U.S. Employment Service. Correlations were computed between two independent raters for each OAI work element. The mean correlation was .53 and the median was .56. Several studies aimed at evaluating the construct validity of the OAI have been conducted, including the comparisons of clusters of occupations obtained with the OAI on several tests and inventories (68 of the 92 measures showed statistically significant discrimination between the clusters), the prediction of mean occupational scores on the General Aptitude Test Battery using OAI factor scores, (median cross-validated multiple correlations were .60 for mental and .24 for motor abilities), bivariate correlations between OAI attribute-requirement estimates and mean scores of job incumbents (statistically significant correlations at the .05 level were found for 38 of 55 analyses), and analyses of variance to relate OAI need-requirement estimates to job satisfaction scores (12 of 15 analyses provided supporting evidence).
The OAI shows generally excellent measurement characteristics, when it is applied in the recommended manner—using college educated, trained analysts. Most of the reported empirical work has been conducted using "paper jobs," that is, written job descriptions from the U.S. Employment Service. It is not clear that it would work as well if used in the field by job incumbents, supervisors, or other occupational experts, many of whom would not be college-trained or be available for special training on the OAI.
A replacement for the OAI, the General Work Inventory (GWI), is shorter and written less technically and could be a more practical alternative for large-scale data collection. This instrument was developed for use by "any literate respondent who is familiar with the job to be analyzed" (Cunningham et al., 1990:34). It has 268 items organized into 8 sections and uses "part of the job" and "extent of occurrence" rating scales, both of which have 9 points and are adjectivally anchored. Research using this inventory in the military showed mean retest reliabilities (for single raters) of .62 across all items, and a mean correlation of profiles of ratings (again, for single raters) of .74, comparable with other similar studies. Ballentine et al., (1992) used the GWI to create a hierarchical structure of Air Force occupations that showed intuitive meaning and corresponded to existing Air Force classifica-