semTools (version 0.4-12)

runMI: Multiply impute and analyze data using lavaan

Description

This function takes data with missing observations, multiple imputes the data, runs a SEM using lavaan and combines the results using Rubin's rules. Note that parameter estimates and standard errors are pooled by the Rubin's (1987) rule. The chi-square statistics and the related fit indices are pooled by the method described in "chi" argument. SRMR is calculated based on the average model-implied means and covariance matrices across imputations.

Usage

runMI(model, data, m, miArgs=list(), chi="all", miPackage="Amelia", 
	seed=12345, fun, nullModel = NULL, includeImproper = FALSE, ...) 
cfa.mi(model, data, m, miArgs=list(), miPackage="Amelia", chi="all", 
	seed=12345, nullModel = NULL, includeImproper = FALSE, ...)
sem.mi(model, data, m, miArgs=list(), miPackage="Amelia", chi="all",  
	seed=12345, nullModel = NULL, includeImproper = FALSE, ...)
growth.mi(model, data, m, miArgs=list(), miPackage="Amelia", chi="all", 
	seed=12345, nullModel = NULL, includeImproper = FALSE, ...)
lavaan.mi(model, data, m, miArgs=list(), miPackage="Amelia", chi="all", 
	seed=12345, nullModel = NULL, includeImproper = FALSE, ...)

Arguments

model

lavaan syntax for the model to be analyzed.

data

Data frame with missing observations or a list of data frames where each data frame is one imputed data set (for imputed data generated outside of the function). If a list of data frames is supplied, then other options can be left at the default.

m

Number of imputations wanted.

miArgs

Addition arguments for the multiple-imputation function. The arguments should be put in a list (see example below).

miPackage

Package to be used for imputation. Currently these functions only support "Amelia" or "mice" for imputation.

chi

The method to combine the chi-square. Can be one of the following: "mr" for the method proposed for Meng & Rubin (1992), "mplus" for the method used in Mplus (Asparouhov & Muthen, 2010), "lmrr" for the method proposed by Li, Meng, Raghunathan, & Rubin (1991), "all" to show the three methods in the output, and "none" to not pool any chi-square values. The default is "all".

seed

Random number seed to be used in imputations.

nullModel

lavaan syntax for the null model. If not specified, the default null model from lavaan is used.

includeImproper

If TRUE, the function will combine the results with improper solutions to get the combined solution.

fun

The character of the function name used in running lavaan model ("cfa", "sem", "growth", "lavaan").

...

Other arguments to be passed to the specified lavaan function ("cfa", "sem", "growth", "lavaan").

Value

The lavaanStar object which contains the original lavaan object (where the appropriate parameter estimates, appropriate standard errors, and chi-squares are filled), the additional fit-index values of the null model, which need to be adjusted to multiple datasets, and the information from pooling multiple results.

References

Asparouhov T. & Muthen B. (2010).Chi-Square Statistics with Multiple Imputation. Technical Report. www.statmodel.com.

Li, K.H., Meng, X.-L., Raghunathan, T.E. and Rubin, D.B. (1991). Significance Levels From Repeated p-values with Multiply-Imputed Data. Statistica Sinica, 1, 65-92.

Meng, X.L. & Rubin, D.B. (1992). Performing likelihood ratio tests with multiply-imputed data sets. Biometrika, 79, 103 - 111.

Rubin, D.B. (1987) Multiple Imputation for Nonresponse in Surveys. J. Wiley & Sons, New York.

Examples

library(lavaan)

HS.model <- ' visual  =~ x1 + x2 + x3
              textual =~ x4 + x5 + x6
              speed   =~ x7 + x8 + x9 '

HSMiss <- HolzingerSwineford1939[,paste("x", 1:9, sep="")]
randomMiss <- rbinom(prod(dim(HSMiss)), 1, 0.1)
randomMiss <- matrix(as.logical(randomMiss), nrow=nrow(HSMiss))
HSMiss[randomMiss] <- NA

out <- cfa.mi(HS.model, data=HSMiss, m = 3, chi="all")
summary(out)
inspect(out, "fit")
inspect(out, "impute")

## Not run: 
# ##Multiple group example
# HSMiss2 <- cbind(HSMiss, school = HolzingerSwineford1939[,"school"])
# out2 <- cfa.mi(HS.model, data=HSMiss2, m = 3, miArgs=list(noms="school"), chi="MR", group="school")
# summary(out2)
# inspect(out2, "fit")
# inspect(out2, "impute")
# 
# ##Example using previously imputed data with runMI
# library(Amelia)
# 
# modsim <- '
# f1 =~ 0.7*y1+0.7*y2+0.7*y3
# f2 =~ 0.7*y4+0.7*y5+0.7*y6
# f3 =~ 0.7*y7+0.7*y8+0.7*y9'
# 
# mod <- '
# f1 =~ y1+y2+y3
# f2 =~ y4+y5+y6
# f3 =~ y7+y8+y9'
# 
# datsim <- simulateData(modsim,model.type="cfa", meanstructure=TRUE, 
# 	std.lv=TRUE, sample.nobs=c(200,200))
# randomMiss2 <- rbinom(prod(dim(datsim)), 1, 0.1)
# randomMiss2 <- matrix(as.logical(randomMiss2), nrow=nrow(datsim))
# datsim[randomMiss2] <- NA
# datsimMI <- amelia(datsim,m=3, noms="group")
# 
# out3 <- runMI(mod, data=datsimMI$imputations, chi="LMRR", group="group", fun="cfa")
# summary(out3)
# inspect(out3, "fit")
# inspect(out3, "impute")
# 
# # Categorical variables
# popModel <- "
# f1 =~ 0.6*y1 + 0.6*y2 + 0.6*y3 + 0.6*y4
# y1 ~*~ 1*y1
# y2 ~*~ 1*y2
# y3 ~*~ 1*y3
# y4 ~*~ 1*y4
# f1 ~~ 1*f1
# y1 | 0.5*t1
# y2 | 0.25*t1
# y3 | 0*t1
# y4 | -0.5*t1
# "
# analyzeModel <- "
# f1 =~ y1 + y2 + y3 + y4
# y1 ~*~ 1*y1
# y2 ~*~ 1*y2
# y3 ~*~ 1*y3
# y4 ~*~ 1*y4
# "
# dat <- simulateData(popModel, sample.nobs  = 200L)
# miss.pat <- matrix(as.logical(rbinom(prod(dim(dat)), 1, 0.2)), nrow(dat), ncol(dat))
# dat[miss.pat] <- NA
# out5 <- cfa.mi(analyzeModel, data=dat, ordered=paste0("y", 1:4), m = 3, 
# 	miArgs=list(ords = c("y1", "y2", "y3", "y4")))
# summary(out5)
# inspect(out5, "fit")
# inspect(out5, "impute")
# 
# ## End(Not run)