According to our final polls-plus forecast, Hillary Clinton had a greater than 99% chance of winning the Michigan primary. ---FiveThirtyEight, 3:43 PM EST, March 8, 2016.
Oops! Or rather, a massive OOPS!!! The FiveThirtyEight blog is considered to be the gold (or rather silver since its founder is Nate Silver) standard in election predictions. So when it is so wildly off the mark as happened last night as Bernie Sanders pulled off an astounding primary victory upset in Michigan over Hillary Clinton, it is definitely quite notable. So how to excuse this error? Actually, credit must go to 538 analyst Harry Enten who did not offer any excuses which would only be laughed at. Instead he publicly ate humble pie:
Bernie Sanders made folks like me eat a stack of humble pie on Tuesday night. He won the Michigan primary over Hillary Clinton, 50 percent to 48 percent, when not a single poll taken over the last month had Clinton leading by less than 5 percentage points. In fact, many had her lead at 20 percentage points or higher. Sanders’s win in Michigan was one of the greatest upsets in modern political history.
Both the FiveThirtyEight polls-plus and polls-only forecast gave Clinton a greater than 99 percent chance of winning. That’s because polling averages for primaries, while inexact, are usually not 25 percentage points off. Indeed, my colleague Nate Silver went back and found that only one primary, the 1984 Democratic primary in New Hampshire, was even on the same scale as this upset. In that contest, the polling average had Walter Mondale beating Gary Hart by 17 percentage points, but it was Hart who won, with slightly more than 9 percentage points over Mondale.
Now that doubt has set in about the accuracy of the polls upon which 538 relies, Enten is wondering if perhaps Sanders now has a chance in the rest of the Midwest:
The question I am asking myself now is whether this means the polls are off in other Midwestern states that are holding open primaries. I’m talking specifically about Illinois and Ohio, both of which vote next Tuesday. The FiveThirtyEight polling average in Illinois gives Clinton a 37 percentage point lead, while the average in Ohio gives her a 20 percentage point lead. If Michigan was just a fluke (which is possible), then tonight will be forgotten soon enough. If, however, pollsters are missing something more fundamental about the electorate, then the Ohio and Illinois primaries could be a lot closer than expected.
Either way, this result will send a shock wave through the press. Heck, I’m a member of the press, and you might be able to tell how surprised I am. This will likely lead to increased news coverage of the Democratic race, which Sanders desperately needs in order to be competitive next Tuesday and beyond.
Another one who was shocked and surprised was Nate Silver himself who as early as 9:25 PM last night weighed in on the historical nature of this astounding upset:
I said earlier today that I had an intuition Sanders could beat his polling in Michigan tonight, but I didn’t expect things to be quite so close. If Sanders winds up winning in Michigan, in fact, it will count as among the greatest polling errors in primary history. Clinton led by 21.3 percentage points in our final Michigan polling average.
And just how great the polling errors were can be seen in a conglomeration of Michigan polls listed at 538 in which none showed Sanders even coming close to winning. In fact, the March 6 Mitchell Research & Communications poll showed Clinton beating Sanders by a whopping 37 points. Of course, even with prognosticators such as 538 noted for its accuracy up until now such as correctly predicting the outcomes in 31 out of 33 Senate races in 2012, predictions can only be as accurate as the polling data upon which it relies.
GIGO. Garbage In, Garbage Out.
