st: Re: biprobit

[Maarten Buis <maartenlbuis@gmail.com>]

--- Junji Xiao wrote me privately:
> I read some of your posts about biprobit commad in stata
> online. It happens that I am also using the command to do
> a project, but I met something weird. After reading your
> posts, I got some idea about the command, but still felt
> puzzled on the results. could you give me some advice?

Such questions must be sent to the Statalist directly rather than its
individual members. The reasons for that are discussed in the
Statalist FAQ: http://www.stata.com/support/faqs/resources/statalist-faq/#private

> I am using a recursive biprobit model, so the correlation
> between the error terms of these two equations are
> important. My results show that /athrho is insignificant
> (/athrho |  -12.15695   431.8511, p = 0.978), but rho is -1,
> and the std of rho is 4.76e-08. Also, the Likelihood-ratio
> test of rho=0:     chi2(1) =  196.543    Prob > chi2 = 0.0000.
>
> I cannot understand such results: if rho is not significant,
> how come the chi2 test rejects the null hypothesis? I
> understand that the program actually is searching over
> /athrho , rather than rho to max the likelihood. Is that the
> reason? In such a case, shall we reject the hypothesis of
> rho = 0 or not?

The correlation between your error terms is in essence -1. If you type
in Stata -di %12.10f tanh( -12.15695)- you will see that the
correlation is -0.9999999999. This is so close to -1 that you cannot
meaningfully distinguish between the two error terms. You can imagine
the kind of problems that will cause for your model. In essence, I
would not trust any results from that model.

-- Maarten

---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

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