I agree very much with the sentiment that polls are usually great at categorizing but not so much as analyzing. If you know anything about sampling errors you will realize that often the categories slant the results exactly because the questions asked force the respondents into predetermined categories and do not allow for those who might not fit any category. Here's an example for the US: Are you A) Republican or B) Democrat? (I've never actually seen one that made this egregious an error, but it could be made). Any poll that did something that clearly wrong is a bad poll, but sadly, would still be reported if is supported this or that point of view. In other words, polls not only have sampling errors, but the sampling errors are ignored whenever the politics warrant ignoring them. Politics is a lot like capitalism in that the morality is by driven by the bottom line in capitalism and the number of votes in politics.
The extreme, and seldom done, poll, is the interview poll. In such a poll people are interviewed to get their opinion. It takes a lot of effort and cost to do one of these and sometimes things are discovered that could not be known otherwise. However, like the predetermined categories of the standard poll, you have the interpretation by the interviewer at the interview -- who may provide leading questions or miss important signals -- and/or the predetermined classifications or analytical categories of the analyst. In other words, if a human is involved you may not actually be getting a report actually reflecting what the interviewed person actually thinks.
In the world of business the story goes that in 1978 a particular computer company interviewed it's customers and determined that if they came out with a personal computer it would sell only 10,000 a year at most. So they decided to make one but to release the design features freely to the market and allow anyone to copy the general design freely. They released their product and received 100,000 orders in the first 90 days. Why did they miss the boat by so much? Because they interviewed their customers -- whom they understood as being representative of the general population -- and their customers were large, multinational corporations who already had big fat computer systems…so this "personal computer" was just a toy. Had they actually interviewed more than the heads of the IT department they would have found the market much greater than they anticipated and probably put out a more proprietary design (and made a lot more money doing so). The point is, a poll is only as good as it's design and execution. Unfortunately, when poll results are reported we are seldom told of any possible design flaws, especially if the poll supports the editorial position of the reporting agency.
So what are we to believe what we "read in the newspaper?" No,…and yes. No, not what we read in THIS paper or THAT paper, but probably what we read in papers holding differing political and social positions. In other words, if you get all your news from one source you may not be getting the balanced and whole picture. Read widely, listen widely, and speak only when you find enough credible sources that you can be confident what you are saying isn't too slanted or just plain fake.
AJ