At the APSA meeting, there was the traditional election forecasting panel, with presentations by 4 scholars or teams of scholars. Mark Blumenthal has a nice write-up here. I have also acquired the presentations of most of the presenters, which are linked below.
Mark hits on the key distinction between the models. Models based on “the fundamentals”—typically, the state of the economy (usually measured with GDP growth in the election year) and presidential approval—predict that the Democrats will lose something close to the historical average (24 House seats in post-WWII elections). In Michael Lewis-Beck and Charles Tien’s model, that translates to a 22-seat loss in the House. In Alfred Cuzán’s model, that translates into either a 27 or 30-seat loss, depending on whether you are modeling all elections or only midterm elections.
Models that incorporate “attitudinal” information, such as the generic ballot, tend to show a larger loss. In Alan Abramowitz’s model, a 5-point pro-Republican gap in the generic ballot—which is close to the current gap—translates into a 49-seat swing in the House and a 4-seat swing in the Senate (which Abramowitz acknowledged was likely too low). Joseph Bafumi, Robert Erikson, and Chris Wlezien’s model predicts a 50-seat swing in the House. (See also their paper.)
James Campbell’s model is a bit different. It incorporates presidential approval, the number of seats controlled by the majority, and the majority’s advantage or disadvantage in the number of “seats in trouble” (based on ratings by the Cook Political Report). Campbell’s prediction is a 51-seat gain. (See also his paper.)
Finally. the panel’s discussant, Gary Jacobson, writes via email:
My model, based on the party of the administration, seats held, presidential approval, the quality of each party’s challengers (using data from the 90 percent of districts where the nominations have been set), and presidential approval currently predicts the Democrats to lose 43 seats. The 95% confidence interval is +/- 27 seats, however.
Below is a graph of the various House predictions. I will add others as they are available. Please email me any other seat predictions that you know of. A key thing to remember is Gary Jacobson’s note about the confidence interval. Although models predicting that the Democrats will lose the majority typically find that the likelihood of this is high (above 60 or 70%) nearly any point prediction for seat shares will have substantial uncertainty.
UPDATE: The graph below has been updated to include Seth Masket’s forecast: