We don’t run corporate ads. We don’t shake our readers down for money every month or every quarter like some other sites out there. We provide our site for free to all, but the bandwidth we pay to do so doesn’t come cheap. A generous donor is matching all donations of $100 or more! So please donate now to double your punch!
Last Wednesday, Nate Silver added to his lore this primary election cycle by publishing a “rose colored glasses” path to victory for Bernie Sanders. It required Sanders to start with a double digit victory in Wisconsin at plus sixteen.
There was only one problem.
The same horrid polls that keep telling us Sanders will lose everywhere suggested Sanders was substantially behind in Wisconsin. At the time, Silver’s FiveThirtyEight gave Clinton a wonkish eighty-four percent chance of winning Wisconsin. Days earlier, FiveThirtyEight contributor Julia Azari commented on those chances to Newsweek. Azari insisted that while Wisconsin might look like “Bernie territory,” in actual fact “things that are rare don’t happen that often.”
The results are in: Bernie Sanders won Wisconsin yesterday by 13.3 points, shaving ten to twelve delegates from Clinton’s lead.
By yesterday morning, FiveThirtyEight had taken account of new polling, adjusting its chances to a 72% chance of victory for Sanders. Still, its Polls Plus model forecast him falling woefully short of the Rose Colored Glasses Path at just a 3.0% margin of victory. By only being wrong by 10.3% this time, FiveThirtyEight may have marginally brought down its overall 12.2% Clinton House Effect outside the South. Bravo!
My projection was Sanders +10.6%, giving me a slight 2.7% Clinton bias. I predicted 100% of Wisconsin’s counties correctly.
The problem is this: early polling in the Democratic primary cycle is astoundingly atrocious. It tends to become somewhat less atrocious as election day nears, but still tends to make errors that should be obvious to the half dozen or so paid staff members at FiveThirtyEight who are literally getting paid to analyze the strength of polling and squeeze more accurate results from them in total then any individual poll could alone.
This phenomenon is well on display already in Pennsylvania. The Rose Colored Glasses post refers to some of the most chuckle worthy polls this season to note that Sanders is “down by twenty-something points” in Pennsylvania. Since Sanders needs a +7% victory: #FeelTheMath. I will not belabor the point again, but a poll with 18% eighteen to forty-four-year-olds has no business anywhere near FiveThirtyEight.
Here’s a little chart I’ve been working on recently. I’ve highlighted the highs and lows based on exit polling so far. The lowest percentage of 18-44 year-olds was in Oklahoma at 33%, almost double the Franklin and Marshall sample size; the highest so far has been Michigan’s share of 18-44 year-olds at 45%, exactly two and a half times the Franklin and Marshall numbers.
*Note: the numbers in parenthesis to the right of figures in the 18-29yo column are the 18-24yo share of the vote according to exit polling.
Early this morning, however, a more rational poll from Quinnipiac is out. It pegs Clinton up just six, with another six percent undecided. But what about Quinnipiac’s age demographics? They’re still off, even if much closer. Quinnipiac’s numbers require that the 18-44yo share of the vote in Pennsylvania will be just 29%. That’s 4% lower than Oklahoma and 6% lower than Florida. It is not entirely impossible that Pennsylvania will have the lowest share of young people and the highest percentage of retirees voting of any state to date. It’s just not all that likely. Pennsylvania does have the second highest share of people 65 and older residing within its borders (nineteen percent for Florida; seventeen Pennsylvania). And, Pennsylvania is stacked up with four other difficult, closed primary states on the same day. Realistically, if we adjust the age demographic in Quinnipiac’s Pennsylvania poll to reflect Florida numbers or, more likely, Ohio numbers, the race is a dead heat. Clinton would be +1% with Florida’s age demographics for voters according to exit polling. Sanders would be +1% if the Ohio numbers are more accurate.
I’ll have a new post up, hopefully later today, with a different route to victory in pledged delegates than FiveThirtyEight’s. It will be based on more realistic attention to demographic contours and, even more importantly, factors like where particular states lay on the election calendar.
Three weeks out, a dead heat in Pennsylvania. The Rose Colored Glasses scenario is looking less and less Rose Colored by the moment.