0
$\begingroup$

I have imputed a data set with mice and want to run a multivariate multiple linear regression on the imputed data. Below is a description of what I have done.

# Imputing data
imp<-mice(df, m=25, maxit=100, seed=1234, meth=initial1$method, 
             pred=initial1$predictorMatrix)

# Running model
fit<-with(imp, lm(cbind(dv1,dv2)+iv3+iv4+iv5))

# Pooling results
res<-pool(fit)

However, when trying to pool the results, I get this error code:

Error in glance.mlm(X[[i]], ...) : 
  glance does not support multiple responses

Any suggestions about how to get it to work?

$\endgroup$
1
$\begingroup$

You could try to apply Rubin's rules directly to the results of the multivariate regressions on each of the imputed data sets. You average the coefficient matrices (from coef() on each model) to get the pooled estimates of the coefficients, calculate variance among the individual coefficient matrices to get the between-imputation variance, and then average the coefficient covariance matrices (from vcov() on each model) to get the within-imputation variance estimates.

Marshall et al show formulas (in Table 2) for significance testing once those matrices are prepared from the individual regressions.

$\endgroup$
0
$\begingroup$

Entered a plus instead of a tilde. Here is the correct code:

imp <- mice(df, m=25, maxit=100, seed=1234, meth=initial1method,
            pred=initial1predictorMatrix)

fit <- with(imp, lm(cbind(dv1,dv2)~iv3+iv4+iv5))

res <- pool(fit)
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.