So a few months ago Chris Stassen, maintainer of the invaluable Stassen.com, emailed me to note that his name was not, in fact, "Harold." I had been calling him Harold for years.
So I don't know if this is revenge or whatever, but yesterday this hit my inbox from Mr. Stassen. It's possibly the most frightening thing I've ever seen. The elderly, children, and pregnant women are advised to stop reading now.
I warned you…
Last year I stumbled across a research paper on the odds of a Vegas favorite winning a game (e.g., a 1-point favorite wins 53% of the time, a 7-point favorite wins 69% of the time). The research was based on Vegas lines and NFL games, so one can argue against applying it to the college game. I think college teams perform less consistently than pro teams (meaning that a big NCAA favorite is less guaranteed to win than an equivalently big NFL favorite)... but probably it gives a decent rough measure.
The other problem I've had is how to get "lines" for games which are purely only hypothetical (e.g., Texas Tech vs Kansas in the Big XII title game), or games which are weeks away and not big enough for Vegas to publish an early line (e.g., Michigan at Purdue). I've finessed this issue by using Sagarin's PREDICTOR ratings (his best estimate of team strength) to compute spreads. Again, this sort of abuses the research paper's results, in that his spreads aren't the same as Vegas lines (though they tend to be pretty similar). Anyway, if we're only looking for ballpark figures, we don't have a lot of other options if we want to do this sort of computation.
There are lots of various ways to throw rocks at this analysis. I did a similar one for Ohio State fans at the start of the season, and computed that Ohio State was twice as likely to lose 3+ games, as they were to go undefeated. That was not too popular with the OSU fanbase, but I think I will be somewhat vindicated by the end of the season.
Anyway, even though there are weaknesses, I think the analysis' results are useful for looking at how a season "should" play out. More important than the spreads and exact per-game probabilities going in, are the ways in which the probabilities of individual games combine into the probability for this or that overall record.
Without further ado, here's the result for the remainder of Michigan's season:
By Sagarin-predictor pointspreads, Michigan is an underdog of more than a touchdown in every remaining game:
+8 at Purdue,
+9 against Northwestern,
+15 against Michigan State,
+16 at Minnesota,
+21 at Ohio State, and
+34 at Penn State
Using these numbers, the composite odds for Michigan's final six games of the regular season are:
6 wins = 0.0% (8-4)
5 wins = 0.0% (7-5)
4 wins = 0.4% (6-6)
3 wins = 3.9% (5-7)
2 wins = 18.7% (4-8)
1 wins = 41.8% (3-9)
0 wins = 35.2% (2-10)
By that calculation, Michigan is more than 75% likely to finish either 3-9 (one win in their remaining games) or 2-10 (no wins), with 3-9 being a bit more likely than 2-10.
They're a 22-to-1 longshot to exceed 4-8, and about a 250-to-1 longshot to be marginally bowl eligible at 6-6.
Individual results [excised for space and horror considerations; it's just the numbers in detail]
The most likely single outcomes are:
(1) Lose all remaining games (35%)
(2) Beat only Purdue (14%)
(3) Beat only Northwestern (12%)
(4) Beat only Michigan State (6%)
(5) Beat only Minnesota (6%)
(6) Beat Purdue and Northwestern (5%)
Anyone still alive after all that? If so: computer ranking are pretty crappy measures of team strength even at the end of the year, and they can be wildly inaccurate with only six games of data. The thing that jumps out at me is the Penn State spread, which is a full ten points higher than the Vegas line. FWIW, the Vegas line is always a more accurate predictor than computer rankings. This is considerably more grim than the facts on the ground. Probably.
But… yeah. I'm building a bomb shelter.