Who are Americans expecting to win in the battleground states?

In our recent polling carried out in the United States on behalf of John C. Hulsman Enterprises, we make no attempts to predict the outcome of the November election. We set out primarily to understand how people in the US felt about other countries around the world, and US foreign policy in general. The results of that polling can be found here.

Alongside those foreign policy questions, we also asked respondents who they thought, regardless of their own preference, would win the key battleground states in the 2020 presidential election. For each of the 12 most hotly contested states, participants were asked to pick whether they expected a Republican, Democratic or Independent victory in the state. “Don’t Know” was excluded as an option in order to force people to give a choice. 

Taken together, the overall “prediction” was the following electoral map, suggesting a Joe Biden victory in the electoral college by 278 to Donald Trump’s 260.

(Non-battleground states in accordance with results from 2016; likely “Don’t Know” responses translated into people suggesting an Independent victory, most likely out of an unwillingness to commit to a prediction. These have been excluded from the final figures. Full tables available here)

Digging deeper into the results – the two closest states in terms of predictions were Ohio and Arizona, where the difference between participants expecting Trump to win over Biden is only 2% and within margin of error. 

The results also show a clear degree of “wishful thinking” from Democrats and Republican voters, towards their own side’s chances. Overall Trump supporters predicted a Republican victory sweep across all the battleground states, while Biden supporters were bullish about Democrat chances everywhere except Texas. 

However even if we remove those who respondents who picked identically for all battleground states (i.e. those who are overly confident on both Democrat and Republican chances in these key states), the only change is to move Ohio and Arizona into the ‘complete toss-up’ category with equal numbers suggesting a Democrat and Republican victory. 

So how well do the American public’s predictions line-up with the high profile election forecasts? In general it seems the public are expecting Trump to do better than the polls/forecasts suggest.  Potentially unsurprising given the (largely unfair) perception in 2016 that the polls got it wrong.

(Forecasts: 538, YouGov)

Notably, the public expects Trump to hold onto North Carolina, Florida, Arizona and Georgia despite the forecasts leaning in the opposite direction. However the public does expect Trump to lose the blue wall of Pennsylvania, Wisconsin and Michigan which propelled him to victory in 2016. 

Taking all this into account it would look like the results that might be most surprising to those in the US are, if forecasts are correct, those in North Carolina, Florida, Arizona and Georgia: the sort of states that would give Joe Biden a comfortable electoral college win. Equally, If Trump manages to hold on to any of the rust belt seats that delivered his victory in 2016 then it’s fair to say both forecasters and the public alike are set to be surprised. Equally if Ohio, Iowa or Texas was to turn blue in a week’s time in a Biden landslide, the upset would be in the opposite direction, (although likely less high profile given the overall Biden victory was in line with expectations).

Do questions like this have any predictive capacity?

Firstly, this is in no way an attempt to predict the outcomes of the election in these states. However, “citizen forecasts” can be somewhat informative about election outcomes, although the research is somewhat mixed depending on how the predictions are transformed into forecasts.

As mentioned these forecasts are subject to a large amount of wishful/partisan thinking. In the latest The Economist/YouGov poll, which regularly asks on a national level who the US public is expecting to win the election, the overall results show a relatively neck-and-neck expectation at the top level (with 41% expecting Biden, and 39% expecting Trump). However, examining the results by vote intention, 74% of Biden voters expect Biden to win, and 83% of Trump voters expect Trump to win. One interesting aspect to this question was that the addition of a “Don’t Know” option means that about 20% place themselves in this group, and a slightly higher proportion of Biden supporters express uncertainty on this point.

One of the possible reasons for this is the consumption of partisan media. A study in 2018 by Searles and colleagues found preliminary evidence that coverage from partisan media that shows the “in-party” set to win the election leads to a greater level of prediction of in-party victory among those who consume that media, and the opposite leads to an adjustment downwards of electoral expectations. A whole range of other aspects are likely to be having an impact beyond this though in 2020: concerns about court battles and voter suppression, as well as the slightly abnormal nature of Trump’s 2016 victory in the electoral college but not the popular vote. Equally, with the pandemic having an impact on ‘the ground campaign’, as well as on the ways in which people are voting this year round, making a “prediction” has become even more difficult than usual.

Despite all this, in an extensive piece of research in 2014, Graefe makes a clear argument for a greater use of voter expectation methods in election forecasting, pointing out that across over 200 surveys analysed, 89% had the majority of respondents correctly predicting the outcome of the election. Equally, the error of these forecasts was found to be considerably lower than the error with traditional intention polling in the final days prior to the election. For this reason, Pollyvote incorporates citizen forecasts into its election prediction model, using an equation to translate these results into a forecast of final vote shares. This equation smooths the results into more plausible vote shares (even if 0% of the sample predicts a Trump victory, the vote share forecast would bottom out at around 41%). Unfortunately this would not apply for state-level predictions, as the smallest vote share it is plausible for a party to achieve in a state would vary hugely.

Without transformations like this, these findings are not particularly helpful in showing the potential of each candidate. When we look at the final predictions in the The Economist/YouGov poll in the run up to 2016, 51% expected Clinton to win, and only 27% expected Trump to win. The 83% expectation of a Clinton victory among Clinton supporters likely goes some way to explaining why there was a widespread perception that the forecasts and analysis completely missed the mark. Equally the interpretation of “winning” is difficult here given Clinton won the popular vote. What is perhaps most interesting is the impact that this perceived upset had on the way US citizens now think about elections. The anticipation of unexpected (or counter-forecast) upsets has clearly grown: Trump supporters now, while reduced in number in the polls than in 2016, are much more confident in a Trump victory, with 59% expecting a Trump win in 2016 up to 83% now. 

Biden supporters, on the other hand, have not shifted too much in the confidence which they had for Clinton in 2016. This confidence is largely more justified this time round, with the polls demonstrating a much clearer lead for Biden than they did for Clinton at this stage, and the possibility of a Trump victory requiring either a major last minute development or a large and systematic bias towards Biden across the vast majority of pollsters. 

In Conclusion

The use of citizen forecasts to predict election outcomes is not without successes, and I expect the analysis from Pollyvote of how close this method was to predicting the outcome to be interesting. In short though, the indication of polling of this sort this year round is that both of the candidates supporters are confident in their choice’s chances, setting it in stark contrast to 2016 when the expectations of Trump’s supporters were relatively muted.  Looking at state level predictions shows where the biggest upsets could occur; if recent polling on Texas is accurate then there is a chance for an unexpected Biden win in the state, and on the (very unlikely) flipside if there was widespread error in the polls which is giving Biden a boost in key “rust-belt” states, the Democratic supporters are no more prepared for an upset than they were in 2016.