The need to pre-specify expected interactions between variables is an issue in multiple regression. Theoretical and practical considerations make it impossible to pre-specify all possible interactions. The functional form of the dependent variable on the predictors is unknown in many cases. Two ways are described in which the data mining technique Multivariate Adaptive Regression Splines (MARS) can be utilized: first, to obtain possible improvements in model specification, and second, to test for the robustness of findings from a regression analysis. An empirical illustration is provided to show how MARS can be used for both purposes.
Adams, Susan M.; Gupta, Atul; Haughton, Dominique M.; and Leeth, John D.
"Data Mining CEO Compensation,"
Journal of Modern Applied Statistical Methods: Vol. 7
, Article 21.
Available at: http://digitalcommons.wayne.edu/jmasm/vol7/iss2/21