[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: Marginal effects for bivariate probit
On Fri, 14 Nov 2003 16:13:16 -0800 (PST) Joseph Capuno
> Hi! I've been running a bivariate probit regression
> and try to obtain the marginal effects using the 'mfx
> compute" option. However, I fail to obtain the
> marginal effects in most of the runs, and in those
> cases when I do Stata does not compute the standard
> I would appreciate it somebody could help me out with
> the following questions:
> (1) Is there a way for me to compute the standard
> errors myself?
You can program the analytical formulae, but it is complicated.
Formulae are given in
Christofides LN, Stengos T, Swidinsksy R (1997)
"On the calculation of marginal effects in the bivariate probit model"
Economics Letters 54, 203-208
with an important Corrigendum by Christofides LN, Hardin JW and Stengos
T (2000) Economics Letters 68, 339, which indicated tersely that many
of the original formulae were wrong. I am not sure that the Corrigendum
resolved everything either (it is very terse).
The help file to Tomas Bartus's program -margin- states that it
calculates marginal effects for -biprobit-, but I have not checked that
it uses the analytical formulae cited above.
> (2) And, how do I interpret Stata's estimates of the
> marginal effects with no standard errors? How much
> confidence can I attach to these estimates?
> (3) Sometimes Stata reports "." when a standard error,
> z-stat, P>|z|, Wald statistic or a chi-squared
> statistic is expected? What does "." mean in this
> context? No estimate?
I suspect both symptoms are related, and connected to an inability to
invert the relevant Hessian matrix (as it is close to singular) --
required to calculate the s.e.s. Also check whether your original model
is well-conditioned (e.g. no very large coefficients)
Professor Stephen P. Jenkins <email@example.com>
Institute for Social and Economic Research (ISER)
University of Essex, Colchester, CO4 3SQ, UK
Tel: +44 (0)1206 873374. Fax: +44 (0)1206 873151.
* For searches and help try: