Is Hillary Clinton pulling away in Iowa?
On twitter, the online political journal The Hill is touting its story headlined “Iowa polls shows Clinton up 29 over Sanders”. The article refers to a survey conducted by Loras College of 500 likely Democratic caucus goers that shows Clinton leading Sanders 59% to 30%. Not surprisingly, more than one of my students (and strong Bernie supporters) has emailed me asking whether this poll can be “trusted”. Their questions give me an opportunity to remind readers again about the difference between political punditry, as exemplified by The Hill article, and political science. As a political scientist who writes about elections, the Loras poll gives me an additional data point, but one that is noteworthy only because it is more recent – not because the results seem to be an outlier. Indeed, there is no reason to accord a poll any additional weight just because of the results – instead it must be judged in the context of all the polling results. As my students have often heard me say, “Never rely on a single poll to gauge how a race is going.” Nor do I pay much attention to “house” effects in an effort to control for the quality of a poll. I know some data sites make a big deal about this, but Drew Linzer was able to forecast the Electoral College results in 2012 using state-level polls without regard for house effects, on the assumption that they would tend to cancel out on average (as long as you controlled for the frequency of the polls). In short, I tend to treat all polls as equals (assuming they meet standard criteria for conducting a truly random sample of a targeted population), and rely on the polling averages rather than the results of any single poll.
For a political pundits, however, the Loras poll, which suggests a huge lead for Clinton, has a “man bites dog” quality that makes it newsworthy precisely because it does seem to be an outlier. Pundits had the same reaction to a recent CNN/WMUR poll that showed Sanders up big in New Hampshire – it was newsworthy enough to generate headline coverage and much pundit speculation about what the numbers meant for her campaign. Of course, it is standard practice in such stories to note – typically way down in the story – that readers should be aware that it is possible the poll being discussed is an outlier, that other polls suggest a different interpretation, etc., etc. The fact that there may be countervailing evidence that should give us pause regarding how to interpret single poll results, however, is not going to stop political pundits from focusing on those polls that they view as more newsworthy. And frankly, if I was in their position – particularly those working in the print media, an industry that is shedding readers at an alarming rate – I would probably make my choice of which poll to highlight based on what is most likely to generate an audience and ad revenues. From the pundits’ perspective, then, all polls are decidedly NOT equal – some are more likely to attract readers than are others, particularly if the results suggest a change in the prevailing narrative.
My point here is not to suggest one approach is superior to the other. It is to remind readers that we have different objectives. Pundits traffic in informed opinion. Their goal is to enlighten and entertain (not necessarily in that order). Political scientists, in contrast, are interested in developing explanations for (generally recurring) events and in testing those explanations with data. If we also entertain, that is an additional benefit. To be sure, there is often overlap between the two, particularly in sites like mine. I have been known to write the occasional post, such as this edited interview with Sarah Palin, that surely cannot be characterized as political science!
So, where does that leave our understanding of Iowa? As it turns out, three subsequent surveys of potential Iowa Democratic caucus goers have come out since the Loras poll. Two show Clinton up by an average of nine over Sanders, but a third CNN poll has Sanders leading Clinton 51-43. What’s a reader to do? Here’s my advice. First, per my discussion above, rely on the averages. In Iowa, the Huffington Post aggregate pollster still has Clinton with a slight lead.
Second, remember that polling in Iowa is notoriously difficult, in part because it is difficult to figure out who is actually going to show up to caucus. As I have discussed previously (and also in a post coming out at U.S. News on Monday), Sanders is polling quite well among younger voters and those who have never attended a caucus before. Pollsters have to make assumptions regarding how many of these younger Sander supporters are actually going to show up. Because pollsters use different methods for predicting who will vote they generate different turnout predictions, different polling samples and different results.
This should also remind us that pollsters can’t account for differences in candidates’ ground game. This is particularly important in a low-turnout event like a caucus. Pollsters don’t know beforehand whether one side will prove significantly more effective than the other in getting its people to show. But these differences can have a discernible impact on the actual results.
There is a final caveat to keep in mind when judging Iowa caucus polls – they don’t always do a good job forecasting results because initial support going into a caucus may not translate into final numbers. Consider the Iowa Democratic caucus. According to the Iowa Democratic Party caucus rules: “For precinct caucuses, preference groups shall be required to have a minimum number of members within their group in order to be considered viable for the purposes of electing delegates to the county convention. The minimum number of members, or viability threshold, a group must have will be determined by the following factors: the total number of eligible caucus attendees at the particular caucus and the total number of delegates the particular caucus is to elect.”
Translated, this means if a candidate does not attract enough initial support at a precinct caucus – a minimum of 15% of caucus goers for precincts selecting four or more delegates, and higher if fewer delegates are chosen – his or her supporters are given time to reallocate themselves to one of the candidates who did clear the minimum threshold of support. So, initial support may be accurately captured by a poll, but by the time the caucusing is done a candidate whose initial support is low may find his or her backing falling well short of what the polls suggest, while frontrunners may see the support grow beyond the initial polling projections. Remember, when the caucusing is over, what gets reported is the number of delegates awarded in each precinct to each candidate – not the share of the vote each candidate initially received by those attending the caucuses, which is what the polls are measuring.
All this is a long way of saying that polling in Iowa is very much an imprecise science. Moreover, history suggests that there can be considerable movement in candidate support in the last 10 days before the caucus, as both Howard Dean in 2004 and Rick Santorum in 2012 found out. Dean saw his polling lead collapse, while Santorum moved from back-in-the-pack to eke out a narrow victory.
Is Hillary Clinton really pulling away in Iowa? I don’t know. And I suspect no one else does either.