I feel like the Electoral-Vote.com analysis of this article is pretty solid, including the assessment of Cahaly.
As long as we’re covering things that make Democrats skittish, let’s talk a bit more about the alleged “shy Trump voters.” Yesterday, Politico Magazine published an interview with Arie Kapteyn of the USC/Dornsife Poll, and Robert Cahaly of the Trafalgar Group about that very subject. That duo was chosen because USC/Dornsife and Trafalgar were the most bullish on Donald Trump’s chances in 2016 than any other pollsters, and so were the only ones to get the election “right.”
If you would like a thoughtful assessment of the possibility of “shy” Trump voters, you should read through Kapteyn’s answers to the questions. He acknowledges that USC/Dornsife didn’t actually do all that well in 2016, since they gave the popular vote to Trump by about 3 points, and he lost it by about 2 points, which means a rather sizable five-point error. Kapteyn also acknowledges that the shy Trump effect, if it exists, is very difficult to measure, and that it may or may not have existed in 2016. To the extent that there is evidence of the effect in 2020, he points to the data that he and the folks at USC/Dornsife have collected in response to the “social-circle” question they ask. In short, when people are asked “Who are you voting for?” then Joe Biden comes out ahead by about 10 points. However, when people are asked “Who are your friends and family voting for?” then Biden’s lead drops to about 6 points. The assumption here is that people are more honest when talking about the behavior of others than they are when talking about their own behavior.
As to Cahaly, he’s a partisan hack who is interested in selling his services to Republican politicians. There is nothing he says in the interview that would do anything to cast doubt on that reputation.
For our part, we remain very skeptical that there is any meaningful shy Trump effect. Here are three major reasons:
The existence of any “shy” effect (whether shy Trump effect, or shy Tory effect, or Bradley effect) is hotly debated, since the effect—if it exists or has ever existed—is invariably subtle enough to potentially also be explained by movement within the margin of error.
In 2016, the final national polling average for Hillary Clinton was 46.8% and for Trump was 43.6%, a gap of 3.2%. When the votes were tallied, the final percentages were 48.2% for Clinton and 46.1% for Trump, a gap of 2.1%. Again, that is a very subtle difference, and could easily be explained by movement within the margins of error and/or a late Comey-inspired break toward Trump that was not captured by polls.
The best evidence of a shy Trump effect in 2016 was that he did about 1 point better in Internet polls than he did in telephone polls. The theory here is that people are more likely to lie to a human being than they are to a computer. But even if that theory is correct, there is no equivalent gap in this year’s polls—Internet and telephone polls are producing nearly identical Trump results in 2020.
With all of this said, we wanted to try to find a different way to approach this problem…
