|Title:||An Adaptive Testing System for Diagnosing Sources of Mathematics Difficulties|
|Principal Investigator:||Embretson, Susan||Awardee:||Georgia Institute of Technology|
|Program:||Science, Technology, Engineering, and Mathematics (STEM) Education [Program Details]|
|Award Period:||4 years||Award Amount:||$1,854,393|
Co-Principal Investigators: Bruce Walker, John Poggio, Neal Kingston, and Edward Meyen
Purpose: Current standards-based state accountability tests typically find that many students do not meet proficiency standards. However, state accountability tests cannot pinpoint specific sources of students' difficulties so that differentiated instruction may be given to remedy the difficulties. The purpose of this project is to build an online and on-demand adaptive assessment system to diagnose sources of students' mathematics difficulties and deficits that are linked to state standards-based assessments.
Project: The researchers will develop an online diagnostic assessment system to diagnose sources of students' mathematics deficits and provide information to teachers to guide instruction.
Products: The outcomes of the project include an adaptive testing system for identifying sources of middle school students' mathematical deficits that are related to state accountability tests, along with published reports.
Setting: The study will be conducted on a representative sample of middle schools in Kansas.
Population: The study sample will consist of a minimum of 3,000 students each in Grades 6 and 7. The students will be representative of the population of Kansas, with a special focus on students with competencies below curriculum-based performance standards, many of whom will be from low income or ethnic minority groups.
Intervention: The assessment system will include a large item bank developed by automatic item generation methods. The assessment system will be administered adaptively and on-demand by online computers for efficient and explicit diagnosis. Items in the assessment system will not only be aligned with national standards for basic mathematics, but will also have well-specified sources of cognitive complexity that are calibrated in an item response theory (IRT) model. Items will be selected adaptively to provide maximum information according to diagnostic IRT models for efficient and valid measurement of each student. The diagnostic assessment system can recommend specific lesson plans, prior knowledge exercises, and online student tutorials that are directly related to Kansas state standards available from the Blending Assessment with Instruction Project. The Blending Assessment with Instruction Project (BAIP) is widely used by Kansas teachers and its direct interface with the diagnostic system allows the seamless collection of validity data.
Research Design and Methods: The diagnostic system for the assessments will be developed by combining standards-based knowledge categorizations of mathematics items with cognitive models of mathematical problem solving. The assessment system will consist of seven components: (1) a diagnostic system; (2) item bank; (3) diagnostic IRT model calibrations; (4) adaptive item selection modules; (5) an interactive online test delivery module; (6) a score report module; and (7) a validity module. A series of studies will be conducted focusing on assessment development and testing. To examine the external and consequential aspects of validity, the relations between the diagnostic scores and performance on online tutorials will be examined using the BAIP system. The adaptive diagnostic assessments will be tested with a stratified, random sample of teachers and their students. Students' adaptive diagnostic test scores will be linked to scores obtained from the summative Kansas state standardized test in mathematics to evaluate the appropriate functioning of the diagnostic assessment system.
Control Condition: There is no control condition.
Key Measures: The key measures for the study include students' scores on the diagnostic mathematics assessments and students' end of year mathematics test score on the Kansas state standardized assessment.
Data Analytic Strategy: Structured item response theory (IRT) models will examine both the response process and content validity of models used to generate diagnostic items. A diagnostic IRT model will be calibrated and implemented for the adaptive testing system. Studies on external aspects of validity will be assessed using hierarchical linear models to examine the relation between diagnostic scores of learning processes and the online tutorials. In addition, structural equation modeling will be used to study convergent and discriminant validity relationships.
Publications from this project:
Embretson, S.E. and Yang, X. (2013). A Multicomponent Latent Trait Model For Diagnosis. Psychometrika, 78: 14–36.
Embretson, S.E. (In Press). Multicomponent Models. In Van Der Linden, W. and Hambleton, R. (Eds.). Handbook Of Item Response Theory: Second Edition. New York: Taylor and Francis Inc.