In the spotlight: Nonparametric regression—Estimation, inference, and effects

Don’t know how your outcome and covariates are related? Don’t worry. You don’t have to force your data into a linear model, logistic model, Poisson model, or any other model that depends on knowing something about the relationship between the outcome and covariates. You don’t even need to decide whether to include squared terms or interactions among your covariates. Nonparametric regression does not depend on knowing the functional form. With Stata 15’s new npregress command, you can use nonparametric kernel estimation to fit your model by simply typing

. npregress kernel y x1 x2

But you will not want to stop with estimation. After nonparametric regression, or any regression, you will want to answer questions based on that model. margins provides those answers.