To Include or Exclude an Explanatory Variable: Beware of Rules of Thumb

In an earlier Foresight article, William Bassin proposed a theoretical rule of thumb to choose between including or excluding an explanatory variable in a regression model. Peter takes a critical look at that rule of thumb, and he shows that it is based on the unrealistic assumption of zero collinearity (correlation) between explanatory variables. He then offers an alternative rule of thumb that accounts for collinearity. But he also warns that rules of thumb in general can be misleading and need to be supplemented by direct testing of forecast accuracy. Copyright International Institute of Forecasters, 2006