Andrew’s talk today about the history of political polling includes:

19. Off-year elections and party balancing

Bafumi et al. (2010) describe balancing as:

the electorate boosts its support for the out party at midterm from a desire for balance in terms of ideology or policy.

Andrew has blogged about this a bunch, e.g. “What I learned from those tough 538 commenters” (2010). Funnily, it includes an example of balancing:

As a reader, when you see someone write that Person X is wrong, one of your first reactions is gonna be, “Hey, maybe not! Let me give X a fair shake.”

Let’s try to understand what all the fuss was about. “Person X” is “Self-described ‘political operative’ Les Francis”, who says in 2010:

I don’t need any polls to tell me that Republicans will do well in November. The “out” party almost always shows significant gains in the first midterm election of a new President.

Yup. Andrew wanted to add that beyond this balancing effect, polls do still help to predict midterm elections. Let:

• polls = % Democrat in polls in midterm years asking respondents which party (“generic”, not a specific candidate) they intend to vote for
• vote = % Democrat in elections in midterm years
• president = 1 if Democrat is in office, −1 if Republican is in office

Bafumi et al. (2010) fit regression models:

vote = b0 + b1 * president + b2 * polls + error

Both coefficients are “statistically significant”.

• b1 is negative: Democrats do better in midterms when there is a Republican president, and vice versa.
• b2 is positive: polls predict election results.

Bafumi et al. (2010) Table 4 shows that b2 doesn’t change much based on the time of the poll, but b1 gets closer to 0 as the polls are closer to election day. Figure 3 shows this regression:

As Andrew points out, based on Bafumi et al. (2010), an election prediction model with both president and polls does better than a prediction model with president alone.

p.s. This Survey Statistics blog series always includes a photo of the polar bear on trail. As Nick Tierney noted on bsky, this may be unexpected. But in this post, there is a connection in that Joseph Bafumi is a professor at Dartmouth, home of part of the Appalachian Trail.