Search

Modeling many political processes requires both a variable selection component and a smoothing component. Recent advances in penalized regression methods allow for variable selection through non-concave constraints and smooth curve fitting through ridge-type constraints. I develop a model that fits both simultaneously, allowing for one set of covariates to be selected through a LASSO-type constraint and another to be smoothed through ridge-type spline constraints. An algorithm, stopping rule, and unbiased degrees of freedom estimate are provided. My method is illustrated first through testing the Tiebout Hypothesis. I fit a smooth curve to a battery of county-level socioeconomic data in the United States, while selecting jumps that coincide with state lines. I next illustrate the method with the 2004 American National Election Study. My method models a survey respondent with a fixed amount of attention, who then selects among several discrete issues and adds them when forming a response. Issues selection is allowed to vary by demographic, but placed under a LASSO constraint. Individual responses are allowed to be arbitrary smooth curves of responses. Simulations and a discussion of further uses conclude the paper. ..PAT.-Unpublished Manuscript [ABSTRACT FROM AUTHOR]