ART - Estimating Final Cost of DOD Acquisition Contract 2011

The document discusses a study that aims to estimate the final cost of Department of Defense (DoD) acquisition contracts. The study utilizes an online database called Purview, managed by OUSD (AT&L), for reporting. The reliability of the contractor's reporting is controlled by their compliance with Earned Value Management System (EVMS) criteria. The study focuses on accurate Estimate at Completion (EAC) values, which closely estimate the final cost of the contract. The Cost at Completion (CAC) is used as the regression response variable to be modeled. To determine the modeled final cost of the contract, the study uses paired t-tests to determine which Actual Cost of Work Performed (ACWP) percentages are statistically equivalent to 100% complete. The study concludes that only the 95% complete range is statistically equivalent to the 99% complete range. Therefore, contracts with a final report indicating a program completion point of 92.5% or higher are considered complete, and the ACWP of the final report is considered the CAC for modeling purposes. The study divides the database into different percent complete ranges (25%, 35%, 50%, 65%, and 75%) and builds predictive models for each range. The models for these percent complete ranges are presented in Table 1. The range of 20% complete to 80% complete is chosen, as data before 20% completion is not available or unreliable, and many contractors discontinue reporting after 80% completion. The study randomly selects 80% of the contracts as a working set and uses the remaining contracts as a validation set for model building. The study normalizes all values reported in DAES to the constant year 2004 using inflation indices. The distribution of the CAC response variable is heavily skewed, so the natural logarithm of the CAC is taken to minimize this skewness. The study uses variables from contractor Cost Performance Reports as predictors for estimating the CAC. It considers interactions between variables and higher-order terms of continuous data. Categorical data, such as Acquisition Category (ACAT) or Contract Type, is entered as binary dummy variables. The methodology chosen for the research is ordinary least squares regression, which minimizes the sum of squared differences between actual and predicted costs. The JMP software is used for building and analyzing regression models. The final regression models for different percent complete ranges (25%, 35%, 50%, 65%, and 75%) are determined based on adjusted R2, data points to variables ratio, p-values, and stepwise regression.