hello. i have some R code that estimates an index popular in the quantitative social sciences called the polychoric correlation coefficient. the difference is that instead of assuming a multivariate normal distribution an elliptical copula distribution is assumed as a model.
the R script works fine, i generate good enough estimates but i cannot, for the life of me, get appropriate standard errors (SE). there are functions in R that estimate this index via maximum likelihood so i have been using these as comparison to evaluate my results. they should be the same (within reasonable rounding error of course) but mines are far, far off.
your job would be to look at my R script (which otherwise works perfectly) and help me fix it to obtain standard errors that match the ones from the already-developed functions in the 'polycor' R package. i know it is impossible for them to be the exact same, but they should be within a reasonable range of which other. ideally, the ones from the copula model should be smaller
PS- as a little bit of a guideline, this is a little more complicated than just grabing the 'optim' function, setting hessian=T and taking the square root of the inverse of the diagonal. i'm wondering if something is wrong with my likelihood function (although i hope not because the estimates i get match the ones i would expect)
i am comfortable around statistics and R but this is definitely escaping my abilities a little bit.
I did my statistics dissertation computation work in R. I think I can solve your interesting problem.
4 freelance font une offre moyenne de $563 pour ce travail
I have very strong hands on R and matlab. I am sure you will be happy to have me working you, and might look forward to re-hire me like many others. Please see pm.
Hi! The first thing I'd try would be to test your functions with normal data in place of elliptical. Thus we could know if the likelihood is good.