Oct. 15, 2024

For many years we, and others, have been analyzing “Unadjusted Exit Polls” (UEP) for US Elections.  These are exit polls that result from asking voters who have already voted who they supported. A state’s UEP results are shown by the journalistic organizations sponsoring the exit poll at the time the state’s polls are scheduled to close.

These UEPs are distinct from the “Adjusted Exit Polls” (AEPs) that are widely reported in the US after UEPs are adjusted to match official vote outcomes hours after the polls close. It should be noted that this universal US practice of conflating UEPs with AEPs is not generally replicated in other countries.

This may be because UEPs reported before official results have been tallied have been highly reliable predictors of official election results all over the world but, for some reason not in the US since at least 2004.  To the contrary, UEP analysis of past US Elections since at least 2004 have shown a consistent “red shift” pattern whereby official election results exceed UEP results for Republican candidates and fall short of UEP results for Democratic candidates.

Mainstream commentators have for many years erroneously claimed that this is a result of “reluctant” Republican exit poll response. But detailed analysis of exit poll data provided by those conducting the 2004 exit poll has demonstrated that Republicans were not more reluctant than Democrats to provide exit poll responses.  

Our UEP analysis of the 2022 US Senate race results is based, as usual, on a comparison of official vote counts with UEP screenshots captured at the close of polling and sample sizes published by mainstream media outlets. The exit polls also include polls of voters who cast absentee ballots or who cast ballots by mail.  

As in the past, we used a standard two-population statistical technique (share of votes, compared to share of non-votes, for the Democratic and Republican candidates). This includes a 30 percent increase in the computed standard deviation to take into account possible polling-site clustering by candidate preference (referred to  in the literature and tables 1-2 as a “cluster factor”).

Table 1 below shows how likely the Official Vote Count (OVC) is, given the UEP, for Democratic Senate candidates in competitive 2022 US Senate races for which exit polling was conducted.

The third column from the right shows that in one state, Florida, there is less than a 1 percent chance that the UEP vote share for the Democratic Senate candidate could be small as it is, if the UEP is assumed to be accurate.

The second column from the right shows the opposite for the Democratic candidate in Georgia, i.e. that there is less than a 1 percent chance that the OVC vote share for could be as large as it is given the UEP, and the final column shows that there is less than a 5 percent chance that the OVC vote share for the Democratic candidate in Nevada could be large as it is given the UEP vote share in that state.

So, there are statistically significant deviations (at the 1 percent or 5 percent levels) of the OVC from the UEP against the Democratic Senate candidate in one state, and in favor of Democratic Senate candidates in two states.

Table 2 below shows how likely the Official Vote Count (OVC) is, given the UEP, for Republican Senate candidates in competitive 2022 US Senate races for which exit polling was conducted.

The fourth column from the right shows that in one state, Florida, there is less than a 1 percent chance that the OVC vote share for the Republican Senate candidate could be large as it is, if the UEP is assumed to be accurate.

The third column from the right shows that in another state, Ohio, there is less than a 5 percent chance that the OVC vote share for the Republican Senate candidate could be large as it is, if the UEP is assumed to be accurate.

The second column from the right shows the opposite for the Republican candidate in Nevada, i.e. that there is less than a 1 percent chance that the OVC vote share for the Republican candidate could be as small as it is given the UEP, and the final column shows that there is less than a 5 percent chance that the OVC vote share for the Republican candidate in New Hampshire could be as small as it is given the UEP vote share.

So, there are statistically significant deviations (at the 1 percent or 5 percent levels) of the OVC from the UEP favoring the Republican Senate candidate in two states, and in against the Republican Senate candidates in two states.

Conclusion

In the 2022 US Senate races in battleground states, there were significant OVC deviations from UEPs favoring Republican Senate candidates in two states: Florida and Ohio, and favoring Democratic Senate candidates in three states Georgia, New Hampshire and Nevada.

This is a marked departure from the consistent pattern of battleground state OVC deviations from UEPs favoring Republicans, or “red shift” patterns, exhibited in prior US elections since at least 2004. Unlike in prior US elections, this more random pattern of OVC deviations from UEPs is closer to what would be expected for more accurate vote counting and exit poll analysis. But it could also result from changes in UEP estimation methodologies that have increased UEP Republican shares to overcome the consistent “red shift” pattern of prior elections, without significant changes in OVC tallying procedures relative to prior elections in the US.

At least one report on 2022 pre-election polling suggested that there were “efforts to adjust the polls to reflect the expected Democratic bias,” or to eliminate an expected “red shift.” Similarly, changing exit polling methodology this way could result in random deviations of UEPs from OVCs even if OVC vote counting had not significantly changed.  This would in effect mean that UEPs estimated before polls close were less “unadjusted” than they have been in the past.    

However, this new pattern could also reflect more accurate US vote counting without significant changes in UEP estimate methodologies, perhaps because of more intense focus on US vote counting due to the Trump campaign claim that there was large scale “blue shift” in the 2020 US Presidential election (without any factual evidence such as UEPs that  consistently suggested the opposite in the 2020 and earlier federal elections).

Without more transparent and public documentation of UEP exit-polling methodologies we are unable to determine whether the 2022 pattern of more random OVC to UEP deviations are a result of significant improvements in US vote counting or adjustments in US exit-polling vote share estimation. (We have over the years contacted the major US exit polling company, Edison Media Research, and before that Mitofsky International, repeatedly to obtain more detailed information on exit-polling methodology, without success.)