Calling people on the "phone" insures only that you get a skewed sample, for a number of reasons.
Donald Trump says not to listen to the polls. And like a stopped clock, he is right about some things, at least twice a day. Polling numbers are hard to quantify these days. One crazy poll that Trump loves to cite, uses the same base of respondents over and over again. They call this a "tracking poll" but it is really tracking very little. Since the base of a few thousand voters includes one black man who is in favor of Trump, it incorrectly suggests that Trump has a huge black following.
The problem with a "tracking" poll also is that if you keep asking the same people their opinion, chances are, they will give you the same opinions. So if your sample is mostly pro-Trump people, it will continue to show a bias toward Trump (or toward Hillary, depending on the initial sample). It only tracks "undecided" voters, who let's face is, are utter morons. When I hear someone say they are "not sure" who to vote for, or that "both candidates are alike" I wonder if they flunked that Kindergarten test of "one of these things is not like the other" (no shame, I flunked it myself).
But telephone polling is another form that is still used today and is utterly obsolete. Bloomberg just released a poll showing Trump with a 2-point lead in Florida. This is not surprising, given that the only people who have landlines anymore or answer the phone are old people. Young people don't have landlines and screen their calls on their smart phones - answering only those calls from people they know, letting the rest bounce to voicemail.
Young people assume - correctly - that most unknown callers are likely scam artists these days. And if anyone calls asking you to complete a "survey" chances are it is also a scam. So young people don't answer surveys as much as old people do.
Also, when doing a telephone survey, who do you call? (and no, it ain't Ghostbusters). The Bloomberg poll apparently uses people in Florida area codes, as evidenced by this first question in the survey:
"First, I just need to ask if you currently live in the state of Florida. (If needed:) People sometimes take their phone number when they move to another state, so we just need to determine what state I’m talking to."
In other words, they are aware that just because you have a 305 area code, it doesn't mean you live in Miami, as you may have moved to Texas and taken your number with you. But they don't consider the opposite effect - that a lot of people who move to Florida, particularly in the last decade, kept their old cell numbers from "up North" even as they live in Florida. A couple I know who live in The Villages still have their 315 area code phones, which is a far cry from 305. One is Syracuse, the other is Miami. They will never be called by Bloomberg unless Bloomberg is polling Central New Yorkers.
So you see the flaw in the Bloomberg poll - it only captures people who have been living in Florida a long time, and moreover, it might never capture newer residents, who have no incentive to change their phone number just because they moved. As I have noted in other postings, area codes mean nothing today, and I am a prime example. Not only are my cell phones in area code 703 (Northern Virginia) but my "landline" (NetTalk VoIP) as well.
Alas, in this modern era, where phone calls are basically dead, the pollster has a difficult problem to solve - how do you contact people and get their opinions? Or more succinctly, how do you get a representative sampling of opinion? Because quite frankly, if Bloomberg called me with its poll, I would probably hang up. They ask far too many questions and I have shit to do. As a result, every poll as a built-in filter in that it samples only people who want to answer pollsters' questions. The longer the poll, the smaller this demographic group is.
What it gets down to is that the only poll that counts is the one on November 8th. Or the 28th, if you are voting for Trump.