data.table is typically much faster than dpylr or custom code for large datasets. Figure out what parts of your preprocessing are taking the most time, and try to rewrite those using data.table.

R has some non-default packages for linear regression [0]. If your usecase is supported by one of those then I'd imagine it would be faster than Stata.

[0] https://stackoverflow.com/questions/25416413/is-there-a-fast...