ABSTRACT
Wald's (1950) theory of statistical decisions is based on the assumption of a particular arid h l ly specified loss. In practice, I-rowcver, it is often difficult to specify the loss fimctiori exactly. Besides, if ;m estirmtor T2 dominittes another estimator Tl with respect to a. loss l,, there is no gnarantce in gcrreral that 7% will still be better than 7; under a different loss l j2. Hwarrg (1985) proposed the critcriori of uriiversu.1 do~~liriatioir (U-dorninatioil) as a rrleans of studying ).low robust the superiority of one estimator over another is with respcct to the loss fimction. 't'his criterion was fiirther developed by Brown arid Hwar~g (1989), whereas, earlier, similar concep1,s were considered in a. less formal inanner by Cohen m d Sackrowitz ( 1 9'iO), Brown (1971) itlltl Rukhin (1987).
