Trial-heat model

The trial-heat model, developed by Jim Campbell, predicts the two-party popular vote based on the state of the economy and the candidates’ polling results.

Variable description

Table 1: Overview of variables used in the trial-heat model
Variable Description 2016 Value
POLL Incumbent party’s candidate two-party support in early September Gallup preference poll 51.2
GDP Annualized real GDP growth rate in the second quarter (April–June) of 2016 as indicated by the Bureau of Economic Analysis’ second estimate released at the end of August 1.1
V Incumbent share of the two-party presidential vote  50.7

Vote equation

The regression model’s vote equation reads as:

V = A + b1 POLL + b2 GDP

2016 forecast

Jim Campbell presented the trial-heat model’s first forecast the at the APSA meeting in Philadelphia. The model then predicted Clinton to gain 52.1% of the two-party vote, compared to 47.9% for Trump.

In its latest update, the model predicts a virtual tie between both candidates, with vote-shares of 50.7% for Clinton and 49.3% for Trump.

Past performance

Since its first application in 1992, the trial-heat model underwent only minor changes. The following chart shows the model’s forecasts and the actual election results for each election since 1992. On average across the six elections, the trial-heat model missed the final results by 2.6 percentage points.


Campbell, J. (2016). The Trial-Heat and Seats-in-Trouble Forecasts of the 2016 Presidential and Congressional Elections. PS: Political Science & Politics, 49(4), 664-668.