Evaluate an expression in a list of imputed data sets

Description

The functions with and within evaluate R expressions in a list of multiply imputed data sets.

Usage

## S3 method for class 'mitml.list'
with(data, expr, ...)
## S3 method for class 'mitml.list'
within(data, expr, ignore=NULL, ...)

Arguments

Details

The two functions are defined as with and within methods for objects of class mitml.list. Both with and within evaluate an R expression in each of the imputed data sets. However, the two functions return different values: with returns the evaluated expression, whereas within returns the resuling data sets. The ignore argument may be used to declare objects that are not to be saved within within.

Value

with: Returns the evaluated expression from each data set as a list (class mitml.result. This is useful for fitting statistical models to multiply imputed data. The list of fitted models can be analyzed using testEstimates, testModels, testConstraints, or anova.

within: Evaluates the R expression for each data set and returns the altered data sets as a list mitml.list. This is useful for manipulating the data prior to data analysis (e.g., centering, calculating cluster means, etc.).

Author(s)

Simon Grund

See Also

mitmlComplete, anova.mitml.result, testEstimates, testModels, testConstraints

Examples

data(studentratings)

fml <- ReadDis + SES ~ ReadAchiev + (1|ID)
imp <- panImpute(studentratings, formula=fml, n.burn=1000, n.iter=100, m=5)

implist <- mitmlComplete(imp, print=1:5)

# * Example 1: data manipulation

# calculate and save cluster means
new1.implist <- within(implist, Means.ReadAchiev <- clusterMeans(ReadAchiev, ID))

# center variables, calculate interaction terms, ignore byproducts
new2.implist <- within(implist,{
  M.SES <- mean(SES)
  M.CognAbility <- mean(CognAbility)
  C.SES <- SES - M.SES
  C.CognAbility <- CognAbility - M.CognAbility
  SES.CognAbility <- C.SES * C.CognAbility
}, ignore=c("M.SES", "M.CognAbility"))

# * Example 2: fitting statistical models

# fit regression model
fit.lm <- with(implist, lm(ReadAchiev ~ ReadDis))

# fit multilevel model using lme4
require(lme4)
fit.lmer <- with(implist, lmer(ReadAchiev ~ ReadDis + (1|ID)))

# * Example 3: manual extraction of variance estimates
require(lme4)
fit.lmer <- with(implist, lmer(SES ~ (1|ID)))

# extract level-1 and level-2 variances
var.l1 <- sapply(fit.lmer, function(z) attr(VarCorr(z),"sc")^2)
var.l2 <- sapply(fit.lmer, function(z) VarCorr(z)$ID[1,1])

# calculate final estimate of the intraclass correlation
ICC <- mean( var.l2 / (var.l2+var.l1) )