The Moderating Effects of Structure on Volatility and Complexity in Software Enhancement

The cost of enhancing software applications to accommodate new and evolving user requirements is significant. Many enhancement cost-reduction initiatives have focused on increasing software structure in applications. However, while software structure can decrease enhancement effort by localizing data processing, increased effort is also required to comprehend structure. Thus, it is not clear whether high levels of software structure are economically efficient in all situations. In this study, we develop a model of the relationship between software structure and software enhancement costs and errors. We introduce the notion of software structure as a moderator of the relationship between software volatility, total data complexity, and software enhancement outcomes. We posit that it is efficient to more highly structure the more volatile applications, because increased familiarity with the application structure through frequent enhancement enables localization of maintenance effort. For more complex applications, software structure is more beneficial than for less complex applications because it facilitates the comprehension process where it is most needed. Given the downstream enhancement benefits of structure for more volatile and complex applications, we expect that the optimal level of structure is higher for these applications. We empirically evaluate our model using data collected on the business applications of a major mass merchandiser and a large commercial bank. We find that structure moderates the relationship between complexity, volatility, and enhancement outcomes, such that higher levels of structure are more advantageous for the more complex and more volatile applications in terms of reduced enhancement costs and errors. We also find that more structure is designed in for volatile applications and for applications with higher levels of complexity. Finally, we identify application type as a significant factor in predicting which applications are more volatile and more complex at our research sites. That is, applications with induction-based algorithms such as those that support planning, forecasting, and management decision-making activities are more complex and more volatile than applications with rule-based algorithms that support operational and transaction-processing activities. Our results indicate that high investment in software quality practices such as structured design is not economically efficient in all situations. Our findings also suggest the importance of organizational mechanisms in promoting efficient design choices that lead to reduced enhancement costs and errors.