Smallish sample size aside, cold calling is a terrible way to conduct political polling. I worked in a call center and was a refusal converter, calling people who already told us to go away and getting them to complete the study anyway. The study I spent the most time with was the CDC's Behavioral Risk Factor Surveillance Study, which I've administered I think four versions of? I also worked on the Sonya Slifka Longitudinal Multiple Sclerosis Study, a tobacco use study for the University of Colorado, and a number of other studies for smaller periods of time on everything from politics to experiences of abuse.

The people who will actually talk to a cold caller tend to fall into one of three categories, by my estimation. They're either lonely, particularly cooperative, or particularly opinionated. These aren't such big confounding factors for a health study, but they're absolutely massive when it comes to politics.

Reporters don't recognize this at all. They see numbers and cited source and just run with it. As someone who helped collect those numbers, I would not be taking them at face value. Social biases may not have much of an effect on how many carrots you eat in a month, but on your vote? I would struggle to come up with a better way to make your sample less representative.