Decomposed Versus Holistic Estimates of Effort Required for Software Writing Tasks
通过两个实验，研究编程学生对自己完成软件项目时间的估计，发现无论分解还是整体估计都过于狭窄，但诱导合理上下限能改善，提示最佳分解水平因任务、判断者和方法而异。
We examine decision analysis’ central “decomposition principle” in the context of work-time estimates of software writers. Two experiments examined the abilities of advanced programming students to estimate how long they would take to complete specific software projects. They estimated their own work times both for entire projects and for their constituent subtasks. Estimates showed varying degrees of overoptimism and overpessimism but all were much too tight, with almost half of actual outcomes falling in the 1% tails of estimated distributions. This overtightness was unaffected by task decomposition, question wording, question order, or training in estimation. It was, however, significantly reduced by a procedure aimed at inducing generous upper and lower plausible limits. An underlying model of incomplete search is used to connect these findings to existing themes in cognition and judgment research, as well as to practical application. The findings suggest that the best level of decomposition at which to elicit work-time estimates may depend on task, judge, and elicitation method.