Elections

Live by the Polls, Die by the Polls

Donald Trump may not be as “wildly unpopular” as the media and pollsters suggest.

No one who cares about the politics of our current historical moment can avoid the obsession with polls. According to most media sources, President Trump is in some difficulty, given his poll numbers. Not fatal difficulty, perhaps, but he is routinely described as “widely unpopular.”

Even excluding CNN and MSNBC from our list of news sources (no one can seriously argue today that CNN is a news source, and MSNBC hasn’t been a news source since long before it hired Rachel Maddow)—the description of Trump as “generally unpopular” is now embedded in the firmament of our political discourse.

RealClearPolitics is the go-to source for the poll numbers, and Democrats routinely describe Trump as “easy to beat,” if only they had the right candidate, or the right message, or the right set of issues with which to challenge him. We all have looked at the RealClearPolitics graph, and the “average” of the polls—and it does look fairly dicey for Trump. His numbers in the RCP average tend to bounce around 42-45 percent (about 44 percent lately), which, for a sitting president overseeing the best economy in three generations, is surely ominous.

It is precisely this RCP average as a political analysis tool that has been driving me to distraction for quite some time. The RCP average appears in many online news stories. As best as I can understand their methodology, RealClearPolitics uses a simple mathematical average of the polls—irrespective of the nature of survey samples. In plain English, the RCP average mixes polls of All Adults (AA), Registered Voters (RV) and Likely Voters (LV) into a single moving “average.” This is separate and aside from any inherent biases of the polls themselves (e.g., oversampling Democrats is a common sin).

This “average” relies on a fatally flawed methodology—at the very least, before averaging anything, RealClearPolitics needs to normalize AA, RV, and LV survey results to the same scale—LV being the most logical one, since the objective here is to predict how the actual vote will go, not how some generic adult “feels” (or will feel) about Trump on some existential level.

In my approach, I’ve adjusted AA numbers up by five points, and RV numbers up by two points—which should be close enough to account for undersampling Republicans in AA/RV surveys relative to LV, based on historic voting patterns. I am well aware that one can debate whether the AA-to-LV adjustment should be five points or four, or 4.5 percentage points, for example, or whether the RV-to-LV adjustment should be two or three points, or 1.5 percentage points, but fundamentally it doesn’t change the big picture.

Ideally, each poll would also be adjusted to account for its left-leaning bias by some scientifically derived factor. As an example, Quinnipiac is a notoriously leftish (and anti-Trump) poll, and its results simply cannot be used directly, at least as they apply to Trump. Recall Quinnipiac’s prediction in June that Biden would beat Trump in Florida by nine points. This made the news at the time, even though the absurdity of that claim is self-evident (and the poll was released just before Trump’s blowout campaign opening rally in Orlando).

Florida tends to be won or lost by very narrow margins. Trump won Florida last time and the state’s Republican governor, Ron DeSantis, is solidly above 60 percent in approval. Florida’s economy is growing as if on steroids. It is not impossible for Trump to lose Florida, given the closeness of the win last time and the demographic shifts in Florida. But the idea of Biden beating Trump in Florida by nine points is ludicrously far-fetched.

The point is not to argue about specific polls, but to take as a given that if we want to compare apples-to-apples, some polls need to be adjusted to account for an obvious anti-Trump or pro-Democratic bias.

Typically when each poll comes out, and especially if the results are “newsworthy” (at least, as the lapdog media see it), a few people dig into the poll methodology and the sample demographics, and often find something fishy—visible undersampling of Republicans, or excessive concentration of respondents in deep blue states, and so on.

In my calculations for the chart below, I’ve adjusted Trump’s approval numbers from Quinnipiac, CNN, Politico, and Reuters up by an additional three points (which, at least with respect to Quinnipiac, seems to give them more of a benefit of the doubt than they deserve; whether CNN is even a legitimate source of poll information at all is also debatable, but I kept it in the mix nonetheless).

Again, one can argue whether three percentage points is the right number, and whether it underestimates or overestimates the bias, and whether the other pollsters need to be adjusted as well (perhaps by a lesser amount—though no one will ever call Washington Post/ABC unbiased, for example), but for now, we’ll stick with a three-point shift upwards just for those four.

Rasmussen is the only major pollster that consistently samples likely voters. Trump’s approval in Rasmussen polling has been more or less steady for many months around 47-49 percent, sometimes a bit less (down to 46 percent at times, in the mid-40s in October, at the height of impeachment fever), occasionally a bit more (at times as high as 50-51), but these are mostly short-term noise artifacts.

Rasmussen is probably the gold standard when it comes to meaningful Trump approval numbers (especially given the larger survey samples). In the chart below, I’ve deliberately excluded Rasmussen’s data, because I wanted to see how the other pollsters are doing, without the chart being pulled upwards by the Rasmussen numbers.

The plot below represents Trump approval from September 1 through January 24, as reported by RealClearPolitics, with the adjustments discussed above:

A number of things stand out: First, through thick and thin, a trendline would show that Trump’s numbers are rock steady, rising slowly from 46 percent to about 47.5 percent. The timeframe covers the Ukraine phone call “bombshell,” the initial Ukraine-related screeching by House Democrats, the impeachment inquiry House vote, House Intelligence Committee Chairman Adam Schiff’s farcical hearings, House Judiciary Committee Chairman Jerry Nadler’s comical hearings, House Speaker Nancy Pelosi’s solemn prayers, the impeachment vote, the withholding of the articles of impeachment gambit, and the start of the trial in the Senate.

And yet, in the big picture, all the drama, all the solemnity, all the prayers, all the posturing, all the hysterics, all the shouting, all the conspiracy-mongering—all of it, together or separately—means exactly nothing. Any changes in Trump’s approval are little more than noise, and his approval hovers around 47 percent, with a visible, though hardly spectacular, steady rise over the past five months:

Second, this is consistent with Rasmussen polling. If Rasmussen is reporting 48-49 percent approval for the president over the long-term, then the chart above represents a good sanity check on their numbers.

Third, this is manifestly not consistent with FiveThirtyEight’s aggregation and analysis (which, incidentally, considers Rasmussen to have a pro-Republican bias):

I do not have sufficient insight into how FiveThirtyEight massages the data to generate its chart (and I presume their weighting algorithm is far more sophisticated than anything I can gin up with Excel at home—a discussion of their approach can be found here), but a difference of four or five percentage points is clearly well outside any noise factors and minor adjustments. Keep this in mind before treating my take on the polls as gospel.

Fourth, this is consistent with the fact that Democratic voters are inefficiently distributed—which is another way of saying that more Democratic voters than necessary are clustered in a handful of states, such as California and New York. The few recent swing state surveys show Trump at rough parity or better, compared to hypothetical opponents.

Yet another way of saying this is that if a national approval rating of 48 percent today really represents “true approval” (e.g., if we had some fantasy poll of 20,000 likely voters, taking a proper weighting of the relevance of the particular states into account, and done by a perfectly neutral, unbiased pollster), then Trump cruises to reelection.

Fifth, this is consistent with anecdotal evidence, such as enthusiasm shown by Trump voters at his rallies, the relative lack of enthusiasm shown for “frontrunner” Biden by his voters (at least judging by meager attendance at his events), Republican fundraising in this cycle, and so on. One can easily get carried away with such anecdotal points, but it certainly helps that the less-quantitative evidence points in the same direction.

Sixth, there is reason to suspect that the poll numbers are actually better for Trump than what we see from the pollsters, due to what might be called “cultural” reasons.

Fewer and fewer people are willing to answer phone calls from unknown numbers; more Republicans than Democrats are likely to refuse to talk to pollsters; perhaps all the polls that I did not adjust at all, in the chart above, could use a shift of 1-1.5 percentage points upwards. It’s also possible that many Republicans who voted for the Libertarian Party in 2016 (and who might be part of the “Trump disapprove” numbers currently) will come home—adding a few percentage points to Trump’s numbers in several states (e.g., Colorado, which is almost certainly out of reach for Trump—but would have a remote shot at winning, should half of the 5 percent Libertarian vote shift to him).

It would not surprise me if the graph I generated underestimates Trump’s performance (not to be confused with “approval of Trump as a person,” or “approval of Trump’s tweeting,” or “approval of Trump’s style”).

Seventh, Barack Obama, despite being always described as “broadly popular” by the sycophantic media, actually had much worse RealClearPolitics numbers at this point in his presidency—his approval rating, using the same flawed RealClearPolitics averaging methodology, was around 44-45 percent. But given the leftish bias of the pollsters (and the resulting pro-Democratic bias of the RealClearPolitics “average”), Obama’s actual approval numbers should be adjusted and normalized downwards for an apples-to-apples comparison—unlike Trump’s.

Without doing the math, I would guesstimate that the same chart done for Obama at this point in his presidency would show his numbers around a long-term 43 percent trendline—compared to Trump’s roughly 47 percent trendline in the chart above. And yet, Obama won reelection relatively easily. Granted, Obama had Mitt Romney for an opponent—it always helps to be lucky in your choice of enemies—but Trump, of course, would be lucky to have Biden or Sanders for an opponent.

Eighth, it is hard to see what else the Democrats can throw at Trump at this point to drive his popularity down.

Impeachment was the heaviest of the heavy artillery—and there is no evidence that it’s working. If Trump’s numbers stayed pretty much steady through it all (and, if anything, have edged upwards a bit), what more can the Democrats do? Impeach him again? And again? Exhume Robert Mueller and wheel him out one more time? Have Stormy Daniels testify before a House committee? If Trump can avoid major unforced errors (never a sure thing with Trump, admittedly), it should all be mostly upslope from here.

Ninth, this is consistent with Trump’s polling gradually shifting from a referendum on Trump to a choice between Trump and his record, and a shrinking group of specific Democrats (and the policies they promote).

I suspect that the more actual voters are asked to choose between Trump and Sanders/Biden/Warren (and their lunatic visions of $100 trillion spending plans and a Venezuela-like future for the United States), the more this comparison benefits Trump.

As the 2020 election gets closer, this is what one would expect to see—thus, the slight upward slope of the trendline is consistent with that theory.

Tenth, does a stable approval rating in the high 40s (let’s call it currently about 47-48 percent, especially if we accept some of the other assumptions above) really translate into “unpopularity”?

Very few national-level politicians in this age of high partisanship and rancor sport approval ratings above 50 percent. Many European leaders, who snigger and sneer at Trump, would kill for a rating even close to the president’s. Nancy Pelosi’s approval rating, incidentally, is far lower than Trump’s—in the 30s.

Of course, Albert Einstein famously said that everything is relative. Pelosi is wildly popular compared to Senate Minority Leader Chuck Schumer (D-N.Y.)—he is in the 20s. If Trump is “widely unpopular” with his high 40s, then what does that make Pelosi and Schumer? The “Trump is widely unpopular” canard is just pro-Democratic media working to advance the leftist anti-Trump narrative.

All of this is a long way of saying that if Democrats think that Trump is beatable based on his allegedly “low” RealClearPolitics approval numbers, and all they need to do is this, that, or the other thing, or maybe also that other thing, they are deluding themselves. The November election will not be a slam dunk, but if I were a betting man, I would put money on Trump being on a path for another four years.