Bioequivalence and Bioavailability Forum

Dear HS,

❝ Do you think that’s possible? RMLE is an iterative process. Even if you find a formula based on the variance-covariance-matrix, you would need to fire up some software to fill it up with numbers first. IMHO no ‘manual’ calculation possible here.

Yes, I think it is possible.
VBA works with double precision reals if the user wants (dim X as double etc) which is definitely good enough. I am rather sure that it is be possible to set up the entire thing in VBA/Excel although I would of course any day prefer a dedicated stats software. There are some practicalities around it, though:
1. VB is interpreted, not truly compiled and executed as binary. This makes it quite slow for complex operations.
2. Setting up the algos in VBA would definitely be more than just a standard day at the office. Need support from underlying matrix routines.

I made some months ago a matrix library from scratch in C and set up the algos for REML estimation and it actually worked fine. Was about 9000 lines IIRC, so it would be the same in VBA. My code was based on doubles, too, and achieved the same as R using d_labes's model specifications (see the legendary freedom thread). We need also to bear in mind that the math operations (plus, minus, log, sine etc) in themselves are never proven wrongly implemented in any VB operation; the wrong implementations come when people use the basic functions in an inappropriate manner to calculate something slightly more complex like standard deviations.

Who's got 1000 hours spare time for this little task?