The influence of training content on the quality of user-developed applications

The development and use of computer-based applications by end users has grown tremendously over the past few years, and is projected to continue increasing in years to come. Previous research has indicated the need to provide training to users that develop their own applications, but has not addressed the impact of specific training content on the correctness of user-developed applications. Integration of previous research on training, application comprehension, and software testing provides a Training-Impact Model that can be used to compare the influence of different training contents. The Training-Impact Model developed in this dissertation links training content to specific types of knowledge, knowledge to application verification and validation, and application correctness (resulting from verification and validation) to decision correctness for structured decision-making applications. Three types of training content are evaluated in this research: spreadsheet training, software testing training, and a combination of spreadsheet and software testing training. The research was conducted as a laboratory experiment using 140 subjects and videotape-based training. Each subject was presented with a spreadsheet that contained intentional defects and directed to verify the spreadsheet template so that a decision could be made based on the template's results. The study found that a significant increase in the number of errors detected occurred for both the combined training group and the testing training group. Furthermore, the results indicate that for the combined training group, the level of software testing knowledge increased as a result of training but that the level of spreadsheet knowledge did not increase significantly. The study also found that many subjects corrected several of the template defects, but very few of the subjects corrected all of the defects. The subsequent decisions made, based on results taken from the spreadsheet template, were often incorrect due to the presence of undetected errors. This suggests the need for future research in end-user error detection behavioral patterns and end-user training.