re: st: PSMATCH2 with Dichotomous Outcomes

["Ariel Linden, DrPH" <ariel.linden@gmail.com>]

Dennis,

PSMATCH2 uses OLS regression to analyze the outcome. I would suggest you use
PSMATCH2 for finding you the matched pairs, and then run logistic regression
on your own using only the matched pairs (ie., logistic Y treat if
_weight==1).

I also noticed that you are using a 0.5 caliper. That is very wide. The
common rule of thumb is 20% of the standard deviation of the propensity
score. I rarely see a caliper that exceeds 0.05 (in fact, most of the time I
see calipers at about 0.02-0.03).

I hope this helps

Ariel

Date: Thu, 12 May 2011 12:17:16 -0400
From: Dennis Kramer <dkramerii@gmail.com>
Subject: st: PSMATCH2 with Dichotomous Outcomes

Can PSMATCH2 be used with a dichotomous outcome and estimate the
average odds ratio of the treatment? I have used the following syntax
for estimating continuous outcomes and I would like to adapt it a
dichotomous platform.

Any suggestions would be greatly appreciated.

set seed 12345
tempvar sortorder
gen `sortorder' = runiform()
sort `sortorder'

global covar LIST OF COVARIATES FOR MATCHING

label define ALL_BLOCKlabel 1 "Treated - Block"
label define ALL_BLOCKlabel 0 "Untreated - Non-Block", add
label values ALL_BLOCK ALL_BLOCKlabel
label variable ALL_BLOCK "treatment"

logit ALL_BLOCK $covar
predict pscore

histogram pscore, start(0.0) width(0.05) by(ALL_BLOCK, col(1))
xsize(1) ysize(1.1) xtitle("Block vs. Non-Block") note("Pr(Block")
scheme(s 1mono) saving(support,replace)

psmatch2 ALL_BLOCK, outcome ( OUTCOME VARIABLES) pscore(pscore)
neighbor(1) caliper(0.5) common

*
*   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/