The key feature of MI over the other methods that are either parametric (e.g., likelihood-based) or non-parametric (e.g., weighting-based) is its versatility in the post-imputation phase as MI can serve multiple analytical goals using the same multiply-imputed datasets. Regardless of the nature of the post-imputation phase, MI inference treats missing data as an explicit source of random variability and the uncertainty induced by this is explicitly incorporated into the overall uncertainty measures of the underlying inferential process. This is accomplished by repeating the same complete-data analysis on the imputed data, and combining the estimates and standard errors under rules defined by Rubin ( 1987), including an explicit estimate of the degree of uncertainty due to the missing-data methodology.

To produce the imputations, some assumptions about the data (typically a parametric model) and the mechanism producing missing data need to be made. The assumed data model should be plausible and should be somewhat related to the analyst’s investigation (Meng 1994). This model forms the basis to approximate the distribution in which the missing data conditional on observed data (i.e. predictive distribution of missing data). The software given in this volume develops computational routines for sampling from this approximate predictive distribution of missing data. The collection of the software presented here adapts roughly three approaches for the sampling.

The first approach jointly models variables subject to missingness, thus jointly samples from the underlying predictive distribution. The software by Yuan (2011) and Carpenter, Gold-stein, and Kenward ( 2011) implement this approach. The second approach has been an excellent alternative to this method. It approximates the joint modeling approach with potentially incoherent variable-by-variable approach (Su, Gelman, Hill, and Yajima 2011 ; Buuren and Groothuis-Oudshoorn 2011 ; Royston and White 2011). While “incoherence” has been a subject of debate, this method has been quite successfully applied in many survey settings where joint approach is essentially not applicable. The final approach pertains to a collection of resampling-based and/or matching-based algorithms (Honaker, King, and Blackwell 2011 ; Yuan 2011).

Successful and thorough development of the MI paradigm allowed many of the current studies in social, medical or educational studies to consider MI. However, the lack of thorough software development as well as appropriate documentation (e.g., user-friendly manuals) has prevented a full incorporation of MI into these subject-matter fields. Most of the current software has only been developed since 2000s. Stand-alone Windows software norm (accompanying Schafer 1997), operating under a multivariate normality assumption, was arguably the first one available as both stand-alone software and S-PLUS library (Insightful Corp. 2003). Additional S-PLUS libraries for categorical and a mixture of categorical and continuous incomplete data later evolved into the engine of the bigger S-PLUS library called missing (Schimert, Schafer, Hesterberg, Fraley, and Clarkson 2000). In 2002, SAS (SAS Institute Inc. 2003) implemented some of Schafer’s normal-based routines along with other routines for monotonic missingness patterns as well as matching-based MI routines (PROC MI and PROC MIANALYZE). This initiation by SAS was an important step in making the modern missing data techniques accessible and visible to a wider group of practitioners. Other major implementations of MI routines are found in IVEware (Raghunathan, Lepkowski, and VanHoewyk 2001) and mice (Buuren and Groothuis-Oudshoorn 2011) which both provide alternative imputation strategies for survey data with variables measured on a diverse set of scales. mice has been implemented as both S-PLUS library and an R package (R Development Core Team 2011) while IVEware is a SAS macro (SAS Institute Inc. 2003).

The second contribution is by Buuren and Groothuis-Oudshoorn (2011) illustrating an increasingly popular approach to producing multiple imputations in settings pertaining to variables that are of varying natures and measured with restrictions. They present the most recent version of their R (R Development Core Team 2011) package called mice which imputes incomplete values by fully conditional specification. This package offers many practical solutions including predictor selection, passive imputation and automatic pooling to combine estimates from the multiply imputed datasets. These features are also extended to the multilevel continuous data. Finally, this version adds a capability of multilevel MI and interactive use with SPSS (IBM Corporation 2011). The third contribution presents an implementation of a similar approach in Stata (StataCorp. 2011). The manuscript by Royston and White ( 2011) describes ice which is the Stata module of the approach using the fully automatic pooling to produce multiple imputation. Royston and White (2011) illustrate this fully-integrated module in Stata using real data from an observational study in ovarian cancer.

Joint modeling approach follows the variable-by-variable approach. Carpenter and his colleagues describe a comprehensive module called REALCOM-IMPUTE of the multilevel model fitting software MLwiN (Carpenter et al. 2011). Variables subject to missing values are modeled under a multivariate latent normal model with random-effects, which is used as a basis to approximate the underlying posterior predictive distribution. The authors use Markov chain Monte Carlo (MCMC) simulation techniques to fit the imputation models and thus draw the multiple imputations. The software also allows for weights to account for sampling design both at level 1 and level 2. A variety of variables can be imputed: continuous, ordinal or nominal. Users can further analyze the imputed datasets under multilevel models and combine estimates using MI rules defined by Rubin ( 1987).

Another increasingly popular package is PROC MI and PROC MIANALYZE procedures of SAS. Yuan (2011) illustrates how to conduct MI inference in SAS. PROC MI implements three major techniques one can adopt to produce multiple imputations. Specific choice of these techniques depends on the missingness pattern and the type of imputed variable. For the problems with monotone patterns of missingness (i.e. a variable missing implies that all subsequent variables to be missing), one can choose from the following three methods depending on the type of the variable(s) to be imputed: matching (using propensity score or predictive mean) or MCMC which draws imputations from a multivariate normal if the underlying variables are continuous. If they are categorical, one can choose logistic regression or discriminant-function-based method to match. For the arbitrary patterns of missingness, one would have to approximate the underlying posterior predictive distribution using a multivariate normal distribution with a set of priors provided by PROC MI (e.g., ridge or Jeffreys prior).

It should be noted that this volume is not intended to be the exclusive source of the multiple imputation software. There are many other software companion to the methods developed so far. Some of the most commonly-used software include R packages Hmsic (Harrell 2011, function aregImpute), norm (Novo and Schafer 2010), cat (Harding, Tusell, and Schafer 2011), mix (Schafer 2010) for a variety of techniques to create multiple imputations in continuous, categorical or mixture of continuous and categorical datasets. Another useful R package for imputing continuous variables in clustered or longitudinal designs is pan (Schafer 2011 ; Schafer and Yucel 2002). There is also a very important package in the form of SAS macro for multiple imputation using a sequences of regression models. This SAS-callable program is called IVEware written by Raghunathanet al. (2001) and very similar to the R and Stata implementation of mice and ice.