Combine Multiple Imputation Inferences

DESCRIPTION:

Combine chi-square or F statistics from multiple imputations.

USAGE:

miChiSquareTest(x, df) 
miFTest(x, df1, df2) 

REQUIRED ARGUMENTS:

x
an miVariable or miList object whose imputations are scalar chi-square or F statistics.
df
degrees of freedom for the chi-square statistics.
df1, df2
numerator and denominator degrees of freedom for the F statistics.

VALUE:

list with components
Fstatistic
F-statistic obtained by combining the statistics.
df1
numerator degrees of freedom; same as the input df or df1.
df2
denominator degrees of freedom (smaller than input df2).
r
estimated average relative increase in variance due to nonresponse.
p
approximate p-value corresponding to the F-statistic. These should be used for screening only; the actual p-value may be larger or smaller by a factor of two.

REFERENCES:



Schafer, J. L. (1997). Analysis of Incomplete Multivariate Data. Chapman & Hall, London, page 115.

Li, Meng, Raghunathan and Rubin (1991). Significance Levels from Repeated p-Values with Multiply-imputed Data. Statistica Sinica 1, 65-92.

Hesterberg, T. (1998). Combining multiple imputation t, chi-square, and F inferences. Insightful Technical Report number 75.

SEE ALSO:

, , .

EXAMPLES:

m.barley <- barley.exposed 
w <- runif(length(barley.exposed)) < .1 
m.barley[w] <- NA 
m.barley <- miVariable(m.barley, 
      data.frame(matrix(rpois(4*sum(w), barley.exposed[w]+.1), 
                        sum(w), 4))) 
ml <- miApply(m.barley, loglin, margin = list(1:2, c(1, 3))) 
miChiSquareTest(miApply(ml, "[[", "pearson"), 
                df = miApply(ml, "[[", "df")) 
fit <- miEval(lm(chol14~., data = cholesterolImpExample)) 
sumfit <- miEval(summary(fit)) 
fstat <- miEval(sumfit$fstatistic) 
miFTest(x =   miEval(fstat["value"]), 
        df1 = miEval(fstat["numdf"]), 
        df2 = miEval(fstat["dendf"]))