I didn’t call Trump a fascist because he’s a Republican. I called him a fascist because he’s a strong leader figure at the head of a political movement based on recapturing past glory through demonization of the other in society. There’s a word for political movements of that sort, and that word is “fascist”.
I also did not say that fascism is a bad thing. If you believe that it is, as you evidently do based on taking umbrage at my use of the label, then you should not have voted for him.
But maybe Trump is already overperforming his poling.
Suppose 47% of Americans say they’ll vote for Romney, and 47% indeed do.
Then suppose 44% of Americans plan to vote for Trump, but only 39% are willing to admit to it. Then Trump could still underperform vs. Romney, but overperform vs. his polling.
I’d made a post last week in a MPSIMS thread on the topic, which I’ll repost here.
(Professional market researcher here)
Exactly so. All of the major political polling companies are market research companies; they do all sorts of other survey research in addition to political polls. They use political polling as a way to build awareness of their brand names, and demonstrate that they’re good at what they do.
Gallup has been the “name brand” in political polling for decades, but, they decided to sit the 2016 elections out entirely. Why? Because all of the conventional wisdom on how to conduct political polling (and how to interpret it) has gone out the window. And, no, it’s not just a function of Donald Trump.
In the past, political polling was a pretty exact science, with some well-established norms on how you did it, and how to model it. It was exclusively conducted via telephone interviews, and home (land line) telephones, at that.
In a poll (or in any market research), it’s a known fact that you generally don’t need a very large sample of respondents, to get fairly accurate results – but there is a very important assumption in that statement. The assumption is that you (the researcher / pollster) are able to construct a sample which is representative of the entire population – that is, you’ll have the right proportions by demographics (age, sex, income, etc.), as well as by political leanings (liberal, conservative, etc.)
If you’re able to get that sample correct, then you only need to interview 300-500 people to get a very reliable result. In political polling, you’ll often see them with sample sizes of 1500 or so, but this is because the pollster wants to be able to show their results among sub-groups within the sample (men, registered Democrats, etc.), and they want to have sufficiently large samples within those sub-groups.
Over the past 15-20 years, market research, as a whole, has really struggled with how to ensure that their samples are representative of the entire population, for a variety of reasons:
We used to use telephone interviewing a lot (and, as noted above, the political polling models were all based around it). Caller ID is now close to ubiquitous in the U.S., and many people won’t answer a call from someone they don’t know.
Fewer than half of U.S. households now have a landline phone. The demographics for people who don’t have landlines generally skew younger, as well, meaning that only surveying people with landlines (as political polls have traditionally done) will yield a sample of people who won’t necessarily look like the broader population.
Surveying via cell phones is certainly possible, but it isn’t necessarily as simple as substituting cell phone calls for landline phone calls. One big issue is cell phone number portability – if I’m a pollster, calling a landline with a 312 area code, I know I’m calling a household / respondent in Chicago. Someone who has a 312 area code on their cell phone could literally be living anywhere – they may have gotten that number years ago, when they lived in Chicago, but they can now keep that number, regardless of where they move to. This makes creating a representative sample (from a geographic standpoint) with cell phones very challenging. In addition, polling on landlines had decades of validation testing behind it, something that cell phone polling doesn’t (yet) have. And, cell phones have the same problem that landlines do, from the standpoint of people screening their calls.
Most market research (i.e., for consumer products, which is what I work on) has shifted to online research. This has its own problems on representativeness regarding people who aren’t online (particularly senior citizens), but more broadly, the bigger issue is that the demographics of people who are willing to participate in online research likely isn’t representative of the entire U.S. population, and the way in which many market research companies get people to participate in online studies (i.e., offer them some sort of points/incentive) undoubtedly attracts a particular type of respondent (i.e., those who are easily incentivized in this way).
For political polling, as I understand it, as with cell phones, there hasn’t been much validation work done to assess how accurate online political polling is, and political polling has traditionally not used respondents who receive incentives to participate.
The people who’ve built predictive models around polling (Nate Silver, Sam Wang, etc.) have done so looking at decades of polling data, which was (nearly) all done using the old paradigms and old methods of political polling…all of which are now up for grabs.
I should also note the L.A. Times poll, which has been mentioned. They took a different approach to a traditional poll, in that they recruited one sample of respondents, and then surveyed them repeatedly over the course of the campaign. This is called a “longitudinal sample”, and it’s not how traditional polling has been done (most polls recruit a new “one and done” set of respondents with each wave of a poll). The L.A. Times poll consistently looked different than most of the other polls (it was nearly always higher for Trump), and folks like Nate Silver had a hard time fitting it into their models (because of its different methodology). As it turned out, in this case, that poll may have been closer to “truth”, at least this time out. It’ll be interesting to see if their approach becomes more widely used.
And, yes, there may well have been a phenomenon of “bashful Trump voters.” But, with Clinton also widely disliked, it’s entirely possible that there were “bashful Hillary voters,” too. I’m not sure that we’ll ever really understand how much of a factor they played in the polls.
This went on longer than I intended! In short: political polling is really, really difficult to get right today, for several reasons. I don’t think that polling is going to go away; I anticipate that there’ll be a lot of analysis that gets done, and a lot of experimentation, in an effort to address the issues. But, if there were easy fixes to be made to get polling “right”, they likely would have been done already.
No, it wasn’t. The L.A. Times poll was a national poll, predicting the national vote. Hillary won the national vote, so the L.A. Times poll was off. That the L.A. times poll happened to correspond with a case where a few state polls were off in the other direction is sheer coincidence - or at least something you can’t distinguish from sheer coincidence.
Fair point, and I should have caught that in my edit. 
The LA Times Poll basically had a single issue that made it lean Trump: a black Trump supporter, which, due to the low number of people involved (much lower since they kept polling the same people), weighted the system heavily towards Trump. That black Trump supporter had to stand in for a lot of black people.
Seeing as Trump didn’t win with higher African American support, I don’t really think the poll was good–even without what squidfood said.
“Higher” is relative, and you didn’t specify what you were comparing it to, but here is how Trump did compared to previous Republican nominees:
Trump won 8% of blacks in 2016
Romney won 6% of blacks in 2012
McCain won 4% of blacks in 2008
No links, but in '04 Bush won 11% and in '00 he won 8%.
Romney: 60.933M
Trump: 62.694M (not the final count but close enough)
That’s apart from the fact that popular vote doesn’t mean much. Voting Republican in California, for example, is an exercise in futility. I am surprised 4.4M people bothered.
There was more enthusiasm where it mattered. 2.913M votes in PA vs 2.619M for Romney in 2012. Or vs. 2.907M votes for Obama in 2012.
Nate Silver was quite explicit about the fact that if a candidate outperformed the polls in one state they would be more likely to outperform in other states, and especially in other states in the same region.