- N.Carolina flips Democrat
- Why are opinion polls so inaccurate?
NORTH CAROLINA FLIPS BLUE
Our 2020 election monitor is now showing Joe Biden extending his lead to 108 electoral college votes overnight from 78 previously.
The change comes as a result of a 3.4% swing to Biden in North Carolina which is worth 15 votes.
If the election were held tomorrow Biden would win by 323 to 215 electoral votes, with an overall 95% probability of winning.
Despite the swing in North Carolina, Biden is still only projected to win by a small margin there, of only 52.27%. Biden’s advantage may be temporary and the state normally votes Republican with a Democrat candidate only having won once in the last 10 elections.
A rising trend in news sentiment for Biden in N.Carolina during late August is probably the reason for the state flipping blue.
More recently sentiment has started to converge for the too candidates suggesting a risk the state could turn red again.
Overnight sentiment fell by 1.1% for Joe Biden, for example, although overall he still enjoys quite a high sentiment rating of 60.3% versus Trump’s 39.7%.
Another input into the model, Media Attention, which measures the proportion of news about each candidate, was also rising for Biden during late August, and is probably a contributing factor to the headline change.
Media Attention has been waning for Biden in September, however, after actually falling by 0.6% overnight, and Trump remains in the lead with 59.7% versus Biden’s 40.3%, so again there is a risk of a flip back to Republican if this new trend continues.
POLLS VS PROJECTIONS
Generally, when the monitor’s projections and poll ratings are in agreement it reinforces their projection. In the case of North Carolina both are now pointing to a Biden victory, although only marginally: in the case of polls Biden is leading by only 1% with 47% of the vote versus Trump’s 46%.
A neat new feature on the monitor means the two can now be compared at a glance by simply switching the “Polls” toggle situated below the interactive map to ‘on’, and then clicking on the state.
Polls have come in for a lot of criticism of late, and it is easy to see why. It is possible to view 2016 results on the monitor which are showing in that election Hillary Clinton led by 47.5% to 46.9% in the polls, even though Donald Trump went on to win the state.
Using another feature on the monitor enables users to compare the model’s projections in 2016 with 2020, again this can be activated simply by switching on the toggle below the interactive map.
At the moment it is showing that in 2016 the model projected Trump to win with a probability chance of 49.8% versus Clinton’s 46.2%, proving as usual that when the two disagree, it is usually our model which is the more accurate!
TRUMP SENTIMENT RISING
It is worth noting that although the monitor headline figures point to Biden extending his chances of winning, other significant news analytics metrics suggest Trump may not be doing as badly as projections perhaps suggest.
The RavenPack news analytics platform is showing overall news sentiment for Trump has been rising and closing the gap with Biden over the last 3 months.
A glance at our platform’s Treemap tile charts below reveals the areas which might be responsible for Trump’s recovery.
The first chart shows the influence of different issues in Trump’s news sentiment. The larger the Tile and the more intense its color, the greater the influence on sentiment; green signifies positive and red negative sentiment.
Below shows the average monthly sentiment tile chart for Trump followed by the weekly sentiment tile chart - comparing long and short term influencing factors.
Note how the negative influence of news about “legal” has fallen in size over time, whilst news about “Government” has turned positive - suggesting these could both be sources of improving sentiment for the President.
What news could be driving the improvement in the spheres of legal and government?
Drilling down it is possible to find the stories driving sentiment, and these tell us that a month ago negative sentiment for ‘legal’ was caused by Chinese social media giant TikTok suing Trump for banning it from operating in the U.S.
More recently, legal issues have been dominated by Biden’s threat to take legal action over Trump’s handling of racial tensions in Wisconsin, and Eddy Grant’s threat to sue the president for using the song “Electric Avenue” in a campaign video, however, as a whole the subject has reduced in significance.
The rise in positive ‘Government’ sentiment, meanwhile, appears to be associated with Trump’s campaign visit to North Carolina - yet it doesn’t ironically appear to have been enough to alter the result of our monitor!
WHY POLLS ARE SO HIT & MISS
Polling as a method for forecasting the result of elections appears to have fallen out of favor after a catalog of high-profile failures, including the 2016 U.S. presidential election, UK’s Brexit referendum, and the 2015 UK general election.
Whilst news analytics-based models, such as those on which our 2020 monitor is based, have done a better job at predicting election results, opinion polls remain the primary tool for election forecasting, and the question remains why this previously highly-respected method has become so unreliable?
According to research following the failure in the U.S. 2016 election and a report commissioned in the wake of the 2015 UK election, there are three primary reasons why polls have failed to accurately predict election outcomes.
The main general reason is because the people who take part in polls are not representative of the people who eventually turn out and vote, and this mismatch causes forecasting errors.
This can be because of a range of factors, however, in the U.S. 2016 election, the primary culprit was what pollsters refer to as ‘nonresponse bias’. This occurs when certain kinds of people systematically do not respond to surveys despite equal opportunity outreach to all parts of the electorate.
For example, the reason why Trump’s supporters may have been unrepresented could have been because the demographics they belonged to might be less likely to respond to a request to be interviewed in a poll.
“We know that some groups – including the less educated voters who were a key demographic for Trump on Election Day – are consistently hard for pollsters to reach,” says a report by the Pew Research Centre.
Trump’s populist anti-establishment appeal may also mean he appeals to segments who show frustration and anti-institutional feelings, and these can be the sort of people who are less likely to engage with pollsters.
In the case of the 2015 UK election, the problem seems to be have been a failure of polls to adequately reflect the size of the Conservative victory because of “unrepresentative samples”, according to the report into the subsequent inquiry
Historically and in 2015, UK polls consistently over-represent labor support, however, it was less of an issue in the past since they usually correctly identified the winning party.
A second key reason for why polls are inaccurate is because of the difficulty of pollsters to adequately predict voter turnout and therefore sample correctly.
Pollsters try to create models to predict how many and what sorts of people will actually turn out to vote on election day. They then use these models to design representative samples for the purpose of polling.
The problem is that there is no reliable way to predict how many and what types of people will eventually turnout so sampling is always a source of error.
Although voters interviewed are asked to predict their chances of turning out, and this is used to assess their likelihood of eventually voting, what people say one day is not always reflective of what they end up doing, and many change their mind on election day.
Using past election data is also problematic since the types of people who turnout change from one election to another and depend on the breadth of appeal of the candidates.
A final reason for the inaccuracy of polls is the idea of so-called ‘shy Trumpers’ or ‘closet Brexiteers’. These are people who don’t like admitting they want to vote for Trump or Brexit because it is seen as somehow ‘socially undesirable’.
Yet whilst this may seem intuitive, there is little hard evidence to support the notion that this is a major influence. Politico and Morning Consult conducted an experiment to see if ‘shy Trumpers’ might account for the failure of polls in 2016 but found that overall, there was little indication of an effect, though they did find some suggestion that college-educated and higher-income voters might have been more likely to support Trump online rather than admit to it over the phone.
Something missing? Tell us what you’d like to see in your US Election updatesSubscribe to US Election Updates