Usually my practice even with small databases has been to pulverize them with a highly normalized schema implemented in #postgres. But this week I discovered GNU #recutils, which seems (together with #emacs #recmode) just right for the early, exporatory phase of an analysis.
https://www.gnu.org/software/recutils/
https://www.gnu.org/software/recutils/rec-mode-manual/rec-mode.html