He's very popular among the age group that isn't registered, or that (due to an unwillingness to commit to the Democratic party) cannot vote in closed primaries, or who just don't bother to show up at all.

It included all non-R, adult respondents. Of course, that's the one poll that showed Sanders ahead.

https://www.democraticunderground.com/?com=view_post&forum=1287&pid=415246

That being said, almost all of the other polls include either registered or likely voters.

They said they registered to vote. Most polls right now are of registered voters. If that person was polled, they will be counted.

it is polling both likely and unlikely voters. Usually, when an election is farther away, such as six months or more away, polling groups are more likely to poll registered voters and less likely to poll likely voters. When an election gets close, likely voters tend to reflect actual voting results more closely than registered voter polling does. However, there is a catch. Likely voter polling depends upon a model to transform registered voter polling into likely voter polling, and the more wrong the model is as compared to what will actually happen on election day, the farther off the results will be.

Let me give you an example. In 2016, a lower percentage of younger voters and Black voters voted than in 2018. If someone built a likely voter model based out of the 2016 model, they would be assuming fewer Black and young voters would show up, so they would give the polling from Black and younger voters less weight. If, on the other hand, they assumed that 2020 voting would be more like 2018 voting, then they would give Black and younger voters more weight.

I wish that all polls would show their original registered voter data, the model they used for conversion, and the likely voter data. Then people could draw their own conclusions.

I will never be polled for example. I don't answer calls unless I know the name. If something is important, they can leave a message.

I also don't use fb, so any polling done there wouldn't reach me. Same for opt-in polling; I don't participate.

This site is about the only place you could poll me, and that is pretty meaningless because of the nature of this forum. It's like polling a maga rally.

I suppose someone could stop me on the street, or knock on my door, but that hasn't happened.

Look at the stats for a poll; the "tab" data. It's amazing how many calls they have to make to reach a single person they are targeting. Several thousand in some cases. That's super crazy.

Usually, the shift from a registered voter to a likely voter estimate is that people weights are used to reflect likelihood of actually voting. That is done via a model that may include responses to questions on how enthusiastic they are or their response to how likely they are to vote and/or information on whether they voted in past elections.

Obviously, they will miss anyone not registered at the point of their poll - and depending on how they create their frame to sample from, they could miss recently registered voters. Consider that some states allow registering up to election day.

When poll results use statistical confidence intervals, it is good to remember that the error terms reflect sampling variation only. It will not pick up any systemic bias due to the frame not representing the true population. In addition, the sampling results are then adjusted to reflect the likely to vote model. That model itself is unlikely to be perfect and it can be very wrong.

When the NYT came out with a set of polls that ALL showed a shift from the Democrats to Trump when compared to the other polls from other places, they also had a podcast on polling. They talked about how their 2016 polls underweighted the non college aged white voter segment. They argued that they needed to correct their model. What they are speaking about there is changing there likely to vote model. However, it is entirely possible that 2020 will be vastly different than 2016 - as 2018 suggests. In 2016, the after the fact "common wisdom" is that those non college educated whites in the rust belt who voted for Trump were more likely to vote than anyone had modeled. However, is it possible that the people who are angry and enthusiastic about voting the President out are now the young and the minority voters who were not that engaged in 2016? (PS that might be that they were relatively unworried about the election - not that HRC was not someone they were willing to have as President.)

However, it is not just the Bernies or Obamas who get new people out. In 2004, Kerry got 9 million more votes than Gore had and exceeded all the campaign's benchmarks in states like Ohio. However, Bush got 11 million more votes - partly because of 911, but also because Cheney and his daughter succeeded in their strategy in pulling out hordes of evangelicals by putting anti LGBT resolutions on ballots in key states.

In summary, polls, all contain embedded assumptions concerning who will vote. The nation is so polarized that I would bet that BOTH sides will have far higher percents of people voting than happened in 2016. I don't know a single Democrat, who would say - as I heard some say in 2004 that it could even help Democrats to have Bush win and have Iraq fail on his watch or as in 2008, say that McCain would not be a bad President and might be able to heal the country. These comments drove me crazy because Bush winning made a huge difference - 2 SCJ and Citizens United - by Bush and McCain would have picked 3 conservative SCJ (McConnell would not have blocked him). I have to say I did not hear any Democrat I knew suggest that Trump was ok.