Error message when fitting robust std error with weights - Zelig

3 Dec 2012

Dear All, 

I try to a fit a poisson model with sandwich estimator for a binary outcome (a technique
to obtain Risk Ratios as found in Cummings, 2009, "The relative merits of riks ratios
and odds ratios").Two questions:
1/ I get similar but still quite different risk ratios with zelig setx and this
technique(in particular with covariates) and I would like to understand why
2/ I get an error message when I try to include weights in my call.
I send a syntax with lalonde data to illustrate the problem (see below or attached text
file)

Many Thanks!!
JBP
 library(MatchIt)
data(lalonde)
library(Zelig)

#Bivariate Risk ratio with zelig setx
z.out=zelig(treat~black,model="logit",data=lalonde)
summary(z.out)
x.low <- setx(z.out, black = 0)
x.high <- setx(z.out, black = 1)
s.out <- sim(z.out, x = x.low, x1 = x.high,num=10000)
summary(s.out)
#RR = 8.36

#Bivariate Risk Ratio with poisson with robust standard errors
obj <- zelig(treat~black, model = "poisson", 
robust=list(method="weave"), data = lalonde)
summary(obj)
 exp(2.106)#=8.215314  somewhat different but similar results

#Now bivariate Risk ratio with covariate married
#setx
z.out=zelig(treat~black+married,model="logit",data=lalonde)
summary(z.out)
x.low <- setx(z.out, black = 0)
x.high <- setx(z.out, black = 1)
s.out <- sim(z.out, x = x.low, x1 = x.high,num=10000)
summary(s.out) 
#RR=7.68

#poisson
obj <- zelig(treat~black+married, model = "poisson", 
robust=list(method="weave"), data = lalonde)
summary(obj)
exp(1.956)#7.070986 quite different results
#In my data, the results are even more striking when including a covariate 
#(as for "married" here, my covariate is a binary variable and setx is taking
the "mean", taking the mode gives even more different results)

#SECOND PROBLEM: fitting poisson model with robust standard error and weight 
length(lalonde$treat)
#creation of false weights 
lalonde$w=c(rep(1,300),rep(2,314))
#fit
obj <- zelig(treat~black+married, model = "poisson", 
robust=list(method="weave"), weights="w",data = lalonde)
summary(obj)
#ERROR message:
#Error in model.frame.default(formula = treat ~ black + married, data = lalonde,  : 
#variable lengths differ (found for '(weights)')

#I need to do that with imputed data so that it should work with the following syntax
#creation of "new" data-frame
lalonde1=lalonde
lalonde1$w=c(rep(0,100),rep(1,214),rep(2,300))
obj <- zelig(treat~black+married, model = "poisson", 
robust=list(method="weave"), weights="w",data = mi(lalonde,lalonde1))
summary(obj)
#observation: when I take out weights from the previous call it works, 
#However, it does not include the sentence: "Robust standard errors computed using
weave". like with non-imputed data. Is it normal?