OR “I don’t have a football blogging problem. I can quit any time I want. Really.”
As we were kindly reminded by an ESPN infographic during Saturday’s game, the Michigan State football team has a reputation for getting off to a good start each season and then collapsing down the stretch. If you take the last five years and split each season into a “good half” and “bad half,” MSU has a record of 22-5 in the opening halves of those seasons and 7-27 in the closing halves.
Of course, many of the games in good halves are nonconference games against weaker opponents (14-4 record in nonconference games). And the conference schedule has tended to be backloaded. So it’s possible that the declines in MSU’s win-loss record from the first half of the season to the second half of the season have been a function of strength of schedule rather than a fall-off in actual football performance.
I thought we’d take a statistical/graphical look at this topic. For each game since 2003 (covering the John L. Smith era and the beginning of the Mark Dantonio era), I’ve computed a game rating for each of MSU’s games based on the sum of the following:
- The (actual) point spread of the game. Positive if MSU wins, negative if they lose.
- The opponent’s final Sagarin rating (current rating for the 2008 calculations). Note that Sagarin ratings are constructed on a point-per-real-world-football-point basis. The point spreads predicted by the ratings generally correspond pretty closely to the Vegas lines.
- An adjustment for whether the game was home or away.
For reference purposes, a game rating of 80 corresponds to playing even with a team ranked between #25-30 nationally on a neutral field. Alternately, beating a team with a Sagarin rating of 70 by 13 points on the road would yield a game rating of roughly 80 (i.e., the win vs. Indiana this season).
There are certainly imperfections in this system: it doesn’t account for players injured for a particular game, final point spreads can be misleading, etc. But I think these ratings are as good an indicator as you can construct based on available data to measure a team’s performance from game to game while accounting for the strength of the opponent and home/road factors.
Below are graphs showing the data for each season, including a polynomial-based trendline (of the second order), along with my brief comments.
- JLS’s first season.
- Went 3-1 in nonconference despite mediocre game ratings.
- 4-0 start in conference play followed by close losses to OSU/UM and then a blowout loss to a not-very-good Wisconsin team.
- Blew out a 3-9 Penn State team and then looked overmatched against Nebrasks in the bowl game.
- Summary: 8-5 record. 7-1 start. 1-4 finish. Close losses to OSU/UM preceded end-of-season decline, although big loss to Wisconsin is main factor in decline.
- Very poor 1-2 nonconference start.
- Only two quality performances: blowout wins against Minnesota and Wisconsin, who finished #27 and #38 in the Sagarin ratings.
- Bad losses to Penn State and Hawaii to end year and eliminate bowl hopes.
- Summary: 5-7 record. 4-3 start. 1-4 finish. 3OT loss to MIchigan preceded late season decline.
- Solid 3-0 nonconference start.
- Very bad loss to a 7-5 NW team following close losses to UM/OSU.
- Summary: 5-6 record. 4-0 start. 1-6 finish. Close losses to UM/OSU preceded late season decline.
- 3-1 nonconference start with mixed results. Only quality conference game rating (80+) was 4-point loss to Penn State to finish season.
- When did JLS get fired during this season?
- Summary: 4-8 record. 3-1 start. 1-7 finish. Team never recovered from blown-lead loss to Notre Dame.
- Dantonio’s first season.
- 4-0 nonconference start; steady but not overwhelming performances.
- Bad conference losses to poor Northwestern and Iowa teams. Generally solid performances in conference games, though–despite close losses again to OSU and UM.
- Summary: 7-6 record. 4-0 start. 3-6 finish. Team finished on slight upswing with decent game ratings in final four games, despite 2-2 record.
- Steady as she goes. Win against Northwestern was first game with rating outside range of 70-90.
- Does the upward trend continue against tougher nonconference opposition?
My take on the data: On the one hand, the first-half/second-half win-loss records are a little misleading. There were some close wins against weak nonconference opponents padding the first-half records and some close losses against good conference opponents bringing down the second-half records.
Under JLS, it was those close losses against good teams that preceded the season-sending declines in performance in the first 3 seasons. 2006 shows a slight uptick at the end of the year, but that was only because MSU was so uniformly bad in conference play following the Notre Dame loss. All-in-all, the data confirm the conventional wisdom on Smith: he couldn’t manage the team’s emotions after disappointing losses against major rivals.
The 2007 and 2008 graphs are certainly encouraging in terms of Dantonio reversing the trend of bad finishes to seasons. We obviously have a very limited data sample, though. Will Dantonio keep the team on an even keel if they lose to OSU or, perish the thought, UM? Or will performance continue to improve over the remainder of the season, leading to a New Year’s Day bowl and the banishing of our late-season ghosts?
Final statistical observation. Below are the standard deviations for game ratings within each of the last six seasons. The data confirm what we all think intuitively: The Dantonio-coached teams have been much more consistent than the JLS-coached teams.
- 2003: 15.3
- 2004: 17.5
- 2005: 18.8
- 2006: 13.9
- 2007: 10.6
- 2008: 8.3
Perhaps some of you with better memories for the details of past football seasons can add your own commentary on the data.