Statistički model efikasnosti zasnovan na Ivanovićevom odstojanju

Abstract

U uvodnom poglavlju se opisuju predmet i cilj istraživanja, navode se polazne hipoteze i metode istraživanja, daje sadržaj i opis disertacije uz navođenje ključnih aspekata na koje će se disertacija usmeriti. Drugo poglavlje je posvećeno konceptu efikasnosti i načinima za merenje efikasnosti. Princip efikasnosti predstavlja jedan od bitnih postulata savremenog poslovnog odlučivanja (Savić, 2011). Efikasnost se definiše kao sposobnost da se minimiziraju ulaganja u ostvarivanju ciljeva organizacionih jedinica uz maksimizaciju rezultata (Amado et al., 2011). Kod organizacija koje koriste jedan ulaz za kreiranje jednog izlaza, efikasnost se definiše kao odnos izlaza prema ulazu. Problem se javlja kod određivanja efikasnosti jedinica koje imaju više raznorodnih ulaza i koriste ih za stvaranje više raznorodnih izlaza (Thanassoulis et al., 2012). Osnovni cilj istraživanja doktorske disertacije je da kroz razvoj novog statističkog modela efikasnosti prevaziđe problem raznorodnih varijabli. Stoga, potrebno je definisati pokazatelj efikasnosti koji će sintetizovati sve indikatore u jednu vrednost. Problemi sa kojima se u tom procesu susrećemo su izražavanje varijabli (ulaza i izlaza) u opsezima koji su međusobno uporedivi i određivanje pondera koji se dodaju pojedinim ulazima i izlazima. Kao jedna od metoda za merenje efikasnosti organizacionih jedinica u disertaciji se navodi analiza obavijanja podataka (Data Envelopment Analysis - DEA). DEA metodu su razvili Charnes, Cooper & Rhodes (1978), da bi merili efikasnost poslovanja organizacionih jedinica i to pre svega onih koje ne stvaraju profit. Tvorci analize obavijanja podataka su predložili neparametarski pristup za izračunavanje efikasnosti, tako što su višestruke ulaze sveli na jedan "virtuelni" ulaz i višestruke izlaze sveli na jedan "virtuelni" izlaz koristeći težinske koeficijente. Problem dodeljivanja težina su rešili tako što su svakoj jedinici dopustili da odredi sopstvene težine sa ciljem da joj se maksimizira efikasnost, uz ograničenje da te težine moraju biti nenegativne vrednosti i da količnik virtuelnog izlaza i virtuelnog ulaza svake jedinice ne može biti veći od 1 (Martić, 1999). Pоrеd DEA mеtоdе zа mеrеnjе еfikаsnоsti, u disertaciji je pоsеbnа pаžnjа usmеrеnа na аnаlizu stоhаstičkih grаnicа (SFA – Stochastic Frontier Analysis). Оvо је аltеrnаtivni pristup оdrеđivаnjа grаnicе еfikаsnоsti kоrišćеnjеm еkоnоmеtriјskih modela. U trećem poglavlju pažnja se posvećuje multivarijacionoj statističkoj analizi. Sam termin multivarijacione analize se koristi da predstavi multivarijacioni aspekt analize podataka, u smislu da su mnogobrojne opservacije izmerene na velikom broju promenljivih. Multivarijaciona statistička analiza obezbeđuje mogućnost analize kompleksnih nizova podataka, tamo gde ima mnogo nezavisnih i zavisnih promenljivih koje su međusobno korelisane na različitim nivoima povezivanja. U okviru trećeg poglavlja, disertacija u značajnoj meri ističe dve ključne statističke tehnike: faktorsku i analizu glavnih komponenata, kao i klaster analizu. Faktorska analiza i analiza glavnih komponenata su statističke tehnike koje se koriste za identifikaciju relativno malog broja faktora koji se mogu koristiti za predstavljanje odnosa između grupa mnogobrojnih, međusobno povezanih, promenljivih.

In section Introduction we define area of research and our main goals, we point out crucial hypothesis and method of research, we provide contents and description of dissertation with emphasising crucial aspects of our work. Second section is committed to the concept of efficiency and methods for evaluating and measuring efficiency. Principle of efficiency is one of the most important parts of contemporary business decision making (Savić, 2011). Efficiency should be defined as a capability to minimize the input means in order to achieve maximum results (Amado et al., 2011). With organizations which use one input to define one output, efficiency is measured as an output divided by input. However, real problem emerges when we have to measure efficiency of organizational systems which use many different types on inputs to create different outputs (Thanassoulis et al., 2012). Main idea of this research was to create novel statistical model of efficiency which can overcome problems with different types of input/output variables. Thus, it is necessary to integrate all indicators into one value. Issue at stake is how to overcome fact that variables are measured in different units, and how to provide appropriate weighting factors. As one of the frequently used methods, Data Envelopment Analysis – DEA will be explained. DEA method was developed by Charnes, Cooper & Rhodes (1978), in order to measure efficiency of organizational units, in particular a non-profit one. Creator of this method suggested a nonparametric approach for evaluating of efficiency. The approach was based on the idea that multiple inputs and outputs are integrated into one virtual input and output. Weighting issue was resolved by allowing each Decision Making Unit (DMU) to determine its own weights, all of this with idea for each of them to maximize its efficiency. Only constrain was that weights had to be a positive number, and that quotient of virtual outputs and virtual inputs cannot be larger than 1 (Martić, 1999). Besides DEA method for measuring efficiency, in dissertation Stochastic Frontier Analysis (SFA) was also elaborated. This is alternative approach towards measuring efficiency with the extended usage of econometrics models. In third chapter, attention is being shifted to the concept of multivariate data analysis. The term has to explain multivariate aspect of data analysis, in a way that many observations are collected on large number of variables. Multivariate data analysis provides us with the opportunity to analyze complex data sets, with many mutually dependent and independent variables occur. In third chapter dissertation is mostly focused on two crucial statistical methods: principal component analysis (PCA) & factor analysis, and cluster analysis. Factor analysis and PCA are most commonly used to identify relatively small number of factors which represent complex interrelations between all variables. These methods are very usefully in identifying hidden dimensions of observed phenomenon. Main difference between these two methods is the way of data examination. Factor analysis is mostly concern about covariance, while PCA is examining variances. Both of them have similar goals and procedures, thus factor analysis can be observed as a special case of PCA (Bulajić, 2002). Also, significant attention is committed to the cluster analysis as a multivariate data analysis method. It is frequently used for grouping objects, so the similar objects IX create group which differs from the other groups.