Understanding Creative Design Practices among Data Artists

• Despite growing professional and institutional interest in artistic visualization, how designers make these works remains poorly understood.The field now includes exhibitions, awards, and professional careers, extending beyond the handful of examples described in 2007.Source: §1 · p. 1
• Prior research describes broad input-driven and output-driven workflows, but gives less detail on how initial directions develop through successive design attempts.The paper focuses on how artists draw on inspiration, test and revise representations, and respond to difficulties.Source: §1 · pp. 1–2
• The study combines 40 public project accounts, artifact-anchored interviews with seven experienced artists, and a design task with eight artists.These approaches provide breadth, contextual detail, and observation of ideation as it unfolds.Source: §1 · p. 2
• Designers develop stories through data exploration, adapt visual inspiration from data and external sources, and use sketches and real-data prototypes to test and reshape ideas.They generally develop one direction at a time, partly because producing alternatives is costly; decisions balance meaning, visual appeal, readability, and feasibility.Source: §1 · p. 2

2.1 Creative Visualization Design

• Creative visualization increasingly refers to personalized, non-traditional forms, but how practitioners design them remains insufficiently understood.Prior work includes sketch-based glyph authoring and image-embedded charts; this study examines the process that leads to the final artifact, extending beyond artifact-focused analysis.Source: §2.1 · pp. 2–3

2.2 Visualization Design Process

• Visualization design frameworks prescribe phases or activities, while practitioner studies describe less systematic work grounded in situated knowledge, precedence, experience, and judgment.Research has also identified inspiration from existing visualizations, real-world phenomena, and personal experiences, and examined writing as an early-stage externalizing method.Source: §2.2 · p. 3
• A key open question is how academic design frameworks align with the actual processes of creative practitioners such as data artists.Understanding this relationship may inform both general visualization designers and creative professionals, as well as future research combining the approaches.Source: §2.2 · p. 3

• The studies are complementary lenses on one phenomenon, spanning breadth, situated reflection, and temporal immediacy rather than serving as independent validations.Source: §3 · pp. 3–4
• Study I analyzes first-person project accounts for patterns in ideation, development, and challenges, while Study II uses artifact-anchored interviews to probe decisions and abandoned directions.Interviews use participants’ project materials and targeted questions, but remain retrospective and subject to memory and later interpretation.Source: §3 · p. 4
• Study III observes eight participants developing a concept during a live session and independent work, documenting ideas, changes, and abandoned possibilities as they unfold.Participants were not required to complete a polished professional project, limiting commitment.Source: §3 · p. 4

3.1 Study I: Public Accounts Analysis

• The corpus comprises 40 visualization-design process articles by 37 authors, drawn from VISAP (5), Nightingale (26), and designers’ websites and books (9), and published between 2018 and 2025.The researchers read 449 remaining articles against the inclusion criteria before finalizing the corpus.Source: §3.1.1 · pp. 4, 6

3.1.2 Codebook Development.

• Three authors developed the codebook through independent coding and discussion, then iteratively revised it and revisited previously coded accounts when changes were made.Source: §3.1.2 · p. 6

3.1.3 Coding Scheme.

• The coding scheme contains 11 top-level codes and 82 sub-codes, allowing events to receive multiple applicable sub-codes and to be ordered by actual event sequence.An activity’s top-level code depends on its role; brainstorming, for example, can be coded as Inspiration or Get unblock.Source: §3.1.3 · pp. 6–7

3.1.4 Analysis of Process Sequences.

• Analysis of 40 project event chains examines activity composition, prevalence, normalized position, transitions, recurring routines, and responses to difficulties.It uses event counts and prevalence to distinguish activity frequency from how widely it appears across projects, and tests differences by output format and stated goal.Source: §3.1.4 · p. 7

3.2.1 Participants.

• Seven experienced data artists with at least five years of professional practice were recruited through targeted email.Participants self-identified as data visualization designers and had public portfolios with several qualifying creative or artistic works.Source: §3.2.1 · p. 7

3.2.2 Procedure.

• Interviews used stimulated recall, anchoring each conversation to one or two participant-chosen projects and their retained materials.Sessions lasted around 60 minutes and explored inspiration, difficulties, externalized ideas, considered forms, and ideal ideation tools.Source: §3.2.2 · p. 8

3.2.3 Analysis.

• Two authors developed interview themes inductively through separate coding and discussion, then compared stable themes with Study I findings.They did not apply Study I’s codebook to the interviews.Source: §3.2.3 · p. 8

3.3.1 Participants.

• Eight of twelve initial participants completed the study; participants had experience in data visualization and backgrounds in design, art, or creative visualization.Recruitment was limited to U.S. residents because of the institution’s payment process.Source: §3.3.1 · p. 9

3.3.2 Procedure.

• In Session 1, participants chose an unfamiliar dataset and spent about twenty minutes developing an open-ended expressive visualization idea.They continued independently afterward, keeping rough and abandoned work and logging notable ideas, inspirations, and obstacles.Source: §3.3.2 · p. 9
• In Session 2, participants reviewed their independent work and discussed idea sources, rejected directions, obstacles, alternatives, and desired ideation tools.The study used two video sessions with independent work between them.Source: §3.3.2 · p. 9

3.3.3 Analysis.

• Researchers reconstructed each case from recordings, transcripts, artifacts, and log cards, then coded eight event chains comprising 132 events.They retained Study I’s codebook and treated generative-AI involvement as an event attribute before comparing findings across all three studies.Source: §3.3.3 · p. 10

4.1 The Overall Design Process

• In 87 of 260 transitions (33.5%), designers returned to an earlier activity, and written accounts and observed sessions both showed a flexible rather than fixed process.Returns often led from refinement to external input, from a visual idea to external inspiration, or from inspiration back to data; designers could skip sketching or prototyping when an idea was clear and return to sketching when a prototype failed.Source: §4.1 · pp. 10–11

4.2 Sources of Inspiration

• Story-finding and data exploration were intertwined in observed sessions, while external references, everyday imagery, personal experience, and data itself all supplied visual inspiration.Other people’s work accounted for 71.9% of external-inspiration events; data exploration also generated visual forms, and designers adapted existing forms for meaning and appearance.Source: §4.2 · pp. 11–13

4.3 Stages Define Problems and Responses

• Problems shifted from unsettled data or visual direction early in projects toward issues with the emerging artifact later.Earlier problems more often prompted a change of direction; later ones more often led designers to adjust an established direction. Outside input also shifted from seeking knowledge or skills earlier to evaluation by others later.Source: §4.3 · pp. 13–14

4.4 Roles of Real Data in Visual Representation

• Sketches helped develop mappings, while prototypes made with real data exposed visual problems and could rule out or redirect ideas before substantial production.Eight of 14 refinements after sketches concerned mapping, while six of 10 after mockups concerned the visual; working with real data also cost time and enthusiasm.Source: §4.4 · pp. 14–15

4.5 Blockers and Getting Unblocked

• Designers matched responses to blockers, while readability and the constraints of physical media bounded which creative directions remained workable.Changing material or technique accounted for 34% of 100 unblock events; designers also paused and sometimes returned to an existing idea with a changed perspective.Source: §4.5 · pp. 15–16

4.6 Sequential but Iterative

• Refinement was common, but designers usually developed one active direction rather than parallel complete alternatives, partly because producing candidates was costly.Refinement accounted for 155 of 682 events (22.7%) and appeared in 38 of 40 accounts; among those 38, five returned to data work, compared with 18 returning to inspiration or another visual idea.Source: §4.6 · p. 16

4.7 Use of AI

• In the design task, AI mainly supported prototyping, data exploration, outside perspective, and technical assistance, reducing the cost of testing alternatives without taking over design judgment.Seven participants used a generative model; the observed use was shaped by task constraints and may not reflect designers’ usual practice.Source: §4.7 · p. 17

5.1 Situated Judgment and Exploration of Alternatives

• Designers often develop one direction at a time, revisiting alternatives when progress stalls; the effort of making options concrete motivates support for testing them.The findings do not establish whether parallel development would improve the work.Source: §5.1 · pp. 17–18

5.2 Inspiration Develops through References and Data Exploration

• References offer meanings or ways to organize and encode data, while data exploration helps designers develop stories and experiment with visual forms.A reference’s usefulness depends on its relevance to the project as well as its appearance.Source: §5.2 · p. 18

5.3 Prototyping with Real Data Helps Develop the Idea

• Sketches help compare metaphors, arrangements, and mappings, while real-data prototypes can expose data properties that prompt revisions to encodings, subjects, or directions.Sparsity, missing values, and unexpectedly dominant marks affected what representations could express and how they appeared.Source: §5.3 · p. 19

5.4 Design Implications for Supporting Creative Visualization

• Design support should link emerging story notes to data checks and connect references, sketches, and selected data for editable trials.These connections could help designers assess whether data supports a story and try drawings with data while preserving the ability to edit them.Source: §5.4 · p. 19
• Reducing repetitive prototype work and retaining versions, mappings, and data selections could support comparison and revision; optional AI should remain controllable, inspectable, and nonessential.The proposed integrated workspace remains future work, with evaluation needed to assess whether it reduces repetitive work and supports meaningful comparison.Source: §5.4 · p. 20

5.5 Studying Inspiration as It Unfolds

• Process records reveal inspiration unfolding beyond finished accounts: blogs show acknowledged sources, interviews explain adaptation, and live observation captures discarded connections.Extending observation to ongoing professional projects could capture inspiration arising during pauses and encounters beyond scheduled sessions.Source: §5.5 · p. 20

5.6 Limitations and Future Work

• Public or self-selected completed projects underrepresent brief explorations and abandoned work; Study III lacked client and delivery pressures, while physical-output insight rests on limited direct access.Study III also lacked a fixed deadline or completion requirement; physical-output understanding drew mainly on the corpus, two interviewees, and one textile-based concept.Source: §5.6 · pp. 20–21

6 Conclusion

• Ideas develop through data exploration, reference adaptation, and successive sketches and prototypes, shaped by production costs, practical constraints, and designers’ judgments.The findings motivate tools connecting stories, data, references, and visual drafts, with alternatives easier to test and revise.Source: §6 · p. 21

A Codebook

• The codebook names blockers such as poor readability, missing data, limited space, technology or material failures, skill gaps, and unsatisfying visuals.It also includes excessive data or workload, complex presentation, and difficulty combining visuals.Source: A Codebook · p. 24
• Data-related codes cover collecting records from public sources, clients, or the author, then exploring, analyzing, selecting, annotating, and preparing data.Exploration may include basic charts or reading records directly; preparation includes cleaning, restructuring, verifying, or hand-assembling data.Source: A Codebook · p. 25
• Other codes capture outside input and inspiration, including client or expert feedback, seeking knowledge, others’ work, metaphors, and features of the data.Inspiration also includes brainstorming, personal preference, physical properties, and forms noticed in intermediate artifacts.Source: A Codebook · pp. 25–26
• Responses to blockers include simplifying scope, reframing the problem, seeking knowledge or help, changing materials or techniques, and testing alternatives.The codebook also records visual and mapping iteration, adding or filtering data, and abandoning or revisiting directions.Source: A Codebook · p. 27

• Explanatory aids add axes, legends, labels, guides, or introductions to help viewers read a piece.Source: A Codebook · p. 27
• Interaction codes what viewers can do or how a piece responds, while material or medium choice records what it is made of or delivered in.Source: A Codebook · p. 28
• Visual ideas are coded as sketches, verbal descriptions, mockups, or physical prototypes, with mockups including drafts made in design software or code.Sketches may use pen and paper or a digital pen; verbal descriptions have no sketch, draft, or prototype, while physical prototypes externalize an idea as a mock-up or test build.Source: A Codebook · p. 28