Font Size: a A A

Identification of desirable design features for statistical expert systems based on the opinion of statistician

Posted on:1990-05-04Degree:D.B.AType:Dissertation
University:Mississippi State UniversityCandidate:Liu, Lai ChingFull Text:PDF
GTID:1476390017453782Subject:Computer Science
Abstract/Summary:
During the last decade, some statisticians have suggested that one possible way to approach the problem of misusing statistical techniques by naive users was to develop statistical expert systems. Although it has been recommended in the literature that statisticians should participate in the process of developing statistical expert systems, formal studies involving their input pertaining to design features of statistical expert systems have not been done. The objective of this study was to identify the design features that statisticians desired for statistical expert systems.;One thousand American Statistical Association members with an expressed interest in statistical computing were randomly selected for this study. A total of 264 usable responses were received. The data were collected using a mail questionnaire constructed to yield information relating to desirable design features for statistical expert systems. Most of the design feature variables were based on the work of Hand.;The results of this study indicated that a large majority of these statisticians had at least heard about statistical expert systems, and very few had used or been involved in designing them. In addition, a large majority of these statisticians indicated that statistical expert systems would be best used for design of studies and selection of statistical techniques.;Factor analysis was used to identify the underlying relationships of the design feature variables. Eight factors resulted. Knowledge base concerns such as modification, representation, and conflicts in the knowledge constituted the most important factor. The other factors were related to techniques selection, assumptions clarification, analysis, branchability, result presentation, explanation, and friendliness.;Multivariate analysis of variance was used to test for differences in the perceived importance of these eight factors among statisticians. The results indicated that these eight factors did not differ among statisticians by major statistical activity; by affiliation; by previous knowledge of, or exposure to, statistical expert systems; or by the preferred stage of the statistical analysis at which statistical expert systems are used. The findings seemed to imply the existence of a model of generic design features for statistical expert systems.
Keywords/Search Tags:Statistical expert systems, Statisticians
Related items