“Face the software costs! The University for Research in Military Procurement in the United States proposes a method for calculating military software costs”
On September 4, 2026, the United States Army Acquisition University released “Research on Cost Estimation Models for Military Software Projects,” aimed at addressing the issues of inaccurate cost estimation and overbudget delays caused by changes in requirements for military software projects. The study integrated historical data from 328 completed software contracts and 43 Department of Defense projects to construct cost estimation relationships based on Equivalent Source Code Lines (ESLOC) and Source Code Lines (SLOC). The findings showed that programming hours were the strongest driver of software costs, while costs for electronic/automated software and fixed-price contract types were relatively lower. Both models had comparable predictive capabilities and could serve as supplementary tools for early cost estimation and cross-verification. The study used ordinary least squares method in JMP Pro 15 to build the models and passed hypothesis tests such as multiple collinearity.
The U.S. War Department’s Software Cost Modeling Project released a study titled “Research on Software Cost Modeling Models for U.S. War Department Projects.” Using 328 completed software contracts and historical data from 43 Defense Department projects, two cost modeling relationships were established based on Equivalent Source Code Lines (ESLOC) and Total Source Code Lines (SLOC). The study concluded that programming hours are the strongest driver of software costs; electronic/automated software and fixed-price contract types have significantly lower costs. Both models performed similarly in predictive capability and can serve as supplementary tools for early cost estimation and cross-verification of military software projects. This research aims to address issues such as inaccurate software cost estimation by the U.S. military and overbudget delays due to changes in requirements. By performing data cleaning and standardization, a model was constructed using ordinary least squares in JMP Pro 15, and it passed hypothesis tests for multiple collinearity.