R/fpval.R

In gausscov: The Gaussian Covariate Method for Variable Selection

#' Calculates the regression coefficients, the P-values and the standard P-values for the chosen subset ind
#
#' @param y The dependent variable.
#' @param x  The covariates.
#' @param ind The indices of the subset of the covariates whose P-values are required.
#' @param q   The total  number of covariates from which ind was selected. If q=-1 the number of covariates of x minus length ind plus 1 is taken.
#' @param inr Logical If TRUE intercept to be included
#' @param xinr Logical If TRUE intercept already included.
#' @return apv In order the subset ind, the regression coefficients, the Gaussian P-values, the standard F P-values.
#' @return res The residuals.
#' @examples 
#' a<-fpval(boston[,14],boston[,1:13],c(1,2,4:6,8:13))
fpval<-function(y,x,ind,q=-1,inr=T,xinr=F){
   	n<-length(y)
	ki<-length(ind)
	x<-matrix(x,nrow=n)
	y<-matrix(y,ncol=1)
	ind<-matrix(ind,nrow=1)
	if(!xinr){
		if(inr){
         		tmpx<-double(n)+1
         		x<-cbind(x,tmpx)
 		xinr<-TRUE
		}
     	}
   	kx<-length(x)/n
 	if(xinr){
		if(max(ind)<kx){ind<-c(ind,kx)}
	}
	kii<-length(ind)
	xx<-x[,ind]
	xx<-matrix(xx,nrow=n)
    	b<-lm(y~0+xx)
	ind1<-(1:kii)[is.na(b$coef)==FALSE]
	ind<-ind[ind1]
    	res<-as.double(b$res)
	result<-summary(b)[[4]]
	kii<-length(result[,1])
	apv<-double(2*kii)
	apv<-matrix(apv,nrow=kii,ncol=2)
#	tvl<-matrix(result[1:kii,3],ncol=1)
    	apv[,2]<-matrix(result[1:kii,4],ncol=1)
# 	tvl<-1-1/(1+tvl^2/(n-kii))
	if(q==-1){q<-kx-kii+1}
    	apv[1:kii,1]<-pbeta(apv[1:kii,2],1,q)
    	if(xinr){apv[kii,1]<-apv[kii,2]
        		ind[kii]<-0
    	}
	beta<-matrix(result[,1],ncol=1)
	ind<-matrix(ind,ncol=1)
    	apv<-cbind(ind,beta,apv)
    	list(apv,res)
}