As the election approaches, I notice reminders that the polls aren’t all that accurate. The problem is that the samples selected for the polls somehow don’t perfectly represent the voting public. Part of the misrepresentation is just demographics – if we look at California, 46.8% identify as Democrats and 23.9% as Republicans. In San Francisco, 85.3% voted Dem in the last presidential election. In Kern County (Bakersfield) 53.9% voted Republican. The populations are set up in a way that can lead to biased samples. That’s not a big deal – the problem is figuring out what the results actually mean.
In this chart, the folks at RealClearPolitics have tried to develop correction factors to deal with polling inaccuracies – showing where past voting records and polls have disagreed, and developing adjustments to better project the results. The trick is to control for the discrepancies – the task isn’t a whole lot different than population projections.
|State||RCP Average||2020||2018||2016||Adjusted 10/21 Poll Average||RCP Projection November 8||Polls Under-Estimated|
|Penn.||Fetterman +2.4||D +5.6||D +16.0||D +6.2||GOP 4.8||Oz +2.4||GOP Hold|
|N.H.||Hassan +5.4||D +11.0||–||D +8.0||GOP 5.8||Bolduc +0.4||Dem Hold|
|Nevada||Laxalt +1.2||D +5.2||R +1.7||D +4.2||Dem 0.8||Laxalt +0.4||GOP Pick Up|
|Arizona||Kelly +2.5||D +4.0||R +0.3||D +1.3||GOP 2.0||Kelly +0.5||GOP Pick Up|
|Georgia||Warnock +2.4||D +2.0||–||R +4.0||GOP 1.4||Warnock +1.0||12/6 Runoff|
|N.C.||Budd +2.8||D +2.7||–||D +2.5||GOP 5.1||Budd +7.9||GOP Hold|
|Florida||Rubio +4.7||D +1.4||Tie||D +4.0||GOP 3.4||Rubio +8.1||GOP Hold|
|Wisconsin||Johnson +2.8||D +6.3||D +10.6||D +7.0||GOP 4.4||Johnson +7.2||GOP Hold|
|Ohio||Vance +2.5||R +0.5||D +16.0||R +0.6||GOP 8.3||Vance +10.8||GOP Hold|
As we look at Pennsylvania, we see Fetterman polling 2.4% above Oz – but the next 3 columns show that historically Dems have polled high, averaging 4.8% more on the polls than in the votes. That correction factor leads them to project Oz the winner by 2.4%.
In New Hampster Maggie Hassan leads by 5.4% – but the 2020 and 2016 voting results suggest that Dem polls show up 5.8% high. The next column shows a correction where Republican Bolduc is up by 0.4% – yet RCP projects a Hassan victory. I suspect the model includes an adjustment for the incumbency advantage.
Georgia’s requirement for a runoff election if neither candidate gets 50% plus one of the total vote shows up in the projection of a runoff on December 6.
Here in Lincoln County, where we don’t have two candidates to pick between, we don’t have to develop a model to project which candidate will win the election. Personally, I’d rather have a poor choice than no choice – particularly when following the law (subsection 7) would provide us a choice.