I'm wondering about your circumstances... Is there a high cost
for assigning persons to "control" as opposed to using the whole
sample? Are there expensive follow-up tests and measures to be
applied, so you want a limited number? - If there is no extra cost,
you should analyze *all* the cases. (Here, I suspect that there is
someone's wrong and improper advice that leads you to want
"equal group sizes" instead of using all cases.)
- For experimental reasons, it can be handier to *stratify* at times...
if, for instance, you want equal numbers of males and females; or
due to low counts, you might want to drop Races that seldom occur,
and over-represent some Race that occurs in barely sufficient numbers.
Here is advice that is off the top of my head (do not consider me expert
on this point) -- When there is prior knowledge that a set of traits are
apt to be responsible for good/bad outcome, for whatever you are studying,
you might take the artificial group formed by the intersection of those
traits as the criterion for "propensity."
--
Rich Ulrich
> Date: Sun, 24 Mar 2013 21:05:41 -0700
> From:
[hidden email]> Subject: matching program in SPSS?
> To:
[hidden email]>
> I was wondering if there was a stratified matching program that I could use
> in SPSS, where I can match participants based on their array of demographics
> (Gender, age, race, SES, etc.), perhaps obtain a similarity score so I can
> locate a "best match" for any one participant. I would pair those
> participants, then randomly assign one of the pair to the treatment group,
> the other to the comparison group.
>
> I am thinking about using a cluster analysis in SPSS to obtain groups of
> participants with similar characteristics, and then pair participants within
> those clusters. I am thinking about using propensity score matching, but am
> I correct that propensity score matching is not quite appropriate because
> the participants are not already assigned to their groups? I am new to
> STATA and am thinking about querying for a macro in STATA that I could use.
>
> Thanks in advance for your response.
> ...
Locked
