Abstract:
|
In the field of educational measurement, many testing programs have transitioned from paper-based to computer-based assessments. This transition enables testing programs to collect data on how examinees engage with the assessment, such as response time to test items, clicks and eye-tracking data. There are many aspects to analyze these big data to advance existing statistical models or improve understanding of cognitive process. This paper focuses on analyzing the student response time data. We build machine learning models to analyze how the response time are related with item characteristics, such as item types, locations and difficulties. We then use these models to predict the response time patterns using real data from two large scale assessments. Using lognormal models, we run simulation studies to demonstrate how to use item response time to improve test constructions. Since examinees are allowed to visit an item multiple times, we also analyze how the response time is related to number of visits and item transition through social network analysis. Timing distributions and their variations across demographic groups will also be compared and presented via data visualization.
|