Why pollsters will keep getting it wrong this year

Bernie Sanders scored an upset victory in Michigan’s Democratic primary that shocked the legions of pollsters who had confidently predicted that Hillary Clinton would walk away the victor. Given the reliance on polls in this political season, it seems worthwhile to take a few moments to reflect on how the polls missed so badly in calling the Michigan contest.

Nearly everyone got it wrong

Nate Silver, the editor of FiveThirtyEight, one of the better-regarded polling sites, said that the missed call was “among the greatest polling errors in primary history.” FiveThirtyEight had given Hillary “a greater than 99% of winning the Michigan primary” on the eve of the election. Oops!

However, Nate was not alone in making the wrong call. The Real Clear Politics polling average for the period of March 3-7 had Clinton ahead by a spread of 21.4 points on the eve of the election. The Clinton camp was in a state of shock after the results came in while Bernie’s supporters were #feelinthebern.

So, what happened? The Wall Street Journal put it this way:

To the extent that Sanders remains a viable candidate, we’d say it is because of two factors of demography—one a strength of his, the other a weakness of Mrs. Clinton’s. The strength has been much remarked upon: Sanders’s overwhelming support, outside the South anyway, from young voters.

Michigan follows the pattern: Under-30 voters favored him 81% to 18%; over-65 voters backed her, 68% to 31%. Every way the pollsters break it down, there is a clear inverse correlation between age and Sanders support.

Problems with obtaining accurate polling data are not new to this election cycle, but the errors do seem to be getting more pronounced. Part of the problem is tied to changes related to an increasingly digital and multicultural world. The ability of pollsters to get hold of people has become more difficult in the cellular age where it is easier to screen calls, and internet polling has not proved reliable. There is also a greater guarding of personal privacy with fewer people willing to share their feelings about a candidate or issue.

Problems with obtaining accurate polling data are not new to this election cycle, but the errors do seem to be getting more pronounced.

One of most known and respected polling companies, the Gallup Organization, incorrectly had Mitt Romney leading Barack Obama in the final days of the 2012 presidential election. In the end, Obama won by a four-point margin proving Gallup’s polls wrong. Following the election, Gallup conducted an in-depth review of its polling procedures in an effort to improve its performance. The analysis revealed three major issues that skewed the results:

1. Misreading how likely voters were to vote.

2. Under-representing some liberal areas within the U.S. in determining the poll model.

3. Too many questions relating to race and ethnicity.

A more detailed reading of the results of the study can be found in this article on Politico.

The Trump effect

Photo: Gage Skidmore

In the current election cycle, questions are being raised about how a “social desirability bias” may be causing supporters of Donald Trump to say one thing to a pollster and then do the opposite in the privacy of the voting booth. Bloomberg Politics reports:

Polls may be underestimating Donald Trump's support, according to intriguing new research that says the Republican front-runner benefits from a "social desirability bias"—some people who plan to vote for him are too embarrassed to admit it.

This has occurred in past elections, most notably in the 1982 election for governor of California where Mayor Tom Bradley, an African-American, led in the polls going into the election but ended up losing. A review of the election concluded that voters were afraid to admit they did not support Bradley lest they be considered racist. This came to be called the “Bradley effect.” Given Trump’s bombastic style there may be a “Trump effect” occurring wherein voters do not want to admit that they find Trump an attractive candidate, thus causing the polls to underestimate his appeal.

The bottom line is that Michigan was not the first election the polls have gotten wrong, nor will it be the last.

Leave a Comment