Claims of fraud in the US elections: the votes ‘subtracted’ from Trump

Just before the supposed ‘showdown’ on January 6th, the Trump campaign’s ‘data arm’ — the hilariously named ‘Data Integrity Group’ came up with yet another claim. Supposedly, the voting machines ‘subtracted’ votes from Trump. Here is one of their videos — about Pennsylvania (an earlier one about Georgia, alleging the same was doing the rounds a few days earlier).

With a catchy title of ‘423k votes disappear, there are two data-related claims made if you strip away the rhetoric.

Claim 1 — Biden got about 90% of the vote in a number of precincts and this is suspicious. This was stated (with bar charts) as a shocking result. The district considered is District 10 in Allegheny county, although some wards in the bar chart below right (like Pittsburgh Ward 1) do not lie in it.

The bar chart on the left, comparing a single inner-city district with all the other districts is clearly misleading —to give a simple example, imagine we have a hypothetical county with 10 districts with the same number of absentee voters in each that vote 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 75% for a candidate. The first 9 districts average to 58%. We can, with the same success, claim the 75% district to be an outlier. Without either giving a spread of absentee vote totals over districts you cannot hope to prove an anomaly. They call it ‘staggering’ and ‘unlikely’. It is in fact entirely likely.

There is a further claim of ‘Trump getting zero absentee votes in some precincts being an anomaly’ without ever justifying this statement, for instance by plotting the Trump votes of every precinct in Allegheny to determine whether these are outliers.

For Philadelphia county, a more straightforward, almost emotive claim of ‘how can any precinct vote 97%+ for one candidate’ is made, again with no comparison with the rest of Philadelphia, no comparison with 2016 (when Clinton did even better in Philadelphia than Biden now) or with previous years.

Claim 2: The Trump vote decrements. The other big claim is that Trump’s vote goes ‘backwards’ in some places. When you consider the ‘live’ updates to the vote total, you would expect them to not be negative, yet sometimes they are. This is their first pair of plots.

The claim is that whilst Biden’s vote total increases continuously (except for two clear errors of zero), Trump’s has a massive update of 145,241 that then gets taken away. This looks every bit like an error by whoever entered the votes (perhaps an extra digit typed by accident), but it is presented as fraud.

The same claim is then made about election-day votes. Shown is their plot for Allegheny, but similar ones are given for Lehigh and Chester Counties. No corresponding plots for Biden are given — are his votes ever ‘taken away’ in the same way?

Unless one can show that Biden’s vote totals always increase, one has to assume these are operator errors and necessary adjustments resulting from human error as counting officials struggle to count and enter results.

The other reason this is not possible is that updates are often tied to a specific physical location, added to the count as a precinct reports. If votes are taken away, if Trump’s count is reduced to near zero, you would expect specific physical locations to have negative vote totals. That does not happen.

An emotive context. The lack of statistical clout is made up by emotive rhetoric and conspiracy theories against the leading Democrats in Pennsylvania.

PS: Georgia. This argument is very similar to one made by the same group regarding the Georgia election. For Claim 1, the authors claim to have used ‘machine learning’ to detect ‘anomalous precincts’. They start off with the clearly wrong claim that ‘receiving 75% or more votes for a single candidate in a precinct is abnormal, receiving over 90% is a marker for fraud’, without justifying that statement at all. It is a ridiculous statement, and one not justified by anything at all. Did they compare with other heavily democrat areas of the nation? Did they compare with 2016? There are no answers given. For Claim 2, likewise, only Trump’s data was shown, nothing from Biden.