How to use this guide

Every data story starts from one of two places:

• You have a question: you're investigating a hypothesis and you go looking for data to test it.
• You have a dataset: you received or found data, and you need to explore it to see what it contains.

Either way, interrogate the source before you start analyzing. Work through each section in order.

Not every move applies to every dataset. Skip sections that don't fit your data. If you don't understand what a section means but you feel like it might be relevant, research it.

Remember: Data tells you what, people tell you why. Data shows you who is counted, but the system decides who isn't. Before you publish, talk to someone who lives inside the numbers.

Interrogate the source

Do this before any analysis. Data is not neutral. It always comes into being with a certain goal in mind. Read any accompanying documentation, methodology notes, or data dictionary.

• Document the data's origin: where it comes from, who recorded it, how, and for what purpose.
• Check the methodology notes for completeness: how were records collected?
• Identify who or what is not included in these figures. Are there things missing?
• Explore the methodological notes for choices that had to be made to generate the data.
• Cross-check key figures against at least one independent source.
• Search for prior analyses and note any known methodological flaws.

First look: understand what you have

Before any analysis, get oriented.

• Define what one row represents (a person, a contract, a country, a day).
• How many rows? How many columns?
• What time period does the data cover? When was it last updated?
• Look at the first and last rows. Do the values look plausible?
• Check data types: are numbers stored as numbers? Are dates recognized as dates?
• Scan for unexpected values: misspellings, mixed formats, ALL CAPS vs lowercase.

Descriptive statistics: what does the data look like overall?

Run these on every column.

• Calculate min, max, mean, and median for numeric columns.
• Where do mean and median diverge strongly? That signals skew or outliers.
• Find the most and least common values in categorical columns.
• Count unique values in categorical columns (countries, companies, categories).

Rankings & comparisons: who leads, who lags?

• Sort each column top and bottom: what's the biggest? The smallest?
• Find who or what appears most often.
• What is the gap between first and last?
• Compare groups: does one category consistently outperform another? Do richer regions get more contracts? Do older passengers survive less?

Context: compared to what?

Raw numbers rarely tell the whole story.

• Express values as rates: per capita, per 1,000, as a percentage of total.
• What is the denominator? A number rarely works in isolation. Is a million a lot or not? Depends on the context.
• Compare to a baseline: national average, previous year, EU norm. A big number in a big country may be less remarkable than a small number in a small one.

Time patterns: is something changing?

Only applies if your data has a date or year column.

• Identify the overall trend: rising, falling, or flat.
• Spot spikes or sudden drops and research what happened at those points.
• Check for seasonal cycles.
• Note whether the rate of change is accelerating or slowing.
• Compare first and last period: what is the % change?
• Check whether a record high or low has been reached.
• Are there sudden breaks or discontinuities? Could these reflect a change in methodology rather than reality?

Geographic patterns: does place matter?

Only applies if your data has a location column.

• Check whether values cluster in certain regions or countries.
• Look for urban/rural or rich/poor splits.
• Identify places that are outliers compared to their neighbours.
• Can you put it on a map?
• Have administrative boundaries changed over the data's time span? (regions merged, renamed, redrawn)

Relationships: does X connect to Y?

Correlation is not causation, but a strong pattern is worth investigating.

• When one value goes up, does another go up or down?

Outliers & anomalies: what doesn't fit?

An outlier may be your story.

• What values are far from the average? (a rough rule: more than 2–3 standard deviations from the mean)
• Are there suspiciously round numbers or uniformly identical values?
• Check whether one row dominates the totals.
• Check whether outliers coincide with known events: elections, policy changes, crises.

Missing data: what isn't there?

Absence of data can be a finding on itself.

• Count empty cells per column.
• Are the gaps random, or do they cluster in certain countries, years, or categories?
• What would you expect to be here that isn't?
• Ask the data provider why records are missing and document the answer.

Verification: before you publish

If something looks to good to be true, it probably is.

• Find a comparable dataset from an independent source and cross-check key figures.
• Speak to an expert who can sanity-check your main findings.
• Consider alternative explanations for the patterns you found, can you rule them out?
• Document your analysis steps so anybody could reproduce them. Publish them as a methodology with your article.
• Identify what the data does not cover, and acknowledge it in your story and your methodology.
• Save a clean, unmodified copy of the raw data with a note of where and when you got it.

The news test

Once you find something interesting, ask these three questions. If yes to all three, you might have a story.

• Surprising? Is this different from what you'd expect?
• Significant? Is it meaningful in scale, impact, or who is affected?
• Showable? Write a one-sentence summary of your finding. Can you also show it in one chart?