On bootstrap confidence limits for possibly skew distributed statistics

Statistics for which confidence limits or tests are calculated by bootstrap techniques frequently have asymmetric distributions. Approaches based only on boot-strapped variance are inadequatein such cases. In a Mte. Carlo study with a markedly skew X2-distributed statistic an approach by Edgeworth expansions using bootstrapped estimates of variance and skewness of the statistic's distribution performed well with respect to size and power and is proposed for variaus applications.