Combining Clickstream Analyses and Graph-Modeled Data Clustering for Identifying Common Response Processes.

Area of Science:

• Educational Measurement
• Psychometrics
• Computer Science

Background:

• Complex interactive test items are increasingly used in computer-administered assessments.
• These assessments generate time-stamped action sequences, offering rich data for understanding examinee behavior.
• Existing research primarily uses action sequences at the item-level, overlooking detailed timing information.

Purpose of the Study:

• To develop an approach that jointly analyzes action sequences and action-level timing data.
• To identify common response processes in complex interactive test items.
• To enhance the fine-grained assessment of examinee behavior.

Main Methods:

• Integration of clickstream analysis, graph-modeled data clustering, and psychometrics.
• Development of similarity measures incorporating both actions and action-level timing.
• Application of cluster edge deletion to identify homogeneous groups of action patterns.

Main Results:

• The approach successfully identifies common response processes by considering both action sequences and timing.
• Demonstrated utility on a complex problem-solving item from PIAAC 2012.
• Provides guidelines for applying the novel analytical method.

Conclusions:

• Joint analysis of action sequences and action-level timing offers a more nuanced understanding of examinee behavior.
• The proposed method enables the identification of interpretable and distinct groups of response processes.
• This approach advances the assessment of complex cognitive tasks in educational and large-scale testing contexts.