I’ve been swimming in Trump supporters arguing that the polls are “rigged” against their man. In every political cycle, there’s always some fringe that argues the polls are understating support for their candidate. In 2012, Republicans argued why Romney would beat his polls, and the Romney campaign itself truly believed this. In 2010 and 2014, as the midterm polls looked better and better (for Republicans), it was Democrats screaming about “oversampling.” I myself indulged in this back in 2012 (but hahahaha oh boy was I wrong).
But it’s worse this year. Trump supporters are being publicly bolstered by their own candidate, who is claiming the polls are “rigged” to anyone who will listen. This has energized a base that is already inclined to believe many less-than-reputable sources because of their (completely justified) distrust of the mainstream media and the incredibly, genuinely dishonest cottage industry of so-called “fact-checkers”. Unfortunately, the simple fact that the MSM is basically untrustworthy does not mean sites of the lunatic fringe (like ZeroHedge.com) suddenly become trustworthy.
They aren’t. This year as in previous years, there is no poll-rigging conspiracy.
Now, it is possible that the polls are wrong. This happens routinely. In fact, I think the media on the whole is greatly underestimating Trump’s chances. By looking at the polls and other factors, FiveThirtyEight’s model at this hour projects Trump with a 17% chance to win. That’s not good, but it’s not doomed, either: it’s the college basketball equivalent of the Marquette Warriors coming back from a 6-point deficit with 15 minutes of play left in the second half to defeat the racist Adolf Rupp. Or, for nerds: it’s the equivalent of rolling a crit on a 17-20/x2 weapon.* It’s not likely, but it still happens pretty routinely. Trump could win this election, and I wouldn’t even be surprised to see the polls proved wrong. (I also would be unsurprised by Hillary beating her polls and bringing home a landslide victory of 12 points or more. Thing about polling error is it goes both ways.)
But what the polls aren’t is rigged.
Here are four false rigged-polling stories I’ve seen in just the past 24 hours:
(1) Monmouth is Collaborating with Clinton!!!
I have no idea where this one came from.
An outlet called PolitiForum made the following claim two weeks ago (under the headline “Wikileaks Proves that the Polls Are Rigged for Clinton”):
The latest batch of files and emails show that Monmouth University was in bed with the Clinton campaign to skew polling data
This claim was attached to this document (at right), which does indeed appear to show Patrick Murray of the Monmouth poll deliberately skewing the polls in order to bolster Hillary.
Problem is, the document is fake. There is no such document in the WikiLeaks archive. There couldn’t be: this document is dated September 2016, but the WikiLeaks archive (at least so far) only goes through March 2016. The headers on this document are fairly clear photoshop jobs, given the giant lines in the page and the bizarre use of a mission-statement image seemingly clipped from the Monmouth website in the header (where no email could print an image). And the rest of the document is just absurd, exactly what you’d expect a Trump supporter with a heavy fever to dream up, from the misspelling of “embedded” to the “lying harpy” line in the summary. “Favored are liberal arts degrees, and, especially, sociology. See attached call files.” Seriously? Does anyone think Monmouth has call files of sociology majors?
This document is a clear fake that is not present in the WikiLeaks archives as claimed, but which somebody pretty clearly photoshopped up instead. Thousands of people are falling for it anyway. I guess I can’t blame them, since Dan Rather did the same thing. But it’s still wrong.
(2) John Podesta Rigged ALL THE POLLS with “oversampling!”
There’s a lot of good dirt in the WikiLeaks dumps, so it’s depressing that I have to focus my attention on this non-story for even a minute while the real scandals go unaddressed.
So here’s verbatim from the Podesta email. This is going to Clinton campaign staffers and others who are associated with the Hillary-for-president effort. “I also want to get your Atlas folks to recommend oversamples for our polling before we start in February. By market, regions, etc. I want to get this all compiled into one set of recommendations so we can maximize what we get out of our media polling.”
Now, I infer from this that we’re not talking about their internal polls. They are attempting here to set out guidelines that they want the media to follow in order to procure and produce certain sets of data. The email includes, this Podesta email “includes a handy, 37-page guide with the following poll-rigging recommendations. …
“For Florida, the report recommends ‘consistently monitoring’ samples to makes sure they’re ‘not too old’ and ‘has enough African American and Hispanic voters.’ Meanwhile, ‘independent’ voters in Tampa and Orlando are apparently more dem friendly so the report suggests filling up independent quotas in those cities first.”
This whole email, if I’m understanding this, is actually a manual for pollsters on how to produce the results Podesta wants.
…And even with these guidelines and suggestions from Podesta on how to rig the polls, what we have to believe is that all of the people that run these polls, ABC/Washington Post, NBC/Wall Street Journal, CBS/New York Times, Monmouth College, you name it, every one of these polls waits for guidelines on who to sample from the Hillary Clinton campaign.
Oi, where do I even start with this one?
Well, let’s start at the very beginning (a very good place to start):
(1) this email is dated January 18, 2008. It has less than nothing to do with this election. Even if there were any wrongdoing here (there isn’t), then it would be Barack Obama who should be frosty about it, because that was who Clinton was running against at the time.
(2) the ATLAS project was an internal polling project (by Democrats, for Democrats), whose data was intended for use in setting up campaign offices and lit drops in the most effective areas. You can see that in another WikiLeaks email, and also by visiting the ATLAS website (which clearly states its progressive loyalties). No ATLAS-originated polls were ever published in the media. So, if there was any attempt to manipulate the results, the only people hurt by it would be the Clinton campaign, not us out in the public. (This sometimes happens, to be fair — as we discussed at the start of this article, Romney was very guilty of bad internal polling in 2012.) They were sampling their own base more thoroughly in order to create a more accurate internal picture of the race for their own internal GOTV operations.
(3) Yes, I said a “more accurate” picture of the race, not a skewed one. “Oversampling” simply does not mean what the pro-Trump press (in classic conspiracy lunatic fashion) arbitrarily decided it meant. “Oversampling” is when you do extra polling of a subgroup in order to ensure that you have the most accurate possible picture of that group.
So, if you’re running a Democratic primary get-out-the-vote campaign, you are going to want a crystal-clear picture of the Latino and African-American electorate because those are your base voters. So you go out and do some extra sampling of those groups. Before the poll is published, the extra samples are weighted back into the overall sample proportionately — so, if you polled a bunch of extra Latino voters for your subgroup oversample, each Latino voter is counted less in the topline results, so that they aren’t overrepresented in the final result. (Put another way: if the overall Black population is 10%, then it doesn’t matter if there are 5 Black voters in your 100-voter sample or 50: they will still be discounted based on the demographic weight so that they count for 10% overall.) Oversampling is essentially conducting an extra poll within a poll, it is a completely statistically valid technique, and it can only enhance the polls overall accuracy.
The Atlantic just published a good article to debunk this ridiculous story and further explain the technique behind oversampling. But you can ask any pollster — including Trump-favorable polls like Rasmussen or IBD/TIPP or LAT/Dornsife — and they will tell you the same thing, because “oversampling” is a well-established statistical practice that only increases the accuracy, especially of subgroup measurements.
(3) Monmouth is lying about their weights!!!
This one is a badly mistaken tweet by somebody named @GOUSAAMER114 that the nutbirds at The Conservative Treehouse (motto: “If Salon.com were pro-Trump”) chose to magnify into a bona fide conspiracy theory. (Poor Monmouth! The best pollsters are the ones getting attacked the worst in this election.)
Today, Monmouth via Patrick Murray presents a presidential poll of Ohio voters. The actual raw data (pdf available here – see pg, 6 and 7) showed Donald Trump with a lead in the result; however, Murray changed the data through weighting to show Hillary Clinton with a lead… It’s obvious Murray didn’t expect to be called out on it – because when confronted with what he did, pollster Patrick Murray flat out lied… You can do the math yourself and see that no “weighting” was done by “region, age, race, or gender” only by party ID.
Sounds damning, but what the Treehouse says is there simply isn’t.
Yes, Monmouth gives unweighted numbers about certain categories of interest. This is useful in demonstrating, for example, that you have more than one black person in your poll — among other things. (The LA Times poll was famously thrown off by a point or so because of a single 19-year-old Black voter in Chicago in its tracking group who supported Trump.)
Then they say they weighted by other demographic information they’d collected and published those results. It would be unusual for them to post all of the crosstabs necessary for us to mathematically prove that they weighted correctly. They didn’t, so I’ve no idea how Treehouse “did the math” as they invite us to do. (Any math they did do was wrong, because we lack, at the very least, the region data required to compute the weighting.) So we have no reason to believe that the weighting was done improperly. True, we can’t prove it was done properly, either, but let’s consider what we’re talking about when we accuse them (with no evidence whatsoever) of “rigging”:
This is an extremely reputable poll that has been doing the same thing for many years. They have been so accurate using this methodology that they have an expected bias of around half a percentage point in the average election (which is very impressive). Their reputation and (ultimately) their livelihoods would be threatened by any attempt to weight the poll inaccurately. Their partisan ID results are, moreover, actually pretty friendly to Republicans compared to the current national average. (I don’t have current figures on Ohio, where this poll was conducted, but I do know that, four years ago, the party ID gap was quite a bit worse for Ohio Republicans.)
The Conservative Treehouse didn’t “catch” Monmouth in a lie; they caught “caught” Monmouth doing exactly the same thing Monmouth was doing in March when Monmouth was telling us about Trump’s upcoming sweeping primary victories in Maryland and the Northeast and the Treehouse was delightedly citing Monmouth’s work.
(4) [Insert any pollster] has a sampling bias toward Democrats!!!
This is a long-standing and universal complaint by partisans who are losing in the polls. I engaged in a heavily qualified version of it in 2012, and I’m really glad I qualified it, because I was completely wrong. This year’s edition is even more virulent.
The gist of the complaint is, “My party typically has better turnout than your poll is predicting, therefore your poll is wrong.” Stated that way, the explanation is obvious: the reason the poll predicts lower turnout is because your party is about to have lower-than-typical turnout in this election. However, conspiracy theorists looking for an excuse to ignore bad poll results have a way of twisting themselves up in knots about this. Here’s a typical example, from ZeroHedge (motto: “correct capitalization and punctuation are tools of neoconservative oppression!!!”) about the recent ABC News poll showing a 12-point lead for Clinton:
Of course, like many of the recent polls from the likes of Reuters, ABC and The Washington Post, something curious emerges when you look just beneath the surface of the headline 12-point lead:
“METHODOLOGY – This ABC News poll was conducted by landline and cellular telephone Oct. 20-22, 2016, in English and Spanish, among a random national sample of 874 likely voters. Results have a margin of sampling error of 3.5 points, including the design effect. Partisan divisions are 36-27-31 percent, Democrats – Republicans – Independents.”
As we’ve pointed out numerous times in the past, in response to Reuters’ efforts to “tweak” their polls, per the The Pew Research Center, at least since 1992, democrats have never enjoyed a 9-point registration gap despite the folks at ABC and The Washington Post somehow convincing themselves it was a reasonable margin.
While it’s true that Democrats have never before enjoyed a 9-point gap in partisan self-identification (aka “party ID”) on Election Day, Mr. Trump seems poised to overthrow that historical norm as he has so many others. (In 2012, I said much the same thing: the biggest Democratic turnout advantage in history was a D+7 win in 2008, and surely they couldn’t match that in 2012, could they? But they did. Historical precedents have a funny way of falling apart the moment you start to depend on them.)
The ABC News poll is indeed weighted by party ID, with an advantage of D+9. However, as explained in its full methodology, the party ID weighting is determined, not by ABC news analysts taking a wild stab at it (as ZeroHedge seems to think), but by other polling. This other polling, for obvious reasons, is not weighted by any partisan factors. You conduct the party ID poll (often with “all adults” instead of a likely voter screen), weight by non-partisan demographics, get your party ID weights from the results, then you go poll the election with just likely voters and apply your weights. This is a typical method for weighting by partisan ID, and it has proved reliable in every recent election.
We haven’t yet seen the partisan ID numbers from ABC’s most cutting-edge polling, but we know that it conducted an all-adults poll not quite two weeks ago… and that poll showed a Democratic party ID advantage of D+8. Bear in mind that, two weeks ago, Trump was doing a bit better, so a D+9 advantage should not be the least surprising.
ABC’s D+9 result is broadly in line with recent polling consensus:
Selzer & Co (best pollster in the country) showed D+6 on October 17.
Ipsos/Reuters showed D+14 on October 17th.
PRRI showed D+7, also on October 17th.
NBC/Wall Street Journal showed D+9 on Oct. 13th.
It’s possible there’s been some tightening here in recent days: NBC/SurveyMonkey have done two polls this week, showing D+3 and D+5, respectively, and Fox had an October 17th poll showing D+4. But that could be noise. Or the ABC News result could be a bit of noise. But what it wasn’t is rigged.
None of the polls are rigged against Donald Trump. Donald Trump simply isn’t doing very well. Conservatism (insofar as it isn’t dead) needs to recognize that, while there are a huge number of different ways to do polls, and while there is a strong liberal media bias (in many ways against Trump), the polls that show Trump losing have been conducted correctly, in scientifically valid ways.
I hope this post doesn’t need to be updated.
*You took Improved Critical on your rapier, nice job. These D&D 3.0 references don’t date me yet, do they?