| This study explored relationships between writing sample features and LMS usage patterns for 366 college students who enrolled in Theology courses, junior-level courses cross-listed with theology courses, or Senior Perspective Program courses in the fall semester of 2012. These hybrid courses were managed inside the Canvas(TM) learning management system. LMS usage log data, containing the URLs visited as well as date/time stamps, were processed to generate session metrics and URL token categories, which were further reduced to eight prominent features (Assignments, Conversations, Files, Grades, Modules, Topics, View, and Wiki). Writing samples were processed using the LIWC software, producing word counts and linguistic particle analysis data. These data were further reduced to writing sample metrics and rollup categories labeled as Pronouns, Articles, Verbs, Social, Cognitive, and Relativity. After students were ranked into quartiles in each category by course frequencies, with outliers receiving boundary rankings, they were pre-classified by their rankings in specific categories as students with Fast Clicker and High Times Out LMS usage patterns. Correlation mining was then performed among and across all writing sample and page views categories. Additional correlation mining was performed among courses classified as Files/Assignments, Assignments/Files, Modules/Assignments, and Files/Modules, and among writing samples classified into four groups based upon average word counts. These correlation tests revealed a clear relationship between writing sample features and LMS usage patterns, suggesting a possibility for building predictive models. This possibility is tempered, however, by the course design and writing sample requirements. The study further explored generating decision trees and applying them to new data from a similar course in the following semester. |
