The new Handbook of Research Methods for Corruption Studies, edited by corruption guru extraordinaire Michael Johnston, offers a wealth of learning about how to fight corruption. None more important than Alina Mungiu-Pippidi’s chapter analyzing the fundamental question in the fight: how do we know if we are winning?

Decades after corruption control rose to the top of the global agenda, assaying how well the fight is going ought to be straightforward. That’s not so as the continuing debate about the validity of TI’s Corruption Perceptions Index, the World Bank’s Control of Corruption measure, and the many other measurement tools on offer shows.

The problem, as Professor Johnston acknowledges in the Handbook’s introduction, is foundational.

“After 40 years of debate about corruption, we are still uncertain about what we are talking about. . . cannot measure it directly. [and don’t know if] a single [corruption] score about a whole country really tells us much.”

What Professor Mungiu-Pippidi shows in her chapter is that the foundational problem is behind our inability to measure progress in fighting corruption. That efforts to gauge success have foundered on misguided attempts to propound a universal, all-encompassing definition of corruption which have produced nothing more than vague, imprecise measures riddled with practical and methodological errors. She explains why these should be scrapped and offers in their stead a series of direct, fact-based indicators for evaluating progress tied to a specific, bounded definition.

Professor Mungiu-Pippidi’s chapter deserves the closest attention by policymakers, evaluators, and citizens who want to know if an anticorruption law, policy, or strategy is making a difference. This post builds on and complements her work in two ways.

First, it provides an additional, persuasive reason why a universal definition of corruption is beyond reach. Second, while her chapter discusses the use (and mostly misuse) of corruption measurement tools to evaluate anticorruption interventions, this post expands the discussion to include other ways in which corruption measures are used and abused.

Defining corruption

Those who negotiated the U.N. Convention Against Corruption invested considerable time and effort trying to define what the convention would be against. Eventually, they gave up, concluding there was no single, all-purpose formulation capable of garnering widespread acceptance. “Corruption,” the Convention’s drafters explained, “is a fluid concept, signifying different things to different people” (here). Accordingly, UNCAC does not define corruption. Rather, it enumerates specific acts that government must criminalize to join the Convention.

The UNCAC drafters’ failure to devise a definition of corruption has not stopped others from trying. Sussex University professor Becky Dobson Phillips lists 117(!) different attempts to date (here). The most widely accepted is Transparency International’s: “the abuse of entrusted power for private gain.” Concise but circular. The dictionary definition of “abuse” is “improper” or “corrupt.” 

The problem is that, like “improper” and “abuse,” corruption expresses a value judgment. A speaker who claims a person acted “improperly” or “abusively” or “corruptly” is, implicitly if not explicitly, comparing the person’s actions against how the speaker believes the individual should have acted. Variously termed evaluative, subjective, or normative, such statements are contrasted with those that provide a neutral, value-free description of a thing or event, descriptions that are objectively verifiable (here).

The confusion arises because “corruption” originated as a descriptive term, and some uses remain descriptive. In many languages its root is the ancient Greek phthora, to decay or die, and the Greeks used it to describe matters such as the condition of a plant (here). A statement that a plant is dying is not a value judgment about the plant. It is a simple assertion of its condition, a fact that can be verified. “Corruption” in English is in some cases still used descriptively. When a file sent over the internet cannot be opened, it is said to be “corrupted.” A readily verifiable fact.

Plato used phthora to describe a government that was in decline, that no longer met the criteria for a just or ideal government (Republic 546a). For him, that was not a judgment about the government’s condition. Plato posited a single, ideal form of government; so any deviation from it – whether it was the rich taking control or the demos ruling — described an objective, demonstrable fact. Now that there is no longer a consensus about what constitutes an ideal government, a speaker calling a government corrupt today is necessarily making a judgment, asserting the government in question deviates from what the speaker considers the ideal or best government.

The judgment inherent in the term “corruption” is obscured because some practices – notably bribery and embezzlement – are universally condemned. Terming them “corrupt” no longer denotes passing judgment on whether they are wrongful or not. Just as with the question of whether a plant is dying or not, bribery and embezzlement are questions of fact. A bribe was paid or it was not; money embezzled or it was not. UNCAC’s drafters may not have appreciated the etymology of the term corruption, but it is behind why they abandoned the search for an all-encompassing definition in favor of a conduct-based one. And why the only two forms of conduct UNCAC mandates parties treat as corruption offenses are bribery and embezzlement. (The drafters discussed other offenses, but lacking a consensus punted, urging parties “to consider” making them a crime.)

Definition matters for measurement

With no universally agreed definition of corruption, attempts to measure it across different nations — without specifying the conduct considered corrupt — are meaningless. Comparisons work only, as the adage goes, when apples are being compared to apples and oranges to oranges. Comparing results using measures that leave it undefined or define it differently, or with survey responses that leave it to the respondent to define it, is making an apples to oranges comparison. 

Some composite measures try to elide the problem by using statistical techniques to combine disparate measures into a single number, an “indicator” of corruption. Best known is the Worldwide Governance Indicator’s “control of corruption” (here). It aggregates responses to questions commercial organizations, media outlets, and NGOs pose about corruption in various countries, questions that range from those asking about instances of corrupt conduct, such as the frequency of bribe requests, to those asking about corruption generally. Responses from all surveys taken in a country are combined to produce a single number. According to WGI, the higher the number, the better job the government is doing curbing corruption.

As Bollen explains, this is one way to define an abstract concept like corruption: a collection of measures that together constitute corruption. But whether in fact the answers to a variety of questions about corrupt conduct define corruption is not something to be taken on faith. Rather, as Melissa Thomas made plain in her critique of the WGI, simply a hypothesis. And like any hypothesis, its acceptance is warranted only if it survives empirical testing.

That it cannot do. For, as explained above, corruption is a value-laden term, one about which individuals differ. Hence, there is no objective, empirical test an indicator claiming to define “corruption” could pass. Just as UNCAC drafters could not find a definition of corruption all could agree on, agreement on what survey responses should be aggregated is out of reach. Corruption is like other value-laden concepts such as “social justice” and “democracy.” It is what Scottish political theorist and philosopher W. B. Gallie dubbed an “essentially contested concept” (here). These are concepts or ideas that cannot, he explained, be validated because reasonable people will never converge on a single meaning. 

Cross-national and within-country corruption measures

One reason to measure corruption is to compare its incidence across different nations. An example is the World Bank’s Enterprise Surveys. Firm managers across some 200 countries and territories are asked whether they had to pay a bribe to obtain permission to do business, obtain a water or power connection, and the like (here). This allows investors to factor in the risk of bribery when deciding where to build a factory or open an office. The results can also prompt countries where bribe requests are high to step up enforcement of their antibribery laws.

A second use is solely for advocacy: to spur authorities in countries scoring poorly on the measure to increase their efforts to combat corruption. This is the raison d’être of Transparency International’s Corruption Perceptions Index. With such measures, accuracy is beside the point; the purpose is to garner attention and stir debate, something on which the CPI has been spectacularly successful. Accuracy only becomes an issue when a country contests its score.

It is one thing to serve as a stimulus to action, and quite another to serve as a measure of the effectiveness of that action. If the CPI and similar measures were used only for advocacy, they would not be problematic. But they are not. The trouble is they are often used to gauge whether an anticorruption policy or strategy is working by tracking year-to-year changes in corruption. This misuse of such measures continues despite warnings by UN agencies (here, here), Sweden’s Quality of Government Institute (here), and numerous academics (examples here, here, and here) that this is not the purpose of these measures and using them for it produces much harm and little if any good.

Measuring corruption within a country obviates one of the biggest challenges with measuring corruption across countries: the context problem. As Professor Mungiu-Pippidi explains in her chapter, measurements within a country eliminate the need to take account of (control for in social science lingo) different conditions in different nations. In authoritarian states, for example, citizens may be reluctant to report corruption and government officials hesitant to act.  In states that strictly enforce laws against bribery, citizens may be less likely to be candid when surveyed about whether they have ever paid a bribe. 

Within-country measures serve two purposes. First, policymakers need to gauge changes over time. The UNODC’s guide to national anticorruption strategies reports that most national anticorruption strategies fail because no effort is made to measure their impact over time (here).  Without measures of what is working and what is not, policymakers do not know whether mid-course corrections are necessary. Similarly with determining the effectiveness of individual policies. Has stepping up audits of large infrastructure projects reduced overbilling? Or simply added to project cost? Moreover, without some measure or measures of change over time, how can policymakers hold an anticorruption agency accountable for combatting corruption?

A second use of within-country measures is to help policymakers allocate resources to combat corruption. A telling, if often overlooked, example is the one Malcolm Sparrow relates about the U.S. Internal Revenue Service’s first effort to measure corruption in a program of cash payments to low-income families. In 1993 the program disbursed close to $15 billion, and the Service’s detection system showing fraudsters had bilked the government of $136 million. Absent any measure of how much fraud and corruption the program suffered, IRS officials had no way to know whether their detection system was adequate.

The Service employed an innovative measure tool that combined surveying recipients with combing through public records of property ownership. It showed losses from fraud and corruption in 1993 were close to $3 billion, 21 times greater than what the existing system had caught. With Congressional approval, the IRS introduced new policies to reduce the program’s vulnerability to fraud, invested millions in new equipment to better detect corrupt schemes, and sharply increased staffing levels in its antifraud and corruption unit.

Types of measures

Corruption measures are of two kinds: i) those based on surveys and ii) those derived from transactional or administrative data.

Survey-based measures divide further into i) perception measures, those where respondents (citizens, businesspeople, or country experts) are asked for their opinions or impressions of an institution or country, and ii) experience measures, those where respondents are asked about their own direct encounters with bribery or extortion.

Transparency International’s Global Corruption Barometer is an example of a perception survey. Respondents are asked: “How much of a problem, if at all, is corruption in the national government?” The surveyor gives respondents a range of answers from “No problem at all” to “A very big problem” to choose from. Although, as explained above, differences in the definition of corruption across countries make comparing answers useful only for advocacy, citizens of the same country are likely to have common views about what is corrupt. Their answers can thus be helpful when comparing a government’s performance over time, one of many measures to include in an evaluation.

The U.N. Office on Drugs and Crime’s well-known crime victimization surveys are an example of experience surveys. Among the questions asked is whether, in seeking a public service in the last 12 months, the respondent “actually made an informal payment, gave a gift, or performed a favor” in return, code words for corrupt conduct. A second example is the World Bank’s Enterprise Surveys, described above. Such “experiential” surveys are generally thought to be more accurate than perception surveys in gauging corruption because they provide information on actual events. Care is required in comparing results across nations, however. Again, as noted above, respondents in one country may be more reluctant to admit paying a bribe than those in another (here).

The number of convictions for bribery or embezzlement, audits showing overbilling on infrastructure, and whistleblower complaints all provide measures of corruption. Termed transactional and administrative measures, they are considered “harder” than those using surveys because they are based on directly observed conduct. While investigations, prosecutions, and convictions for bribery or embezzlement are the most used, criminal investigations have produced other useful data: in one a bribetaker left a paper trail of illicit payments received, in another a maritime company listed bribes paid at a port of entry. Also useful are discrepancies between two independent data sources that in a corruption-free world would match. Examples here include i) trade statistics between exporting and importing countries, ii) funds disbursed by a central government against funds received by local schools, and iii) bid prices before and after an anticorruption crackdown. (The UNODC’s Statistical Framework to Measure Corruption contains a comprehensive list of sources of administrative data (here).)

Survey measures are cheaper and more easily replicated across countries and over time, which explains their popularity. But to the problems noted above there is the added one of bias. A landmark study by Daniel Treisman found there is often little statistical relationship between the level of corruption surveys of outside experts — principally from wealthy, liberal democratic regimes – report, and actual citizen experiences paying bribes, a result he explains because perceptions are heavily swayed by a country’s economic wealth and level of democracy rather than actual illicit acts (here). Similar findings hold for a set of eight African countries. Experts surveyed, again from wealthy democratic states, thought bribery was far more prevalent than citizens reported (here).

Measures based on transactional and administrative data can be very precise. In a widely praised example, Benjamin Olken commissioned a study of the cost of road construction in Indonesia where engineers took samples to assess the amount of gravel and other materials actually used, workers were surveyed to determine wages paid, and suppliers were queried on prices for materials. The results allowed him to pinpoint which roads were corrupted and how (here).

Measures using transactional and administrative data have their own drawbacks, however. Some, which simply collate already existing data, as Uganda has done (here), are relatively inexpensive; others, which require combing databases from different sources, and collecting new data, are often time-consuming and expensive. Also, they measure specific, often narrow and limited phenomena — overbilling on government contracts found through audits, for example. There are also completeness issues: How many bribes went undetected?

Choosing the right measure

All measurement problems start from the same point. What do we want to measure and why? With corruption, the first hurdle is definition. As decades of experience have  shown, and Professor Mungiu-Pippidi reiterates, the drafters of the UN Convention Against Corruption got it right in the first place. Corruption has meaning only when tied to specific forms of conduct: bribery, embezzlement, and other forms of wrongful conduct that the Convention identifies.

The second step is to determine the purpose of the measure. To provide investors information about bribery risks in different countries. To prompt a government or governments to crack down on corruption. Or is the purpose to evaluate the effectiveness of a policy or set of policies. Or determine whether an agency or sector needs to commit more resources to corruption prevention.

Once these questions are answered, the challenge is to choose the right measurement tools and adapt them to the task at hand. As the UNDP/Global Integrity measurement guide explains (here), to do so policymakers and their advisers must bring creativity, technical skill, and commitment to the task. If corruption is to be overcome, citizens must demand nothing less.

In addition to the links in the text, this note relies on these GAB posts —

• Bello y Villarino, J.-M., “The Problem with Anticorruption Diagnostic Tools Is Not (Primarily) Too Much Standardization,” Global Anticorruption Blog, August 14, 2018 (guest post).
• Messick, R., “Corruption Measurement: A Primer,” Global Anticorruption Blog, June 18, 2014.
• Messick, R, “Why Is Corruption So Hard to Define?” Global Anticorruption Blog, August 5, 2015.
• Stephenson, M., “In Bribery Experience Surveys, Should You Control for Contact?” Global Anticorruption Blog, September 19, 2017.
• Shipley, T., “The Need for Better Monitoring and Evaluation of Anticorruption Projects,” Global Anticorruption Blog, July 16, 2024 (guest post).

For more information see —

• UNODC & UNDP, Manual on Corruption Surveys: Methodological Guidelines on the Measurement of Bribery and Other Forms of Corruption through Sample Surveys (Vienna, 2018). A guide to measuring bribery and other corruption through population- and business-based sample surveys, including methodological guidance on SDG indicators 16.5.1 and 16.5.2.
• UNODC, Statistical Framework to Measure Corruption (Vienna, 2023). Presents some 150 transactional and administrative indicators covering criminal offenses under UNCAC, preventive measures, and the broader enabling environment, developed through consultation with national statistical offices and academic and international experts.
• UNDP/Global Integrity, User’s Guide to Measuring Corruption and Anti-Corruption (2015). In-depth discussion of different measures and how they have been used by different countries.
• Maslen, C., How-to Guide for Corruption Assessment Tools 3^rd ed. (U4, 2023). Overview of publicly accessible tools assessing country-level corruption/governance performance, including newer additions like V-Dem, the Basel AML Index, Berggruen Governance Index, FATF Mutual Evaluations, and the TRACE Bribery Index.
• Hart, E., Guide to Using Corruption Measurements and Analysis Tools for Development Programming (U4, 2019) emphasizes value of transactional and administrative data, targeted surveys, and bespoke proxy indicators.
• De Jaegere, Samuel, and Nils Taxell. Forthcoming. “Measuring the Impact of Anti-Corruption Agencies: A Guidance Note for Anti-Corruption Agencies and National Statistical Offices.” International Anti-Corruption Academy, draft 30 May 2026.

On the use of indicators to measure abstract concepts, there is no better guide than the articles collected in –

Elkins, Zachary (2026). Working with Concepts: Foundational Essays by David Collier, with Research Notes on Innovation in the Field, Cambridge University Press.