DuckDB is incredible for large stats projects using R. R can't really handle large datasets and simply runs out of memory if you load everything. You get around this by splitting your data into a bunch of RData files but there's quite a bit of overhead maintaining those, recreating them when you need more variables, etc.

With DuckDB you can just load it once into the database then use dplyr/dbplyr to transparently access data with the queries pushing down to the DB afaik. And it runs anywhere and doesn't need local admin (like most database servers).

(SQLLite can't handle many columns without a custom build)