4 Comments

Many thanks for the charts!

Expand full comment

I omitted 1992 because of the obscuring Perot factor, the most consequential 3rd party candidacy since 1912. In looking for patterns, it helps to go back a bit further than the presidential year, since that election is often itself an outlier -- 1966 for the most part returns to 1962 and earlier, for example, as does 1982

Expand full comment

In 2020, the Democrats retained their house majority while losing seats (13 compared to 2018 elections) and regaining the presidency. That's an unusual combination, indicating some cross-currents. In the postwar period, a similar pattern only happens twice, with the Democrats in 1960 and the Republicans in 2000. In the subsequent midterm elections, the president's party lost (election to election) 6 seats in 1962 and gained 8 seats in 2002. Not exactly strong historical support for massive Democratic losses in 2022 -- which may happen, just not because of an inevitable historical repeat.

Expand full comment
author

Douglas,

I hear you, but I think your comment is at risk of overfitting to past patterns — and ones that may be outliers at that. There are some obvious reasons 2002 might not be instructive of future outcomes, for example. And Democrats lost 9 seats in 1992, similar to their 2020 loss, — and then went on to lose 54 more in 1994!

As with every relationship, there are many ways to talk ourselves into believing an outcome should fall toward the extremes of the prediction interval. But the best bet is usually the mean or median of our generated predictions. And after all, that’s why we have an error term!

Expand full comment