Sorry-there is something here I dont understand. How would you match
1:1 and then identify the control with the closest propensity score
and use it for subsequent analyses-ie cox regression, logistic
regression ??
Something akin to the greedy-routine for SAS....
Regards,
M
On Thu, Nov 12, 2009 at 6:29 PM, Martin Weiss <[email protected]> wrote:
>
> <>
>
> What exactly is your problem with this output? -psmatch2- does not change
> the marginal distribution of treated and untreated, and it is not supposed
> to:
>
>
> *************
> webuse labor, clear
> gen byte wc =we > 12
> //see distribution
> ta wc
> //distribution reemerges after every command
> psmatch2 wc wmed wfed, outcome(ww) neighbor(1) noreplace
> psmatch2 wc wmed wfed, outcome(ww) kernel kerneltype(epan)
> psmatch2 wc wmed wfed, outcome(ww) llr kerneltype(tricube)
> *************
>
>
>
> HTH
> Martin
>
>
> -----Ursprüngliche Nachricht-----
> Von: [email protected]
> [mailto:[email protected]] Im Auftrag von moleps islon
> Gesendet: Donnerstag, 12. November 2009 17:29
> An: [email protected]
> Betreff: Re: st: RE: psmatch2 and stcox
>
> Thx - for the first suggestion. However despite using the
> noreplacement option I get the same output, ie not 1:1 matching :
>
> xi:psmatch2 treatment `var',noreplacement neighbor(1) common out(dead)
>
>
> | psmatch2:
> psmatch2: | Common
> Treatment | support
> assignment | On suppor | Total
> -----------+-----------+----------
> Untreated | 628 | 628
> Treated | 338 | 338
> -----------+-----------+----------
> Total | 966 | 966
>
>
> Any clues?
>
>
> Regards,
>
> M
>
>
>
>
> On Thu, Nov 12, 2009 at 3:23 PM, Villa Lora, Juan Miguel
> <[email protected]> wrote:
>> Hi!
>> I've got suggestions:
>> 1. Try to estimate the propensity score firstly (with a probit or logit)
> and set the prediction in the option p(your_prediction). This saves you time
> when running the command several times.
>> 2. Specify the option "noreplacement" which yields a 1-1 matching as
> wanted.
>>
>> JM.
>>
>> -----Original Message-----
>> From: [email protected]
> [mailto:[email protected]] On Behalf Of moleps islon
>> Sent: Martes, 10 de Noviembre de 2009 03:28 p.m.
>> To: [email protected]
>> Subject: st: psmatch2 and stcox
>>
>> I´m trying to match 336 exposed patients from a cohort with a total of
>> 966 patients using psmatch2:
>>
>> xi:psmatch2 E `var', outcome(out) neighbor(1) common
>>
>> I thought this would give me 1:1 matching, but I get 336 exposed and
>> 662 control patients.
>>
>> Then doing
>>
>> ttest -different variables-, by(_t)
>>
>> doesnt really match very well despite reasonable overlapping propensity
> distributions. Is this due to omitting the weighting option?
>>
>> I´d like to use these propensity matched data for a logrank test, but I´m
> uncertain to the type of weight I should use. I assumed it should be
> pweight, but I found another posting here where they said fweight leaving me
> uncertain....
>>
>> Regards
>>
>> M
>>
>> *
>> * 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/
>
*
* 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/
