BmaSamples {bfp}

Bayesian model averaging over multiple fractional polynomial models

Description

Draw samples from the Bayesian model average over the models in saved in a BayesMfp-object.

Usage

BmaSamples(object, sampleSize = length(object) * 10, postProbs =
posteriors(object), gridList = list(), gridSize = 203, newdata=NULL,
verbose = TRUE, includeZeroSamples=FALSE)  

Arguments

object

valid BayesMfp object containing the models over which to average

sampleSize

sample size (default is 10 times the number of models)

postProbs

vector of posterior probabilites (will be normalized within the function, defaults to the normalized posterior probabilities)

gridList

optional list of appropriately named grid vectors for FP evaluation, default is a length (gridSize - 2) grid per covariate additional to the observed values (two are at the minimum and maximum)

gridSize

see above (default: 203)

newdata

new covariate data.frame with exactly the names (and preferably ranges) as before (default: no new covariate data)

verbose

should information on sampling progress be printed? (default)

includeZeroSamples

should the function and coefficient samples include zero samples, from models where these covariates are not included at all? (default: FALSE, so the zero samples are not included)

Value

Return an object of class BmaSamples, which is a list with various elements that describe the BayesMfp object over which was averaged, model frequencies in the samples, the samples themselves etc:

priorSpecs

the utilized prior specifications

termNames

a list of character vectors containing the names of uncertain covariate groups, fractional polynomial terms and fixed variables

shiftScaleMax

matrix with 4 columns containing preliminary transformation parameters, maximum degrees and cardinalities of the powersets of the fractional polynomial terms

y

the response vector

x

the shifted and scaled design matrix for the data

randomSeed

if a seed existed at function call (get(".Random.seed", .GlobalEnv)), it is saved here

modelFreqs

The table of model frequencies in the BMA sample

modelData

data frame containing the normalized posterior probabilities of the models in the underlying BayesMfp object, corresponding log marginal likelihoods, model prior probabilities, posterior expected covariance and shrinkage factors, coefficients of determination, powers and inclusions, and finally model average weights and relative frequencies in the BMA sample.

sampleSize

sample size

sigma2

BMA samples of the regression variance

shrinkage

BMA samples of the shrinkage factor

fixed

samples of the intercept

bfp

named list of the FP function samples, where each element contains one FP covariate and is a matrix (samples x grid), with the following attributes:

whereObsVals

where in the scaled grid are the originally observed covariate values? (integer vector of the indexes)

scaledGrid

numeric vector with the positions of the scaled grid points, corresponding to the columns of the samples matrix

counter

how often has this covariate been included in the BMA sample? (identical to the number of rows in the samples matrix)

uc

named list of the uncertain fixed form covariates, where each element contains the coefficient samples of one group: in a matrix with the attribute counter as number of samples in the rows, and the columns are appropriately named to correspond to the single design variables.

fitted

fitted values of all models in object, in a matrix with layout models x observations.

predictions

samples from the predictive distribution at the covariates given in newdata

predictMeans

means of the predictive distribution at the covariates given in newdata

See Also

BmaSamples Methods, BayesMfp

Examples

## construct a BayesMfp object
set.seed(19)

x1 <- rnorm (n=15)
x2 <- rbinom (n=15, size=20, prob=0.5) 
x3 <- rexp (n=15)

y <- rt (n=15, df=2)

test <- BayesMfp (y ~ bfp (x2, max = 4) + uc (x1 + x3), nModels = 200, method="exhaustive")

## now draw samples from the Bayesian model average
testBma <- BmaSamples (test)
testBma

## We can also draw predictive samples for new data points, but then
## we need to supply the new data to BmaSamples:
newdata <- data.frame(x1 = rnorm(15),
                      x2 = rbinom(n=15, size=5, prob=0.2) + 1,
                      x3 = rexp(n=15))
testBma <- BmaSamples(test, newdata=newdata)
predict(testBma)

## test that inclusion of zero samples works
testBma <- BmaSamples (test, includeZeroSamples=TRUE)
testBma