The long-range presidential election forecast model developed by Jay DeSart is designed to predict the election outcome up to a year in advance. The multiple regression model is based on state electoral histories, national polling data, and two variables that attempt to estimate next year’s election context. The model generates forecasts of state-level outcomes using the following vote equation:
Vi = A + b1 Previous resulti + b2 Prior national polls + b3 Home statei + b4 Regime age
| Table 1: Overview of variables used in the DeSart model | |
| Variable | Description |
|---|---|
Previous result |
State i’s result from the previous election |
Prior national polls |
Average of all national head-to-head matchup polls taken in month X prior to the election |
Home state |
1 if state i is is the home state of the Democratic candidate, -1 if state is is the home state of the Republican candidate, and 0 otherwise |
Regime age |
Number of terms the party currently occupying the White House has done so |
| V | Democratic share of the two-party vote in state i |
| A | Constant |
The forecast of the national popular vote is then calculated by weighting each state by its proportion of the total number of votes cast in the previous election. On his website, Jay DeSart provides more details about the approach as well as conditional model forecasts depending on hypothetical match-ups of candidates. PollyVote identifies the most likely nominees based on forecasts from PredictWise and incorporates the respective model forecast in its combined forecast.
Reference
DeSart, J. A. (2015). State Electoral Histories, Regime Age, and Long-Range Presidential Election Forecasts: Predicting the 2016 Presidential Election. 2015 Iowa Conference on Presidential Politics, Dordt College, Sioux Center, IA, October 29-31.