Data and Investigative Journalism

Introduction and Context

Data and investigative journalism are complementary practices. Data journalism uses structured information—such as databases, spreadsheets, statistical releases, maps, and administrative records—to discover, verify, explain, and visualize patterns. Investigative journalism is a systematic process of examining concealed, complex, or consequential matters in the public interest. An investigation may use data, interviews, observation, leaked documents, public records, or all of these together.

The combination is powerful because datasets can reveal patterns that individual testimonies do not show, while investigative reporting supplies the institutional, human, and ethical context that numbers alone cannot provide. A database may show that a local authority repeatedly awards contracts to a small group of companies; interviews and records may explain how the system operates, who benefits, and whether the pattern reflects corruption, legal procurement, administrative failure, or legitimate specialization.

Data is not automatically objective. It is produced through human decisions about definitions, collection, classification, omission, and publication. Investigative journalists therefore treat data as evidence requiring scrutiny rather than as unquestionable fact.

Learning Objectives

By the end of this chapter, students should be able to:

1. Distinguish data journalism from investigative journalism and explain how they reinforce each other.
2. Formulate an evidence-based investigative question and identify relevant datasets and records.
3. Obtain information through public databases, records requests, freedom-of-information processes, and source relationships.
4. Clean, document, analyze, and verify data using spreadsheets and basic statistics.
5. Apply document analysis, source mapping, triangulation, and investigative frameworks.
6. Communicate findings accurately through narrative, tables, charts, and transparent methodology.
7. Recognize ethical, legal, privacy, security, and governance risks.

1. Core Concepts and Terminology

Data journalism

Data journalism is journalism in which data materially contributes to discovering, testing, explaining, or communicating a story. It includes relatively simple work, such as calculating changes in a public budget, and advanced work, such as joining several datasets to investigate environmental inequality. The defining feature is not visual design or technical complexity but disciplined use of evidence.

Investigative journalism

Investigative journalism is original reporting that examines matters of public significance through sustained research. It normally involves a hypothesis or question, documentary and human sources, verification, and an effort to reveal information that is not readily available or understood. An investigation can expose wrongdoing, but it can also reveal institutional neglect, conflicts of interest, systemic discrimination, unsafe conditions, or failures of policy.

Public interest

Public interest concerns information that helps communities understand matters affecting rights, safety, public resources, democratic accountability, or institutional power. Public curiosity is not the same as public interest. A person’s private life may attract attention without justifying publication.

Dataset and metadata

A dataset is an organized collection of observations, records, or measurements. A row may represent a person, transaction, inspection, or event; a column records a variable such as date, location, amount, or status. Metadata describes the dataset: its creator, collection method, definitions, time period, units, revisions, and limitations. Good reporting preserves metadata because it allows others to interpret the numbers correctly.

Unit of analysis

The unit of analysis is the entity represented by one observation. It might be a hospital admission, not a patient; a payment, not a supplier; or a complaint, not a complainant. Confusing units can produce false conclusions. For example, ten complaints may concern one incident, and ten payments may belong to one contract.

Population, sample, variable, and indicator

The population is the complete set relevant to a question. A sample is a subset examined when the full population is unavailable or impractical. A variable is a characteristic that can differ across observations. An indicator is a measured value used to represent a condition, such as the percentage of inspections completed on time.

Data provenance

Provenance is the documented history of data: where it originated, who handled it, what transformations were applied, and which version was used. Provenance is essential for reproducibility and correction.

2. From Public-Interest Question to Investigative Hypothesis

Strong investigations begin with a question rather than a preferred conclusion. “Is public money being allocated fairly?” is a broad starting point. It becomes more researchable when narrowed to a place, period, institution, and measurable issue: “Did a city’s emergency procurement during 2022–2023 disproportionately benefit suppliers with political or familial connections, after accounting for contract value and supplier capacity?”

A useful investigative hypothesis is provisional and testable. It identifies:

• the suspected pattern or mechanism;
• the population and time period;
• the evidence that could support or weaken the claim;
• alternative explanations;
• the public-interest significance.

Journalists should write a hypothesis log. Each entry records the date, claim, supporting evidence, contrary evidence, confidence level, and next test. This reduces confirmation bias—the tendency to notice evidence that supports an initial belief while discounting disconfirming evidence.

A practical framing model is:

Question → Claim to test → Evidence needed → Sources → Verification → Public explanation

The evidence plan should distinguish primary evidence from secondary evidence. A procurement contract, court filing, inspection record, or original dataset is generally primary evidence. A press release or another news report may identify leads but should not automatically establish the underlying fact.

3. Finding and Obtaining Evidence

Open sources and public records

Potential sources include government budgets, procurement portals, corporate registries, legislative records, court documents, land records, inspection databases, environmental permits, election-finance disclosures, academic repositories, scientific data, and annual reports. Search systematically using alternative spellings, organization names, registration numbers, dates, addresses, and associated individuals.

Do not assume that a document’s presence online proves completeness or accuracy. Check the publishing agency, date, version, scope, and definitions. Download original files where possible and preserve the URL, access date, and file hash or local archive according to newsroom policy.

Freedom-of-information processes

Freedom-of-information (FOI), access-to-information, and public-records laws differ by jurisdiction. A responsible request identifies the legal authority, describes records precisely, specifies a date range and format, and avoids asking an agency to create a new analysis unless the law requires it to do so.

A useful request may ask for:

• the underlying records rather than a summary;
• machine-readable format, such as CSV, where available;
• data dictionaries and codebooks;
• correspondence concerning a defined decision;
• records of exemptions, deletions, or redactions;
• the retention schedule if records are said not to exist.

Keep a request log containing submission date, statutory deadline, agency contact, clarification requests, fees, response, appeal deadline, and outcome. Narrow, well-defined requests usually produce better results than vague demands for “all information.” Journalists should understand exemptions protecting personal privacy, national security, confidential commercial information, law-enforcement methods, and privileged communications. An exemption should be examined, not simply accepted; appeals and public-interest tests may be available.

Human sources and source protection

Human sources can explain institutional practice, identify missing records, and interpret technical evidence. Source mapping should record a source’s role, access, expertise, motivation, relationship to the subject, possible bias, and corroborating evidence. A source’s motive does not make information false, but it affects how the information should be tested.

Promises of anonymity require careful editorial approval. Before granting confidentiality, establish what can be published, what identifying details may create risk, whether the source has direct knowledge, and whether the information can be independently corroborated. Secure communication, access controls, encrypted storage, and minimal retention reduce risk. Never promise protection that the journalist or organization cannot realistically provide.

4. Data Preparation and Spreadsheet Practice

Analysis begins before calculation. Preserve the raw file as read-only and create a working copy. Record every transformation in a data diary. Common preparation steps include:

1. Inventory: list files, fields, dates, formats, and missing elements.
2. Standardization: make dates, names, categories, currencies, and units consistent.
3. Deduplication: identify repeated records without deleting legitimate repeated events.
4. Validation: check ranges, impossible dates, negative values, totals, and category codes.
5. Documentation: retain original values when correcting or recoding fields.
6. Reproducibility: use formulas, queries, or scripts that can be rerun and reviewed.

A spreadsheet should have separate sheets for raw data, cleaned data, lookup tables, calculations, and outputs. Avoid typing calculated results over source values. Use stable identifiers, not row numbers, when joining files. A join can create duplicate rows if one identifier appears multiple times in either table; journalists must inspect the relationship before interpreting totals.

Missing data requires classification. A blank may mean “not recorded,” “not applicable,” “withheld,” “zero,” or “unknown.” These meanings are not interchangeable. If a dataset records only reported incidents, it cannot establish the true incidence of all incidents.

Basic statistical measures

The arithmetic mean of n values is:

Here, xᵢ is an observation, n is the number of observations, and x̄ is the mean. The mean is appropriate when values are measured on a meaningful numerical scale and extreme values do not distort interpretation.

The median is the middle value after sorting observations. It is often preferable for skewed data, such as salaries or property prices. The range is the maximum minus the minimum. The percentage change from an original value to a new value is:

The original value must not be zero, and the direction of comparison must be stated. A rise from 80 to 100 is a 25% increase, while a fall from 100 to 80 is a 20% decrease; the denominators differ.

A rate relates an event to a relevant population or exposure:

Here, k is a reporting multiplier such as 1,000 or 100,000. The numerator and denominator must refer to compatible places, periods, and populations. Rates are often more informative than raw counts, but small denominators can produce unstable results.

A weighted average is appropriate when observations contribute unequally:

Here, xᵢ is a value and wᵢ is its weight. The journalist must explain why the weights are appropriate. A simple average of district percentages may misrepresent a citywide percentage if districts have different population sizes.

Worked example: interpreting a percentage

A public agency recorded 240 workplace inspections in one year and 300 the next. The percentage change is:

((300 − 240) ÷ 240) × 100% = 25%

Thus, inspections increased by 25%. This does not prove that workplace safety improved. The number of violations, inspection resources, workplaces, selection criteria, and reporting practices must also be examined. The example demonstrates a central principle: a correct calculation can still support an incorrect interpretation if the denominator and causal context are ignored.

5. Document Analysis and Source Mapping

Investigations often depend on large collections of documents. Begin with a document inventory containing title, date, author, recipient, document type, file location, relevance, and verification status. Read strategically: identify repeated names, dates, decisions, unexplained gaps, changes in language, and references to missing attachments.

A chronology is one of the most effective investigative tools. Place decisions, payments, meetings, warnings, incidents, and public statements on a common timeline. Chronology can expose contradictions—for example, a public claim that a decision was made in March when internal records show that negotiations began in January.

Entity resolution links different references to the same person, company, property, or institution. Names may vary because of abbreviations, transliteration, mergers, or spelling errors. Use additional identifiers such as registration numbers, addresses, directors, dates, and bank details. Never merge entities solely because names look similar.

A source map represents relationships among people, organizations, transactions, documents, and events. It can be drawn manually or built with a spreadsheet. Each connection should have a source and confidence level. A visually impressive network is not proof; it is a way to identify relationships requiring verification.

Triangulation means testing a material claim with independent forms of evidence. For example, a claim about an unsafe building may be examined through inspection reports, photographs, tenant interviews, permit records, and expert assessment. Repetition of one unverified claim across many outlets is not independent corroboration.

6. Investigative Frameworks and Reasoning

The evidence matrix

An evidence matrix connects each intended statement to supporting material.

The matrix distinguishes what is known from what is suspected. It also prevents a dramatic allegation from being published before the necessary evidence exists.

The “who, what, when, where, why, and how” model

The traditional reporting questions remain useful, but investigative work adds mechanism and accountability. Who made or benefited from a decision? What happened? When and where? Why did the system permit it? How did the action produce harm or advantage? Which rule, duty, or public expectation applies? Who was asked to respond, and what evidence did they provide?

Alternative hypotheses

For every suspected pattern, list plausible alternatives. Concentrated contracting might result from favoritism, but it might also reflect a specialized market, emergency rules, or the absence of qualified competitors. Testing alternatives strengthens fair reporting and reduces overstatement.

7. Communicating Data-Driven Findings

A published investigation should answer the audience’s central question early, then show how the conclusion was reached. Explain the dataset’s origin, period, unit of analysis, exclusions, calculations, and limitations. Distinguish reported facts, analysis, allegations, and verified conclusions.

Choose a form that fits the evidence. A table may be clearer than a chart when exact values matter. A line chart can show change over time; a bar chart can compare categories; a map can reveal geographic distribution but may mislead when areas differ greatly in population. Avoid truncated axes, decorative 3D effects, excessive precision, and color schemes inaccessible to readers with visual impairments.

If a chart shows rates, label the denominator and multiplier. If a visual compares money, specify nominal or inflation-adjusted values and currency. Explain suppressed values, missing records, revised figures, and uncertainty. A methodology box or linked notebook can make the work auditable without overwhelming the main narrative.

Right of reply is both an ethical and practical safeguard. Contact subjects with specific, evidence-based questions and a reasonable deadline. Do not reveal confidential sources unnecessarily. Record the response accurately, distinguish refusal from denial, and publish material corrections promptly.

8. Ethics, Law, and Governance

The core ethical principles are accuracy, independence, minimizing harm, fairness, accountability, and transparency. Data work creates distinctive risks:

• Privacy: combining harmless fields can identify individuals.
• Re-identification: supposedly anonymous records may become identifiable when joined with other datasets.
• Stigmatization: maps and rankings can unfairly label neighborhoods or groups.
• False precision: exact-looking figures may conceal measurement error.
• Consent and vulnerability: victims, children, whistleblowers, and marginalized sources may face disproportionate risk.
• Security: leaked records may expose passwords, addresses, medical information, or source identities.

Data protection law may regulate personal data, sensitive categories, retention, transfer, and automated processing. Defamation law may apply when a publication harms reputation through an untrue factual assertion. Copyright, confidentiality, court restrictions, and secrecy laws may also affect publication. Legal advice should be sought for high-risk investigations; journalists should not assume that public availability eliminates privacy or defamation concerns.

A public-interest justification does not excuse careless handling. Apply data minimization: retain and publish only what is necessary. Remove irrelevant personal details, restrict access, encrypt sensitive files, and establish deletion rules. Editorial leaders should document risk assessments and publication decisions.

9. Technology, Automation, and AI

Modern investigations may use optical character recognition, spreadsheet queries, geographic information systems, database tools, version control, and programming languages. Automation can sort millions of records, identify duplicate names, extract dates, or flag anomalies. It cannot by itself establish motive, legality, causation, or truth.

AI systems can assist with transcription, classification, translation, document search, and anomaly detection. However, they may hallucinate facts, reproduce bias, misread scans, leak confidential information, or obscure the reasoning behind a result. Journalists should:

1. use approved tools and never upload confidential material to an unknown service;
2. preserve the original document and generated output;
3. verify every material claim against primary evidence;
4. disclose significant automated assistance where editorial policy requires it;
5. test error rates on representative samples;
6. maintain human responsibility for publication decisions.

A model’s probability score is not proof. Automated flags are leads for human investigation, not conclusions.

10. Applied Case: The Emergency Procurement Files

This is a fictional illustrative case designed for learning.

Background

The fictional city of Riverton published a spreadsheet of emergency medical-supply contracts issued during a disease outbreak. A reporter noticed that one supplier, Northstar Medical, received 38% of the recorded contract value.

Problem

The initial observation raised questions but did not establish favoritism. The reporter needed to determine whether Northstar’s share was unusual, whether the data was complete, how contracts were awarded, and whether any public harm resulted.

Analysis

The reporter preserved the original file and obtained the procurement policy, contract notices, amendments, supplier-registration records, bid evaluations, meeting minutes, and audit correspondence through public sources and a targeted FOI request. The spreadsheet contained 412 rows, but inspection showed that some rows were amendments rather than new contracts. After defining the unit of analysis as the original contract, the reporter grouped amendments with their parent contracts and checked totals against the city’s published expenditure report.

The reporter then created an evidence matrix and compared Northstar with other suppliers by contract count, total value, delivery period, product category, and emergency-procurement justification. A high contract share remained after corrections, but the supplier specialized in a product category that represented a large proportion of emergency spending. This was an alternative explanation.

Document analysis identified that a procurement official had previously worked for Northstar. That fact alone did not prove improper influence. The reporter sought appointment records, conflict-of-interest declarations, evaluation-panel minutes, emails, and interviews. Records showed that the official disclosed the former employment but did not participate in the final evaluation. However, two required declarations were missing from the file, and an audit memo noted inconsistent documentation across several awards.

The reporter contacted the city, Northstar, the official, medical staff, and an independent procurement expert. The city supplied additional records and corrected two publication errors. Northstar explained its capacity and provided delivery documentation. Medical staff confirmed that most supplies arrived on time, although one shipment was late.

Outcome

The defensible story was not “official secretly awarded contracts to a former employer.” The evidence supported a more precise conclusion: emergency procurement was highly concentrated in one specialized supplier; disclosure and recordkeeping were inconsistent; and documentation gaps made independent oversight difficult. The report published the methodology, distinguished verified facts from unresolved questions, included responses, and avoided publishing unnecessary personal information.

Learning points

1. An anomaly is a lead, not a verdict.
2. Unit-of-analysis errors can substantially change findings.
3. Alternative explanations improve accuracy.
4. Conflict of interest requires evidence about disclosure, participation, influence, and rules—not merely association.
5. Corrections and responses increase credibility rather than weaken a well-supported story.

11. Challenges and Best Practices

Common challenges include incomplete records, incompatible formats, changing definitions, delayed FOI responses, inaccessible PDFs, deliberate obfuscation, small samples, survivorship bias, and pressure to publish before verification. The following practices reduce risk:

• define the question and unit of analysis before calculating;
• preserve raw data and document transformations;
• inspect a sample manually before trusting automation;
• check totals against an independent source;
• separate counts, rates, percentages, and monetary values;
• test alternative explanations;
• seek expert review for technical claims;
• maintain a source and request log;
• protect vulnerable sources and personal data;
• offer meaningful right of reply;
• publish methodology, limitations, and corrections;
• use the least sensational accurate wording.

Chapter Summary

Data and investigative journalism combine structured evidence with sustained public-interest reporting. Effective work begins with a focused, testable question and proceeds through source discovery, records access, data preservation, cleaning, analysis, document review, triangulation, and fair communication. Journalists must understand units of analysis, missing data, denominators, percentage change, rates, means, and weighted averages. They should use evidence matrices, timelines, source maps, hypothesis logs, and alternative explanations to control bias and distinguish suspicion from proof. Ethical practice requires privacy protection, source security, proportionality, transparency, right of reply, and correction. Technology and AI can expand investigative capacity, but human verification and editorial accountability remain essential.

Key Terms

Data journalism: Journalism that uses structured data to discover, test, explain, or communicate findings.

Investigative journalism: Sustained original reporting that examines concealed, complex, or consequential matters.

Public interest: Information relevant to rights, safety, accountability, public resources, or democratic decision-making.

Dataset: An organized collection of observations or records.

Metadata: Information describing a dataset’s origin, definitions, structure, and limitations.

Unit of analysis: The entity represented by one observation or row.

Variable: A characteristic that can differ among observations.

Indicator: A measured value used to represent a condition or trend.

Data provenance: The documented history and transformations of data.

Freedom of information: A legal process for requesting records held by public bodies.

Data cleaning: Detecting, correcting, standardizing, and documenting data problems.

Triangulation: Testing a claim through independent evidence or methods.

Source mapping: Systematic recording of sources, roles, access, motives, and corroboration.

Entity resolution: Determining whether different references identify the same entity.

Evidence matrix: A table linking claims to required evidence and verification status.

Denominator: The reference population or quantity used in a rate or percentage.

Re-identification: Linking supposedly anonymous data to identifiable individuals.

Right of reply: A fair opportunity for subjects of reporting to respond to material allegations.

Data minimization: Collecting, retaining, and publishing only necessary information.

Confirmation bias: The tendency to favor evidence supporting an existing belief.

Review Questions

Short Answer Questions

1. What distinguishes data journalism from investigative journalism?
2. Why is the unit of analysis important when analyzing a dataset?
3. State two purposes of metadata.
4. What is triangulation, and why is it valuable?
5. Identify two risks associated with publishing personal data.

Long Answer Questions

1. Explain the process of developing an investigative question into a testable hypothesis and evidence plan.
2. Describe a rigorous workflow for cleaning and documenting a spreadsheet used in an investigation.
3. Explain how FOI requests, human sources, and public records can complement one another.
4. Discuss how a journalist should communicate statistical findings without overstating their meaning.
5. Evaluate the ethical and editorial safeguards required when using AI in investigative reporting.

Analytical / Application Questions

1. A city reports 120 complaints in one year and 150 the next. Calculate the percentage change and explain what additional information is needed before concluding that the underlying problem worsened.
2. A supplier receives 60% of emergency contracts but operates in a product category representing 70% of total emergency spending. Identify at least three alternative explanations and design a verification plan.
3. You receive a dataset containing names, addresses, dates, and complaint categories. Propose a publication strategy that communicates the public-interest finding while reducing re-identification and stigmatization risks.

Multiple Choice Questions with Answers

Question 1

Which statement best describes the relationship between data journalism and investigative journalism?

A. Data journalism replaces the need for interviews and documents.

B. Investigative journalism uses only confidential sources.

C. Data can help discover and test patterns, while investigative reporting supplies context, verification, and accountability.

D. The two practices are identical in method and purpose.

Correct Answer: C. Data can help discover and test patterns, while investigative reporting supplies context, verification, and accountability.

Question 2

A dataset contains one row for each payment, but a reporter treats each row as a separate contract. What is the main risk?

A. Using a wrong color scale

B. Confusing the unit of analysis and overstating the number of contracts

C. Calculating a median instead of a mean

D. Using too many source interviews

Correct Answer: B. Confusing the unit of analysis and overstating the number of contracts

Question 3

A value rises from 80 to 100. What is the percentage change using 80 as the original value?

A. 20%

B. 25%

C. 80%

D. 125%

Correct Answer: B. 25%

Question 4

Which practice most improves reproducibility in spreadsheet-based reporting?

A. Overwriting the original file with corrected values

B. Keeping raw data, documenting transformations, and using repeatable calculations

C. Relying on memory for cleaning decisions

D. Publishing only the final chart

Correct Answer: B. Keeping raw data, documenting transformations, and using repeatable calculations

Question 5

Why can a rate be more informative than a raw count?

A. A rate relates events to a relevant population or exposure.

B. A rate always proves causation.

C. A rate eliminates all uncertainty.

D. A rate does not require a denominator.

Correct Answer: A. A rate relates events to a relevant population or exposure.

Question 6

What is the principal purpose of an evidence matrix?

A. To rank journalists by productivity

B. To connect intended claims with supporting evidence and verification status

C. To encrypt confidential files

D. To replace the right of reply

Correct Answer: B. To connect intended claims with supporting evidence and verification status

Question 7

Which is the strongest example of triangulation?

A. Repeating an allegation from several articles that cite the same source

B. Using one anonymous social-media post

C. Comparing an inspection record with photographs, interviews, and an expert assessment

D. Assuming a government press release is complete

Correct Answer: C. Comparing an inspection record with photographs, interviews, and an expert assessment

Question 8

What does data minimization require?

A. Publishing every field in a dataset

B. Collecting, retaining, and publishing only information necessary for the legitimate purpose

C. Deleting all source material immediately

D. Avoiding all public records

Correct Answer: B. Collecting, retaining, and publishing only information necessary for the legitimate purpose

Question 9

Why should an AI-generated anomaly be treated as a lead rather than a conclusion?

A. AI systems cannot process structured data

B. An anomaly score establishes legal guilt

C. Automated systems may contain errors or bias and require human verification

D. AI output is always confidential

Correct Answer: C. Automated systems may contain errors or bias and require human verification

Question 10

Which factor is essential when interpreting a percentage or rate?

A. The denominator and the period or population to which it refers

B. The visual size of the chart

C. The number of paragraphs in the article

D. Whether the figure has a decimal point

Correct Answer: A. The denominator and the period or population to which it refers