Nonparametric meta-analysis of independent samples of records

Consider k≥1 independent samples, each with a certain number of record values. The resulting pooled sample of records can be re-ordered to develop exact nonparametric inferential procedures. The distribution theory of such ordered multi-sample record values is developed in this paper. The ordered record sample is then used to present exact distribution-free confidence intervals for quantiles of the population, and also to develop exact prediction intervals for future record values. A real data set is finally used to illustrate the results developed here.