Corruption is designed to be invisible. Bribes come without receipts, and both sides have every reason to keep quiet. So how can anyone say that one country is more corrupt than another, or that corruption is rising? Researchers have built three broad kinds of measure, each useful and each flawed.

Perception indices

The best known is the Corruption Perceptions Index (CPI), first published by Transparency International in 1995. It ranks around 180 countries and territories on a scale from 0, highly corrupt, to 100, very clean. The CPI does not count bribes. It measures perceptions of public sector corruption by combining about a dozen surveys and expert assessments, from sources such as risk-rating firms, development banks and surveys of business executives. A country needs at least three sources to be scored. The World Bank's Worldwide Governance Indicators, developed by Daniel Kaufmann, Aart Kraay and colleagues, include a similar composite called Control of Corruption.

A low CPI score means corruption is perceived as widespread. Year after year, the Nordic countries, New Zealand and Singapore sit near the top, while countries torn by war or state collapse, such as South Sudan and Somalia, sit near the bottom. The index has real strengths. It covers almost every country, it is easy to understand, and it has pushed governments and journalists to talk about a subject many preferred to avoid.

Limits of perceptions

Perceptions are not facts. Experts may judge a country partly by its economic success, assuming a rich country must be clean. They may react to headlines, which creates a paradox: exposure raises perceived corruption even when it shows accountability working. A country that prosecutes a corrupt minister may see its score fall, while a country that hides its scandals looks fine. Small year-to-year movements are often within the margin of error, so a change of one or two points usually means little.

The CPI also focuses on bribery inside a country's public sector. It says little about the banks, lawyers and property markets of wealthy countries that help others hide stolen money. Denmark regularly tops the index, yet the Estonian branch of Danske Bank, Denmark's largest bank, handled roughly 200 billion euros of payments from 2007 to 2015, much of it suspicious and linked to Russia and other former Soviet states. Clean-scoring countries can still host enormous dirty money flows.

Experience surveys

A second approach asks ordinary people what has happened to them. Transparency International's Global Corruption Barometer asks whether people personally paid a bribe recently, typically in the past twelve months, when using services such as police, courts, schools or clinics. These surveys capture petty corruption well, and the results can be surprising: in some countries with poor CPI scores, few people report paying bribes, and the reverse also happens. But experience surveys miss grand corruption, which ordinary people never see, and respondents may not admit to breaking the law. Researchers use indirect question techniques to reduce that problem. Comparing the two kinds of measure is itself informative: where experts see heavy corruption but citizens report few bribes, the problem may sit at the top, in contracts and budgets, rather than at the clinic counter.

Hard measures

A third approach looks for physical traces of theft. In a famous study of more than 600 village road projects in Indonesia, the economist Benjamin Olken sent engineers to dig core samples from finished roads and estimate the materials actually used. He measured corruption by comparing reported spending with engineers' estimates of materials used. On average, about a quarter of the reported spending could not be accounted for. When villages were told that the chance of a government audit would rise from 4 percent to 100 percent, the missing share fell by about 8 percentage points. The study is valuable precisely because it does not depend on anyone's opinion or memory; the road itself is the witness.

Another clever study used parking tickets. Until late 2002, diplomats at the United Nations in New York could ignore parking fines because of diplomatic immunity. Economists Raymond Fisman and Edward Miguel found that diplomats from more corrupt countries left more fines unpaid, even though nobody could punish any of them. Norms, it seemed, traveled with people. Level 3 returns to what happened when enforcement arrived.

Worked example

Suppose a government announces that its CPI score rose from 38 to 41 after a new anticorruption law, and claims victory. A careful reader asks several questions. Is a three-point rise outside the margin of error? Which underlying surveys moved, and were they surveys of foreign investors, who may react mainly to economic news? Do experience surveys show fewer people paying bribes at clinics and police stations? Have audit reports found less missing money in public works? If the experience surveys are flat and the audits still find a quarter of road budgets vanishing, the higher score probably reflects mood rather than change. The reverse case deserves equal care: a falling score after a wave of prosecutions may mean the system has finally started catching people. No single number settles the question. The strongest conclusions come when perceptions, experiences and hard evidence all point the same way.