We live in a pretty evenly divided country politically. But despite that, twice as many Americans (61 percent) live in states where both legislative chambers have Republican majorities than where both chambers are blue (30 percent) — with the rest (9 percent) living in states where one party controls one chamber each. Republicans swept into many state legislatures in 2010 — most notably in Wisconsin, Michigan, Ohio, Pennsylvania and North Carolina — and haven’t been dislodged since.
In layperson’s terms, how exactly do you forecast a state legislative outcome? What factors do you take into account?
You take things that can be measured before an election, and, via statistical modeling, look at how those things are related to election outcomes in the past. Then you take those same factors, measure them in the current election, and in light of their relationship to past elections, forecast this election.
My analysis is at the district level. That allows you to take the translation of votes into seats into account. Because of gerrymandering and just where people happen to live, the Democrats commonly get more than 50 percent of the votes for state legislature, but not a majority of state legislative seats.
The starting point is the partisan split in the vote in a district in the last election, and then you see what factors are associated with a change in the vote. One such factor is incumbency — including whether an incumbent retired as well as whether an incumbent is now running but wasn’t in the last election.
Another important factor is the “partisan tide.” Whether people say they’ll vote for a Republican or Democrat in U.S. House elections — a.k.a. the “generic ballot” — is the main part of this. That percentage has been associated with a shift across the country in how many votes each party gets in state legislative races.
What are the highlights of your current forecasts? How many state legislatures might shift in a consequential way?
My most recent forecasts say that the Democrats will take seven chambers, while the forecast two months ago was they would take nine chambers. The most important pickups for the Democrats could be in the Michigan House and Senate and the North Carolina Senate, which are toss-ups according to the model.
A Democratic takeover of the Michigan state legislature would be particularly noteworthy, since they are also favored to win the gubernatorial race. There have only been two times since 1974 that a complete partisan shift in both legislative chambers and the governor’s mansion has occurred: Maine and Wisconsin, both in 2010.
In addition, the Maine and New York Senates are very likely to flip blue, while the Arizona, Colorado and New Hampshire state senates are toss-ups. The state houses in Arizona, New Hampshire and West Virginia are also toss-ups.
Many people have noted that a good showing for Democrats might position them to control the redistricting process after 2020? Do you see any signs that this could happen?
I think it’s likely nonpartisan redistricting will pass in Michigan this year, so Democratic gains in the legislature probably won’t matter for redistricting. The same is true in Arizona, where a nonpartisan commission already draws district lines. In North Carolina, a Democratic takeover of either chamber would be extremely important for redistricting, as the governor can’t veto redistricting bills.
For the most part, the maps Republicans drew after their 2010 sweep are effective enough that the GOP is likely to maintain control in the Florida, Pennsylvania and Wisconsin state legislatures, unless one of these states institutes nonpartisan redistricting. Chances of this reform basically died in the Pennsylvania legislature earlier this year, however.
And now the obligatory post-2016 questions: What makes you most cautious about your forecasts?
This is my favorite question! Here’s one important thing.
For the final forecast, I use a strict decision rule to pick the generic ballot number: use the last survey that ended before Oct. 29 that used either registrants or all adults as its sampling frame. This helps ensure consistency with historical polls and avoids issues with how to determine “likely voters” or how exactly to aggregate different polls with different methodologies. Most importantly, using a consistent rule means you can’t cherry-pick the poll you want.
In this case, the last poll I picked was an Oct. 21-23 Economist/YouGov poll. I calculated the percentage who chose the Democratic House candidate among those intending to vote for a Democrat or Republican (thus excluding undecided voters or those who planned to vote for a third-party or other candidate). This figure was 53.4 percent. But the comparable statistic from the 538 poll aggregation on Oct. 22 was 54.5 percent, suggesting a larger Democratic advantage.
So it is possible that this forecast is too pessimistic for Democrats. But I think it’s important to stick to a consistent decision rule regardless.