Date:         Thu, 22 Dec 2011 17:36:19 -0500
Reply-To:     Sas Beginner <cep@STUNTZ.ORG>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Sas Beginner <cep@STUNTZ.ORG>
Subject:      creating propensity score. quasi separation error

hello.

i created a dataset of 6-month time blocks. within each 6-month block, i am
running proc logistic to create propensity scores. i then use greedy
matching to match the exposed subjects to reference subjects (within this
6-month block).

it was all looking good until i hit a warning on one of the 6-month blocks.

here is my sas code.
 PROC LOGISTIC DATA=inpatient_229 Descend;
     MODEL initiator = age alcoholism_prior beta_blocker_2 bis_2
                       bmi_prior diuretics_2 dvt_prior estrogen_2 gc_2
                       heart_failure_prior
                       high_fall_risk_prior hyperthyroidism_prior
                       neuro_impair_prior pul_emb_prior
                       smoking_prior statins_2
      /SELECTION = STEPWISE RISKLIMITS LACKFIT RSQUARE PARMLABEL;
      OUTPUT OUT= Propen229 prob=prob;
 RUN;

here is the sas log:

NOTE: PROC LOGISTIC is modeling the probability that initiator=1.
WARNING: There is possibly a quasi-complete separation of data points. The
maximum likelihood estimate may not exist.
WARNING: The LOGISTIC procedure continues in spite of the above warning.
Results shown are based on the last maximum likelihood iteration. Validity
of the model fit is questionable.
NOTE: There were 3484 observations read from the data set V20FRAC.INPATIENT_229.
NOTE: The data set WORK.PROPEN229 has 3484 observations and 81 variables.

my questions:
- is there anything i can do to resolve the warning?
- should i just not include this specific 6-month block in my final
analysis dataset?

many thanks!  happy holidays.