| Comprehensive evaluation is a procedure of evaluating and sorting subjects according to multiple scales. For a specific synthetical problem, a variety of evaluation methods can be selected, of which the basic idea is to transform multiple scales to a single comprehensive one reflecting the overall situation in order to realize sorting of subjects evaluated. These methods vary in the ways that problems are solved, data are standardized and evaluation information is composed, therefore evaluation results of different methods are not completely consistent. Unfortunately, there are no definite criteria for selecting evaluation methods and appraising evaluation results, indicating that problems exist theoretically and practically and need to be solved in urgency. The purpose of this study is to solve the problem of method selection, so that practitioners are able to choose an appropriate method in specific situation; furthermore, the problems of grading and evaluation computation are also to be solved.The contents of this study are as follows:Firstly, compare comprehensive evaluation methods. All the methods are categorized into nonparametric evaluation method and multivariate statistical method. Each of them is established with special background and significance and has its own application scope. In other words, there is no absolute criterion for the judgment of different evaluation methods. Methodological comparison will be conducted in this study for similarities and differences, advantages and disadvantages and application scope. Besides, some improvements will be conducted if possible.Secondly, explore new ways to solve the problem of grading subjects. In the practice of comprehensive evaluation, sometimes the evaluators are not only concerned about how subjects are sorted but also expect to know how they are graded; however, there is no answer to this question till now. Cluster analysis of multivariate statistical analysis, especially ordered sample cluster, will be introduced in this study so that hopefully subjects can be graded more scientifically and rationally.Lastly, realize the computational problem with software. Since the computation of evaluation methods are complicated and no relative program has been provided by popular statistical software, the research and application of these methods are restricted to some extent. SAS is authoritative software in data management and statistical analysis in the world with characteristics of great flexibility and function. Users are able to deal with complex problems by writing programs with powerful SAS language, such as SAS macro, SAS array, SAS function and SQL. This study focuses on writing automatic programs with great intelligence and generalities based on SAS software so that when they are used in other similar situations, only a small change for the program is needed to fulfill comprehensive evaluation rapidly and precisely.The main conclusions of this study are as follows:First, the evaluation results of different methods are often consistent when dealing with the same data, and Kendall coordination coefficient is used to measure this consistency. Spearman's rank correlation coefficient contribute to the final selection from a variety of results, however this final result still depends on professional interpretation. The difference between the results of different evaluation methods is caused by mythological differences, especially when the population of subject is large, or the variation of the data is small, or the number of evaluation scales is massive, or they are seriously multi-correlated. Thus, nonparametric evaluation methods are suitable for simple situations. Both principal component analysis and factor analysis are multivariate statistical methods simplifying evaluation scales and extracting data information based on the idea of dimension-reduction, which can solve the problem of information overlapping caused by interrelated scales in the comprehensive evaluation system, therefore they can be applied to more complex situations. Nevertheless, they belong to exploratory methods with no fixed evaluation strategies, and the final evaluation result is not always scientific and reasonable, because the fitness of a model depends considerably on the characteristics of data. Principal component analysis can only extract the 1st principal in comprehensive evaluation, while factor analysis is able to extract more common factors. Furthermore, factor analysis can try different ways of modeling and factor rotation for more satisfactory evaluation results. As a ralatively new nonparametric method in multivariate statistics, the idea of statistical depth function is to transforms multiple variables to a single one which is another solution to comprehensive evaluation.Second, cluster analysis is capable of solving the problem of grading. Disordered sample cluster can be used for exploratory study which preliminarily discusses the classification of subjects, while ordered sample cluster can be applied to continued grading based on the sorting result of comprehensive evaluation. Ordered sample cluster reflects the actual difference among subjects and presents the optimal number of grading, which possesses unique practical value and broad application prospect.Third, this study writes programs for routine evaluation methods based on SAS software,including RSR method, Topsis method, entropy method, efficacy coefficient method, principal component analysis, factor analysis, statistical depth function and cluster analysis. The result of practical examples shows that the SAS programs are very accurate, reliable and easy to understand and use, which possesses fairly promotional value.In summary, the systematic comparison between several typical methods in this study provides reference for practical application. The idea of cluster analysis can be used to perform evaluation grading. Furthermore, the corresponding SAS programs written in this study can solve the problem of evaluation computation. Therefore, this study enriches and develops the theoretical system of comprehensive evaluation. |
PDF Full Text Request