Out of the Margins: US Election Polls and Policy

If the polls had been right about the US election, British businessman and Donald Trump supporter, John Mappin, would not have found himself USD 124,000 richer on 8 November. Mappin placed a series of bets in favour of Trump, putting money on an outcome that some pollsters had only given a two percent chance of occurring.

In the US, as elsewhere, polls play a crucial role in campaign coverage during election periods with many experts basing their predictions and analysis on them. Although there is little evidence to suggest that polls influence how people vote, they can imit political uncertainty, allowing for better future planning. For this reason, pollsters have endured significant derision following Trump’s victory in the November election, something few people – seemingly not even the Trump team themselves – were prepared for. The immediate aftermath of the election is evidence of this, with global stock markets as well as the Canadian immigration website, temporarily crashing. Many have criticised pollsters and pundits for incorrectly, and almost unanimously, predicting a Hillary Clinton victory. However, the picture is more complex and the limitations behind polling in the US are also linked to the broader structure of the political system.

One of the most important reasons why polls did not correctly predict the outcome resides not solely in polling methodology, but also in the structure of the US electoral system; namely, the existence of the Electoral College. This is important because many polls, especially the ones with more statistically sound methodology, are conducted nationally. In this sense, they more accurately correspond to the popular vote, which Clinton is set to win comfortably, as she holds over two million votes ahead of Trump. Most of the national polls therefore accurately reflected what they were measuring - the national vote - or at least did this within a normal margin of error.

The problem therefore lies more in the limited and poor polling at a state level, where Electoral College votes, and thus the presidency, are decided. These polls are relatively fewer in number and tend to have greater errors. To understand this dynamic, it is crucial to know how polls are modelled. Generally, they rely on two things: demographic composition of the electorate and predictions for how each group is expected to vote. Pollsters will give greater weight to individual voters, especially for groups that are less likely to respond, such as young Hispanic males. For example, if a pollster needs to poll 10 young Hispanic men to get a representative sample of a particular area but they only get seven responses, they will often give greater weight to those seven responses. Polling errors can stem both from getting an unrepresentative sample of respondents within each group, or from incorrectly predicting how many of each voter group will show up at the election.

The latter appears particularly important in this case. White, non-college educated voters turned out to vote for Trump in much greater numbers, especially in the Midwest, than was predicted. Conversely, there was an apparent overemphasis on how minorities would vote, both in terms of candidate choice and turnout. One of the reasons for this is that in the previous two elections President Obama was able to effectively mobilise minority voters, especially African Americans and Hispanics. However, it is clear that this election took place within a vastly different national and international context.

One reason why these dynamics are not well accounted for relates to how technology has made it more difficult for pollsters to get statistically representative sample sizes. Previously, polls were conducted using telephone directories where numbers could be randomly chosen. Reliance on mobile phones and online polls (like that of SurveyMonkey) has two effects; these methods are generally less responsive and skewed to particular segments of the population. It is therefore difficult to collect data reflecting the correct demographics and random samples from these methods.

Although some of these issues exist at the national level, they are more salient at the state level where there are less polls and fewer dedicated resources. In addition, at the state level, different pollsters tend to face similar challenges; for example, the same levels of non-responsiveness. They therefore tend to make errors in the same direction and, as such, having multiple polls does not reduce the margin of error.

This is why poor polling in states like Wisconsin and Michigan turned out to be so important. Clinton was expected to win in these states, and thus less systematic polling was done there. Polls here, along with other Midwestern swing states, underestimated Trump’s lead by four points on average. Overall, state polls did not adequately capture how white, non-college educated voters would vote, underestimated their turnout, and overestimated the turnout of minorities, especially those expected to vote for Clinton.

Although some commentators have pointed to the idea of a “shy Trump voter”, data analysis following the election calls this into question. The theory is based on the concept of ‘social desirability bias’, where voters feel that answering honestly as to who they would vote for would result in external judgement. However, the fact that Trump underperformed in the polls in Democratic strongholds like New York, where social desirability bias would presumably be stronger, points to the limitations of this theory. Rather, the growing disconnect between the Electoral College and the popular vote is a more powerful explanation

Politically, the result of this oversight is that many in Washington DC, and elsewhere, were unprepared for a Trump win. This, combined with his dearth of governing experience and controversial policies, has resulted in increased political uncertainty. Despite the fact that Trump has been able to defy expectations since winning the Republican primary, once in office he will face many typical political challenges, including resistance from within Congress, despite its Republican majority, and at various levels of government. Cities, such as New York City and Los Angeles for example, have become progressive strongholds with their mayors indicating a refusal to carry out Trump’s proposals, such as deportation of illegal immigrants and creating a registry for Muslims.

Over and above these constraints, Trump faces myriad challenges that restrict policy implementation. His plans for Obamacare are a good example of this. Trump has promised to keep two of the most popular provisions of the Affordable Care Act, namely that children can stay on their parents’ health insurance until the age of 26, and that insurance companies cannot deny coverage, or charge exorbitant premiums, because of preexisting conditions.

Health insurance experts, however, point to an obvious problem with such a plan: it is not financially viable for insurance companies to have people take out health insurance when they have a serious condition. In order for the provision to work, everyone has to take out some form of insurance every year. To ensure that poor people can afford this, there has to be a tax to subsidise them. Thus, Trump’s policies are internally contradictory; the provisions of the Affordable Care Act that Trump wishes to keep are dependent on those that he plans to repeal. Trump will therefore be constrained by the same challenges that face any politician when they have to shift from promises to policy, and from plans to implementation.