軟體產業在少量樣本下之量化專案管理方法

Abstract

近年來，量化專案管理(Quantitative project management)的管理模式逐漸受到軟體產業的重視，其主要目的是以量化的方式來協助管理軟體發展專案，以達到專案既定的品質及流程績效目標。有中外研究指出一些已經成功應用在其他產業（如：製造業）的統計品管方法與最佳化技術，如：修華特管制圖和隨機最佳化模型(Stochastic Optimization Model, SOM) 等，亦可以用來增進軟體組織效益並改善軟體流程績效，進而協助軟體組織達到量化專案管理的目標。然而罕見有文獻探討如何在軟體專案上實際執行這些量化技術的細節，與這些量化技術在軟體產業上的適用性，尤其在實務應用修華特管制圖及SOM方法時，需要相當大量且穩定的流程績效數據，但是在軟體發展流程中是不太可能取得大量且穩定的流程績效數據。當僅利用少量數據來進行軟體專案之量化管理時，這些原應用於製造業之品管及最佳化技術，所提供之量化管理資訊可能會產生相當大之偏誤。因此，本研究之主要目的是提出一套適用於軟體產業少量數據特性的量化專案管理方法，利用三種無母數方法與適用於新速(short-run)製程的Q管制圖，以取代傳統修華特管制圖和SOM方法，使軟體組織能有效地達成量化專案管理的目的，並分別以模擬案例與實務案例來驗證本研究的有效性與實用性。研究結果指出，使用本研究所提出的方法，僅需要少量樣本的數據，即能夠達成與需要大量樣本的其他量化技術同樣的目的（包括：不確定性估計、風險分析、敏感度分析及流程績效監控），且本研究方法不僅可用於小樣本，亦可用於大樣本。本研究可以協助只有少量流程績效數據之軟體組織以量化的方式有效地管理專案，以改善軟體發展流程績效，提升產品品質。

Quantitative Project Management (QPM) has recently garnered considerable attention in the software industry. The purpose of QPM is to quantitatively manage the project to attain the project’s quality and process-performance objectives. Stochastic optimization model (SOM) and Shewhart control charts have also attracted substantial interest in helping software organizations to achieve the requirements of QPM. While many study focused on the benefits of using SOM and control charts, only few studies discussed the implementation details of these quantitative techniques in QPM. Particularly, when applying SOM method and Shewhart control charts in QPM, an adequate amount of process data from a homogeneous source of variation are required. But, a large dataset is typically unavailable in software development processes (SDPs). Consequently, if the sample size is too small to verify that the process performance possessing a certain distribution and to construct valid control limits, the effectiveness of SOM and control charts may be diminished. Therefore, this study considers the situation that only a small amount of process data can be obtained in the SDP and develops a novel approach based on nonparametric techniques and the short-run Q chart to replace SOM and Shewhart control chart. Simulations and case studies are performed in this study to verify the effectiveness of the proposed approach. In the result of simulation study, the proposed method is demonstrated to be effective using small samples in uncertainty estimation, risk analysis, sensitivity analysis and process monitoring. Moreover, the proposed approach can be utilized for both small and large dataset. Finally, with the benefits from this study, it is expected that software practitioners can perform QPM without collecting a large amount of data, and begin monitoring process performance with a limited amount of data to produce high-quality software products within a short period of software development lifecycle. Software organizations can also be benefit from this study using the proposed approach to improve SDPs and product quality.