By Kevin Reuning (@KevinReuning)
The 2020 Democratic Primary features a large number of candidates sending signals regarding their ideological commitments, with many of them attempting to claim the label of being the most “progressive.” But what do they mean when they say “progressive,” and which of them is, in fact, the most progressive? While being progressive is inherently vague and context-dependent, we can provide more than mere guesswork and armchair assumptions. The Washington Post recently started providing data on what stances each of the Democratic candidates are taking on a variety of issues. We can use this resource to compare candidates across issues and provide an overall metric of where each candidate stands comparatively.
For those of you interested in the details, I explain in full how I use a latent variable method to estimate candidate progressivism further below. In general, I combine the positions collected by the Post along with a data collected from DFP volunteers on how progressive positions are relative to each other to place the candidates in a single dimension of ideology. This is done by assuming that each candidate selects the position that is closest to their own ideology.
Based on this analysis, the most progressive candidates are Elizabeth Warren and Cory Booker. They are just ahead of a handful of candidates including Tusli Gabbard, Kristen Gillibrand, and Julian Castro. This is no surprise as Warren is the most liberal senator (and Booker is the third-most liberal) according to NOMINATE -- a measure constructed using a similar method using roll-call votes. Though to be clear, this outcome is likely in part attributable to the issue areas on which the Post have recorded positions and which they have not. For example, the Washington Post has not collected any LGBTQ or foreign policy positions. In addition, these data do not track things like whether a candidate is joining a picket line, whether a candidate had unionized their staff or other important metrics of progressivism. These positions don’t encompass the breadth of a candidate’s career or their fundraising sources.
Joe Biden is right in the middle of the pack (either the 11th most progressive candidate or the 13th least progressive, depending on which way you look at it). He also has some of the largest credible intervals, meaning that his position is relatively hard to pinpoint. These two characteristics, that he is in the middle, but could be more progressive or more conservative, are likely features and not bugs of his campaign, as journalists covering the race have frequently noted that he has been less committal on policy than many of his rivals on the trail.
The most conservative candidates are John Hickenlooper, Steve Bullock, and John Delaney. These three are not just conservative in comparison to the most progressive but are substantially more conservative even compared to the more moderate candidates like Beto O’Rourke or Eric Swalwell.
Along with being able to identify who is more progressive, we can also estimate what issue positions are more or less progressive. For each issue, Washington Post recorded either two or three positions (excluding the No Clear position). The two figures below show where each of these positions are placed on the same scale as the candidates. They are placed next to the other potential positions that could be held on the same issue.
There are a lot of different policy positions, so here I just highlight a few. Endorsing the Green New Deal is clearly one of the more progressive policy positions that a candidate can take right now. Other similarly progressive positions include banning fossil fuel exports, packing the Supreme Court, repealing penalties for those apprehended crossing the border, and providing health care to undocumented immigrants as part of a government-run plan.
The conservative positions are not necessarily just the opposite sides of these. For example, supporting E-Verify requirements (but only if it is improved) is one of the more conservative positions, and is also likely one of the reasons that Bernie Sanders does not score higher on the candidate scale (as well being open to ending the filibuster when other candidates unequivocally support it). In addition, maintaining the Electoral College, opposing a fracking ban (without at least wanting to regulate it) and continuing to lease land for fossil fuel extraction are all identified as very conservative stances.
To estimate these models I started with the data provided by the Washington Post. For some of these policies, there is not a clear position that is more progressive (nuclear energy, for example, is a very disputed policy domain) and across policies, it isn’t clear which policy is more progressive (is eliminating the Electoral College more progressive than term limits for the Supreme Court?).
The first step then is to provide some outside information on policy positions. I do this two ways: through priors (which I will address later) and through a survey of DfP volunteers. The survey asked participants to indicate which of two policy positions is the most progressive position. Positions were compared both within a single policy and across policies. From this data, it is possible to build a model of how the survey respondents selected which policy is more progressive assuming that each position has a position on a latent progressive scale. This model looks like:
i is 1 if policy position j was labeled as more progressive and 0 if k was. θ
j,m and θ
k,n were the latent progressivism of position m on policy j and position n on policy k. If k==j then this is a comparison of positions on the same policy. Finally, Β
i is a parameter for each survey respondent that captures how well they did overall in identifying how progressive policies were and is a logit transformation. The priors come back in by placing very loose priors on each theta under the assumption that lower coded policies were more conservative. This looks like: θ
j,1 ~ N(−0.5, 0.5) and θ
j,1 ~ N(0.5, 0.5) if there are two positions or θ
j,1 ~ N(−0.5, 0.5), θ
j,2 ~ N(0.0, 0.5), and θ
j,3 ~ N(0.5, 0.5) if there are three positions. These loose priors do not constrain the model from finding that the third position is actually the most conservative policy.
The next step is to connect the candidates θc latent position to these policy positions. Here we assume that each candidate selects the position that is closest to them among positions. This looks like:
In brief, this says that the probability of selecting a position is a function of how far away a candidate is from that position compared to how far they are from the other positions as well. The Βj term that controls how indicative this policy domain is on the overall positions. The candidate ideology variable (θc) is given a standard normal prior.
To see the benefit of this process we can compare the estimates generated by the latent variable model to those estimated by a simpler scale. One of the simplest ways to measure something like ideology is to add up the number of progressive positions a candidate takes. In this case, not all candidates take every position so instead of adding we take the mean score where a 0 is the most conservative position, a 1 is the most liberal, and when there are three positions 0.5 is the moderate position.
In some cases the estimates are similar, Elizabeth Warren and Cory Booker are both near the top on each scale, while Hickenlooper is at the bottom of both scales. Other candidates, like Steve Bullock, move from being moderate to being identified as very conservative. Bullock does not have many policies recorded by the Washington Post (in fact only environmental policies), and he does take some more progressive stances. He would re-enter the Paris Accords with strengthened targets, supports a carbon tax and at least wants to regulate fracking. In the naive, mean estimates, these positions are counted as equally as meaningful as his opposition to the Green New Deal or the end to fossil fuel leasing. The latent variable process, however, finds that opposition to fossil fuel leasing should carry more weight as a very conservative stance, while the positions he supports are more moderate.
The conservative media apparatus likes to discuss the Democratic field as if there is a lot of uniformity––that all candidates are extremely far left. Here, I have shown not only which candidates, as of now, appears to be the most progressive, but also the degree of variation in the field. Democrats have a lot of choice in 2020. For those interested in supporting the most broadly progressive candidate, Warren appears to be a good choice. Not only is she taking strongly progressive positions, but recent polling has shown her support growing.
Kevin Reuning (@KevinReuning) is an assistant professor of political science at Miami University.