Read more
This study tested the hypothesis that defense cost variances reported on the Cost Performance Report are normally distributed. The DOD requires that all defense cost variances which breech a pre-specified threshold be investigated. The present variance investigation model has been criticized because it can prompt frivolous investigations. In theory, statistical models could reduce the number of frivolous investigations, but they are not used because they require too much information about the cost variance, including its distributional form. Often such models assume a normal distribution, but researchers have shown that the models do not work properly if the assumption is fallacious. Two prior studies have investigated the normality of cost variances with mixed results, and neither investigated defense cost variances. Here, fifty series of cost variances from two defense contracts were extracted from Cost Performance Reports and evaluated using four popular tests of normality (Bowman-Shenton, Shapiro-Wilks, Kolmogorov-Smirnov, and Chi-square). The results show that the vast majority of the series of cost variances were not normally distributed. These results were insensitive to the normality test used and to the effects of inflation. The statistical variance investigation models may still be used, but normality should not be assumed.
