 |
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: dprobit and lincom
Melanee Thomas <[email protected]> asks how to get -lincom- to
work with the values reported in the output from -dprobit-:
> I have a question about dprobit and lincom. Specifically, when I run lincom
> after dprobit, the returned coefficient is a probit coefficient, rather than
> the dF/dx reported by dprobit. Is there an option I can use to transform the
> probit coefficient returned by lincom to the dF/dx reported by dprobit?
Maarten Buis <[email protected]> replied mentioning -mfx- for Stata 10
users and the new -margins- command in Stata 11.
Austin Nichols <[email protected]> replied with some substative advice
and a couple examples involving data manipulation.
Maarten then replied with his own example using -probit-, -summarize-, and
-nlcom-.
I'll just comment the use of -lincom- after -mfx- and -margins-.
While -mfx- does compute marginal effects, it does not provide an easy way of
posting its results so that -lincom- will work with the marginal effects.
The -margins- command computes marginal effects and has an option that will
allow -lincom- to work with the resulting marginal effects.
Here is a brief example of how this can be done.
First we'll establish that -margins- can reproduce the marginal effects from
-dprobit-.
Using the auto data, we'll use -dprobit- to fit a probit regression of the
'foreign' indicator variable on 3 predictors. -dprobit- reports the marginal
effects at the overall mean of the 3 predictors.
. sysuse auto
. dprobit for mpg turn trunk, nolog
Next we'll refit the model using -probit-, and use the -margins- command to
reproduce the marginal effect calculations.
. probit for mpg turn trunk, nolog
. margins, dydx(*) atmeans
Here is what Stata reported:
***** BEGIN:
. sysuse auto
(1978 Automobile Data)
. dprobit for mpg turn trunk, nolog
Probit regression, reporting marginal effects Number of obs = 74
LR chi2(3) = 38.76
Prob > chi2 = 0.0000
Log likelihood = -25.652611 Pseudo R2 = 0.4304
------------------------------------------------------------------------------
foreign | dF/dx Std. Err. z P>|z| x-bar [ 95% C.I. ]
---------+--------------------------------------------------------------------
mpg | -.0089624 .0094573 -0.96 0.337 21.2973 -.027498 .009574
turn | -.0755808 .0188995 -4.28 0.000 39.6486 -.112623 -.038538
trunk | -.0062585 .0137758 -0.45 0.654 13.7568 -.033259 .020742
---------+--------------------------------------------------------------------
obs. P | .2972973
pred. P | .1438596 (at x-bar)
------------------------------------------------------------------------------
z and P>|z| correspond to the test of the underlying coefficient being 0
. probit for mpg turn trunk, nolog
Probit regression Number of obs = 74
LR chi2(3) = 38.76
Prob > chi2 = 0.0000
Log likelihood = -25.652611 Pseudo R2 = 0.4304
------------------------------------------------------------------------------
foreign | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
mpg | -.0395317 .0411346 -0.96 0.337 -.1201541 .0410907
turn | -.3333765 .0779801 -4.28 0.000 -.4862148 -.1805383
trunk | -.0276053 .0616797 -0.45 0.654 -.1484953 .0932847
_cons | 13.37647 3.500776 3.82 0.000 6.515074 20.23786
------------------------------------------------------------------------------
. margins, dydx(*) atmeans
Conditional marginal effects Number of obs = 74
Model VCE : OIM
Expression : Pr(foreign), predict()
dy/dx w.r.t. : mpg turn trunk
at : mpg = 21.2973 (mean)
turn = 39.64865 (mean)
trunk = 13.75676 (mean)
------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
mpg | -.0089624 .0094573 -0.95 0.343 -.0274982 .0095735
turn | -.0755808 .0188995 -4.00 0.000 -.1126232 -.0385385
trunk | -.0062585 .0137758 -0.45 0.650 -.0332586 .0207416
------------------------------------------------------------------------------
***** END:
Now that we've established that -margins- can reproduce the marginal effects
reported by -dprobit-, all we need is to use -margins-' -post- option to post
the marginal effects and their covariances to -e(b)- and -e(V)- so that we can
use -lincom- with them.
. margins, dydx(*) atmeans post
. lincom mpg - turn
Here is what Stata reports:
***** BEGIN:
. margins, dydx(*) atmeans post
Conditional marginal effects Number of obs = 74
Model VCE : OIM
Expression : Pr(foreign), predict()
dy/dx w.r.t. : mpg turn trunk
at : mpg = 21.2973 (mean)
turn = 39.64865 (mean)
trunk = 13.75676 (mean)
------------------------------------------------------------------------------
| Delta-method
| dy/dx Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
mpg | -.0089624 .0094573 -0.95 0.343 -.0274982 .0095735
turn | -.0755808 .0188995 -4.00 0.000 -.1126232 -.0385385
trunk | -.0062585 .0137758 -0.45 0.650 -.0332586 .0207416
------------------------------------------------------------------------------
. lincom mpg - turn
( 1) mpg - turn = 0
------------------------------------------------------------------------------
| Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
(1) | .0666185 .0163695 4.07 0.000 .0345348 .0987022
------------------------------------------------------------------------------
***** END:
The output from -lincom- is the difference between the conditional marginal
effects of 'mpg' and 'turn'.
--Jeff
[email protected]
*
* 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/