Black women didn’t increase their support for Trump, and neither did white women — why you shouldn’t believe the viral “55%” claim
Did Black women really more than double their support for Donald Trump between 2016 and 2020? That’s what you’d have to believe in order to believe the more common claim that his support among white women went up between the elections. The numbers come from exactly the same sources, and the claims are equally flawed.
By the day after Election Day, you had probably seen the viral claim that 55% of white women went for Trump in 2020, up from 52–53% in 2016. It’s a pretty devastating claim, and one that right wing sources wasted no time in exploiting as a means of dividing the left. (See, e.g., Spiked (www.spiked-online.com/2020/11/05/white-women-are-the-new-white-men/amp/): “After spending the last few years lambasting white men, the woke mob has turned its attention to a new target: white women.”) The problem is, exit polling data couldn’t support this claim in an ordinary year, and the dramatic increase in absentee voting appears to have skewed the exit polling results.
The 55% claim was surprising, particularly in a year in which so many people knew white women who were becoming allies in the struggle against police violence, but people seemed to nevertheless accept it at face value as the truth. Essence Magazine opened its article the day after the election (https://www.essence.com/news/politics/55-percent-white-women-trump-election-2020/) with “More white women voted for President Donald Trump in 2020 (55%) compared to the 53% (or 52% some polls found) who voted for him in 2016, according to the New York Times exit poll.” Pundits drew the obvious, depressing conclusion: far from being allies, white women were even more of a problem now than they were in 2016. Teen Vogue ran an article a week after the election (https://www.google.com/amp/s/www.teenvogue.com/story/white-women-support-trump/amp) subtitled “White women have to answer for backing the Republican nominee yet again.” The following day, the Chicago Tribune for featured an opinion piece (www.chicagotribune.com/columns/dahleen-glanton/ct-trump-women-voters-20201111-rmuaw5q4gvgyfbt4fxbqpmbcpi-story.html%3foutputType=amp) entitled “Black women and Latinas are poised to start a political revolution without white women.” Both articles based their arguments on the purported increase in Trump support among white women compared with 2016, and both cited the same source for the 55% statistic: a New York Times article (https://www.nytimes.com/interactive/2020/11/03/us/elections/exit-polls-president.html) from Election Day, which reported results of exit polling performed by Edison Research for the National Election Pool.
What these articles didn’t pass along was the fine-print disclaimer in the New York Times article, which showed that the data couldn’t be used in order to make comparisons like 52% in 2016 vs. 55% in 2020. The unidentified NYT author explained that “[t]he numbers on this page are preliminary estimates from exit polls conducted by Edison Research for the National Election Pool,” and cautioned that “[t]he polls are not precise enough to distinguish between, say, 53 percent support or 50 percent support from a certain group.” In other words, even in an ordinary year, there isn’t enough precision in these exit polls to identify a three-point change, such as the one purportedly identified for white women.
But this wasn’t a normal year, and errors in estimating absentee voting may have systematically skewed the results even further. Exit polls have a hard time modeling absentee voting in general, and an even harder time in elections with higher than ordinary levels of absentee voting. Edison Research uses telephone surveys to assess absentee voting, and then combines the phone survey data with in-person survey data, using a proprietary formula to arrive at its overall exit polling results. It’s the combining process that creates the problem. As Nate Silver (https://fivethirtyeight.com/features/ten-reasons-why-you-should-ignore-exit/) explains, combining the data “requires the polling firms to guess at the ratio of early voters to regular ones, and sometimes they do not guess correctly.” On Election Day 2020 (the day the New York Times published the Edison Research exit polling data), no one knew how many absentee ballots there would be in many states. Without the correct ratio, the demographic data can’t be used in a combined form, even if the two methods were each reliable.
Edison made its guess, as it always does, based on voting behavior in prior years. But none of those years included an election during a pandemic, let alone an election with a partisan split between voters on the issue of whether in-person or absentee voting was safer. In the 2020 election, it will surprise no one to learn that in-person voting favored Trump, while absentee voting favored Biden. Thus, we would expect in-person exit polling to show greater support for Trump across nearly all demographic categories, and telephone surveys to show greater support for Biden. Combining those results based on Edison’s Election Day guess is what gives us the 55% statistic. But if Edison underestimated the amount of absentee voting, that would mean the 55% is too high, by an unknown amount. This kind of underestimation happened in the 2000 presidential election in Florida, where, Silver notes, pollsters significantly underestimated the absentee vote, which skewed Republican, leading to an overestimation of Gore’s share of the vote.
Is there evidence that Edison underestimated the proportion of absentee voting? Yes. In addition to the fact that absentee voting happened at unprecedented levels in this election, Edison’s results for other demographics suggest an underestimate. For example, the same data that show a purported increase of 2–3% in support for Trump among white women also show an increase in support among Latina voters from 25% in 2016 to 30% in 2020. These same data sources also purport to show that support for Trump among Black women more than doubled between elections — from 4% in 2016 up to 9% in 2020. Ask yourself if you believe that increase, before you decide to believe the supposed increase among white women.
To summarize: First, exit polling isn’t precise enough to distinguish between 52% and 55% even in an ordinary year. Second, in this extraordinary year, we have no reason to believe that Edison correctly guessed the proportion of absentee ballots on Election Day, before most of those ballots even arrived. Third, the results of Edison’s Election Day guess are themselves unbelievable, given that they predict that Black women were more than twice as likely to vote for Trump in 2020 as they were in 2016.
The 55% claim is getting amplified not only by folks on the left, but also by groups that would hate for the left to build any kind of coalition, and would rather maintain the “Dems in disarray” trope. It’s a divisive message to lob at the left because it carries with it the subtext that all of the solidarity you thought you saw from white women over the summer was a lie — they were getting more conservative that whole time, because statistics. It’s a message that isn’t based on reliable stats, and one we should stop spreading.