Anna Dijkstra <[email protected]> :
I would think you could match on some observables (also google
"statistical matching") but you can try:
clear all
tempfile master using
input linkidx dupersid rxrecidx
020183 02019 1
020152 02019 2
110161 11010 1
110161 11010 3
end
save `using'
clear all
input linkidx evntidx eventyr eventmm eventdd
020183 1 2006 8 6
020152 1 2006 8 6
110161 5 2006 4 10
110161 2 2006 7 19
110161 8 2006 5 8
end
g obs=_n
save `master'
set seed 101
use `master', clear
joinby linkidx using `using', unm(both)
drop _m
format *idx %9.0f
egen N=count(linkidx), by(linkidx obs)
g u=ceil(uniform()*N)
bys obs linkidx: g ok=u[1]
by obs linkidx: keep if (_n==ok)
drop N ok u
sort obs
li linkidx evntidx rxrecidx, noo
On Fri, Jul 31, 2009 at 3:11 AM, Michael I. Lichter<[email protected]> wrote:
> Anna,
>
> If this doesn't do what you want, you need to be more specific about your
> needs:
>
> ------
> use file2, clear
> set seed 20090730
> gen myorder = runiform()
> sort LINKIDX myorder
> tempfile file2tmp
> save `file2tmp'
> use file1
> merge LINKINDX using `file2tmp', sort
> drop myorder
> ------
>
> Michael
>
> [email protected] wrote:
>>
>> Hi all,
>> I'm a relatively new STATA user, and I'm trying to merge a couple of
>> large datasets where neither the master nor the using dataset has a unique
>> key. The data comes in this format:
>> Dataset 1: (note that LINKIDX is not unique)
>> EVNTIDX LINKIDX EVENTYR EVENTMM EVENTDD
>> ...
>> 1. 300020190021 300020190083 2006 8
>> 6
>> 2. 300020190021 300020190052 2006 8
>> 6 3. 300110100795 300110101161 2006 4
>> 10
>> 4. 300110100822 300110101161 2006 7
>> 19
>> 5. 300110100808 300110101161 2006 5
>> 8
>>
>> Dataset 2: (note that LINKIDX is not unique) LINKIDX
>> DUPERSID RXRECIDX ...
>> 1. 300020190083 30002019 300020190083001
>> 2. 300020190083 30002019 300020198849002
>> 3. 300110101161 30011010 300110101161001
>> 4. 300110101161 30011010 300110101161003
>>
>> I have already performed a merge where I have limited dataset 1 to only
>> the unique observations of LINKIDX, and linked them to the multiple
>> observations in dataset 2 (using a one-to-many merge). In the case of the
>> above datasets, it would involve linking observation 1 in dataset 1 to
>> observations 2 and 3 in dataset 2. However, I would like to perform a
>> random link for the remaining observations. That is, for observations 3-5 in
>> dataset 1, which match the LINKIDX for observations 3 and 4 in dataset 2, I
>> would like for STATA to randomly pick a LINKIDX in dataset 1 to merge with
>> each matching LINKIDX in dataset 2. I am not sure whether I should simply
>> use the merge function, because it may result in systematic selection of one
>> observation in dataset 1.
*
* 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/
