Over at my Go To election campaign coverage website at Electoral-Vote.org there was this yesterday (Jan 13):
Why Have Recent Polls Been So Wrong?
Polls in places as disparate as Israel, the U.K., and Kentucky were way off last year. Pollsters still want to know why. Pew Research Center, which also does polling, has carried out an investigation and come to some conclusions. What it did was look at one of its 2014 polls more closely. In particular, from the polling data, it knows who it selected out as a likely voter and who was labeled an unlikely voter. Then it got the file of people who actually voted and matched it against its own poll to see how well its likely voter screen worked.
It appears that the main source of error in the poll was the likely voter screen. Asking people if they plan to vote turns out not to be a good predictor of whether they will vote or not. In particular, Democrats have a tendency to tell pollsters they will vote and then don’t vote, which leads to “surprising” Republican victories. Also a factor is people who changed their mind after talking to the pollster, but that probably is a smaller factor. Knowing what the problem is doesn’t solve the problem going forward, of course. On the other hand, it is good to know that low response rates and the large number of people who don’t have a landline weren’t the core problems. (V)
The voter screen, the voter screen, the voter screen, yaddah, yaddah, yaddah…
[First of all, I don’t give a rat’s ass about polling for primaries. I am talking about polling for Election Day.]
[We all have some curiosity about voting machine reliability, but we also are not talking about THAT today.]
I will simply address the polling non-reliability based on my last post. In that, the voter turnout for people who were actually citizens AND registered to vote in the last Presidential year, 2012, was 86.81% nationally. States varied from West Virginia’s 70.26% to the 90.26% in Colorado, a swing state. The median was 87.75% in Hawaii.
With 7 out of every 8 REGISTERED voters actually casting a ballot in 2012, I cannot see one reason AT ALL why anyone has a problem with polling and getting good results. You find 1,000 registered voters, sorted only for demographic balance (which no one admits to being their problem) , and of those you can expect 87% to vote come Election Day. And if 87% vote, your poll is going to be damned close. With only 1 out of 8 registered voters NOT voting, how can you possibly get it far wrong?
SO, IT IS ONLY NECESSARY TO GO BY REGISTERED VOTER LISTS. The proper demographics within that population should be a piece of cake – men, women, blacks, Democrats, Republicans, white, Asian, Hispanic – everything but dogs and cats mating in the streets. Using the latest percentages of each to select which voters to poll – that should be about the easiest thing in the world.
Adjustments per state should, of course,be made, too. Basically that means that in West Virginia the uncertainty is higher, but other than that, WTF is the problem?
Now, a full 40 million fewer voters voted in the off-year 2014 election, down from 132 million to 92 million nationwide. That put the natinoal average at 64.89% of REGISTERED voters. So in off-year elections, the uncertainty gets larger.
But I dare any of the polling firms to come up with more reliable indicators of WHO will actually go to vote.
I mean, WHAT THE HELL? We read that this polling firm prefers phone polls and that one prefers online polls. HOW can that POSSIBLY be more reliable than building a database of ACTUAL voters from previous elections, adding new registered voters to that database – and then use THAT to select who to poll?
I tells ya, folks, this ain’t perzackly rockit science, ya know?