THREE ESSAYS ON THE ECONOMICS OF SOFTWARE MAINTENANCE

Abstract

This thesis comprises three essays that contribute to software maintenance economics literature. They aim to quantify, model, and control the huge cost of software maintenance. This thesis distinguishes between two types of software maintenance cost: internal and external. Internal cost refers to the cost of resources allocated to software maintenance by the IS department while external cost refers to the opportunity cost associated with long software maintenance lead time experienced by the IS users. The first two essays address the internal cost issues while the third addresses the external cost issue. The three essays deepen our understanding of software maintenance economics in the follow ways: The first essay conceptualizes the software maintenance task as a knowledge-intensive production process. It recognizes the critical role that a programmer's human capital may play in determining his effort expended on the task, and provides first empirical study that investigates the impact of both general and firm-specific human capital on maintenance productivity. The general human capital includes both programming knowledge and functional domain knowledge while firm-specific knowledge refers to application-specific knowledge, which is measured by the number of times a programmer has modified the application. Using experimentation and the estimation method of maximum likelihood, it finds that (1) both types of human capital can significantly increase maintenance productivity, (2) programming knowledge has a stronger impact than domain knowledge, and (3) a programmer who has modified a program once took 33% less time than one who is new to the program. The second essay broadens the scope beyond a maintenance task and considers a sequence of maintenance tasks to be performed on an application over its life-cycle. Empirical evidence suggests that maintenance effort of a software increases over time because the software tends to deteriorate, with frequent modifications. This leads to dramatic increase in the total maintenance effort of the software as it ages. Besides controlling maintenance productivity, another means to control this total effort is to replace the aged software with a freshly written one. Software replacement, however, involves huge initial investment in rewriting effort that must be weighed against the potential saving derivable from the fresh software. The problem is to determine the optimal timing to replace a software so that the total saving outweighs the rewriting effort by the maximum value. The second essay develops a new normative model of software replacement that address this problem. The model shows, among other things, that software replacement must be accompanied by stringent control of maintenance quality for it to be effective. The third essay further broadens the scope of analysis to include the IS department-users interface. This interface highlights the servicing nature of the software maintenance process. It also highlights the gap between the service time and level time of a maintenance request which represents external user opportunity cost. Although results from classical literature suggest that the delay can be very extensive, then is no quantitative evidence on the extent of this delay in software maintenance. Using field data of 1424 maintenance requests, the third essay represents the first study to quantify the magnitude of software maintenance lead time and delay. It also attempts to derive a clear understanding of the delay phenomenon by estimating the distributions of service times and lead times in software maintenance. It is found that the average delay of a request is 114.91 days, which is more than ten times the actual service time required. In combination, these three essays contribute to knowledge in software maintenance economics by employing a blend of research methods, including experimental study, normative modeling, econometric estimation, and field study to address both the internal and external software maintenance cost problems. In doing so, this thesis provides a more comprehensive view and deepen our understanding of the software maintenance cost.