Case Study 2 (Deep Dive): The Challenger Data Display

This case is real history. We keep strictly to the verifiable record—the cold-temperature O-ring concern, the night-before teleconference, and the failed persuasion. We invent no quotations, dates, or casualty details. Edward Tufte's analysis of the actual charts is the source for the data-display reading (Tufte, Visual Explanations, 1997, Tier 1).

What happened

On the night before the Space Shuttle Challenger launched in January 1986, engineers at Morton Thiokol—the contractor that built the solid rocket boosters—were worried. The forecast for launch morning was unusually cold, and they had reason to believe cold compromised the O-rings, the rubber seals in the joints between the booster segments. They had data from previous flights associating low temperature with O-ring damage. In a late-night teleconference with NASA, they raised the concern and presented charts to support it. They did not persuade the decision-makers. The launch proceeded the next morning in record cold; an O-ring failed; the shuttle was lost and its seven crew members died.

The engineers were not ignorant of the danger. The information that mattered existed in their data and their documents. The failure—this is the part that belongs in a chapter on data display—was that the way the data was presented did not make the one decisive relationship visible and unmissable.

The data-display reading

Edward Tufte studied the charts the engineers actually used and concluded that their structure obscured the very point they were trying to make. We stay at the level his analysis supports:

The decisive relationship was never isolated. The point that should have dominated the entire presentation was a single relationship: as temperature falls, O-ring damage rises—and the forecast launch temperature was colder than any prior flight had experienced. That relationship existed in the data. But it was spread across multiple charts and tables rather than concentrated into one clean graphic that showed it at a glance.

The data wasn't ordered to reveal the pattern. When you want a reader to see a relationship between two variables, you sort and plot the data so the pattern leaps out—for a temperature-versus-damage relationship, that means putting temperature on an axis and letting the eye follow damage as temperature changes. The engineers' displays did not foreground the trend this way; the reader had to assemble the relationship from scattered pieces rather than perceive it directly.

The crucial comparison was missing from view. The most persuasive single fact—we have never flown anywhere near this cold—is a comparison between the forecast temperature and the entire range of prior flight experience. A display that placed the forecast temperature against all prior launches would have made the extrapolation obvious: the launch was being attempted far outside the conditions under which the hardware had ever been tested. That comparison was not made unmissable.

The decision-makers, reading under time pressure late at night, did not reconstruct the buried relationship from the available exhibits. The data did not fail. The visualization did.

What a clean display would have shown

Here is the figure that the existing data could have supported—and that this chapter's principles would produce. (Described in words, with alt-text, since this book is text.)

Figure (the display that should have existed), described: A single scatterplot. The x-axis is temperature, in degrees Fahrenheit, running from cold on the left to warm on the right, covering the full range of prior launches and extending left to the forecast launch temperature. The y-axis is a measure of O-ring damage. Every prior flight is one point, placed by its launch temperature and the damage it sustained. Because the data is sorted naturally along the temperature axis, the eye follows damage increasing as temperature drops. A single clearly marked point or vertical line sits far to the cold (left) end, labeled "forecast launch temperature"—visibly outside the cluster of all prior experience, in the region where damage is worst. Alt-text: a scatterplot of O-ring damage versus launch temperature, every prior flight as a point, showing damage rising as temperature falls, with the forecast launch temperature marked far colder than any previous flight—making the danger of launching in unprecedented cold immediately visible.

One figure. One relationship. The point unmissable: we have never flown this cold, and cold is exactly when the seals fail. Its interpretive caption writes itself, in the Level-3 form from §9.5:

Figure X. O-ring damage rises sharply as launch temperature falls, and tomorrow's forecast is colder than any flight on record—we would be launching far outside all prior experience, into the conditions where the seals have failed before.

That caption states the finding, gives the comparison, and names the implication (do not launch). A decision-maker who saw only that figure and that caption would have the decisive fact, isolated and unmissable. The data for it existed. The display did not.

The lesson, and the line to the rest of the book

This is the same case we read structurally in Chapter 4, now at the level of the chart. There, the buried conclusion; here, the buried relationship in the data. Both are the same failure wearing different clothes: the necessary information was present, but no single element made the decisive point impossible to miss.

The warning the case leaves for your own work is exact: "the data was in there" is not the same as "the reader saw the point." When you present data that informs a decision—especially a high-stakes one—accuracy is not the bar you have to clear. Isolating the one relationship that matters into a single clean figure that a time-pressured reader cannot miss is the bar. A chart that scatters a critical pattern across many exhibits, or drowns it in decoration, has not merely committed a craft error. In the situations that matter most, it has failed the people who depended on it.

We return to this case one final time in Chapter 38 (ethics), where the responsibility dimension—designing communication so the truth is received, not merely included—is the whole subject. For now, hold the data-display lesson: make the decisive relationship unmissable, or accept that it may not be seen.

Discussion questions

1. The chapter distinguishes accuracy from making the point unmissable. Using only the verifiable facts above, explain how the Challenger charts could be fully accurate and still fail.
2. Why does sorting the data along the temperature axis matter so much for revealing the relationship? What happens to a reader's ability to see a trend when the same data is presented unsorted or split across tables?
3. The "clean display" puts the forecast temperature against the full range of prior experience. Why is that comparison more persuasive than simply stating the forecast temperature as a number?
4. This case appears in Chapters 2, 4, and 9, each framing it as a different failure (audience, structure, data display). Are these three separate failures, or one failure seen from three angles? Defend your answer.
5. A care note for discussion. This is a real tragedy in which people died. How should a technical writer handle a case like this in their own teaching or documentation—what's the line between learning the lesson and exploiting the event? (We take this up directly in Chapter 38.)