Aisyah,

The following easy code is very efficient for accurate missing data imputation:

##

Robust sequential imputation of missing values

require(rrcovNA)

library(rrcovNA)

data(bush10)

> head(bush10)

V1 V2 V3 V4 V5

1 111 145 188 190 260

2 113 NA 187 190 259

3 NA 150 195 192 259

4 110 147 211 NA 262

5 101 136 240 200 266

6 93 125 262 203 271

impSeqRob(bush10) # impute squentially missing data

Little and Rubin (2002) give good discussion of missing data issues.  For discrete data, some individuals have used EMCAT or other modules in R.  You build the model (essentially fitted contingency table).  As a record with missing data arrives, you find all cells in the table that agree with the record that has missing data.  To preserve joint distributions, you draw one of the match records probability proportional to size (where the size measure of a cell occurring in the contingency table).

Thank you very much william,

But, the thing i wonder is that EMCAT does not listed in this link:

http://cran.r-project.org/web/packages/available_packages_by_name.html

however, i would  apply your recommended package in my analysis

Treatment of missing data in Structural equation models (SEM) or linear regression models can be done by Full Information Maximum Likelihood (FIML) estimation. Given missing completly at random or missing at random this leads to unbiased estimations. The package "lavaan" is, to my knowledge, the most complex SEM package for R. It offers the option to fit models by FIML.

But be aware that FIML is only applicable for models with continious outcomes. If you have categorical outcomes or mediators in your model it will not work.

For these conditions multiple imputation methods are the best solution. There will surely exist R packages on these too.

If your study question may be handeled by propensity score analysis, there also other methods that can be used.

Dear Daniel, 

Thank you for your suggestion.

But, do you mean that multiple imputation is applicable for the various data types?

I have not yet done multiple imputation, but to my knowlodge there are ways to handle different kinds of data. One way is to use approximations to a normal model. Good paper on the topic is:

Graham JW. Missing data analysis: Making it work in the real world. Annual review of psychology. Palo Alto: Annual Reviews; 2009. p. 549-576.

thank you very much Daniel

The paper may be useful to you.

Conference Paper Comparative Statistical Algorithms for Imputation of Missing...

I would be very careful about using imputation to fill in missing data, especially if the percentage of missing data is large.

If there is some structure to your missing data, for example if they are "left censored" (e.g., below an instruments limit of detection) or "right censored" (e.g., a clinician is unable to follow a patient through to the end of a study), then NADA is an R package that provides MLE estimates (using the cenmle function).   I'd be happy to share working R code if your missing data follow this pattern.

Al

 Thank you watheq..appreciate with your shared article.

Thank you Albert Parker, let me try the NADA in R. Currently im trying to test with Lavaan but seem it is still in beta phase and i got some errors with the code.

If you could share your code in R, it is highly appreciated ;)

I would like to share the infor that i found for missing data packages in R and as for my future reference:

## Some R packages relevant to missing value imputation

# Amelia Amelia II: A Program for Missing Data (Version: 1.5-4, Published: 2011-08-23)

# ForImp Imputation of missing values through a forward imputation algorithm (Version: 1.0, Published: 2011-06-15)

# mi Missing Data Imputation and Model Checking (Version: 0.09-14, Published: 2011-04-25)

# missForest Nonparametric Missing Value Imputation using Random Forest

# mitools Tools for multiple imputation of missing data

# VIM Visualization and Imputation of Missing Values

# vmv Visualization of Missing Values

# arrayImpute Missing imputation for microarray data

# arrayMissPattern Exploratory analysis of Missing patterns for microarray data

# MissingDataGUI A GUI for Missing Data Exploration

## Some R packages for analysis with missing values

# cat Analysis of categorical-variable datasets with missing values

# missMDA Handling missing values with/in multivariate data analysis (principal component methods)

# mlmmm ML estimation under multivariate linear mixed models with missing values

# monomvn Estimation for multivariate normal and Student-t data with monotone missingness

# mvnmle ML estimation for multivariate normal data with missing values

# NestedCohort Survival Analysis for Cohorts with Missing Covariate Information

# norm Analysis of multivariate normal datasets with missing values

Aisyah,

I apologize for not following up sooner, I missed your request for the R code to implement the NADA package when some data are left-censored. 

Package nada in R is explained at:

   http://cran.r-project.org/web/packages/NADA/NADA.pdf

To perform a 1-way ANOVA wrt groups in group:

   cenmle(resp,censored,group,dist="gaussian")

Inputs: resp is a data vector, censored is logical vector with TRUE for left-censored data and FALSE otherwise, and group specifies levels of some factor.

To perform n-way ANOVA or regression:

  cenreg(Cen(resp,censored) ~ x + y*z,dist="gaussian")

where x, y and z are explanatory var’s, covariates, or interactions.

I hope this is helpful,

Al

Thank you Albert Parker,

I would try to implement NADA in R ;)

Aisyah,

The following easy code is very efficient for accurate missing data imputation:

##

Robust sequential imputation of missing values

require(rrcovNA)

library(rrcovNA)

data(bush10)

> head(bush10)

V1 V2 V3 V4 V5

1 111 145 188 190 260

2 113 NA 187 190 259

3 NA 150 195 192 259

4 110 147 211 NA 262

5 101 136 240 200 266

6 93 125 262 203 271

impSeqRob(bush10) # impute squentially missing data