Introduction to lasso using Stata

The lasso command package is available as of Stata 16 and has enhanced features in Stata 17. You can use lasso to perform model selection and prediction for continuous, binary, and count outcomes, plus use state-of-the-art methods to make inferences on the variables of interest, while lasso selects the control variables. This presentation provides an introduction to lasso using Stata. In lasso for prediction, some conceptual aspects of the methods will be addressed. An example of continuous variable prediction comparing lasso, elastic net, square-root lasso, and ordinary least squares (OLS), among other things, will be presented, showing how to use the options to select the value of the lasso penalty parameter. In lasso for inference, some of the theory will be covered. Likewise, it will provide an example for a linear model using the double-selection, partialing-out, and cross-fit partialing-out lasso methods (also known as double machine learning), and finally, it will show how to customize individual lassos and how to use some control selection techniques.