| This factor-analytic study of a performance-based identification protocol for young, socioeconomically disadvantaged high-ability learners investigated the issues of reliability, test equivalency, and bias. A group-administered, performance-based set of instruments was designed in a joint project between the Center for Gifted Education and the State Department of Education, South Carolina. These instruments went through a series of processes of review and refinement leading to their use in a field test in fall 1999. The outcome of this field test administration is the subject of exploratory and confirmatory factor analysis in this study.; Reliability of the instruments was established on the pilot study data which were gathered from a heterogeneous sample of 1425 students. Statistical anchoring using linear transforms was used to address the status of the two forms of the test instruments. The Cronbach alpha values ranged from 0.71 to 0.78, values lower than desirable for psychometric instruments, but acceptable in view of the special purpose of this test. Exploratory factor analysis on a randomly chosen half of the field test data (N = 1800 students) lead to structural equation modeling of both a priori and exploratory factors on the second half of the field test data.; The exploratory factor analyses did not support a construct of high-ability learning. All emergent factors accounted for less than a majority of the variance in the relevant sub-samples. Nonetheless, the structural equation models demonstrated that there was no evidence of bias on the basis of gender, ethnicity, or socioeconomic status. Project STAR did indeed exhibit the ability to discriminate in an unbiased way among young, socioeconomically disadvantaged high-ability learners.; The overriding implication of this study is that performance-based identification should be utilized as part of the testing battery available to school districts seeking to assess potential for learning. At the same time, the failure to detect a strong factorial structure in the results of a performance-based test specifically designed around a factorial schema implies that there are layers of complexity inherent in this testing protocol that deserve close attention. Further research arising from increasingly standardized implementations is expected to shed more light on what has been called in this study the "elusive factor" issue. |
