Overall, this approach has performed well. In most years, it predicts the average underperformance of Democrats in Washington state vis-a-vis the previous Democratic presidential candidate’s performance quite well; 1992, 1998, and 2016 are outliers, but the overall error is almost exactly zero.

What would this tell us today about Washington? The model suggests that Democrats should run, on average, about 3.8% behind Joe Biden’s vote share from 2020. That would reflect one of the larger expected midterm drop-offs in our dataset, akin to 2014 and 1994, but not what we saw in 2010. (Notably, though, Barack Obama won in 2008 by a larger margin that Biden won in 2020 or Clinton won in 1992.) We should also remember that error margins are real, and that although the model performs well, it is still imperfect.

Finally, can we extrapolate from this to the national vote? If Washington Democrats are expected to run more or less 3.8 points behind Biden in their districts, does that mean that all Democrats are expected to run more or less 3.8 points behind Biden in their districts? If so, this would actually translate to modest Republican gains; they’d win the popular vote by about three points. The final table shows how far Democrats nationally ran behind the previous Democratic performance, what the model predicted for Washington state, and the differences.