Perceptions of Candidate Ideology

By Kevin Reuning (@KevinReuning)

As discussed in a previous post, ideology is a tricky concept. There are no clear measures of ideology as there are deep and important disagreements over how issues fit together into a cohesive world view. One way to avoid these tricky issues is to instead rely on voter’s perceptions of ideology. Perceptions of candidate ideology are likely just as important as any measure from objective candidate positions, as voters will vote (and donors will donate) based on which candidates they perceive as being close to them. Using the 2018 Cooperative Congressional Election Study (CCES), It is possible to measure these perceptions of ideology for a large set of candidates that held Congressional or Gubernatorial office in 2018.

In the 2018 CCES, 60,000 respondents were asked to provide an assessment of the ideology of Donald Trump, the Republican Party, the Democratic Party, the Supreme Court, and the incumbent and challenger in their local races for Senator, Governor, and House of Representatives. Respondents placed each of these politicians on a 7 point scale from Very Liberal to Very Conservative. The responses by themselves are potentially problematic as conservative voters will see everyone to the left of them as liberal while conservatives are moderate. However, a statistical technique known as Aldrich-Mckelvey scaling (named for the political scientists who developed it) accounts for this bias by assuming that each voter’s response for a candidate is a function of the candidate’s true ideology, and the voter's ability to perceive ideology (a slope parameter) and a bias that pushes voters perceptions to the left or right (an intercept). The parameters can be estimated as long as there are multiple overlapping ratings from each voter. Since every voter scored Trump, both parties and the Supreme Court, we can estimate their slope and intercept and use it to generate estimates of other candidates’ ideologies. Because of the large number of parameters and field, I use Stan’s Variational Bayes algorithm to estimate the parameters (meaning that these are more approximate than most estimates).

 The figure below shows the estimates for each Presidential candidate in the data, as well as the four bridging ratings (the court, the parties, and Trump). The scores fit our general expectations. Bernie Sanders is perceived to be the most liberal candidate while Tulsi Gabbard is the most conservative. Interestingly the Democratic Party is viewed as being in about the middle of the 2020 candidate pack. The Supreme Court is viewed as slightly right of center while Trump is viewed as more conservative than the Republican Party as a whole.


Along with viewing the candidates against each other, this data allows us to compare the perception of the 2020 field with all the 2018 candidates and incumbents included in the data. What is striking is that the Democratic field is relatively well spread across the entire range of 2018 candidates/incumbents. Bernie Sanders Is perceived as being slightly further out to the left that might be expected given the general distribution, but the rest are generally within the overall shape of the party. Donald Trump was also solidly to the right of the bulk of Republican candidates and incumbents.


Finally, we can compare the field to how voters perceive themselves. Again, the candidates are not far outside the bounds of the variation in voter ideology. Voter perceptions of themselves are more extreme than their perceptions of the 2018 candidates so this pushes the 2020 field towards looking relatively moderate, even Donald Trump. It is worth noting that part of this might be a result of a limitation of the model. The extreme lumps on either side of the plot are likely the result of voters rating everyone as being very far to one side while they are far to the other side. So the blue, very liberal set of voters, placed each candidate as being Very Conservative and themselves as Very Liberal. This pushes their ideology to the extreme and makes It hard for the model to parse out exactly how far they are from the mean voter. The edges then should be taken with a grain of salt.


As I’ve discussed before, there is no reason to believe that the entire 2020 Democratic field is extremely liberal. Across the potential candidates, on multiple measures, there is a range of ideological positions. Importantly, this shows that the ideologies within the 2020 Presidential field are not out of step with previous candidates or the voters. There is always a need in the media to make things sound new, but the 2020 field is not a radical break from the past. 

Kevin Reuning (@KevinReuning) is an assistant professor of political science at Miami University.

The VB estimates were compared to NUTS estimates where only the scores of the Presidential candidates and the bridging items were included. The estimates were similar across both methods. If you are curious, the NUTS estimates are here.

Code available on github for all estimates and plots. 

Subscribe to our mailing list

* indicates required