I have alluded to this several times, but I'll try to go into more detail...
The polling areas are well-known. Both parties' pollsters know what the typical makeup of any given group of people is. Let's say we have an area made
of blocks, numbered 1-20. They know that
Block 1 will poll 30%D, 20%R
Block 2 will poll 25%D, 20%R
Block 3 will poll 33%D, 14%R
Block 4 will poll 29%D, 27%R
and so on. Notice that all the blocks poll more Democrat than Republican. That's not because there are more de
Democrats than Republicans; it's because these areas have been chosen, most many, many years ago, by Democrats. Of course they were
attracted to Democratic areas. It is expensive to analyze a completely new area so the same areas are used over and over again. That's why most people
never get a single poll call, but others seem to get them regularly.
I should mention that although I used the word "block" to indicate a portion of a larger polling area, these are not necessarily geographic areas. A
block is just a list of phone numbers gathered over time. Maybe someone responded to a political ad 20 years ago... maybe they called the DNC or
RNC... anything can be used to gather the numbers. The point is that these are known numbers, some Republican, more Democrat, and those percentages
OK, now let's say Reuters wants to run a poll. They develop their questions, wording them very carefully to add just a hint more bias without it
sounding like it is biased, and send that list of questions to a polling company. This company is really nothing more than someone well-connected who
has an office with dozens of little cubicles and all of these cubicles connected to a phone. Employees set in these cubicles looking at a computer
screen with a list of questions to be asked, and when they are ready the computer calls the next random number from the block. Their phone rings, they
click a button to connect, and the phone on the other end rings them through. All with a mouse click.
"Good morning/evening Sir/Madam. This is [my name] from XYZ Polling Services. Do you have just a moment to take part in our latest poll?"
They ask the questions, the person answers, and as they do the caller fills in an online form that is connected to that phone number. When enough
questionaires are filled out, the results go back to Reuters. The analysists at Reuters go to work with their magic math... they analyze that data six
ways from Sunday. They then collate their results and do a couple more magic calculations (think witches over a bubbling cauldron dropping in
ingredients if it makes you feel better... the math gets pretty intense and is probably even scarier than the witches for most people) and voila! They
can now tell, based on the known percentages of the various blocks used, exactly what the average American thinks.
And we never hear about it. That's just the baseline. It's not published.
Now the election season is underway and Reuters does another poll. The questions are similar to those used the first time, adjusted slightly to
include current events. But one thing is different: this time, Reuters specifies which blocks are to be polled. XYZ Polling Center does their calling
thing and sends the results back.
Reuters chose which blocks to use based on their original poll (that no one else ever saw). Let's say their results showed that people responded to
Question D in the following way: 28.4% yes, 66.9% no, 4.7% other (don't care/no opinion). But they don't want the public to see that! The pollsters at
Reuters want the public to see the situation is neck-and-neck. So they chose blocks which show that particular question as fairly equally answered.
Now the data is in and it is coallated for publication, then checked to ensure it says about what it was supposed to (otherwise it will be held up),
and finally sent out to all the media outlets. That's what we all see.
Now comes the tricky part with the bubbling mathematical caudron.
The analysts at Reuters are not finished. They toss a few drops of Spittle of Serpent into the cauldron and compare the numbers from their published
poll to those from the baseline poll. Out of the math cauldron pops the actual numbers which would have been the result had they not selectively
chosen which blocks to use. Another round of calculations and they know, based on previous baseline polls and election results, exactly how the people
are leaning. Believe it or not, they can get those numbers to within less than 0.1%!
It's all the magic of math. I'll bet there's a lot of people right now reading this who wish they had paid more attention in math class instead of
trying to catch a glimpse of Betty Lou's panties.
Anyway, that's why the polls are always skewed. Reuters knows that a poll that favors a particular opinion will tend to cause a shift in the undecided
portion of the public. They publish these polls to do just that: sway public opinion. Internally, they have the real numbers hidden in data files
buried within their network. Those who need access have easy access; those who do not need access do not. Everything is encrypted to assure that the
true numbers do not leave the intranet they control.
What has happened lately, though, starting in 2016, is that the baselines are skewed as well. Ever since the rhetoric has become so abrasive and
threatening, people are not answering truthfully. "How likely are you to vote?" is answered "very likely" when in reality the person answering has
little interest in voting... they just don't want people hassling them to convince them to go vote. They will answer "Who would you vote for today?"
based not on who they would vote for today, but based on who they think their neighbors will be voting for... no sense being ostracized over a phone
The truth is that it is possible to find out who answered what question with what answer, but it is also tremendously expensive. The data is tied to
the phone number, and the phone number can be reverse looked up to identify the owner of that phone line. Don't think that just because your number is
unlisted or unpublished that this is not the case... somehow, sometime, someone catalogued that number. It's just not worth Reuter's time or XYZ
Polling Center's time to find you. It's the same principle behind cell phone positioning; the phone company can pinpoint your location from GPS
closely enough that they can tell you what side of the street you standing on... but they rarely do. It's much more expensive. That's only used for
law enforcement, and then only when the phone company is feeling generous.
But the fact that it is possible to tie poll call answers to a specific household is enough to make people nervous, especially when the rhetoric
becomes so intense. So the baseline is skewed, this time by an unknown amount. Right now, I promise you, there are mathematicians across the country
sitting in cubicles in front of massive computers, furiously analyzing data to try and develop an algorithm that will give them better numbers again.
The problem is that rhetoric and the effect it has on the human psyche is notoriously hard to quantify.
Anyway, hope that clears up some of the mystery.