They tout the consumer confidence from Michigan or wherever and tell you that people are in a dour mood. They survey CEOs about their outlook ("feelings") for the year and months to come and tell you how to interpret it to invest money in bonds, stocks, etc. All these surveys are sold as financial news.
These surveys are meaningless. They don't tell you anything.
Humans talk and act. Their actions many times have not much to do with their talk--they will keep saying they wont smoke or eat more or buy a Gucci bag but will keep doing those activities anyway. It is a fundamental problem with all surveys (including pseudo subjects like psychology, sociology, etc. which are heavily survey dependent)-a human animal just because of it's ability to talk is assumed to know what it wants to do-and this is a wrong assumption. Look at the number of smokers and dieters around you who are addicted to cigarettes and are fat to convince yourself that what we say is very often not followed through. Actions speak louder than words, someone has said---and that is the right view from a biologist/scientist-animals which don't speak also show their preferences by their actions. This is also true for savage people with rudimentary language skills living in the Amazonian jungles, the aborigines in Australia, NZ and many parts of Africa and Asia, and numerous other "uncivilized" human beings which still live today. They may not talk much---but they are still humans, at least in the present classification of species. And you trust their actions for finding out about their preferences, not what they say. This also holds for kids-who talk a lot of things but you can tell what they really want much better by watching what they do.
Polls for who wins in Hillary vs. Trump are everywhere. Most are leaning toward Hillary. What they are not considering is that many people might be lying in the polls-just because they say they will not vote for Trump today doesn't mean that when election time in November, they will follow through on this. They will in the end vote for him for other reasons, or justifications. They may not like associating with him today on a poll service, but where it really counts, when the actual vote is done, they will vote for him.
Note that what I am saying here is different from the standard deviation of surveys and sampling problems. When probabilities are tight, you need a lot of samples to figure out what's going on in the underlying distribution. But here I am talking about why even if the samples were representative of the underlying distribution, they are still imperfect, because people will say one thing on the survey, e.g. I will vote for Clinton today; but when the voting ballot is actually cast, they will vote for Trump. Everyone knows that that's the one which counts.