RE: st: MIXLPRED: marginal effects after-MIXLOGIT-

[nagi kayo <kayonagi@hotmail.co.jp>]

Dear Arne (if I may)

 

Thank you very much for your quick reply and for your question and advice.

I am really sorry that I did not provide you with enough information on my application.

 

first, please let me explain the data I used.

although the number of "id" used in my estimation is about 10,000, 

the following is the data on the first two "id." 

 

id alt  d  d1  d2  d3 d1inc  d2inc  d3inc  d4inc      p

1   1   1   1   0   0   665       0     0      0  0.214

1   2   0   0   1   0     0     665     0      0  0.186  

1   3   0   0   0   1     0       0   665      0  0.381

1   4   0   0   0   0     0       0     0    665  0.219

2   1   0   1   0   0   779       0     0      0  0.553

2   2   1   0   1   0     0     779     0      0  0.301

2   3   0   0   0   1     0       0   779      0  0.107

2   4   0   0   0   0     0       0     0    779  0.039

 

id:    households

alt:   1=households are worried about their retirement life 

         because the pension benefit is NOT enough.

       2=households are worried about their retirement life 

         for some reasons other than pension.

       3=households are NOT worried about their retirement life 

         because the pension benefit is enough.

       4=households are NOT worried about their retirement life 

         for some reasons other than pension.

d:     dummy which equals one if households choose "alt" in the same row.

       (id 1 chose alt 1, and id 2 chose alt 2.)

d1-d3: intercepts

d1inc-d4inc: real households income (units of 10,000yen)

             Mean=531

             Std.Dev.=413.4245

             Min.=0

             Max.=6079.027

p:     an alternative specific variable

 

using the above data, i did the estimation as follows:

 

************************************************

mixlogit d d1 d2 d3 d1inc d2inc d3inc, group(id) rand(p)
mixlpred pred_base

preserve
quietly replace d1inc=d1inc+1
quietly replace d2inc=d2inc+1
quietly replace d3inc=d3inc+1
quietly replace d4inc=d4inc+1
mixlpred pred_inc1
gen dif_inc1=pred_inc1-pred_base

sum dif_inc1 if alt==1

sum dif_inc1 if alt==2
sum dif_inc1 if alt==3
sum dif_inc1 if alt==4

restore
************************************************


the results are 

  mean of dif_inc1 is -2.11e-12 if alt==1

  mean of dif_inc1 is -1.05e-12 if alt==2 

  mean of dif_inc1 is  1.87e-12 if alt==3

  mean of dif_inc1 is  1.29e-12 if alt==4

 

i think these marginal effects are too small...

 

so as you kindly advised me, i saw Richard Williams's reply to your earlier post

on calculating marginal effects after -mixlogit- and increased by 0.001 and divided

the difference by 0.001.

 

************************************************

mixlogit d d1 d2 d3 d1inc d2inc d3inc, group(id) rand(p)

mixlpred pred_base

preserve
quietly replace d1inc=d1inc+0.001
quietly replace d2inc=d2inc+0.001
quietly replace d3inc=d3inc+0.001
quietly replace d4inc=d4inc+0.001
mixlpred pred_inc2
gen dif_inc2=(pred_inc2-pred_base)/0.001

sum dif_inc2 if alt==1

sum dif_inc2 if alt==2
sum dif_inc2 if alt==3
sum dif_inc2 if alt==4

restore

************************************************


the results are 

  mean of dif_inc1  7.83e-7 if alt==1

  mean of dif_inc1  3.85e-7 if alt==2 

  mean of dif_inc1 -6.64e-7 if alt==3

  mean of dif_inc1 -5.04e-7 if alt==4


I think these marginal effects are still too small.

 

finally, I increased the Std.Dev. of the regressor divided by 1,000

and divided the difference by the Std.Dev. of the regressor divided by 1,000.

 

************************************************

mixlogit d d1 d2 d3 d1inc d2inc d3inc, group(id) rand(p)

mixlpred pred_base

preserve
quietly replace d1inc=d1inc+0.4134245
quietly replace d2inc=d2inc+0.4134245
quietly replace d3inc=d3inc+0.4134245
quietly replace d4inc=d4inc+0.4134245
mixlpred pred_inc3
gen dif_inc3=(pred_inc3-pred_base)/0.4134245

sum dif_inc3 if alt==1

sum dif_inc3 if alt==2
sum dif_inc3 if alt==3
sum dif_inc3 if alt==4

restore

************************************************

 

the results are 

  mean of dif_inc1 -2.70e-9 if alt==1

  mean of dif_inc1 -1.42e-9 if alt==2 

  mean of dif_inc1  2.24e-9 if alt==3

  mean of dif_inc1  1.89e-9 if alt==4

 

still, the marginal effects are so small.

 

actually, i also controlled the other variables like age and wealth

and calculated the marginal effects like above,

but their marginal effects are too small, too.

so I am wondering if there is something wrong with my calculation.

 

I greatly appreciate it if you would advise me again. 

 

with best wishes,

Kayo

 

 

 

---------------------------------------- > Date: Wed, 29 Feb 2012 11:01:57 +0000 > Subject: Re: st: MIXLPRED: marginal effects after-MIXLOGIT- > From: arnehole@gmail.com > To: statalist@hsphsun2.harvard.edu > > Dear Kayo, > > It's difficult to help without knowing more about your application. > Most importantly: what is the unit of measurement of the income > variable? If income is measured in dollars or euros (or yen), for > example, you would expect the marginal effect of a one unit increase > in income to be small. See Richard Williams' reply to my earlier post > on calculating marginal effects after -mixlogit-. > > Arne > > 2012/2/29 nagi kayo <kayonagi@hotmail.co.jp>: > > Dear Professor Arne Risa Hole and all > > > > > > > > I read the thread Professor Hole kindly wrote on "Mon, 6 Feb. 2012 13:18:34" about calculating marginal effects after -mixlogit- and tried to calculate marginal effects using the following commands. in my estimation model, the number of alternatives 
 is four, and "inc" is a demographic variable, and "price" is an alternative specific variable. > > > > > > > > mixlogit d d1inc d2inc d3inc, group(id) rand(price) > > mixlpred pred_base > > > > replace d1inc=d1inc+1 > > replace d2inc=d2inc+1 > > replace d3inc=d3inc+1 > > replace d4inc=d4inc+1 > > mixlpred pred_inc > > > > > > however, unexpectedly, "pred_inc" is quite similar to "pred_base," and thus the difference between the two is almost zero, which means that the marginal effect of income is almost zero. > > > > although i also tried the other explanatory variables, the results are not changed (in all cases, marginal effects are almost zero). > > > > is my calculation of predicted probabilities wrong? > > > > > > > > i greatly appreciate it if you would give me your advice. > > > > > > > > with best wishes, > > > > Kayo > > > > > > * > > * For searches and help try: > > * http://www.stata.com/help.cgi?search > > * http://www.stata.com/support/statalist/faq > > * http://!
 ww
 w.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/