BayesPen.refit: Bayesian Penalized Credible Regions Solution Path Refit

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

Description

Refits the solution path given by BayesPen.

Usage

1
BayesPen.refit(y, x, fit, joint, max.refit, ...)

Arguments

y

A n-vector of responses. If fit is a list from BayesPen.lm or BayesPen.lm.confounders then y is not required.

x

A n x p design matrix that includes all potential covariates. In the confounder selection case this includes the exposures and confounders, i.e. cbind(x,u). If fit is a list from BayesPen.lm or BayesPen.lm.confounders then x is not required.

fit

A list returned from BayesPen, BayesPen.lm, or BayesPen.lm.confounders.

joint

For variable selection this indicates if the joint or marginal solution path should be used. Joint must be TRUE for confounder selection.

max.refit

The maximum number of models to be refit.

...

These are additional terms passed to glm to refit a glm other than linear. The default is linear.

Details

This refits each model in the solution path with the frequentist model using the glm function.

Value

coefs

A matrix of regression coefficients for each model in the solution path. The regression coefficients for parameters omitted from a model are set to 0.

SSE

SSE of each refitted model.

dev

Deviance of each refitted model.

df

Error degrees of freedom from each refitted model.

joint

Returns the logical joint.

Author(s)

Ander Wilson, Howard D. Bondell, and Brian J. Reich

References

Bondell, H. D. and Reich, B. J. (2012). Consistent high-dimensional Bayesian variable selection via penalized credible regions. J. Am. Statist. Assoc. 107, 1610-1624.

Wilson A., Reich B. J. (2014). Confounder selection via penalized credible regions. Biometrics 70: 852-861.

See Also

BayesPen

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
######################
#Variable Selection
set.seed(1234)
dat <- SimExample(500,model="BR1")
X <- dat$X
y <- dat$y

#fit the full model assuming flat priors on beta
betahat <- solve(t(X)%*%X) %*% t(X) %*% y
cov <- solve(t(X)%*%X) * sum((X%*%betahat-y)^2)/(length(y)-length(betahat))

#find solution path
fit.BayesPen <- BayesPen(beta=betahat, beta_cov=cov)

#refit the model
refit <- BayesPen.refit(y,X,fit.BayesPen)

#plot it
BayesPen.plot(refit)

AnderWilson/BayesPen documentation built on May 5, 2019, 4:56 a.m.