Logistic and Multiple Regression

This study seeks to predict cost growth in major DoD acquisition programs using logistic and multiple regression. Specifically, this research uses logistic regression to determine whether or not cost growth will occur in a program and if so, then uses multiple regression to determine to what extent that cost growth will occur. We compile data from all defense departments using the Selected Acquisition Reports presented between 1990 and 2002. We combine the efforts of previous research and focus our study on cost growth in research and development dollars for the Engineering Manufacturing Development phase of acquisition. For the logistic regression portion of our research, we produce a seven-variable model that accurately predicts 72 percent of our randomly selected validation data. For multiple regression, we produce a six-variable model that accurately predicts the amount of cost growth incurred for 91 percent of those programs that do incur cost growth. We conclude that the two-step regression methodology offers a significant advantage over traditional methods by removing those data points that do not incur cost growth. We further conclude that there is no significant advantage gained by either isolating each cost variance category individually or by combining these categories.