Date of Award
Master of Science in Cost Analysis
Daniel V. Ferens, PhD
This study attempted to analyze the effect of calibration on the performance of the SASET computer software cost estimating model. Data used for input into the model were drawn from the most current USAF SMC Software Database (SWDB). Once all the records to be used for analysis were identified, the DBMS-Calibration tool (which is part of SASET) was used to perform regression analysis on the relationship between program size (measured in SLOC) and the effort required to develop the program (measured in man-months). Productivity information reported from this tool was then input into equations used to calculate the Productivity Calibration Constants (PCC) and Software Class Multipliers. A comparison was then made between the model's accuracy before calibration and its accuracy after calibration. This was done using records which were not used in calibration (referred to as validation points). Several measures such as mean, variance, mean magnitude of relative error (MMRE), and the percentage method were used to describe accuracy. The majority of the results agreed with previous studies that calibration does improve a model's prediction performance. However, emphasis is placed on the fact that calibration is most useful when the group of calibration data points are homogenous.
DTIC Accession Number
Vegas, Carl D., "Calibration of the Software Architecture Sizing and Estimation Tool (SASET)" (1995). Theses and Dissertations. 6514.