dchibarsq: Mixed chi-squared distributions

Description

Calculates "mixed" chi-squared distributions (mixtures of chi-square(n) and chi-square(n-1)); useful for Likelihood Ratio Tests when parameters are on the boundary

Usage

dchibarsq(x, df = 1, mix = 0.5, log = FALSE)
pchibarsq(p, df = 1, mix = 0.5, lower.tail=TRUE, log.p = FALSE)
qchibarsq(q, df = 1, mix = 0.5)
rchibarsq(n, df = 1, mix = 0.5)

Value

Vectors of probability densities (dchibarsq), cumulative probabilities (pchibarsq), quantiles (qchibarsq), or random deviates (rchibarsq) from Goldman and Whelan's "chi-bar-squared" distribution.

qchibarsq uses simple algebra for df=1 and uniroot for df>1.

Arguments

x: numeric vector of positive values
p: numeric vector of positive values
q: numeric vector of quantiles (0-1)
n: integer: number of random deviates to pick
df: degrees of freedom (positive integer)
mix: mixture parameter: fraction of distribution that is chi-square(n-1) distributed
log: return log densities?
log.p: return log probabilities?
lower.tail: return lower tail values?

Author

Ben Bolker

References

N. Goldman and S. Whelan (2000) "Statistical Tests of Gamma-Distributed Rate Heterogeneity in Models of Sequence Evolution in Phylogenetics", Mol. Biol. Evol. 17:975-978. D. O. Stram and J. W. Lee (1994) "Variance Components Testing in the Longitudinal Fixed Effects Model", Biometrics 50:1171-1177.

Examples

Run this code

x <- rchibarsq(100)
plot(density(x,from=0))
curve(dchibarsq(x),add=TRUE,col=2,from=0)
if (FALSE) {
library(lattice)
print(qqmath(~ simdist,
       distribution=qchibarsq,
       panel = function(x, ...) {
         panel.qqmathline(x, ...)
         panel.qqmath(x, ...)
            }))

}
## create first line of table in Goldman and Whelan 2000
round(qchibarsq(c(0.01,0.05,0.9,0.95,0.975,0.99,0.995),df=1),2)
## check second line of table
round(pchibarsq(c(3.81,5.14,6.48,8.27,9.63),df=2),3)
## create middle column
round(qchibarsq(0.95,df=1:10))

Run the code above in your browser using DataCamp Workspace