As Donald Trump’s presidential campaign sputters onward, a number of denialists have emerged ― including Trump’s own pollster-turned-campaign manager, Kellyanne Conway ― to claim that the GOP nominee is doing better in online polls than in polls conducted over the phone by live interviewers. These “poll truthers” argue, incorrectly, that this supposed discrepancy proves there’s actually a big segment of uncounted voters who will turn out to vote for Trump on Election Day. They’re just afraid to admit that to pollsters ― or so the argument goes.
Evidence shows that something very different is happening. Trump’s support doesn’t change much depending on how the polls are conducted. But Hillary Clinton’s does. And like most other aspects of current polling, it’s not good news for the Trump camp.
In telephone polls conducted by live interviewers, Clinton averages an 11-point lead over Trump in the HuffPost Pollster average. She’s approaching 50 percent, whereas Trump is struggling below 40 percent. These polls are averaging only 5.5 percent undecided and 4.4 percent for other candidates.
Polls conducted either online or using automated voice technology over the telephone, without a real person on the other end of the line, tell a different story. In these polls, Clinton averages just under 44 percent support, a 5-point drop from the live telephone polls, while Trump loses less than 1 percentage point, still sitting at about 38 percent.
That 5 percent who supported Clinton in live telephone polls, but not in these other polls, seems to have gone to the “undecided” column. In online and automated phone polls, undecideds increase by 5 percent, growing to over 10 percent of voters. Support for “other candidates” increases slightly, to 5.8 percent.
How would that happen? It’s likely that this change is what pollsters call a “mode effect,” which means that there’s a difference in results based on how the poll was conducted. In live interviewer telephone surveys, respondents usually aren’t given an “undecided” option. If someone does say they’re undecided, most of the time the interviewer will ask a follow-up question along the lines of, “If you had to choose today, which candidate would you pick?” or “Do you lean toward any candidate?”
A lot of the time, voters who are undecided do nevertheless have some idea of what they might do. And they’ll often choose a candidate on the follow-up question.
Contrast that with an automated phone or online poll: Respondents are given an option to select “don’t know” or “undecided” on the first question, and in many cases there isn’t a follow-up question to ask whether the respondent is leaning toward a candidate. With the explicit option given, more respondents will choose “don’t know” or “undecided” in the first place. If there’s not a follow-up question to ask undecided voters which way they’re leaning ― or even if there is ― the numbers for “undecided” will be higher in polls conducted this way.
Since it’s Clinton’s support that increases when undecideds are pushed in live telephone surveys, the polls indicate that undecided voters tend to lean more in Clinton’s direction. The “shy Trump voter” theory could still be argued ― perhaps these voters say they lean toward Clinton, but really they lean toward Trump and don’t want to tell an interviewer that. But if that were true, we’d still expect to see Trump’s support drop in live interviewer polls ― and it doesn’t.
Trump’s support is stuck somewhere in the upper 30s regardless of poll method. The “shy Trump voter” theory doesn’t hold up, no matter how you slice the numbers.