I am using cem to match treatment and control group members from 5 multiply
imputed datasets from Amelia. I've encountered two types of problems.
First, when I run a logit model after calling cem on the multiply imputed
datasets:
> run <- att(new1, Pass ~ SEQ + O_GPA + O_HRS_EARNED + AGE + PELL1 +
LOG_CA_SCORE + LOG_CM_SCORE, data = imputed, model = "logit")
I get the following warning messages, one for each multiply imputed
dataset:
Warning messages:
1: In eval(expr, envir, enclos) :
non-integer #successes in a binomial glm!
2: In eval(expr, envir, enclos) :
non-integer #successes in a binomial glm!
3: In eval(expr, envir, enclos) :
non-integer #successes in a binomial glm!
4: In eval(expr, envir, enclos) :
non-integer #successes in a binomial glm!
5: In eval(expr, envir, enclos) :
non-integer #successes in a binomial glm!
My dependent variable had no missing cases, so the values for it were not
imputed for any observations, and I have checked that the values of it are
either 0 or 1. Any ideas why I am getting this warning? I've noticed from
searching online that this warning arises when weights are included in a
binomial glm model, so I wonder if it might have something to do with how
cem is weighting the control observations.
(2) The second issue relates to displaying the output of the model. When I
type:
> run
I get:
Logistic model on CEM matched data:
SATT point estimate: 1.509872 (p.value=0.001640)
95% conf. interval: [0.570028, 2.449717]
which is fine, but when I type:
> summary(run)
I get:
Treatment effect estimation for data:
NULL
Logistic model estimated on matched data only
Coefficients:
Error in symnum(pv, corr = FALSE, na = FALSE, cutpoints = c(0, 0.001, :
'x' must be between 0 and 1
Any idea why I'm getting this error?
Thanks for any help you can provide,
Bill
Hello,
I am wondering how I can get a composite weight from cem_weights and the sampling weight (the inverse of the probability of a unit to be selected) from my survey data.
Shall I simply use cem_weight*sampling_weight as a new sampling weight? If not, is there any proper way to deal with it?
I am using STATA. cem_weight is supposed to be used as "iweight" while sampling weight is supposed to be used as "pweight". I am worried about the compatibility issue to use cem_weight*sampling_weight as a weight.
Thanks.
Best.
Sifan Zhou
Ph.D. student
Economics Department
University at Albany, SUNY