Cost Drivers - A Survey and Assessment

Abstract

Cost estimating depends on a good understanding of the cost drivers – for cost model developing and for cost predicting. Cost estimating begins with a determination of “what drives cost for my system,” i.e., the cost drivers. Not to be confused with cost elements, cost drivers are the technical and program parameters of cost according to determinable cost estimating relationships (CERs). A distinct sub-discipline of cost estimating involves the collection and validation of historic data and the subsequent derivation of statistically-based CERs. Since cost estimating (or cost engineering) is widely accepted as a subset of systems engineering, it seems natural that disciplined cost estimators depend upon engineering-based, and mathematically-correct, CERs. While CERs may not enjoy as much stature as do “laws of nature,” they are mathematically precise, roundly debated, and systematically applied. This paper reports on a recent survey performed by the author on his 20 years of experience in collecting spaceprogram cost data, deriving sets of CERs from such data, applying those CERs to predicting the cost of future development and production, and then validating those CERs through comparing those cost predictions with actual costs. The traditional approach to model development (defined as an aggregate of CERs that populate the entire work breakdown structure, or WBS) is to first establish a candidate list of cost drivers and their logical relationship to cost. This is called the heuristic approach to model building as opposed to the purely statistical approach which is to dump a mass of data into a number-crunching program which determines the most robust relationships. In a sense, the heuristic approach assures “engineering sense” in the final relationships. Then, historic cost data are collected from developers and producers and linked to one or more cost drivers. For this step, some un-linkable cost drivers are dropped in favor of others. A traditional cost driver is size. Early hardware models were based on weight (pounds or kilograms); early software models were based on lines of code. Subsequent model drivers have shunned such coarse choices and have sought costs associated with data rate, detector sensitivity, bandwidth, and other characteristics. My survey of established space-based cost estimating models provides insight into the early optimistic phases of cost model development where obvious and noble cost drivers were established. The paper concludes with a robust set of recommended space-system cost drivers.