Beware the Iowa polls

It sounds self-congratulatory (OK, it is self-congratulatory), but I say it with purely pedagogical intent: Four years ago, about the only truly accurate polls of the Iowa caucuses were ours.

It sounds self-congratulatory (OK, it is self-congratulatory), but I say it with purely pedagogical intent: Four years ago, about the only truly accurate polls of the Iowa caucuses were ours.

The secret to success here is simply stated, though less easily implemented: Poll the right people. Surveying Kansans won’t tell you much about the Iowa caucus results, and neither will polling the 2 million adult Iowans who will not attend a caucus. Yet that is exactly who most Iowa caucus polls survey.

Voters who take the time and make the effort to attend a caucus are different. At a minimum they are more likely to be interested in politics and committed to a candidate, but they are also more likely to have been contacted by one or more of the campaigns. Indeed, candidates will spend millions reaching out to these voters in every conceivable way, which further distinguishes them from the balance of the electorate, who glean all their information from news reports and paid ads.

Most polls that purport to provide Iowa caucus results survey the nearly 3 million residents of Iowa, asking whether each respondent is registered to vote and then either whether they have participated in or are likely to attend one party’s caucus or the other. For a variety of reasons people tend to say yes, and as a result, responses to these questions are so highly inflated as to be worthless. If everybody who said they had attended or would attend a caucus actually showed up, turnout would be about five times higher than it was at its highest.

I know the details on the Democratic side more intimately, so I will focus there, but the argument should be similar for the GOP caucuses as well.

The only accurate way to poll the Iowa caucuses starts with the party’s voter file, which provides some data on past caucus participation. Four years ago the Democrats charged about $65,000 for this vital tool. Not one public pollster purchased this database, despite the fact that it is impossible to develop a useful sample without it. Yet poll after poll purported to tell us which candidate stood where in the caucuses.

That list is valuable to campaigns because it helps guide their voter contact programs. The most likely caucus attendee is one who has been before.  Campaigns lavish attention on those voters, rendering them a different species from the ordinary Iowa registrant.
However, only polling those who have attended previously is not sufficient either. The caucuses do not simply recycle the same participants year after year. Our analysis of the Iowa voter file indicated that 56 percent of those who attended the 2000 caucus had never been to one before. That tracks fairly well, though not precisely, with exit polls from 2004 (actually entrance polls), which found that 45 percent had never been to a caucus before. (There are ways to poll these first-time participants more precisely, too, but I’m not giving away all the secrets here).

The Iowa caucuses present an almost unique polling challenge. Roughly half the participants will come from a very small group of voters who have participated in caucuses before. Without the party list there is no way to know who those folks are, and thus they will be vastly underrepresented in any statewide “caucus” poll. The other half will be voters who are particularly motivated, but who have not caucused in previous years, another hard-to-locate segment.

So as you attempt to sort through a raft of Iowa polls over the next year, remember that most of them have not gone to the expense of trying to be accurate and are often just plain wrong.

Mellman is president of The Mellman Group and has worked for Democratic candidates and causes since 1982, including Sen. John KerryJohn Forbes Kerry2020 Dem contenders travel to key primary states When it comes to Colombia, America is in a tough spot 36 people who could challenge Trump in 2020 MORE (D-Mass.) in 2004.