Claims of US election fraud: the bellwethers.
The US consists of 50 states and Washington DC. Within those are 3243 counties and county-equivalents. Following an election, the vote totals for each county are publicly available.
A small number of those counties are the so-called bellwethers — they have the distinction of having backed the winner of every presidential election since Ronald Reagan was elected for the first time in 1980. Or, more correctly, had that distinction before last November: of the 19 bellwethers running a ‘correct streak’ since 1980, all but one — Clallam County in Washington — backed Donald Trump in this election. Two of those had far longer streaks — Valencia County in New Mexico and Vigo County in Indiana went back to 1952 and 1956 respectively. Furthermore, Trump’s wins in those counties were not narrow: he carried the bellwethers by a double-digit margin on average.
As a result, those 18 ‘failed bellwethers’ are frequently cited as evidence of election fraud. The argument goes that the election was fixed in the big ‘Democrat cities’, stealing the Trump win that the bellwethers are pointing to. Indeed, if we do suppose that ballot-stuffing and editing of vote totals did take place in the ‘Democrat cities’, this is what we would expect to see. However, there are a number of reasons to believe a far more mundane explanation: this is a result of the changing alignment of the Democrat and Republican-supporting coalitions in the US, with the Democrats gaining among some demographics and losing out in others.
A comparison to 2016. Consider a more systematic set of bellwethers, allowing comparison between elections. Let us look at all the counties that were ‘wrong’ exactly once since 1980:
We see that 14 counties picked the right winner in every election between 1980 and 2020 except for 2016. So, both Trump’s and Biden’s victories represented ‘the only wrong guess’ for a similar number of counties: 14 vs 18. (Two counties — Alamosa County in Colorado and Valverde County in Texas — went with Clinton in 2016 and Trump in 2020 having picked the winner in 1980–2012. A total of 11 counties were ‘wrong’ once in 1980–2012 but correct in 2016 and in 2020).
In the above context, the ’18 out of 19 bellwethers’ statistic looks consistent with 2016. That does not stop Trump and his online activists posting very selective lists of bellwethers in order to skew this picture: observe the following image doing the rounds on twitter:
OK, but perhaps 2016 was anomalous as well? How is it that out of the 43 counties to predict all but one election in 1980–2020, 32 failed to predict 2016 or 2020? Here, we need to look at shifts in the demographic make-ups of the Democrat and Republican vote bases. In 2016, Trump gained among white men without college degrees, including blue-collar former Democrat voters. In 2020, Biden hit back among the formerly Republican suburban white voters.
A bellwether is typically distinguished by its fine balance between Republicans and Democrats. Any nationwide shifts could well upset this balance: for instance in Vigo County unionised formerly Democrat voters started deserting to Trump, which turned the county from one in balance into a Republican stronghold. Bexar county in Texas on the other hand is a bellwether containing the city of San-Antonio. One of the fourteen to break their streak and go with Clinton in 2016, it stayed blue in 2020. This can be explained by the suburbs becoming more blue throughout America.
Are bellwethers a reliable predictor anyway? Bellwethers are more a statistical curiousity than reliable election predictor. This was known as far back as the 1970s. This 1975 paper cites three elections — 1940, 1960 and 1968 — in which previous bellwether countries were worse at predicting the eventual election result than a randomly chosen county in the nation! This is to say that the percentage of previously bellwether counties to back the overall winner in those contests was smaller than the percentage of counties nationwide to do so.