First off, we have the mode of interview. What I mean here is:was the poll conducted over land-line, cellphone, or internet. Most polling firms use a combination of two of those three modes. PPP, for example, uses land-line (80%) and internet (20%) among those without a land-line. Survey USA implements a cell-phone sample as well. The days of conducting an accurate poll with land-lines alone are very much over. According to the National Center for Health Statistics, 41% of Americans are "cell-phone" only. No poll can simply ignore 41% of the population and expect to get a representative sample of the electorate. Particularly, since this group tends to skew much more to the left than the land-line only group.
Second, we have question wording. By question wording, I also mean the answers that are given as options. A prime example of this Presidential approval ratings. In particular why Rasmussen and Gallup and others always tend to be at odds with each other. Rasmussen gives two degrees of approval and disapproval. Gallup, on the other hand, presents the answers as black and white approve or dissaprove. This makes a difference. A respondent onm the fence would be more likely to say "somewhat approve" than to say "approve" when given the option. It allows a respondant to respond "approval" without being too thrilled about it. This is also why Rassmussen has a lower level of "no opinion" responses on this question than Gallup does. Rasmussen is generally 1-2% and Gallup is anywhere from 5-9% generally. The largest differentiation between these polls is the level of approval and not disapproval. Therefore, one can conclude that there is a relationship between the "no-opinion" respondents for Gallup and the "somewhat approve" for Rasmussen.
Tomorrow, I will discuss demographic sampling and adjustments based of historical data.