Probability Methods for Cost Uncertainty Analysis, 2nd Edition

367 Building Statistical Cost Estimating Models

The coefcients for all cost drivers should also be analyzed. If the result-

ing relationship between cost and a cost driver is different than would be

expected from engineering judgment, then this model is most likely not one

that should be used to predict future cost. Correlations between cost drivers

should also be calculated. Often two or more cost drivers could provide very

similar relationships with cost. Including correlated drivers together in a cost

model could yield inconsistent results in a regression model. In statistics,

this is known as multicollinearity—and