Chart Selection and Common Visualization Traps

Chart Selection and Common Visualization Traps

Quick Answer: Choose charts by message and data type: scatter for correlation/relationship between two quantitative variables, bars for categorical comparison, lines for time series, tables when exact values matter, and use pies sparingly. Avoid dual-axis illusions, 3D distortion, truncated baselines, and overcrowded multi-series charts.

Competency 4.6 is incomplete if you only know “make it pretty.” CBDA scenarios frequently test whether you pick the right chart for the insight—especially when options include scatter vs bar vs pie, or when a trap chart flatters a weak story. This section is your decision tree and trap catalog.

Matching Chart to Message and Data Type

Start with the question the visual must answer, then the data shape.

Scatter for correlation (IIBA sample-style emphasis)

When the business question is whether two continuous measures move together—marketing spend vs revenue by market, wait time vs satisfaction, temperature vs demand—a scatterplot makes the relationship visible as a cloud of points. Pattern language stakeholders can grasp:

• Upward slope cloud → positive association,
• Downward → negative,
• Formless blob → little linear association,
• Clusters/outliers → segments or data quality issues to investigate.

Scatter does not prove causation. Pair with Domain 3 discipline: correlation is a finding about association, not automatic mechanism. Still, for displaying correlation, scatter beats a pie or a dual-axis bar illusion.

Bars for comparison

Bars excel when the eye must compare lengths across categories: products, regions, channels, risk tiers. Sort by value for “which is biggest?” questions; keep a meaningful category order (e.g., process sequence) when sequence is the point.

Use clustered bars for a few series; beyond that, prefer small multiples or a different structure rather than eight side-by-side colors.

Lines for time

Lines communicate continuity and rate of change. Best practices:

• Consistent time grain on the axis,
• Limited series (often ≤3–5 without faceting),
• Markers only when points are sparse or need emphasis,
• Annotations for interventions and external shocks.

Do not use lines to connect unordered categorical axes (product A–F) as if they were temporal—that invents a false continuum.

Tables when exact values matter

Tables win when stakeholders must:

• Look up a precise number (contractual SLA, invoice, compliance threshold),
• Compare many metrics per entity where position encodings would sprawl,
• Audit or reconcile to source systems.

Enhance tables with conditional formatting carefully (accessible colors, not color-only). For a board vote on a single KPI trend, a line still beats a 24-row monthly table. For finance sign-off on twelve account balances, the table is the product.

Pie pitfalls

Pies encode angle/area, which humans judge poorly for close values.

Acceptable: 2–5 slices, large differences, part-of-whole is the only message, labels include % and category.

Problematic:

• Many slices (legends with 15 colors),
• Ranking tasks (“which region is 3rd?”),
• Comparing two pies across time (harder than grouped bars or lines),
• Exploded 3D pies (distortion stacked on distortion),
• Slices not summing to a meaningful whole.

If the question is comparison or ranking, bars usually win.

Common Visualization Traps

Dual axes

Plotting revenue on the left axis and a satisfaction score on the right, then aligning scales so lines “move together,” can manufacture a story of linkage. Viewers assume shared meaning that does not exist.

Safer alternatives: two aligned charts with independent clear scales; index both series to a common base period; scatter if the unit of analysis is paired observations; explicitly teach that dual axis is advanced and easily abused.

3D charts

3D bars, pies, and surfaces add perspective distortion. Occlusion hides values; angles mislead. For CBDA business communication, prefer 2D. 3D is almost never the best answer unless the data themselves are inherently spatial surfaces—and even then, caution rules.

Truncated baselines (again, as a selection trap)

Even with the “right” chart family, a truncated bar axis can become the real message. Selection + scale must both be honest.

Too many series

A line chart with 18 products is a spaghetti plot. Nobody sees structure.

Remedies:

• Highlight one series; gray the rest,
• Top N + Other,
• Small multiples (one panel per category),
• Aggregate to a higher category first,
• Interactive filter with a smart default, not a wall of color.

Other frequent traps

Decision Flow for Exam Scenarios

When a stem asks which chart to use, run this quick filter:

1. What is the research question? Relationship, comparison, trend, composition, distribution, or exact lookup?
2. What is the data type? Two quantitatives → scatter candidate; time → line candidate; categories → bar candidate; need exact cells → table.
3. Who is the audience and medium? Slide vs dashboard vs report density.
4. Which option minimizes distortion? Eliminate 3D, dual-axis gimmicks, and overloaded pies.
5. Does the chart make the answer pre-attentive? If they must study a legend for minutes, redesign.

Worked mini-scenarios

A. “Is advertising spend associated with weekly sales across stores?”
→ Scatter (store-week points or store aggregates), optionally color by region. Not a pie. Not a dual-axis “spend line vs sales line” as the first choice for correlation.

B. “Which three product lines have the highest defect rate this quarter?”
→ Sorted horizontal bars. Table acceptable if many metrics per line; pie is poor for ranking.

C. “How did ticket volume change after the May release?”
→ Line (or column) over time with annotation at release. Dual axis with NPS only if carefully justified—and never as fake proof of causation.

D. “What is the contractual on-time % by carrier for invoice disputes?”
→ Table (exact values) possibly with bar for visual scan. Executives may still want a bar for the top dispute; legal/finance needs the numbers.

E. “Show share of revenue by five segments.”
→ Simple pie/donut or 100% bar; if comparing shares across years, grouped bars or small multiples beat two pies.

Integrating Chart Choice with Story and Ethics

Chart selection is part of Competency 4.5–4.6 together:

• The story sets the question and the fair comparison,
• The chart makes that comparison perceptually accurate,
• The narrative prevents overclaiming from a pretty pattern.

A scatter that shows weak association should not be captioned as “proves drivers of growth.” A bar chart of a non-significant A/B win should not use truncated axes to look like a landslide. Traps are often ethical failures wearing design clothing.

Scenario Walkthroughs

Scenario A — Correlation request from a VP

VP: “Show me that employee training hours cause higher NPS.” Data only support a cross-sectional association. Chart: scatter of training hours vs NPS by team, with clear title “association, not proven causation,” and narrative that experiments or longitudinal designs would be needed for causal claims. Choosing a dual-axis line that forces the two series to track visually would be a trap.

Scenario B — Too many series in a product review

Product review deck has one line chart of weekly active users for 22 SKUs. Replace with small multiples for strategic SKUs, or a ranked bar of change, or highlight top movers. Spaghetti is not “comprehensive”—it is unreadable.

Scenario C — Pie for ranking

A pie of 11 regions is used to find the third-largest. Switch to a sorted bar. If part-of-whole is still needed, add a single “share of total” column in a table beside the bars.

Exam Focus Checklist

Prefer the option that:

1. Matches intent (scatter for correlation, bar for comparison, line for time, table for exact values),
2. Avoids known traps (dual axis illusion, 3D, truncated bars, pie overload, series overload),
3. Serves the audience and decision,
4. Keeps encoding honest and accessible,
5. Supports a narrative that states what the chart does—and does not—prove.

Chart selection is a high-frequency Domain 4 skill: treat it as decision design, not decoration.