The 2016 U.S. election cycle has been among the longest, nastiest, and most divisive in the country's history. The partisan divide between Democrats and Republicans seems wider and more insurmountable than ever. Members of each camp routinely express bafflement and disbelief that anyone would even consider voting for the opposing candidate. Virtually the only matter that both sides can agree on is that they're looking forward to the whole affair finally wrapping up next week.
Why is it that this election has proven so dramatically polarizing? This question is immensely complex by nature, and academics will no doubt continue to study it for decades. But it's not too early to use data to draw some conclusions about what determines voter preferences. In an effort to better understand this important subject, we at ClearerThinking.org recently used machine learning techniques to analyze a survey of voters and the beliefs that drive their choices. One finding in particular struck us as revealing:
Trump supporters and Clinton supporters tend to have substantially different understandings of honesty.
This analysis is a part of a series of three studies that we conducted by recruiting over a thousand participants participants online. In it, we asked people who had already decided their vote a range of questions about demographics, party affiliation, policy preferences, and their beliefs regarding how leaders should act and think. Then, we used a machine learning model to predict support for each candidate based on participant responses to these questions. (You can read more about the study design at the bottom of this piece.)
Among a number of interesting results, we found that perceptions of honesty were one of the biggest differences between Clinton and Trump supporters. This isn't that surprising by itself. Honesty and candidate character have been among the most visible and frequently discussed issues of this campaign season, and our study found that large majorities of each voting bloc in our sample — 86% of Clinton voters and 96% of Trump supporters — view the opposing candidate as dishonest.
However, each side tends to use dramatically different metrics to determine which candidate is the more honest of the two. In short, Clinton voters tend to place a higher value on factual veracity (e.g. making empirically correct claims) when considering how honest a candidate is, while Trump supporters tend to focus more on authenticity (e.g. saying what you really think at a given moment).
Let's take a closer look at how voters perceive each of these candidates, starting with Clinton. The media often portrays Clinton as a wonkish, excessively prepared candidate who's extremely guarded about her true feelings and who tends to speak in memorized talking points. Voters from both camps appear to agree with this portrayal to one degree or another. In our survey, 69% of Clinton voters said that she's less likely to say what's really on her mind than Trump, and 86% of Trump voters agreed. Clinton tends to earn higher marks for factual accuracy than her rival, though there's more disagreement on this subject. A full 93% of Clinton voters said that she makes factually incorrect statements less frequently than Trump, and 26% of Trump voters agreed that she's factually correct more often than their candidate. (This means that most Trump voters think he's the more factually accurate candidate, but Clinton voters believe in her veracity much more strongly than Trump voters believe in his.)
By contrast, Trump's media persona is that of an impulsive, off-the-cuff speaker who almost can't help but say what he really thinks, even at the expense of factual accuracy. As noted above, the voters in our survey largely backed this image — they broadly agreed that Trump is more likely to say what's on his mind, but is also more likely to miss the factual mark.
We explicitly asked our survey respondents about this distinction, and voters' feelings about it broke down predictably but somewhat unevenly across party lines. As you can see in the graphic below, a full 85% of Clinton supporters said that citing inaccurate facts and statistics to the public is worse than relying on talking points in an effort to avoid speaking one's mind. 52% of Trump supporters agreed with them. By contrast, 15% of Clinton supporters and 48% of Trump supporters said that relying on canned talking points is more dishonest than making factually incorrect statements.
These findings represent part of the differing attitudes about honesty between Clinton and Trump supporters, but they don't tell the whole story. You may be wondering, for example, why it is that Trump supporters perceive their candidate as so much more honest than Clinton, while simultaneously appearing to be less certain of Trump's command of the facts, and more ambivalent about what constitutes real dishonesty.
Two further tidbits from our survey may help explain this paradox. The first involves another high-profile issue from this election cycle: media bias. Judging by mainstream fact-checking media outlets, Clinton broadcasts false information far less often than Trump. PolitiFact, for instance, has gauged 70% of the Trump statements it's fact-checked as "mostly false" or worse. By contrast, just 27% of the Clinton statements it's looked into have earned such rebukes. In both cases the numbers are disturbingly high, but if you trust Politifact to assess such things accurately, a Trump claim is far more likely to be false than a Clinton claim.
But findings such as this one aren't likely to affect Trump supporters' perception of their candidate, because they overwhelmingly distrust the media outlets that perform such fact-checking functions. A whopping 92% of Trump supporters in our survey reported that they don't at all trust the mainstream media, compared to 61% of Clinton supporters. (That both numbers are so high does not augur well for a return to fact-based political discourse in the near future.)
As a result, Trump is fairly impervious to hemorrhaging support when the news media reports that he's said something untrue. Recall that 74% of his supporters think he's led a more factually accurate campaign than Clinton has, despite vehement disagreement from media fact-checkers such as PolitiFact.
And significantly, his supporters also broadly think that factual precision is a secondary issue when it comes to honesty. Consider the chart below, which shows levels of agreement on a -3 to 3 scale, with Trump supporters in red and Clinton supporters in blue. As you can see, 83% of Trump voters say that it's important for leaders to say what's really on their minds, to one degree or another. By contrast, just over half of Clinton voters — 53% — say that it's important for leaders not to say what's on their minds, to some degree. Qualitative answers given by our survey participants suggested that these Clinton voters believe that saying what one really thinks can be a sign of impulsiveness or recklessness, and that these qualities are not attractive in a presidential candidate.
There's one final factor to consider here — distrust of politicians in general. The graph below shows how strongly the voters in our sample agreed or disagreed with the notion that "all politicians are liars," on a scale of -3 to 3. The overall proportion of voters from each camp that agree to some extent with this idea is fairly similar, with 62% of Trump voters and 56% of Clinton supporters signing on. However, a much larger proportion of Trump supporters agree strongly with this notion — 39%, compared to just 15% for Clinton voters. Since Trump positions himself as a political outsider and Clinton's campaign emphasizes her experience in government, this portion of Trump voters is more likely to trust their candidate much more than Clinton by default.
And there you have it. Judging by this survey, Clinton voters tend to believe that their candidate is more likely to be factually accurate while on the campaign trail, which they consider the most important metric for honesty. While they concede that she doesn't always say what she's really thinking, many of them consider that behavior appropriate for a leader. By comparison, Trump voters believe their candidate is more honest about his inner thoughts, and while they're less certain that he's factually accurate all the time, they tend to distrust the organizations most likely to call him on apparent falsehoods. This disagreement over the nature of honesty seems to magnify the typical distrust for the opposing party, which helps to explain why questions of honesty are so critical to this election.
More about these studies:
This data came from three studies, which we conducted via the online recruiting platform Amazon Mechanical Turk. In all three cases, we only examined data from people who said they planned to vote in the U.S. presidential election and who said that they had already decided to either vote for Donald Trump or Hillary Clinton, paying each participant a small fee for their work. Study 1 was a small pilot study, in which we asked a long string of questions and required free-form text responses. In Study 2, we collected 138 variables on each of 440 Clinton supporters and 360 Trump supporters. (This approach yielded 800 data points in total, though there is actually some extra data we collected as well which we did not have access to yet when we performed our analyses.) Study 2 examined far more factors than any previous work we have seen on the topic of why people support Trump vs. Clinton, which usually only consider a handful of variables or basic demographic information.
These 800 data points were used to build a machine learning model (logistic regression with L2 regularization) designed to predict support for Trump vs. Clinton from the other variables, by testing which factors were most useful in predicting who each person supported. The accuracy of the model was confirmed on a clean withheld sample of data (160 data points not used to train the model, achieving 91% accuracy). This was done to verify that the model was good and that overfitting had not occurred.
Finally, we conducted Study 3 to confirm/double check some of our prior findings from Study 2 and to gain deeper insight into why we were finding them. In particular, we wanted to check that certain associations held — indeed, we were able to confirm them successfully — and also dig deeper into why participants were answering each question the way they were. So for Study 3, we used a much smaller sample size, but asked participants to explain their answers to most of the questions (in free-form text responses that we could then read through for deeper understanding, which meant that data collection took a lot longer per question). The results discussed in this particular post pertain to one particularly interesting strong finding from Study 2: that Trump supporters really prefer a candidate who says the things that pop into his mind, rather than using filtered talking points. Other data in Study 2 and Study 3 allowed us to investigate this issue further, which led to the results presented here. The data we collected is not a nationally representative sample, so has its limitations. However, it is a good sign that our data confirmed every finding we were aware of that had come from nationally representative samples and large surveys. Namely, both these polls and our results agree that on average Trump supporters are (relative to Clinton supporters) more male, white, aged, Republican, socially conservative, fiscally conservative, anti-immigration, uneducated, married, rural-dwelling, authoritarian, and religious. This is true if we examine the individual variables in our data — in each case, a higher percentage of Trump supporters we polled have these traits than the Clinton supporters — and it is also true if we look at the coefficients in our machine learning model (which tries to predict who each person will vote for).
Please note that it is very hard to infer causality from survey data, so any causal claims that we make should be considered tentative. They are merely our best inference about the truth based on the data that we collected.