[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
st: RE: Redpace
"Nick Cox" <firstname.lastname@example.org>
st: RE: Redpace
Wed, 2 Jun 2010 11:45:06 +0100
The missing reference is
SJ-6-2 st0106 . MSL random-effects dynamic probit mod. with autocorr. errors
(help redpace if installed) . . . . . . . . . . . . . . M. B. Stewart
Q2/06 SJ 6(2):256--272
introduces new command, redpace, to be used in MSL
estimation for random-effects dynamic probit models
with autocorrelated errors
As often noted on this list, minimal author (year) references are regarded as incomplete.
I would be happy to be wrong but this looks like a case in which
1. The author and expert is not a member of Statalist. He may be able to advise better but you would need to contact him directly.
2. This is a specialised command on which there may not be much experience among active list members.
3. The question of what will work better with your data is difficult to answer in abstraction. For example, there is no indication of what commands you are trying, including how complicated a model you are attempting to fit. Banal though it is, some of us find ourselves repeating very simple advice: if your model won't converge, try a much simpler one.
I am trying to run -redpace (Stewart, 2006).
I am following a procedure to begin with small number of replications
(R=5, 10 etc) and then use these results as
starting values for larger R (option -from).
However, after a certain number of R (about R=20) I could not find concave
Any suggestion will be very helpful.
Did the halton option or the antithetic improve the convergence.
* For searches and help try: