jackafterboot {MCE} | R Documentation |
Efron and Tibshirani have introduced a method for estimating Monte Carlo error for the estimates generated from the bootstrap. This method uses ideas similar to a jacknife method.
jackafterboot(data, B = 100, type = "", truth)
data |
vector. replication results |
B |
numeric (integer). number of bootstrap samples to be taken |
type |
character. type of simulation result including "mean", "SE", "RE", and "PB" for mean, standard error, relative efficiency, and percent bias. |
truth |
numeric. if type="PB" the true value. |
This function takes original data and perfoms the bootstrap as well as estimates the error in the bootstrap estimates. For an original data set with N points and B bootstrap samples, the first estimate is calculated using only the replicates that do not contain the first point of the data. Then the second estimate is from only the replicates that do not contain the second point, and so on until all N points have been removed. An estimate is calculated from these replications without the nth point. Then the standard deviation from the N estimates should estimate Monte Carlo error.
Jackknife-After-Bootstrap estimate of MCE
Efron notes that this technique runs into trouble if one point appears in all B bootstrap samples, but claims this is rare when N >= 10 and B >= 20.
Elizabeth Koehler
Efron, B. and R. Tibshirani (1993). underline{An Introduction to the Bootstrap}. New York: Chapman & Hall.
data1<-rnorm(1000) jackafterboot(data1,100, type="mean")