Trump underperforming his polls?
News coverage of polling in the Republican primaries pointed out that Trump underperformed his polling averages. In the set of ten states where FiveThirtyEight had enough polling data to compute a polling average, Trump’s polling margin against Nikki Haley was 7 percentage points better on average than his eventual vote margin (see table below). But how much of this is Trump’s poor performance, and how much of this is the result of primaries being famously hard to poll?
One way in which primaries are hard to poll is that different states have different rules on who can participate. “Closed” primaries and caucuses (like those in Iowa and South Carolina) and require that voters be registered with a party in order to participate in that party’s primary. In other “open” primaries, like those in Michigan and Virginia, anyone can choose to vote in any primary. Finally, other states have a variety of rules that fall between these two extremes, the most common of which is allowing voters registered with no party to vote in either party’s primary, such as in New Hampshire, which we will call “semi-closed” for purposes of simplification.
Breaking out Trump’s under- and over-performance by these categories tells a somewhat more nuanced story. In the three closed primaries, Trump’s average underperformance on his margin against Haley was only 2 percentage points. In the three semi-closed primaries, Trump’s average underperformance was 5 percentage points (median is 7). And in the four open primaries, Trump’s average underperformance was 13 percentage points.
A reasonable inference from this pattern is that more open primaries drew in more independents and Democrats to the Republican primary, given the uncompetitive nature of the Democratic primary, and that these voters disproportionately favored Haley over Trump. Because more open primaries create more uncertainty about who to sample for primary surveys, it is unsurprising to see more polling error in such circumstances. And that adds additional difficulty to the challenge that polling in lower turnout primary and special elections presents in identifying who is likely to vote, as Politco’s Steven Shepard has pointed out. This pattern may also reflect a dislike of Trump among independent voters, but it would be a mistake to think that all of these extra Haley voters were Republican voters in the first place.
