As has already been mention The Don't, it would be based on statisitical analysis of a Sample Space. I'm not sure how big the Sample Space would be though, and that would obviously affect accuracy. Also the diversity of the sample space could affect the result too. If the sample space was almost all one city, or one area, then it would obviously be swayed one way or another. If it was done well, and proportionate to that nations population distribution, it could be a reasonabley accurate guide.
But then there is also questions such as whether a mean or median average should be used. A mean in these cases can be inaccurate, as abnormally large or small can mean that the mean is thrown out. In cases like this, medians often make more sense. But then, if different minority groups of the population are one way or the other, then they can end up totally different than the median. Then you can do things like a mean of the median 20%...hmmm