Hand Counts Vs. Machine Counts in the 2020 Democratic primaries
Background
Several exit polls, conducted by Edison Research and published by CNN at poll closing time have been analyzed by Theodore de Macedo Soares and shown to deviate substantially from official election results for many of the early Democratic primaries. As of March 16, 2020, exit polls have been conducted in 17 primaries. Of those, according to Soares’ analyses, 11 have shown large deviations from official election results, beyond the statistical margin of error (at the 95% confidence level), with the election results favoring Biden over Sanders, compared to the exit polls, and one showing a deviation beyond the statistical margin of error in the opposite direction — i.e. favoring Sanders. I will subsequently refer to the deviations in favor of Biden as “red shifts”, which mean that the official results favored the more conservative or right-wing candidate in the official results compared to the exit polls. Deviations in the opposite direction are referred to as “blue shifts”, and these are extremely rare.
It should not be considered highly suspicious for an occasional red shift or blue shift to be demonstrated. In the absence of any substantive reason for them, such as election rigging for example, they will occur 5% of the time, just on the basis of “random error” — i.e. chance. But an imbalance that demonstrates such deviations in one direction in 11 of 17 analyzed elections, as is the case here, is extremely unlikely in the absence of a substantive underlying reason. Furthermore, we saw the same process occurring in the 2016 primaries, with all in the same direction. In 2016, 12 of the 27 primaries with exit polls showed red shifts in favor of Clinton over Sanders (although two of those were still under investigation at the time that Soares published his analysis), whereas none showed a blue shift — i.e. in favor of Sanders in the official results.
The 11 statistically significant red shifts in favor of Biden that have thus far been published by Soares include South Carolina (red shift = 5.1%), Massachusetts (red shift = 8.4%), Texas (red shift = 4.4%), Vermont (red shift = 10.8%), California (red shift = 7.7%), Michigan (red shift = 7.5%) and Missouri (red shift = 9.6%). In addition, New Hampshire also demonstrated a red shift (2.9%) — in favor of Buttigieg over Sanders, at a time when Buttigieg was Sanders’ nearest competitor. Soares is continuing to work on articles regarding the other states with red shifts and will be publishing them soon.
However, because many people are skeptical of exit polls (a skepticism that is unwarranted and is primarily the result of aggressive mainstream media efforts to squelch all talk of election fraud in the United States), despite their routine use in many other countries for the purpose of monitoring their elections, I felt that it would be useful to perform additional analyses for the states implicated by Soares’ exit poll analyses, to either add confirmation for or shed doubt against Soares’ analyses.
If the exit poll discrepancies from official election results that Soares has identified are indeed the result of election rigging, then it would be highly likely that Sanders would have performed better in hand counted than in machine counted voting jurisdictions, because vote counts produced by machines are far more susceptible to vote rigging than hand counting. Hand counting is susceptible to small errors, but not errors large enough to result in statistically significant exit poll discrepancies from official results.
My analyses compare election results for the New Hampshire, Vermont, and Massachusetts Democratic primaries of 2020, for machine-counted vs. hand-counted townships. Soares’ analyses found red shifts of 2.9%, 10.8%, and 8.4%, in New Hampshire, Vermont, and Massachusetts, respectively, all beyond the statistical margin of error, against Sanders. In an effort to do similar analyses, I also called the election divisions of South Carolina, Michigan, and Texas. I was told by them that South Carolina and Michigan do no hand counting of paper ballots whatsoever (though they are potentially available for counting under certain circumstances), and I was told by the state of Texas that they didn’t know if such information was available, or if hand counting of paper ballots is ever done anywhere in Texas, and that I would have to check with each township in Texas in order to find out (a task beyond plausibility). I made no effort to do the same thing in California, because the votes are still being counted there, and I made no effort to do the same thing in Missouri because I just found out about the red shift in Missouri this morning (March 28).
Methods
For all three states, I obtained the official election results from the website of their Election Division, and I also obtained a list of hand-counted vs. machine-counted townships from their Election Division (In Massachusetts and Vermont, I could not find that information on their website, so they sent me the list by email).
In all three states I compared total hand-counted vs. machine-counted results, testing for statistical significance at the 95% confidence level, using the standard two-tailed comparison of two proportions. In Vermont and New Hampshire, I also compared hand counted vs. machine counted results specifically for small townships, which were by definition differentiated from large townships according to the total number of votes counted in the township. A comparison of hand-counted vs. machine-counted results in large townships in those states was not possible because no large township counted their results by hand.
In New Hampshire, the comparison was done only for the percentages of the two-person vote totals between Buttigieg (Sanders’ nearest competitor at the time) and Sanders. In Vermont and Massachusetts, the comparison was done between Sanders and Biden, utilizing the percentages of the total, rather than the two-person vote count. In Massachusetts, the percent of the total vote count was also compared between Biden and Warren, because Warren was also a major candidate in that race.
Results
The main result in all three states, depicted in the Table below, was that Sanders performed substantially better in his total hand-counted vote percentage compared to his machine-counted vote percentage, whereas his nearest competitor (Buttigieg in New Hampshire, Biden in Vermont and Massachusetts) performed worse in in the hand counted townships. In all three states, the Sanders percentage difference between the hand-counted vs. machine-counted votes was statistically significant at the p < .0001 level. In New Hampshire and Massachusetts, the p value was statistically significant at the p < .00001 level.
When the analysis was confined just to small townships, Sanders also demonstrated a significant difference between hand- vs. machine-counted townships in Vermont. The margin for the hand-counted vs. machine counted townships was 3.0% for townships with less than 300 total votes (p = .0002) and 2.5% for townships with 300–499 total votes (p = .005). In New Hampshire, no statistically significant difference was demonstrated between the hand- vs. machine-counted percentages for the Sanders vote.
In Massachusetts, Warren also performed significantly better in hand- vs. machine-counted townships — by 2.8%, which differed from Biden’s negative margin by 5.1%
1. Positive numbers denote higher percentages for hand- vs. machine counted townships, and vice versa.
2. Percentage is from 2-person vote of Sanders and Buttigieg
3. Percentage is from total vote
4. p < .00001
5. p < .0001
Interpretation/Discussion
Sanders demonstrated highly statistically significant better performances for hand-counted vs. machine-counted voting townships, versus his nearest competitors (Buttigieg in New Hampshire, and Biden in Vermont and Massachusetts) in all three states. This supports the exit poll findings of Sanders under-performing in the official vote count compared to what was predicted in the exit polls (in all three states), by suggesting manipulation of the electronic vote count against him.
There could have been factors other than vote manipulation that accounted for the differences in Sanders’ worse performance in the machine-counted townships. The machine-counted vs. hand-counted jurisdiction are likely to differ in many ways. The most obvious way that they are likely to differ is in population density. If that is the main difference, then Sanders’ underperforming in the machine-counted jurisdictions is even more significant, because results from the 2020 Iowa caucus demonstrated a much better Sanders performance in Iowa’s high population centers compared to rural areas. Sanders won by landslide proportions in all of the highest population density jurisdictions, whereas Pete Buttigieg beat him soundly throughout most of the rural areas. That would predict that Sanders would have done much worse in the rural areas of Vermont and New Hampshire, compared to the cities, but instead, Sanders did better in the hand-counted jurisdictions, which were overwhelmingly rural. It seems difficult to think of a reason for this other than electronic vote manipulation — which would have occurred predominantly in relatively high population density areas, since the low-density areas were much more likely to count their votes by hand. That would also explain why Sanders’ relative underperformance in the machine counted jurisdictions in Vermont increased when the analysis was confined to comparison of more similarly sized voting jurisdictions.
None of this constitutes proof of election rigging in any of these states. There could be benign explanations for this that I cannot think of. But certainly these findings are highly suggestive of rigging against Sanders when viewed in the context that these results are all highly consistent with the massive red shifts in exit polls shown in so many primary states in both 2016 and in 2020 — not to mention abundant other evidence of cheating, such as massive voter suppression in Sanders areas and a clearly fraudulent audit in Chicago. In total, between the primary seasons of 2016 and 2020, there have been 23 statistically significant red shifts and only 1 statistically significant blue shift, when comparing Sanders with his various opponents. The odds against that happening by chance are astronomical.
It seems to me that a country that calls itself a democracy should be willing and eager to recount paper ballots by hand in elections that are highly suspicious and important. Our news media often notes the importance of having paper trails available for our elections. But, as noted by the Brennan Center for Justice, “Paper records of votes have limited value against a cyberattack if they are not used to check the accuracy of the software-generated total to confirm the veracity of election results”. We will never know for certain whether these elections were rigged or not (or the 2016 primary elections, though the evidence from both exit polls and other sources was overwhelming), unless and until hand recounts of paper ballots in machine-counted jurisdictions are performed.
Dale Tavris, MD, MPH