Beyond Left and Right: Cluster Analysis of German Party Platforms

A few weeks ago, I wrote for FiveThirtyEight explaining the 2017 German federal elections in the context of American politics and used data from the Manifesto Project to show where the German and American party platforms have fallen on the left-right spectrum since 2000. As some pointed out, however, there is more to understanding the similarities and differences between parties than this one-dimensional measure; ultimately, that measure was used because it was easy to understand at a glance. Here, I want to illustrate another way I used the Manifesto Project data to compare parties that I think can be useful for more nuanced cross-country political comparisons.

The number that I used from the Manifesto Project is a “RIFLE” score, a number calculated from the percentage of the party platform dedicated to certain themes. However, the Manifesto Project captures much more than these dimensions; its coders evaluate the entire party platform and categorize each discrete statement according to the project coding system, which can be found in their codebook. Using all of the basic dimensions that the Manifesto Project uses, I clustered the German parties since 2000 into two and then four clusters. I then predicted which of these clusters the American party platforms would fall into for this same time period.

subway chart@2x.png

When I used two clusters, the three most right-leaning German parties fell in one cluster, and the more left-leaning ones — the Greens, the Left, SPD, and the Pirates — fell in the other. The American parties generally fell in with the conservative German ones, with the exception of the Democrats in 2012. However, when I used four clusters, CDU, FDP, and AfD all wound up in their own clusters, while the liberal parties remained in one.

This captures some important points that a purely left-right comparison misses: AfD has more significant differences from these other “right-leaning” parties than is indicated by its position very near the CDU on right-left spectrum. Additionally, while the RIFLE score indicates that the US Republican Party is far out to the right of any German one, taken on the whole in the clustering exercise, their platform most closely resembles the moderate CDU.

While selecting a subset of issues can unfairly amplify the differences between parties when those issues may or may not be a significant portion of the party platform, it is quickly understandable to a US audience accustomed to thinking in terms of marquee issues and “left” vs. “right.” Meanwhile, a clustering analysis captures a more comprehensive picture of similarities and differences, but requires background knowledge to tell you anything about the ideology of those clusters. In order to have it both ways, perhaps the best solution is to present a simpler metric — such as a left-right measure or particular issue positions — along with the results of a more complex analysis to provide both context and nuance.

Anna Hazard