Robust test limits

Willem/Wim Albers, W.C.M. Kallenberg, G.D. Otten

    Research output: Contribution to journalArticleAcademicpeer-review

    Abstract

    Because of inaccuracies of the measurement process inspection of manufactured parts requires test limits which are more strict than the given specification limits. Test limits derived under the assumption of normality for product characteristics turn out to violate the prescribed bound on the consumer loss drastically when normality fails, as often happens in practice. The ratio of the true density and the normal density at the specification limit is the key-point. A new test limit is presented using techniques of nonparametric density estimation. The new test limit has the desired robustness property: a small loss under normality and a large gain in case of nonnormality in comparison to the normal test limit. Monte Carlo results show that asymptotic theory is in agreement with moderate sample behaviour.
    Original languageUndefined
    Pages (from-to)416-439
    Number of pages24
    JournalMathematical methods of statistics
    Volume6
    Issue number4
    Publication statusPublished - 1997

    Keywords

    • EWI-13157
    • bandwith
    • MSC-62G07
    • MSC-62G35
    • MSC-62N10
    • Specification limit
    • Monte Carlo experiments
    • Density estimation
    • Robust test limit
    • Second order unbiasedness
    • Normal test limit
    • Deconvolution

    Cite this