Data in Journalism

The Rise of Data Journalism

Data journalism uses computational methods to find, analyze, and tell stories hidden in datasets. The field was pioneered by outlets like The Guardian's Datablog, FiveThirtyEight, and ProPublica, and has become standard practice at most major news organizations. The Panama Papers investigation — which analyzed 11.5 million leaked documents to expose offshore tax evasion — is perhaps the most ambitious data journalism project ever.

Good data journalism combines the rigor of statistical analysis with the storytelling power of traditional journalism. It can reveal patterns invisible to traditional reporting: racial disparities in police use of force, geographic patterns in healthcare access, systematic differences in how judges sentence defendants.

How to Evaluate a Data-Driven Story

When reading data journalism, apply these checks. Where does the data come from? Government statistics, academic datasets, and data obtained through Freedom of Information requests are generally more reliable than self-reported surveys or proprietary corporate data.

Is the methodology transparent? The best data journalism shows its work — publishing the underlying data and analysis code. ProPublica, The Markup, and FiveThirtyEight regularly publish their datasets and methodologies. If you cannot see how the analysis was done, be cautious.

Does the visualization match the data? Charts can mislead through truncated axes, misleading scales, or cherry-picked timeframes. Check whether the visual representation accurately reflects the numbers.

Are alternative explanations considered? A correlation between two variables does not prove causation. Good data journalism acknowledges confounding factors and alternative explanations. If a story presents a correlation as if it were causation, the analysis is incomplete.

A Worked Example: How a Chart Misleads

Suppose a news graphic shows unemployment 'soaring' from 5.0% to 5.4% over three months, using a bar chart whose Y-axis starts at 4.9% instead of zero.

With the axis truncated at 4.9%, March's bar is five times taller than January's, implying a dramatic surge. With an honest axis starting at zero, the three bars are nearly identical — a change of just 0.4 percentage points. Same data, opposite impression. The actual change is 0.4 points, an 8% relative rise: meaningful, but hardly the crisis the truncated chart implies. A data-literate reader checks the axis before reacting to the shape of the bars.