Hi Peter, Kosuke is right; full matching essentially uses all observations and so your outcome models have to take into account the structure of the matching in some way, either by using the subclasses that are formed (e.g., through random or fixed effects) or by using the weights that matchit produces (which are essentially frequency weights).  I am not sure your zelig code below does either of those things; it looks like it is just taking a simple difference in means of the outcome in the matched samples (which will be the same as the original samples). Ben Hansen discusses the subclass-specific fixed effects approach in his 2004 paper (http://www.stat.lsa.umich.edu/%7Ebbh/hansen2004.pdf), <http://www.stat.lsa.umich.edu/%7Ebbh/hansen2004.pdf%29,>  or you can also see description of both approaches in my paper on full matching (Stuart, E.A., and Green, K.M. (2008). Using Full Matching to Estimate Causal Effects in Non-Experimental Studies: Examining the Relationship between Adolescent Marijuana Use and Adult Outcomes. /Developmental Psychology/ 44(2): 395-406.) All of that said, I do also agree with Kosuke that more work is needed to better understand the best approaches for doing analyses following full matching.  But I hope this helps. Thanks, Liz On 9/11/09 9:55 AM, "Kosuke Imai" &lt;kimai(a)Princeton.EDU&gt; wrote:    Full matching tries to use as many observations as possible, and    so if you    just do diff-in-means you will get a poor estimate (this is close    to not    doing any adjustment).  I think that it is an open research    question as to    how to do model-based adjustment after full matching, and so I    don't want    to make any strong suggestion here.  One way to do it however is    to use    either fixed or random effect type models so that the strata    created by    full matching can be incorporated into the model.  But, others may    have    better ideas.    Kosuke    --    Department of Politics    Princeton University    http://imai.princeton.edu    On Wed, 9 Sep 2009, Peter Steiner wrote:    > Hi,    >    > after briefly outlining my problem with the full-matching option    in Zelig and    > MatchIt to Imai, I post it in more detail and with the LaLonde data.    > Unfortunately, I was not able to figure out how to correctly    implement    > full-matching with Zelig using all your documentation.    > Moreover, the help in R seems to be sometimes inconsistent (the    help file for    > match.data() differs for the MatchIt and Zelig Package; one    suggests that    > weights are allowed (MatchIt) the other does not show an option    for weights    > (Zelig))    >    > However, I'm interested in estimating ATE using full matching. A    > straightforward way seems to be the following:    >    > m.out <- matchit(treat ~ age + educ + black + hispan +    >  nodegree + married + re74 + re75, data = lalonde,    >  method = 'full', discard = 'both')    > z.out1 <- zelig(re78 ~ age + educ + black + hispan + nodegree +    >  married + re74 + re75, data = match.data(m.out, "control"),    >  model = 'ls')    > z.out0 <- zelig(re78 ~ age + educ + black + hispan + nodegree +    >  married + re74 + re75, data = match.data(m.out, "treat"),    >  model = 'ls')    > x.out1 <- setx(z.out1, data = match.data(m.out, "treat"), cond = T)    > x.out0 <- setx(z.out0, data = match.data(m.out, "control"), cond = T)    > s.out1 <- sim(z.out1, x = x.out1)    > s.out0 <- sim(z.out0, x = x.out0)    > ate.all <- c(s.out1$qi$att.ev, -s.out0$qi$att.ev)    > c(mean(ate.all), sd(ate.all))    >    > This results in an ATE of 1335.5, which seems at least plausible.    But here is    > my problem. Using only full-matching without additional    covariance adjustment    > by setting    >    > z.out1 <- zelig(re78 ~ 1, data = match.data(m.out, "control"),    >  model = 'ls')    > z.out0 <- zelig(re78 ~ 1, data = match.data(m.out, "treat"),    >  model = 'ls')    >    > results in an ATE of 86.88, indicating a very poor performance of    full    > matching. (Using Ben Hansen's R package alone, full matching    would reduce    > almost all bias.) So I tried to use the weights generated by the    matchit()    > command. Using weights in match.data(..., weights = 'weights')    doesn't change    > the estimates very much. And whenever I include weights in    zelig() the sim()    > command produces an Error.    >    > So my question is, what is the correct way to perform    full-matching in Zelig?    >    > I have another question, concerning nearest neighbor matching    (1:1 matching,    > for instance). What does ATE exactly estimate in Zelig? ATE of    the original    > (unmatched) sample or ATE of the matched sample only (note that    the covariate    > distribution of the original and matched sample usually differ    considerably).    > I assume the latter.    >    > Thanks a lot,    > Peter    >    >    --    MatchIt mailing list served by Harvard-MIT Data Center    List Address: matchit(a)lists.gking.harvard.edu    Subscribe/Unsubscribe: http://lists.gking.harvard.edu/?info=matchit    MatchIt Software and Documentation: http://gking.harvard.edu/matchit/    Browse/Search List Archive:    http://lists.hmdc.harvard.edu/lists/matchit/