Date of Completion

8-8-2017

Embargo Period

2-3-2018

Keywords

Gifted Identification, Bayesian Data Analysis, Simulation

Major Advisor

Mccoach, D. Betsy

Associate Advisor

Swaminathan, Hariharan

Associate Advisor

Rogers, Jane

Associate Advisor

Loken, Eric

Field of Study

Educational Psychology

Degree

Doctor of Philosophy

Open Access

Open Access

Abstract

This study applied statistical simulation techniques to posit a practical situation of gifted identification based on students’ performance on the intelligence test and academic achievement tests of math and reading. Three tests were generated based on the one-parameter item response theory model. The marginal reliabilities were .95, .90, and .80 for observed intelligence and observed abilities in math and reading, respectively. Totally, 1,000,000 pairs of true and observed abilities in each of the three domain areas were generated. Results suggest that using different combination rules is conceptually aimed to identify different gifted populations. Using the conjunctive rule to combine a high standard for all three measures is aimed to identify the gifted population who are gifted in all three domain areas; however, using the complementary rule is aimed to identify the gifted population who are gifted in at least one domain area. Therefore, the consequences on the identified gifted group and the performance of gifted identification using different combination rules should not be compared directly. Given the differences in the reliabilities and correlations, results suggest that the conjunctive and compensatory rules identify more students who are gifted in the domain area measured with higher reliabilities and more correlated with the other domain areas; and, the complementary rule favors the students who are gifted in the domain area measured with higher reliabilities. Using any test as the conclusive test favors the students who are gifted in the domain area more correlated with the domain area measured by the conclusive test. In general, gifted identification using different combination rules may perform relatively well in terms of the positive predictive rate (PPR) or sensitivity but rarely both. This study explored new methods of gifted identification under the Bayesian framework to systematically address the measurement error issues. Results suggest that using the posterior probability of being gifted given multiple observed abilities to identify gifted students has the potential to simultaneously improve the PPR and sensitivity of gifted identification.

COinS