Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


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

Re: st: Outreg for asclogit


From   John Luke Gallup <jlgallup@pdx.edu>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Outreg for asclogit
Date   Thu, 4 Oct 2012 02:29:27 -0700

Wameq,

A more flexible solution to the problem of getting the marginal effects for -asclogit- is to move the estimates from -estat mfx-, which is the idiosyncratic command to create the marginal effects, into the r() matrices in which they are placed by -margins-, the command that calculates marginal effects for almost all other estimation commands.  Then one can use -outreg- directly to put marginal effects in a formatted table, with access to all of -outreg-'s options.

There are a couple of hoops: only rclass programs can place values in r() matrices and the marginal effects from -estat mfx- must be reorganized.

The following code does the trick:

webuse choice, clear
qui asclogit choice dealer, case(id) alternatives(car) casevars(sex income)
qui estat mfx, at(sex=0 dealer=1)
  
capture program drop Fill_rbV 
program Fill_rbV, rclass
	args b V
	return matrix b = `b'
	return matrix V = `V'
	return local cmd "margins"
end

local k = e(k_alt)
forvalues i = 1/`k' {
	tempname b`i' V`i'
	mat `b`i'' = r(`e(alt`i')')
	mat `V`i'' = diag(`b`i''[.,2..2])
	mat `V`i'' = `V`i'' * `V`i''
	mat `b`i'' = `b`i''[.,1..1]'
}

outreg, clear
forvalues i = 1/`k' {
	Fill_rbV `b`i'' `V`i''
	outreg, margin merge ctitle("", "", "`e(alt`i')'") nodisplay
}
outreg using ascmfx, replay replace

The table looks like this:


                                         -------------------------------------------------
                                                               American   Japan   Europe 
                                         -------------------------------------------------
                                          dealer    American    0.017    -0.012   -0.004 
                                                               (2.29)*   (2.34)*  (1.66) 
                                                    Japan       -0.012    0.017   -0.004 
                                                               (2.34)*   (1.83)   (1.03) 
                                                    Europe      -0.004   -0.004   0.009  
                                                                (1.66)   (1.03)   (1.30) 
                                          casevars  sex*        0.026    -0.161   0.135  
                                                                (0.38)   (2.03)*  (1.76) 
                                                    income      -0.008    0.006   0.002  
                                                               (2.95)**  (1.96)   (1.14) 
                                          N                      885       885     885   
                                         -------------------------------------------------




John

On Oct 2, 2012, at 9:39 PM, John Luke Gallup <jlgallup@pdx.edu> wrote:

> Wameq,
> 
> -outreg- (and probably -outreg2- and -estout-) don't handle marginal effects after -asclogit- because the results are saved in a very non-standard format.
> 
> -estat mfx-, which is only for use after -asclogit- saves the marginal effects in r() matrices named after the logit categories.
> 
> Things you can't do in -outreg- you can still do with a bit of work with -frmttable-, which is the formatting engine distributed with -outreg-.
> 
> The following code creates an -outreg- compatible table (which can be merged, etc.) showing the marginal effects with t-statistics below them and asterisks for significance.
> 
> First I calculate the marginal effects:
> 
> webuse choice, clear
> asclogit choice dealer, case(id) alternatives(car) casevars(sex income)
> estat mfx, at(sex=0 dealer=1)
> 
> Then I create a Stata matrix holding the marginal effects and t-stats (`b') and an indicator matrix for where the stars go (`stars', using the mata command).  These are passed to -frmttable- to create a Word table (or TeX table) with the results.
> 
> tempname b stars
> loc ctitl `""","""'
> forvalues i = 1/`e(k_alt)' {
> 	tempname b`i'
> 	mat `b`i'' = r(`e(alt`i')')
> 	mat `b' = (nullmat(`b'), `b`i''[.,1..1], `b`i''[.,3..3])
> 	mat `stars' = (nullmat(`stars'), `b`i''[.,4..4])
> 	loc ctitl `"`ctitl' ,`e(alt`i')'"'
> }
> mata: st_matrix("`stars'", (abs(st_matrix("`stars'")):<0.05) + ///
> 	(abs(st_matrix("`stars'")):<0.01))
> frmttable using ascmfx, statmat(`b') substat(1)	///
> 	annotate(`stars') asymbol("*","**") ctitle(`ctitl') ///
> 	replace note("* p<0.05; ** p<0.01")
> 
> You could adapt this to report standard errors, p-values, or change the stars, etc. if you want.  
> Here's the table:
> 
>                                        ---------------------------------------------------
>                                                              American   Japan    Europe  
>                                        ---------------------------------------------------
>                                         dealer    American     0.02*     -0.01    -0.00  
>                                                               (2.29)*   (-2.34)  (-1.66) 
>                                                   Japan       -0.01*     0.02     -0.00  
>                                                               (-2.34)   (1.83)   (-1.03) 
>                                                   Europe       -0.00     -0.00    0.01   
>                                                               (-1.66)   (-1.03)  (1.30)  
>                                         casevars  sex*         0.03      -0.16    0.13   
>                                                               (0.38)*   (-2.03)  (1.76)  
>                                                   income      -0.01**    0.01     0.00   
>                                                               (-2.95)   (1.96)   (1.14)  
>                                        ---------------------------------------------------
>                                                        * p<0.05; ** p<0.01
> 
> 
> John
> 
> 
> On Oct 2, 2012, at 1:38 PM, Wameq Raza <wameq.r@gmail.com> wrote:
> 
>> Hi everyone,
>> I was wondering if anyone could kindly tell me if there's a way to get
>> the marginal effects of a asclogit model that I'm running
>> into a table using outreg.
>> More specifically, I'm generating the marginal effects using: estat
>> mfx, var (variables); i'm doing this for both alt specific and
>> case specific vars.
>> Any help would be much appreciated!
>> Best
>> Wameq
>> 
>> --
>> W A M E Q   R A Z A
>> *
>> *   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/
> 


*
*   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/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index