≡ Menu

Does pre-season strength of schedule matter?

NFL.com posted an article yesterday looking at the strength of schedule for each team in 2013. We have known each team’s opponents since the end of the regular season, and while the full schedule won’t come out until April, it’s simple to calculate a team’s strength of schedule for 2013. Usually, the media reports this by looking at the win-loss record of each opponent from the prior season. Here are the projected SOSs for each team next season:

TeamSOSOpponent record
Carolina Panthers0.543138-116-2
Detroit Lions0.539138-118-0
New Orleans Saints0.539137-117-2
St. Louis Rams0.539137-117-2
Baltimore Ravens0.535137-119-0
Green Bay Packers0.533136-119-1
Arizona Cardinals0.52131-121-4
Miami Dolphins0.52133-123-0
San Francisco 49ers0.52132-122-2
Minnesota Vikings0.516132-124-0
Seattle Seahawks0.516130-122-4
Cincinnati Bengals0.508130-126-0
Jacksonville Jaguars0.508129-125-2
New England Patriots0.508130-126-0
Atlanta Falcons0.504128-126-0
Chicago Bears0.502128-127-1
Tampa Bay Buccaneers0.5127-127-2
Washington Redskins0.498127-128-1
New York Jets0.496127-129-0
Philadelphia Eagles0.496127-129-0
Cleveland Browns0.492126-130-0
Pittsburgh Steelers0.496126-130-0
Tennessee Titans0.488124-130-2
New York Giants0.48123-133-0
Dallas Cowboys0.48121-134-1
Buffalo Bills0.473121-135-0
Houston Texans0.473120-134-2
Kansas City Chiefs0.473121-135-0
Oakland Raiders0.469120-136-0
Indianapolis Colts0.461117-137-2
San Diego Chargers0.457117-139-0
Denver Broncos0.43110-146-0

So does this mean that the Panthers job isn’t so attractive anymore? Not really. Strength of schedule absolutely matters: teams like the 2010 Kansas City Chiefs and 2012 Indianapolis Colts made the playoffs in large part based on easy schedules. But can you predict in the off-season which teams will have the hardest and easiest schedules?

I looked at the preseason strength of schedule for every team from 2003 to 2012. The correlation coefficient between a team’s predicted SOS and actual SOS measured at the end of the season was just 0.06 in 2012 and 0.24 over the ten-year period from 2003 to 2012. If you were to run a regression, the best fit formula to predict end-of-year strength of schedule from predicted strength of schedule is:

Actual SOS = 0.369 + 0.262*Proj_SRS

In English, that means we would project the Broncos (pre-season SOS of 0.430) to have an actual SOS of 0.482, while the Panthers (pre-season SOS of 0.543) to have an SOS of 0.511 by the end of 2013. Another way of thinking of that is that all strengths of schedule regress to the mean (of 0.500), with only 26% of a team’s actual SOS being dictated by their projected strength of schedule. However, the R^2 is just 0.06, indicating no meaningful relationship.

If instead we use the Simple Rating System to give us a projected strength of schedule, the best fit formula becomes 0.500 + 0.009*SRS.1 That means if your average opponent was 3 points better than average in Year N-1, you could expect that team’s strength of schedule at the end of the year to be equal to 0.527. If you tried to use pre-season SOS based on both Year N-1 win-loss records and Year N-1 SRS results, the Year N-1 win-loss records variable is not statistically (or practically) significant. You can view the projected SOS based on win-loss records and SRS (using Year N-1 data) and the actual SOS based on win-loss records and SRS (using Year N data) for all 320 team seasons here.

Let’s wind the clock backs one year. At the time, the story was that the Giants and Broncos had the league’s hardest schedules, while the Packers and Patriots were gifted the easiest slate of opponents. How did that turn out?

TmProj. SOSProj. SRSAct. SOSAct. SRS

Denver ended up having the 4th easiest schedule in the league, while the Packers schedule ended up being slightly harder than average. If you look at the SRS ratings, the results held a bit better. So what are the projected end-of-season strength of schedule ratings for each team in 2013 using the SRS (and the regression formula above) instead of win-loss records as the input?

TeamProj. SRSAdj. Proj W/L SOS
Carolina Panthers2.60.522
New Orleans Saints2.40.521
Minnesota Vikings2.30.52
Green Bay Packers2.10.517
Detroit Lions20.517
Atlanta Falcons20.517
St. Louis Rams1.90.516
Tampa Bay Buccaneers1.40.512
Arizona Cardinals1.20.51
Baltimore Ravens1.20.51
Chicago Bears1.20.51
Philadelphia Eagles0.90.508
San Francisco 49ers0.90.507
Washington Redskins0.60.505
Seattle Seahawks0.50.505
Dallas Cowboys0.10.501
New York Giants00.5
Miami Dolphins00.5
Cincinnati Bengals00.5
Cleveland Browns-0.40.497
New York Jets-0.40.497
Pittsburgh Steelers-0.50.495
New England Patriots-0.60.495
Buffalo Bills-0.70.494
Jacksonville Jaguars-20.483
Tennessee Titans-20.483
Indianapolis Colts-2.30.48
Houston Texans-2.30.48
Kansas City Chiefs-2.60.478
Oakland Raiders-2.70.477
San Diego Chargers-3.30.472
Denver Broncos-3.90.467

I’ll note that I’m hardly breaking new ground showing that a team’s projected strength of schedule based on the prior year’s win-loss record is meaningless. Doug wrote a study on this back in 2006, and Football Outsiders has shown that their DVOA stat is more highly correlated with future SOS than win-loss record. But it’s good to break out the data and confirm this every now and again.

Here’s another way to think of it, building off Doug’s study. I looked at the regular season records of the 320 teams from 2002 to 2011. If you knew only a team’s record in Year N-1, the best projection of their record in Year N would be:

Year N Win% = 0.353 + 0.295*YrN-1Win%

The R^2 in this formula is just 0.08, indicating not much of a relationship. What if we added as an input each team’s projected strength of schedule? That variable turns out to be statistically (and practically) insignificant, and the R^2 does not change.

  1. The R^2 in this case is only 0.12, so the relationship in general is still very weak. []
  • We were obviously in our own little corner of the internet back then, but here’s an SOS analysis I did for NN in January of 2011 that shows how the correlations have changed over time, and why even a significant ProjSOS-ActSOS relationship has little practical application:


    That was an update of this post (http://www.ninersnation.com/2009/4/28/856882/movin-on-up-in-the-2010-draft-how), which I’m only linking to here because it’s a perfect example of how sometimes we create methods for an analysis on the fly that — ignored by us at the time — actually had more far-reaching implications than the results of the analysis itself when you look back 4 years later.

    • Chase Stuart

      Good stuff.

      The CCs jump around year-to-year due to the small sample size, but in general, looking at projected SOS using last year’s W/L records has no practical value. It does look like SRS and DVOA and probably some other advanced stats have some merit, but even then, it’s very limited.

  • Libertarian soldier

    Why am I not seeing NE on the list? Something wrong with my browser?

    • Ryan

      Agreed soldier, I don’t see NE in the 2nd table.

      • Chase Stuart

        Thanks guys. The table now has them.

  • mrh

    As Danny’s linked post implies, SoS for year N based on year N-1 overall W-L pct (or SRS) is such a crude measure. Assume your team is a [morally] 8-8 team. Would your rather play a schedule of 16 games against all [morally] 8-8 teams? Or would your rather play a schedule of 4 x 16-0 teams, 8 x 5-11 teams and 4 x 6-10 teams [again, all moral records]? Both schedules are collectively .500, but your chances of winning 10-11 games and making the playoffs are much higher with the 2nd schedule.

    Obviously, this is an extreme example but it seems there might be some value in looking at a season N schedule game by game, estimating the chances of winning and losing each game for a morally .500 team (using season N-1 W-L pct, SRS, DVOA, or whatever your preferred stat is), and then then using that W-L pct to rate SOS in season N. More work than I could do but I’d read someone else’s study. 😉

  • George

    Less than a week into the off-season and we’ve already broken out regression analysis and (conceptually) a projection towards next season (or proving that an SOS based on the previous year’s win/loss – which generally is regarded as a worthwhile ranking by those that haven’t dug into a little bit – is actually pointless). Great stuff.

    • Chase Stuart

      Thanks George. Don’t worry, Football Perspective isn’t going anywhere in the off-season. But I already knew I could count on you and Richie as daily visitors.

  • Richie

    I’m hardly breaking new ground showing that a team’s projected strength of schedule based on the prior year’s win-loss record is meaningless.

    Too bad so many commentators don’t understand this.