Exploring visualization support for early-career decision making

Student loans: Interview study

Our first case examines decision making on student loans. We conducted an interview study with students who were considering or had recently made borrowing decisions. The goal was to understand how individuals at an early stage of forming decision making strategies approach this process, what criteria they take into account, what challenges they encounter, and what tools they currently use, in order to identify opportunities for visualization support that could better assist student loan decision making.

Participants: We recruited 12 young students or recent graduates (6 female, 6 male), aged 20–25, through convenience sampling supplemented by snowballing until thematic saturation was reached. ⁹¹ Nine participants were pursuing Master’s degrees and actively using student loans, while three had recently started working and were about to begin repayment, reflecting our aim to involve real decision makers and capture the realities of occasional rather than habitual financial decisions. According to the authors’ institutional ethics review procedure, all studies reported in this paper were classified as low-risk and did not require a full ethics or privacy assessment.

Interview method and procedure: We used the critical incident technique, ⁸³ asking participants to recall the moment they determined the amount of their student loan. This was followed by semi-structured questions exploring their decision-making processes in greater detail. Participants were asked to describe the tools they used, their effectiveness, the key information they sought, the adequacy of its presentation, any frustrations they experienced, and their overall satisfaction with and desires for improvements in available loan support tools. Interviews were conducted in person after obtaining informed consent, each lasting approximately 15 min. The full interview protocol is available at the OSF link in the abstract.

Interview analysis and results: All interviews were audio recorded and transcribed. The transcripts were analyzed using inductive coding through qualitative content analysis. ⁹² One author identified themes relating to students’ loan decision making processes, existing tool usage, and expectations for visualization design, which were later reviewed and confirmed by another author. Key findings from this analysis are summarized below.

Students’ loan decision making process: Participants described a surprisingly simplistic process, typically limited to calculating the difference between income and expenses. For example, P11 said, “I calculated everything: tuition, rent, bills, food. I also factored in my part-time job earnings and my student grant. The money I was short, I decided to borrow” (R9 (Bracketed tags indicate which design requirements (Rs) and visualization goals (Gs) are grounded in corresponding Phase 1 insights (see Table 1)). Most treated their income and expenses as static values, rarely considering short-term changes or long-term consequences. None explored scenarios such as working additional hours, reducing expenses, or varying repayment periods (R2). Only two participants, when asked directly, mentioned interest rate effects as factors they later wished they had considered (R10). These patterns suggest that students tend to overlook both near-term flexibility and long-term implications when deciding how much to borrow (R10), which may partly explain overborrowing (see “Domain characterization” under “Student loans”).

Job seeking: Interview study

Our second case examines decision making on job seeking. We conducted an interview study with job seekers who were actively searching for jobs or had recently done so. The goal was to understand how individuals at an early stage of forming decision making strategies approach this process, what criteria they consider, what challenges they encounter, and what tools they currently use, in order to identify opportunities for visualization to better support job-seeking decisions.

Participants: We recruited six participants (four female, two male), aged 25–54, through convenience sampling until thematic saturation was reached ⁹¹ ; two held bachelor’s and four held master’s degrees. The sample included recent graduates entering the job market for the first time, unemployed individuals re-entering the workforce, and young employed participants, all actively seeking employment, reflecting our aim to involve real decision makers and capture the diverse realities of occasional rather than habitual job-seeking decisions.

Interview method and procedure: Similar to the student loan study, we used the critical incident technique, ⁸³ asking participants to recall a time when they were actively searching for a new job, particularly a moment when they felt lost during the process. Additional semi-structured questions explored their decision making in greater depth, including the job-search context, the quantitative and qualitative factors considered, the tools or resources used, their effectiveness and satisfaction, and desired future support or functionality. Interviews lasted 45–60 min and were longer but fewer than in the loan study, likely reflecting the broader and more personal nature of job-seeking discussions. Participants often provided extensive context about aspirations and prior experiences, resulting in richer narratives that nonetheless appeared to reach saturation relatively quickly, as contextual variation did not substantially alter the main observations. We also acknowledge that, beyond domain-related factors, perceived richness and saturation may have been influenced by differences in interviewer style and interpretation, as the job-seeking interviews were conducted by a different author. This variation aligns with reflexive qualitative practice, ⁸¹ which recognizes researcher perspectives and contextual interactions as integral to interpretive validity rather than biases to be eliminated.

Interview analysis and findings: All interviews were audio recorded, transcribed, and analyzed using inductive qualitative content analysis, ⁹² following the same procedure described in the student loan study (Phase 1). Key findings are summarized below.

Job-seeking decision making process: Unlike the student loan context, participants considered a broader mix of qualitative and quantitative criteria (R1), weighing richly categorical aspects (e.g. team fit, culture, values) alongside numeric factors (e.g. salary, distance; R5). Preferences ranged from binary deal-breakers to graded judgments (e.g. P6 contrasted yes/no for “usefulness” with continuous scale for “suitability” and “realistic expectations”). Constraints were often applied early (e.g. distance, salary), while values and fit served as key differentiators later (R2). For example, P4 emphasized “the vibe in the company” and described team composition as a cutoff; P6 rejected otherwise attractive roles on moral grounds.

Access to current job-seeking tools and support: Similar to the student loan context, no participant reported using a dedicated visualization or decision-support tool. Most (4/6) relied on job platforms (e.g. LinkedIn, Glassdoor) or recruiters for discovery and basic filtering, but support diminished when structuring criteria, comparing alternatives (R3), or tracking evolving judgments (R7). In response, several participants created their own structured resources (e.g. Google Sheets) to compare postings and record notes (R7, R8), echoing decision makers’ preference for tabular layouts^20,85 and prior work on personal data construction.^19,88

Desired features for visualization support in job-seeking: Unlike student loan participants, who appreciated straightforward recommendations, job-seeking participants emphasized intuitive interfaces that manage complexity while preserving control. They wanted support for adjusting priorities (R5), understanding “why A over B?” (R6), and carrying priorities across stages of the search process, from discovery to shortlisting to interviews, noting that priorities evolve as values and cultural fit become more salient (see Table 1). They requested comparison views that support both hard constraints (R2; e.g. P4 treated team composition as a binary cutoff when comparing similar roles) and graded trade-offs (e.g. P6 distinguished categorical vs scalable criteria (R1) and asked for workflows enabling side-by-side comparison (R3) and progressive narrowing (R1). Participants also sought mechanisms to quantify qualitative aspects without over-constraining them (R4), such as reordering criteria or adjusting importance weights. In addition, they valued practical signals such as source links, recency status (e.g. “mostly recently posted,” P3), and application counts, and requested tracking integrated into the same evaluation workspace (R11). Finally, they asked for free-form profiling and reflection (e.g. post-interview notes like “team was fun; strong growth potential,” P4) to keep personal goals visible during comparison (R7, R11).

Synthesizing design requirements and visualization goals

To make the connection between Phase 1 findings and probe design explicit without committing to domain-specific implementations, we structure the synthesis as a progression from concrete observations to abstract requirements and goals, followed by re-concretization in domain-specific probes. We distinguish four levels: (i) descriptive, domain-specific Phase 1 insights (reported in Student loans: Interview study and Job seeking: Interview study); (ii) normative, domain-agnostic design requirements (R; see Table 1); (iii) technical, domain-agnostic visualization goals (G; see Table 1); and (iv) domain-specific probes described in Probe for student-loan decisions and Probe for job-seeking decisions, which instantiate the abstract goals through concrete interface components.

The design requirements are grounded in two complementary sources: empirical observations from the Phase 1 interviews (see the corresponding interview sections) and structural characteristics of the respective domains (see Domain characterization under Student loans and Job seeking). Some requirements stem directly from participants’ expressed needs (e.g. side-by-side comparison, priority adjustment, scenario exploration). Others arise from observed mismatches between current decision practices and structural demands of the domains (e.g. long-term, interest-sensitive consequences in student loans or evolving elimination strategies in job seeking). We therefore treat the requirements as analytic abstractions synthesizing lived practice and domain constraints rather than verbatim feature requests.

We group the design requirements into three higher-level visualization goals reflecting recurring dimensions of decision support across both domains: (G1) unified and flexible multi-criteria comparison workspace, (G2) expressive and inspectable preference articulation, and (G3) constructive and contextual decision modeling (see Table 1).

Visualization probes

The following visualization probes are instantiations of the design requirements (Rs) and visualization goals (Gs) synthesized from Phase 1 insights and domain characteristics (see Table 1). Requirements under G1 were operationalized as persistent comparison workspaces with alternatives as rows and criteria as sortable, filterable, and inspectable columns. Requirements under G2 were operationalized through mechanisms for explicit preference articulation and ranking rationale, including weighted aggregation, visible criterion-level contributions, and non-equidistant scoring for categorical attributes in the job-seeking probe. Requirements under G3 were operationalized through functionality for constructing, reshaping, or contextualizing the decision space, including loan-plan creation, repayment trajectories, historical interest-rate exploration, job-search goal statements, and criterion-specific notes. Next, we describe the probes, which serve as elicitation instruments in Phase 2 to examine how non-expert decision makers compare alternatives, articulate preferences, and reason about consequences using structured visualization support.

Probe for student-loan decisions

We introduced a visualization probe (Figure 1) to operationalize the requirements identified for the loan domain, particularly long-horizon reasoning, alternative construction, and transparent comparison of financial consequences. The probe supports iterative what-if exploration of repayment plans while making temporal trajectories, interest sensitivity, and trade-offs between monthly and total costs directly inspectable.

OPEN IN VIEWER

Figure 1.

The visualization probe for the loan study consists of two coordinated components: an input widget on the left (a) and a visualization widget on the right. Before reaching this view, users complete short financial forms (not shown) specifying income and expense details, which are carried over into the interface. In the input widget, users can adjust parameters such as current debt, monthly borrowing amount, loan duration, repayment start date, and repayment rate. A short narrative summary explains how each change affects monthly obligations and total repayment. The system suggests a baseline monthly loan amount based on the entered data, while also allowing users to configure early repayment. Each configuration can be added as a new loan plan, represented by a distinct color consistent across all coordinated views. The visualization widget integrates three linked components, (b) the repayment chart, showing total outstanding debt over time for all active plans, (c) the interest explorer, displaying historical student-loan interest rates and allowing interactive adjustment via a draggable handle or numeric input, and (d) the loan comparison table, a bar-encoded tabular visualization listing all user-generated plans with attributes such as interest rate, monthly loan amount, monthly repayment, total paid, interest proportion, and final repayment date.

Specifically, the student loan visualization probe consists of:

Loan comparison table (R1, R2, R3, R6): (Figure 1(d)) Serves as the central view for inspecting and comparing multiple loan plans, implemented as an interactive, bar-encoded tabular visualization, ⁹³ known for its effectiveness in decision making^20,85,93 (see also the review of existing MCDM tools in the related work section; R3). Each row represents a user-generated loan configuration, and columns encode attributes such as interest rate, monthly loan amount, monthly repayment, total amount paid, interest proportion, and final repayment date (R1). Bars encode magnitude while preserving exact numerical values (R1). The table supports sorting by individual attributes, range filtering through sliders in column headers, and brushing across multiple columns to identify plans meeting compound conditions (e.g. low total payment and short repayment time; R2, R3, R6).

Input widget (R6, R7, R8, R9): (Figure 1(a)) Displays income and expense data carried over from earlier forms and suggests a baseline monthly loan amount when expenses exceed income (R9). Users can specify their current debt, desired monthly loan amount, and loan duration (R7). Repayment settings include (i) a slider defining the delay between graduation and the start of repayment, and (ii) radio options for either the default 35-year term or a custom duration (R7). The system automatically computes the corresponding monthly repayment (R6). By selecting Add to comparison table, users can save the current configuration as a distinct loan plan (R8).

Repayment chart (R3, R10): (Figure 1(b)) Displays total outstanding debt (y-axis) over time (x-axis) for all active plans (R10). This view allows users to compare repayment trajectories, identify payoff timelines, and observe how interest-rate changes affect debt evolution (R3,R10).

Interest explorer (R3, R10): (Figure 1(c)) Displays historical student-loan interest rates from 2000 to the present (R10). A black horizontal handle encodes the current interest rate for a plan before it is added (R10). Users can drag this handle up and down to set the rate or type a value in the paired numeric field (R10). When plans are added, each plan’s chosen rate is displayed as a colored horizontal line, enabling comparison to both historical data and other active plans (R3, R10). This direct manipulation makes the interest-rate control explicit and discoverable, supporting users in relating their choices to probability intuitions grounded in historical trends (R10).

Together, the components form a coordinated workspace for constructing (R8), comparing (R3) and evaluating alternative repayment scenarios (R2, R3, R10). A consistent, color-blind–safe palette preserves visual correspondence across views, enabling users to track each loan plan throughout the interface (R3). All components are dynamically linked, changing repayment settings, income and expenses automatically recompute the current plan and propagate updates to the repayment chart, comparison table, and interest explorer (R6, R7, R8, R10). This design supports iterative what-if exploration and transparent inspection of temporal consequences (R6, R10). Collectively, the probe operationalizes a persistent comparison workspace for alternative construction and trade-off analysis, while supporting uncertainty-aware reasoning through temporal projection.

Probe for job-seeking decisions

We introduced a visualization probe (Figures 2 and 3) to operationalize the requirements identified for the job-seeking domain, namely mixed-type comparison, qualitative preference articulation, and iterative narrowing of alternatives. The probe provides a persistent workspace for comparing self-assembled job alternatives while enabling the integration of categorical variables into structured aggregation.

Job comparison table (R1, R2, R3, R5, R6, R11): (Figure 2(d)) Implements a LineUp-based view ³ for ranking (R5) and exploring multiple job alternatives across user-defined criteria (R1). Each row represents a job listing and each column a criterion (e.g. salary, location, flexibility, team fit; R1). Attributes are encoded as normalized bar lengths with precise values accessible on demand (R1). Users can sort by any attribute to perform criterion-wise comparison and prioritize a single dimension when needed (R3, R5). Users can reorder and group columns to structure the workspace around currently salient criteria (R6). Users can resize columns to allocate attention to key criteria during inspection (R5). Users can adjust criterion weights through header controls to articulate and revise priorities (R5). The aggregate score is shown as a stacked bar to make per-criterion contributions explicit and support “why A over B” rationale (R6). Filtering and brushing support applying constraints to narrow alternatives (R2). All interactions update rankings in real time to provide immediate feedback about trade-offs under revised priorities (R5, R6). Changes from the previous ranking are shown in an adjacent view, where connecting lines indicate rank shifts (R11). Animated transitions help maintain orientation when ranks change due to sorting or weighting.

Attribute scoring widget (R4, R5): (Figures 2(b) and 3) While LineUp ³ supports nominal attributes for grouping, coloring, and sorting, it does not natively allow categorical preferences to contribute directly to the aggregate score. Categorical values must be numerically pre-encoded before import, limiting users’ ability to express unequal or relative preference gaps (e.g. remote >> hybrid > on-site). The Attribute Scoring Widget extends LineUp with a non-equidistant scoring mechanism,^90,91 allowing users to map categorical levels to numerical values through direct manipulation (R4). Each category (e.g. on-site, hybrid, remote) is represented by a draggable point on a horizontal slider, whose spacing encodes perceived preference strength (R5). Adjustments are instantly synchronized with the job comparison table, integrating the resulting encodings into the aggregate ranking. This mechanism supports partial ordering, unequal preference gaps, and flexible re-positioning, enabling users to express nuanced categorical judgments and bridge qualitative reasoning with quantitative aggregation.

Attribute details view (R7): (Figure 2(c)) Lists all available attributes such as salary, skill fit, industry, and company size, and provides space for explanations and comments to help users reflect on how each criterion applies across job alternatives (R7)

Profile widget (R7,R11): (Figure 2(a)) Allows users to articulate and iteratively refine job-search goals in free text (R7). Participants can enter a brief description of what they seek in a new role (e.g. “remote UX position in a collaborative environment”), externalizing priorities that serve as a reference during evaluation (R7, R11).

OPEN IN VIEWER

Figure 2.

The visualization probe for job seeking shows four coordinated components: (a) The profile widget allows participants to define and iteratively refine their job-search goals by entering short free-text notes, (b) The attribute scoring widget enables interactive quantification of categorical variables using non-equidistant sliders (see Figure 3), (c) The attribute details view lists all available criteria and provides space for explanations or comments to support reflection on how each applies across alternatives, (d) The LineUp 3 comparison view ranks job alternatives across weighted attributes such as salary, location, and skill fit, supporting sorting, weighting, and aggregation, (e) shows an example of how a newly added categorical criterion (e.g., flexibility) is integrated into the aggregate ranking through non-equidistant attribute scoring (see Figure 3).

OPEN IN VIEWER

Figure 3.

Sliders used by job seekers to quantify preferences for “flexibility” levels (“on-site,”“hybrid,”“remote”). ⁸⁹

Together, the components establish a persistent comparison workspace (R1, R3) that supports iterative preference articulation (R5), structured aggregation (R6), and progressive narrowing of alternatives (R2). The LineUp-based comparison view provides the central workspace for inspecting (R3) and ranking job alternatives (R6), while the attribute scoring widget enables user-defined quantification of categorical preferences (R4, R5) that directly influence the aggregate ranking. The profile widget and attribute details view support the articulation of goals and reflective interpretation of criteria, helping users externalize and revise their priorities during decision making (R7). Real-time updates in the comparison table provide immediate feedback on how changes in weights (R5) or quantification of categorical attributes (R4) affect rankings and trade-offs (R6), while the adjacent rank-change view exposes changes in iterations (R11). Collectively, the probe operationalizes comparisons including mixed-type criteria and inspectable decision processes by integrating qualitative preference expression with quantitative aggregation in a unified workspace.

Student loans: Visualization study

This study invited participants to explore the loan visualization probe described in the previous section (shown in Figure 1). The goal was to elicit insights into how visualization tools can support student loan decisions, an occasional rather than habitual decision type where prior experience offers limited guidance. All user study materials are available via the OSF link in the abstract.

Participants: We recruited 11 young students or recent graduates (8 female, 3 male), aged 22–30, following the same sampling procedure based on thematic saturation and participant profile as in Phase 1, including 7 active borrowers and 4 recent graduates about to begin repayment.

Pre-study: Participants provided informed consent and received a short walkthrough of the visualization probe by the interviewer. They then completed a short training session based on a printed scenario describing a hypothetical student’s income and expenses. Using these values, participants performed five analytical tasks, such as filling in data, calculating total repayment, adjusting interest rates, and sorting or filtering loan alternatives, to ensure full understanding of the probe’s functions without implying decision strategies. All participants completed the tasks successfully on their first or second attempt, after brief clarification when needed.

Study task: Participants used the loan visualization probe to construct and select a student-loan plan aligned with their goals. They entered their current financial situation or recalled data from when they were borrowing, interacted freely with the probe (adding and modifying plans, adjusting monthly loan amount, interest rate, duration, and repayment), and narrated their actions and reasoning using a think-aloud protocol.

Post-study: Participants took part in a short semi-structured interview designed to elicit reflections on their decision-making process and their experience with the visualization probe. They were asked to comment on how specific components (such as the comparison table, sorting and filtering features, and the interest explorer) influenced their reasoning, what elements supported or hindered understanding, and which parts of the tool they would keep or change. The session concluded with a five-item financial self-efficacy questionnaire ⁸⁷ to assess how the tool affected participants’ confidence in managing and planning financial decisions. We used a seven-point Likert scale (item three reverse-scored, and the score computed as the item average).

Analysis: All study data were audio-recorded and subsequently transcribed. The decision task was analyzed using an insight-based evaluation, ⁸⁶ acknowledging that personal financial goals, risk tolerance, and long-term impacts make a single “correct” decision difficult to define. ⁹⁴ This approach allowed us to assess whether participants’ intermediate reasoning and justifications were coherent with their interactions and the visual evidence, rather than evaluating the objective correctness of their final loan choice. Insights were identified from the transcripts and categorized as singular (about one plan or attribute) or comparative (relationships across plans). Singular insights relate to specific attributes of a single loan plan, for example, “This loan plan has a monthly repayment amount of X,” whereas comparative insights discuss relationships between different loan plans, for example, “How an increase in interest rate affects the monthly repayment compared to a previous loan plan.” For each insight, we annotated the evaluation attribute (e.g. total paid, interest proportion, monthly repayment, final repayment date); for comparative insights, we also marked the alteration attribute (the parameter modified by the user, such as monthly repayment, loan amount, interest rate, or duration) and the evaluation attribute (attributes used to make decisions about loans). Insights describing basic tool functionality (e.g. “This slider changes the interest rate”) were excluded since comprehension was verified in the pre-study. Interaction logs (number of plans added, number of unique interest rates and monthly amounts explored) were derived from screen recordings. Responses to open-ended questions about participants’ experience with the probe were analyzed using inductive coding and qualitative content analysis, ⁹² consistent with the interviews in Phase 1.

Job seeking: Visualization study

This study invited participants to explore the job-seeking visualization probe described in the previous section (Figures 2 and 3). The goal was to elicit insights into how visualization tools can support job-seeking decisions, an occasional rather than habitual decision type where prior experience offers limited guidance. All user study materials are available via the OSF link in the abstract.

Methodologically, Phase 2 of the job-seeking study was shaped by insights from its preceding interviews, which showed that job seekers relied on qualitative, self-defined criteria (e.g. team culture, flexibility, career growth), created their own datasets rather than working with fixed variables, and preferred to retain control over their decisions instead of relying on automated recommendations. These findings motivated a different evaluative focus from the loan study. In the loan study, whose Phase 1 interviews revealed that participants relied on narrow, short-term calculations and overlooked long-term trade-offs, Phase 2 evaluated how visual comparison could foster more systematic reasoning through an insight-based analysis and self-efficacy measures. In contrast, the job-seeking study approximated an authentic decision experience in the employment context, involving participants in constructing the dataset used in the study and focusing on how they engaged with the newly introduced attribute-scoring mechanism. Translating inherently qualitative criteria into numeric values within a structured visualization can be counterintuitive, since such judgments are often made implicitly, in relative rather than absolute terms, and may lose nuance when expressed through numerical scales. This uncertainty made it important to center the evaluation on how participants perceived, rationalized, and enacted this quantification step, whether it supported or conflicted with their natural decision-making process, and how it shaped the way they articulated and justified their choices through the probe.

Participants: We recruited seven participants (five female, two male), aged 25–35, following the same sampling procedure and profile as in Phase 1, including two university students and five young professionals actively seeking employment, with two bachelor’s, four master’s, and one doctorate degree.

Pre-study: Each session began with informed consent and a short introduction to the study. Participants were asked to provide links to five job advertisements they had recently considered. During an initial interview, they collaboratively constructed a personal dataset by defining the criteria they used in their job search (e.g. salary, skill fit, flexibility, company values). These criteria were documented by the researcher and translated into a structured dataset of 20 positions, comprising the 5 listings supplied by the participant and 15 additional listings sourced from LinkedIn that matched their interests. For missing categorical values, placeholder entries were added based on the participant’s defined preferences. This dataset was then imported into the visualization tool used in the main study. Before beginning the main tasks, participants received brief training with the LineUp interface using an unrelated dataset (Forbes Top 2000) to familiarize them with sorting, filtering, grouping, and weighting operations.

Study task: Participants engaged in a two-stage think-aloud session designed to examine job-seeking decision making with visualization and the influence of quantifying qualitative criteria. In the first stage, participants defined their job-search goals within the probe and explored their personalized dataset to make an initial job selection, iterating freely with the LineUp table to compare and prioritize alternatives. In the second stage, they were introduced to the attribute-scoring mechanism, which enabled them to assign numerical values to qualitative factors. They then revisited the same dataset and reconsidered their earlier choices, now integrating both qualitative and quantitative information within a shared scoring scale. This structure allowed us to examine how quantification within the visualization shaped participants’ decision-making processes and their strategies for evaluating and refining job choices.

Post-study: After completing the tasks, participants took part in a short semi-structured interview focusing on their experience with the tool. Questions addressed the tool’s intuitiveness and clarity (e.g. “Which parts felt intuitive or unclear?”), perception of the scoring mechanism (“How did you feel about converting qualitative factors into numerical scores?”), and its perceived impact on their decision making (“Did the tool influence how you viewed or prioritized jobs before and after scoring?”). Sessions lasted approximately 30–45 min in total.

Analysis: Audio recordings from both the usability and post-study interviews were transcribed and analyzed using open coding. One co-author conducted the initial coding, which was reviewed by another. Codes captured participant reactions, feedback on the tool’s usability, and reflections on their decision-making process. Through iterative refinement, codes were grouped into broader themes, including intuitiveness and clarity of use, perception of the qualitative–quantitative scoring, patterns of decision making, and emotional or reflective responses. Unlike the loan study’s structured insight-based analysis, this phase adopted an inductive, interpretive approach to capture how participants evaluated the attribute-scoring mechanism and how it shaped their decision strategies.