End User Learning Behavior in Data Analysis and Data Modeling Tools

The research examined naive user analysts' learning of data analysis skills; namely. (1) the difficulty of learning data analysis, (2) the differential learning rates among development tools, and (3) the dimensions of the tools contributing to the learning differences. A total of fifty-six students participated in two experiments. The experiments involved repeaied trials of practice and feedback in drawing application-based data models. On average, the participants were experienced end users of computer systems in organizations. The two tools examined in the experiments were the logical data structure model (LDS), which is based on the entity-relationship concept, and the relational data model (RDM). The correctness of the models improved over the trials in both LDS and RDM groups with LDS users performing better than RDM users, particularly in terms of representing relationships. LDS users were found to be more top-down motivated in their method of analysis than RDM users. The study suggests that among end users, the LDS formalism is more easily learned than the RDM formalism. The results also imply that end-user training should stress conceptual top*wn analysis, not bottom-up output directed analysis. INTRODUCTION languages, users (i.e., naive analysts) primarily need to describe the data and relationships for The development of high quality systems by end an application (Harel and McLean, 1985). Little users requires effective training and tools that empirical research exists on how people learn to support and improve the users' problem-solving conceptualize, analyze, and design data. approaches (Davis, 1982; Alavi, 1985). Yet to provide effective training and tools, we must unThe objective of this research is to investigate derstand the skills needed to perform analysis, how available development tools support naive design, and programming tasks. Much of the analysts in learning data analysis. The key existing behavioral work on development has inquestions of the research are: (1) How difficult vestigated programming (Pennington, 1982; is data analysis for naive analysts7 Does the rate Sheil, 1981). Few studies have addressed analyat which naive analysts learn vary for different sis or design (Jeffries, et al., 1980; Vitalari and tools? And if so, which dimensions of a tool Dickson, 1983). Within analysis and design, contribute to learning differences? The next data instead of procedure specification is of parsection presents the theoretical concepts underticular importance because with non-procedural lying the research and the propositions studied.