In 2016 a referendum and a national election both defied their respective polling consensus, as rising nationalist sentiment and economic disenfranchisement broke the models of British and American political opinion polling. On June 23, 2016, 52 per cent of United Kingdom (UK) voters chose to leave the European Union (EU), driven by a desire to “take control” back from the political and economic union. Subsequently, on November 8, 2016 President Donald Trump won the United States (US) presidential election with 304 of 538 electoral votes, pledging to “make America great again”. These were not the first elections to defy pollster predictions, but their underlying dynamics beg the question: how can we forecast rising nationalism?
Why the Polls Were Wrong
Poll-based forecasts like FiveThirtyEight’s use historical models of voting patterns and samples of current electorate opinion (the polls themselves) to predict who will actually come out to vote, and which candidate they will choose. A glaring weakness of this method was exposed in recent elections: if actual voting patterns differ widely from historical ones, here due to the dynamic of nationalism, they will be underestimated by the model. This was true for the US election, where models greatly underestimated turnout of Republican voters and overestimated turnout of Democratic voters, while models greatly overestimated youth turnout in the UK referendum.
Beyond their dependence on historical models, there are inherent weaknesses to making predictions based on polls. These forecasts are plagued by uncertainties that make it difficult to trust the results of any poll or aggregation. Many pollsters mourn the loss of ubiquitous landlines, which made it relatively easy to get a representative sample of a state’s population. Forecasters like Allan Lichtman, who has accurately predicted the result of the last nine US presidential elections, criticize poll-based forecasts for basing predictions on polled “snapshots” of electorate opinion. The predictive power of polls is therefore limited by the uncertainty of changing opinion in the future. Forecasting guru Nassim Taleb, famed for foretelling a financial “black swan” preceding the 2008 financial crisis, argues that this uncertainty is so significant that election predictions should not stray far from 50/50 until vote counting begins.
The Earthquake Method: Stability vs. Upheaval
Lichtman has taken a different approach to predicting US presidential elections that was successful both before and during the rise of nationalism. His model is based on techniques used to predict earthquakes by assessing geophysical measures of stability and upheaval. The “13 Keys to the White House” are based on broad political and economic trends, and are used to assess the stability of the incumbent political party. If there are too many indications of upheaval, recorded as true-or-false values, the model predicts a change in leadership. While this model seems simple, it is based on Lichtman’s immense experience in American political history that informs careful selection and definition of the relevant keys and the threshold for upheaval. The benefit of his model is its generality; unlike poll-based models that are extremely fine-tuned to account for statistical biases, Lichtman’s model is resilient to confounding factors like nonresponse in polling or changing demographic support due to rising nationalism.
What Comes Next?
If anything is certain, it is that poll-based forecasts are uncertain. In many Western countries where globalism has been the incumbent paradigm for decades, the rise of nationalism is destroying the assumptions that these models rely on. However, this does not mean that the future itself is entirely unpredictable. Lichtman’s model of governmental stability might offer some insight for current governments facing destabilizing forces of nationalism. If economic and political trends are pushing for upheaval, it may be wise to listen and shake up the country on your own terms. Perhaps if previous UK Prime Minister David Cameron had expected upheaval he would not have offered such a clear opportunity through a referendum. Similarly, the democratic party may have had more success with a more radical presidential candidate like Senator Bernie Sanders.
For the average citizen, it is important not to put too much faith in any model or prediction. Data science may provide new tools for analyzing macro trends, but it is error prone in the absence of contextual understanding and informed judgement. Rather than smugly sharing or angrily denouncing the latest election predictions, remember that as far as anyone knows the vote is evenly split until election day. Instead, keep informed on the issues at stake, engage in discussion to test your views, and use that context to shape your predictions of what comes next.