Today I tested having get_sample() return a matrix rather than a data.frame() but found no performance benefit. The speed gain in running get_sample() was lost due to the additional time required to access columns from the matrix in do_nmb_sim(). (This was surprising to me but I ...