I agree. People have gotten into this mind set—and I have, too—that we put faith in the polls, when the press really should reveal in indepth stories post- AND pre-election the accuracy of their predictive power, or lack thereof. Many of these pollsters should be sued for false advertising.
The reason polls appear to be reliable is that in the vast majority of cases they are highly accurate. It's sort of like quantum mechanics - science is based in part on observation, yet at the quantum level observation effects the results. Likewise, polling is statistically valid, yet if we introduce the concept of "moral hazard", all bets are off. Here's a comment I posted the other day that expands on this theme.
olling is statistically sound; where it can be corrupted is if a choice has a high (perceived/actual) sanction value. If there is no or limited sanction value, as in the 2006 mid-term elections or various state & local choices, then polling can be highly accurate.
But, introduce the concept of (personal) moral hazard, and results can be widely skewed. For example, how many attendees of the Colosseum waited until the emperor gave his thumb up/down before roaring their approval? Likewise, how would you answer if asked during the period Galileo was imprisoned whether you believed the earth revolved around the sun?
In our current era, there are certain moral imperatives that are expected to be widely embraced, including gay rights & affirmative action. The inherent danger in opposing this "cultural norm" can range from implied rebuke to ostracism to outright violence. That's why polling results for both Prop 8 out here in Calif (opposition to gay marriage) and Obama are horribly corrupted.
And it's not just lying - it's decline-to-answers & hang-ups. If a pollster calls 1,000 people and only 900 were willing to answer, what can we surmise about the 100 (10% - actual estimates are 80%) who declined? If they are split 75/25 for McC, then that's a net of 5 pts.
Right now, the shorthand for this effect is called the "Bradley (or Wilder) Effect", but it is really much greater than that - these candidates are merely symptoms of the phenomenon. Mark my words, many doctoral theses are going to be written analyzing/explaining situations where polling results can be invalidated when the environment for personal expression of opinion has an associated high risk factor.