Toward Predicting Test Score Gains With Online Behavior Data of Teachers

As technology continues to disrupt education at nearly all levels from K–12 to college and beyond, the challenges of understanding the impact technology has on teaching continue to mount. One critical area that yet remains open, is examining teachers’ usage of technology by specifically collecting detailed data of their technology use, developing techniques to analyze that data and then finding meaningful connections that may show the value of that technology. In this research, we will present a model for predicting test score gains using data points drawn from typical educational data sources such as teacher experience, student demographics and classroom dynamics, as well as from the online usage behaviors of teachers. Building upon prior work in developing a usage typology of teachers using an online curriculum planning system, the Curriculum Customization Service (CCS), to assist in the development of their instruction and planning for an Earth systems curriculum, we apply the results of this typology to add new information to a model for predicting test score gains on a district-level Earth systems subject area exam. Using both multinomial logistic regression and Näıve Bayes algorithms on the proposed model, we show that even with a simplification of the highly complex tapestry of variables that go into teacher and student performance, teacher usage of the CCS proved valuable to the predictive capability in average and above average test score gains cases.