Re: st: Bivariate probit model with partial observability and survey data

[Stas Kolenikov <skolenik@gmail.com>]

2011/8/21 Eduardo Andrés Alfonso Sierra <ealfonsosierra@googlemail.com>:
> It starts Fitting comparison model and it finished after 339 iterations.
> Then, in Fitting full model, it keeps on iterating, quite in the same
> log likelihood value after 700 iterations and always showing backed
> up.
> Now, after more than 4 days running!, it keeps on going.
>
> I am aware of the poor convergence properties of this model, but I
> really don´t get why the weights and adjustments for survey design
> might change it so radically, so any idea on how to proceed would be
> greatly appreciated.

There are two issues with weights:

1. If there is a lot of variability (a coefficient of variation of say
10, and the ratio of the largest to the smallest weight of say a
1000), then the convergence (or the lack of it) and the estimates are
essentially determined by the few largest weights. Look at -summarize
weights- to see how bad the situation is.

2. -ml maximize- does not work well with weighted data if the scale of
the weights is of the order say 1e4 or 1e5. I reported this earlier:
http://www.stata.com/statalist/archive/2011-05/msg01412.html, and
suggested the shortcuts to circumvent it. I should've recalled this
discussion in the first place! I have been trusting -ml- to behave
well so sincerely... but in my case, and apparently in your case as
well, it does not.

-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/