Jackknife estimate of the variance of a parameter estimate.
c=covjack(x,T) [c,y]=covjack(...)
a matrix of doubles
a function which computes the empirical estimate from x
a 1-by-1 matrix of doubles, the estimate of the covariance matrix
a m-by-n matrix of doubles, the parameter estimates of the resamples, where m is the size of the parameter estimate and n is the length of x.
The function computes T(x) with one observation removed at a time and uses the result to compute an estimate of the variance of T(x) assuming that x is a representative sample from the underlying distribution of x.
If T is multidimensional then the covariance matrix is estimated.
The function T must have the following header:
p=T(x)
x
is the sample or the resample
and p
is a m-by-1 matrix of doubles.
In the case where the parameter estimate has a more general
shape (e.g. 2-by-2), the shape of p
is reshaped
into a column vector with m
components.
See "T and extra arguments" for details on how to pass extra-arguments to T.
Note that the jackknife method does not work for some functions T that are not smooth enough, the median being one example.