Combine Multiple Imputation Inferences

DESCRIPTION:

Average estimates, and compute combined standard errors or covariance estimates from multiple analyses on imputed data sets.

USAGE:

miMeanSEDefault(estimate, se, df = NA, n = <see below>, sse = F, cov) 

REQUIRED ARGUMENTS:

estimate
a multiple imputations object whose imputations are numerical atomic objects with the same dimensions. These should contain complete-data estimates (i.e. estimates obtained by filling in missing values with multiple imputations).
se
like estimate, but containing complete-data standard errors for the estimates.

OPTIONAL ARGUMENTS:

df
integer, or vector whose length is the same as the number of estimates; degrees of freedom for complete-data analyses. Set df=Inf (infinity) to obtain degrees of freedom for problems in which complete-data inferences would use normal or chi-square rather than t or F distributions.
n
integer, or vector whose length is the same as the number of estimates; sample size for data. The default value is inferred to be df plus the number of parameters in estimate.
sse
logical, if TRUE then se is assumed to contain squared standard errors (variances) rather than standard errors.
cov
like estimate, but containing complete-data covariance matrices for the estimates, whose diagonal elements are squared standard errors. If cov is supplied then se is ignored.
...
additional arguments, currently ignored.

VALUE:

a list, whose components are described in the help file for . The first two components are:
est
average of the coefficients across imputations.
std.err
standard errors for the coefficients.

DETAILS:

This function performs core calculations for combining estimates and standard errors, and is called my miMeanSE and its auxiliary function, including miMeanSEAux and its methods, and miMeanSEList and miMeanSEMatrix.

If df or n contain non-identical imputations, warnings are issued and certain averages used. If either of these are not supplied, or contain missing values ( NA), then some components of the output will contain missing values.

SEE ALSO:

, , , , , , , .

EXAMPLES:

fit <- miEval(lm(chol14~., data = cholesterolImpExample)) 
sumfit <- miEval(summary(fit)) 
coefs <- miEval(coef(sumfit)) 
coefs[[1]]  # note estimates and std errors 
miMeanSEDefault(miEval(coefs[,1]), miEval(coefs[,2])) 
# May use miMeanSE as a front end 
miMeanSE(miEval(coefs[,1]), miEval(coefs[,2])) 
# May pass additional information, e.g. residual degrees of freedom 
miMeanSE(miEval(coefs[,1]), miEval(coefs[,2]), 
         df = nrow(cholesterolImpExample) - 3)