Graduation Year

2025

Document Type

Dissertation

Degree

Ph.D.

Degree Name

Doctor of Philosophy (Ph.D.)

Degree Granting Department

Educational Measurement and Research

Major Professor

John Ferron, Ph.D.

Committee Member

Eunsook Kim, Ph.D.

Committee Member

David Lamb, Ph.D.

Committee Member

Tony Tan, Ph.D.

Keywords

geography, multilevel modeling, spatial analysis, virtual schools, assessment

Abstract

This methodological case study introduces a novel approach to investigate the impact of student, teacher, and geographic factors on the academic achievement of virtual school students in Florida. Traditional linear regression models, and even multilevel models, often overlook spatial dependencies, which can lead to biased estimates and incomplete inferences. This research addresses this limitation by comparing a cross-classified multilevel model (CCMM) with a novel cross-classified spatial model (CCSM) that incorporates coefficients from geographically weighted regression (GWR). Utilizing secondary data from a statewide virtual school, including student demographics, teacher characteristics, and standardized test scores, along with U.S. Census Bureau data, the first goal was to compile a comprehensive dataset. From this dataset, descriptive analyses revealed regional variations in test scores, with lower performance observed in northeast and north-central Florida, and higher scores in some major metropolitan areas. CCMM results indicated that student-level characteristics were the primary drivers of variance in test scores, consistent with existing literature on achievement gaps related to gender, race, socioeconomic status, and disability. Teacher-level factors, such as the number of assigned students and years of employment, also significantly predicted achievement. While initial CCMMs showed that geographic-level characteristics (e.g., proportion of adults without a high school diploma) also influenced test scores, the fundamental assumption of geographic independence was violated. To address this issue, GWR models were employed, and showed that the relationship between geographic predictors and student achievement varied significantly within and across Florida regions. The novel CCSM integrates these GWR-derived coefficients into the CCMM framework. Findings suggest that this hybrid approach refines the interpretation of geographic fixed effects, particularly regarding racial composition, by accounting for local spatial processes. For instance, the perceived impact of a region’s racial makeup on test scores is altered when the non-stationarity of poverty and low education is taken into account. This study makes significant methodological contributions by demonstrating the benefits of combining multilevel and spatial regression techniques in the field of education. It highlights the importance of accounting for both nested data structures and spatial dependencies to gain a better understanding of the multiple influences on student achievement. Although limited by the availability of secondary data and the generalizability to a single virtual school, this research provides a framework for future studies on the contextual influences on educational outcomes, particularly within virtual learning environments.

Download

Share

COinS