The Exaggeration of Political Polarization in America

For a few years I’ve been fascinated by the idea that, in American politics, the perception of polarization is larger than polarization itself.

I’ve been interested this ever since reading Fiorina’s book which made the point that, if the positions of the Democratic and Republican parties are far apart, then, even if most voters are in the middle, they will perceive a world of polarization. Suppose, for example, that Democratic officeholders are at – 2 (on some scale) and Republican officeholders are at +2, with the average Democratic and Republican voter being at – 0.5 and +0.5, respectively. Then the average Democrat voter will (correctly) see the Republican Party as far-right, and the average Republican voter will (correctly) see the Democratic party as far-left. But the voters are not so polarized.

The challenge is how to measure the perception polarization, to go beyond these sorts of theoretical models to separate perceptions from reality. I recently came across an excellent paper by Jacob Westfall, John Chambers, Charles Judd, and Leaf Van Boven (now that’s a great name) that does just that! Here’s the abstract:

The study uses data collected in the American National Election Studies between 1970 and 2004 to examine Americans’ perceptions of polarization between Democrats and Republicans. Respondents reported their own attitudes on partisan issues, such as whether the government should increase spending and provide more services, and they estimated the attitudes of Democrats and Republicans. Over the years and across issues, people generally exaggerated polarization, overestimating the mean difference between Democrats’ and Republicans’ attitudes. The tendency to exaggerate polarization was larger among people who were more partisan themselves — those who identified more strongly as either Democrat or Republican and held relatively extreme attitudes. Strength of partisan identification and attitude extremity were independently predictive of respondents’ exaggerated polarization. Also, exaggerated polarization was predictive of civic actions (e.g., voting in national elections, making campaign contributions, attempting to persuade others), independent of people’s own partisan identification and attitude extremity. These findings demonstrate the importance of social psychological theory for understanding how Americans perceive the political landscape, and how such perceptions predict political behavior.

Here were my review comments:

I like this paper. I don’t know the literature well enough to know if this analysis of NES has been done before, but if this is a new analysis, it will be very helpful in our understanding!

The paper is generally well structured but the intro is mostly pretty obvious stuff and could be cut. For example, you could get rid of everything on page 3. Get right to the results.

You should also look at this paper by Goel et al which is highly relevant to your argument. That paper discusses differences between polarization and perceptions of polarization by looking at perceived attitudes of friends:

Also this paper by Gelman and Cai which looks at perceptions of presidential candidates’ issue stances, comparing respondents of different ideologies and parties:


Bottom of page 9: forget the chi-squared tests and p-values. Give estimates and confidence intervals. What’s of interest is how much the difference is, comparing real to perceived polarization.

Same on bottom of p.10, p.11, etc.: Just give the results and c.i.’s for differences, please no chi-squares etc. This is distracting. What’s important is the magnitude of the differences, not whether or not they are zero!

Figure 2 is misleading and should be removed, or replaced by raw data. The quadratic curve is not real, it’s just fit from a quadratic model. Without the actual data I think it is misleading.

[Cross-posted at The Monkey Cage]

Andrew Gelman

Andrew Gelman is a professor of statistics and political science and director of the Applied Statistics Center at Columbia University.