Election 2016 Social Desirability Bias / Johnson and Stein: Third Party Candidate Bias and the Hidden Voters
The 2016 US presidential election is heavily influenced by Social Desirability Bias. We have seen how Clinton and especially Trump have been impacted in polls. But what about the third party candidates, are they impacted by Social Desirability Bias as well?
Both Johnson and Stein have greatly benefited from bias in 2016. These candidates rate significantly higher in live polls than in anonymous polls. It seems like many people see them as non-objectionable candidates, especially in comparison to a candidate like Trump who has received tremendously negative commentary by the media, and disproportionately chose them in live polls to avoid social upset.
For instance, let’s assume an individual is leaning towards Trump. But following the election news and her social media stream makes her realize that Trump is viewed by many as simply a bad decision and completely socially unacceptable. Supporting him publically could really change her life – in an emotionally-charged election climate, she could lose friends or even diminish job prospects as she lives in an urban area that traditionally votes Democrat. She does not want to say that she is undecided as people will continuously try to convince her to support a particular candidate. The easiest path, socially, for her is to declare support for a third party candidate. They are seen as fairly neutral in the sense that they do not inspire the type of visceral response that a candidate that Trump might. Also, her social network would remain presumably unaffected. So, she declares support for Stein (or Johnson) when people ask, providing some generic and socially acceptable reasons to do so.
The previous example is more or less how Social Desirability Bias, an emotional election, social media and successful political branding have combined to make third party candidates look much more viable than they really are.
The last point of political branding is fairly important but actually difficult to quantify. Reading liberal media and how it discusses Trump and how conservative media discusses Clinton, we can more or less determine how the opposing parties have branded their main competitor.
Non-Clinton supporters have successfully braded Clinton as a dishonest sneaky career-politician with criminal tendencies. By associating with this brand, many people will see you in this light. Some might think, ‘wow, (person’s name) is very pro-Clinton, I guess he is really ok with so much deceit, I wonder if he has ever lied to me, is that guy a lier?’
Non-Trump supporters have successfully branded Trump as a fascist racist misogynist money-grubbing capitalist. By associating with this brand, many people will see you in this light. Some might think, ‘wow, (person’s name) is very pro-Trump, I guess they would have voted for Hitler too, I wonder if she is secretly racist, I never see her socializing with other races, hmm’.
Third party candidates on the other hand have fairly clean slates. In fact, many realize these candidates and parties exist but really don’t know much about them. Such negative political branding simply has not reached these third parties. In this sense, they are much more socially acceptable, especially in certain social circles.
Table 1: Favorable Ratings for US Presidential Candidates Sept 2016
|
Very Favorable |
Somewhat Favorable | Somewhat Unfavorable | Very Unfavorable |
Don’t Know |
|
| Clinton | 22% | 24% | 10% | 43% | 2% |
| Trump | 19% | 17% | 11% | 51% | 2% |
| Johnson | 4% | 23% | 24% | 14% | 35% |
| Stein | 5% | 18% | 20% | 18% | 40% |
Source: The Economist/YouGov Poll
As the previous table shows, both Clinton and Trump are highly disliked. Trump leads in this category, partially explaining the high negative bias that exists for him in live polls.
In contrast, Johnson and Stein lead in the ‘Don’t Know’ category. This is not something you want to lead in during a normal election cycle. But in the current election cycle, there appears to be a sweat spot between being known enough but not being that well known which tends to attract a positive bias in live polls for those looking for ‘cover’.
Another way of looking at the data is to add together the Favorable and the Don’t Know categories to get a neutral to positive rating. In order to avoid potential social condemnation, people would seek out these categories over the Unfavorable category candidates. In this sense, those sensitive to potential social condemnation would avoid supporting Trump in public. Especially those not overly interested in politics would certainly consider publically supporting Johnson or Stein as they lead in the neutral to positive rating.
Table 2: Comparing Aggregate Favorable and Don’t Know Categories for the four main US presidential candidates
|
Favorable + Don’t Know |
|
|
Clinton |
48% |
| Trump |
38% |
| Johnson |
62% |
| Stein |
63% |
Source: The Economist/YouGov Poll
When people look for ‘cover’ from social condemnation, they don’t need the perfect candidate, just one that will be more socially acceptable than the one they really prefer and that will fit their own unique circumstances. Many people really are not that into politics (those reading this might find this unusual but it’s true!) and would not mind a type of filler candidate for a poll or even to tell friends who is more unknown. The main focus is to avoid the socially unacceptable candidate – which is clearly Trump in this election but could also be Clinton in certain circumstances as well given her rather high unfavorable rating.
Measuring Social Desirability Bias is rather straight-forward. Live polls contain an element of social pressure as a live interviewer conducts them. Anonymous polls, relying on robocalls, IVR, and internet collection, contain less social pressure. The difference between these two provides us our bias rating. The following chart shows the bias for Johnson and Stein from June to September, which is the entire data series of acceptable data.
Chart 1: Social Desirability Bias for Johnson (Libertarian) and Stein (Green)
Source: RealClearPolitics
The consistency of the positive bias for both third party candidates is fairly impressive which reinforces the thesis that an underlying variable is impacting the superficial attractiveness of these candidates. If the bias appeared very volatile, a rethink might be necessary, but it appears very consistent. The bias should diminish as election-day approaches but will likely not go to zero. Assuming it does stay around these levels, a potential election-day surprise could occur as the bias gets allocated to votes for other candidates.
The size of the bias is actually fairly large and leaves a lot up for grabs. The combined bias, on average, amounts to around 4 percentage points which is actually a lot in this election. In comparison to the aggregate poll results of these two candidates in September 2016, the bias amounts to over 40% which is substantial to say the least. Also, recall that the 2012 election was won by a margin of less than 4 percent of the popular vote.
Another interesting comparison is to look at how social media metrics forecast popular vote. In terms of Social Desirability Bias, live polls tend to be less accurate than anonymous polls in emotional elections. Even more accurate could be social media and on-line activity as people in general do not believe their actions are being analyzed and tend to not modify their actions.
The following table shows a comparison of these three main methods for forecasting popular vote in the election. The poll data has been adjusted to correct for the fact that polls allow for undecideds whereas the social media based forecast does not. The adjustment is very straight-forward in that we assume that the undecideds are divided equally between the candidates on a pro-rata basis of their current poll results. The adjusted numbers for Johnson and Stein are slightly higher than what you would have seen in the previous chart and if calculating straight averages from the data.
Table 3: Forecast Popular Vote for 2016 US Presidential Election for Johnson (Libertarian) and Stein (Green), Comparing Live Polls, Anonymous Polls and Social Media Influence Index
| Johnson | Stein | |
| June-2016, Live Polls | 9.4% | 5.0% |
| June-2016, Anonymous Polls | 6.6% | 4.5% |
| June-2016, Social Media Influence | 2.5% | 1.3% |
| July-2016, Live Polls | 9.8% | 4.6% |
| July-2016, Anonymous Polls | 6.4% | 3.1% |
| July-2016, Social Media Influence | 3.4% | 3.3% |
| Aug-2016, Live Polls | 10.2% | 4.4% |
| Aug-2016, Anonymous Polls | 8.0% | 3.1% |
| Aug-2016, Social Media Influence | 3.1% | 1.6% |
| Sep-2016, Live Polls | 10.8% | 3.4% |
| Sep-2016, Anonymous Polls | 8.1% | 2.8% |
| Sep-2016, Social Media Influence | 2.6% | 1.4% |
Source: RealClearPolitics, ZettaCap
Comparing these forecasts further confirms that Social Desirability Bias is at work. Higher degrees of a structured and pointed political environment equates to higher support for third party candidates. Higher degrees of anonymous and unfiltered responses equates to lower levels of support for third party candidates.
Assuming that the best forecasting tool is social media, real support for third party candidates is actually very low. This is not a media friendly topic, but much of the support found in polls appears to be from individuals wanting to avoid social condemnation from peers.
The advantages of looking at social media and on-line activity for forecasting elections are covered elsewhere but, summarizing, people tend to be fairly unfiltered when on-line which would make such data, in aggregate, more accurate in an emotional election. Additionally, third party candidates have a built-in advantage when it comes to social media as they overwhelmingly attract younger supporters in comparison to the main parties which should make their social media based metrics look more favorable than those of the main parties. In reality, what we see is that the third party candidates perform very poorly on this metric implying that though they appear to poll well, their actual level of support is far lower.
Additionally, it should be pointed out that both Johnson and Stein headed their respective party’s ticket in 2012. In that election, they jointly received 1.35% of the popular vote. The parties have not changed and the top of the tickets have not changed. In general, it seems like their messages are almost exactly the same as in 2012. And, we are supposed to accept that they will, according to live polls, receive 14.2% of the vote in 2016. This seems very far off the mark. The SMI (social media-based metric) forecasts a joint popular vote of 4.0% which is still a considerable increase over 2012. In short, looking at historical performance of these parties and acknowledging that they are running essentially the exact same ticket as previously makes you wonder if the polls really are exaggerated.
This large difference between what common sense and social media activity tells you and what live polls tells you can be most fully explained by Social Desirability Bias – saying one thing in public and doing another in private in order to comport with inferred social norms.
If we assume the SMI is correct, it means that there is an amazing 10 percentage points of popular vote up for grabs that is hidden in superficial support for third party candidates. Where will these votes go to on election-day? As explained in other posts, the largest chunk should go to Trump as the least socially desirable candidate. Additionally, it is a little difficult to accept that Clinton who has been a fixture in American Politics for a few decades would grab these votes given that in other elections voters normally know fairly early if they intend to vote for the most well-known candidate. Again, these arguments are more fully made in other posts.
The main point is that there appears to be an exceptionally high level of voters taking cover from the overly emotional election by superficially supporting third party candidates. Much of the reason for feeling they need to superficially support a candidate different from their preferred candidate is that there is a very intense negative brand attached to Trump causing him to be seen as socially undesirable by many making them change their public-facing response to election questions. Further, these voters will likely move disproportionately to Trump on election-day given that he is the most socially unacceptable candidate. In short, look for third party candidates to substantially underperform polls and for Trump to outperform them on election-day as many voters superficially supporting third parties shifts to Trump in the privacy of the voting booth.