As far as the polling numbers go, Donald Trump is losing big. Perhaps. Do those numbers have any meaning whatsoever? Absolutely not.
There are two ways the public is being told how things are tracking in the current US Presidential Election. The first is the personal perception of political pundits (based on their attitudes toward events, speeches and policy proposals), and the second is polling. The relationship between the two is certainly intersectional and symbiotic. Media opinion informs people’s positions – and these positions might appear as polling data. Polling data influences media commentary – and the interactional dynamics proceed.
I have some serious concerns about the veracity of the polling data we are consuming, as well as what the polling doesn’t show, or arguably “disguises”.
Here are three problems:
- Polling often oversamples political affiliation and then fails to wash these biases out.
- Polling presents a precarious assumption about “voter turnout”.
- The “likely voter” model is outdated for the current cycle.
The first issue is the most striking and certainly the most obvious problem. The vast majority of lay-people do not interrogate the polling numbers they see. People just take the number for granted. For example, Reuters says that Clinton is up by five points – not many people will actually go and have a look at the polling data. But they should. I know this sounds really “politically-nerdy” but really there’s one essential thing you want to know about every poll: “what is the political affiliation of the people polled”?
The above Reuters poll, for example, sampled 680 Democrats and 578 Republicans. This is a 9% differential, or 17% more Democrats than Republicans in the poll. It doesn’t need to be elaborated that in a highly polarised political landscape, oversampling from one party is problematic. Be aware that this poll is by no means an outlier – this happens all the time (and mostly reflecting a deficiency of sampled Republicans).
Brexit polling got it wrong and…the US presidential polls are missing, or disguising, far more variables and dimensions. On this basis alone, any rational observer has plenty to be skeptical about.
I find the second issue far more interesting. Polling gets a 100% vote rate. They call people sitting at home, or engage on-line, and each and every respondent “votes” or contributes to the data. This necessarily assumes that all these people will come out to vote. But that’s simply not accurate. The highest US presidential election turnout in the last 100 years was 63% in 1960. Even Obama’s 2008 “Hope & Change” Election only had a 57% voter turnout.
Some people might suggest that this figure should wash-out against both parties, so if both parties have 57% of voters turnout, the polling preferences still bare out. But is this a fair comment in the current season? I’d suggest not. I’ve written about this before: the voter turnout in the Democratic primaries was at historic lows, while the Republicans’ was at record highs. Does this translate into likely general election voters? The jury is out. The past four elections say yes, but beyond that there is no clear correlation. One should also consider the number of people attending the respective Donald Trump and Hillary Clinton rallies. Trump regularly gets 5,000 – 15,000 attendees while Clinton often speaks to hundreds rather than thousands. Will this enthusiasm gap mirror voter turnout in November? Maybe yes, maybe no – but I think it’s hard to argue that it’s a “good” thing for Hillary Clinton.
The last thing I want to touch on is the “likely voter” model. Many people don’t give this much thought, but pollsters have an assumption about how many of the people polled will come out and vote. They have built-in assumptions based on age, geography, income, education etc. The issue is that these models have not been updated to reflect Donald Trump’s candidacy. I’m trying to give Trump more credit than he deserves, it’s simply that his appeal is very much outside the bounds of election cycles of recent memory. Ask yourself this question: do you think that the people who will vote for Trump in 2016, are all the same people as voted for Mitt Romney in 2012? If you answered yes, then for you the models are fine. If however, you think they might be different people, then, Huston: we’ve got a problem.
On the one hand, pundits regularly deride Trump voters as “low-information”, “low-income”, “low-education”, and white. On the other, they say the average Republican voter is conservative, middle-income, and older. What we know for certain is that Trump is not a conservative. If he was, we wouldn’t have the “Never Trump” movement and we wouldn’t have some obscure third-party candidate lauded by this “purist conservative” crowd. In any case, the incongruities are blatant. Trump is a disruptive force in many, many ways – not least in terms of the “type” of voters to whom he appeals.
Also on The Big Smoke
There are three reasons I suggest we should be wary of using the current polling data that is emerging from the US presidential election, as a reliable guide to the likely outcome. First, it habitually oversamples Democrats. Second, it either assumes that all respondents will come out and vote, or that there is direct equivalence between Democratic and Republican turnout. Third, the “likely voter” models are based on outdated assumptions and voter characteristics that may not be as applicable to Trump voters.
A word of advice: take the polls with a grain of salt. The Brexit polling got it wrong and the main thing those polls missed was the differential in likely-voter across the age spectrum. The US presidential polls are missing, or disguising, far more variables and dimensions that simply age. On this basis alone, any rational observer has plenty to be skeptical about.