October 24, 2022

Presidential Election in 2020

Unadjusted Exit Poll (UEP) analysis of the 2020 US Presidential and Senate elections shows the same “red-shift” pattern of Republican favoring discrepancies from Unadjusted Exit Polls (UEPs) particularly in battleground states that has prevailed in every general US Presidential election since 2004.[1]

UEPs are samples of voter responses taken after they vote in-person or by absentee ballot. These are obtained from screen shots of exit polls reported by US media right before, or soon after, polls close, and include UEP candidate vote shares and sample sizes. These differ from the exit polls that are adjusted to match official election results, or Adjusted Exit Poll (AEP) candidate vote shares and sample sizes, that are widely reported in US media.

In the United States, an adjusted exit poll is created by the exit pollsters after the polls have been closed for several hours. The adjustments from the unadjusted exit poll to the adjusted exit poll are based on the assumption that the reported results must accurately reflect how the ballots were cast. When sending observers to watch and report on foreign elections the State Department does not make this assumption. Whether or not intended, in the US the AEP serves to obscure the meaning of UEP when the official vote count lies outside the margin of error of the unadjusted exit poll. The meaning is that the official vote count may not accurately state how the voters voted.

In foreign nations, when a candidate seeking election pays for all versions of an exit poll (unadjusted and adjusted), the candidate is free to obtain an AEP that more closely matches the reported vote count than does the original UEP. When sending observers to watch a foreign election, the U.S. State Department pays for one exit poll - an Unadjusted Exit Poll. The State Department does not pay for an AEP. The State Department observes the foreign election for the purpose of checking on the officially reported results and not for the purpose of legitimizing an inaccurate vote count.[2]

These UEP and AEP data were captured for all 22 states, and the nation, nine General Election Senate races, and the two Georgia run-off Senate races, where exit polling was conducted. Screen shots with Presidential UEP shares and sample sizes can be found here.[3] UEPs are calculated from gender vote shares except for Kentucky where regional weighted average vote shares were calculated, as shown in table displayed in the link above. Presidential VCs can be found here. National Presidential AEP and sample sizes can be found here. State Presidential AEPs and sample sizes can be found by searching NYTimes 2020 Presidential exit polls for each state. For example, Presidential AEP with sample size for Kentucky can be found here.

For each of these races Republican Vote Counts (VCs) are compared to Republican UEPs, and Democratic VCs to Democratic UEPs. In almost all cases Republican VCs are higher than their UEPs and Democratic VCs lower than their UEPs, leading to a “red shift,” or Republican minus Democratic VC differences that are larger than Republican minus Democratic UEP differences

A statistical analysis of Trump VCs minus UEPs for 22 states and the nation for which UEPs were conducted is presented in Table 1 below. Note that in Table 1 the ratios of UEP/AEP sample sizes are 93% or more for 20 of 22 states, and 96% or more for 18 of these states. Exceptions are SC 64% and WI 78%. This suggests that, except for SC and WI, UEPs are from samples that are almost identical, or very close, to AEP samples. Table 1 also shows that the national UEP sample is also very close (91.3%) to the size of the complete AEP sample which is 15,590.

Table 1 below shows Trump favoring UEP discrepancies, or VC vote share greater than UEP vote share (VC – UEP > 0), in 21 of the 22 states. The statistical odds of an error in the same direction in 21 out of 22 cases is one out of 4,194,304.

Table 1: US 2020 Presidential General Election Trump Unadjusted Exit Poll Analysis (see attached file)

Table 1 shows that in all 8 of the states (AZ, FL, IA, KY, ME, MT, NV, and WI) where the VC – UEP discrepancy was statistically significant (or outside of the statistical margin of error at the standard 95% Confidence Interval (CI)), this discrepancy was a “red shift” for Trump. Five of these states (AZ, FL, IA, NV, and WI) were “battleground” states. These highly statistically significant discrepancies, particularly in in IA and WI where Trump Vote Count shares had odds of less than one in 10,513 and 431,627, respectively, of occurring through random chance, strongly suggest possible vote miscount and should be forensically investigated. Trump’s national VC red shift of 2% is also well outside of the UEP 95% margin of error and thus highly statistically significant, with odds of less than one in 8,076 of occurring through random error.

A similar statistical UEP analysis of UEP discrepancies against Biden in Table 2 below shows that Biden got a lower VC than UEP share in all but one (CA) of the 22 states for which exit polls were conducted. Again, the statistical odds of an error in the same direction in 21 out of 22 cases is one out of 4,194,304.

Table 2: US 2020 Presidential General Election Biden Unadjusted Exit Poll Analysis (see attached file)

Table 2 shows that in all four of the states (IA, KY, MI, and WI) where the UEP – VC < 0 discrepancy was statistically significant, or outside of a 95% margin of error or Confidence Interval (CI), this discrepancy was a “red shift” against Biden. Of these 4 states, 3 (IA, MI and WI) were “battleground” states and in all three the shift against Biden was highly significant with odds of less than 1 in 2,523, 542, and 365 respectively. Table 2 also shows a national (USA) red shift of 1.9% against Biden that is well outside of the UEP 95% CI and thus highly statistically significant, with odds of less than one in 5,795 of occurring through random error.

Fortunately, in this election (unlike in 2004 and 2016, see references above) these red-shifts for Trump and against Biden did not change the final outcome of the national election, though Biden’s UEP share was larger than Trump’s in two additional states (IA and NC), and in IA the Trump favoring and Biden disfavoring UEP-VC discrepancies were both highly significant with odds of less than one in 10,513 and one in 2,523 of occurring because of random error. 

More generally this pattern of consistent pro-Republican UEP – VC < 0 disparities (like similar red-shift patterns in prior US elections) strongly suggests politically motivated vote miscount and should be forensically investigated. This is particularly the case for states where these discrepancies are statistically highly significant or highly improbable - for example in WI and IA, where Trump VC shares had odds of less than one in 431,627 and 10,513, respectively, of being so much larger than UEP shares, and Biden VC shares had odds of less than one in 19,700 and 2,523, respectively, of being so much smaller than UEP shares, because of random error.[4]

Similarly suspect is the fact that in the national (USA) UEP Biden’s margin of victory was almost 3% larger than in the official VC (8.4% rather than 5.5%), suggesting a popular vote win by more than 13 million rather than 8.5 million votes.

Senate Elections in 2020

UEP and AEP Senate data were captured for nine states (AL, GA, IA, KY, ME, NC, NH, SC, TX), covering nine General Election Senate races and the two Georgia run-off Senate races, where exit polling was conducted. Screen shots with Senate UEP shares, and sample sizes for states other than AL, IA, ME, NH, and TX can be found here.[5] Also, again, UEPs are calculated from gender vote shares except for Kentucky where regional weighted average vote shares were calculated as shown in table displayed in the link above. The AL Senate UEP sample size is estimated to be equal to the AL Senate AEP sample size. IA, ME, NH, and TX Senate UEP sample sizes are estimated to be equal to Presidential UEP sample sizes for these states. State U.S. Senate VCs can be found here. State U.S. Senate AEPs and sample sizes can be found by searching “CNN election 2020 exit polls” for each state. For example, the AEP and sample size for the 2020 Alabama Senate race can be found here.

Again, for each of these races Republican U.S. Senate Vote Counts (VCs) are compared to Republican U.S. Senate UEPs, and Democratic U.S. Senate VCs to Democratic U.S. Senate UEPs. As with the Presidential election in most cases U.S. Senate Republican VCs are higher than their UEPs and U.S. Senate Democratic VCs lower than their UEPs, leading to a “red shift”, or Republican minus Democratic VC differences that are larger than Republican minus Democratic UEP differences.

Table 3: US 2020 General Election Republican minus Democratic Senate Unadjusted Exit Poll Analysis (see attached file)

Table 3 displays a statistical analysis of Senate Republican VCs minus UEP discrepancy for 11 Senate races in 10 states for which UEPs were conducted, and for the 2 later Georgia runoff races. Note that in the table the ratios of UEP/AEP sample sizes are 95% or more for all except the ME Senate race where this ratio is 79.2% again suggesting that UEP samples are in almost all cases very close to complete AEP samples.

As in the prior Tables, most of the UEP discrepancy is red shift, or higher VC than UEP, for Republican U.S. Senate candidates. There is red shift in 8 of the 10 General Election races (not Georgia Runoff) in the top 10 rows of Table 3. The exceptions are the GA-Warnock/Loeffler and NH Senate races. In four of these races (AL, GA (Ossoff/Purdue), KY, and SC) the UEP discrepancy is statistically significant (outside the 95% CI). Only one race, the GA-Warnock/Loeffler race, shows a statistically significant “blue shift,” or Loeffler receiving fewer votes than her UEP. However, this one case of statistically significant blue shift (2.9% against Loeffler) in all of the races for which UEPs were taken in the 2020 elections, is negated by the more than offsetting red shift (5.1% for Loeffler) against Warnock displayed in Table 4 below.

Table 4: US 2020 General Election Democratic minus Republican Senate Unadjusted Exit Poll Analysis (see attached file)

Table 4 depicts U.S. Senate Democratic VC – UEP < 0 discrepancies 2020 for U.S. Senate Democratic candidates.

Again, almost all of the discrepancies in the General Election (9 out of 10) show red shift, the lone exception (NH) shows statistically insignificant blue shift. In this case all six of the statistically significant discrepancies (AL, GA (for Loeffler), IA, KY, NC, and SC) are red shifts. Moreover, the statistically significant red shift against Cunningham (for Tillis) in NC suggests that Cunningham may have won when the voters marked their ballots as Cunningham had a larger UEP Vote Share than Tillis.

Finally, Tables 3 and 4 both show blue shift for Ossoff and Warnock in the GA Senate runoff races, but as these blue shifts are in all cases within the 95% CI they are more likely to be a result of random statistical (or other) UEP error.

This pattern of consistent pro-Republican UEP – VC discrepancies (like similar red-shift patterns in prior US elections) strongly suggests politically motivated vote miscount and should be forensically investigated, particularly in cases where the discrepancies are statistically significant and may have changed the final election outcome as in the Senate race in NC.

About the authors:

Ron Baiman is an Associate Professor of Economics teaching courses in economics and statistics at Benedictine University in Lisle, Illinois.

Peter Peckarsky is the National Director of the Election Software Project of the Columbus Institute for Contemporary Journalism. He was a 2022 candidate for the Democratic nomination to be a U.S. Senator from Wisconsin..

Jonathan Simon is Executive Director of the Election Defense Alliance.

[1] See Baiman 2017: U.S. 2016 Unadjusted Exit Poll Discrepancies Fit Chronic Republican Vote-Count Rigging, not Random Statistical Patterns: https://www.opednews.com/articles/U-S-2016-Unadjusted-Exit-by-Ron-Baiman-2016-Elections_Exit-Polls-161208-153.html , Updated, Expanded and Corrected Affidavit Version: U.S. 2016 Unadjusted Exit Poll Discrepancies Fit Chronic Republican Vote – Count Rigging, not Random Statistical, Patterns: https://www.researchgate.net/publication/319205877_Updated_Expanded_and_Corrected_Affidavit_Version_US_2016_Unadjusted_Exit_Poll_Discrepancies_Fit_Chronic_Republican_Vote_-_Count_Rigging_not_Random_Statistical_Patterns?channel=doi&linkId=599b2d5ca6fdcc500349b9a5&showFulltext=true ; Dopp et al. 2005: History of the Debate Surrounding the 2004 Presidential Election ; and Hartmann 2020: The Hidden History of the War on Voting, p. 87-93, for more background.

[2] https://columbusfreepress.com/article/why-united-states-state-department-would-not-certify-trump%e2%80%99s-election-legitimate

[3] Note: Zoom-in magnifying function is required to make out some of the data displayed in this pdf.

[4] A documented example showing that miscount may have contributed to changing the final outcome of the 2004 Presidential race can be found here: https://freepress.org/article/official-states-electronic-voting-system-added-votes-never-cast-2004-presidential-election

[5] Note, again: Zoom-in magnifying function is required to make out some of the data displayed in this pdf.