Polling is an exact science. The problem is not with polling -- as the article makes clear. The problem is with connecting two different populations: the people you reach in a poll, and the people who vote on election day.
If you poll 1000 adults truly at random, their opinion on a yes no question will have a 95% confidence interval of around +/- 3% for the entire population of 250 million adults.
That is exact a statistical consequence of the Central Limit Theorem and there is no aberration from it; if there were, so many aspects of the world -- including the basic physics of the universe -- would fall apart.
Next, if you randomly examine the ballots of 1000 people who have just voted, you will be able to predict with 95% confidence what the percentages of votes are in the entire voting population to within about +/-3%. Again, there is really no arguing with this.
When FReepers say, "How can these polls be right, they involve so few people?" This is nonsense, pure and simple. It is as dumb as thinking that you have to flip a coin ten million times before you will get close to 50% heads. There is a 95% chance you will get 50% heads +/- just 0.01% by flipping the coin only 10,000 times. It is not the polls per se that break down. What breaks down is the predictive model that attempts to say how the actual voting population on election day compares with the population answering the poll right now. That is where the error comes in. Not in the math, but in the various theories that attempt to relate two different populations -- which are never the same -- to each other.
Thank you.
I thought I was the only one to to take a statistics class in university
Another area where error could creep in is if the sample is not truly random. This does not have to be insidious but may may be respondents being swayed by body language, vocal tone or the connotative meanings of the words used in the questions. At times the researcher is not even aware of this bias.
Then, of course there can be simply a push poll designed to influence an outcome rather than predict it.
Your argument assumes pollsters are honest people.
There is a Reason 94% of the population does not trust media of which the polls are part of.
I think you are confusing the math with the polling itself.. Polling is far from an “exact science”.
Statistics is only a tool of polling methodology. That is the easy part that every schoolboy leans in a beginner course in statistics and probability. You don’t “poll” a jar of multi-colored jellybeans. Jelly beans aren’t undecided if they are blue or red. Jelly beans don’t say they lean blue when the lean red. The inexact nature of “polling” is how many of each color to place in the jar before you select a small sample and apply the simple math, which , as you say, is exact but only to the extent that it provides an exact margin of error and confidence level.
IT is much like the computer model Global Warming science. Garbage in, garbage out. If you are allowed to juice your model’s input, you can make any claim you wish with a very high degree of exactness.
Polling is not an exact science and never can in a predictive sense. Too many uncontrollable attributes that help to hide internal bias.
That is why they apply empirical averaging, to make assumptions from their sample and blend them with theoretical infinity.
Having said that the science is exactly enough, when they truly are disciplined in applying it, for polling by competent people.
Damn you, made me think this morning...I just got up...
Polling is not an exact science and never can in a predictive sense. Too many uncontrollable attributes that help to hide internal bias.
That is why they apply empirical averaging, to make assumptions from their sample and blend them with theoretical infinity.
Having said that the science is exactly enough, when they truly are disciplined in applying it, for polling by competent people.
Damn you, made me think this morning...I just got up...