First try did not post.
*****
Maarten,
I'm sorry if I was not clear.
The example you provide yields, e.g., P(y =1) and P(y=2) with the
standard errors for each.
And it does yield P(y=1 | y = 2) by adding the two.
But it does not yield se[P(y = 1 | y = 2)]. Rather, it yields 2
standard errors, se[P(y=1)] and se[P(y=2)].
The esttab portion, shown below, concentrating on the foreign effect
only, has an se for (1) of .0280 and an se for (2) of .0553, but it
does not have an se for (1) or (2). Staying with this example, call
the rep78 values of 1 and 2 poor repair records, 3 an average repair
record, and 4 and 5 good repair records. I want to know the effect of
foreign on the probability of having a poor repair record (either y =
1 or y =2), and I need a standard error for that quantity.
--------------------------------------------------------------------------------------------
(1) (2) (3)
(4) (5)
rep78 rep78 rep78
rep78 rep78
--------------------------------------------------------------------------------------------
rep78
foreign (d) -0.0393 -0.156** -0.482***
0.233* 0.444**
(0.0280) (0.0553) (0.0901)
(0.0982) (0.136)
length -0.000650 -0.00274 -0.00746
0.00765 0.00320
(0.000541) (0.00154) (0.00404)
(0.00412) (0.00184)
mpg -0.00277 -0.0117 -0.0318*
0.0326* 0.0136
(0.00221) (0.00597) (0.0158)
(0.0157) (0.00751)
--------------------------------------------------------------------------------------------
N 69 69 69
69 69
--------------------------------------------------------------------------------------------
******
I do not understand, getting the standard errors is equally trivial
this way. See the example below.
-essto- and -esttab- are not necessary, they just help put the result
in one table at the end. These commands are part of Ben Jann's -estout-
package which can be downloaded by typing -ssc install estout-.
*------------ begin example ----------------
sysuse auto, clear
ologit rep78 foreign length mpg
qui levelsof rep78
local k : word count `r(levels)'
forvalues i = 1/`k' {
eststo mfx`i' : mfx, predict(p outcome(`i'))
}
esttab, margin se
*-------------- end example ------------------
(For more on how to use examples I sent to the Statalist, see
http://home.fsw.vu.nl/m.buis/stata/exampleFAQ.html )
Hope this helps,
Maarten
On Fri, Jan 9, 2009 at 4:35 PM, Eric Lawrence <[email protected]> wrote:
> <>
> I've searched the archives and not found an answer to my query.
> Is there a straightforward way to obtain the standard errors for the
> marginal effects for multiple categories in ordered logit/probit
> models?
>
> More precisely, after ologit/oprobit, typing:
>
> mfx, at(mean) predict(outcome(1))
>
> yields the marginal effects for the dependent variable taking on the
> value of 1. [y = Pr(y==1) (predict, outcome(1))]
> What I would like is something like:
>
> mfx, at(mean) predict(outcome(1 2))
>
> that would produce the marginal effects and the standard errors for
> the dependent variable taking on the value of 1 or 2, y = Pr(y==1 |
> y==2).
>
> It's trivial to get the marginal effects by repeated calls to mfx.
> The standard errors are not as obvious.
>
> Thank you.
>
> Eric Lawrence
> Dept. of Political Science
> George Washington University
>
*
* 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/
