Desmond Howard Ripped FiveThirtyEight Over Michigan One Percent Playoff Projection
By Ty Duffy
FiveThirtyEight is assigning Michigan a 1 percent chance of reaching the College Football Playoff. Desmond Howard was asked about this during a Sportscenter hit. He pointed out that FiveThirtyEight was wrong about the election.
We’d defend Silver and Co. on that charge. Their model was much less bullish on Clinton than others and gave a substantial probability to Clinton winning the popular vote but losing the election. FiveThirtyEight did not conduct the polls. The election result was within the boundaries of a standard polling error.
That said, FiveThirtyEight’s “model” setting Michigan’s probability odds of reaching the playoff at one percent seems ludicrous. Here is the justification for it.
"Michigan needs chaos. After losing The Game to Ohio State, the Wolverines’ CFP odds plummeted from 37 percent to 1 percent. And what’s that 1 percent situation? Pure chaos. Michigan needs all of the teams ahead of them in the committee rankings to lose this weekend,2 and they probably also need Oklahoma State to upset Oklahoma. Even all that probably won’t be enough — our model would still only set their CFP odds somewhere north of 10 percent — but one thing the model doesn’t explicitly know is that Michigan has the head-to-head edge over both Big Ten title game participants. Since the committee has shown that it tends to re-sort from scratch in its final rankings, that one factor might be enough for Michigan to build a case from, in the event of upsets at the top."
The model is based on the past behavior of pollsters and the committee. Pollsters weren’t setting a four-team playoff. We have a sample size of two college football playoff committees. The most consistent playoff committee behavior has been adjusting its justifications ad hoc to meet the decision. We’d question what value a mathematical probability model provides for projecting the decisions of 10 individuals without polling them.
Moreover, the model apparently does not take into account head-to-head results. So, the model doesn’t “explicitly” know Michigan beat Wisconsin, beat Penn State by 39, beat Colorado, who may win the Pac 12 title, by 17, and lost to Ohio State in overtime, in the Horseshoe, on a contentious 4th down spot.
We’ve seen two broad committee decision-making patterns. They tend to prefer conference champions. They also tend to prefer the strength of schedule, as measured by wins over teams high up the rankings. The former rationale would favor Wisconsin or Penn State. The latter would favor Michigan. If either Clemson or Washington loses, those rationales will collide.
If this model has a face, assigning Michigan a probability of “one percent” seems destined to blow up in it.