[Attaullah Shah]

You posted the same question here, to which I have replied.

[Attaullah Shah]

Dear Bjarte Aagnes,

Thank you for your detailed post and the time you’ve taken to benchmark the performance of various options. Your insights are valuable to us and the community. My reply is late but I saw the post today when Huong tagged me in his post.

I would like to clarify that asreg was primarily designed to write its output to the dataset, rather than creating matrices. The additional code you used, and the fact that asreg creates additional variables in the datasets and writes to them, seem to have increased the timing, (my guess).

While your matrix-based solution offers speed advantages in the specific scenario you tested, it’s important to remember that asreg was designed with a broader set of use-cases in mind. It aims to provide a balance between performance and flexibility, allowing for a wide range of covariates without the need to manually specify them.