This week’s model depicts a “what-if” scenario wherein Donald Trump wins the election; this scenario quantifies the dynamic of “social desirability bias.” If you would like to skip the blog and go right to that scenario, click here
Polling is as much of an art as it is a science, particularly when it comes to forecasting elections. Polls conducted the same week can generate different results based on the people who are polled (e.g., likely voters versus eligible voters, tracking poll), the types of questions (e.g., yes/no, range of choices), mechanism (e.g., telephone interviews, push-button) or the polling organization (e.g. pro-Trump, pro-Clinton). Each results reflects some inherent bias source from one or more of these factors (read more here).
For that reason most elections watchers rely on the data from multiple polls to minimize those factors.Our election series does so as well. We rely on the polls from www.270towin.com, which uses several polls, including those that report figures for third-party candidates. The latest data, presented in our interactive format, show Clinton with a substantial and growing lead.
Does that mean the election is over? Do these latest polls tell the definitive story of who will win the election?
No. Not by a long shot. Recent history suggests that the polls can get it wrong and get it wrong “bigly.” (read: Brexit)
And why do polls get it wrong?
Simply, people lie.
Are We Seeing the “Bradley Effect” in Effect?
If respondents are dishonest, then polls are useless.
There is historical precedent for this and it’s called “The Bradley Effect.” The Bradley Effect refers to the 1982 California gubernatorial race in which Los Angeles Mayor Tom Bradley, an African-American, lost the election to George Deukmejian, who was white, despite final election polls AND exit polls on Election Day showing Bradley with a sizable lead. Adherents of the Bradley effect believe that some voters will tell pollsters that they are undecided or likely to vote for a minority candidate but will vote against the minority candidate on Election Day.
The Bradley Effect is a specific version of what psychologist call “social desirability bias,” which “refers to the fact that in self-reports, people will often report inaccurately on sensitive topics in order to present themselves in the best possible light (Fisher, R. J. (1993). “Social desirability bias and the validity of indirect questioning“. Journal of Consumer Research, 20, 303-315.).
In other words, people lie to pollsters to make themselves look good.
Now this does not mean that the system is “rigged” against any particular candidate; rather it means that polls are imperfect and they involve human beings, who are subject to all kinds of real and perceived pressures.
“What-if” Scenario: Do Current Polls Hide Massive and Widespread Social Desirability Bias?
Is there evidence of “social desirability bias” in this election? Specifically, are people telling pollsters that they are against Trump but will vote for him come election day? Writers debate this here, here and here.
At Trefis, we are indifferent to whether social desirability bias exists or not; in fact, we are inclined to assume it does exist, along with the many other biases mentioned at the top of this article. What we care about are data.
Clinton leads Trump by 172 electoral votes. She has 352; he has 180 (Utah’s 6 votes are up for grabs). Ignore the solidly blue and red states and focus on the closest races where Clinton leads Trump by less than 5% points. Those races are (Clinton %, Trump %):
- Nevada (44%, 41%)
- Arizona (43%, 41%)
- Florida (45%, 42%)
- North Carolina (45%, 42%)
- Ohio (44%, 40%)
You would have to believe that enough bias exists in these states to flip these states in Trump’s favor if, in fact, respondents are lying to pollsters. When we create our own scenario and “tip” these states to Trump, what happens?
Trump nets a total of 79 electoral votes (Nevada – 6; Arizona – 11; Florida – 29; North Carolina – 15; Ohio – 18) giving him 259. Clinton would have 273 electoral votes.
- Clinton still wins.
Therefore, other states must break Trump’s way.
Could Trump Still Win The Election?
He sure could. The rest of the data in our scenario show a pathway to victory. Here’s what else would have to be true if you believe widespread social desirability bias skews current polling results.
What-if he pulls out Utah, currently a neck and neck race between Trump and newcomer Evan McMullin, and its 6 electoral votes?
- Not enough: Clinton wins 273 to 265.
He needs at least one more state. The next closest race in terms of Clinton’s percentage lead is New Hampshire (44%,39%). Net 4 electoral votes.
- Still not enough: That results in a 269 – 269 tie.
Trump would need, let’s say, Colorado (44%, 38%) and its 9 electoral votes to flip as well.
- Final tally: Trump 278, Clinton 260.
In this scenario, Trump needs these states to flip:
- North Carolina
Therefore, if social desirability bias exists and it accounts for Clinton’s lead in these closer races… AND NOTHING ELSE CHANGES between now and election day, Trump may yet win the election.
Is it possible? Yes. It is probable? That’s up to you to decide. Here’s the model.