Case Study 1: Minard's March on Moscow

Case Study 1: Minard's March on Moscow

The most praised data visualization in history was not praised for being pretty. It was praised for answering six questions at once — and for showing, in a single image, exactly why the chart type you choose must match the questions you are asking.

The Situation

In 1869, a retired French civil engineer named Charles Joseph Minard published a small chart in a pamphlet about the transportation of goods. He had spent his career designing maps of commerce — grain shipments, coal flows, wine exports — and had developed an unusual instinct for encoding quantitative information directly onto geographic space. The chart in the 1869 pamphlet was a departure from that commercial work. It depicted not trade but war: Napoleon Bonaparte's catastrophic 1812 invasion of Russia, and the retreat that destroyed his army.

The history was known. The invasion had begun in June 1812 with a Grande Armée of over 400,000 soldiers crossing the Niemen River into Russian territory. The army pushed east through scorched Russian countryside, engaged in the bloody Battle of Borodino, and reached Moscow in September — only to find the city abandoned and largely burned by its own inhabitants. Without winter quarters, without supply lines, and without a negotiated peace, Napoleon ordered a retreat. The return march from Moscow to the Niemen, in the brutal Russian winter, turned into one of the most devastating military catastrophes in European history. Of the 400,000-plus soldiers who had crossed into Russia, fewer than 10,000 returned.

The story was told and retold in military histories, memoirs, and popular accounts for fifty years before Minard picked up his pen. What Minard added was not new facts. What he added was a chart — one chart, on one page — that answered so many questions at once that Edward Tufte would later call it "probably the best statistical graphic ever drawn" (The Visual Display of Quantitative Information, 1983). The praise has been contested, celebrated, and quoted for four decades since, but no one has seriously proposed a replacement. Minard's chart remains the archetype of a visualization that does exactly what this chapter argues charts should do: match its form to the questions being asked.

The Data

The underlying data for Minard's chart was a compilation of military records, historical accounts, and weather observations. For each stage of the campaign, Minard needed:

The geographic route — the path the army took across Russia, from the Niemen River eastward toward Moscow and back
Army size — how many soldiers were present at each point along the march, from the initial 422,000 down to the final survivors
Temperature — the air temperature during the retreat, measured in degrees Réaumur (a historical temperature scale) on specific dates
Time — the dates corresponding to the army's positions during the retreat
River crossings and city locations — Moscow, Smolensk, the Berezina, the Niemen — as geographic landmarks that anchored the narrative

That is six variables: two spatial coordinates (latitude and longitude), one quantitative variable (army size), one temporal variable (date), one additional quantitative variable (temperature), and a directional marker (advance versus retreat). Six variables is a lot. Most textbook treatments of multidimensional visualization warn that beyond three or four variables, charts become unreadable. Minard packed six variables into a single chart and made them all legible.

How he did it is the central question of this case study. The answer is: he chose the chart type after he identified the questions he wanted to answer. The chart type followed the questions, not the data.

The Visualization

The chart itself is best described as a flow map — though the term "flow map" did not exist as a category when Minard drew it. The main visual element is a thick band that represents the Grande Armée, drawn on top of a simplified map of the route from the Niemen River to Moscow. The width of the band at any point is proportional to the number of soldiers still alive at that point. The band is drawn in two colors: a tan color for the eastward advance and a black color for the westward retreat. Below the map, a separate panel shows temperature data from the retreat, plotted against the same longitude axis as the map above. River crossings, key battles, and city names are annotated directly on the chart.

Unpack the encoding choices and the genius becomes clear:

Encoding 1 — Longitude/latitude as position. The army's route is drawn on real geographic space. East is east. Moscow is where Moscow is. The viewer can see at a glance how deep into Russia the army penetrated and can connect the geography to any mental map they already carry. Geographic position is the most natural encoding for spatial data, and Minard used it directly.

Encoding 2 — Line width as army size. The thickness of the band at any longitude directly encodes how many soldiers were present at that point. Length and width are high on Cleveland and McGill's encoding hierarchy, so the viewer can perceive differences in army size pre-attentively. When the band is thick, there are many soldiers. When the band is thin, there are few. When the band shrinks dramatically between two points, something devastating happened between those points. This is visible without reading any numbers.

Encoding 3 — Color as direction. The tan band is the advance. The black band is the retreat. The two colors use different hues (identity encoding) to distinguish two fundamentally different phases of the campaign. A viewer who has never seen the chart before can immediately tell which portion is advance and which is retreat — not because of a legend, but because the two colors diverge at Moscow and the black band runs west alongside the tan band that runs east.

Encoding 4 — The separate temperature panel. Minard could have tried to encode temperature into the main map — perhaps as a color gradient on the band, or as a shaded background. He chose instead to put the temperature data in a separate panel below the map, using the same horizontal axis (longitude) as the map. This is a form of small multiples: two charts stacked vertically, sharing a horizontal scale. The viewer reads the main chart for army size and geography, glances down to the temperature panel to see how cold it was at that longitude, and integrates the two pieces of information in working memory. The temperature panel shows temperatures dropping to -30 degrees Réaumur (-37 degrees Celsius, -35 Fahrenheit) during the retreat through Smolensk and the Berezina crossing. The viewer can see, without doing any mental math, that the dramatic shrinking of the black band at the Berezina corresponds to the coldest days of the retreat.

Encoding 5 — Direct annotation. The names of cities, rivers, and battle sites are written directly on the chart at the locations where they happened. No legend lookup required. No cross-referencing. The viewer's eye can move from "Moscow" to the width of the band at Moscow to the temperature panel below Moscow — all in a single visual movement.

Encoding 6 — The implicit time dimension. Time is not explicitly plotted, but the viewer can infer it from the combination of geographic position and direction. The advance happened in summer and autumn; the retreat in winter. The temperatures in the bottom panel are labeled with dates, making the time sequence visible for the retreat phase. Tufte praised this as "weaving time and space into a single image" — the chart does not choose between geography and chronology; it provides both simultaneously.

The Impact

The 1869 pamphlet in which Minard first published the chart had limited immediate circulation. Minard was well-regarded in French engineering circles but was not famous outside them. The chart itself did not transform military history or policy — Napoleon's Russian campaign was already a century old, and the analytical question of "what happened and why" had been settled by historians long before Minard drew his map.

The chart's influence came later, and came from a different direction. In 1983, Edward Tufte published The Visual Display of Quantitative Information and made the Minard chart one of his central examples. Tufte reproduced the chart at full size, wrote about it with unusual fervor, and called it "probably the best statistical graphic ever drawn." He argued that it did six things at once — showing army size, two-dimensional geographic location, direction, temperature, time, and the sequence of events — in a form that remained coherent and readable. The claim has been debated. Some critics argue that other historical visualizations (Playfair's time-series charts, Nightingale's coxcombs, W.E.B. Du Bois's sociological charts of 1900) deserve comparable praise. Others argue that "best" is a meaningless superlative for any cultural artifact. But the chart's stature as a teaching example has been unshakable since Tufte's book appeared.

In the decades since, Minard's chart has become the single most-reproduced data visualization in textbooks, lecture slides, and popular science articles about information design. A quick search turns up reproductions on museum walls, on t-shirts, on book covers, and in the opening slides of more university visualization courses than can reasonably be counted. The chart is the canonical answer to the question "what does a great chart look like?" — and, more importantly for this chapter, it is the canonical answer to the question "what does it look like when the chart type matches the questions being asked?"

Why It Worked: A Chart-Selection Analysis

From the perspective of this chapter's framework, Minard's chart succeeded because it began with a clear list of questions and chose a form that answered all of them at once.

Question 1: Where did the army go? This is a spatial question. The answer is a map.

Question 2: How big was the army at each point? This is a quantitative question paired with a spatial dimension. The answer is a line (or, better, a band) whose width encodes the quantity along the spatial path.

Question 3: Did the army advance or retreat, and when did each phase happen? This is a direction-and-time question. The answer is color (for direction) plus implicit sequencing (for time).

Question 4: How cold was it? This is a quantitative question linked to location and time. The answer is a secondary panel aligned with the primary chart's horizontal axis.

Question 5: What happened where? This is a narrative-annotation question. The answer is text written directly on the chart at the relevant locations.

Question 6: What is the overall story? This is the synthesis question. The answer is everything above, arranged so the viewer can see it all at once.

Notice what Minard did not do. He did not plot longitude against army size on a scatter plot. He did not make a line chart of army size over time. He did not make a choropleth map of Russia colored by army presence. He did not make a bar chart of casualties by stage of the campaign. Any of these charts could have been drawn from the same underlying data. Each would have answered one or two of Minard's questions, but none would have answered all six at once.

The key insight is that the chart type was chosen to match the combination of questions, not any single question in isolation. A scatter plot would have answered "how did army size relate to longitude" but lost the geography. A line chart would have answered "how did army size change over time" but lost the space. A map would have answered "where was the army" but lost the quantity. Minard needed a form that encoded spatial position, quantity, and direction simultaneously — and he invented (or adapted) the flow map to do exactly that.

This is the principle of Section 5.4 in action: start with the question. When you have multiple questions, ask whether they can be answered by a single chart type or whether they require separate charts. Sometimes the honest answer is "separate charts," and small multiples or a dashboard is the right response. Sometimes the questions have enough shared structure that a single, carefully designed chart can carry all of them. Minard's campaign data had enough shared structure — everything was linked to position along the route — that one chart could do the work of five or six.

The Limits of the Chart

Intellectual honesty requires acknowledging that Minard's chart is not perfect and not universally applicable.

The chart has a specific narrative. It tells a story: a large army marched east, shrank along the way, reached Moscow, retreated, and was destroyed in the cold. The chart supports this narrative strongly — almost too strongly. A viewer who wanted to contest the story ("was the cold really the main cause of the losses?") would find the chart's framing difficult to argue against, because the visual juxtaposition of shrinking band and falling temperature is so pre-attentively compelling. This is a mild form of the ethical concern raised in Chapter 4: every chart is an argument. Minard's chart makes a strong argument. Strong arguments are not inherently wrong, but they require the chart maker to be sure the argument is defensible.

The chart is not directly reproducible with modern software defaults. No mainstream plotting library (matplotlib, seaborn, Plotly, Altair) has a built-in "Minard flow map" function. Drawing Minard's chart requires custom work: calculating the band width at each point, drawing the band as a filled polygon, overlaying the geographic base, aligning the temperature subplot, annotating the cities by hand. This is one reason the chart has become an inspiration rather than a template — it is easy to admire but hard to copy.

The chart's information density is high, and viewers need time. Minard's chart is not a 5-second chart. A viewer needs at least a minute, probably several minutes, to take in all six encoded variables. This is fine for an educational poster or a book illustration, but it is not the right form for a dashboard tile or a broadcast graphic. The chart is optimized for slow reading, not quick scanning. Different contexts demand different chart types. Minard's chart was designed for a pamphlet where the reader had time to study the image.

Not every dataset has Minard's structure. The chart works because all the variables share a natural organizing principle: location along the route. Army size, direction, temperature, and event names all line up against the same horizontal axis. Most datasets do not have this kind of internal structure, and trying to force them into a Minard-style visualization produces confused charts. The lesson is not "draw flow maps for everything" but "find the organizing principle that lets your questions share a single chart, if one exists."

Lessons for Modern Practice

Minard's chart is 155 years old, but the principles it demonstrates are the principles this chapter has been arguing for.

Start with the questions, not the data. Minard had a list of things he wanted the viewer to see. He designed the chart around that list. If you start with the data and ask "what chart should I make?" you will default to the chart type you know best. If you start with the questions and ask "what chart will answer these?" you will often end up with a different chart — sometimes a better one.

Look for organizing principles that let questions share a chart. Minard's success depended on finding that every variable in his dataset could be related to longitude along the route. Not every dataset has such an organizing principle, but when one exists, it lets you fold multiple questions into a single chart. The same principle operates in small multiples (where the shared x-axis is the organizing dimension) and in carefully designed dashboards.

Match the encoding to the information type. Minard used position for geography, width for quantity, color for direction, and text for names. Each encoding fit the type of information it was carrying. When you design a chart, ask for each variable: what kind of value is this, and what encoding is most appropriate for that kind of value? Chapter 2 built the theoretical framework for this; Chapter 5 is where you apply it in chart selection.

Do not be afraid of unusual chart types when the standard types do not fit. Minard's flow map was not a standard chart type in 1869 — he effectively invented it for this application. If your questions do not fit any standard chart type, the right response is sometimes to design a new one, not to cram the questions into a chart that was not built for them. Custom visualization is a real option. Most practitioners underuse it because it is harder than selecting from a gallery.

Remember that slow reading is a legitimate use case. A chart that rewards extended study is not the same thing as a chart that fails at quick scanning. Some charts should be read in 5 seconds, and others should be studied for 5 minutes. Both are valid, but they require different designs. Before you design a chart, know how long the viewer will spend with it and design accordingly.

Annotate directly on the chart whenever possible. Minard put the city names, battle locations, and key events directly on the chart. No legend. No cross-referencing. Modern charts often separate the annotations into legends, captions, or callout boxes — which is fine in context, but Minard's approach shows what happens when annotations live where the data lives. The reader's eye does not have to move between the chart and an external key.

Discussion Questions

On "best chart ever drawn." Tufte's claim is famously strong. Do you think Minard's chart deserves the superlative? What criteria would you use to compare it to other candidates — Nightingale's coxcombs, Playfair's time-series, W.E.B. Du Bois's sociological charts, or modern work like the IPCC climate graphics? Is "best" even a meaningful category for information design?
On matching chart to questions. The central lesson of this case study is that Minard chose the chart type to match the list of questions he wanted to answer. Think about your own work. When you start a new visualization, do you begin with a list of questions, or do you begin with the data and the default chart types your tools provide? What would change if you always started with questions?
On slow charts and fast charts. Minard's chart is designed for extended study. Modern dashboards and broadcast graphics are designed for quick scanning. Both approaches have legitimate uses. How do you decide, for a given project, which style is appropriate? What does it mean to design "for the time the viewer has"?
On custom chart types. Minard effectively invented the flow map for this application. Modern visualization practice tends to rely on a fixed vocabulary of chart types supported by standard libraries. When, if ever, is it worth the extra effort to build a custom chart type? What are the costs (time, reproducibility, maintenance) and what are the benefits (fit to the questions)?
On the chart as argument. Minard's chart makes a strong implicit argument: the cold killed Napoleon's army. The visual juxtaposition of shrinking band and falling temperature is nearly impossible to ignore. Do you think Minard was honest about the argument his chart was making? How should a chart maker handle the tension between designing a chart that tells a clear story and designing a chart that does not overstate the evidence?
On the threshold concept. The chapter's threshold concept is "Question Before Chart." Minard's work is often treated as a triumph of visual design, but the underlying lesson is about disciplined thinking — he decided what questions to answer before he drew a single line. In your own practice, how can you enforce this discipline? What would it take to make "list the questions" a mandatory first step in every chart you create?

Minard's chart has been praised for 150 years, reproduced in hundreds of textbooks, and studied by more visualization students than any other single image in the discipline. The praise is not, in the end, about the chart's beauty, although the chart is beautiful. The praise is about the alignment between the chart's form and the questions it answers. When you read this chapter's threshold concept — "You do not choose a chart based on what looks interesting; you choose it based on what question you are answering" — Minard is the example. The discipline he practiced in 1869 is the discipline you should practice every time you make a chart.