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: How to make a code faster - alternatives to egen var = concat(vars) ?,


From   Antoine Terracol <terracol@univ-paris1.fr>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: How to make a code faster - alternatives to egen var = concat(vars) ?,
Date   Thu, 17 Jun 2010 23:17:58 +0200

This might not be the most efficient way, but it works:

capture program drop mymean
program define mymean, byable(recall)
syntax varname
marksample touse
local groupname ""
foreach var of local _byvars {
	local a =`var'[_byn1()]
	if "`a'"=="." {
			local a "missing"
			}
	local groupname "`groupname'`a'"
	}
su `varlist' if `touse'
scalar mean`groupname'=r(mean)
end

scalar drop _all
sysuse auto, clear
bysort foreign rep78 : mymean price
scalar dir


Antoine


On 17/06/2010 16:56, Tiago V. Pereira wrote:
Thank you so much again, Antoine!

Yes, this is a very efficient way! However, I could not figure out how I
can save the combination of the categorical variables that a specific
meanX refers to.

For example, the commands

  sysuse auto, clear
  bysort foreign rep78 : mymean price
  scalar dir

show a list of scalars containing the mean of the ith combination, but I
don't know if the mean10 refers to the combination  "foreign = Foreign,
rep78 =4"   or  "foreign = Foreign, rep78 = 5"


[Actually I do in this specific case if I take a look at each value from
the output.]

Nevertheless, assuming a very large number of categorical variables
(n>10), I cannot write a loop and say that mean2451 refers to the
combination x1==0 x2==0 x3==2 x4==0 x5==2 and x6==1. I want to summarize
the mean of this combination in group 1 and generate a separate variable
for group 2.


for example

bysort x1 x2 x3 x4 x5 x6 : mymean score if group==1

*/ yes, -mymean- needs further amendments to have this option

replace score = mean2451 if group==2&x1==0&x2==0&x3==2&x4==0&x5==2&and&x6==1


So, In this case I know that mean2451 comes from the combination
x1==0&x2==0&x3==2&x4==0&x5==2&and&x6==1 from group 1 and I replace its
value for all subjects from group 2 having an identical combination.


This is getting tough, but you have any additional tips, I will be really
very grateful!

Thanks again.

Tiago





















Tiago,

You would have to define a -byable- -program-, such as:

capture program drop mymean
program define mymean, byable(recall)
syntax varname
marksample touse
su `varlist' if `touse'
local a= _byindex()
scalar mean`a'=r(mean)
end

sysuse auto, clear
bysort foreign rep78 : mymean price
scalar dir

Antoine

On 17/06/2010 15:47, Tiago V. Pereira wrote:
Thanks, Antoine!

But for each combinations, I want to save a local containing the
r(mean).
Is it possible to do that using -bysort-?

Tiago


---------------
Dear statalisters,

I am working on a stata code, and I need some advice.

I have n categorical variables that assumes values equal to 0, 1 or 2.
My
objective is to summarize a continuous variable (say, age) by all
possible
combinations of these categorical variables.


For example, suppose I have 5 categorical variables (x1, x2, x3, x4 and
x5):


sum age if x1==0&x2==0&x3==0&x4==0&x5==0

then

sum age if x1==0&x2==0&x3==0&x4==0&x5==1

then

sum age if x1==0&x2==0&x3==0&x4==0&x5==2

and so forth.


What I am doing is the following: (1) I generate a string of the
categorical variables

egen combination = concat(x1 x2 x3 x4 x5)

(2) convert them to numeric

encode combination, gene (y)

and loop over the values of the new variable y to summarize the
continuous
variable

forvalues i = 1/`some_max_value' {

sum age if y=`i'

}

This naïve solution works very well for small samples (_N<1000) and
small
number of categorical variables (5 to 7). But when I need investigate in
a
larger sample with a larger number of categorical variables, this code
is
highly inefficient (e.g. slow).

Do you have any suggestions to make this procedure faster in larger data
sets?

Thanks in advance!

Tiago






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