from Bill Still
I’ve been reporting on the USC Daybreak poll, which still has Clinton ahead of Trump by seven-tenths of one percent.
Then someone emailed me a website called LongRoom. It compares the various statistical skews or biases of all the polls and removes them.
As we saw in Still Report #1096 – How Clinton Rigs the Polls, at Nate Silver’s “538” office, pollsters congregate in their break room and talk about “skews” and biases.
Applying skews to data can be completely legitimate. For example, let’s say you get opinions from 200 people in city A and only 100 people from city B.
If you know the historic voting patterns, then you know what weight to give one city over the other. This is a statistical skew.
They are necessary because if historically, twice as many people tend to vote for Democrats in City A then you know you have to apply that skew to that city’s data to get a fair view – make them equally weighted.
The alternative would be to try to get an exact number of people from each city to answer your questions based on their previous biases – an impossible task.
But the downside is that polls are products that someone buys and pays for and they expect certain results. So, the easiest way to provide what the customer wants is to manipulate the skews.
Longroom understands this and gives you data that unbiases the skews, so you get an honest look at the polling data coming in from all the pollsters.
[Insert]
For example, here you see that the average media polling bias today is 4.2% favoring the Democrats. Then it lists the biases of 5 of the top polls.
You’ll notice that I was on the right track. The USC Daybreak poll is the least biased with only 1.4% bias towards the Dems. CNN, however has an outrageous 7.1% bias.
If we look at today’s Daybreak poll, Clinton leads with 44.7% and Trump has 44%. That means that Trump’s real number – when you add in 1.4% of his initial 44% should be 44.60% and Clinton’s number should go down to 44.07. That puts Trump in the lead by over one-half point, instead of down by .7 percent.
So while this does show the relative direction of the votes, we all know that the “Monster” vote is going to be a huge factor and it is entirely outside of this statistical system.
No matter how accurately these polls are cleansed of the skews or biases, the Monster vote is coming from people who just not been counted in the past. They either haven’t voted in the past, or would have been counted as either independents or Democrats, but now will be crossing over en masse.
Bottom line – these polls can say whatever they want, but no one knows how big the monster vote will be – 10%, 20% or more, it is literally the 600-pound gorilla in this room this year.
I’m still reporting from Washington. Good day.