'll be doing for our 2012 record at the end of this post - but, for now, one more table.
- Year : Normalized Wins
- 2007 : 4
- 2008 : 6
- 2009 : 10
- 2010 : 7.5
- 2011 : 11
Going through the past five years and looking at our Normalized Wins (i.e., our wins once our RiCG is adjusted to 50%), we see another upward trend for San Francisco.
Even accounting for the "luck" of 2011, in which we went 6 - 2 in close games, we still win 11. But, temper that with our "unlucky" years, such as '10, where we went 1 - 4, and '09 where we went 2 - 6, and what you get is a "normalized" record that accounts for luck, much like Pythagorean Wins does.
And, again, you see a trend.
Improvement every year, except in 2010. Just like DVOA and point differential and Expected Wins, our Normalized Wins suddenly revert in '10 to '08 levels. Then, just like in every other table, our Normalized Wins pick right back up on that upward trend in '11 and the arrival of Jim Harbaugh.
Having the Proper Perspective on 2010
What we have established then, first of all, is an obvious upward trend in significant statistical categories. They all show the exact same trend - that being that 2010 was a major disappointment filled with unsustainable unluckiness.
Yes, read that again: unsustainable unluckiness. Let's copyright that phrase here at NN and make it our rallying cry for the Singletary Era.
And let me state here, it's not that Singletary himself was unlucky. No, Singletary was perhaps the worst head coach in NFL history - or, put another way, the difference between Singletary and Harbaugh might be the greatest head coaching change in NFL history. Regardless, it was not Singletary being unlucky, it was this team being unlucky in having Singletary as a coach. Such unluckiness in happening to have one of the worst head coaches in history was unsustainable.
Thus, it was not that our accomplishments in 2011 were unsustainable. No. Instead, it was our terrible run of luck in 2010, both in RiCG, and in having Singletary for a head coach, and in losing Alex Smith and Singletary deciding to start a different Smith who in 2011 backed-up Eric Crouch (who?) for the UFL (what?) team Omaha Nighthawks (heh?) and had a chance to start when Crouch got injured, but the team instead decided to start Jeremiah Masoli (...um?) - it was all this dreadful,dreadful, unsustainable unluckiness that made the let-down that was 2010 an unsustainable year.
2011 was statistically bound to be better. And it was. No surprise. No shock.
If 2010 had not been marred with such unsustainable unluckiness (you'll never forget that phrase), 2011 would have just been another good year on top of the steady improvement of the past five.
SIDE NOTE: And right here I really, really, really want to say that Alex Smith's career has been marred with perhaps the greatest run of unsustainable unluckiness in history. Such a hypothesis would help us understand his under-performance for the beginning of his career, and his sudden improvement once he actually got a good head coach and stayed healthy. It will also help us understand what I predict to be his continued improvement in 2012 now that this offensive system is sticking around, and it's Jim Harbaugh's system, no-less. But all that for another post...
So what I would like to do, now, is use our projected DVOA / mean wins via FO to predict a point differential; and then use other point differentials to predict our Expected Wins. Then, I'd like to finish with some possible "normalized" records for RiCG variance.
In other words, I'd like to show you a realistic "range" of records using similar data as above.
Predicting the Future
FO has us at a mean of 7.5 wins, which roughly equates to about 46.8% of games. The Expected Wins equivalent of that is for us to give up something like 329 points, and score 311.
Now, there are different variations of that. For example, we could score 406 and give up 429, but that seems rather unrealistic. The only "realistic" way we could be projected to win 7.5 games is, in my opinion, to give up 329 and score 311.
To judge how "realistic" this most "realistic" option is, let's take a look at our scoring for the last five years.
- YEAR : PF; PA
- 2007 : 219 : 364
- 2008 : 339 : 381
- 2009 : 330 : 281
- 2010 : 305 : 346
- 2011 : 380 : 229
So we would need to give up 100 more points than last year, and 40 more points than '09. We would have to revert to about '10 levels when we were coached by Singletary.
Is that possible? 100 more points means about 6 more a game, and that means around 20.5 per game for the season. Let's make a bold assumption and say the Saints, Patriots, and Packers score an average of 28 against us. That would amount for 42 more points than last year's average right now - so where do the rest of those points come from?
I'm not sure.
And what about the offense? Remember, even if we somehow give up 100 more points than last year, we'd also have to score 69 less. Given how average our offense was in 2011, that would require a big drop to one of the bottom ten offenses in the league.
Alex Smith would have to turn into nothing, despite finally having a good head coach and the same offensive system for a second year. And our run game would need to disappear. Not only that, but we scored 380 points last year despite all our field goals. If our red zone percentage improves, but our general ability to sustain a drive does not, do we really score 69 less points? Inversely, if we actually put drives together, but our red zone percentage remains one of the worst in the league, do we score 69 less? And if we do, does our defense somehow still give up 100 more?
I just don't see either of those things happening.
For the sake of fun, let's go ahead and give us a PF total that is a reflection of the past couple of years - such as 340.
This implies a 40 point drop in offense. In order to expect 7.5 wins, we would need to give up 360 points. Again, that just doesn't make sense. It would mean 131 more points given up than last year. Our defense would have to fall off a cliff.
Well, what if we score 40 points less, but our defense gives up those extra 42 points on the road to GB, NE, and NO, and averages it's 14 a game total from last year in every other game, for a total of 271 on the year. That gives us 10.08 wins.
As far as I can tell, this would be the most realistic drop possible, and we still end up at 10 wins. If we give us an unlucky year in close games - such as a 2 - 6 record - then we could go 8 - 8. But if we go .500 in close games, then we win 10 - and if we do better than .500, we win 11 or even 12.
A point differential of 69 would put us at roughly '09 levels; and our expected wins and normalized wins would also revert back to '09 levels.
If we accept the premise that 2010 - and not 2011 - was an anomaly, then reverting to '09 levels is the only realistic regression we could honestly expect.
And we would still win 10 games.
Which is why I'll keep doing what I've been doing, and putting the over/under at 10 games. I expect us to go OVER that, because I think our offense will improve more than our defense regresses; but, at worst, based on the drop off I've detailed above, we would win around 9 - 11 games depending on our RiCG.
Not only that, but 10 games matches the roughly 0.5 regression of our win expansion from '10 to '11. The same regression that other rookie head coaches in the past couple of years have had when improving their team by six or more wins.
- John Harbaugh, in his rookie year (2008), improved the Ravens by six games to 11 wins. In '09, they beat the .500 mark of regression, by winning nine games (half of six is three, but Baltimore regressed only two).
- Mike Smith improved the Falcons by seven wins, also in '08, to 11 wins. The following year, they only regressed by two - again, beating the half-mark of regression, or 3.5.
- Sean Payton's rookie season marked a seven win improvement as well, to 10 wins. The next year, 2007, they won seven games, for a regression of three, which also beats the half-mark of 3.5
- In 2000, Jim Haslett led the Saints to a similar seven win gain, to 10 wins. In 2001, they also regressed three wins, to seven - still beating the 3.5 mark.
- In 2008, rookie coach Tony Sparano led the Dolphins to 10 more wins, totalling 11. In '09, they regressed four wins to finish 7 - 9. Again, this also beat the half way mark of five.
These coaches are the only ones in history to ever take a losing team and improve it by at least six wins in their rookie season. In 2011, Jim Harbaugh joined them, by taking the 49ers to a seven win improvement - all the way to 13.
If we set the goal at the half-way regression mark (i.e., 3.5 more losses), then we would expect the 49ers to win 9.5. This matches well, once again, with setting the over/under at 10 wins. And if Harbaugh does what every other coach on this list did, he will beat that mark by only regressing three games or less (i.e., 10 wins or better).
10 wins is a low-end estimate, but 10 wins is a realistic low-end estimate. If we have a particularly unlucky season, we could win 9 or 8. But, if we go .500 in close games or better, then we will win at least 10 games.
And again, this all assumes my low-end estimate of points for and points against. I personally believe we will do better than that, as our defense will stay elite and our offense will show improvement.
But that's not what I want to argue here. I just want to argue that 7.5 wins is silly and unrealistic. You would have to expect us to have a similar amount of unsustainable unluckiness as 2010 to finish with seven, or even eight, wins.
The realistic regression is 10.
But, knowing Jim Harbaugh...
we'll go 19 - 0.
Let's start, today, by beating Aaron Rodgers and the Green Bay Packers, because Rodgers is gonna look soooooo great in an Alex Smith jersey.
WHO'S GOT IT BETTER THAN US!?
To end, I just saw a commercial for college football where people were talking about "the 12th man." It got me thinking, everyone in every stadium refers to their fans as "the 12th man"; but our team is the most complete team in the NFL, and it takes "all 53" of them to win games.
Remember Schembechler, "The Team. The Team. The Team."
So I make a motion that we call 49er fans "the 54th man." Eh? Ehhhh!?!?! You know you like it.