impNorm: Imputation and prediction for incomplete multivariate normal...
In norm2: Analysis of Incomplete Multivariate Data under a Normal Model

Description Usage Arguments Details Value Author(s) References See Also Examples

Simulates or predicts missing values from their predictive distribution given the observed data under a normal model with fixed parameters.

impNorm(obj, ...)


## Default S3 method:
impNorm(obj, x = NULL, intercept = TRUE, param,
   seeds = NULL, method = "random", ...)


## S3 method for class 'formula'
impNorm(formula, data, param,
   seeds = NULL, method = "random", ...)


## S3 method for class 'norm'
impNorm(obj, param = obj$param, seeds = NULL, 
   method = "random", ...)

`obj`	an object used to select a method. It may be `y`, a numeric matrix, vector or data frame of responses to be modeled as normal. Missing values (`NA`s) are allowed. If `y` is a data frame, any factors or ordered factors will be replaced by their internal codes, and a warning will be given. Alternatively, this first argument may be an object of class `"norm"` resulting from a call to `emNorm` or `mcmcNorm`; see DETAILS.
`x`	a numeric matrix, vector or data frame of covariates to be used as predictors for `y`. Missing values (`NA`'s) are not allowed. If `x` is a matrix, it must have the same number of rows as `y`. If `x` is a data frame, any factors or ordered factors are replaced by their internal codes, and a warning is given. If `NULL`, it defaults to `x = rep(1,nrow(y))`, an intercept-only model.
`intercept`	if `TRUE`, then a column of `1`'s is appended to `x`. Ignored if `x = NULL`.
`formula`	an object of class `"formula"` (or one that can be coerced to that class): a symbolic description of the model which is provided in lieu of `y` and `x`. The details of model specification are given under DETAILS.
`data`	an optional data frame, list or environment (or object coercible by `as.data.frame` to a data frame) containing the variables in the model. If not found in `data`, the variables are taken from `environment(formula)`, typically the environment from which `impNorm` is called.
`param`	assumed values for the model parameters. This must be a list with two named components, `beta` and `sigma`, which are numeric matrices with correct dimensions. In most circumstances, the parameter values will be obtained from a run of `emNorm` or `mcmcNorm`; see DETAILS.
`seeds`	two integers to initialize the random number generator; see DETAILS.
`method`	if `"random"`, the missing values in each row of `y` will be simulated from their joint predictive distribution given `x` and the observed values in `y`. If `"predict"`, missing values will be replaced by regression predictions given the observed values. See DETAILS.
`...`	values to be passed to the methods.

This function is used primarily in conjunction with mcmcNorm to draw multiple imputations by the multiple-chain method. In those instances, the simplest way to call impNorm is to provide an object of class "norm" as its first argument, where that object is the result of a call to mcmcNorm. The parameter values stored in that object will then be passed to impNorm automatically.

Alternatively, one may call impNorm by providing as the first argument y, a vector or matrix of data to be modeled as normal, and an optional vector or matrix of predictors x. Missing values NA are allowed in y but not in x.

A third way to call impNorm is to provide formula, a formula for a (typically multivariate) linear regression model in the manner expected by lm. A formula is given as y ~ model, where y is either a single numeric variable or a matrix of numeric variables bound together with the function cbind. The right-hand side of the formula (everything to the right of ~) is a linear predictor, a series of terms separated by operators +, : or * to specify main effects and interactions. Factors are allowed on the right-hand side and will enter the model as contrasts among the levels. The intercept term 1 is included by default; to remove the intercept, use -1.

norm2 functions use their own internal random number generator which is seeded by two integers, for example, seeds=c(123,456), which allows results to be reproduced in the future. If seeds=NULL then the function will seed itself with two random integers from R. Therefore, results can also be made reproducible by calling set.seed beforehand and taking seeds=NULL.

a data matrix resembling the original data y, but with NA's replaced by simulated values or predictions.

Joe Schafer Joseph.L.Schafer@census.gov

Schafer, J.L. (1997) Analysis of Incomplete Multivariate Data. London: Chapman & Hall/CRC Press.

For more information about this function and other functions in the norm2 package, see User's Guide for norm2 in the library subdirectory doc.

emNorm, mcmcNorm

## run EM for marijuana data with ridge prior
data(marijuana)
emResult <- emNorm(marijuana, prior="ridge", prior.df=0.5)

## generate 25 multiple imputations by running 25 chains
## of 100 iterations each, starting each chain at the 
## posterior mode
set.seed(456)
imp.list <- as.list(NULL)
for(m in 1:25){
   mcmcResult <- mcmcNorm(emResult, iter=100)
   imp.list[[m]] <- impNorm(mcmcResult)}