These Democratic Senators Should Be Afraid for Their Jobs

Chelsea Manning’s recent announcement that she would challenge incumbent Ben Cardin in Maryland’s Senate primary set off a flurry of debate about the legitimacy of her challenge. Beyond the specifics of Manning’s run, there’s a larger question: Should progressives try to unseat Democrats they judge to be to be too centrist, or should they hold off in the name of party unity? Already two other Senate Democrats, Dianne Feinstein (California) and Joe Manchin (West Virginia), face primary challenges from Kevin de Leon and Paula Jean Swearingen, respectively, both of them left-wing insurgents. How should progressives determine where to allocate time, attention, and money when choosing incumbent Democrats to challenge from the left?

Let’s pursue an answer to this question in the most progressive way possible: by looking at a bunch of data.

There are a number of reasons why progressives might want to primary a particular member of Congress. But for the sake of our analysis, we’re going to assume that for any given seat, the Democratic electorate will want to nominate the most progressive candidate who can regularly win it. By this standard, the best targets are incumbent Democrats who are significantly more conservative than a generic Democrat from their constituency would need to be in order to win reelection.

We can start by comparing incumbent Democrats who are from the same state, and as a result are in theory accountable to the same general electorate. By this standard, it isn’t shocking that Cardin drew a primary challenge. According to FiveThirtyEight’s Trump Score, Cardin votes with the Trump administration noticeably more often than Chris Van Hollen, Maryland's other senator (26 percent of the time for Cardin, compared to 21 percent for Van Hollen). This difference between the Maryland senators is not limited to the Trump era: Cardin’s first dimension DW-NOMINATE score, a measure of relative ideology over the course of a member’s career, (-.328) is far more conservative than Van Hollen (-.399). Cardin also drew ire from the progressive base when he voted against the Iran deal, Obama’s signature second-term foreign policy achievement.

All this suggests that Cardin, who hails from a safely Democratic state, has room on his left flank such that a primary challenge isn’t all that surprising. The challenge to Feinstein is also predictable. Feinstein votes with Trump 31 percent of the time, while her Democratic colleague from California, Kamala Harris, votes with Trump only 17 percent of the time.

However, this simple method won’t work for all Senators or House members. Since Joe Manchin’s counterpart in West Virginia, Shelley Moore Capito, is a Republican, it wouldn’t make sense to directly compare their roll-call voting to see if primarying Manchin is a good use of the base’s time and energy. Moreover, it could be the case that Cardin and Feinstein, while being the more conservative of their states’ two Democratic senators, aren’t out of step with their states’ electorates—perhaps Van Hollen and Harris are more liberal than their constituencies.

To address these potential issues, we've specified a model to help progressives understand who it’s most worth their time to primary. Though there are many measures, such as Progressive Punch and DW-NOMINATE, we use the Trump Scores for votes cast in 2017, something that is immediately relevant to the Democratic base. Our model predicts Trump Score using four constituency-level variables: Trump’s margin of victory (or loss), white share of the voting age population, the percent of the constituency over the age of 25 with a bachelor’s degree, and median household income. For each senator, we temporarily removed them from the dataset, re-specified the model using all other Democratic senators, and used that re-specified model to generate a predicted Trump Score for the missing senator. The result is an estimate of how often a hypothetical replacement Democrat that emerged from any given electorate would vote with Trump based on that electorate’s political, racial, educational, and economic profiles.

Comparing these replacement-level predictions to reality reveals which constituencies could likely sustain more progressive senators. The chart below shows the results: The x-axis represents each senator’s replacement-level prediction, the y-axis is their observed Trump Score, and the black line represents replacement-level voting behavior. Democratic senators who are 5 percentage points above their replacement-level Trump Scores are labeled:

So that gives us a list of targets that looks like this:

• Angus King of Maine has a predicted Trump Score of 27 and an actual score of 45. (+18)
• Mark Warner of Virginia has a predicted Trump Score of 27 and an actual score of 43. (+16)
• Patrick Leahy of Vermont has a predicted Trump Score of 12 and an actual score of 27. (+15)
• Dianne Feinstein of California has a predicted Trump Score of 19 and an actual score of 31. (+12)
• Benjamin Cardin of Maryland has a predicted Trump Score of 21 and an actual score of 27. (+6)
• Tim Kaine of Virginia has a predicted Trump Score of 28 and an actual score of 34. (+6)
• Bill Nelson of Florida has a predicted Trump Score of 32 and an actual score of 38. (+6)
• Joe Donnelly of Indiana has a predicted Trump Score of 42 and an actual score of 47. (+5)
• Brian Schatz of Hawaii has a predicted Trump Score of 22 and an actual score of 27. (+5)
• Michael Bennet of Colorado has a predicted Trump Score of 26 and an actual score of 30 (+4)

King is a special case in that he runs for office as an Independent, only joining the Democratic caucus once the congressional session convenes; his score is closer to an estimate of the anti-Trump value progressives would realize by having a Democrat beat him in a general election. King and Warner are deficit hawks, with both touting the 2010 Bowles-Simpson plan that would have made deep cuts to Social Security. According to the 2016 American National Election Study, 68 percent of Democratic primary voters want to expand Social Security, and only 3 percent support cuts (the rest want spending to remain the same).

Bennet also appears on our list—progressives may remember his vote against the filibuster of Gorsuch. Then there’s Leahy, with a Trump Score of 27—far higher than Bernie Sanders (14), who represents the same electorate. Manchin, despite the ire he raises from progressives, does not crack the top ten in terms of Trump Score Above Replacement. While his Trump Score is higher than any other Democrat, he also represents a more conservative electorate than any other Democrat, putting his Trump Score almost exactly in line with what our model predicts a hypothetical replacement would have. Tom Carper, a senator from Delaware who is quite conservative on Social Security, barely misses our list at number 11.

• Cory Booker of New Jersey has a predicted Trump Score of 28 and an actual score of 14. (-14)
• Jeff Merkley of Oregon has a predicted Trump Score of 26 and an actual score of 12. (-14)
• Kirsten Gillibrand of New York has a predicted Trump Score of 21 and an actual score of 9. (-12)
• Tammy Baldwin of Wisconsin has a predicted Trump Score of 33 and an actual score of 21. (-12)
• Ron Wyden of Oregon has a predicted Trump Score of 26 and an actual score of 18. (-8)
• Elizabeth Warren of Massachusetts has a predicted Trump Score of 18 and an actual score of 11. (-7)
• Tom Udall of New Mexico has a predicted Trump Score of 30 and an actual score of 23. (-7)
• Richard Blumenthal of Connecticut has a predicted Trump Score of 26 and an actual score of 20. (-6)
• Sherrod Brown of Ohio has a predicted Trump Score of 36 and an actual score of 30. (-6)
• Kamala Harris of California has a predicted Trump Score of 21 and an actual score of 16. (-5)

The list of more-progressive-than-replacement Democratic senators doesn’t surprise. Booker, Merkley, and Gillibrand, who have distinguished themselves with their anti-Trump votes, stand out. Booker and Gillibrand are both seen as presidential contenders (as is Harris, whose Trump Score Above Replacement is limited by California’s liberalism). Merkley has opposed Trump particularly on civil liberties grounds. Baldwin, Udall, and Brown are liberal Democrats who represent reasonably swingish states. Warren and Blumenthal mostly don’t appear higher on the list due to how liberal their electorates are.

Our model should not be the end of discussion. For one, measures of roll call voting are imperfect for analyzing aggregate liberalism, and the Trump Scores are limited in many respects. Specifying models with scores like Progressive Punch or DW-NOMINATE could produce different results.

Even if progressives agreed on who to primary, finding viable candidates could be a key problem. For instance, while Angus King is the worst member of the Senate Democratic caucus according to our model, there isn’t a sizable bench of progressive candidates who could run statewide in Maine—a problem that doesn’t exist in Florida, full of ambitious relatively young politicians eager to move up.

In addition, politicians could be strongly above-replacement on issues that they take seriously, or be deeply effective legislators, or have other assets we have not accounted for in our data. Cardin, for instance, was instrumental in requiring dental benefits be covered in the CHIP program after Deamonte Driver, a 12-year-old boy from Maryland, died when a cavity became a deadly brain disease. And, on the flip side, a candidate who is otherwise a round replacement but took one particularly conservative vote may be worth a challenge.

Finally, it’s worth at least noting that a similar method of identifying Republican candidates suitable for primary challenges would be impossible because Senate Republicans’ Trump Scores are not meaningfully associated with their general election constituencies. So it may be that Democrats are acting irrationally by allowing their voters to determine their votes—that LOL nothing matters and every Democrat should vote like Gillibrand and Booker. Here's a graph of Trump Score and Trump's statewide margin of victory in the Senate broken down by party:

With these caveats, our model suggests that there are several Democrats ripe for primary challenges in the Senate. Even if they are not ultimately unseated, facing pressure from their left may bring them back in line with the constituencies from which they have emerged. Indeed, Dianne Feinstein has made some more anti-Trump actions since her challenger announced, such as releasing the Fusion GPS transcripts (some have attributed the move to her facing a primary). By reframing primaries away from simply who are the least anti-Trump Democrats to who are the Democrats whose seats could sustain a more anti-Trump senator, progressives can win primaries and influence policy.

Sean McElwee is a researcher and writer based in New York. Follow him on Twitter.

Jon Green is a PhD student in political science at Ohio State University. Follow him on Twitter.