Chapter 8: Layout, Composition, and Small Multiples

"At the heart of quantitative reasoning is a single question: Compared to what? Small multiple designs, multivariate and data bountiful, answer directly by visually enforcing comparisons of changes, of the differences among objects, of the scope of alternatives." — Edward Tufte, Envisioning Information

Every chapter so far in Part II has assumed that the output is a single chart. A single decluttered bar chart. A single line chart with a well-written action title. A single scatter plot with one annotation calling out the key point. These single-chart examples are the right place to build the foundational skills — decluttering, typography, annotation — because a single chart is the smallest meaningful unit of visualization practice.

But most real-world visualization work does not stop at a single chart. A business report includes six charts across four pages. A dashboard shows eight charts on one screen. A news article embeds three charts supporting three different parts of the story. A scientific paper has a figure with four panels, each showing a different aspect of the experiment. A climate report arranges twelve small country maps on a single page to let the reader compare warming across regions. In each of these cases, the question is not "how do I make this chart?" but "how do I arrange these charts on the page so they work together?"

Composition — the arrangement of multiple visual elements into a coherent whole — is the subject of this chapter. It is a different skill from single-chart design, and it is the skill that separates practitioners who produce one good chart at a time from practitioners who produce coherent visual arguments across multiple charts. The difference matters because the reader's experience of a report, a dashboard, or an article is not a sum of individual charts — it is a single integrated impression of the whole page, and that impression depends as much on the arrangement as on the individual charts.

The central idea of this chapter is the small multiple: a set of charts that share the same visual encoding, the same scales, and the same design grammar, differing only in which slice of the data they show. Edward Tufte famously called small multiples "the best design principle" for information visualization, and most modern data journalism has vindicated his claim. When you need to show the same relationship across many groups, the same trend across many categories, the same pattern across many time periods, the same distribution across many subgroups — the right answer is almost always small multiples. A single-panel chart trying to cram everything into one picture usually fails; a small-multiple layout succeeds because it lets the reader compare by looking at adjacent panels that share a common visual language.

This chapter covers small multiples in depth, but it also covers the broader craft of composition: the application of Gestalt principles to layout, the choice of aspect ratios, the principles of reading order, the distinction between small multiples and dashboards, and the question of how to build a visual hierarchy when a figure has multiple panels. By the end of the chapter, you will know how to arrange multiple charts on a page so that they tell a coherent story, not just sit next to each other.

No code in this chapter. The matplotlib subplots(), GridSpec, and constrained_layout APIs are waiting for you in Chapter 13, where we cover multi-panel matplotlib figures in detail. This chapter is about the design principles that will guide your choices when you reach those functions. As with every chapter in Part II, the principles outlast the library — they apply whether you implement them in matplotlib, seaborn's FacetGrid, Plotly's subplots, Altair's faceting, or a dashboard framework like Dash or Streamlit.

8.1 Composition: Why Layout Matters

What Composition Is

Composition is the arrangement of visual elements — charts, text, titles, annotations, logos, whitespace — on a page or screen. It is what turns a collection of individual elements into a coherent whole. A document with five separate charts on five separate pages has no composition; a document with five charts carefully arranged on a single page has composition, and the quality of that composition affects how the reader experiences the document.

Composition is usually invisible when it is done well. The reader does not notice that the charts are aligned, that the whitespace is generous, that the hierarchy is clear, that the reading order flows naturally. The reader only notices that the document feels "polished" or "professional." What they are experiencing is the cumulative effect of a hundred small compositional decisions that the designer made in service of their reading experience.

Composition is also painfully visible when it is done badly. A page with charts of random sizes scrambled in no particular order, with inconsistent typography across panels, with misaligned elements and cramped whitespace, feels cluttered, rushed, and amateurish — even if each individual chart on the page is correct. The reader's impression of the document is dragged down by the bad composition, and the credibility of the underlying analysis suffers.

The lesson: composition matters enough that you should pay attention to it, even when it feels like "just layout." The reader's experience of your work depends on both the charts and the arrangement of the charts, and the arrangement is a design decision that deserves deliberate thought.

The Four Jobs of Composition

Composition does four specific jobs in any multi-chart figure or document:

1. Enable comparison. When the reader wants to compare two charts — to see whether two regions are performing differently, to see whether two years had different distributions, to see whether two treatments produced different outcomes — the composition is what makes the comparison possible. Aligned axes, consistent scales, adjacent placement: these are the compositional choices that let the eye move between panels and see differences at a glance. Without good composition, comparison becomes a mental computation that the reader has to perform explicitly, and most readers do not bother.

2. Establish hierarchy. Not all charts in a figure are equally important. Usually one chart is the hero — the main finding, the most prominent visual — and the others are supporting details that provide context, caveats, or zoomed-in views. Good composition makes this hierarchy visible: the hero chart is larger, more prominent, in the dominant position; the supporting charts are smaller, less prominent, in subordinate positions. The reader sees the hierarchy without needing to be told which chart to look at first.

3. Guide reading order. Multi-chart figures have an implicit sequence — the order in which the reader is expected to look at the panels. Western audiences read left to right, top to bottom, which means the chart in the top-left is seen first and the chart in the bottom-right is seen last. Good composition aligns the reading order with the logical order of the argument: context first, then the main finding, then the details. Bad composition scatters the panels in an order that fights the reader's natural scanning pattern.

4. Create visual unity. A figure with multiple panels should feel like one figure, not five disconnected images that happen to be on the same page. Visual unity comes from consistent typography across panels, consistent color choices, consistent line weights, consistent aspect ratios for similar chart types, and aligned edges that create a visible underlying grid. When a figure has visual unity, the reader perceives it as a single coherent artifact. When it does not, the reader perceives it as a collage.

These four jobs — comparison, hierarchy, reading order, visual unity — are the criteria against which any multi-chart composition should be evaluated. A composition that achieves all four is a good composition. A composition that fails at any of them is a candidate for redesign.

When Composition Is Not the Problem

Composition is a real skill, and it is worth developing, but it is not the answer to every problem. Sometimes the problem is not composition but content: a figure with three confusing charts does not become less confusing if you arrange the charts more beautifully. The charts themselves need to be clearer.

Before you invest in compositional polish, apply the Chapter 6 and Chapter 7 disciplines to each individual chart. Declutter it. Give it an action title. Add annotations. Format the axes. Make each chart stand on its own as a decluttered, self-explanatory, single chart. Only then is composition worth worrying about — because composition amplifies what is already there, but it does not rescue what is broken.

The analogy: composition is like the page layout of a magazine. A beautifully laid-out page with badly written articles is still a badly written magazine. Fix the articles first, then lay out the pages.

Check Your Understanding — Open a recent report or slide deck you produced that contains multiple charts. Evaluate it against the four jobs of composition: does it enable comparison between related charts, establish clear hierarchy, guide a sensible reading order, and create visual unity? Which of the four is weakest?

8.2 Gestalt Principles Applied to Layout

Chapter 2 introduced the Gestalt principles — proximity, similarity, enclosure, continuity, common fate — as laws of visual perception that describe how the human visual system groups elements into wholes. We looked at them in the context of single charts (pop-out effects, grouping of data points). In this section, we apply the same principles at the level of the whole page, where they govern how the reader perceives multi-chart figures as coherent compositions.

The application is not metaphorical. The same perceptual mechanisms that group red dots with red dots in a scatter plot also group aligned panels with aligned panels in a multi-panel figure. The visual system does not distinguish between "inside a chart" and "across multiple charts" — it perceives the whole visual field as one scene and applies the same grouping rules throughout.

Proximity: Charts That Belong Together Should Be Near Each Other

The proximity principle says that elements close to each other are perceived as related. In a multi-chart figure, this means the reader will automatically group charts that are near each other and treat them as "belonging together." Charts with more space between them are perceived as separate.

The implication for layout: put related charts near each other. If two charts tell two parts of the same story, place them close enough that the proximity conveys the relationship. If two charts tell unrelated stories, place them farther apart — or, better, put them on different pages. Uneven proximity (some charts clustered, others isolated) creates visual groupings that correspond to the conceptual groupings in your argument.

This sounds obvious, but the default behavior of most layout tools is to distribute charts with uniform spacing, which flattens the proximity signal. The reader sees four charts in a row, each the same distance from its neighbors, and has no cue about which charts go together. A better layout might put two charts close together, a small gap, then two more charts close together — visibly signaling "these two go together" twice.

Alignment: Shared Edges Create Visual Order

Alignment is the other side of proximity. When elements share a common edge — the left side of each chart lines up at the same x-coordinate, the top of each chart lines up at the same y-coordinate — the visual system perceives them as organized, related, and intentional. When edges do not align, the elements feel random and the composition feels sloppy.

For multi-chart figures, the most powerful alignment is the shared axis. When two charts share the same x-axis (same time range, same tick positions), the reader can compare values across the two charts at the same x-position without any mental computation. The alignment does the work of comparison. Without shared axes, the reader would have to match x-positions visually between the two charts, which is slow and error-prone.

Shared y-axes work the same way but are less common in practice. A pair of charts showing different quantities against the same time range usually have different y-axes (different units, different scales), but they should still have aligned y-axes in the sense that the plotting areas are the same size and the y-axes are drawn at the same x-coordinate. Visual alignment is the foundation of comparison, and alignment at the pixel level matters — a two-pixel mismatch between panel edges is visible to the reader as "something is slightly off."

The practical implication: when you lay out a multi-chart figure, align everything to a grid. Same left edge, same right edge, same top edge (for charts of the same height), same plotting-area dimensions. Default plotting libraries often do not enforce this alignment — each chart is drawn independently, and their edges fall where they fall. You have to override the defaults to enforce alignment, and the override is usually worth it.

Similarity: Same Design for Same Role

Similarity says that elements with the same visual characteristics (color, shape, size, texture) are perceived as related. In a multi-chart figure, this means charts with the same design grammar — same fonts, same colors, same line weights, same aspect ratios — feel like they belong together, while charts that look different feel separate.

The layout implication is that you should use the same design for every chart in a related set. If a figure has three time-series panels showing different variables, all three should use the same line weight, the same color scheme (or complementary colors), the same axis formatting, the same typography. If one panel uses a different style — a thicker line, a different font, a different color palette — the reader perceives it as different and may assume the difference has meaning that it does not.

This is why small multiples, which we discuss in Section 8.3, work so well: by definition, small multiples use the same design for every panel, differing only in the data. The similarity principle is automatically satisfied. The reader can compare across panels because the visual grammar is identical; only the data varies.

The similarity principle also explains why it is a mistake to mix chart types in a multi-panel figure. A figure with a bar chart, a line chart, a scatter plot, and a pie chart in four quadrants is visually chaotic, because the four chart types have different visual grammars and the reader's perceptual system cannot find a unifying structure. If you need different chart types to tell your story, consider whether they belong on the same page at all, or whether they should be separated into distinct figures.

Enclosure: Boxes Group What They Contain

The enclosure principle says that elements inside a shared boundary are perceived as grouped. Drawing a box around three charts groups them together visually, separating them from the charts outside the box. Subtle shading behind a set of charts has the same effect with a gentler visual impact.

Enclosure is a strong grouping signal. Use it sparingly, because overuse creates a cluttered, over-framed look. The best use of enclosure is to signal a high-level structure: all the charts in this box are "before the intervention"; all the charts in that box are "after the intervention." The reader sees two boxes, each containing a set of panels, and understands the grouping instantly.

Enclosure is also the principle behind margin and padding around a figure. A figure with generous whitespace around it is visually enclosed by the whitespace, distinguishing it from the surrounding text or adjacent figures. Cramming a figure against its neighbors violates the enclosure signal and makes the figure feel "mixed in" with things that should be separate.

Continuity: The Eye Follows Lines

Continuity says that the eye tends to follow lines and smooth paths. In layout, this means that aligned elements create implicit lines that guide the reader's eye. A row of charts with aligned tops and bottoms creates a horizontal band that the reader's eye scans across naturally. A column of charts with aligned left and right edges creates a vertical band that the reader's eye descends through naturally.

The continuity principle is what makes grid-based layouts feel ordered. The grid creates implicit alignment lines, the eye follows those lines, and the reader experiences the layout as structured rather than scattered.

The implication is that you should design your layouts around implicit or explicit grids. Even a two-chart figure benefits from having both charts the same width or the same height, because the shared dimension creates an alignment line the eye can follow. Three-chart figures should use a 1×3, 3×1, or 2+1 arrangement, each of which produces clean alignment lines. Four-chart figures can use 2×2, which is the most visually stable arrangement. More complex layouts (5, 6, 8, 12 panels) should still be based on a grid, usually 2×3, 3×2, 2×4, or 3×4.

Check Your Understanding — Sketch a rough layout for a figure with four panels: two time-series line charts showing temperature and CO2, and two bar charts showing emissions by sector for 1990 and 2024. How would you arrange them? Which Gestalt principles would you apply?

8.3 Small Multiples: Tufte's "Best Design Principle"

What Small Multiples Are

A small multiple (also called a trellis display, a faceted chart, or a panel chart) is a set of charts that share the same visual encoding — same chart type, same axes, same scales, same design — differing only in which slice of the data they display. Each panel shows the same kind of relationship applied to a different subset, and the reader compares the panels to see how the relationship varies across the subsets.

The canonical example, from Tufte's Envisioning Information: a grid of twelve small line charts, one for each month of the year, each showing the same variable (e.g., temperature) across the same hours of the day. The twelve panels share the same axes, the same line weight, the same color, the same design. The only difference between panels is which month's data they show. The reader can see at a glance how the temperature pattern differs between January and July, between November and May — because the panels are directly comparable.

Small multiples are powerful because they decompose complex comparisons into simple side-by-side readings. Instead of asking the reader to mentally split a single cluttered chart with twelve overlapping lines (a "spaghetti chart"), the small-multiple layout gives each line its own clean panel and lets the reader see the comparison by moving their eyes between panels. The comparison is the point, and the shared design enables it.

Tufte called small multiples "the best design principle" for information visualization, and while "best" is hard to defend for any single principle, the claim is not unreasonable. Small multiples have resolved more visualization problems than any other single design technique in the modern era. News graphics use them. Scientific figures use them. Dashboards use them. When a single chart cannot show the comparison cleanly, the answer is almost always small multiples.

When to Use Small Multiples

Small multiples are the right answer when:

1. You need to compare the same relationship across many groups. Temperature over time for 50 states. Revenue growth for 12 product lines. Vaccination rates for 30 countries. Each group gets its own panel; the reader compares by looking at adjacent panels. A single chart with 50 overlapping lines would be unreadable; the 50-panel small multiple is clean.

2. You need to show how a pattern changes across time periods. Monthly sales distributions for 12 months. Weekly case counts across 52 weeks. Yearly trajectories across 20 years. Each period gets its own panel; the reader sees the progression by scanning across panels.

3. You need to show the same variable at different scales or zoom levels. A full time series of stock prices, plus zoomed-in panels for specific events (the 2008 crash, the 2020 pandemic, the 2022 rate hikes). Each zoom level is a panel; the reader can see both the big picture and the details in a single figure.

4. You need to show the same chart with different subgroups highlighted. A chart of pandemic cases across all 50 states, repeated five times with a different state highlighted in each panel (the five states the report is focused on). Each panel uses the same background of gray state lines with one state colored in bright accent. The reader sees each highlighted state in context.

5. You need a consistent way to present many charts in a report. A quarterly business review with twelve pages of charts benefits enormously from using consistent small-multiple layouts rather than twelve unique chart designs. The consistency reduces the reader's cognitive load and creates a coherent report feel.

When Small Multiples Fail

Small multiples are not always right. A few failure modes:

1. Too many panels. A figure with 200 panels — one per zip code — is dense to the point of unreadability, even if each panel is individually readable. At some point, the number of panels overwhelms the reader's ability to scan them, and a different strategy (aggregation, ranking, interactive filtering) is better.

2. Panels that require different scales. If the twelve groups have wildly different ranges (one with values from 1 to 10, another with values from 100 to 10,000), forcing them onto a shared scale hides the variation in the small-range groups and cramps the large-range groups into a thin strip. Free scales (each panel scaled to its own data) solve the scaling problem but sacrifice comparability. There is no easy answer; the right choice depends on which you care more about: absolute comparison (shared scale) or relative pattern (free scale).

3. Too little data per panel. If each panel has three data points and the comparison across panels is essentially about those three points, a single chart with all the points on one axis may be more readable. Small multiples shine when each panel has enough data to show a meaningful pattern. With too little data per panel, the multiplication adds structure without adding information.

4. Heterogeneous chart types. Small multiples work only if every panel uses the same chart type. If your story requires a line chart for one variable, a bar chart for another, and a map for a third, you cannot use small multiples — you need a dashboard or a multi-figure document. The similarity principle from the previous section is violated.

Design Principles for Small Multiples

Good small multiples follow a few specific design rules:

1. Use the same scale across all panels unless there is a compelling reason not to. Shared scales enable direct comparison. Free scales should be used deliberately and labeled clearly so the reader knows they cannot compare magnitudes across panels.

2. Use the same chart type for every panel. No mixing of bar charts and line charts in the same small multiple. The visual grammar must be consistent.

3. Use a consistent color scheme. If a panel uses blue for data, every panel should use blue for data. If different panels highlight different elements, use color subtly to highlight only the differences.

4. Label the panels clearly. Each panel should have a short title identifying what subset of the data it shows (e.g., "California," "Texas," "Florida"). The title is usually above the panel in a small, unobtrusive font.

5. Keep the panel design simple. Small multiples work because the individual panels are simple. Do not try to pack each panel full of annotations, multiple series, or decorative elements. Each panel should be a minimal chart that shows one thing clearly.

6. Arrange panels in a meaningful order. Alphabetical order is rarely the best choice. Better orderings: by value (ranked from highest to lowest), by geography (states in rough spatial layout), by time (chronological), by category (grouped by family). The order should support the comparison the reader wants to make.

7. Provide a "guide panel" if needed. For very small panels, the reader may need help reading the chart. A single labeled panel at the top-left can serve as a guide, showing the reader what the axes and encoding mean. The other panels can then be unlabeled (or minimally labeled), because the guide panel provides the reference.

A Worked Example: The Climate Small Multiple

For the progressive project, we transform the climate plot from a single line chart into a three-panel small multiple. The three panels:

• Panel 1 (top): Global temperature anomaly, 1880–2024. Y-axis: temperature anomaly in degrees Celsius. Line in a warm color.
• Panel 2 (middle): Atmospheric CO2 concentration, 1880–2024. Y-axis: CO2 in parts per million. Line in a neutral color.
• Panel 3 (bottom): Global sea level, 1880–2024. Y-axis: sea level in millimeters relative to 1993 baseline. Line in a cool color.

All three panels share the same x-axis (year), aligned at the same x-coordinates. Each panel has its own y-axis with its own units — this is a "free y-axis, shared x-axis" design, which is appropriate because the three variables have different units but the same time range. The shared x-axis lets the reader see the temporal relationship: where CO2 rises, temperature follows; where temperature rises, sea level follows. The spatial alignment of the three panels makes the causal story visible without any annotation explicitly stating it.

An overall figure title runs above all three panels: "Three Measurements of a Warming Planet, 1880–2024." A subtitle identifies the data sources: "NASA GISS temperature; Mauna Loa CO2; CSIRO sea level." A source attribution runs along the bottom. Each panel has a small title ("Temperature Anomaly," "CO2 Concentration," "Sea Level") just above or to the left of the panel. No legend is needed, because each panel shows only one variable.

The result is a figure that tells the climate story across three related measurements. A single-panel chart trying to combine all three on one y-axis would be unreadable (three different units cannot share one axis honestly). A dual-axis chart would be dishonest for exactly the reasons Chapter 4 discussed. A small-multiple layout is the only solution that preserves the integrity of each variable while letting the reader see the relationships.

Check Your Understanding — Take a chart you have made or seen that shows "too many" series in a single spaghetti layout. Redesign it as a small multiple, one panel per series. How many panels do you have? What ordering would you use? What stays the same across panels?

8.4 Aspect Ratios: Wide, Square, and Tall

The aspect ratio of a chart — the ratio of its width to its height — is a design choice that most practitioners never think about. They accept the default from their plotting library, which is usually a slightly-wider-than-tall rectangle (matplotlib's default is 6.4 × 4.8 inches, or 4:3). The default is reasonable for many charts, but the right aspect ratio depends on the chart type and the data, and thinking about aspect ratio deliberately is one of the easier ways to improve a chart.

Wide for Time Series

Time series charts — line charts with time on the x-axis — usually look best in a wide format: the x-axis considerably longer than the y-axis. A wide format gives the time dimension room to breathe, lets the reader see the full temporal pattern without the chart feeling cramped, and matches the horizontal reading direction that most audiences use.

Cleveland's banking to 45 degrees rule provides a principled way to choose time-series aspect ratios: the average slope of the line segments in the chart should be around 45 degrees, because human perception is most accurate at distinguishing slope differences near 45 degrees. If the chart is too tall, the slopes appear too steep and all variations look similar. If the chart is too wide, the slopes appear too shallow and the variations become invisible. The 45-degree rule is a heuristic, not a law, but it gives you a principled starting point.

For practical purposes, a wide aspect ratio of roughly 3:1 to 2:1 (three times as wide as tall, or two times as wide as tall) is a reasonable starting point for most time-series charts. For very long time ranges (decades or centuries), the ratio can go wider. For short time ranges (days or weeks), the ratio should be closer to square.

Square for Scatter Plots

Scatter plots work best in a roughly square format. The reason is symmetry: the x and y axes represent two continuous variables that are typically equally important, and the reader is looking for relationships between them. A square plotting area gives both axes the same visual weight, and neither is emphasized over the other. A wide or tall scatter plot implicitly tells the reader that one axis is more important, which is usually not the intended message.

The exception is scatter plots where one axis has a much larger meaningful range than the other. If x values range from 0 to 100 and y values range from 0 to 3, a square chart will waste most of its vertical space on empty area above the data. A slightly wider (say, 3:2) aspect ratio in this case makes the data use the plotting area efficiently. But the square default is a reasonable starting point.

Tall for Rankings and Lists

Horizontal bar charts and dot plots that rank categories usually work best in a tall format: more height than width. Each category gets its own row, and the tall format gives you room for many rows without crowding. A wide horizontal bar chart with many categories has thin, short bars; a tall horizontal bar chart with the same categories has bars of readable width and clear labels.

The exception is when there are very few categories (fewer than five). With three or four categories, a tall chart feels empty, and a more balanced aspect ratio (closer to square) is better. The rule is: use tall when you have more categories than you have horizontal room.

Aspect Ratio in Small Multiples

For small multiples, the aspect ratio of each panel becomes a design decision with compounding effects across the figure. If you have twelve panels in a 3×4 grid, and each panel is wide (3:1), the whole figure becomes extremely wide. If each panel is tall (1:3), the whole figure becomes extremely tall. Neither extreme usually works well.

A good rule of thumb: the aspect ratio of each small-multiple panel should roughly match the aspect ratio of the overall figure, scaled by the number of panels. If the overall figure is 12 × 8 (wide), and you have a 3×4 grid, each panel should be 4 × 2 (wide). If the overall figure is 8 × 12 (tall), and you have a 3×4 grid, each panel should be 2.67 × 3 (close to square). Match the panel shape to the figure shape.

When the natural aspect ratio of the chart type (tall for rankings, wide for time series, square for scatter) conflicts with the aspect ratio that fits the figure, you usually have to compromise. Sometimes this means using a different chart type. Sometimes it means using fewer panels. Sometimes it means accepting a slightly non-optimal aspect ratio.

Check Your Understanding — Take three charts from your recent work: one time series, one scatter plot, and one ranked bar chart. What aspect ratio does each currently use? What aspect ratio would be better based on the principles in this section?

8.5 Reading Order: Z-Pattern, F-Pattern, and Hero Placement

How Readers Scan Multi-Panel Figures

When a reader encounters a multi-panel figure, their eye does not land on all the panels simultaneously. It starts somewhere and moves in a sequence. Understanding the sequence — and designing the figure to match it — is how you make the layout work with the reader's natural scanning pattern rather than against it.

For Western audiences (who read left-to-right and top-to-bottom), the dominant scanning patterns are the Z-pattern and the F-pattern.

The Z-pattern: The eye starts at the top-left, scans across to the top-right, drops diagonally to the bottom-left, and scans across to the bottom-right. This pattern is characteristic of image-heavy pages where the reader is scanning for visual content rather than reading text. Multi-panel figures, dashboards, and magazine layouts often get a Z-pattern reading.

The F-pattern: The eye starts at the top-left, scans across the top, drops down the left side (reading less as it descends), and occasionally sweeps across to the right for features that catch attention. This pattern is characteristic of text-heavy pages and lists.

Multi-panel figures usually get the Z-pattern. The top-left panel is read first. The top-right panel is read second (or nearly simultaneously). The bottom-left and bottom-right are read third and fourth. Panels outside this rough sequence — middle panels, off-grid panels — are read last, if at all.

Hero Placement

The implication is that the most important panel should be in the top-left quadrant of the figure. This is where the reader's eye lands first, and this is where you get their freshest attention. The panel in the top-left position should be the hero — the main finding, the most important chart, the panel you most want the reader to remember.

This is a strong principle, and it is often violated. Default layouts (including matplotlib's subplots(2, 2)) treat all four panels equivalently. The top-left is no more prominent than the bottom-right. If you accept the default, you are relying on the reader to find the hero panel on their own — and in a Z-pattern scan, they will find the top-left first regardless of which panel is actually the hero. The mismatch means the reader's attention is spent on whichever panel happens to be in the top-left, not on the one you wanted them to focus on.

The fix is to place the hero panel in the top-left deliberately, and to size the panels so that the hero is visibly larger or more prominent. A common pattern is the hero plus supporting layout: one large panel taking up the top half of the figure (the hero), with three smaller supporting panels across the bottom. The reader sees the hero first, absorbs its message, and then scans the supporting panels for context.

Matching Layout to Narrative Order

The reading order of the layout should match the logical order of the argument. If your argument goes "context → main finding → implications," the panels should follow that order:

• Context panel (top-left): Background, baseline, what the data looks like in the normal state.
• Main finding panel (top-right or top-center): The headline result, the thing the reader should remember.
• Implications panels (bottom row): What the main finding implies, what the context says about the finding, the subgroup breakdowns that confirm or qualify the main finding.

This sequence follows the Z-pattern naturally. The reader absorbs context, then sees the finding, then reviews the implications. A figure designed around this sequence reads intuitively because the narrative order and the reading order are the same.

If your argument has a different logical order — "main finding → context → qualifier" — lay out the panels in that order. The principle is: the sequence of panels should match the sequence of ideas. A figure whose panels are in a different order from the argument's logic will fight the reader's natural scanning pattern.

When All Panels Are Equal

Small multiples are an exception to the hero-placement rule. In a small multiple, every panel has the same importance — the point is to compare across panels, not to emphasize one panel over the others. In this case, the layout should treat panels equally: same size, same design, same margins, arranged in a regular grid. There is no hero; every panel is an equal part of the comparison.

For small multiples, the ordering decision is about which panels go where in the grid. The order should support the comparison: panels arranged by value (highest to lowest) let the reader see a ranking; panels arranged by time let the reader see a progression; panels arranged by geography let the reader see a spatial pattern. Alphabetical order is rarely the best choice because it rarely supports any meaningful comparison.

Check Your Understanding — For the climate small-multiple example from Section 8.3, does the three-panel arrangement (temperature, CO2, sea level) follow a meaningful order? What would change if the panels were reordered as CO2, temperature, sea level?

8.6 Dashboards vs. Small Multiples: Different Jobs

The Distinction

A small multiple shows the same chart type applied to different slices of the data. Every panel is the same kind of chart. The comparison is across slices of a single dataset or variable.

A dashboard shows different chart types applied to different aspects of a system. A typical dashboard might have a line chart of key metrics over time, a bar chart of performance by region, a table of top performers, and a pie chart of resource allocation. The charts are diverse because they answer different questions about the same system.

The distinction matters because the two layouts have different design rules. Small multiples depend on consistency — same scale, same chart type, same design — because the whole point is comparison across panels. Dashboards depend on diversity — different chart types for different questions — and the design challenge is making the diverse elements feel coherent despite their differences.

Designing Small Multiples

The rules for small multiples are covered in Section 8.3: same chart type, consistent scales, consistent design, meaningful ordering. The design discipline is high, but the rules are clear. Once you know you are making a small multiple, the design decisions largely follow from the principle of enforced comparison.

Designing Dashboards

Dashboards are harder. You have to make heterogeneous elements feel unified, which means establishing visual grammar that spans different chart types. Some principles:

1. Consistent typography across all panels. Same font family, same size hierarchy, same colors for text. This is the strongest unifying signal for a dashboard with diverse chart types.

2. Consistent accent color. Even if the chart types differ, using the same accent color across the dashboard creates visual unity. Blue bars, blue lines, blue highlights — the repetition of color ties the panels together.

3. Consistent whitespace and alignment. Every panel should have the same margin, the same padding, the same alignment to the grid. Uneven spacing makes the dashboard feel chaotic.

4. Clear visual hierarchy. A dashboard usually has a primary question (the thing you most want the viewer to monitor) and secondary questions. The primary chart should be larger or more prominent; the secondary charts should be smaller and supporting. Treat every chart as equal and you get visual flatness.

5. Group related charts together. Charts that answer related questions should be placed near each other, following the proximity principle. Charts that answer unrelated questions should be separated by whitespace or visual breaks.

6. Consistent time range across time-series panels. If the dashboard has multiple time-series charts, they should usually share the same time range so the reader can compare events across charts. A dashboard with one chart showing "last 30 days" and another showing "last year" forces the reader to mentally rescale, which is a cognitive cost.

7. Limit the number of panels. A dashboard with thirty panels is usually too many. The reader cannot monitor thirty things at once. Pick the most important five to ten and make those prominent; move the rest to a secondary view.

When to Use Which

Use a small multiple when: - You want to compare the same thing across many groups. - You have a single dataset with a clear grouping variable. - The comparison is the point.

Use a dashboard when: - You need to monitor or present several different aspects of a system. - The viewer needs different kinds of information for different questions. - You want to show a comprehensive snapshot of a complex state.

In practice, many real documents use both. A quarterly business report might have a dashboard summary at the front (diverse chart types showing different KPIs) and small-multiple appendices (one chart type per section, showing regional breakdowns). Knowing which technique to use for which job is part of the craft.

8.7 Bringing It Together: The Climate Figure as a Small Multiple

By the end of this chapter, the progressive climate project has evolved substantially. Here is the full evolution, tracing the design decisions from Chapter 1:

• Chapter 1: Ugly default matplotlib line chart. Single series. Default title, default colors, default everything.
• Chapter 6: Same chart, decluttered. Top and right spines removed. Gridlines lightened. Background cleaned. Still single series.
• Chapter 7: Same chart, with words. Action title stating the finding. Subtitle with context. Annotations on the 2016 and 2023 peaks. Source attribution.
• Chapter 8: New composition. Three panels showing temperature, CO2, and sea level. Shared x-axis. Each panel retains the declutter and typography discipline from Chapters 6–7. One figure title ties them together.

The final composition:

Overall figure title (action, left-aligned, 20pt bold): "Three Measurements of a Warming Planet, 1880–2024"

Overall subtitle (14pt regular, medium gray): "Global temperature anomaly, atmospheric CO2 concentration, and sea level relative to baselines."

Panel 1 (top, wide aspect ratio, most prominent): - Title: "Temperature Anomaly" - Y-axis: "Degrees Celsius above 1951–1980 average" - Line: warm red-orange - Annotation: "2023: +1.18°C — warmest year on record"

Panel 2 (middle, same dimensions as Panel 1): - Title: "CO2 Concentration" - Y-axis: "Parts per million (ppm)" - Line: medium gray or neutral color - Annotation: "2024: ~420 ppm, 50% above pre-industrial"

Panel 3 (bottom, same dimensions as Panels 1 and 2): - Title: "Sea Level" - Y-axis: "Millimeters relative to 1993 baseline" - Line: cool blue - Annotation: "Rising roughly 3.4 mm/year since 1993"

Shared x-axis: Year, 1880–2024, with ticks every 20 years. The three panels are aligned so that each year column is visually aligned — you can draw a vertical line from any year on the bottom panel up through the middle panel to the top panel and see the three variables at that year.

Source attribution (8pt, muted, bottom of figure): "Sources: NASA GISS (temperature), NOAA Mauna Loa (CO2), CSIRO (sea level). Last updated: December 2024."

The composition works because:

• Proximity groups the three panels as a single figure.
• Alignment (shared x-axis) enables temporal comparison across panels.
• Similarity (same chart type, same design grammar, same axis formatting) creates visual unity.
• Reading order (top to bottom: temperature → CO2 → sea level) follows the causal story.
• Aspect ratios match the chart type (wide for time series) and the figure shape.
• Hero placement puts temperature at the top, where it gets the most attention, because it is the variable most readers already know about.
• No dashboard-style diversity — every panel is the same chart type, telling the same kind of story about a different variable.

A reader looking at this figure for five seconds can understand: the world is warming (top panel), CO2 is rising in parallel (middle panel), and sea level is rising as a consequence (bottom panel). The three variables together tell a story that no single variable alone could tell, and the shared x-axis makes the temporal relationship visible without any explicit annotation stating it. This is what small multiples do at their best: they let the data show the comparison, rather than relying on the chart maker's narration.

Chapter Summary

This chapter covered the composition of multi-chart figures: how to arrange charts on a page or screen so that they work together as a coherent whole. The central idea is that composition is a real design skill distinct from single-chart design — a figure with five perfect individual charts, arranged badly, is still a bad figure.

The four jobs of composition are enabling comparison, establishing hierarchy, guiding reading order, and creating visual unity. A composition is evaluated against these four criteria.

The Gestalt principles introduced in Chapter 2 apply at the layout level just as they apply within a single chart. Proximity groups related panels; alignment creates comparison-enabling structure; similarity ties heterogeneous elements together; enclosure signals high-level groupings; continuity lets the eye follow alignment lines naturally. Good composition uses these principles deliberately.

Small multiples are the most powerful composition technique for multi-chart figures. A small multiple shows the same chart type applied to different slices of the data — same encoding, same scales, same design, different data. The power comes from enforced comparison: because every panel uses the same visual grammar, the reader can compare across panels without mental effort. Tufte's claim that small multiples are "the best design principle" is overstated, but not by much.

Aspect ratios matter more than most practitioners realize. Time series want wide aspect ratios; scatter plots want square ratios; ranked bar charts want tall ratios. Cleveland's "banking to 45 degrees" rule provides a principled starting point for time-series aspect ratio decisions.

Reading order follows the Z-pattern for most multi-panel figures: top-left first, then top-right, then bottom-left, then bottom-right. The most important panel (the hero) should be in the top-left, where the reader's eye lands first. The order of panels should match the logical order of the argument.

Dashboards and small multiples have different design rules. Small multiples depend on consistency across panels (same chart type, same scale). Dashboards depend on creating unity from diverse elements (consistent typography, accent color, whitespace, and alignment across different chart types). Knowing which technique fits which job is part of the craft.

The threshold concept is that comparison requires consistency. You can only compare panels that share the same visual encoding and the same scales. Violate the consistency, and the small multiple becomes a gallery of incomparable pictures. The discipline of maintaining consistency across panels is what makes small multiples work.

Next in Chapter 9 (the final chapter of Part II): storytelling with data. Once individual charts are clean (Chapter 6), self-explanatory (Chapter 7), and well-composed (Chapter 8), the final question is how to sequence them into a narrative that walks the reader through a complete argument. Storytelling is where the design principles of Part II meet the communication practice of data journalism and business communication.

Spaced Review: Concepts from Chapters 1-7

These questions reinforce ideas from earlier chapters. If any feel unfamiliar, revisit the relevant chapter before proceeding.

1. Chapter 2: Gestalt principles (proximity, similarity, enclosure, continuity, common fate) were introduced as laws of visual perception within a single chart. How does the chapter extend these principles to multi-chart layouts? Is the extension metaphorical, or is the visual system really applying the same principles at both scales?

2. Chapter 3: The chapter recommends using a consistent color scheme across panels in a small multiple. How does this recommendation interact with the color palette principles from Chapter 3 (sequential, diverging, qualitative)? Should small-multiple panels use the same sequential palette, or should the palette vary?

3. Chapter 4: Dual-axis charts were discouraged in Chapter 4 because they manufacture apparent correlations. Small multiples with free (different) y-axes are similar in that they put different-scaled variables in the same visual field. Are small multiples with free axes ethically equivalent to dual-axis charts? What is the difference, if any?

4. Chapter 5: The chapter argues that small multiples are "the best design principle" for many visualization problems. How does this interact with Chapter 5's chart selection framework? Is "small multiples" a chart type, a meta-strategy, or something else?

5. Chapter 6: Decluttering applies to each individual panel in a small multiple. But does the declutter principle also apply at the figure level? Should the whitespace between panels be minimized (more data per square inch) or generous (more visual breathing room)?

6. Chapter 7: Each panel in a small multiple needs a title, and the figure as a whole needs a title. How do these two levels of titles interact? When the panel titles are "California," "Texas," and "Florida," what does the overall figure title need to add?