Thread by @WeAreRLadies

First up: The LASSO family

Perhaps the most popular ML technique in economics is LASSO - a variable selection technique. The R package 'glmnet' gives users a range of distributions of the response variable to choose from: normal, binomial, poisson, multinomial,cox and others!

Interaction terms in the model are common in econ (& other soc. sciences). Want to select interactions along with the main effects? R package 'hierNet' implements Hierarchical LASSO. Users can choose the kind of hierarchy condition based on the research Q!

In grouped data, we might want to select variables from among groups as well as within group level. Check out Sparse-Group LASSO using R package 'SGL'.

Does your data have an underlying index (e.g. time?) You might want to select covariates honoring the undelying index. Check out fused LASSO in the R package 'extlasso'.
#rstats #econtwitter

Another common type of data structure is time series. LASSO in a time-series setting can be implemented using ordered LASSO in R package 'orderedLASSO'.
#rstats #econtwitter

Finally, want more control over how your variables are coded, which interactions to include and how to group your variables in LASSO? Check out the R package 'glinternet' which provides more flexibility!
#rstats #econtwitter