Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Obtaining 95%CI for marginal effect


From   Steven Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Obtaining 95%CI for marginal effect
Date   Sun, 6 Feb 2011 11:08:13 -0500


The method I showed will work for indicator variables too, but I, for one, don't understand what is represented when indicators for variables with >2 categories are set to their means. (-svy: prop- won't help, as for k categories, it gives k proportions). Much better, I think, to set the other categorical variables to typical values. e.g., for the auto data set:

 xi: svy: reg mpg weight i.foreign i.rep78
 // adjust to rep78=3
adjust weight=3138.575 _Irep78_2=0 _Irep78_3=1 _Irep78_4=0 _Irep78_5=0, by(foreign)

For non-linear models like -logistic-, setting variables to their means can produce unexpected results. See: http://www.stata.com/statalist/archive/2010-07/msg01596.html and Michael Norman Mitchell's follow-up.


Note: Your original formulation in -mfx- looks incomplete. You specified "at_Imedgrp_1=0 _Imedgrp_2=0" But medgrp had four levels,( 0,1,2,3), since -xi- produced three indicator variables. Your specification was equivalent to saying: at medgrp==0 or medgrp==3. Is this what you intended?

Steve


On Feb 6, 2011, at 9:59 AM, Nur Hafidha Hikmayani wrote:

Thanks Steve.
I'm afraid however that I'm not clear enough when some independent
variables are categorical. You gave an example in which weight and
turn are numerical variables - in my case, there is only 1 numerical
IV. Suppose foreign is the main IV of interest and other covariates
are mostly categorical, can we use -adjust- too (perhaps with
-svy:prop- beforehand instead)?

hafida-

On Sun, Feb 6, 2011 at 9:12 PM, Steven Samuels <sjsamuels@gmail.com> wrote:
Nur-

I apologize. I checked and discovered -adjust- after -svy: reg- does not compute predictions at the weighted means of the covariates, only at the unweighted means. As a work-around, you could substitute the weighted means
by hand.

****************************
sysuse auto, clear
drop if rep78==.
svyset rep78 [pw=head]
svy: mean weight turn   //get survey weighted means
xi: svy: reg mpg weight turn i.foreign
adjust weight= 3138.575 turn=40.33816, by(foreign) ci se
*****************************

Steve
sjsamuels@gmail.com


Nur-

Use -adjust- with the -ci- option. The fitted value of y is not a "marginal effect"; for -regress- or (-svy: regress-) the default marginal effects are
the regression coefficients.

*********************
sysuse auto, clear
xi: reg weight price turn i.foreign
adjust price turn, by(foreign), se ci
******************

Steve
sjsamuels@gmail.com

On Feb 6, 2011, at 7:02 AM, Nur Hafidha Hikmayani wrote:

Dear all,
I've been running some regression models using -svy- and estimating
its marginal effect using -mfx- (I use Stata 10.1).
I wonder how can I get the 95% CI for the marginal effects (y)?

The output for regression and its marginal effect are as follows:

. xi: svy: reg GH i.medgrp exgrp chronic nummed gp
------------------------------------------------------------------------------
         |             Linearized
GH | Coef. Std. Err. t P>|t| [95% Conf. Interval] ------------- +---------------------------------------------------------------- _Imedgrp_1 | 4.429839 3.56263 1.24 0.214 -2.564628 11.42431 _Imedgrp_2 | 8.333728 3.633545 2.29 0.022 1.200035 15.46742 _Imedgrp_3 | 10.05818 3.773961 2.67 0.008 2.648813 17.46755 exgrp | -8.916839 1.89046 -4.72 0.000 -12.62835 -5.205324 chronic | -10.31767 1.936802 -5.33 0.000 -14.12017 -6.515169 nummed | -1.063043 .3452676 -3.08 0.002 -1.740902 -. 3851831 gp | -4.347845 1.773649 -2.45 0.014 -7.830027 -. 865663 _cons | 83.19335 3.520136 23.63 0.000 76.28231 90.10438
------------------------------------------------------------------------------

. mfx, at(mean _Imedgrp_1=0 _Imedgrp_2=0)
Marginal effects after svy:regress
  y  = Fitted values (predict)
     =  53.085624
------------------------------------------------------------------------------
variable | dy/dx Std. Err. z P>|z| [ 95% C.I. ] X --------- +--------------------------------------------------------------------
_Imedg~1*|   4.429839     3.56263    1.24   0.214  -2.55279  11.4125
0
_Imedg~2*|   8.333728     3.63354    2.29   0.022   1.21211  15.4553
.379185
_Imedg~3*|   10.05818     3.77396    2.67   0.008   2.66136   17.455
0
exgrp*| -8.916839 1.89046 -4.72 0.000 -12.6221 -5.21161 . 762312
chronic*|  -10.31767      1.9368   -5.33   0.000  -14.1137 -6.52161
 .83752
nummed | -1.063043 .34527 -3.08 0.002 -1.73975 -.386331 6.69376 gp*| -4.347845 1.77365 -2.45 0.014 -7.82413 -.871558 . 472814
------------------------------------------------------------------------------
(*) dy/dx is for discrete change of dummy variable from 0 to 1


Any help is much appreciated,
Thanks,
hafida-

*
*   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/

*
*   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/


*
*   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/

*
*   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index