Election 2016          Social Desirability Bias / Very High Undecideds:  Eventually to Benefit Trump

As compared to historical US presidential elections from 1936 to 2016, the current level of undecideds is at the all-time high for this time in the cycle.  Some assumptions, which will be explained later, have been made, so there are caveats involved.  Historically, high undecided levels tended to hurt the socially undesirable candidate in the early polls but to help this candidate at the election-box as hidden supporters come out and vote.  Put another way, when polls tell you there is a very high level of undecideds and there is a clear candidate that people likely believe is socially embarrassing to publically support, it will be that very candidate who will benefit in the end – meaning the undecideds when they eventually ‘decide’ will lean towards the socially undesirable candidate to a greater extent.

In the current election, as we have reviewed elsewhere, Trump is the socially undesirable candidate.  There was a clear Social Desirability Bias against him in the primaries.  And, there clearly is a bias against Republicans on a national level.  There is, in contrast, a positive bias in favor of Clinton with certain demographic groups including women and minorities.  The unusually high undecided level will most likely benefit Trump as supporters come out of hiding on election-day.  Though not covered in this post, many who appear to support her in polls will also not vote for Clinton due to the positive Social Desirability Bias that she is enjoying.

In order to calculate the level of undecided voters four-way race data (includes two main parties plus Libertarian and Green Parties) from RealClearPolitics was used.  By taking the sum of the four candidates and subtracting the total from one, we know the implied percent undecideds.  Then, average percent undecided is calculated for a number of time periods.

For historical polling data, Gallup data is used, which is taken from Wikipedia.  This data does not provide exact dates for polls and instead uses monthly categories.  For some years an August/September category is used.

For comparison purposes, August/September data was averaged from the RealClearPolitics data to calculate undecideds for 2016 so that we could compare against all the historical data.  We only have partial data from September 2016 currently, but feel confident that the same conclusions can be made given the very high level of undecideds in August and through the first 10 days of September.  This is the first caveat – that we don’t yet have complete data from September.

The second caveat is that anonymous poll data is preferred to live poll data for 2016.  The main reason for this is that, as this election cycle has been so unusual, using anonymous data seems more accurate.  Also, as most historical elections did not suffer from such a great level of Social Desirability Bias, it seems like comparing the anonymous data, which is much less open to such bias, from 2016 makes more sense.

Chart 1:  Percent Undecided in US Presidential Races, Average Percent in August / September

Source:  Gallup, Wikipedia, RealClearPolitics

The August/September 2016 average using live polls is 6.4% and using anonymous polls is 12.4%.

The average for the entire non-2016 dataset was 7.4% and its high, from 1976, was 12.0%.

So, it really depends on if you chose to use live or anonymous poll data.  Using live poll data, 2016 looks like it has fewer undecideds than average while using the anonymous data it is slightly above the all-time high.

First and foremost, we believe the level of undecideds should be at or near the highest range of historical data due to the assumed considerable Social Desirability Bias in 2016.  All other things held constant, a higher level of Social Desirability Bias would result in higher undecided levels in polls.

As a backdrop, let’s review some of the factors that might contribute to people declaring themselves as undecideds when the election is rapidly approaching.  First, both main candidates have the highest unfavorable ratings of any US presidential candidate on record which would tend to push people’s decision back.  Second, Trump is the first US presidential candidate who has neither served as an upper echelon US military commander nor been previously elected to political office.  In addition, he is often referred to as a ‘reality-TV star’ in degrading terms and regularly called a bigot, racist, and misogynist by the media.  So, people might push off declaring support for Trump for these and other reasons.  On the other hand, Clinton has been under investigation for a variety of scandals and new evidence continues to drip out with more potential bombshells (from Wikileaks perhaps) expected prior to the election – more valid reasons to push off declaring.  Then, there are two additional valid parties, Libertarian and Green, disputing the election and some might want to learn more from them before declaring.

In addition to the aforementioned reasons, a few other points jump out from the historical data that support the idea that 2016 undecideds are likely on the higher end of the range.  First, election years with valid third party candidates tended to have slightly higher undecideds at this stage of the election.  Second, elections that did not include an incumbent tended to have higher undecideds as well.  As 2016 qualifies for both, we would from the start expect a higher level of undecideds.  Then, pile on the fairly unusual nature of the candidates and it is understandable why undecideds could be at or near an all-time high.

Who would a high level of undecideds benefit?  Using historical examples it seems like it tends to benefit the more socially undesirable candidate or party, or Trump/Republicans in this case.  Common sense confirms this assumption as by August/September most likely voters know enough to declare in a poll their vote intention.  Many declare undecided simply because they feel social pressure to keep their true leanings to themselves, to declare their support for a socially acceptable candidate or even to declare for a socially neutral candidate (like a third party candidate).

Let’s review two of the historical elections with the highest undecideds at this stage of the race to see if there could be commonality with 2016.

The highest non-2016 undecided level occurred in 1976.  This year saw Ford versus Carter.  Ford had the unfortunate responsibility to run as a Republican after scandals had rocked the party during the previous years.  Spiro Agnew, Nixon’s initial VP, resigned in 1973 due to tax evasion and was replaced with Ford.  Then Nixon under threat of impeachment resigned in 1974 making Ford president.  Ford in an unpopular move quickly pardoned Nixon.  The Republican Party had a cloud hanging over it in the 1976 election.  Most certainly there was a large element of social stigma involved for those openly supporting the Republican Party at the time.  Although no clear data exists, it seems clear that a Social Desirability Bias played strongly against Ford in the polls such that many who actually supported him and ended up voting for him declared themselves as undecideds in August/September.

In 1976, Ford as the more socially undesirable candidate benefited greatly from the unusually high level of undecideds.  The election results showed that, from the August/September period, Ford’s numbers increased 12 percentage points, Carter’s numbers decreased 2 percentage points, and 2 percentage points went to other smaller parties.

The second highest level (two-way tie for second place) of undecideds occurred in 1980.  This election saw a three way race between Carter, Reagan, and Anderson.  Carter was the incumbent.  Normally in elections with incumbents the level of undecided is not very high, but Carter was not that popular and was struggling with an Iran hostage situation.  Anderson was running as an independent, but had served as a Republican in the House of Representatives.  Reagan was running as the Republican candidate and had been the Governor of California but he was most well known nationally as a B-movie actor and for appearing in cigarette ads.  He was often made fun of and criticized as not being a serious politician.

In 1980, Reagan as the socially less desirable candidate benefited greatly from the unusually high level of undecideds.  The election results showed that, from the August/September period, Reagan’s numbers increased 10 percentage points, Carter’s numbers increased 5 percentage points, and Anderson’s decreased 6 percentage points.

Reagan was the socially undesirable candidate of 1980.  A B-movie actor who is often poked fun of is not normally the type of candidate who would attract positive coverage.  Likely, many voters leaned in Reagan’s direction but felt that he was not the socially acceptable candidate and hid in the undecided category and/or with Anderson, who had served in the House as a Republican and likely seen by many as a more socially acceptable candidate than Reagan.  But on election-day, these hidden voters came out in favor of Reagan.

Multiple elements from the 1976 and 1980 elections are present in 2016.  For instance:

• Damaged Party Brand – the Republican Party had a serious problem in 1976 with its image as it does in 2016. Just prior to 1976, there were major scandals large enough to tarnish the entire party.  More recently, the Republican Party has been hit by an identity crisis and enough negative coverage to create a negative bias against the entire party and not just Trump as some assume.  In other words, the Social Desirability Bias against the Republicans did not start with Trump as many ‘traditional’ Republicans would like to hear.  As we have seen there were significant issues with Social Desirability Bias against the Republicans in the mid-terms, or well before Trump came on the scene.  In both 1976 and 2016, we saw the implied impact of Social Desirability Bias as the percent of the populace registered as Republicans plummeted in the run-up to these elections (PEW Research and Gallup have some interesting research on longer term trends) but the party still continued to win many national and local elections, implying that many did not want to publicly be associated with the party but continued to vote for its candidates.  Even Romney’s ‘terrible choke’ in 2012 was lost by less than 4% of the popular vote, a small historical margin when running against an incumbent and being the candidate with a shrinking base.  The point being that the Republican Party brand in both 1976 and 2016 had been so tarnished that many viewed it as socially undesirable enough to change their answers in polls, but not on election-day (at least we assume that will be the case in 2016).
• Butt of Joke ‘Non-Serious’ Candidate – In 1980, Reagan (like Trump in 2016) was the butt of jokes and left many wondering how the electorate could take him seriously as a national politician. For many, it just seemed inconceivable that a ‘B-movie actor’ (or a ‘Reality-TV star’ in 2016) could run for president.  Not only did attacks such as these mark the candidate but also the candidate’s supporters as the underlying message was that there could be something wrong with the judgment of someone supporting such a candidate.  In other elections, supporters of candidates might be attacked for a candidate’s policy or stances, but rarely are supporters’ judgment questioned simply for preferring a candidate.  In the case of Reagan and Trump, their supporters’ ethics, morals, and judgment were derided before election-day creating an environment where a negative Social Desirability Bias could thrive.  Surely Reagan suffered from Social Desirability Bias as does Trump currently as neither was/is seen as being a totally socially acceptable choice.
• Still Undecided even though everyone knows the Leading Candidate – In 1980, everyone knew Carter, just like everyone knows Hillary Clinton in 2016. Carter was the sitting president running for reelection.  He clearly had an advantage as the incumbent as the nation knew him already.  If a candidate is very well known and undecided levels are extremely high, there is a major problem.  There are not too many explanations for this phenomenon other than undecideds are particularly high as people know they don’t want to support the most well-known choice but for some reason are having trouble declaring support for the other choices.  In 2016, the ‘Carter’ is ‘Clinton’.  Hillary Clinton has been a fixture in national-level US politics for about 24 years.  She was one of the country’s most visible First Ladies, a senator from arguably the most powerful state in the country, and the Secretary of State.  Also, Forbes has her at #2 on the “The World’s 100 Most Powerful Women” list, and she holds no official government office at present.  Clinton certainly is an incredibly recognized and well-known person.  The fact that there are so many undecideds in an election which has such an individual as the odds-favorite for the past year does not bode well.  It seems like on election-day, many of the undecideds will break against the better known candidate in 2016 as they did in 1980.

These elements have come together in 2016 to create an extremely high level of undecideds.  Trump essentially has inherited a similarly tarnished party image to 1976 and a similar ‘non-serious’ candidate tag as Reagan in 1980.  These elements help to create a rather stubborn Social Desirability Bias against Trump in the polls.  Additionally, we have the phenomenon, similar to that of 1980, of high undecideds in a race which boasts an incredibly well-known and experienced candidate.  This environment should produce a combination of 1976 and 1980 where undecideds break for the socially less desirable candidate, or Trump, on election-day.