Just cause you lean don’t make you a pushover: Party Identification Measurement in Polling

Party identification, i.e. self-identification as a Republican, Democrat, Independent or something else, is one of the most valuable pieces of information modern electoral pollsters collect about their respondents. Armed with party ID information, we can make pretty accurate guesses about how people feel about issues, what stories will appeal to them and, ultimately, how they will vote. It is an essential poll metric for undertaking any type of socio-political analysis of a population. Some research organizations even use it as a weighting variable. However there are two challenges inherent to measuring party ID: 1) there is no industry-standard, foolproof way of identifying party ID, and 2) there are no “true population” statistics for party ID against which we can benchmark our measurements.

We will address the latter issue first: there are no true population statistics for party identification because party ID is a value or opinion, and not a demographic characteristic. Party ID, by definition, means which party you ‘most identify with’. As the definition implies, this identification is not fixed; it can change over time and is subject to the opinions of the respondent. While much existing research (Campbell, Converse, Miller and Stokes, 1960) has shown that for most people party identification is something they ‘inherit’ from their parents and keep throughout their lives, other research (Fiorina, 1981; Brody & Rothenberg, 1988; Franklin & Jackson 1983) has shown that population-wide, party ID ebbs and flows based on a number of factors.

As a community, we do have some guidance based on the very high quality work done by the General Social Survey (link) and the American National Election Studies (link). The ANES provides estimates of party ID statistics based on annual surveys of the US population. However, since the ANES is only conducted once a year and is usually slightly behind current, it can only provide a macro estimate for party ID and does not provide much guidance on the monthly and weekly variations we often see. Below is data from the ANES on party ID from 1968 through 2008. Note the variation year over year.

In the absence of a standard metric for party identification, the research community has developed a multitude of ways to measure party identification. This goes back to the first challenge mentioned above: no standard industry question. Every pollster has their own recipe for the question wording, response options, follow-up questions and reporting practices. Here is how Ipsos asks party ID:

One of the most important parts of our question, and the part that causes much of the variation in reported party ID splits, is the follow-up question for Independents and others. We call this our “push” question. It functions by “pushing” people who nominally identify themselves as Independents to pick the side (Democrat/Republican) they most agree with or the one they “lean” towards. When we report our final party ID figures we include these “leaners” with our Democrats and Republicans. This causes our Democrat and Republican populations to appear high and our Independent population to appear low compared to other pollsters who do not push independents. The results, both pre and post push question, from our January through April 2012 party ID question is:

We push our leaners for two important reasons. First, we’ve been doing it for a decade and want to keep our trend data. Second, and more importantly, we have discovered that “leaners” actually behave very similarly to regular partisans when it comes to casting a vote. When we focus on horserace questions (such as Obama vs. Romney), separating leaners from the true Independents and including them with partisans actually yields a more accurate depiction of likely voting behavior. Data from our horse race question, Jan-April 2012 divided by our full party ID spread:

This data indicates that leaners are actually more similar to regular partisans than towards true Independents. Our take is that reporting the “unpushed” independents obscures this important data and 1) makes the independent population seem larger than it really is and 2) makes the election appear more volatile than it really is.

However, we continue to receive numerous inquiries into our Independent population and party ID splits. Therefore, moving forward Ipsos is going to publish both our “pushed” version of the party ID and the unpushed numbers. We feel this will offer clarity to the situation, and it also further reiterates Ipsos’ commitment to a transparent polling process.