Amelia + CEM + Zelig错误

Question

I am currently trying to write a paper where I need to match on key pre-treatment covariates contained in a multiply imputed data set, and then run a linear regression on this data. I've been able to write code to get zelig to run a linear regression on the multiply imputed data that I got from amelia . Yet, when I try to use cem to match on key pre-treatment covariates, I cannot get zelig to run the model. Here is the output rom my console:

> # Run Amelia, 5 imputations!      
>   complete <- amelia(clean, m = 5, ts = "wave", cs = "country", idvars = c("idnum", "strata", "weight"))     
-- Imputation 1 --

  1  2  3  4  5  6  7  8  9 10 11 12

-- Imputation 2 --

  1  2  3  4  5  6  7  8  9 10 11

-- Imputation 3 --

  1  2  3  4  5  6  7  8  9 10 11 12

-- Imputation 4 --

  1  2  3  4  5  6  7  8  9 10 11 12

-- Imputation 5 --

  1  2  3  4  5  6  7  8  9 10 11 12

>   
> # Estimate naive and controlled linear models with MI data sets using Zelig
>   
>   lm.out.naive <- zelig(dv ~ treat*year + treat + year,  data = complete, model = "ls")

>   lm.out.control <- zelig(dv~ treat*year + treat + year + age + gender + ideo + married, data = complete, model = "ls")

> # Estimate CEM linear model
> 
>   # First, generate matched data set
>     imputed <- complete$imputations[1:15]
>     matched <- cem("treat", drop = c("pais", "wave", "year", "idnum", "strata"),
+                         datalist = imputed, data = clean)
> 
>   # Then, run regression
>     lm.out.match <- zelig(dv ~ treat*year + treat + year + age + gender + ideo + married, data = matched, model = "ls")

Error in resolve_vars(new_groups, tbl_vars(.data)) : 
  unknown variable to group by : by

I am not understanding what the last error means and why I can't run a regression on the matched data set after using multiple imputation. Any insight on this issue would be highly appreciated!

Answer 1

I recently experienced this error using zelig 's logit on a multiply imputed data frame. My solution was to properly specify the data = parameter. You might try this:

lm.out.match <- zelig(dv ~ treat*year + treat + year + 
age + gender + ideo + married, data = mi(matched$imputations), model = "ls")

I recall that an old version of zelig did not require this format, but instead seemed to handle the "amelia" class object as-is.