https://www.dau.edu
A few things are apparent from Tables 4 and 5. First, the lognormal and Weibull distributions are both relatively good at fitting the cost growths for all the WBS elements except for the Level 2 WBS of Data. (Note: For sensitivity analyses, we redid the analysis with none of the outliers removed via the ATF process, and our conclusions were identical.) For that element, either the triangular or beta distribution is the preferred distribution of choice. Second, although we listed the lognormal as the mode choice for Table 4, the Weibull is equally preferred. As a visual example of how closely these two distributions fit the cost growths, we display Figures 3 and 4. (Note: The other WBS elements were comparable.) The curve fits are very comparable. Figures 5–8 display the quantile-quantile plots of the lognormal and Weibull fits for those WBS elements with the most datapoints. These also show the lognormal providing a better fit than the Weibull, although they all illustrate deviation in the right tail. Lastly, and except for the Level 2 WBS of Data, we do not consider or recommend using either the Rayleigh, beta, or triangular distributions for the remaining WBS elements.
FIGURE 3. LOGNORMAL FIT TO WBS ST&E COST GROWTH AIC = 139.08
Histogram of Data Rayleigh Distribution
0.8
0.5
0.4
0.2
0.0
0 1
2
3 4 5 6 7
Note. AIC = Akaike Information Criterion; ST&E = System Test and Evaluation; WBS = Work Breakdown Structure.
43
Defense ARJ , Spring 2025, Vol. 32 No. 1
Made with FlippingBook - Online Brochure Maker