Posted by
Andrews on Wednesday, October 08, 2008 4:25:56 PM
We see every day new polls showing that the voters have shifted one point for Obama or two for McCain. Every day it seems the public is slightly adjusting in one direction or another. Neither really seems to gain much, as they are always withing the margin of error of a tie, but the numbers continue to bounce back and forth.
I have one question for my readers. Does anyone seriously believe there are a significant number of uncommitted voters still? Are there really many people out there who have no opinion about the election and could vote for either? Or is it more likely that "shifts" in the polls are simply polling artifacts?
My thought is that most voters are pretty settled in their opinions, and have been for some time. A major revelation, or a bad performance in a debate, may shift some, but, barring a big revelation, a major misstep, or a significant event in the world, I just can't believe there are many undecided voters out there.
Given that, why do the polls shift so often? The answer is simple. When you poll 1000 people, each one of those people is 0.1%. That means simple variation in the group selected could make a large difference. Let us suppose a city has a 20% black population. You pick 100 people to determine the number of black citizens. It is likely you could get anywhere between 15 and 25 black people in a random sample, simply by chance. Which means black population could fluctuate from the true measure by as much as +/-5%.
An even more basic example. Take a group of 100 people at random. How often do you end up with precisely 50 men and 50 women? Most often you get somewhere between 40 and 60 of either sex. So, judging by those results, the percentage of men in the population can vary +/-10% from the true value. Are the proportion of sexes changing between these samplings? Or is it an artifact generated by using a small sample?
In statistics, the solution to this is taking multiple samples. You may even take a smaller group but by taking a larger number of smaller groups, you have greater confidence in the results. But political polls do not work that way. They tend to take one poll of 1000, or maybe even 2000 people to represent tends of millions of voters. And then poll again a day later. This leads to polls showing 1 or 2 point "changes" which are likely more artifacts of the polling methodology than any real change in the poll.
So, next time you hear that Obama or McCain gained or lost 1 or 2 points in a poll, recall that means that there were 10 more supporters in this group of 1000 than in the last. Obviously chance can explain that.
Polls can be useful, and by watching for trends over time using multiple polls, they can sometimes indicate an overall trend. But the minute attention paid to one or two point shifts is quite misplaced.
And, one other thing to recall. Just because a polling subject says they are a registered voter, a likely voter, or anything else, it does not mean they will actually vote. Nor are answers given binding contracts. Even with the best designed polls, and the best selection of subjects, polls can still turn out to be completely wrong. So take any poll results with a huge grain of salt.
POSTSCRIPT
Before anyone from the left accuses me of writing this "just because McCain is losing", please check my previous articles. I have cautioned against relying on polls both when McCain was ahead and when he was behind. I have been and continue to be skeptical about polling. The number of times polls have failed to predict elections allows me to hold no other belief.
On the other hand, I have mentioned polls in passing, and even used them as a general indicator of mood. Even though they are far from reliable, and I think they are often subject to misleading results due to poorly constructed questions or poorly selected subjects, they are the only tool available, and so I am forced to use them.