As we talked about previously: with such a large field (and with it being this early) in the 2020 Democratic primary campaign, it is difficult to measure candidate support in a useful way. This problem is even more acute when looking at differences across groups, since the data becomes even harder to parse as the number of observations decreases. None of this has stopped pundits from discussing age gaps, ideology gaps, and race gaps between supporters of various candidates. This analysis relies on comparing topline support between different demographic groups, which means that it is generally only comparing voters’ first choices and is therefore ignoring the broader contours of support.
In the previous blog, I used survey data and a random utility model to identify support for candidates beyond the number-one choice. I extend this method here, using new data from a Data for Progress/YouGov Blue and a MaxDiff model to estimate variation in support across different subsets of voters. The Data for Progress/YouGov Blue survey ran from May 30 to June 3, and includes 625 voters whoindicate they will probably vote in the Democratic primary. To measure support, they gave voters sets of five candidates to compare and asked them to select their top and bottom choice from those five candidates. I then modeled that selection as a function of latent candidate support, which itself was a function of voter demographics: ideology, race, sex, age, income, and education. And the end results show candidate support for different demographic groups while holding all else constant. Formally, this is a Hierarchical Bayes MaxDiff model.
Let’s start with the least surprising result: candidate preference breaks down by voters’ ideological identities. The figure below shows the latent level of support for different ideological groups, holding all else constant. Joe Biden performs best among liberal and moderate voters, while having the lowest level of support among very liberal and conservative voters. In contrast, Harris, Warren, and Sanders all perform the best among very liberal voters, with a slight drop off when moving to liberal voters. On the other side, Michael Bennet performs the best among more conservative voters and those who identified as “other.”
Once controlling for ideology, there is little variation by race—in contrast to the conventional wisdom that political pundits frequently put forward. Black voters tend to support Joe Biden more than nonblack voters, but the difference is not stark. Beto O’Rourke and Julian Castro also receive a large amount of support from Hispanic voters. Although not displayed here, the only candidates that receive a disproportionately large amount of support from White voters are Seth Moulton, Steve Bullock, and John Hickenlooper.
In contrast, the breakdown by age generally tracks with common narratives. Joe Biden receives high support from older voters and is at his worst among those in the 30-44 age group. Bernie Sanders is almost an inverse of this, with younger voters supporting him more. Interestingly, Kamala Harris follows Biden’s pattern more closely, while Elizabeth Warren has a consistent level of support across all age groups.
Finally, there are only small differences among voter sex. The biggest gap is for Elizabeth Warren, who receives substantially more support from women than from men. Among the top of the field, almost all receive more support from women, and it is not until you get lower in the field that the relationship flips.
This analysis finds support for some common narratives regarding the 2020 field. Joe Biden receives high support from moderate voters, while much of the rest of the field is garnering support from liberal voters. There are definitely some differences across age as well. However, once controlling for other factors, there aren’t huge differences in support by race or sex—cutting against these common narratives. Where differences exist, they are not always what we would expect: women appear to generally be more supportive of the top of the field, including Biden.