| ess {epinet} | R Documentation |
Calculate the Effective Sample Size
Description
Calculate the Effective Sample Size for a marginal posterior sample obtained via MCMC
Usage
ess(x, ignoreBurnin = FALSE, burninProportion = 0.1)
Arguments
x |
a numeric vector of length N assumed to be samples from a Markov chain |
ignoreBurnin |
logical indictating whether or not the first burninProportion of vector x should be ignored |
burninProportion |
if ignoreBurnin == TRUE, the first burninProportion*length(x) samples are removed from x before the ess is calculated |
Details
Calculates the effective sample size of x based on an estimate of the lag autocorrelation function. Details of the method are in Section 11.5 of Bayesian Data Analysis, Third Edition, 2013, Andrew Gelman, John B. Carlin, Hal S. Stern, David B. Dunson, Aki Vehtari, Donald B. Rubin.
Value
Returns the estimated effective sample size for the last (1-burninProportion) samples in x.
Author(s)
David Welch david.welch@auckland.ac.nz, Chris Groendyke cgroendyke@gmail.com
References
Gelman, A., Carlin, J.B., Stern, H.S., Dunson, D.B., Vehtari, A., Rubin, D.B., 2013 Bayesian Data Analysis, Third Edition, (Section 11.5), Boca Raton, Florida: CRC Press.
Examples
set.seed(8)
x <- runif(1000)
# expect ESS of close to 900 as samples are iid
ess(x, ignoreBurnin = TRUE)
# no burnin to ignore so ess is actually close to 1000
ess(x, ignoreBurnin = FALSE)
# ESS is a rough measure at best
ess(1:1000,ignoreBurnin = FALSE)