Abstract
A comparison is made between two types of specification of monthly wintertime surface temperatures over the United States. The specifications are obtained by multiple regression of station temperature anomaly at each of 37 stations onto 700 mb height anomalies represented by 1) grid-point values selected by a forward stepwise screening procedure, and 2) coefficients of the dominant empirical orthogonal functions (EOF's). Various measures of skill show that specifications derived from the pointwise screening are superior in both developmental (dependent) and independent samples. The differences in the skill levels are interpreted as a disadvantage of the spatial generality inherent in the EOF representations.
