ApolloChoiceModelling forum

Sir in a E-MDCEV model there are two coefficient matrix
1. Estimated parameters with approximate standard errors from BHHH matrix:
Estimate BHHH se BHH t-ratio (0)
bm 1.14780 0.20237 5.6717
bh 0.65524 0.19768 3.3147


2. Unconstrained optimisation.
Estimates:

Estimates:
Estimate s.e. t.rat.(0) Rob.s.e. Rob.t.rat.(0)
bm 1.14780 0.135865 8.4481 1.28327 0.89443
bh 0.65524 0.138205 4.7411 1.17424 0.55801

which one should 've consider for the results. and should we consider the t6 ratio or the rob t ratio for the confidence intervals. ??

Hi

Apollo always returns BHHH standard errors first as an approximation. This is to give you an initial idea about significance while the full covariance matrix is computed. Once that computation is done, then you should use the full covariance matrix one. And you should generally use the robust standard errors

Stephane