Trending

Newsletter

Why Most Polls Failed to Predict Trump’s Win

By: Luis Henrique Ball Jr. - Nov 9, 2016, 12:11 am
photo-1
Why most polls failed to predict Trump’s win in the 2016 US presidential election.

Most polls failed to predict Trump’s win over Hillary Clinton on November 8. If Brexit, seeming the catalyst for all things in 2016, had weakened some of the confidence once typically placed on polling, the 2016 US election has all but ridiculed the practice. If the presidential candidates are to be believed, electoral polling should never again be trusted to deliver accurate estimates. They say polls have simply become too biased.

Bias, however, does not appear to be a reason polls are less accurate, and more volatile, than ever. Many pundits and pollsters are quick to point out that, instead, we should look to the technological advances that have fundamentally reshaped both polling methods and the country’s demographics.

The iPhone and the internet have, together, not only removed newspapers as a reliable funding source for polling, but also altered respondent behavior to the point where they’ve stopped taking part in call-in polls. The drop in funding and respondents has led to shortcuts, and while the business of polling was never lucrative, now it’s become both cost ineffective and inaccurate. To illustrate the sub-par nature of current polling techniques, consider the fact that in the 370-plus polls that FiveThirtyEight uses to make its election projections, only 34 got an A+ grade.

article-2 (Data from Washington Post article on Anzalone Liszt Grove Research)

Part of the problem is that reaching the electorate with these new phone habits is hard. Most pollsters have neither the money nor the sophistication to pull it off. Since mobile phone users lean democrat by 11 percentage points and landline users only four, failing to include a good mix of both in a poll can lead to drastically different biases. Over half, 204, of all the polls used in FiveThirtyEight’s modeling do not include live caller respondents with a cellphone.

“Without reaching the cell-only population you are missing up to 40 percent of eligible poll respondents for a national poll” — Scott Clement and Peyton Craighil

To make matters worse, even if one removed all the polls that failed to include Americans who only use cellphones, the respondent pools change so frequently that only the most interested of poll-watchers would catch subtle shifts in the data.

Reuter’s election poll ran into some controversy over this very issue when it was accused of oversampling voters with a college degree. It’s widely accepted that education levels often vote in blocs, high school education level or less voting one way and, often, college educated voting another. In this election, voters with a college degree have shifted dramatically in support of Democrats, while voters with a high-school degree or less overwhelmingly in support of Republicans.

photo-3
The Wall Street Journal, Voters’ Education Level a Driving Force This Election

Since oversampling college-educated voters (Reuters had it about 53 percent of those polled), would potentially bias the poll toward Hillary, complaints by the conservative media were not entirely without merit. Reuters has since changed its polling pool to distinguish how many of those polled received any college education and those who got a college degree, clarifying the “oversampling” may very well have come from polling those who had not finished their degrees.

Despite the number of polls that conservatives claim show similar bias, most pundits agree that the polls aren’t rigged. Instead, the consensus seems to be that, as a result of changing technology, a decrease in funding, and a highly partisan electorate, polls have simply become less accurate.

Instead of looking to public polls for campaign guidance, candidates have instead turned to private polls. Once established as an ‘A+’ pollster, many polling companies that once publicly shared their data began selling it instead to political campaigns. Pollsters specialize by political party, state, demographic, and also methodology. Some choose to stick to tried and true polling methods, sticking to a sizable sampling of responders, both on landline and cellphones, that they poll throughout the race. Others market themselves with their cutting edge techniques, which entail adjusting polling results for previously observed responder voting behavior, referred to as voter-recall. Both are indisputably more accurate than publicly accessible polls.

The advantage for pollsters to going private comes down to “[hedge funds] paying a lot more than any news outlet or school would pay for more precise data that isn’t always publicly available”. The losses many companies suffered as a result of Brexit has only further fueled the practice. Despite the obvious upsides to having so much cash thrown at polling, it comes a little benefit to the general public, who will never see the results.

The divide between what the public knows and what especially motivated private parties know should be concerning. Unless the average voter can afford access to more accurate polling, planning, financial or otherwise, around the potential results of the election will remain out of reach. Not has the average citizen lost dozens of sources of local news, but with them the ability to reliably make assumptions about their future after the election.

The sad truth however, is that maybe the average voter no longer cares about neither the news or the election, and instead looks to inform him/herself so far as to further affirm that which he already knows.

The game is ‘rigged.’ Why bother.

Luis Henrique Ball Jr.

Luis H. Ball Jr. is a writer and technology-evangelist. You can read his musings on the industry at elPundit.com. Follow him on Twitter @luqven