|Title:||Cross-Classified Structural Equations Model: Development of an OpenMX Module and its Application to Multiyear Assessment and Intervention Data in Literacy Research|
|Principal Investigator:||Mehta, Paras||Awardee:||University of Houston|
|Program:||Statistical and Research Methodology in Education [Program Details]|
|Award Period:||3 years||Award Amount:||$702,393|
|Goal:||Methodological Innovation||Award Number:||R305D090024|
The proposed project aims to: (1) develop a software library for fitting cross-classified structural equations models (CC-SEM); (2) analyze a number of large literacy-related datasets with common methodological issues; and (3) document and disseminate the software library.
Multilevel modeling (MLM) is the preferred approach for modeling longitudinal and nested educational datasets. However, multiyear student data are no longer hierarchically nested within a single teacher, but are partially cross-classified due to dispersion of students from the same classroom to multiple different classrooms. In addition, student outcomes may be partially clustered within multiple teachers. Although current MLM software packages allow cross-classified data to a certain extent, estimation is computationally challenging for such nonhierarchically nested data. In addition, current MLM software packages allow outcome data only at the lowest level of hierarchy. Multi-level structural equations modeling (ML-SEM) has more recently become available and it allows for latent variables at each level measured by multiple indicators at that level. The current proposal aims to develop a ML-SEM software library that will allow latent variable models for arbitrarily cross-classified and multiple-membership data.
The CC-SEM library will be developed as an extension to OpenMx, an open source SEM project integrated into the R statistical modeling language. Integration of CC-SEM with R and OpenMX is to make the software: (1) fully programmable and extensible; (2) free and opensource; (3) multiprocessor and cluster enabled; and (4) user-friendly with multiple front-ends (e.g., matrix, simple script based and graphical interfaces). The library will use sparse matrix methods for formulating CC-SEM models as restricted linear mixed-effects models and for computing the likelihood and its derivatives. Potential users of CC-SEM include substantive users interested in analyzing data, methodologists interested in evaluating models and methods, and statistical developers interested in further extending its functionality.
In addition, the project will analyze a number of large educational datasets with common methodological issues that are of interest to educational researchers. The methodological issues include: (1) longitudinal (multiple years, grades, and cohorts) student language and literacy outcome data with multiple within-year and end-of-year assessments with cross-classification of responses within teachers from different grades; (2) multiple teachers within a grade; (3) repeated teachers across years;(4) pullout instruction for a subset of students; (5) multiple latent student constructs of interest; and (6) teacher, school and district level constructs of interest.
The project will fully document every aspect of the software library as well as the CC-SEM modeling framework using the R documentation standards. A user-friendly manual for the CC-SEM software will be developed that will include examples from publicly available datasets as well as the datasets used in the secondary data analysis.
Related IES Projects: Language and Literacy Abilities in Spanish Language Speaking Children (R305A100272) and The Roles of Instruction and Component Skills in Reading Achievement (R305A120785)
Publications from this project:
Branum-Martin, L. (2013). Multilevel Modeling: Practical Examples to Illustrate a Special Case of SEM. Applied Quantitative Analysis in Education and the Social Sciences. Rutledge.
Mehta, P.D. (2013). nLevel Structural Equation Modeling. Applied Quantitative Analysis in Education and the Social Sciences. Rutledge.