Reply #12: Exactly

[Stuckinthebush]

However, inferential methods have a major assumption, that the sample is a true random representation of the population. When you poll, the only feasible way to get results is to use telephone numbers. Once connected, there are ways to randomize who you talk to, however, the initial method of getting the sample is seriously flawed. This is well known in the industry, but there is little that can be done to solve it. In inferential statistics, if one of your assumptions is violated, there are consequences to interpretation of results.

Technically, polling methods are not inferential but descriptive, however the underlying sampling theory applies.