Defense Acquisition Research Journal Issue 95

Technology Trust

https://www.dau.edu

TABLE 16. PRINCIPAL COMPONENT ANALYSIS Model Inputs: VAR23:VAR33 PU1, PU2, PU3, PU4, PEOU1, PEOU2, PEOU3, PEOU4, IU1, IU2, IU3 * indicates negative values Cum Proportions: 55.05% 75.51% 85.26% 90.59% 94.76% 96.74% 97.92% 98.97% 99.57% 99.87% 100.00% Eigenvectors:

0.3537 *0.2475 *0.1379 *0.0953 0.1383 *0.3637 *0.0508 *0.4055 0.5314 *0.0935 *0.4195

0.3592 *0.2186 0.0260 0.1763 0.1098 *0.1431 *0.7667 0.1402 *0.1323 0.2344 0.2811

Eigenvalues (Arranged and Ranked):

6.0552 2.2509 1.0725 0.5861 0.4586 0.2184 0.1292 0.1157

0.0666 0.0320 0.0148

A traditional ordinary least squares multivariate regression also does not make too much sense in that no one-to-one correspondence is detected among the data rows. That is, different active-duty military from the same unit participated in the three experimental stages. This means that the responses of one soldier will not correspond to the same perception of another soldier testing another system during a different stage. This explains partly the low predictability of Post-experiment results using Pre-experiment data.

Having additional information on paper, without the ability to perform hands-on experimentation, yields little difference

and only minor benefits.

Additional sophisticated methods were performed, such as bootstrapping the regression, where an empirical bootstrap of the data was nonpara metrically simulated and bootstrapped, then regression models were run. The process was repeated thousands of times. Figures 3, 4, and Table 17 illustrate the results. Only 9% to 12% of the time will a single variable be considered statistically significant, and the goodness-of-fit predictabil ity levels vary widely, from 18% to 95%, depending on the specific issue under study. No consistent and valid predictive power is apparent in the Pre-experiment data. This concurs with the two-variable T -tests and MW tests shown previously where we do see significant and valuable insights exist when hands-on experimentation is performed, which means without these experiments, paper-based cursory system knowledge is insufficient to identify the true value and risks of a system.

30

Defense ARJ, January 2021, Vol. 28 No. 1 : 2-39