Konzistentnost izbora metoda viseatributivne analize

In this paper we analyze the consistency of the results of Multi-Attribute Decision Making (MADM) methods. Two conditions of consistent choice are defined, due to change in units used for measuring different attributes, and due to framing of attributes (whether they are defined as benefit or as cost...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Ekonomski anali 2002, Vol.43 (155), p.59-82
1. Verfasser: Pavlicic, Dubravka
Format: Artikel
Sprache:eng
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:In this paper we analyze the consistency of the results of Multi-Attribute Decision Making (MADM) methods. Two conditions of consistent choice are defined, due to change in units used for measuring different attributes, and due to framing of attributes (whether they are defined as benefit or as cost attributes, when both forms are possible and are normatively equivalent). It is shown by several examples that two popular MADM methods (SAW and TOPSIS) violate both conditions. We conclude that the main causes for inconsistent choices are: 1) normalization of data, conducted in the first step of the observed methods, and 2) the usage of Likert scales for measuring qualitative attributes. Finally, these conclusions are generalized to all MADM methods which are based on normalized ratings and which treat Likert scales as interval scales. . U radu se razmatra problem objektivnog ocenjivanja metoda viseatributivne analize (MADM). Definisana su dva uslova konzistentnosti izbora ovih metoda, kojima se ispituje osetljivost njihovih rezultata s obzirom na promenu jedinice mere u kojima su iskazane vrednosti pojedinih atributa, i s obzirom na formulaciju atributa, u slucaju kada se isti atribut moze prikazati na dva normativno ekvivalentna nacina. Na ilustrativnim primerima je pokazano da dve popularne metode (Metoda ponderisanog zbira i Metoda slicnosti sa idealnim resenjem) ne zadovoljavaju oba uslova. Kao osnovni uzrok nekonzistentnosti njihovih rezultata navedeni su: normalizacija empirijskih podataka i primena Likertovih skala za merenje kvalitativnih atributa. Dobijeni rezultati su zato uopsteni na sve metode MADM koje koriste normalizovane vrednosti atributa i u kojima se Likertove skale tretiraju kao intervalne skale.
ISSN:0013-3264
1820-7375
DOI:10.2298/EKA0205059P