Mathematical word problems have long drawn educational researchers’ attention. Through word problems, students have the opportunity to apply and demonstrate their mathematical thinking to real-life problem-solving. However, not all math problems are created equally: there are many potential factors, including item structure, number properties, and linguistic features within a word problem that can negatively affect students’ responses, irrespective of their true mathematical understanding. This study proposes network analysis modeling as a novel approach for quantifying item process data at student, item, and student-item hybrid levels. We demonstrate these methods using 30 elementary students’ action sequences (eight per item) on two comparable math items that differed primarily in type of verbal information presented.