Claims of US election fraud: geographical variations in the Biden vote part 1.

In the US, it is common to analyse election results at county level and compare with past elections to see how support for the two main parties changes with time. In 2020, it also seems to give rise to people scouring the county-level data in the ‘battleground’ states — the ones whose results were contested by Trump — to see if they can find a few ‘abnormal’ results in which they can claim election fraud to have taken place. For instance consider this analysis by Scott Hounsell — one of a three-part piece covering Pennsylvania, Michigan and Wisconsin.

All three compare Biden’s results to Obama’s in 2012 on the country level. (They also compare Trump’s against Romney’s.) Instead of considering the most commonly used comparison metric — the percentage swing to either the Dems or the GOP, they look at the percentage improvement of the vote totals for both candidates. This is the quantity obtained by taking the votes in a county for Biden (or Trump) in 2020, dividing them by the votes in that county for Obama (or Romney) in 2012, subtracting 1 and expressing the resulting fraction as a percentage.

The argument. To sum up the exact argument being used:

  • In 57 out of the 67 Pennsylvania counties, Trump’s percentage improvement over Romney was better than Biden’s over Obama. Indeed, in 35 out of the 67 Pennsylvania counties, Biden got fewer votes than Obama did in 2012.
  • In four counties — the suburban Montgomery and Chester and the Republican strongholds of Lancaster and Cumberland — Biden increased his vote by 30–40% compared to Obama in 2012. Those counties also saw overall voter turnout increase by 3–6%.
  • Four other countries — the metropolitan Allegheny, Delaware and Dauphin plus Centre, containing Penn State University — saw Biden increase his vote by about 20%, with Trump also increasing his vote by a smaller amount.
  • The remaining two counties are not discussed. One is the deep red Butler to the north of Pittsburgh, the other the ‘purple’ Bucks to the north of Philadelphia that edged a little bit into being more blue.
  • This implies that Biden’s results in Lancaster, Cumberland, Montgomery and Chester are ‘abnormal’, both when considered against the rest of the state and when considered against their geographical neighbours.
  • Trump’s increases over Romney are even throughout the state and are therefore ‘consistent’.

I will use percentage swings instead of the metric Hounsell chose. It is more mainstream and treats red counties the same as blue ones. Improving from 200,000 to 300,000 votes would count as ‘50% improvement’ for Hounsell but improving from a 10,000 to 20,000 as ‘100% improvement’. In Philadelphia County, Barrack Obama grabbed 85% of the vote there in 2012, with 588,806 votes. In 2020, 741,377 voters cast their vote in this county. Even if Biden, improbably, won every single one of those votes, he would have only improved by 26% over Obama. (In practice, he only improved by 2%, an increase far smaller than Trump’s over Romney in the county. However, Obama was super-popular in the inner cities.)

Chester and Montgomery — the suburban counties. The swings in Chester and Montgomery are indeed striking — over 17% in the former and 12% in the latter compared to 2012. Both are suburban counties around in the city of Philadelphia, and the higher turnout plus swing for Biden is consistent for suburban areas nationwide. Hounsell needed to compare those to other similar areas in the country, but he did not do so. Moreover, he ignored the fact that in 2016, Philadelphia suburbs were already swinging hard towards the Democrats. There were already warning signs back in August that the swing would continue.

Lancaster and Cumberland compared to the rest of the county. In Lancaster, despite Hounsell’s claims, the actual swing is not big — 3% to Biden. In Cumberland, it is just under 8%. That in itself is surprising, given that many similar Pennsylvania counties swung hard towards Trump.

However, let us look at the spread of those swings to Trump. In Warren County, it is a gigantic 22%. In Schuylkil County, an even greater 26.5%. It could very well be — something for me to verify later — that a typical non-metropolitan county swung on average 10–15% to Trump, with some big outliers on either side. That in itself is not evidence of foul play.

How a proper analysis should have been done. To do the analysis even semi-properly, one has to split the Pennsylvanian counties into two parts — the metropolitan/suburban and the rest. In the first bracket, you would have Philadelphia, Delaware, Allegheny, Dauphin, Chester, Montgomery, Bucks and Centre. This should be compared to other metropolitan and suburban areas in the US, not to the rest of Pennsylvania. Geographical proximity is not a guarantee of uniformity of voting behaviour.

In the second bracket, you will have all the other counties and analyse them together to see if the three that swung towards Biden — Lancaster, Butler and Cumberland — are anomalous or not. I suspect the first two are not and the last one would merit closer examination — perhaps the state capital of Harrisburgh is slowly creeping further into it, making it more suburban and hence less Republican.

PS: Hounsell uses the argument ‘Biden is not as popular as Obama’ repeatedly. In that case, how does he explain the former getting 81 million votes? Is he alleging localised fraud in just a few suburban counties or a great nationwide election-fixing effort giving millions of votes to Biden?

PPS: Hounsell’s argument for Wisconsin and Michigan is similar: pick a few suburban counties in which Biden improved and allege foul play. Because Biden’s vote is concentrated in the metropolitan areas and his biggest gains in the suburbs, he is able to allege that ‘Trump’s gains are uniform and Biden’s are not’ by assuming uniformity over geographically proximate areas and not bothering to check whether demographically similar areas behaved in a similar way during the election.

For Wisconsin, he identifies the suburban counties of Washington, St. Croix, Dane, Waukesha, and Ozaukee. In Michigan, the counties of Antrim, Grand Traverse, Livingston, Leelanau, Washentaw, and Kent were supposedly abnormal. They vary from Kent and Washentaw containing the cities of Grand Rapids and Ann Arbor respectively to coastal counties such as Leelenau and Grand Traverse. Once again, he includes Antrim, that swung away from the Dems between 2012 and 2020 as something suspicious. Grand Traverse only swung to the Democrats by 3% and Livingston by less than 1%. Even Washentaw was only a 5% swing to the Democrats. The only double digit swings to the Democrats among the counties Hounsell highlights are Kent and Leelenau, both with around 12%. However, the latter is said to have had ‘an influx of affluent retirees from Detroit’.