Login  Register

Re: propensity scoring

Posted by Dennis Deck on Sep 01, 2006; 6:55am
URL: http://spssx-discussion.165.s1.nabble.com/variables-fro-factor-scores-tp1070665p1070668.html

Sometimes stratification, covariate, or weighting make more sense than
matching.

* Here we save the propensity score (CDp), modeling assignment to the
College Dreams program (CD=1 for tx group and 0 for comparison).

LOGISTIC
   Vars= CD with Male LowInc Minority NonTrad Transfer Absences HighRisk

                 GPA5 AvgAch
  /Save=PRED(CDp) /Classplot /External .


* Here we calculate quintiles (CDq) on the propensity score for use in
  stratification and evaluating overlap of the group distributions .

RANK CDp  /NTiles(5) INTO CDq  .


* Here we compute the inverse probability of treatment weights (IPTW)
based
  on the estimated propensity score .

DO IF (CD=1) .
+  COMPUTE iCDp = 1 / CDp .
ELSE .
+  COMPUTE iCDp = 1 / (1-CDp) .
END IF .

* Here we apply the weights in an outcome analysis predicting the odds
of HS
  graduation for treated vs untreated students  .

WEIGHT BY iCDp .
LOGISTIC
   Vars= Graduation with CD Male GPA5 HighRisk
     /Save= Pred(GradP) /Classplot /External .
WEIGHT OFF .


There is more to good propensity score analysis but the syntax is
straightforward - at least when using the PS to define strata,
covariate, or weights.   Matching is more complicated.

Dennis Deck, PhD
RMC Research Corporation
[hidden email]

-----Original Message-----
From: Todd McDonald [mailto:[hidden email]]
Sent: Thursday, August 31, 2006 10:13 AM
Subject: propensity scoring

Does SPSS have the ability to create matched groups using propensity
scores?

  Todd





---------------------------------
How low will we go? Check out Yahoo! Messenger's low  PC-to-Phone call
rates.