There's a "tiny" glitch I came across when running the ClarkLDF method on LARGE
triangles. Since the origin year parameters are large while the growth curve
parameters are small, the ____number (forgot the term) of the matrices over
which the optimization is run is too small. Behind the scenes I scale the
observations to avoid that situation (a common fix), and the solution is found.
However, the vcov method does a recalculation "from scratch" and the
_____number error is raised at that time. An easy fix would be to calculate the
vcov matrix with the scaled data and save the unscaled matrix in the result.
Since no one besides me (that I know of) has come across this glitch, it's low
on my priority list.
Original issue reported on code.google.com by chiefmur...@gmail.com on 10 Nov 2012 at 10:06
Original issue reported on code.google.com by
chiefmur...@gmail.comon 10 Nov 2012 at 10:06