From generation to judgment: the multidimensional structure of design thinking mindset and differences across learning stages in AIGC-supported fashion film design learning

Synthesis: Liu, Guo, Mu, Rao, Ni, Shen and You surveyed 409 first- and third-year students at three Chinese universities who had used AIGC tools in fashion film design projects, and tested whether design thinking mindset holds its five-dimensional structure across year groups. It did: the five dimensions formed a distinguishable, correlated structure with measurement invariance by year, and third-year students scored higher on all five latent means, from 0.622 for Empathy to 0.865 for Abductive Reasoning. The authors' argument is that generative capacity is not the same thing as design judgment, so a program that only teaches students to produce more alternatives leaves the mindset untested.

Key Findings

1. Five DTM dimensions held as a distinguishable structure. Uncertainty and Risk, Empathy, Experimentation, Abductive Reasoning and Creative Confidence were modeled as correlated first-order factors over 16 items and fit well (CFI = 1.000, TLI = 1.004, RMSEA = 0.000, SRMR = 0.021), where a one-factor model did not (CFI = 0.779, RMSEA = 0.170).
2. The structure was comparable across year groups. Measurement invariance testing supported comparison of latent means between first- and third-year students, which is what makes the year difference interpretable.
3. Third-year students had higher latent means on all five dimensions. Standardized differences ran 0.718 (Uncertainty and Risk), 0.622 (Empathy), 0.721 (Experimentation), 0.865 (Abductive Reasoning) and 0.760 (Creative Confidence), all Holm-adjusted p < 0.001.
4. Abductive reasoning and creative confidence showed the largest estimates. The authors note that no formal contrast between dimension-specific differences was tested, so the ranking is descriptive.
5. Observed-score tests converged with the latent model. Welch t tests on observed means were significant across all five dimensions after Holm adjustment, with Cohen's d from 0.594 to 0.719.
6. University membership did not explain the year difference. A sensitivity analysis adjusting for university shifted the year coefficients by roughly 1.3–3.5%, and all five remained significant.
7. The sample was balanced and clean. 411 questionnaires yielded 409 usable cases after excluding two straight-lined responses; 201 first-year and 208 third-year students, with no missing values on the 16 items.

Generation is not judgment

The paper's framing is the useful part for program design. AIGC lets students produce and compare a much wider range of visual alternatives than they could make by hand, but the availability of alternatives does not tell you whether a student can interpret, evaluate, select among or commit to one. The five DTM dimensions are the psychological orientations that do that work — tolerance for uncertainty, empathy for an audience, willingness to experiment, abductive reasoning toward a plausible direction, and confidence in one's own creative judgment.

That reframes what an AIGC-heavy studio course is assessing. If the deliverable is a set of generated alternatives, the course measures generation, which the tool now does. If the deliverable is a defended selection under uncertainty, the course measures the mindset the authors are trying to get at.

What a year-group difference can and cannot mean

Third-year students scoring higher on all five dimensions is consistent with a design education that develops these orientations over time, and it is equally consistent with selection — students who lacked them leaving the program, or a cohort difference unrelated to teaching. The design is cross-sectional with two year groups, so the paper cannot separate development from selection, and the authors say so.

The measurement work is what earns the comparison. Invariance testing across year groups is the step that makes a latent mean difference meaningful rather than an artifact of students interpreting items differently as they advance, and the authors report it for the five focal dimensions rather than asserting comparability.

Scope and instrument

The instrument adapts Vignoli et al.'s DTM framework into 16 items across five dimensions, translated into Chinese and reviewed by four experts, with a 27-student pilot excluded from analysis. Tools named across projects included ChatGPT, DeepSeek, Grok, Jimeng, Nano Banana, Kling and Midjourney, and the authors make no attempt to compare them — the study is about the learners, not the tools.

The questionnaire measures self-reported tendencies, not observed design behavior or judged creative performance, which the authors state as a limit on what the dimensions can support.

What this means for practice

• Assess selection and justification, not output volume. Generated alternatives are cheap now, so the gradeable work is the reasoning that picks among them and the audience reasoning behind the pick.
• Teach tolerance for uncertainty explicitly. It was the dimension with one of the larger year differences (0.718), and it is coachable through briefs that leave the right answer genuinely open.
• Build abductive reasoning into studio critique. The largest year difference was in forming plausible creative directions (0.865), which suggests critique time spent on "what could this become and why" rather than on execution quality.
• Do not read year differences as proof of curriculum effect. The design compares two cohorts at one time point; use a within-cohort pre/post or a longitudinal design to test whether the program causes the change.

Limitations

• Cross-sectional with two year groups. Developmental interpretations of the year differences cannot be separated from cohort or selection effects.
• Self-report only. The 16 items measure reported tendencies, not cognitive engagement, design judgment or creative performance.
• Three universities, purposive recruitment. First- and third-year students in digital media and image design programs at three Chinese universities, recruited online with researcher assistance.
• No formal comparison across dimensions. The ranking of standardized differences (largest for Abductive Reasoning and Creative Confidence) is descriptive; no dimension-specific contrast was tested.

Citation

Liu, Q., Guo, H., Mu, C., Rao, F., Ni, S., Shen, Z., & You, Y. (2026). From generation to judgment: the multidimensional structure of design thinking mindset and differences across learning stages in AIGC-supported fashion film design learning. Frontiers in Psychology, 17.