Problems with products? Control strategies for models with interaction and quadratic effects

• DOI: 10.1017/psrm.2020.17
• Publication Date: 2020-05-18T10:22:55Z
• Authors: Janina Beiser-McG...; Liam F. Beiser-Mc...
• Keywords: Overfitting

Abstract Models testing interactive and quadratic hypotheses are common in Political Science but control strategies for these models have received little attention. Common practice is to simply include additive variables, without relevant product terms, into with interaction or terms. In this paper, we show Monte Carlos that terms can absorb the effects of other un-modeled non-linear analogously, included reflect omitted interactions non-linearities. This problem even occurs when do not share any constitutive We Carlo experiments regularized estimators, adaptive Lasso, Kernel Regularized Least Squares (KRLS), Bayesian Additive Regression Trees (BART) prevent misattribution interactive/quadratic effects, minimize problems efficiency loss overfitting, low false-positive rates. illustrate how inferences drawn change used strategy using a recent paper. Implementing recommendations paper would increase reliability conditional relationships estimated many papers literature.