Why might defining piece-wise linear variables perform better than approximation using dummy variables? More accurately modeling complex/flexible functional forms Accounts for omitted variable bias Less concern about multicollinearity Coefficients become easier to interpret