Modeling the Past: A Note on the Search for Proper Form

As historians move beyond the computation of predictive measures to fashion regression models of past behavior they need to consider not only the estimation of regression parameters, but also the specification of a functional form for the model itself. The choice of a particular form for expressing the relationships among variables often has important implications for interpreting the historical issues under investigation. This paper presents methodology developed by J. B. Ramsey and other theorists for comparing distinct functional forms of a regression model according to criteria other than differences in the capacity to account for variation in the dependent variable. Ramsey's procedure directly tests the hypothesis of correctly specified functional form through scrutiny of the pattern of residuals that would be expected for properly specified models. The methodology can thus provide useful information on the choice of computing models especially when measures of explained variation are too close in value to indicate a clear preference for a given equation.