Abstract
Higher order asymptotic aspects of the jackknife-t method are studied. Emphasis is placed upon constructing higher order accurate confidence intervals for parameters which are smooth functions of multivariate means. Explicit formulae for two-term Edgeworth corrections for both cumulative distribution functions and percentiles are provided. These formulae can be automatically evaluated given specific distributions. Jackknife-t confidence intervals having coverage error of O(n-3/2), n being the sample size, are obtained by twice inverting certain Edgeworth expansion. A numerical example is given in the case of estimating the coefficient of variation from normal populations. Bootstrap intervals are also discussed.
