Low-storage quantile estimation (Q1965993)

Many different approaches have been taken to the problem of low-storage quantile estimation. One approach is based on stochastic approximation. Others, such as minimax and recursive median, are based on trees. The authors compare their computational properties such as storage demands, the ease and speed of computation, as well as statistical properties such as asymptotic behavior and robustness. An approach to low-storage quantile estimation based on the sample histogram, which has good computational and statistical properties, is then described. A simulation study is presented where the efficiency of various low-storage quantile estimators are compared. It is shown that quantile estimates based on the sample histogram are similar to the sample quantile and have greater accuracy than the estimates produced by other more complex and computationally intensive methods.