RE: st: RE: psmatch2 with the if condition

["Nick Cox" <n.j.cox@durham.ac.uk>]

In effect you raise two new points: 

1. Whether -psmatch2- promises, implicitly or explicitly, to match
companies [in your case] in a specified subset only with others in the
same subset. 

2. How to achieve that if that is what you wish. 

On 1, the best answers will come from the program authors, who may or
may not be currently members of this list. 

A solution 2 that is possible: You could temporarily -drop- observations
that you don't want to be included from the dataset. That way,
-psmatch2- will not see any others. 
However, I wouldn't implementing that until you have a good answer on 1.


I haven't read the -psmatch2- help. But the looping you quote looks
sound, and yours is buggy, so I recommend that you base your code more
closely on that example. 

Nick 
n.j.cox@durham.ac.uk 

Roberto Mura

Nick,
Thank you for your kind reply. The loop is not the main issue. Even 
if I do not specify any loop but simply code:
use "C:\AAA.dta", clear
catenate condition1 =  country sic2
gen quotedcondition1=1 if quotedcompany==1
replace quotedcondition1=0 if quotedcompany==0 & diversification_orig==1
gen quotednumerator=.
gen quoteddenomenator=.
gen privatenumerator=.
gen privatedenominator=.
egen g=group(condition1)
bysort condition1: gen Ncondition1=_N

psmatch2 quotedcondition1 if g==1, pscore(size) noreplacement

I still find that psmatch2 uses companies which do not belong to 
group 1 to look for matches. As a matter of fact, it is assigning _id 
values outside the range of group 1 (there are just 15 firms in it 
but psmatch2 reports values like 877). Also, I am not blaming 
psmatch2 for anything and I never said I discovered a bug. I am just 
trying to understand what I am doing wrong. Also, please note that my 
code is very similar to the one reported in the help of psmatch2 
which I report below for simplicity:
Matching within strata
     The following code illustrates how to match within exact cells 
and then calculate the average effect for the whole population.

         g att = .
         egen g = group(groupvars)
         levels g, local(gr)
         qui foreach j of local gr {
                 psmatch2 treatvar varlist if g==`j', out(outvar)
                 replace att = r(att) if  g==`j'
         }
         sum att

At 11:22 22/03/2010, you wrote:
>You refer to various user-written programs here. Remember that you are
>asked to explain where user-written programs mentioned in a post come
>from.
>
>-catenate- is from SSC. It has long since morphed into, and been
>superseded by, the official -egen- function -concat()-. However, its
use
>here is not problematic.
>
>More importantly, -psmatch2- is a very often used command from SSC. I
>haven't used it, but my guess is very much that you are
misunderstanding
>something rather than that you have discovered a bug.
>
>Note that Stata is not doing anything independently of the commands you
>issue, contrary to some of your wording here.
>
>I haven't tried to understand your detailed logic, but what seems much
>more likely than a problem with -psmatch2- is that you are
>misunderstanding what your -while- loops are doing.
>
>You have two loops
>
>local j=1
>while `j' <= g {
>         ...
>         local i=1
>         while `i' < Ncondition1 {
>                 ...
>         }
>}
>
>Both involve looping until a value stored in a variable is reached, -g-
>in the outer loop and -Ncondition1- in the inner loop. With this kind
of
>programming what Stata does is look at the value of the named variable
>_in the first observation_. This is documented at
>
>FAQ     . . . . . . . . . . . . . . . . . . . . .  if command vs. if
>qualifier
>         . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
J.
>Wernow
>         6/00    I have an if command in my program that only seems
>                 to evaluate the first observation, what's going on?
>
http://www.stata.com/support/faqs/lang/ifqualifier.html
>
>So, it's as if you had programmed
>
>local j=1
>while `j' <= g[1] {
>         ...
>         local i=1
>         while `i' < Ncondition1[1] {
>                 ...
>         }
>}
>
>It's evident that is definitely not what you want and leading to your
>loop terminating before the work you want is done and that you are
>unjustly putting the blame on -psmatch2-. It may be that your loop just
>goes round once and does not get beyond -g[1]- which is 1.
>
>Nick
>n.j.cox@durham.ac.uk
>
>Dr. Roberto Mura, Ph.D. [edited]
>
>I have a cross section of companies, some listed some private.
>I am trying to find a matching private (undiversified) firm for each
>listed one, on the basis of company size. Also, since my dataset is
>cross country, I am trying to control for country and SIC code as
>well. To this end, I have created a condition, called -condition1-, to
>divide the sample into a set of clusters:
>
>catenate condition1 =  country sic2
>
>Now, once a match is found, I need Stata to report some accounting
>data for the matched private firm on the same row as the  listed firm.
>I have tried to create a "self-fulfilling" nested loop so that I
>don't need to specify beforehand the total number of clusters
>(-condition1-) and also because the total number of firms changes with
>different clusters (-Ncondition1- below).  The problem that I am facing
>is that -psmatch2- does not seem to understand the -if- condition.
>I have uploaded the dataset (for version 11 and for version 9 of
>Stata) and the do file in my website under
>www.robertomura.com/psmatch2/ if you want to take a look.
>Here is what I do:
>
>use "D:\AAA.dta", clear
>catenate condition1 =  country sic2
>gen quotedcondition1=1 if quotedcompany==1
>replace quotedcondition1=0 if quotedcompany==0 &
diversification_orig==1
>gen quotednumerator=.
>gen quoteddenomenator=.
>gen privatenumerator=.
>gen privatedenominator=.
>egen g=group(condition1)
>bysort condition1: gen Ncondition1=_N
>
>local j=1
>while `j' <=g {
>psmatch2 quotedcondition1 if g==`j', pscore(size) noreplacement
>replace quotednumerator=fnetincome if _nn==1
>replace quoteddenomenator=shareholdersfundsnew if _nn==1
>replace privatenumerator=fnetincome if _n1==_id
>replace privatedenominator=shareholdersfundsnew if _n1==_id
>
>local i=1
>while `i' <Ncondition1 {
>replace privatenumerator=fnetincome[_n+`i'] if _n1==_id[_n+`i']
>replace privatedenominator=shareholdersfundsnew[_n+`i'] if
>_n1==_id[_n+`i']
>replace privatenumerator=fnetincome[_n-`i'] if _n1==_id[_n-`i']
>replace privatedenominator=shareholdersfundsnew[_n-`i'] if
>_n1==_id[_n-`i']
>local i = `i' + 1
>}
>local j = `j' + 1
>}
>
>
>The problem seems to lie in
>
>psmatch2 quotedcondition1 if g==`j', pscore(size) noreplacement
>
>What seems to be happening is that the matching is done only for the
>first 3 firms of the first cluster and then Stata is mis-assigning
>the _id since the numbers do not follow the clustering but rather,
>Stata is applying the -psmatch2- considering the sample as a whole. If,
>for instance, I change the if condition into:
>
>psmatch2 quotedcondition1 if g==1, pscore(size) noreplacement
>
>then Stata is not matching inside the cluster (condition1) but again
>it is treating the sample as a whole. In other words the "if"
>condition does not appear to be used at all by Stata.
>
>I have tried both version of psmatch2. The one available through SSC
>(version 3.1.5 2may2009) and the more recent one available from
>http://leuven.economists.nl/stata (4.0.1)
>I have Stata/MP 11.0 for Windows (64-bit x86-64)  running on XP Pro
>64. I also tried with the older version I had, Stata/SE 9.2 for
>Windows 64-bit x86-64and things don't change.

*
*   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/