If you’re addicted to politics like I am, every four years you also get addicted to political polls. And since my undergraduate minor was stat, beginning in August I begin every day by looking at the numbers and trying to figure out whether the polls are really saying what they appear to say.
Right now what they appear to say is that Joe is ahead in the national popular vote by somewhere between 3 and 13 percent. Wait a minute, you say, all the aggregate pollsters show Joe with a lead of 8 or 9 points, somewhere around 50-51% for Biden, 42-43% for Trump. But if you read the fine print at the bottom of the page, you’ll notice that virtually every poll says something about having a ‘margin of error’ of 5 percent. If you deduct 5% from Joe’s number he’s down to 48%. Add 5% to Trump’s numbers and he’s up to 45%. Get it? Good.
A 48% to 45% margin for the popular vote is exactly how things ended up for Hillary and Trump in 2016. But that’s the easy part. Now things get a little more complicated.
You’ll notice that every poll says that the respondents were either ‘registered’ voters or ‘likely’ voters, but never both. The registered polls tend to be earlier in the season, the polls answered by likely voters are mostly what starts to be published right around now. I guess this is because asking someone whether they plan to engage in any particular activity four months before it is scheduled to take place, is asking them to make a prediction about their own behavior which may or may not actually occur.
I happen to think that grouping voters in either category is basically a waste of time, if only because the fact that someone says they are intending to vote doesn’t mean that when they show up they will actually spend more than two seconds thinking about which lever to pull down. And I have yet to see any pre-election poll which asks respondents to divulge just how much time – hours, minutes, seconds – they have spent thinking about how they are going to vote.
Then there’s a bigger problem when we look at the daily voting trends published by aggregators like Silver’s 538, 270towin, Real Clear Politics (RCP) or Huffington Post. Very few pollsters run their polls more than once a week, sometimes even less frequently than that. So unless you aggregate the results of numerous polls together, you won’t have any new data to publish except once a week. Which means your website will go stale very quickly, which means your clicks will start dropping like flies.
On the other hand, if you aggregate a whole bunch of polls every week, how do you know that the pollsters aren’t measuring responses by using duplicate answers from the same respondents, whether they even know it or not? If you’re an aggregator, how do you adjust the numbers to take into account an unintended bias that will impact your trends from duplicate counts? You can’t.
I try to deal with these problems by trusting one pollster who polls very frequently, in this case I reference Morning Consult, which claims to run a national poll every day. Here’s their aggregate national poll from May through today:
Joe is at 51%, Trump at 43%. Note how little the results have changed over the last 90 days. I’m hardly the first person to notice that the numbers for both candidates have been basically unchanged over this period of time. In 2016, Hillary was under 42% on July 1st, pulled back up to 45.7% at the beginning of August, was just over 42% as September came around and moved between 45% and 46% over the last 30 days. Here’s how Clinton and Trump looked from the beginning of June:
I have left the numbers for Johnson and Stein off the graph, but note that not only does this graph show much greater day-to-day variation for Clinton and Trump, but more than 13% of the queried voters said they wouldn’t vote for either of them at the end of the campaign. Right now the Sedaris dog-shit number for 2020 in half of what it was in 2016.
That’s where things stand at the national level but it’s the state-level polls that we still need to see. That’s tomorrow’s column. Stay tuned.