Developments in data analysis systems for the social sciences

A discussion of the rapid growth of & changes in computer programming systems for soc sci data in the past decade. The field is characterized by a great variety of systems & programs, each using a diff language & cliff methods that are not transferable from one system to another. This entails great losses because changing from one system to another means learning all over again. Nevertheless, some convergence is beginning to take place. This partial convergence is seen in the development of a specialized language for standardized statistical or numerical data analysis. 3 systems are examined: BEAST, developed by G. Sadowski et al at the Brookings Instit; Data-Text, developed by A. S. Couch, the author & others at Harvard; & SPSS, developed by Norman Nye et al at Stanford. The similarity between the 3 systems in input-output & labeling flexibility, in the use of transformational language, & in the methods of statistical computations is shown. Each system has developed a method for dealing with missing observations, which is said to be the most troublesome problem in soc sci data analysis. The systems differ in the syntax of their languages, in the use of cliff computers, in the extensiveness of file saving, manipulation & updating, & in the degree to which the procedural language is complete. It is concluded that problems to be worked on in the future are job control language, interfacing procedures for user-supply routines, language standardization, exportability from one system to another, & cooperation between soc sci'ts working in computer systems. A. Peskin.