Loess: a nonparametric, graphical tool for depicting relationships between variables

Loess is a powerful but simple strategy for fitting smooth curves to empirical data. The term “loess” is an acronym for “local regression” and the entire procedure is a fairly direct generalization of traditional least-squares methods for data analysis. Loess is nonparametric in the sense that the fitting technique does not require an a priori specification of the relationship between the dependent and independent variables. Although it is used most frequently as a scatterplot smoother, loess can be generalized very easily to multivariate data; there are also inferential procedures for confidence intervals and other statistical tests. For all of these reasons, loess is a useful tool for data exploration and analysis in the social sciences. And, loess should be particularly helpful in the field of elections and voting behavior because theories often lead to expectations of nonlinear empirical relationships even though prior substantive considerations provide very little guidance about precise functional forms.