Case Study 2: The Financial Times Log-Scale Pandemic Trajectory Chart

Case Study 2: The Financial Times Log-Scale Pandemic Trajectory Chart

In March 2020, the world needed a chart that had never been drawn before. The Financial Times designed one in a week. Millions of people learned to read it in a month. It is a master class in choosing a chart type for a specific question — and in breaking the rules with reason.

The Situation

In the first weeks of March 2020, the novel coronavirus now known as SARS-CoV-2 was spreading across the world at a rate that surprised even the epidemiologists who had been warning about it. Case counts were doubling every two or three days in some countries, every five or six days in others. Governments were making decisions — school closures, lockdowns, border restrictions — that depended on understanding how their own trajectory compared to countries that were further along the outbreak.

The question was specific and urgent: Is our country's outbreak growing faster or slower than other countries' outbreaks, controlling for the fact that different countries started at different times and different population sizes?

Standard chart types struggled with this question. A simple line chart of cases over calendar time showed several problems at once. First, it compressed the curves of earlier-hit countries (like China and Italy) off the left side of the chart while stretching the curves of later-hit countries across the right. Second, it placed countries at vastly different stages of their outbreaks side by side without aligning them, so a viewer could not easily tell whether the United States on March 15 was tracking Italy's trajectory or diverging from it. Third, it used a linear y-axis that made exponential growth look like a J-curve — dramatic but hard to interpret quantitatively, because the slope on a linear scale changes with time even when the growth rate is constant.

Public health researchers had long used logarithmic scales for exponential growth because logarithms have a mathematical property that matters here: on a log scale, exponential growth plots as a straight line, and the slope of the line corresponds to the growth rate. A line on a log chart going up at a 45-degree angle means the quantity is doubling over some fixed interval. A flatter line means slower doubling. Two lines with the same slope mean the same growth rate. A viewer who can read log scales can compare growth rates across countries simply by comparing line slopes.

But log scales were (and are) unfamiliar to most general audiences. News outlets had avoided them for years because readers often misread them — perceiving the same distance on the y-axis as representing the same quantity, which is true on a linear scale but false on a log scale. The traditional view was: if you want general readers to understand your chart, use a linear y-axis.

In March 2020, the Financial Times graphics desk — led by John Burn-Murdoch — made a different call. The questions they were trying to answer required a log scale. A linear chart simply could not communicate the growth-rate comparison at the heart of the pandemic story. So the FT built a log-scale line chart, added an alignment trick to handle the different start times, and taught their readers to read it. The chart went viral. It was republished by newspapers around the world, adapted by health departments, copied by amateur analysts, and referenced in academic papers. It became, for a few critical weeks in the spring of 2020, one of the most influential pieces of data visualization in history.

The Data

The underlying data was publicly available from Johns Hopkins University's Center for Systems Science and Engineering (CSSE), which aggregated case counts from national health agencies worldwide and published daily updates through a GitHub repository. For each country and each date, the dataset provided:

Cumulative confirmed cases
Cumulative confirmed deaths
Cumulative recovered cases
The date of each observation

The data had several properties that shaped chart selection. First, it was a count that only grew (cumulative cases cannot decrease). Second, the growth was approximately exponential in the early weeks, which meant the raw count spanned several orders of magnitude across countries at any given moment. Third, countries entered the dataset at different times, because the outbreak hit different countries at different times. Fourth, the data was noisy — reporting was inconsistent, testing capacity varied, and weekend reporting delays were visible in every country's series.

The question-to-data fit was the central problem. The question was about growth rates — how fast each country's outbreak was growing relative to the others. The data was absolute counts spanning several orders of magnitude. A chart that showed absolute counts on a linear scale could not make growth rates visible. A chart that normalized counts (say, by population) would answer a different question. The FT needed a chart that preserved the absolute counts but transformed the axis so that growth rates became visible as slopes.

The Visualization

The FT chart was a line chart. Each country was one line. The x-axis was "days since the country passed some threshold" — typically "days since 100 confirmed cases" or "days since 10 confirmed deaths." The y-axis was the cumulative count on a logarithmic scale. Reference lines were drawn diagonally across the chart showing what a doubling-every-2-days curve, a doubling-every-3-days curve, a doubling-every-week curve, and a doubling-every-month curve would look like.

Every design choice served a specific question.

Choice 1 — Line chart. The question was about change over time. Line charts are the signal chart type for change-over-time questions. This was the easy call.

Choice 2 — Logarithmic y-axis. The central question was about growth rates, not absolute counts. On a log scale, exponential growth plots as a straight line, and the slope of the line corresponds directly to the growth rate. A viewer looking at two lines can compare their slopes and immediately see which country was growing faster. This is not possible on a linear scale, where the slope at any point depends on the current level and the rate interact in a way the eye cannot decode.

Choice 3 — "Days since threshold" alignment. This was the clever choice. Instead of plotting each country against calendar time, the FT plotted each country against days since it passed 100 cases (or, in some versions, 10 deaths). This aligned the start of every country's outbreak at day zero, so the viewer could compare trajectories from matching starting points. Italy's curve from its day 14 (two weeks after 100 cases) could be compared directly to the United States' curve from its day 14. The alignment eliminated the confusing offset that calendar-time charts had been producing.

Choice 4 — Diagonal reference lines. The FT drew dashed diagonal lines at fixed doubling rates — every 2 days, every 3 days, every week, every month. A country's line could be read by comparison to these reference lines. If a country was tracking the "doubling every 3 days" line, it was doubling every 3 days. If it curved downward, it was slowing. If it fell between the "week" line and the "month" line, it was no longer in rapid growth. These reference lines made the log-scale readable without requiring the viewer to compute logarithms or estimate slopes numerically.

Choice 5 — Direct labeling. Countries were labeled directly at the end of each line, not via a legend. This is a technique discussed in the chapter (and expanded in Chapter 7): when there are more than three or four series, direct labeling reduces the eye movement required to identify each line. On the FT chart, the viewer saw "Italy" written next to the Italy line, "United States" next to the US line, and so on. No legend lookup.

Choice 6 — Color restraint. The chart used a limited color palette, with grayed-out lines for countries not highlighted and a small number of emphasized lines (typically the viewer's country plus a few reference countries). This let the viewer focus on the comparison of interest without the visual noise of 50 distinctly colored lines.

Choice 7 — Thick and thin lines to signal status. Lines for countries with more complete data or more recent updates were drawn thicker; lines for countries with sparse or delayed data were drawn thinner. This subtle weight variation gave the viewer a visual cue about which trajectories to trust more.

The Impact

The FT trajectory chart spread faster than most viral images in the spring of 2020. Within days of its first publication in early March, it was being shared on Twitter, embedded in news articles across multiple languages, and reproduced (with varying degrees of fidelity) by health ministries, university research groups, and amateur data analysts. John Burn-Murdoch, the lead author at the FT, found himself briefly at the center of a global conversation about how to visualize the pandemic — giving interviews, explaining the log-scale decision, and responding to criticism from readers who were seeing log-scale line charts for the first time.

The chart changed how millions of people understood the pandemic. Before the chart, most news coverage of case counts had used linear-scale line charts that made every country's growth look like an undifferentiated vertical spike. After the chart, viewers could see that some countries were tracking the "doubling every 3 days" reference line while others were tracking "doubling every 6 days" — a difference of a factor of two in how fast cases were accumulating. The chart made the consequences of faster or slower spread visible in a form that numbers alone could not convey. Policy analysts, public health officials, and ordinary viewers used the chart to reason about the urgency of intervention.

The chart also survived contact with criticism. Some readers complained that log scales were misleading because they made rapid growth "look smaller" than on a linear chart. The FT responded directly, in articles and on social media, explaining that the log scale was the whole point — it let the viewer see whether the growth rate was constant (straight line), slowing (curve bending down), or accelerating (curve bending up). The criticism was a fair one in general — log scales can be confusing — but the FT chart was built with extensive annotation, reference lines, and explanatory text that taught readers how to interpret it. The chart did not just use a log scale; it taught its audience how to read a log scale, which is a rarer and more demanding design accomplishment.

In the months that followed, variants of the FT chart appeared everywhere. The Economist adopted similar conventions. Our World in Data (owd.ourworldindata.org) built interactive versions that let readers select countries and thresholds. The US Centers for Disease Control and Prevention and the UK Health Security Agency included log-scale trajectory charts in their pandemic reporting. A design choice that had been considered too technical for general audiences in February 2020 became a standard convention of pandemic reporting by April.

Why It Worked: A Chart-Selection Analysis

The FT chart is a textbook example of matching chart type to question. Walk through the framework step by step.

Step 1: Classify the question. The question was specifically about growth rates — how fast each country's outbreak was growing, and how that rate compared to other countries. This is not a simple "change over time" question (which a linear chart would answer) but a more specific "rate of change over time" question. Identifying this precisely was the critical move. A less precise framing — "show case counts over time" — would have led to a linear chart that did not answer the real question.

Step 2: Classify the data. The data was a count (continuous, positive, monotonically increasing) indexed by date (temporal) and country (categorical). Multiple series, one per country.

Step 3: Consult the matrix. For change-over-time questions with multiple continuous series indexed by category, the matrix points to line charts with one line per category, possibly faceted into small multiples if the series overlap heavily. Line chart was the right base choice.

Step 4: Apply the context check.

Dataset size: 200+ countries would produce a spaghetti chart. Mitigation: emphasize a handful of countries and gray out the rest.
Range of values: Cumulative cases span several orders of magnitude across countries at any moment. Mitigation: log-scale y-axis.
Alignment problem: Countries started outbreaks at different times. Mitigation: align to "days since threshold," not calendar time.
Audience: General newspaper readers, not epidemiologists. Mitigation: add reference lines for standard doubling rates, teach the reader how to read the chart through annotation.
Medium: Digital, with the ability to update daily and show animation if desired. Mitigation: static version for headline use, interactive version for reader exploration.
Purpose: Explanatory communication with a specific policy-relevant insight to convey. Mitigation: emphasize the comparison that matters for the reader's country.

Every design choice followed from the combination of question type, data properties, and contextual constraints. Nothing was decorative. Nothing was default.

The rule the FT broke. The chart violated the conventional wisdom that log scales should not be used for general audiences. The violation was justified by the questions the chart was answering — growth-rate comparisons require a log scale — and by the extensive effort the FT put into making the log scale readable. This is the chapter's point from Section 5.8 in action: rules can be broken, but only with reason, and the reason has to be defensible in terms of the questions being asked.

The Counter-Examples: Charts That Failed at the Same Question

For contrast, it is worth noting what the alternative charts looked like and why they failed.

Linear-scale line chart with calendar dates. This was the default choice in the first weeks of pandemic reporting. Each country was a line. The y-axis was cumulative cases. The x-axis was calendar date. The chart showed "which countries have more cases" at any given date but obscured the growth-rate question. Countries at different outbreak stages looked incomparable. Growth rates were invisible — a country doubling every 2 days and a country doubling every 6 days both looked like upward curves, and the difference in slope was unreadable because of the scale compression at the low end and expansion at the high end.

Bar chart of cumulative cases by country. Some outlets used a horizontal bar chart updated daily, showing the countries with the highest cumulative case counts ranked from most to fewest. This answered "which country has the most total cases right now?" but said nothing about growth rates or trajectories. A country that had peaked and was leveling off looked the same as a country still in rapid growth. The chart could not support policy decisions about intervention timing, because the ranking did not distinguish between past severity and current trajectory.

Map of cases by country. Choropleth maps of cases per capita or cases per 100,000 population answered "where is the outbreak?" but again did not capture growth rates. A country with a flat trajectory at 1,000 cases per 100,000 looked darker than a country with rapid growth currently at 200 per 100,000 but doubling weekly. The map form was spatial, but the spatial question was not the most important question in early March 2020.

Linear chart with different y-axis scales per country. Some outlets tried to solve the "different scales" problem by creating one small-multiple panel per country, each with its own auto-scaled y-axis. This preserved the shape of each individual country's curve but made cross-country comparisons impossible, because the y-axes were incomparable. A sharp curve on one panel might be at vastly smaller absolute levels than a gentle curve on another panel.

Each of these alternatives was a reasonable chart for some question. None was the right chart for the question the FT was answering. The lesson is the same lesson as Case Study 1: the chart type follows the questions. When the question is "compare growth rates across countries aligned at matching starting points," the answer is a log-scale line chart with days-since-threshold alignment. When the question is different, the answer is different.

Lessons for Modern Practice

The FT pandemic chart is not a template you will copy. It was built for a specific question in a specific moment. But the process by which it was built is a template for your own chart design.

Start by articulating the question in a sentence. The FT team could have said "show pandemic data" and produced any of a dozen standard charts. They said (approximately) "help readers compare growth rates across countries controlling for different outbreak start times." That sentence is specific enough to rule out most chart types and point toward the one that fits. Before you start any chart, write the question in a sentence that is specific enough to rule out bad answers.

When the standard chart types do not fit, examine why. The standard chart types for "cases over time" would have been a linear line chart. The FT team noticed that the linear line chart did not answer their specific question and examined why. The why — that growth rates are visible as slopes only on log scales — pointed them toward a log scale. Noticing the gap between the standard chart and the specific question is the beginning of better chart design.

Be willing to teach your audience. The log scale was unfamiliar to most FT readers. Rather than avoiding it, the FT team built extensive annotation, reference lines, and explanatory text to teach readers how to read the chart. This is harder than using a familiar chart type, but it is legitimate, and the payoff is a chart that answers the real question. If your question requires an unfamiliar chart type, consider whether you can invest the extra effort to teach it rather than substituting a worse chart your audience already knows.

Use alignment tricks when calendar time confuses the comparison. The "days since threshold" alignment was a small choice with a huge effect. Aligning multiple series at a meaningful event (first 100 cases, first case of a disease, start of a treatment) often makes comparison possible where calendar-time alignment would obscure it. This is a general technique worth keeping in your toolkit.

Annotate the reference framework. The doubling-every-N-days reference lines on the FT chart turned an unfamiliar log-scale into a readable chart by giving viewers a built-in conversion tool. If you use any chart where the axis is unfamiliar (log, probit, logit, percentile, etc.), include reference marks that let viewers translate positions into concepts they understand.

Update and iterate. The FT chart was updated daily for months. Each iteration fixed small issues, improved annotations, and adapted to the evolving story. Visualization is not a one-and-done activity for communicated data; it is a continuous improvement process. Plan for iteration, not for a single perfect chart.

Break rules with reason, not convenience. Rules like "do not use log scales for general audiences" exist because log scales are genuinely harder to read. The FT broke this rule because the question required it and because the team was willing to invest in teaching. Rule-breaking of this kind is legitimate and sometimes necessary. Rule-breaking of the "I did not feel like following the rule" kind is not. Know which kind you are doing.

Discussion Questions

On teaching the audience. The FT explicitly taught its readers how to read a log scale, something most news outlets had previously considered too technical for general audiences. Was this the right call? When, in your own work, is it appropriate to invest in teaching rather than choosing a more familiar chart type? What are the risks of the "teach the audience" approach?
On the question-first discipline. The FT team's critical move was articulating the question precisely enough to rule out linear charts. Think about a recent visualization you made. Could you have articulated the question in a single sentence before you started? If yes, did that sentence point to the chart you actually made? If no, what would change if you had required yourself to write the sentence first?
On breaking rules with reason. The FT broke the "no log scales for general audiences" rule because the question required it. What other visualization rules might be worth breaking under the right conditions? What makes a rule-breaking defensible versus a rule-breaking that is just sloppy?
On alignment tricks. The "days since threshold" alignment was the clever choice that made the comparison possible. Think about a dataset in your own work where calendar time alignment obscures the story. What alignment (days since an event, percentage of a process completed, relative to a baseline) might make the comparison clearer?
On responsibility during a crisis. In March 2020, the FT chart was being used to inform policy decisions that affected millions of lives. Does the high-stakes context change the ethical obligations on chart design? What would you do differently when designing a chart that will influence crisis decisions versus one that will be seen by three colleagues in a business review?
On the temporal window of a chart's relevance. The FT trajectory chart was dominant for roughly two months in the spring of 2020. By the summer, the questions had changed — the growth-rate comparison became less central as countries diverged in their responses — and the chart form evolved. How do you know when a chart type has outlived its question? What signals should trigger a redesign?

The FT pandemic chart was not a piece of art. It was a response to a specific question under a specific constraint in a specific moment. But the process that produced it — starting with the question, identifying why standard charts failed, choosing a form that fit the question, teaching the audience to read the form, iterating over time — is the process this chapter has been arguing for. In 2020, the world needed a chart that had never been drawn the same way before. The process of matching form to question produced one in a week. That process will produce the right chart for your questions, too, if you give it the chance.