Claims of US election fraud: the riddle of the voting machines favouring Biden

A particularly bold claim about voting machines surfaced a few days ago. An anonymous account on Rumble claimed they built a predictive model. This predictive model took every one of the of the 3243 US counties and county-equivalents in the US and compared the Biden vote share predicted by the model with the actual result.

The conclusions of the analysis: Biden vote share in counties in which Dominion and Hart machines were used were significantly under-predicted by the model.

Then, it pulled out its ‘trump’ card, marking in red the counties where particular voting machines (Dominion and Hart) were used. And — surprise surprise — they ‘showed’ that in 72% of the counties that used the Dominion voting machines, the actual Biden vote share compared to the prediction was higher. What is more, if you subtracted 5% from the Biden vote share in each of those Dominion-using counties, this data aligned exactly with the data from the non-Dominion counties.

Provided we could see and examine the model, this would be strong evidence that Dominion voting machines added extra votes to Biden’s vote totals (the author estimates a 5% boost to Biden). The choice of voting machine should be entirely independent of the degree of over- or under-prediction, given that counties of very different demographics and political leanings use Dominion. Therefore, even if the predictive model is inaccurate, the kind of pattern shown above would still be a warning sign provided the model is not ‘doctored’ to generate this effect. By ‘doctored’ I mean amended after the election (or with knowledge of what voting machines are used where) in order to produce the sort of pattern observable above.

If the model is doctored, then all bets are off — the X-coordinate of every point in the plot above is arbitrary and could be in a different place from where it actually is. You can pretty much show anything that way.

Thus far, we have no reason to believe that the model has not been doctored. The author is Russell J. Ramsland Jr., who in a separate signed affidavit confused Michigan and Minnesota as well as citing false turnout statistics as a similar claim comes up here). His model is not detailed in any paper and not verifiable — we are only told it is 90% accurate which, given that the vast majority of US counties are either red or blue strongholds, is not a mark of great quality. He has not shown it running on other election data. All the info we have is the video above and a short description here. However, viewing the diagram on page 6, we can see that this ‘accuracy’ is questionable, with points in the main ‘point cloud’ being wrong by up to 20% and the ‘outliers’ by more.

The predictive model predictions compared with the actual results. Blue line is the linear fit to the data. The model accuracy for each point is the difference between its X- and Y-coordinate: it is clear this is considerable for many points.

Furthermore, one has to ask questions regarding the obvious avenues of research not taken. Why use only the predictive model to compare the election results against? Why not also put the 2016 vote percentages on the X-axis and the 2020 ones on the Y-axis? Or perhaps 2012 ones, before Trump appeared on the scene? Dominion fraud giving Biden 5% would show very clearly on such a diagram. Yet, this is not done, and I would guess not done because it does not show that Dominion voting machines gave Biden more of a gain, on average, compared to 2016 or 2012, which pretty much destroys the whole argument.

The court in Georgia concurred with this view, noting that ‘his (Ramsland’s) report is inadmissible because it utterly fails to disclose the data or methodology he (or others) used’.