BFDP: Bayesian false-discovery probability

In gap: Genetic Analysis Package

Bayesian false-discovery probability

Description

Bayesian false-discovery probability

Usage

BFDP(a, b, pi1, W, logscale = FALSE)

Arguments

a	parameter value at which the power is to be evaluated.
b	the variance for a, or the uppoer point (RR_{hi}) of a 95%CI if logscale=FALSE.
pi1	the prior probabiility of a non-null association.
W	the prior variance.
logscale	FALSE=the orginal scale, TRUE=the log scale.

Details

This function calculates BFDP, the approximate P(H_0|\hat\theta), given an estiamte of the log relative risk, \hat\theta, the variance of this estimate, V, the prior variance, W, and the prior probability of a non-null association. When logscale=TRUE, the function accepts an estimate of the relative risk, \hat{RR}, and the upper point of a 95% confidence interval RR_{hi}.

Value

The returned value is a list with the following components: PH0. probability given a,b). PH1. probability given a,b,W). BF. Bayes factor, P_{H_0}/P_{H_1}. BFDP. Bayesian false-discovery probability. ABF. approxmiate Bayes factor. ABFDP. approximate Bayesian false-discovery probability.

Note

Adapted from BFDP functions by Jon Wakefield on 17th April, 2007.

Author(s)

Jon Wakefield, Jing Hua Zhao

References

\insertRef

wakefield07gap

See Also

FPRP

Examples

## Not run: 
# Example from BDFP.xls by Jon Wakefield and Stephanie Monnier
# Step 1 - Pre-set an BFDP-level threshold for noteworthiness: BFDP values below this
#          threshold are noteworthy
# The threshold is given by R/(1+R) where R is the ratio of the cost of a false
# non-discovery to the cost of a false discovery

T <- 0.8

# Step 2 - Enter up values for the prior that there is an association

pi0 <- c(0.7,0.5,0.01,0.001,0.00001,0.6)

# Step 3 - Enter the value of the OR that is the 97.5% point of the prior, for example
#          if we pick the value 1.5 we believe that the prior probability that the
#          odds ratio is bigger than 1.5 is 0.025.

ORhi <- 3

W <- (log(ORhi)/1.96)^2
W

# Step 4 - Enter OR estimate and 95% confidence interval (CI) to obtain BFDP

OR <- 1.316
OR_L <- 1.10
OR_U <- 2.50
logOR <- log(OR)
selogOR <- (log(OR_U)-log(OR))/1.96
r <- W/(W+selogOR^2)
r
z <- logOR/selogOR
z
ABF <- exp(-z^2*r/2)/sqrt(1-r)
ABF
FF <- (1-pi0)/pi0
FF
BFDPex <- FF*ABF/(FF*ABF+1)
BFDPex
pi0[BFDPex>T]

## now turn to BFDP

pi0 <- c(0.7,0.5,0.01,0.001,0.00001,0.6)
ORhi <- 3
OR <- 1.316
OR_U <- 2.50
W <- (log(ORhi)/1.96)^2
z <- BFDP(OR,OR_U,pi0,W)
z

## End(Not run)

gap documentation built on Aug. 26, 2023, 5:07 p.m.