Impact of programming and application-specific knowledge on maintenance effort:A hazard rate model

We empirically test the relative impact of general programming knowledge and application-specific knowledge on maintenance productivity. One hundred undergraduates participated in a quasi-experiment that required them to perform two maintenance tasks in sequence on an inventory control application. Each maintenance task involved a modification to the original application and the hours needed to complete each maintenance task are used to measure productivity. Since subjects may submit modifications that do not meet all the user requirements, the person-hours spent can be less than the actual hours required if modified applications were to meet the user requirements completely. That is, the observed time effort censored the actual required time effort. To overcome the challenge of this censored time problem, we use a proportional hazard model to model the effect of human capital on productivity. The method of maximum likelihood estimation was used to estimate the model parameters. Our study enables us to draw several implications for formulating hiring policies relating to software maintenance.