A few weeks about I threw together some thoughts on the "fundamentals" of this Democratic primary election: primarily, demographics. I wrote, in part:
“Sanders deserves credit for being able to build a campaign that allowed him to reach his maximum feasible support in the Democratic primary. But fundamentals are that for a reason: you either harness a winning demographic coalition or you find a way to change the demographic makeup of the electorate.
Sanders was not able to do either and that, more than anything to do with the much-hawked “process,” is the reason he is not the presumptive Democratic nominee.”
Last night, Hillary Clinton managed dominating wins in New Jersey and California, and with them secured the Democratic nomination.
The biggest delegate prize last night was California. Polls showed the state tightening over recent weeks, and the most recent set of polls all showed Clinton’s lead over Sanders closing to a roughly 2-point margin (It’s a different story all-together, but there’s certainly some evidence to suggest the polling failure in CA was a result of poll herding, defined here.).
Meanwhile, the demographic model I’ve been using this cycle to assess the viability of Bernie Sanders always showed him at a significant demographic disadvantage in the state. In fact, it was predicting an outcome similar to that of the New York primary, which makes sense – it’s a very large, very diverse state (New York, unlike California though, was a closed primary). And indeed, the demographic model was correct – Sanders only managed to get 43% in California, after camping out in the state for the last two weeks.
Now that the primary is wrapped up (minus the contest in Washington, DC next week), I wanted to use this as an opportunity to look back at the outcome of the demographic-based model. As a refresher, the model was pretty straight-forward, with state-based predictors including percentages of white men, white women, African Americans, self-identified Democrats, self-identified Liberals, and non-college white voters in the electorate.
In the end, the model incorrectly predicted 4 states: Arkansas, Ohio, Massachusetts, and Missouri. Notably, all four are states that Sanders was modeled to win. On the flip side, there were no states that Clinton was modeled to win that she lost. The average error across all states was -2.65% (which includes Arkansas, where a -22% underperformance by Sanders skews the overall average, and West Virginia, where Sanders vote margin was impacted by protest vote for lower-tier candidates).
What’s notable, as I wrote about last week, is that – relative to the demographic model – Sanders performed the same in later primaries as he did in the earlier primaries. Fundamentally, Sanders maintained his core base of younger, white voters throughout the primary, but was unable to remedy his substantial disadvantages among non-white voters. That’s the lesson on this campaign.