To increase the power of a hypothesis test.

To provide an estimate of the bias and variance of a statistic by systematically leaving out one observation at a time.

To replace the need for parametric assumptions when the sample size is extremely large.

To compute the exact p-value for non-normal data.