General Nonparametric Jackknife

DESCRIPTION:

Performs delete-one jackknifing of observations from specified data. Calculates jackknife statistics for parameters of interest and produces an object of class jackknife.

USAGE:

jackknife(data, statistic, args.stat=NULL, seed=0, group.size=1,  
          assign.frame1=F) 

REQUIRED ARGUMENTS:

data
data to be jackknifed. May be a vector, matrix, or data frame.

OPTIONAL ARGUMENTS:

statistic
statistic to be jackknifed. It may be a function which accepts data as the first argument and returns a vector or matrix, or a call referring to the data which evaluates to a vector or matrix. If it is a function, other arguments may be passed to the function through args.stat. If it is an expression and the data argument is the name of an object, then the object should also be referred to by name in the expression. If the data is constructed within the call to jackknife, the data should be referred to as data in the expression. (See help for bootstrap for examples of usage.)
args.stat
list of other arguments, if any, passed to statistic when calculating the statistic on the resamples.
seed
seed for randomization done by statistic, and for random assignment of observations to groups if group.size is not equal to one. May be a legal random number seed or an integer between 0 and 1000 which will be passed to set.seed.
group.size
integer giving the number of observations to remove in each resample. If group.size=1, the standard delete-1 jackknife is performed. Otherwise, the observations are divided into floor(n/group.size) groups of equal size and these groups are jackknifed. Although this is similar to delete-d jackknifing, all possible subsets of the specified size are not used, and the jackknife statistics treat the replicates as a standard jackknife sample of size floor(n/group.size). This is provided primarily to allow grouped jackknifing when calculating acceleration for BCa confidence intervals.
assign.frame1
logical flag indicating whether the resampled data should be assigned to frame 1 before evaluating the statistic. This may be necessary if the statistic is reevaluating the call of a model object. If all jackknife estimates are identical, try setting assign.frame1=T. Note that this will slow down the algorithm.

VALUE:

an object of class jackknife which inherits from resamp. This has components call, observed, replicates, estimate, B, n, dim.obs, and seed.start. The data frame estimate has three columns containing the bootstrap estimates of Bias, Mean, and SE.

SIDE EFFECTS:

If assign.frame1=T, the user must be sure that this assignment does not overwrite some quantity of interest stored in frame 1.

DETAILS:

Performs nonparametric jackknifing of observations for a wide scope of statistics and expressions.

REFERENCES:

Efron, B. and Tibshirani, R. J. (1993). An Introduction to the Bootstrap. San Francisco: Chapman & Hall.

Shao, J. and Tu, D. (1995). The Jackknife and Bootstrap. New York: Springer-Verlag.

SEE ALSO:

, , , , , .

EXAMPLES:

jackknife(stack.loss,var) 
# See help(bootstrap) for more examples of syntax.