Description and advocacy

One basic need of everyone interested in corruption is to describe how pervasive it is around the world. Non-governmental organizations trying to raise awareness of corruption need data with which to justify and illustrate their arguments to take corruption seriously. Assessing the prevalence and impact of corruption is vital to detect vulnerabilities in national systems and target responses in an effective and sustainable manner. Policy makers need to know how pervasive corruption is in their country in order to identify it as a priority for government action. It may also be helpful to distinguish among how different groups in society—citizens, small business owners, corporations, public officials, and/or country experts—perceive the pervasiveness of corruption differently; this may indicate who is most affected by corruption and how.

For this purpose, a measure should be simple to understand and interpret for everyone regardless of technical background; this helps the measure achieve its purpose of raising awareness. It should correspond to the overall pervasiveness of corruption in a state, or perhaps the pervasiveness of several broad types of corruption (such as petty bribery for services or large-scale diversion of government resources), to give the user a sense of the nature and severity of the problem. The measure should have wide coverage around the world, to allow for maximum relevance to the largest number of people. And it must be freely and easily available to anyone so as to maximize how often and how prominently it is used.

On this front, although certainly not beyond criticism, several existing measures of corruption have been extremely successful. Chief among these is Transparency International’s Corruption Perception Index (CPI). The strengths and weaknesses of the CPI will be discussed in greater depth later in the module; at present, it suffices to say that the CPI has the advantage of covering nearly every country in the world but does not disaggregate types of corruption or which sectors are more prone to corruption within a country, nor does it provide much policy actionable information about why or how a country is corrupt. Andersson and Heywood (2009, 747), an article that is generally critical of many uses of the CPI, is worth quoting at length:

[T]he CPI has rightly been seen as an immensely important step in focusing attention on the issue of corruption, offering for the first time a systematic basis on which to compare perceptions of corruption across a range of different countries, year by year. Without the CPI, it is doubtful whether many secondary studies which seek to identify the causes of corruption would have been undertaken, since the index offers an ideal large-n basis for analysis.We should therefore not underplay its significance in the fight against corruption: its value goes beyond the stimulation of research activity, since the publication of the CPI each autumn has generated widespread media interest across the world and contributed to galvanising international anti-corruption initiatives, such as those sponsored by the World Bank and the OECD.

Comparative institutional research

Most political scientists and economists believe that corruption is (at least in part) a product of institutions (North 1990). In this context, “institutions” doesn’t just mean the formal rules and structures of governance, such as laws dictating the distribution of political power and property rights. The less formal norms and beliefs that inform how people respond to more formal rules and laws are considered institutions as well. According to an institutional view, the overall degree of corruption in a state is not determined by individual-level resistance to (or acquiescence toward) corrupt behavior. It is instead determined by the information and incentives created for those individuals by the system as a whole, and the strategic interaction among many individuals influencing each other, that produces high or low corruption (Persson, Rothstein, and Teorell 2013). For this reason, academics in this tradition are interested in research that compares units with different institutions to one another. These comparisons allow scholars to determine which institutions allow corruption to flourish.

A measure that is useful for description and advocacy may have some features that make it useful for comparative institutional research. For example, because much of the variation in institutions is at the national level, a national measure of corruption is often appropriate (Mungiu-Pippidi and Fazekas 2020). Similarly, a measure that aggregates all forms of corruption into a single measure can be suitable because we often expect national institutions have powerful and widely-distributed impacts on corruption of all kinds. But some of a measure’s qualities are probably more important for research than they are for advocacy. For example, institutional variation exists not just across space but over time. If we want to know whether institutional change is effective at reducing corruption, we need measures that are available for a long span of years and comparable over time as well as across countries. The long time span is particularly necessary because institutional change is slow and infrequent, its effects sometimes equally slow and accruing long after the instigating event. Awareness of corruption as an important issue might benefit from a half-century of reliable data, but institutional research benefits much more.

The questionable validity of trends in some corruption measures is a focus of criticism of those measures. For example, a critique from Heywood and Rose (2014, 513–14) finds that the most recent values of two important measures of corruption (from years 2011 and 2012, contemporaneous with the paper) are extremely well predicted by eleven-year lags of the same variables; only ten to fifteen percent of the variation in the contemporary values is not predicted by the decade-long lag. They conclude that “while it is the case that scores in these indices vary from year to year… these changes cannot sensibly be viewed as part of a systematic change” (p. 515): the remaining variation is likely to be noise. Standaert’s (2015) findings are broadly consistent with this conclusion: though many popular measures of corruption are very closely correlated with each other, their similarity derives from the fact that they classify the same countries as more or less corrupt. By contrast, their changes over time are not as closely related. However, there is evidence that these changes over time are not merely noise. About 40% of the variance in within-country changes in five influential measures of corruption are explained by a common factor; this contrasts with 90% of the total variance in the measures including both between-country and within-country variation (Dalton and Esarey 2024, 26).

Anti-corruption policy research

Government officials, civil society organizations, and scholars (especially those studying policy and public administration) are all interested in fighting corruption through policy interventions. They need to be able to track changes in corruption before and after policy implementation to assess that policy’s effectiveness. They also need to identify sectors, organizations, and/or transactions that are particularly susceptible to corruption to be able to target interventions accordingly. Governments also need measures to demonstrate that their anti-corruption efforts are in earnest and not just for public relations purposes.

In some cases, the policies of interest are very large-scale institutional reforms. Such reforms include the creation of a Cabinet-level anti-corruption ministry, broad changes in the diversity and freedom of media, or significant shifts in the fundamental system of governance. When a change of that magnitude is made, the same measures that are suitable for comparative institutional research may suffice to track associated changes in corruption aggregating over every agency and level of government nationwide.

In general, however, those measures do not suffice. Many policy changes cannot be targeted at every form of corruption or at every part of the government, and we would not expect all effective reforms to be total systemic overhauls. For example, if a country makes the procurement and purchasing procedures for local governments more transparent, we would expect there to be an effect on procurement-related corruption by local officials, not necessarily a massive change in corruption of every kind and at every level of government. An aggregated national-level measure would likely understate the effects of such a policy, or maybe miss them entirely. Even targeted measures of corruption would not necessarily capture the extent to which the government is contemporaneously exerting significant effort on anti-corruption initiatives. Even major policy changes have effects that are only realized years later because of the strong hysteresis of bureaucratic behavior. A policy’s effects on corruption may also not be immediately reflected in some measures because of a temporal disconnect between latent (unobservable) corruption and its visible consequences.

The mismatch between many influential measures of corruption (which are aggregated national-level corruption metrics) and the need for more targeted and policy-relevant measures is a frequent topic in the corruption literature. For example, the former Minister of Governance and Anti-Corruption in Tunisia Kamel Ayadi and his co-author Todd Foglesong (2023, 5) argue:

Officials with an anti-corruption mandate say they need to know whether they are doing well, so that they do not believe that they are succeeding when they are failing. But global measures of corruption are not fit for these purposes: they do not help public officials diagnose, design, and implement sustainable policies of anti-corruption and then monitor and evaluate their effects; they do not help them appreciate changes in the scale and character of corruption, prioritise and coordinate the interventions of auditing or investigative bodies, or devise remedies for the side effects of anti-corruption campaigns; they do not help government leaders and corporate CEOs understand or communicate with the public about the accomplishments or setbacks in efforts to curb, contain, or combat corruption. And yet measures that serve these purposes are sorely needed to avoid the inclination of public institutions to claim success and declare progress where there might be regress and failure.

Heywood (2017) also makes this point, in greater depth and from a scholarly perspective. Like Ayadi and Foglesong (2023), his starting point is a concern that despite twenty-five years of intensive research “the jury remains out on whether such an increase in productivity has had any meaningful effect in terms of actually reducing corruption” (p. 22). He identifies three problems with the contemporary measurement of corruption:

the way in which corruption has been conceptualized in much mainstream academic research, resulting in “magic bullet” solutions based on institutional reconfiguration (hocus-pocus); the tendency of much research and anti-corruption advocacy to concentrate on nation-states as the primary unit of analysis (locus); and the lack of sufficient disaggregation of different types and modalities of corruption beyond crude binary divisions that do not recognize the complexitites of an increasingly transnational world (focus).

Heywood (2017) explicitly highlights the institutional focus of existing measures and their inappropriateness for anti-corruption research, as we noted above. Mungiu-Pippidi and Dadašov (2016) also echoes this criticism, noting the “non-actionable nature of the corruption indicators” (p. 417). They attribute this shortcoming to two other shortcomings: a “lack of validity of underlying theoretical concepts and of a unitary theory of corruption and governance” and the “lagging nature of governance indicators” (p. 416):

…changes in the assessments of governance might reflect corrections of errors done in the past. A notable example in this context is the worsening of the [Worldwide Governance Indicators] Control of Corruption scores in Greece, Spain, Italy, and Portugal in the aftermath of the recent financial crisis. This development might have resulted from different individual sources reassessing the governance context in these countries.

After studying the reliability and validity of many international corruption indices (ICIs), Ko and Samajdar (2010, 531) conclude that they are useful for studying corruption generally and cross-nationally but not as useful for designing and monitoring the progress of anti-corruption policies:

…we need to redirect our effort towards developing more measures of specific types of corruption instead of being content with measures of the general perception of corruption. It is especially important to devise problem specific measures, i.e., measures of variants of corruption that are especially prevalent in a particular country or group of countries. This is critical if we want to utilize corruption indexes to design and evaluate various anti-coruption programs that are meant for controlling specific types of corruption in particular policy areas and regions.

Some of the corruption measures developed since Ko and Samajdar (2010), such as the procurement corruption measures developed by Fazekas and Kocsis (2020), address these critiques. But these approaches come with their own disadvantages. We will discuss those trade-offs in detail later in the module.

Behavioral and sociological corruption research

As Simpser (2020, 1373) points out, “the notion that ideas, attitudes, beliefs, and norms—often referred to as ‘culture’—are also important causes of corruption has… received considerably less attention” compared to institutional explanations. Simpser (2020) attributes this disparity to scholars’ strong reliance on rational choice theory that emphasizes the importance of costs and benefits in determining behavior. We might also attribute it to a common conception that corruption is a collective action problem, where “the rewards of corruption… should be expected to depend critically on how many other individuals in the same society that are expected to be corrupt” (Persson, Rothstein, and Teorell 2013, 450). In that conception, the institutions that coordinate individual behavior are often more important than the characteristics of the individual themselves.

Nevertheless, a substantial body of research studies the relationship between corruptibility and individual characteristics. These characteristics can be psychological, sociological, and/or cultural. For example, Simpser (2020) studies whether the cultural attitudes toward bribery that individuals inherit from their family persist in new environments. Studies like this, where the individual is the unit of analysis, obviously require individual-level measures. Measures designed for comparative institutional or anti-corruption policy research are typically measured at a group level (such as country or agency).

Sometimes, individual-level studies can use some of the components of group-level measures. There are many surveys that ask individuals about their experience with or attitude toward corruption, particularly bribery. For national-level measures, their responses are averaged by country. However, the individual responses can be studied as well. This is how Simpser (2020) finds that second-generation immigrants from countries with higher tolerance for bribery are more likely to tolerate bribery themselves, despite living in a new environment.

Much of the behavioral research about corruption is experimental. For example, Dorrough et al. (2023) have over 5,000 subjects play a bribery game with one another online to determine whether a subject’s country of origin influences others’ willingness to offer them bribes. They find that subjects from countries with a reputation for high corruption are more likely to be offered bribes, but less likely to accept them. For a study like this, the key dependent variable is whether a participant offers or accepts a bribe. While this behavior usually cannot be directly observed, within an experiment it can be. In experiments like this, a bespoke corruption measure must usually be created to satisfy the requirements of the specific design.

What is corruption? What counts as corruption?

As noted in the introduction, there have been many attempts to measure corruption. Of course, it is difficult to agree on how to measure something if we cannot agree on what it is. This is one important reason why corruption has proved difficult to measure. The situation is hardly unique to corruption; for example, the United States Supreme Court once struggled with a similar problem in trying to create a rule that distinguished illegal obscenity from constitutionally protected free speech and artistic expression. In a concurring opinion for Jacobellis v. Ohio (1964), Justice Potter Stewart wrote that in defining obscenity as “hardcore pornography” he could not be more specific:

I shall not today attempt further to define the kinds of material I understand to be embraced within that shorthand description, and perhaps I could never succeed in intelligibly doing so. But I know it when I see it…

Such an amorphous standard is impossible to apply consistently or understand systematically. Having failed to construct a set of structured rules that could identify obscenity, the Court eventually (about a decade later) defined it as a function of popular perception. Specifically, speech could not be obscenity unless “the average person, applying contemporary community standards would find the work, taken as a whole, appeals to the prurient interest” (Miller v. Ohio, 1973). As we will see, such perception-based measures are also widespread in the measurement of corruption, and for much the same reasons.

Perhaps the most widely accepted definition of corruption is one used by Transparency International (2023b): “the abuse of entrusted power for private gain”. The definition is venerable, appearing in a slightly different form as far back as 1935 in an Encyclopedia of the Social Sciences (Senturia 1935). Hawken and Munck (2008) note that it forms the basis for “substantial agreement on a working definition of corruption (p. 74). But they also point out that”this definition does not settle all questions,” perhaps an understatement of the debate concerning the definition’s suitability. As Philip (1997) points out, such a definition leaves undefined “which view of the character and scope of public office or public interest should be accepted” (p. 441). If a legislator accepts large campaign donations from a business in their constituency and then (without an explicit quid pro quo agreement) champions that business’s interests over those of other constituents, is this corrupt or merely politics as usual? If it is corrupt, then corruption in many Western-style democracies (and in particular in the United States) should probably be assessed as more widespread compared to how it typically is. We might appeal to the law as an arbiter by saying that only abuses of public office for private gain that are illegal should be considered corrupt (Fisman and Golden 2017, 26–29). Yet this approach also has problems:

To ask whether a politician acts corruptly we must be aware that the characterization of public office will inevitably point beyond the compliance with rules to the principles underlying those rules—principles which come into play to cover cases on which formal rules are silent. (Philip 1997, 445)

Thus, although such campaign contributions are (generally) legal in the United States, they might still be considered corrupt. This precisely why Philip (1997) argues that “definitional disputes about political corruption are linked directly to arguments about the nature of healthy or normal condition of politics” (p. 446). Any measurement of corruption will need to make choices among these definitions, and will embody these choices. These issues are discussed in greater depth by the UNODC module on defining corruption; there is also a thorough discussion of the history of definitions of corruption and their (lack of) overlap in Rose (2018).

Corruption is not directly observable

Corruption would be easier to measure if it could be seen. But corruption is not directly observable for a variety of reasons. First, it is typically illegal even in jurisdictions where it is common and subject to opprobrium even when legal. Ergo, corruption is most effective when hidden. This gives the perpetrators, and possibly even administrators who want to encourage foreign investment by corruption-averse businesses, a powerful incentive to hide corruption from the public.

Corruption is also difficult to measure because most of its victims are not directly involved. If we were trying to measure a different crime, like robbery or fraud, the perpetrators would have a strong incentive to cover it up. The victims, however, would not. Victims want justice and restitution, which typically involves reporting the crime to a government authority. For this reason, other crimes intrinsically create evidence of their own commission (if not of the perpetrator’s identity). But corruption often does not. When a government official accepts a bribe in exchange for special treatment, the public at large is harmed—but they are not party to the transaction. Unless someone has been personally asked for a bribe, they only know that the government is inefficient or that its outcomes are unequal. And even if they have been personally asked for a bribe, they do not necessarily know how pervasive such requests are.

Finally, behaviors and expectations about corruption are linked in ways that make measurement challenging. For example, corruption creates court cases related to the prosecution of corruption; these cases can be counted as a measure of the extent of underlying corruption. However, a small number of prosecutions can indicate different things. There may be little underlying corruption to prosecute. There may be a great deal of corruption, so much that corrupt officials are shielded from consequences. Or the government may simply not have the resources to effectively detect and punish corruption (Brooks et al. 2013, 28). Indeed, citizens who do not trust the government may simply refuse to report corruption because they do not expect the offender to be punished (Amini, Douarin, and Hinks 2022). More generally, an actor’s incentives for tolerating corruption depend on their expectations of what other actors are doing (Persson, Rothstein, and Teorell 2013), meaning that the systemic level of corruption and reactions to it are simultaneously determined.

For these reasons, it is intrinsically difficult to link corruption to something we can observe, either direct or indirectly. But any measure by definition needs to rely on something observable, ideally something that cannot be easily influenced by people and institutions that have an incentive to cover up corruption. The choice of what we will observe as evidence of corruption is one important way in which corruption measures differ from one another. We begin by considering quantitative approaches to this problem, and end by considering qualitative approaches.

Unit of analysis

Measures need to choose the boundaries within which they are going to record the level of corruption. Some of those boundaries are geographic. For example, corruption can be measured at the subnational level, in individual cities, counties, prefectures, and so on. It can also be measured at the national level, averaging over an entire country. Supranational or regional measurements might be needed if the locus of corruption is an international organization.

The boundaries might also be organizational. Corruption can be measured in specific industries, sectors, or government agencies, separating corruption in (for example) defense procurement from corruption in the provision of healthcare or (as another example) corruption in the legislature from corruption in the judicial system. Alternatively, a measure can try to capture the pervasiveness of corruption in the polity or economy as a whole.

Other boundaries are temporal. Corruption can be measured cross-sectionally, taking a snapshot of corruption at a particular point in time. It can also be measured longitudinally, typically (but not necessarily) annually. We can measure corruption contemporaneously, producing estimates in real time, or we can measure it retrospectively, creating ratings for the past based on information available in the present.

Type of corruption

Any measure must decide what kinds of corruption will be included or excluded in the measurement. One common distinction in the literature is made between petty corruption, or corruption perpetrated by low-level officials in exchange for providing basic government services, and grand corruption by high-level officials generally involving positions of great influence, highly consequential decisions and large sums of money (Fisman and Golden 2017, 37). A measure might try to target one or both of these forms of corruption. Examples of petty corruption include bribes paid to civil servants to obtain permits or licenses or paid to police officers to avoid enforcement of traffic laws. Grand corruption includes misuse of government resources for personal gain and accepting bribes in exchange for favorable legislation or regulatory treatment. Petty corruption tends to be more visible to the public than grand corruption because it victimizes them directly. But grand corruption can be even more harmful to a country than petty corruption because it distorts policy and depletes funding needed to provide public goods.

In some cases, multiple dimensions of difference have been combined into a more complex typology. As an example, Ang (2020) distinguishes among four different types of corruption (p. 10):

• Petty theft refers to acts of stealing, misuse of public funds, or extortion among street-level bureaucrats
• Grand theft refers to embezzlement or misappropriation of large sums of public monies by political elites who control state finances
• Speed money means petty bribes that businesses or citizens pay to bureaucrats to get around hurdles or speed things up
• Access money encompasses high-stakes rewards extended by business actors to powerful officials, not just for speed, but to access exclusive, valuable privileges.

Recognizing these distinctions can also help us discover meaningful differences among countries that more aggregated measures of corruption would miss. Ang (2020) uses expert surveys to measure the four different types of corruption for 15 different countries, finding that some countries’ level of corruption may be over- or under-stated by highly aggregate measures such as single-question overall corruption perception surveys (pp. 35-36). It also uncovers qualitative differences in how corruption manifests in different countries, even if those countries have similar aggregate corruption (pp. 37-38). Ang argues that different forms of corruption have different effects on economic development; she would not be able to detect that difference if she was not able to separately measure different kinds of corruption.

Perception, experience, and other referents

Because corruption is not directly observable, we have to measure some visible consequence of corruption rather than corruption itself. This choice creates another dimension of difference among corruption measures.

One of the major proxies for corruption is perception of corruption. Perception-based measures ask some group of people to report how widespread corruption is in the targeted place and time. The precise form of the perception-based measure can vary widely. First, different measures ask different groups for their perceptions. These groups might represent the resident population, firms doing business in the targeted area, academics or other knowledgeable experts (who may or may not be resident in the targeted area), or other subgroups that are thought to be particularly exposed the form of corruption being studied. Second, different measures may ask about specific types of corruption; respondents may be asked about their perception of corruption in a particular institution (like the police or the national legislature) or about a particular type of corruption (such as vote-buying or bribery). They might also be asked about some indicator of corruption’s importance other than its frequency, like the scale of monetary losses created by corruption. As one example of a perception-based measure, Wave 7 of the World Values Survey (Haerpfer et al. 2022, Question 118) asks:

We want to know about your experience with local officials and service providers, like police officers, lawyers, doctors, teachers and civil servants in your community. How often do you think ordinary people like yourself or people from your neighbourhood have to pay a bribe, give a gift or do a favor to these people in order to get the services you need? Does it happen never, rarely, frequently or always?

Perception-based measures of corruption have widely recognized weaknesses. A commonly recognized weakness is that perceptions and reality are not always the same (Heywood and Rose 2014, 510–11). Studies have revealed that a country’s perceived level of corruption is systematically different depending on whose perception is being solicited (Hawken and Munck 2011, 4–9). Even worse, perceptions of corruption have repeatedly been shown to be subject to various forms of forms of bias (Bello y Villarino 2021; Corrado et al. 2023; Jahedi and Méndez 2014; Ko and Samajdar 2010). These biases may be particularly influential when tracking change in corruption over time (Andersson and Heywood 2009, 755–56; Heywood and Rose 2014, 512–16), which may explain why such measures are typically much more strongly correlated with each other when explaining variance between countries as compared to when explaining change over time (Standaert 2015, 788).

Despite these limitations, studies have also vindicated the usefulness of perception-based measures for comparative institutional research about corruption. A study by Charron (2016) finds “strong counter-evidence is found to the prevailing pessimistic claims in the literature—the consistency between actual reported corruption, as well as citizen and expert perceptions of corruption, is remarkably high and such perceptions are swayed little by ‘outside noise’ (p. 147). In some cases, the same studies that found evidence of bias in perception-based measures nevertheless strongly support their scientific value. For example, Jahedi and Méndez (2014, 109) find that:

…subjective measures can perform better because of their ability to measure unobservables. In your experiment, general subjective measures of crime effectively captured both explicit and implicit events, where objective measures were only able to capture the latter. A major implication of our study is that subjective measures can be valuable, independent of their correlation to objective measures. Many other studies have tried to validate subjective measures by showing that they match up well with obejctive measures. …Somewhat ironically, we find that subjective measures are in fact most useful in those situations where they disagree with their objective counterparts.

Another often-used observable referent for corruption is direct experience with corrupt activities. Experience-based measures ask a group of people to report whether or how often they have been personally exposed to corruption. For example, the International Crime Victims Survey (Van Kesteren 2018, Question B12) asks:

In the last five years, has anyone such as a police officer, other government official (for example an inspector or a customs officer), a doctor, or teacher asked you, or expected you to pay a bribe [or backhander] for his or her services?

As before, the respondents to this question can be a sample from the general population (as in the ICVS) or some subset of the population whose experiences are directly relevant (e.g., business owners). Such experience-based surveys also have limitations; most importantly, they are likely to underestimate or ignore grand corruption because the vast majority of citizens are not directly involved (and those who are have strong reasons to conceal their behavior).

It is sometimes possible to use a behavioral referent that measures how people react to the presence of corruption. For example, the extent of corruption in a country can be estimated by examining how the stock prices of politically-connected companies in that country react to crises that endanger the current regime (Fisman 2001). Or we might be able to estimate corruption by studying the behavior of diplomats; officials accustomed to ignoring regulations (such as parking prohibitions) might continue to do so while serving abroad (Fisman and Miguel 2007). Chen and Kung (2019) use the differential in the price of public land sales to politically connected buyers compared with ordinary buyers to measure the extent of the corruption premium afforded to powerful people; this also allows them to track how this premium falls over time in response to anti-corruption initiatives. However, such measures are rarely available for many units over many time periods, making longitudinal comparative research challenging.

We might also try to measure concepts closely associated with corruption, such as forms of corruption that present a risk of government instability. For example, the Political Risk Service’s measure of corruption (Political Risk Services Group 2018) uses experts to assess the degree to which corruption creates political risks for their clients, “institutional investors, banks, multinational corporations, importers, exporters, foreign exchange traders, shipping concerns, and a multitude of others” (p. 1). Therefore, their measure may lean toward assessing corruption in light of the stability concerns that it creates or the harm it inflicts on international businesses rather than its impact on the public as a whole (pp. 4-5). The obvious shortcoming of this approach is that it is does not directly measure corruption, but political risk to investors that is created by corruption. It may also weight some forms of corruption (e.g., the sort of corruption that endangers investments) more strongly than others.

Finally, June et al. (2015) makes a distinction between input and output proxies for measuring corruption. Input measures of corruption are “measures of the existence and quality of institutions, rules, and procedures” while output measures are “measures of what those mechanisms lead to in practice” (p. 14). One example of an input measure is the existence of an independent anti-corruption agency in the government; such an agency is (presumably) in a position to detect and punish corruption and therefore may reduce it, but its existence does not directly tell us how much corruption is present in the country. An example output measure is a survey question asking businesses whether they have paid or been solicited for a bribe by a government official in the last 12 months; this is a direct measure of how prevalent bribery is (at least in the business community). The distinction between inputs and outputs to corruption is also recognized in the UNODC’s Statistical Framework to Measure Corruption (UNODC 2023, 3–4) where “types of corruption” include criminal offenses under the United Nations Convention Against Corruption treaty (UNCAC), such as bribery and embezzlement, but also include preventative measures (like a merit-based civil service system and and judicial independence) and the strength of reporting and punishment mechanisms (such as the resources allocated to anti-corruption activities and protection for whistleblowers).

Input measures of corruption are strongly theory-laden, leading to potential difficulties in testing those same theories using such measures. An input measure for corruption is only accurate if we understand the theoretical connection between those inputs and corruption well enough to have confidence that the former represents the latter. Continuing a previous example, the existence of an independent corruption agency is only an indicator for lower corruption if we know that such agencies are effective. Output measures are also often indirect proxies for underlying corruption and assume a theoretical connection between themselves and the target concept of corruption: even if we directly measure a plausible output of corruption (like prosecutions for corruption), we must assume that convictions rise and fall in concert with the degree to which corruption is widespread in society. As noted above, this is not always the case.

Scale of the measure

A quantitative measure of corruption necessarily creates a scale, and different measures can have different scales. The scale of a measure¹ can be categorical, a qualitative grouping of units that share similar characteristics without being ranked; for example, a measure might classify corruption in a country as mostly petty corruption, mostly grand corruption, both, or neither. A measure of corruption can be ordinal, a ranking of units by the scope of corruption they experience without specifying how much more or less corrupt one rank is than another. A question asking survey respondents to rate their home country on a five point scale, where 1 means “corruption is absent” and 5 means “corruption is pervasive,” would be an example of an ordinal measure. It can be interval, a quantitative ranking of units where the measurement tries to capture the relative magnitude of corruption among those units. For example, we might ask survey respondents to report how important various problems (including corruption) are to them on a 100 point feeling thermometer scale, where 0 indicates the least importance and 100 indicates an urgent priority. Or a measure can be ratio, the most precise type of quantitative ranking where the zero point of the scale has a meaningful interpretation. The number of convictions for corruption-related offenses in a country in the last year is a ratio-level measure of corruption.

Data source

The underlying data for corruption measures can be generated in different ways. One major distinction is whether the measure comes from primary or secondary data. A measure generated using primary data involves the collection of new information by the researcher involved in the measure’s creation, while a measure generated with secondary data involves the use (and typically the extensive statistical processing) of data already collected from other sources.

Among primary measures, there are a few common sources for data. One such source is a survey instrument. We could survey a large, representative group of people from the target to answer questions about their experience with or beliefs about corruption, the strength of institutions, or any other indicator related to corruption. We could also limit a survey to firms or other institutions, or the leaders thereof, who we think are particularly knowledgeable about or sensitive to corruption due to their activities. Finally, we could impanel a group of experts on the unit of analysis (such as academics who specialize in the study of a particular country) to report on the level of corruption in the target based on their expertise.

Another common source of information for primary measures is a government or other administrative record of corruption-related activity. For example, when studying regional differences in corruption within the People’s Republic of China, Dong and Torgler (2013) use “the number of registered cases of corruption per 100,000 people each year” as their dependent variable, which they measured “by collecting the number of annual registered cases of corruption in the procurator’s office by region (listed in the China Procuratorial Yearbooks)” (pp. 154-155). Data from this source is sometimes touted as more “objective” compared to that from other sources (Ayadi and Foglesong 2023, 11–12). However, regardless of whether this measure is less subjective than others, it is still an indirect indicator of the true level of corruption and not corruption itself. Indeed, as noted above, we have reason to suspect that cases and/or convictions for corruption could move in a direction opposite of the trend of the actual level of corruption. Enforcement of anti-corruption laws can be selective or negligent, particularly in cases where corruption is widespread; on the other hand, a surge in cases or convictions might indicate an institutional crackdown (and thus a reduction) in corruption.

Many of influential measures of corruption are indexes created by combining information from many sources of secondary data. One of the earliest and most influential of these indexes, the Control of Corruption Estimate from the World Bank’s World Governance Indicators, compiles information from more than 30 different underlying sources (World Bank Group 2023b). These sources include surveys of firms and populations from the target country, “views of country analysts at the major multilateral development agencies,” ratings created by nongovernmental organizations, and “commercial business information providers, such as the Economist Intelligence Unit, Global Insight, and Political Risk Services” (Kaufmann, Kraay, and Mastruzzi 2010, 6–7).

Measurement model

Finally, raw data must be somehow converted into a processed measurement of corruption by the researcher. This can be done in a variety of ways. For many measures, a simple average or other summary statistic can be employed. As an example, a survey of popular experiences with corruption can generate a measure of the proportion of respondents who report being asked for or paying a bribe in the last twelve months; this proportion then becomes a measure of how widespread corruption is in the unit of analysis. For other measures, a count or weighted count is more appropriate. For example, measures involving the number of cases or convictions for corruption are counts, possibly reweighted using a denominator (such as population) to allow fair comparison across units.

Other corruption measures have considerably more complex measurement models, particularly those measures that involve aggregation of raw data from many independent sources. These models often take a number of inputs, reweight those inputs according to a variety of criteria, attempt to correct for bias and unreliability in some sources, and/or attempt to extract a common corruption dimension from a number of inputs that might be influenced by both corruption and non-corruption factors. The Varieties of Democracy project (Marquardt 2024), for example, takes opinions from over 3,700 individual country experts and then:

converts these manifest items (expert ratings) to a single continuous latent scale and thereby estimates values of the concept. In the process, the model algorithmically estimates both the degree to which an expert is reliable relative to other experts, as well as the degree to which their perception of the response scale differs from other experts. Similarly, we use patterns of overlapping coding—both in the form of experts who code multiple countries and experts who code hypothetical cases (anchoring vignettes)—to estimate the degree to which differences in scale perception are systematic across experts who code different sets of cases.

Transparency International’s Corruption Perception Index

Transparency International, a non-governmental organization aiming to bring attention to and develop solutions for the problem of corruption around the world (TI 2024b), first developed their Corruption Perception Index (or CPI) in 1995 (TI 2024a). It “draws upon 13 data sources which capture the assessment of experts and business executives on a number of corrupt behaviors in the public sector” as well as “mechanisms available to prevent corruption in a country” (such as the country’s ability and willingness to enforce laws against corruption) (Transparency International 2023a), and is thus an expert perception-based measure that also incorporates perceptions about institutional effectiveness alongside more direct perceptions about corruption. It standardizes the scales of these inputs, uses imputation to fill in missing values, and then averages the standardized and imputed components to generate an overall interval-level corruption score that varies between 0 and 100 (with lower values indicating greater corruption). These scores are calculated contemporaneously at the country-year level and are designed to be comparable between countries and (since 2012) over time. The map below shows TI CPI scores for 2022, rescaled so that higher values indicate more corruption. Current and previous values for the CPI are available at https://www.transparency.org/en/cpi/2023.

As discussed above, the TI CPI is a measure designed to capture the overall pervasiveness of country’s public sector and to identify which countries have the highest and lowest aggregate levels of corruption in a given year. It is therefore a very useful measure for description and advocacy, highlighting the places where corruption needs the greatest attention. It can also be useful for comparative institutional research, since it facilitates the comparison of countries to each other and (since 2012) to themselves over time. However, the TI CPI is considerably less useful for measuring the efficacy of most anti-corruption reform. It changes very slowly over time (Heywood and Rose 2014) and aggregates over all aspects of a country’s government, making it less-sensitive to sector-specific reforms and slow to respond to changing realities.

The World Bank’s Control of Corruption Estimate

The World Bank, a large and influential international organization primarily concerned with making loans to low- and middle-income countries to promote development (World Bank Group 2024), also conducts research and provides consulting services to client countries. One of their research products is the Worldwide Governance Indicators (WGI), a suite of six variables designed to measure key concepts related to economic development; these indicators were first developed by Daniel Kaufmann and Aart Kraay and are measured at the country-year level between 1996 and 2022 (World Bank Group 2023c). One of the six indicators is Control of Corruption, which “captures perceptions of the extent to which public power is exercised for private gain, including both petty and grand forms of corruption, as well as ‘capture’ of the state by elites and private interests” (World Bank Group 2023a). More than fifty variables from twenty-four sources serve as the inputs to their measurement model and include both popular, business executive, and expert perception of corruption, both in specific agencies or sectors and overall, some questions about personal experience with corruption, some questions related to institutional effectiveness in fighting corruption, and some other aggregate indexes measuring corruption (including the V-Dem Political Corruption Index, the Economist Intelligence Unit Riskwire and Democracy Indices, and the World Justice Project Rule of Law Index). The map below shows WBGI CCE scores for 2022, rescaled to range between 0 and 100 so that higher values indicate more corruption. The CCE is available as a part of the WGI data set at https://databank.worldbank.org/source/worldwide-governance-indicators.

The WBGI CCE shares most of the same strengths and weaknesses as the TI CPI. It captures a somewhat broader view of corruption than the CPI because it includes many more sources, including popular perceptions alongside expert perceptions. It also has a more formalized measurement model designed to identify differences in corruption across countries and over time, and to separate these differences from related but conceptually distinct characteristics such as political stability or freedom of expression. Their measurement model is also designed to be robust to missing data, allowing it to incorporate information from sources that are only available for some countries. However, it is still a highly aggregated measure and not designed to pick out subnational or sector-specific changes in corruption. It also assumes that the overall worldwide level of corruption is not changing over time; this is a flaw that is addressed by the Bayesian Corruption Index.

The Bayesian Corruption Index

Standaert (2015) developed the Bayesian Corruption Index (BCI) as an extension of the World Bank CCE’s unobserved components model. It is based on “69 variables coming from 18 different sources and cover[ing] 211 countries and regions” (p. 783); these sources are surveys of perceived corruption from households, firms, and experts from NGOs, think tanks, commercial rating agencies, and governments (p. 784) and are shared with the CCE. While the CCE’s estimation procedure fixes the overall level of global corruption as constant over time, the BCI’s Bayesian state-space model permits simultaneous measurement of both individual country-level trends as well as changes in the aggregate level of corruption around the world. This measurement model also allows imputation of missing values in its source components as an intrinsic part of the process, meaning that uncertainty about relationships among the input measures is incorporated into the imputation and uncertainty about the accuracy of the imputation is reflected in the final corruption index. The map below shows BCI scores for 2021, rescaled to range between 0 and 100 so that higher values indicate more corruption. The data for the BCI are available from https://users.ugent.be/~sastanda/BCI/BCI.html.

As noted above, the BCI is designed to improve upon two aspects of the WBGI CCE: it allows changes in the overall pervasiveness corruption worldwide, and it uses a more sophisticated approach to missing data that incorporates uncertainty. However, other than these changes, its advantages and disadvantages are similar to those of the CCE.

The Varieties of Democracy Project’s Political Corruption Index

The Varieties of Democracy (V-Dem) project has the primary aim of “conceptualizing and measuring democracy” (Varieties of Democracy 2024). Toward this end, it takes a much more complex and granular approach than other attempts to measure democracy, separately estimating the freedom and fairness of elections, the protection of civil liberties, freedom of the press, and other components of governance. One of the aspects of governance assessed by V-Dem (and not directly incorporated into any of their democracy index measures) is corruption. The highest-level V-Dem indicator of corruption is the Political Corruption Index (PCI), which averages measures of executive corruption, legislative corruption, judicial corruption, and public sector (bureaucratic) corruption (Coppedge et al. 2024, 305–6). Each of these measures is also separately available. It aims to measure “how pervasive is political corruption.” The component measures are designed to capture both petty and grand corruption in multiple forms.

Corruption is measured at the country-year level, with five experts per country-year observation typically consulted about the ratings (Marquardt 2024). These ratings are ordinal, but converted via an item-response model into a continuous interval measure of the latent concept of corruption. Country-years are assessed retrospectively for much of the temporal coverage of the V-Dem indicators, which range from 1789 until 2023 (Coppedge et al. 2024, 306). The map below shows V-Dem Political Corruption Index scores for 2022, rescaled to range between 0 and 100 so that higher values indicate more corruption. The PCI and other indicators from the V-Dem data set are available at https://v-dem.net/.

There are two great advantages of using the V-Dem data. First, the data provide a much greater degree of institutional disaggregation compared to the CPI, CCE, and BCI. Corruption measures for the executive, legislative, and judicial branches as well as the bureaucracy are all separately available. This makes it possible for researchers and policy-makers to have a more targeted, sector-specific view of corruption in a country. As an example, the judicial component of the index is shown in the map below. Although the pattern of judicial corruption generally mirrors the pattern of overall political corruption, there are some places where judicial corruption is performing better or worse than the system as a whole.

The other great advantage of the V-Dem data set is its long temporal coverage: data for most countries is available from the year 1789 up to the present day. This long coverage is possible because the measure is created using expert panels, and therefore can be retrospective. That is also the source of the V-Dem’s key limitation: it depends very strongly on the judgments of country experts, who may have biased or incomplete knowledge of the pervasiveness of corruption in the target country or institution.

TRACE Bribery Risk Matrix

The TRACE Bribery risk matrix is a measure of “the risk of encountering public sector bribery in 194 countries, territories, and autonomous regions” (International 2024a, 1). It has been collected annually since 2016 (and in 2014 before that). Rather than trying to measure all forms of corruption, it focuses specifically on business bribery risk (p. 3). The aggregate risk index is a weighted combination of four domain indexes (quoting the names from pp. 1-2):

1. business interactions with government;

2. anti-bribery deterrence and enforcement;

3. government and civil service transparency; and

4. capacity for civil society oversight.

Each domain index is made up of the average of subdomain scores; each subdomain is the average of normalized variables from publicly available data sets. These data sets include some of the same data sets referenced above, including the Varieties of Democracy data set, the World Justice Project Rule of Law Index, business surveys conducted by the World Bank, and the Bertelsmann Transformation Index. Subdomain, domain, and final scores all range from 1 and 100 (International 2024b, 3–7); the aggregate bribery risk index values for 2022 are shown in the map below.

2022 subscores for anti-bribery deterrence and enforcement are shown in the map below. As with the V-Dem judicial subscore, the countries that are rated as high bribery risks also tend to be rated poorly on deterrence and enforcement.

Index of Public Integrity

The Index of Public Integrity (IPI), developed by Mungiu-Pippidi and Dadašov (2016), takes an approach to measuring corruption that is considerably different than the other four country-level panel measures of corruption. Their approach attempts to balance a focus on “inputs” and “outputs” to corruption, reduce dependence on perception-based indicators, and uses input-oriented measures of corruption that are more directly under the control of governments trying to fight corruption (or not trying to). It has been collected once every two years since 2015. The IPI combines six different components into a single measure (pp. 423-425):

1. administrative burden, or the degree of regulation of the economy, a proxy for the scope of possible corruption enabled by government intervention into the economy;

2. trade openness, another proxy for the scope of corruption enabled by (possibly selective) enforcement of government intervention;

3. budget transparency, a measure of how easy it is to hide corrupt government expenditures;

4. judicial independence, a proxy for the ability of the legal system to fight corruption without interference from the legislature or executive;

5. e-Citizenship, a measure of social media use and other online tools to disseminate information about corruption and organize civic action; and

6. freedom of the press, a proxy for the ability of traditional media organizations to freely investigate and publicize potential instances of corruption in government.

Unfortunately, relative to the four measured we noted above, the IPI is available for fewer countries and time periods. The 2021 values of the measure are shown in the map below, with blank spaces for those countries where no data is available. Note that the IPI is coded so that higher values indicate more public integrity (that is, less corruption).

The 2021 values of the judicial independence IPI subscore are shown in the map below; as before higher values indicate more public integrity (in this case, greater judicial independence).

Gallup World Poll

Many of the composite measures of corruption profiled above use popular survey data measuring perceived corruption of government as one of their inputs. However, we can (of course) study these survey measures directly as a measure of (perception of) corruption as well. One such measure comes from the Gallup World Poll. Gallup “continually surveys residents in more than 150 countries… using randomly selected, nationally representative samples” of about 1000 people per country (Gallup 2017, 34). There are two questions on the Gallup World Poll relevant to corruption. Country-averaged answers to the first question, “is corruption widespread throughout the government in (country), or not?”, are shown in the map below (Gallup 2017, 34). The map shows the proportion of respondents from the 2016 survey answering “yes” to this question. There are some countries for which data are not available; composite indicators (like the World Governance Indicators) combine multiple survey measures with varying availability using multiple imputation to achieve worldwide coverage.

Country-averaged answers to the second question, “is corruption widespread within businesses located in (country), or not”, are shown in the map below (Gallup 2017, 34). As in the previous map, the proportion of “yes” responses in each country is depicted. There is a strong correspondence in answers to these two questions at the national level.

Where the data allow it, survey indicators of the perception of corruption can be localized to more specific regions within countries. This can be important where there are reasons to suspect heterogeneity in corruption within a state. This heterogeneity could exist, for example, where sub-national governments have substantial autonomy and may exhibit different levels of corruption. For example, the two maps below show averaged responses to the two Gallup World Poll questions within the states of the United States. Because a much smaller sample size is available for American states in the poll, the data must be averaged over a five year period (2013-2017) to achieve adequate samples for sparsely populated states. The poll results reveal substantial differences in perceived corruption among American states, with higher levels in the states of the deep South, the desert Southwest, and upper New England.

Risky Procurement Practices

A relatively new approach to measuring corruption involves examining public records for procurement contracting and recording the extent of practices that facilitate corruption. Rather than attempting to holistically measure all types and sources of corruption in a country, this approach targets corruption in the executive branch and pertaining to the supply of goods and services to the government (typically a form of grand corruption). Dahlström, Fazekas, and Lewis (2021) presents such a measure for the United States government between 2003 and 2015. Although US federal law imposes mandates procedures to ensure market competition when fulfilling government contracts, these procedures can be set aside in certain circumstances at the discretion of the procuring agency (Dahlström, Fazekas, and Lewis 2021, 655). Setting aside these procedures raises the risk that government officials may be favoring certain providers and/or overpaying for goods and services in exchange for bribes or other individual benefits. The measure examines data from “all regulated federal contracts in the United States” via www.usaspending.gov and reports the frequency with which these practices occur.

The map below presents the proportion of contracts over $150,000 USD in value made under a non-competitive procedure in 2015, organized by the location (state) of the office that originated the contract. This is an input measure of corruption, as it measures the prevalence of rules and procedures that allow corruption to take place; it therefore does not directly measure the existence of corruption. However, Dahlström, Fazekas, and Lewis (2021) find that politicized offices are more likely to use non-competitive procedures than more politically insulated agencies, and that politicized agencies also award more such contracts in battleground states during an election (p. 652). Their finding suggests the construct validity of this measure of corruption, as the measure predicts patterns we would expect to find if it was actually tracking corruption.

Similar measures also exist for the 27 countries of the European Union plus Norway, as described by Fazekas and Kocsis (2020). They describe six indicators or red flags that a contract might exhibit as a red flag for corruption (p. 159):

1. only a single bid was received for the contract;
2. the call for bids was not widely published;
3. “less open and transparent procedure types” were used;
4. a short “advertisement period (number of days between publishing a tender and the submission deadline)” was used for the contract;
5. vague evaluation criteria for contract completion were specified; and
6. there was a very short or long time for choosing among submitted bids for the contract.

Two of these indicators, the proportion of single-bid contracts and the proportion of contracts using restricted procedures, are shown in the maps below for the year 2022. Fazekas and Kocsis (2020) also validate their measures like Dahlström, Fazekas, and Lewis (2021), finding that the deviation between bid and final price for a contract is higher when red flags are present.