09.12.2012 Views

I__. - International Military Testing Association

I__. - International Military Testing Association

I__. - International Military Testing Association

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

An AT0 is organized by duties and tasks within duties. All<br />

tasks on the TI survey are listed under the appropriate duty.<br />

The TI values are used to weight the duties and tasks. To<br />

accomplish this weighting, the TI values for each task are<br />

squared and summed within a duty. The weight for each duty is<br />

the sum of the squared TI values across all tasks within the duty<br />

divided by the sum of the squared TI values across all duties.<br />

These weights are the percentages of test questions to be<br />

selected to cover the required knowledges to successfully perform<br />

the tasks within each duty.<br />

The TI value of each task within a duty is reflected by a<br />

letter from A to D. Tasks are designated as r'AV' tasks if their -.<br />

TI values are at least one standard deviation above the mean of<br />

the TI values or if their TI values are at least 6.0. Similarly,<br />

tasks are designated as 'ID" tasks if their TI values are more<br />

than one standard deviation below the mean of the TI values:<br />

however, all tasks with TI values of at least 4.00 are designated<br />

as qVC'* tasks. Of the remaining tasks, the upper 50% are<br />

designated "B" tasks and the lower 50% are designated "C" tasks.<br />

SMEs are required to write at least one item to test the job<br />

knowledge required for every '@A" task and to write no more than<br />

three items for a single task. Procedures are available to<br />

override these restrictions; however, they require written<br />

justification. Items can be written on "D1@ tasks only with the<br />

group facilitator's approval.<br />

PROCEDURE<br />

Tasks for each of 26 AFSs for which testing importance<br />

indices were available (914X0, 753X0, 423X3, 791X0, 423X4, 792X1,<br />

915X0, 908X0, 392X0, 231X2, 542X2, 674X0, 552X0, 324X0, 542X1,<br />

427X3, 321XlE, 112X0, 121X0, 274X0, 321XlG, 241X0, 431X0, 275X0,<br />

566X0, and 231X0) were randomly divided into two samples--one<br />

sample designated the Validation sample" and the other sample<br />

designated the "cross-validation sample." First, the IIAtt tasks<br />

were randomly split between the two samples, such that each<br />

sample contained approximately an equal number of l*AVV tasks. The<br />

IIB, II IICII and I'D“ tasks were split between the two samples in the<br />

same manner. Regression equations were computed separately for<br />

each validation and cross-validation sample with TI as the<br />

criterion and PMP, PTM, AG, TD, and TE as the predictor<br />

variables. The two sets of regression weights computed for each<br />

AFS were applied to the predictor scores for the cross-validation<br />

sample to generate predicted testing importance (PTI) values.<br />

The predictive efficiency of each set of weights can be measured<br />

by the Pearson coefficient of correlation (r) between TI and PTI.<br />

If the shrinkage in r using the two sets of weights on the<br />

cross-validation sample is statistically nonsignificant (Walker t<br />

Lev, 1953), then the data for both samples can be combined for a<br />

hierarchical clustering analysis. In this procedure, the number<br />

Of regression equations is reduced by one at each stage of the<br />

clustering by combining AFSS into groups and combining their<br />

corresponding regression data. The two most similar groups are<br />

combined at each stage, as measured by the resulting loss of<br />

312<br />

,<br />

i

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!