§3.2

The Chart Atlas

The chart atlas is not a gallery. It is a translation guide, and the thing it translates is a sentence: this chart compares ___ with ___. Section 3.1 argued that the comparison, not the chart, is the real decision. This section is the vocabulary that decision draws on — twenty-six visual forms, each tied to the managerial question it answers, the finding it produces in real data, and the way it can be made to overclaim. A histogram, a line chart, an indexed line, a heatmap, and a coefficient plot are not interchangeable styles. Each answers a different question and carries a different risk.

The executive question: which visual form matches the question I actually have?

The atlas is organized by evidence family rather than by chart name, because a manager does not arrive knowing she needs a lollipop chart. She arrives with a question about spread, or ranking, or change over time, and the family is what routes her. Figure 1 is the routing table.

Figure 1. Eight evidence families. Find the row whose question sounds like yours, then use the family filter in Figure 2 to see only those forms. The names of the charts matter far less than knowing which row you are in.
FamilyThe question it answersForms in the atlas
DistributionWhat does the spread look like — is the average safe to use?Histogram, density curve, box plot, strip plot
ComparisonWhich category is largest, and by how much?Sorted bar, dot / lollipop, grouped bar, stacked and 100% bar, slopegraph
TimeWhat changed, when, and starting from where?Line, indexed line, area and stacked area, small multiples
RelationshipDo two things move together, and is the pattern clean?Scatterplot, bubble plot
GeographyDoes location itself change the decision?Choropleth, tile map
MultivariateCan I scan many pairs or many cells at once?Correlation matrix, heatmap, treemap
UncertaintyIs this difference large enough and precise enough to act on?Interval plot, coefficient / forest plot
Business BridgeHow do components add up to the number leadership tracks?Pareto, waterfall, funnel

Every miniature chart in the atlas is drawn from the real case data for this part of the book, not from screenshots or invented numbers, so that chart choice is learned from evidence rather than from style guides. Soup carries the Part II through-line — seasonality, price, volume, share. County supplies cross-sectional and geographic examples. Zillow supplies long state time series. Small teaching arrays appear only where the form is a managerial accounting bridge with no natural field in the three datasets; those cards are badged so the distinction stays visible.

The atlas

Read each card from top to bottom:

  1. The family names the evidence problem the form is built for.
  2. The chart is live — hover for values, and where two forms answer the same question, toggle between them.
  3. The finding states what this specific data example actually reveals.
  4. Use / Question / Trap say when the form should and should not be reached for.
  5. The link, where one appears, opens the section of Part II that treats the form in depth.

Filter by family or search by question — typing "drop-off", "composition", or "over time" will find the form faster than remembering its name.

How to read the atlas

Start from the comparison, then choose the chart.

Each card moves from business question to visual form to misuse risk. Charts are live: hover for values, and where two forms answer the same question, toggle between them. Filter by family or search the questions below, and follow a card’s link when a form gets a full treatment elsewhere in Part II.

Soup panel

88,409 rows

2,042 stores, 2001-06 to 2006-12

County cross-section

3,111 counties

Demographics, votes, density, region, and state geography

Zillow time series

51 states

2000-01 to 2026-04

Atlas scope

26 chart forms

8 evidence families, several interactive

Soup

88,409 store-month rows from 2,042 stores; the Part II through-line for seasonality, price, volume, and share.

County

3,111 counties for cross-sectional distributions, relationships, maps, intervals, and coefficient previews.

Zillow

51 state time series from 2000-01 through 2026-04 for lines, indexes, and heatmaps.

Teaching

Small invented arrays only where the chart is a business bridge form rather than a natural field in the three cases.

26 of 26

Distribution

Shape, spread, outliers, and typical units.

4 charts

Distribution

Histogram

Soup
Finding in this data: The median store-month sells 659 units, while the 95th percentile sells 5,476.

Use: Show the shape of one numeric variable with counts.

Question: Are most store-months small, or do a few stores dominate volume?

Trap: Do not compare many groups with overlapping histograms.

Soup: Progresso store-month volume is long-tailed.

Shape before summary → §3.4

Distribution

Density Curve

County
Finding in this data: The county median is 69.5%, higher than the mean of 66.1%.

Use: Show a smooth distribution shape when exact bins would distract.

Question: Where is the typical county in the vote distribution?

Trap: Do not imply precision in thin tails.

County: Trump vote share clusters heavily above 50 percent across counties.

Shape before summary → §3.4

Distribution

Box Plot

County
Finding in this data: South has the highest regional median county share; East has the lowest.

Use: Compare distributions across groups compactly.

Question: Which regions have higher or lower typical county vote shares?

Trap: Do not use it when stakeholders need individual values.

County: regional boxes reveal different medians and spreads.

Distribution

Strip / Jitter Plot

Illustrative
Finding in this data: Western store-months spread widest and carry the longest right tail; the Northeast is tighter.

Use: Show every observation when the distribution itself is the point and groups are few.

Question: How spread out are store-month volumes within each region, not just the average?

Trap: Do not jitter thousands of points into an ink-blob; sample or switch to a box plot at scale.

Soup: a regional sample of store-month volume, one dot per store-month.

Shape before summary → §3.4

Comparison

Rank or contrast a small set of categories.

5 charts

Comparison

Sorted Bar Chart

County
Finding in this data: Midwest averages 69.2%, versus 51.8% in East.

Use: Rank a small set of categories by one metric.

Question: Which region has the highest average county vote share?

Trap: Do not use unsorted bars when rank is the message.

County: regions sorted by average Trump vote.

Comparison

Dot / Lollipop Plot

Zillow
Finding in this data: Ohio rose 56.2% since January 2020; Texas rose 35.6%.

Use: Compare values precisely without heavy bars.

Question: Which states had the largest home-value increase since 2020?

Trap: Do not use it for continuous time paths.

Zillow: state home-value change from January 2020 to the latest month.

Comparison

Stacked & 100% Bar

Illustrative
Finding in this data: Core soups dominate every quarter, but premium and broth share rises noticeably in Q4.

Use: Show part-to-whole composition across a few ordered groups.

Question: Does the soup category mix shift across quarters, in units or in share?

Trap: Do not stack many categories; only the bottom segment is easy to compare across bars.

Soup: category units by quarter, toggled between absolute and 100% share.

Comparison

Grouped Bar Chart

Illustrative
Finding in this data: Side-by-side bars make Q4 the clear peak for every category, which stacking can hide.

Use: Compare a few categories within each group side by side.

Question: How do core, premium, and broth units compare within each quarter?

Trap: Do not group so many series that bars become too thin to read.

Soup: the same quarterly category data shown as clustered bars instead of stacked.

Comparison

Slopegraph

Illustrative
Finding in this data: Most regions edged up, but the East slipped slightly — a divergence a bar chart buries.

Use: Compare two points in time across categories and emphasize who moved.

Question: Which regions shifted between the two elections, and in which direction?

Trap: Do not use it for more than two time points; it stops being a slope.

County: average regional vote share in two elections, connected by a line per region.

Time

Order, timing, baselines, and growth paths.

4 charts

Time

Line Chart

Zillow
Finding in this data: The same housing cycle is visible, but dollar levels remain very different across states.

Use: Show how a metric evolves over ordered time.

Question: How did state home values move through the housing cycle?

Trap: Do not use it for unordered categories.

Zillow: California, Texas, Florida, and New York home value paths.

Baselines and indexes → §3.1

Time

Indexed Time Series

Zillow
Finding in this data: Indexing shows relative growth: Ohio leads the selected states after 2020.

Use: Compare growth paths with different starting levels.

Question: Which housing markets grew faster after 2020?

Trap: Do not read index values as dollars.

Zillow: each state equals 100 in January 2020.

Baselines and indexes → §3.1

Time

Small Multiples

Zillow
Finding in this data: The post-2020 surge appears in every panel, but amplitude and level differ sharply.

Use: Repeat the same chart across groups with common scales.

Question: Do housing cycles look similar across states?

Trap: Do not vary scales silently when magnitude matters.

Zillow: four state panels reveal timing and amplitude.

When faceting earns its space → §3.3

Time

Area / Stacked Area

Illustrative
Finding in this data: The stacked view shows the combined four-state market roughly tripling since 2000.

Use: Emphasize magnitude over time, or composition over time when totals matter.

Question: How large did each state market get, and how do they sum over time?

Trap: Do not stack area when readers need each series read precisely; only the bottom is honest.

Zillow: California alone as a filled area, then four states stacked.

Relationship

Two-variable patterns before model claims.

2 charts

Relationship

Scatterplot

County
Finding in this data: College share and county Trump vote have a strong negative correlation (r = -0.54).

Use: Show the relationship between two numeric variables.

Question: Do counties with more college graduates vote differently?

Trap: Do not call a slope causal without a design.

County: education share and Trump vote share.

Reading a slope before the equation → §3.4

Relationship

Bubble Plot

County
Finding in this data: The largest-vote counties cluster in higher-density space, where county Trump share is usually lower.

Use: Add magnitude to a two-variable relationship.

Question: Do the largest-vote counties sit in different density/vote space?

Trap: Do not let bubble area overwhelm the x-y comparison.

County: vote count sizes each point.

Geography

Spatial pattern only when place changes action.

2 charts

Geography

Choropleth / Tile Map

County
Finding in this data: The tile map makes broad regional pattern visible while deliberately hiding within-state county variation.

Use: Location is part of the decision.

Question: Where is the regional pattern strongest?

Trap: Do not map data just because a place field exists.

County: state tile map summarizes county vote patterns.

Geography

True Choropleth Map

Illustrative
Finding in this data: Growth concentrates across the Midwest and Southeast, not the highest-priced coasts.

Use: Location is part of the decision and the geographic shape itself carries meaning.

Question: Where did home values grow fastest since 2020, on a real US map?

Trap: Do not map raw counts on a choropleth; area distorts them — map rates or changes.

Zillow: state home-value growth since 2020 on an albers-usa projection.

Multivariate

Dense scans across two dimensions or many pairs.

3 charts

Multivariate

Correlation Matrix

County
Finding in this data: College share correlates -0.54 with county Trump share; white share correlates 0.52.

Use: Scan many pairwise relationships at once.

Question: Which demographic variables move together before modeling?

Trap: Do not treat correlation as effect size or causality.

County: demographics, density, income, and vote share.

Multivariate

Heatmap

Zillow
Finding in this data: The largest selected state-year jump is Arizona in 2021 at 26.2%.

Use: Show a matrix of values across two ordered dimensions.

Question: Which states and years show the biggest housing jumps?

Trap: Do not use color when exact values are the decision.

Zillow: annual home-value changes by state and year.

Multivariate

Treemap

Illustrative
Finding in this data: Core and premium soups fill two-thirds of the canvas; the long tail is visibly small.

Use: Show part-to-whole for many categories where rank and rough share matter.

Question: How is revenue split across product families at a glance?

Trap: Do not expect precise comparisons; area is read far less accurately than length.

Teaching data: revenue share across six product families as nested rectangles.

Uncertainty

Estimates plus the range that should qualify them.

2 charts

Uncertainty

Interval Plot

County
Finding in this data: The low- to high-density decile gap is about 34.8 percentage points.

Use: Show estimates with uncertainty in the same visual.

Question: Does the density-vote pattern look stable enough to discuss?

Trap: Do not hide the denominator behind a precise dot.

County: mean Trump vote by population-density decile.

What an interval is about → §3.4

Uncertainty

Coefficient / Forest Plot

County
Finding in this data: The largest standardized association is White share (+9.03 points).

Use: Summarize model estimates and intervals.

Question: Which predictors remain visually important after adjustment?

Trap: Do not show coefficients before explaining units and controls.

County: standardized regression preview.

What an interval is about → §3.4

Business Bridge

Managerial decomposition from components to action.

4 charts

Business Bridge

Pareto Chart

Teaching
Finding in this data: The top two product families account for two-thirds of the illustrative total.

Use: Show which categories contribute most of a total.

Question: Which product families explain most revenue?

Trap: Do not use it when categories are mutually ordered by time.

Teaching data: revenue concentration across product families.

Concentration, measured properly → §4.2

Business Bridge

Waterfall Chart

Teaching
Finding in this data: Discounts, margin, and trade spend turn a 100 list-price index into 61 net revenue points.

Use: Explain how components bridge from start to finish.

Question: What moved gross price to net revenue?

Trap: Do not use it for unrelated categories.

Teaching data: price bridge from list price to net revenue.

The dashboard arc → §4.1

Business Bridge

Funnel Chart

Illustrative
Finding in this data: The biggest drop is from price-eligible to test-launched — roughly half are filtered out there.

Use: Show sequential drop-off through an ordered pipeline.

Question: Where does the pricing-test pipeline lose the most stores?

Trap: Do not use a funnel for non-sequential categories; the narrowing implies an order.

Teaching data: stores moving from review to a confirmed price rollout.

The dashboard arc → §4.1

Business Bridge

Pie Chart (Trap exemplar)

Anti-pattern
Core soups: 42%42%Premium soups: 24%24%Meal kits: 13%13%Broths: 9%9%Seasonal: 7%Other: 5%
Core soupsPremium soupsMeal kitsBrothsSeasonalOther
Finding in this data: Try to order Broths, Seasonal, and Other by eye — the pie makes it a guess; a bar would not.

Use: Almost never for analysis; at most a single two-to-three slice part-to-whole.

Question: Which product family is biggest — and can you rank the rest by eye?

Trap: Do not ask readers to compare angles; rank and small differences are nearly unreadable.

Teaching data: the same revenue mix as the treemap, shown as six pie slices.

Figure 2. The chart atlas turns chart choice into managerial choice: each visual form is tied to a question, a real finding, and a misuse risk.

Choosing between near neighbors

Most real chart decisions are not "bar or map." They are choices between two forms in the same family that look interchangeable and are not. These are the four that come up constantly.

Bar or dot plot? Bars encode value as length from zero, so they need a zero baseline and they get visually heavy past a dozen categories. Dot plots drop the bar and keep the position, which lets you rank twenty or thirty items legibly and lets the axis start somewhere other than zero when the interesting variation is in a narrow band. Use bars when the magnitude relative to zero is the point; use dots when the ranking is the point.

Stacked or grouped bars? Stacking answers a part-to-whole question and answers it honestly for exactly one segment: the bottom one. Every segment above it starts at a shifting baseline, so comparing "premium soup across quarters" in a stacked chart is guesswork. Group the bars when readers need to compare a specific series across groups; stack them when the total and the rough mix are what matter; use the 100% variant only when the total is deliberately irrelevant.

Line or slopegraph? A line chart is for the path — the shape between the endpoints carries information. A slopegraph is for the move — two points in time, many categories, and the question is who rose and who fell. Using a line chart for two time points wastes the form; using a slopegraph for a monthly series destroys the seasonality.

Map or bar? A map is worth its considerable space only when location itself changes the decision — territory assignment, logistics, regulation, store siting. If the question is "which five states grew fastest," a sorted bar or dot plot answers it more precisely in a quarter of the room, because the eye reads position far more accurately than it reads color. And whatever the form, map rates and changes rather than raw counts: on a choropleth, area silently multiplies the count.

Saying no with the atlas

The atlas is at least as useful for declining a chart request as for filling one. A map is useful only when location changes the decision. A density curve is useful when shape matters more than counts. A correlation matrix is useful for scanning many relationships at once, and a poor executive answer when the decision needs one ranked comparison. A pie chart is in the atlas as an anti-pattern precisely so the argument against it can be made from the page rather than from taste: try to rank the three smallest slices by eye, then do the same on a bar.

Simple charts are not inferior. Mismatched charts are.

How the rest of the book reuses this page

Later chapters point back here instead of re-teaching chart basics. The pricing chapters use the scatterplot and coefficient-plot vocabulary to introduce elasticity. The causal chapters use interval plots and small multiples to explain effect heterogeneity. The machine-learning chapters reuse heatmaps, lift charts, calibration views, and residual displays. The AI-evaluation chapter borrows the same uncertainty discipline without pausing to define every form from scratch. Within Part II, three forms get their own treatment immediately: indexed lines as the baseline decision, small multiples as the heterogeneity test, and intervals and statistical charts as the boundary of what description can claim.