Moby Dick Is A Useless Guide To Polling, Volume 2
Someone had to say it and I said it and I'll say it again
Welcome, friends, to the second of a multi-part ethering discussion of media coverage of polls and how we can survive and thrive in spite of it. Stick around, won’t you?
In my last post on this subject, I praised Herman Melville for wisely admonishing us not to look too long into the face of polls, lest we be transfixed and beguiled by terrifying apparitions that lead us astray. I wish to reiterate that rumors that Melville was not talking about polls, but instead about gazing into open flame, are lies spread by our enemies to divide us. Pay them no heed!
To recap: Discomfort and distress around polls during this time of the election year is about how they are covered, not about how they are conducted (usually). When confronted with a poll of any kind, we must ask ourselves “what segment of the electorate, if any, is changing?” and the necessarily follow-up, “why?”. Polls can tell you the first; they are not designed to answer the second, although they can hazard a try. Not all media coverage of polls speaks to the first question and precious little even tries to address the second. Onward!
**
Let me extend my particular thanks to the New York Times for serving up a prime example of the effect I just described. This week, the NYT released a poll under the headline Trump Shows Signs of Strength in Sun Belt Battlegrounds, Polls Find. Grim news for Democrats! Let’s take a gander at one of those Sun Belt states, shall we?
The NYT Sienna poll shows Donald Trump opening up a 5 point lead over Kamala Harris in Arizona, 48-43%. Alarming! Especially alarming since the previous NYT Sienna poll, in August, had Harris up by 5%, 50-45%. A ten-point swing! In a battleground state! In a single month! This, by any reasonable measure, is an earthquake. How does the NYT account for this staggering shift in the intentions of the Arizona electorate?
Cleverly, it doesn’t. The coverage mutters a bit about Trump’s popularity with Latinos (on which more momentarily), but basically presents the poll’s findings as if they simply sprang fully-armed from the head of Dash Sulzberger. There’s even a piece about how a lot of Arizona voters must be ticket-splitting because of how well Reuben Gallego is doing in his US Senate matchup with Kari Lake. It is for us, the readers, the dig into the crosstabs a bit, and see that the Harris’s reversal of fortunes in the state is to do with a decline in support among Hispanic voters and a collapse in support amongst women, a full 10% of whom appear to have defected from Harris to Trump.
All of which naturally calls the question “what the fuck?”
Let’s be clear: it is possible that the NYT Sienna poll is right, and that if the election had been held on the days the poll was conducted, Trump would have romped in Arizona. What I’m objecting to is the way that the NYT presents a tectonic change in a critical state based on swings toward Trump in two segments (women and Hispanic voters) the scale of which would not so much defy national trends as openly spit in their face, without attempting to explain that at all. What are we to believe happened? That women and Hispanic voters in Arizona watched the September 10th debate and - alone among all constituencies in the country - swung from Harris to Trump? Who’s to say? Certainly not the commissioner and analyst of the poll, the nation’s (notional) principal organ of record.
Voters make decisions for reasons that are nuanced and that can be unpredictable, but there’s nothing magical about their process. Something happens - an act or acts by a candidate; a change in environment; a change in the voters’ degree of political engagement (as typically happens post Labor Day) - that makes a group of voters identify more with one candidate than another, and polls reflect that shift in opinion. Contra to the way the NYT presented this poll and the way that the news media typically covers these things, Number Does Not Just Go Up Or Go Down, and to present polls as if that were the case is to misuse them. If you can’t even guess as to the process by which a shift appears to have occurred, it’s worth stopping and asking yourself if the shift has occurred in the way it appears to have, or indeed at all.
For what it’s worth, I’ll surmise what’s happening with the NYT polls. The September poll - the one with Trump up five in Arizona - is an outlier (not my analysis, just an observation). The August poll - the one with Harris up five in AZ - was also an outlier (same), but a little less clearly so because a lot of polls were showing a trend in Harris’s direction after her accession to the top of the ticket. Most polls, then and now, have Harris with a slight lead in Arizona, within the margin of error and in ways that appear to largely accord with national trends. My guess is that the first poll was a bit broken, and the second poll was a bit broken, and the ways in which they were each broken highlighted how broken they both were.
It is, of course, entirely possible that either the August or September poll was right. But I submit that it is essentially impossible for both NYT Sienna polls to have been right; either the August poll was off, or the September poll was off, or both, but it strains credulity past breaking point to accept a swing of that magnitude, against national trends, toward the candidate who didn’t have a national convention and came off worse (even by the evaluation of his own side) during the biggest political spectacle to occur between one poll and another, amongst two segments that have demonstrated consistent hostility toward that candidate since the race began.
The right call, faced with the second of two outlier polls the results of which are so contradictory as to cast doubt on the validity of both, could be to run coverage of the second with a heavy emphasis on trying to account for the difference. It might be to simply pull the poll altogether, or deemphasize coverage while you work with the pollster to figure out what’s rattling under the hood. But to blithely present the results as straightforward findings is just embarrassing.
And this happens all the time (although the recent NYT example is probably the most egregious I’ve seen); blithe presentation of findings as fact, absent context and - critically - any serious analysis of the drivers behind those facts. And that is how you find yourself writing in the Grey Lady about a very-likely-imaginary legion of Trump-Gallego ticket-splitters i mean good lord guys get it together.
**
One more thing about that poll, and most polls in general, that comes back to the question of “who is moving?” Again, the September NYT poll showed a serious collapse in Harris’s Hispanic support in Arizona, which runs against most national findings (again, I think this is the result of two polls that got the results slightly wrong in opposite ways, creating the illusion of a major shift, but that’s just speculation).
Let us be clear: there is a very evident trend line in Hispanic voting; we see it in polls and we see it in voting behavior. Hispanic voters, over the previous two decades, have trended slowly but noticeably Republican. These results are not evenly distributed across the Hispanic population and varying by geography, generation, and specific cultural legacy within the broad definition of “Hispanic”. This is not a secret. This is not a truthbomb. This observation should be as controversial and titillating as a high school physics lesson.
AND YET
Every election cycle, and especially this one, we are treated to reams of breathless coverage about the startling trend in Hispanic voters who are going Republican; young Latinos “flock” to Trump, we are told; Trump’s appeal to Hispanic voters is underestimated, huge, and so on. This coverage is disordered and insults both its subject and its audience.
Again: there is no sharp trend among Hispanic voters toward the Republican Party, but rather a slow, steady movement in that direction. Young Latinos tend to be more conservative than young Latinas, but there doesn’t seem to be much evidence that that is translating into strong support for Trump himself (my firm’s work with this segment suggests that Trump might in fact be more a liability than an asset in this specific respect). And Trump’s appeal to Hispanic voters, in his third straight presidential election, does not appear to have deviated all that much from expectations. He might do surprisingly well, or surprisingly poorly, with this large constituency, possibly enough to win or lose the election itself, but the idea that Trump has some powerful, hidden appeal with Hispanic voters is an assertion without evidence.
We can speculate about why this flock of canards pops up in coverage with such regularity. The core reason, I suspect, is that it appears paradoxical for Hispanic voters to support an overtly racist presidential candidate who has made open xenophobia toward Hispanic immigrants a load-bearing plank of his platform. But for this to seem paradoxical requires a flattened conception of Whiteness, migration, and the politics of identity and material interest. It might not to be reasonable to expect every member of a media audience to be current on these issues, but it is reasonable and indeed necessary to expect it of the media covering and commenting on Hispanic voters in elections.
The size of the Hispanic demographic means that even slow movement toward the Republican Party within some of its segments is a various serious matter for Democrats. Progressive political and advocacy organizations are devoting an increasing amount of time and resource to understanding what is happening and why (my firm’s own studies are part of this larger effort); someday we might be able to get into some details here in this substack. For the moment I’ll say that Republican drift within some segments of the larger Hispanic demographic has a complex but definite internal logic; it is not, in any sense, just sort of happening, and certainly is not the result of some hidden pull toward Trump, whose depiction in media coverage as a kind of Pied Piper for Hispanic voters, who are forever described as about to line up behind him in unprecedented numbers, is unsupported and frankly revolting. If I could strangle one canard about voter behavior with my bare hands, it might be that one.
**
There’ll be at least one more in this series on how to consider polls. Until then: when you’re next confronted with a poll, remember to ask “who is moving?” and “why?”. If the answer to both isn’t clear, take it with a grain of salt.
And if the answer to either is “Trump’s huge appeal to Hispanic voters” without prominent and rigorous data support, toss it in the fucking trash where it belongs.