Exploring the connection between deep learning and learning assessments: a cross-disciplinary engineering education perspective

The paper investigates how assessment methods relate to deep learning within multidisciplinary engineering education. Deep learning, first distinguished by Marton and Säljö (1976), emphasizes understanding, integration, and linking new concepts to prior knowledge, in contrast to surface approaches focused on rote memorization. Prior studies suggest self/peer assessment and feedback can promote deep learning, which correlates with better performance, retention, and transfer of knowledge. In contemporary blended and online learning contexts, designing an effective learning environment and content is crucial. Engineering employers seek graduates with systems thinking, innovation, interdisciplinary and communication skills, making deep learning essential in engineering curricula. However, limited research has examined which factors foster deep learning in engineering and how assessment types (summative vs. formative) relate to those factors. This study aims to identify the dimensions of deep learning factors and learning assessments, and to test the relationships between assessment types and deep learning factors for engineering students.

The literature review outlines the deep learning concept and its associated components, focusing on factors educators can influence. Deep learning entails comprehension, critical engagement, connection of ideas, and application in real-life contexts, while surface learning relies on memorization. Warburton (2003) posits that learning environment and course content/design are key factors affecting deep learning for engineering students, while individual student factors are harder to manage. Learning environment components include student-teacher relationships, satisfaction, support, relevance, resource availability, effective assessment strategies, and real-world applications. Efforts by teaching staff to make units interesting, effective communication, appropriate workload, good teaching, and availability of online/offline resources enhance deep learning. Innovative learning environments and real-world opportunities are positively associated with deep learning. Course content and design should emphasize key concepts, varied learning opportunities, and relevance to engineering applications, incorporating peer/group learning, deep theoretical understanding, involvement of industry professionals, industry visits, flexible teaching, and student choice/voice. Real-world tasks and diverse instructional methods enhance engagement and outcomes. Learning evaluation (assessment) is central to the environment and includes multiple methods (problem-based assignments, open/closed-book problem solving and exams, multiple-choice tests, seminars/presentations). Assessments aligned with learning objectives foster deep learning. Two assessment types are distinguished: summative (e.g., final exams, major reports, presentations) and formative (problem-based, in-class, with feedback enabling improvement). Prior work indicates assessment type influences learning strategies, suggesting both summative and formative assessments may impact deep learning factors. Hypotheses propose positive relationships between summative/formative assessments and two deep learning factors (learning environment; course content and design).

A structured, paper-based questionnaire was administered to 600 randomly selected second-, third-, and final-year Civil and Mechanical Engineering undergraduates at Queensland University of Technology (QUT); 243 complete responses were received (response rate ~40%), an adequate sample for SEM (recommended 200–400). Demographics: ~70% domestic, ~30% international (9 countries), 82.6% male, average age 24 (range up to 39), GPAs 4–7 (scale 1–7). The instrument items for factors affecting deep learning and perceptions of assessment methods were derived from the literature. Section one captured demographics; section two used a 5-point Likert scale (1=strongly disagree to 5=strongly agree) to rate items covering learning environment, course content/design, and assessment methods. Reliability was assessed via Cronbach’s alpha: learning environment (4 items, α=0.819), course content and design (4 items, α=0.751), learning evaluations (4 items, α=0.764), summative assessment (3 items, α=0.764), formative assessment (2 items, α=0.792), indicating good internal consistency. Structural equation modeling (SEM) using AMOS 25 was employed to test the hypothesized relationships, following a two-step approach (measurement then structural models). Exploratory factor analysis (EFA) preceded SEM to identify factor structures: PCA with varimax rotation, eigenvalue>1, and factor loadings ≥0.65; KMO and Bartlett’s tests verified suitability. Cross-loaded/low-loading items were removed to refine constructs. Ethical procedures ensured anonymity and confidentiality to mitigate common method bias.

• Response and structure: 243 valid responses (~40%). EFA identified two deep learning factors—Learning Environment and Course Content & Design—explaining 61.5% variance; and two assessment factors—Summative Assessment and Formative Assessment—explaining 74.3% variance. Some deep learning items (university learning environment, student choice/voice, flexibility, field trips/industry visits) and two assessment items (problem-based assignment, multiple-choice test) did not load adequately and were excluded. Selected item means (N=243 unless noted): efforts to make units interesting (M=4.24, SD=0.886), availability of resources (M=3.94, SD=0.893), effective assessment strategy (M=3.99, SD=0.935), real-life examples/videos (M=4.11, SD=0.864), peer learning (M=3.35, SD=1.178), group learning/tutorials (M=3.83, SD=1.059), deep explanation of theories (M=3.79, SD=0.894), involvement of industry in lectures (M=3.71, SD=0.954). Assessment item means: open-book in-class problem solving (M=3.14, SD=1.174, N=220), open-book final exam (M=4.02, SD=1.007, N=221), seminar/presentation (M=3.67, SD=1.045, N=222), close-book in-class problem solving (M=3.13, SD=1.256, N=221), close-book final exam (M=3.93, SD=1.005, N=219).
• Measurement/model fit: SEM fit indices indicated good model fit: χ²(60)=249.944, χ²/df=4.166 (<5), RMSEA=0.074, CFI=0.921, IFI=0.923, TLI=0.898, GFI=0.925, AGFI=0.887.
• Path estimates (all significant): • Summative Assessment → Learning Environment: β=0.36, t=4.208, p<0.001. • Summative Assessment → Course Content & Design: β=0.31, t=3.573, p<0.001. • Formative Assessment → Learning Environment: β=0.29, t=3.295, p<0.001. • Formative Assessment → Course Content & Design: β=0.28, t=3.128, p=0.002.
• Interpretation: Both assessment types significantly and positively influence the two deep learning factors, with summative assessment showing the strongest effect on the learning environment. All hypothesized relationships (H1a, H1b, H2a, H2b) were supported.

The study confirms that deep learning in engineering education is shaped by two educator-influenced dimensions: learning environment and course content/design. Both summative and formative assessments significantly and positively influence these dimensions. Summative assessments (e.g., open/close-book final exams, end-of-term presentations) appear to most strongly shape the learning environment by directing students’ engagement with resources, attention to lecturer efforts, adherence to assessment strategies, and use of real-life examples. Knowledge of exam formats and expectations leads students to optimize environmental features for success. Summative assessment also promotes group work, theory comprehension, and engagement with industry perspectives—key aspects of course content/design. Formative assessment (e.g., in-class open/close-book problem solving) fosters ongoing engagement, feedback, and opportunities for improvement, positively affecting both the learning environment and course design features such as peer/group learning and deeper theoretical understanding. Together, these findings address the research question by demonstrating that assessment design is a lever for promoting deep learning. Practically, integrating both assessment types provides complementary benefits: formative assessments guide learning during the course, while summative assessments consolidate achievement and motivate strategic engagement. These results align with prior work showing that well-aligned, varied assessments enhance deep learning, and they underscore the need for intentional assessment design to cultivate environments and curricula that support sustained, meaningful learning.

The study develops and empirically tests a theoretical model linking assessment types to deep learning factors in engineering education. Two deep learning dimensions (learning environment; course content and design) and two assessment dimensions (summative; formative) were identified. SEM results show that both summative and formative assessments significantly and positively influence both deep learning dimensions, with summative assessment exerting the strongest effect on the learning environment. Contributions include: (1) a validated multidimensional model connecting assessment methods to educator-controlled deep learning factors; (2) evidence from a sizeable undergraduate engineering sample; and (3) practical guidance that combining formative and summative assessments supports deep learning. Future research should extend beyond a single-institution survey, incorporate experimental or longitudinal designs, include additional variables (e.g., deep learning outcomes), and examine demographic moderators.

• Design and data: Cross-sectional, perception-based survey from a single university limits causal inference and generalizability. Experimental or longitudinal studies could better capture causal mechanisms and pre/post effects.
• Common Method Bias: Only anonymity/confidentiality was used to mitigate CMB. Additional procedural and statistical remedies (e.g., multi-method data, temporal separation, marker variables) are recommended.
• Measurement scope: Some hypothesized deep learning items (e.g., student choice/voice, flexibility, field trips/industry visits) showed poor loadings, possibly reflecting contextual constraints; further refinement and contextual validation are needed.
• External validity: Results are specific to QUT engineering students; replication across institutions, disciplines, and cultural contexts is needed.
• Unexamined moderators/outcomes: Relationships with demographics (gender, age, nationality, year of study) were not analyzed; deep learning outcomes (performance, transfer) were not included as dependent variables.