The Goldilocks Principle: Avoiding Pitfalls in Interpretation of Regression Coefficients

Abstract:     
Regression coefficients estimate the effect of a one-unit increase in the independent variable X on the dependent variable Y. Although a one-unit increase is mathematically convenient, it is an arbitrary choice for interpreting the substantive significance of a coefficient. Depending on the type of variable and its level and distribution, a one-unit increase is too big for some independent variables, too small for others, and just right for yet others. This paper discusses theoretical and empirical criteria for choosing the right size contrast for each Xi to convey the size and shape of its relationship with Y. Transformation of variables and model specification are also considered. Example sentences are used to illustrate how to interpret coefficients in the context of the research question and data.

Number of Pages in PDF File: 15

JEL Classification: A23, C10

working papers series