Computational Analysis of Scoring Models for R and D Project Selection

Several authors have proposed using scoring models for prescriptive analysis of the R and D project selection decision problem. This research indicates that these models do not meet with important practical requirements. For example, many authors recommend a multiplicative index, over an additive index, in order to generate a wide range of project scores. The additive index is shown to have important advantages over the multiplicative index. The most serious shortcoming in the models, however, is the relatively arbitrary fashion in which the models have been constructed and the failure of the model builders to recognize the impact of certain structural considerations on resulting project scores. Comparative analyses relating project rankings produced by scoring models to rankings produced by a profitability index and by a linear programming model demonstrate that the performance of a scoring model is highly sensitive to decisions made during the development of the model. Considerations such as (1) the underlying distributions of project data, (2) time preferences, (3) the number of ranking intervals or categories, and (4) the width of the intervals, all have important implications for final project scores and associated rankings.