4.1.1 Empirical Analysis

In this chapter, we estimate the relationship between corruption and median incomes in US states using annual data between 1984 and 2021. Because median income is a lot more sensitive to the underlying distribution of wealth in a state, we use median household incomes as our dependent variable instead of per capita personal incomes.

The results of an Arellano–Bover/Blundell–Bond type system GMM (Generalized Method of Moments) estimation that assumes corruption is endogenously determined are presented in Table 4.1. In our regressions, following the growth literature, we also control for education measured as the share of population with a college degree or above, urbanization measured as the share of farm employment in total employment and the government size/scope measured by the Fraser Institute’s Economic Freedom Index (EFI). We use both CCI and CRI as our measures of corruption and find evidence of a negative effect of corruption on median incomes across states over time. On average, a 1 standard deviation increase in CCI causes median incomes in a state to decrease by approximately 1.75 percent. To put that in perspective, consider states such as Alabama and Mississippi, on the one hand, and Vermont and Wisconsin, on the other. If corruption in Alabama were as low as in Vermont, median incomes would have been $1,000 more annually. Since more than 15 percent of Alabama’s population live below the poverty line ($30,000 for a household of 4), a $1,000 difference is not trivial. In Mississippi, where the poverty rate is close to 20 percent, the situation is even worse. Reducing corruption in Mississippi to its level in Wisconsin would mean approximately $1,250 more for each household. Although the estimated standardized coefficient of CRI is smaller than that of CCI, it is still statistically and economically significant.

4.2.1 Empirical Analysis

To our knowledge, there are only two studies analyzing the relationship between corruption and economic inequality in US states. While Apergis, Dincer, and Payne (Reference Apergis, Dincer and Payne2010) use the Gini index of income inequality to measure economic inequality, based on data from the Current Population Survey of the Census Bureau, Dincer and Gunalp (Reference Dincer and Gunalp2011) use various Atkinson indices as well as the Gini index. Both studies find a positive relationship between CCI and income inequality across fifty states over the last two decades of the twentieth century. In this chapter, following Dincer and Gunalp (Reference Dincer and Gunalp2011), we use the Gini index as our income inequality measure, constructed from the American Community Survey covering annual data between 2010 and 2021, and both CCI and CRI as our corruption indices. Because income inequality changes very slowly, current inequality is likely to be affected by inequality in the previous period. Owing to this dynamic nature of inequality, we estimate the effects of corruption on economic inequality with an Arellano–Bover/Blundell–Bond type system GMM estimation, assuming that corruption is endogenously determined as we have done in our growth regressions earlier. In our regressions, we control for states’ median incomes too. The results are presented in Table 4.2. According to our estimations, corruption causes income inequality to increase. On average, a 1 standard deviation increase in CCI causes a state’s Gini index to increase by approximately 0.3 standard deviations, indicating that approximately 8.5 percent of the difference in Gini index between Alabama and Vermont and 12.5 percent of the difference between Mississippi and Wisconsin is explained by the differences in CCI. Although the estimated standardized coefficient of CRI is smaller than that of CCI, it is still statistically and economically significant. Taken all together, our evidence strongly suggests that more extensive corruption in a state leaves its citizens poorer and increases income inequalities among them. As we shall see in Chapters 4–8, those factors contribute to further disparities in public health and other public policy areas.

4.3.1 The Social Foundations of Politics

Democratic politics and institutions function best in a setting of limited, conditional trust and judicious skepticism with respect to both government and other people (for a classic account, see Lane, Reference Lane1962). Citizens who see their leaders break laws and enrich themselves at public expense – or who just believe such activities are commonplace – are likely to trust them less. Such trust can spill over into apathy or cynicism if people conclude that voting the scoundrels out, and political participation generally, is futile (de Sousa and Moriconi, Reference De Sousa and Moriconi2013) or that reform appeals serve mostly to distract attention from the crooks. Indeed, where corruption is perceived to be pervasive or where past reform efforts have failed, venal officials may enjoy an atmosphere of impunity.

Evidence from several countries, however, points to a variety of connections among corruption (both as perceived and as experienced), trust (in government and among citizens), and political participation – linkages that vary with contrasting contexts and can also reflect variations in research methods (Školnik, Reference Školnik2020a). Giommoni (Reference Giommoni2021) finds that exposure to news about corruption scandals tends to depress citizen participation in Italian municipal elections and discourages candidates from challenging incumbents, making it easier for the latter to retain their hold on office. Morris and Klesner (Reference Morris and Klesner2010: 1278), analyzing Americas Barometer data from Mexico, found “widespread perceptions of corruption, low levels of interpersonal and political trust, and some pessimism regarding the efforts of the government to address the problem.” While interpersonal trust was largely unrelated to more public aspects of the distrust-and-corruption connection in their findings, the data suggest that corruption contributes to a climate of opinion inhibiting reform mobilization. Školnik’s (Reference Školnik2020b) data from Colombia suggest that while perceptions of extensive corruption reduce citizen participation in many political activities, personally experiencing it produces an increase. In Senegal, by contrast, field experiments by Inman and Andrews (Reference Inman and Andrews2009) found that perceptions of corruption increased citizens’ likelihood of both voting and protesting. Olsson (Reference Olsson2014), using survey data from thirty-three countries, shows that corruption lessens citizens’ sense of political efficacy and thus reduces participation. Reciprocity – a sustaining factor in many accounts for strong civil societies and democratic accountability, and central to sustaining collective action (Ostrom, Reference Ostrom1998) – can in practice cut in two different directions depending upon institutional quality and levels of mutual trust:

Many of these connections make intuitive sense. But liberal democracy American-style may be particularly vulnerable on grounds of trust and participation – not just because of the scope of legal or “influence market” corruption the country seems to experience (see Chapters 1–3 of this book) but also with respect to citizens’ expectations regarding their place, and rights, in the system.

As Zakaria (Reference Zakaria1997: 25–26) observes, for a century and more liberal democracy (as against illiberal varieties offering elections but little else) has promised what he calls the tradition of “constitutional liberalism,” derived from the Greeks and Romans. It

Underlying that scenario is a grand social bargain that is both fundamental and surprisingly fragile. Liberal democracy, and in particular Americans’ idealized variant of it, does not promise results but is presumed to offer individual liberty and a voice in decisions affecting one’s life. Two aspects of that bargain that often receive relatively little discussion are important here. First, unlike social democracies, socialist, or nationalist systems – much less authoritarian regimes – outside of major national crises, American democracy asks relatively little of citizens. Loyalty to the constitution or political order is defined, on a day-to-day basis, in terms of compliance with laws and norms justified – ideally – as maintaining openness and fairness, not in terms of all-consuming loyalty to particular leaders, parties, or overarching national identity. Those loyalties are limited and conditional; citizens remain free to choose identities, religions, political parties, and lifestyles – in effect, free to be themselves within a broad and diverse social arena. Moreover, Americans are comparatively lightly taxed (OECD, 2022; Tax Policy Center, 2020) and – while not all might agree – lightly regulated, at least in their personal lives. In a system where individual liberty is so central, the importance of trust in that political order is clear. Second, and equally important, American democracy promises relatively little. The economy remains largely in private hands and is widely regarded as offering opportunities based on individual initiatives, but few if any ensured outcomes. Compared to many other affluent democracies, American governments provide only modest direct social benefits, and the public role in providing those that do exist – tax subsidies for home ownership, for example (Desmond, Reference Desmond2023), or Social Security as social insurance rather than as a personal savings program (Markowitz, Reference Markowitz2023) – is often de-emphasized or misunderstood. Inequalities in the economy are immense and continue to grow yet are tolerated in a normative climate where wealth is seen by many as an outcome of hard work and diligence in taking advantage of opportunities (but for a perspective on the growing tensions within that climate, see Hacker and Pierson, Reference Hacker and Pierson2020).

That American democracy rests in part upon longstanding myths yet falls short of its ideals in many ways is both true and, for our immediate purposes, beside the point. The ideas outlined earlier are widely accepted justifications for the system overall, not claims about day-to-day realities. They are also a shorthand version of values many Americans view as defining their role in the larger system – at the risk of overstating the point, as their democratic birthright.

4.3.2 Corruption and the Social Bargain

Corruption threatens to disrupt and discredit precisely those values. Recall Warren’s (Reference Warren2004) argument that in a democracy, the essence of corruption is the “duplicitous exclusion” of people from decisions that affect their lives and from the institutions and processes that are supposed to protect their liberties and individual dignity. The social bargain outlined earlier can be undermined in significant ways if people perceive that laws and institutions have been captured by others seeking, or already holding, unfair advantages, and that their opportunities to defend their interests have been taken away. If interpersonal trust is in a shaky condition, suspicions that others are seizing such advantages may become all the stronger, while expectations that others will cooperate in reform will likely suffer (“Why should I be the honest loser?”). Rothstein and Uslaner (Reference Rothstein and Uslaner2005: 53, and passim.) argue that inequalities – both material and of opportunity – undermine trust and weaken social capital and that “dishonest government undermines trust at least indirectly.”

These connections have found a definite place in the contemporary anti-corruption debate. Indeed, Transparency International (2022) builds its entire analysis on a definition of corruption as “the abuse of entrusted power for private gain.” That is a strong definition in terms of breadth, and because it highlights certain expectations between those with power and those without. It encounters serious problems, however, where power that has been inherited, bought, seized by violence, or otherwise not “entrusted” by anyone, as is often the case in seriously corrupt systems. Still, in the vast majority of instances we would call corrupt, someone has acted in untrustworthy ways, at least with respect to the broader public and its interests.

But corruption can also thrive when there is too much trust and vigilance is relaxed, when trust is uncritical, or where there is little or none. If I do not trust an official to follow the rules, I may try to buy some predictability, either from the official or via a seemingly helpful middleman (Khanna and Johnston, Reference Khanna and Johnston2007). Trust and the standards that flow from it can have strong situational aspects, as a Polish sociologist friend explained to one of us (Johnston) in connection with his youthful days working on a construction crew under the old communist regime. Theft of tools and materials was routine – almost expected – for if they belonged to the state, then they belonged to everyone, and if they belonged to everyone, then they belonged to no one. But when it emerged that one crew member had stolen from another’s personal stash, that was a very different thing, and direct measures were called for – probably after work, in a back alley.

Corruption can also underwrite a kind of in-group trust (You, Reference You and Uslaner2017: 476). Seeing another as predictably corrupt might be useful knowledge in some situations; alternatively, we might conclude that “He’s crooked but I can adjust our financial data and take that into account.” Corrupt group schemes (Gong, Reference Gong2002), or even just knowledge of them, can place people in shared jeopardy, creating interdependence or even a highly pragmatic kind of solidarity: consider the notion of “honor among thieves,” or the strong senses of obligation that can accumulate over time in a patronage network (Springborg, Reference Springborg1979). Political patronage or similar favors, when common enough to become a routine expectation, exemplify Axelrod’s (Reference Axelrod1990) well-known findings about repeated successful transactions as a source of trust. We might even imagine corrupt in-group trust eventually becoming a source of Putnam’s (Reference Putnam2000) “bonding” social capital.

There are broader systemic connections between corruption and trust, too. Democracy envisions citizens as formal equals but, as noted, often exists amid significant material and social inequalities, and vests control over resources in a few with the expectation that such authority will be accepted by the many. But if we believe adverse outcomes and events are the doing of corrupt officials or rigged procedures, democratic institutions and processes can become all but unworkable. As we may be seeing in contemporary American politics, such effects can be long-lasting, for trust is difficult to build and easy to destroy.

4.3.3 “Who Ya Gonna Believe? Me or Your Own Eyes?”Footnote ²

Globally, higher-quality governments tend to enjoy somewhat broader, if still limited and conditional, political trust (Uslaner, Reference Uslaner2017: 302; Rothstein, Reference Rothstein2011). Causality is complex because those governments are also more likely to have resources, institutions, and political support that help sustain good policy and implementation. The United States, thanks to a history of generally sound public institutions (but on their recent effectiveness, see Fukuyama, Reference Fukuyama2014), long enjoyed solid if not overwhelming public trust. But that is no longer the case: according to Pew Research, while 77 percent of poll respondents in October 1964 said they “trust the government in Washington to do what is right ‘just about always’ or ‘most of the time,’” by May 2022, after nearly six decades of mostly-declining trust, that figure had dropped to 20 percent (Pew Research, 2022). By comparison, when in 2020, Ipsos (2022) asked a comparable question about government trust in the thirty-seven countries that were Organisation for Economic Co-operation and Development (OECD) member states at that time, just short of 46 percent overall expressed trust in their national governments.

It would be a mistake to exaggerate levels of trust in the past, or to attribute the recent decades’ decline solely to corruption (for a partial dissent see Uslaner, Reference Uslaner2017: 303) or to any other single cause. Inequality, domestic tensions, and a generalized sense of unfairness in society can also weaken trust (Uslaner, Reference Uslaner2017: 302–303, 305), as can policy issues as diverse as border protection and government aid for various segments of the population. Specific events have interrupted the decline at times: trust in government enjoyed a marked, if brief, increase to 60 percent in the wake of the terrorist attacks of September 11, 2001 (Pew Research, 2022). Finally, trends in trust may well be influenced by the sharply contrasting worldviews on display in various news media. As You (475) notes, few of us have a clear idea of how much corruption actually occurs; We depend for that upon the news media, yet their reporting is also shaped by financial and, increasingly, partisan considerations.

4.3.4 Do We Trust Each Other?

Interpersonal trust is part of this picture too. Trust among neighbors and people we encounter in everyday life is an essential part of social capital, which in turn figures prominently in most discussions of collective action (Putnam, Reference Putnam2000; You, 479). As noted, if I do not trust my fellow citizens, I am more likely to look askance at any advantages they gain and at any anti-corruption commitments they proclaim, particularly if their schemes require me to forgo benefits of my own. Adding to the complexity, our earlier discussions of state political cultures suggest that we might well find substantial contrasts in mutual trust between, say, Moralistic and Individualistic states.

Measuring interpersonal trust, however, can be complicated. Since the late 1950s, many pollsters have at least occasionally asked respondents some variation on the following question: “Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?”Footnote ³ (Ortiz-Ospina and Roser, Reference Ortiz-Ospina and Roser2016). That question has often been criticized – some might say the two statements are simultaneously true – but it has been used in surveys for a long time in part because it has been used in surveys for a long time, creating an extended time series. If we take that sort of item as a benchmark, the United States emerges with middling or slightly better scores: An early 2022 survey of thirty countries found 33 percent of American respondents said most people can be trusted, compared to 11 percent in Brazil, 47 percent in Norway, and 56 percent in both India and China (IPSOS, 2022). A broader cross-section based on World Values Survey data showed that in 2014, 38.2 percent of US respondents said most people can be trusted, much the same as the result for 1998 (39.6 percent). At the top of the table in 2014 were The Netherlands at 66.2 percent, Sweden (63.8), and China (62.7),Footnote ⁴ and on the lower end were the Philippines (2.8 percent), Trinidad and Tobago (3.2), and Colombia (4.1) (Our World in Data, Reference Ortiz-Ospina and Roser2023). Many of the most apparently trusting countries – for example, the Netherlands and Sweden – score well on corruption and procedural fairness indices too (Uslaner, Reference Uslaner2017: 304; Grimes, Reference Grimes, Zmerli and van der Meer2017) but, as the example of China shows, that pattern has some striking exceptions.

Overall, the United States may not be experiencing a comprehensive crisis of trust: while political trust is in sad shape, interpersonal trust is at least moderately intact. But what do we find at the state level?

4.3.5 Empirical Analysis

Connections between trust and corruption are not limited to law-breaking by public officials. Particularly when we consider Warren’s argument regarding “duplicitous exclusion,” they are shaped by citizen expectations regarding their place and rights in society and by their perceptions of the political and social clout of money and elite privilege. A 2015 Rutgers-Eagleton Poll of 843 adult respondents in New Jersey,Footnote ⁵ for example, found 54.8 percent strongly agreed, and 24.3 percent somewhat agreed, with the statement: “When it comes to politics, people like me get overruled by the big campaign contributors.” Among those strongly agreeing, 42.1 percent strongly disagreed with the statement that “today’s economy offers most people a fair chance,” and another 23.2 percent somewhat disagreed. In that same survey, 56.7 percent strongly agreed, and 19.6 percent said “somewhat agree,” with the statement: “More and more there is one set of rules for the rich, and another for the rest of us.” A May 2015 The New York Times surveyFootnote ⁶ found that just 31 percent of adults agreed that “all Americans have an equal chance to influence the elections process,” while 66 percent (and 55 percent of Republicans) said “wealthy Americans have more of a chance.” In the same survey, 55 percent said “candidates who win public office promote policies that directly help the people and groups who donated money to their campaigns” “most of the time,” and 30 percent responded “sometimes”; only 13 percent said “rarely” or “never.” A 2019 poll found 82 percent saying “large corporations,” and the same percent saying “people who are wealthy,” have “too much power and influence in today’s economy” (Pew Research, 2020). Those are an important qualification upon the country’s favorable whole-country corruption index scores (see Chapter 1 of this book) and suggest that assessments of American corruption must include more than outright bribery. They also underline the importance of cases such as that of Clarence Thomas, the most senior Justice on the United States Supreme Court in terms of time served, whose rewarding relationships with a Texas billionaire have raised concern among many citizens (Robin, Reference Robin2023; Pilkington, Reference Pilkington2023).

To investigate relationships between corruption and trust in government in greater detail and to emphasize the diverse influences identified in Chapters 1–3, we focus now on the fifty states using data from the 2016 Cooperative Congressional Election Study (Milyo, Reference Milyo2019). Respondents were asked, “How often can you trust your state government to do what is right?” We assign a score of 1 for “hardly ever,” 2 for “some of the time,” 3 for “most of the time,” and 4 for “just about always.” According to that survey, trust in state governments is unimpressive at best, as 73 percent of the respondents chose “hardly ever” or “some of the time.” Only 3 percent responded “just about always.”

What factors help account for such low levels of trust? We estimate a multilevel regression (Table 4.3) with ordered probit controlling for respondents’ demographic characteristics and their ideological beliefs. The CCI is our corruption measure, and because there is an unknown time lag between corrupt actions and convictions, we use five-year moving averages. We start by estimating a parsimonious model of trust with CCI as the only independent variable and end with a comprehensive model including, inter alia, income inequality by state, an important variable affecting trust in government in previous studies (see, e.g., Uslaner, Reference Uslaner2017).

The estimated coefficient of CCI is negative and statistically significant in all estimations, indicating that people are less likely to trust their state governments where corruption is more common. The signs of the estimated coefficients of the control variables are mostly consistent with previous studies such as Wolak (Reference Wolak2020). As income inequality, measured as the income share of the top 5 percent in a state, increases, trust in government decreases, albeit at a decreasing rate. Older people and women are less likely to trust their state governments, taking the other independent variables into account. People who define themselves as liberal are more likely to trust their state governments if they also regard the governor of their state as liberal. Finally, trust in state government is likely to be higher among people who have a postgraduate degree.

The overall picture emerging from Table 4.3, as in our earlier discussions, is that trust in state governments reflects a range of influences, much as we would expect, but can be significantly disrupted by corruption. Given the fact that on any given day citizens are more likely to interact with state agencies and authorities (and with the local governments the states charter and regulate) than with their federal counterparts, these are important findings. We will say more about the implications of that finding later, and in Chapters 8 and 9.

5.2.1 Police and Society

Sarat et al. (Reference Sarat, Douglas and Umphrey2011) characterize the law as a means of punishment and of regulation. Therein lies a chronic hazard: laws would seem to have little value in either sense unless backed up by coercion or its credible threat, and yet that coercion itself must be regulated by law. Punishments are difficult to regulate, for reasons ranging from the ways police encounters often take place (outside of public view, in situations marked by confusion, vulnerability, and ambiguity) to the human failings and outright biases of those responsible for enforcement. All too often the measured and proportional application of coercion gives way to violence – or death. Similarly, maintaining necessary social order can easily give way to social control of more sinister sorts, including, but hardly limited to, the perpetuation of class, racial, and other inequalities, and the imposition of political, policy, and cultural preferences, again facilitated by coercion. What constitutes excessive force by police is a controversial issue. Killings, by tragic contrast, are unambiguous outcomes, and we have data on their occurrence. Those data, in turn, are indicators of a larger dilemma – one our analysis seeks to draw out by emphasizing a range of possible causes.

Using annual data from fifty states covering the period between 2013 and 2019, we find a positive relationship between corruption and police killings that remains robust when alternative influences are taken into account. We discuss, first, the channels through which corruption and other political variables – political culture, party competition, ideology, and the power of special interest groups – could affect police killings of Black Americans. We then describe our data, methodology, and results and examine important issues raised by the research.

5.2.2 The Thin Blue Line?

Civilians’ risk of being killed by police in America is considerably higher than in similar countries, but for Blacks, the risk is approximately three times higher than for Whites (Buehler, Reference Buehler2017; DeGue et al., Reference DeGue, Fowler and Calkins2016). The American Public Health Association (APHA) argues that even when no one has been killed, the extended consequences of police violence include mental and physiological harm disproportionately affecting Black individuals and communities and their ability to achieve positive health outcomes (APHA, 2018; see also Alang et al., Reference Alang, McAlpine, McCreedy and Hardeman2017; Sewell, Reference Sewell2017; Bor et al., Reference Bor, Venkataramani, Williams and Tsai2018; Das et al., Reference Das, Singh and Kulkarni2021).

Social scientists in several disciplines have investigated these issues from the standpoint of conflict theory, racial threat hypotheses, and a range of demographic and economic variables (see, e.g., Tolliver et al., Reference Tolliver, Hadden and Snowden2016; Edwards et al., Reference Edwards, Esposito and Lee2018; and Moore et al., Reference Moore, Robinson and Clayton2018). Racial diversity and segregation are consistently found to be important factors explaining variations across counties, cities, and states in police killings of Blacks (Nicholson-Crotty, Nicholson-Crotty, and Fernandez, Reference Nicholson-Crotty, Nicholson-Crotty and Fernandez2017; Johnson et al., Reference Johnson, Vil and Gilbert2019). Size of the Black population and the degree of segregation in a community are positively related to police killings of Blacks, pointing to racial bias, be it explicit or implicit (Siegel et al., Reference Siegel, Sherman and Li2019; Mesic et al., Reference Mesic, Franklin and Cansever et al.2018). Experimental evidence shows that being Black is strongly associated with others’ perceptions of threat (Eberhardt et al., Reference Eberhardt, Goff, Purdie and Davies2004) and affects police decisions to shoot (Correll et al., Reference Correll, Park and Judd2002, Reference Correll, Park and Judd2007). Class conflict is important as well (Jacobs and O’Brien, Reference Jacobs and O’Brien1998); according to Blalock (Reference Blalock1967), Turk (Reference Turk1966), and Quinney (Reference Quinney1970), poor people are more likely to be perceived as threatening. The more economically unequal the society, the greater the likelihood that dominant interests enforce their control through coercion (Chambliss and Seidman, Reference Chambliss and Seidman1982). Coercion is a key factor sustaining de facto systems of order in unequal societies (Jacobs and Britt, Reference Jacobs and Britt1979).

The role of police in maintaining such hierarchies is exemplified in many ways. During his 2015 presidential campaign, Donald Trump referred to the police as “the force between civilization and total chaos” (Chammah and Aspinwall, Reference Chammah and Aspinwall2020). Such “thin blue line” imagery is particularly powerful for some: black-and-white American flags with a blue line dividing an upper segment from the lower were displayed by white supremacists during the 2017 “Unite the Right” march in Charlottesville, Virginia, and are a common sight elsewhere. Defenders of the symbol describe it as only expressing respect for law enforcement, but others see it as symbolizing the police as a force suppressing a segment of society, or as openly racist (Chammah and Aspinwall, Reference Chammah and Aspinwall2020). Whatever one’s interpretation, there is no doubt that deeper questions of race, justice, accountability, and the proper role of coercion are embedded in controversies regarding law enforcement in America.

5.3.1 Corruption

Corruption, particularly of the structural sort, is not merely a matter of misconduct by individuals. Rather, it is a manifestation of broader “bad governance” in public institutions (Rose-Ackerman and Palifka, Reference Rose-Ackerman and Palifka2016). According to Rothstein and Teorell (Reference Rothstein and Teorell2008), “good governance” requires political equality on the input side to be complemented by impartiality on the output side of the political system – that is, in the exercise of authority. Rose-Ackerman and Palifka (Reference Rose-Ackerman and Palifka2016) argue that “good governance” requires accountability as well. Corruption impedes both accountability and impartiality in public institutions including police departments, and thus could contribute to police killings of Blacks.

When corruption takes hold, the uses of police powers – including coercion – can become more arbitrary, negotiable, and discriminatory in ways that reflect officers’ own interests and preferences, not the law and justice. According to the Mollen Commission, formed to investigate New York Police Department (NYPD) corruption in the early 1990s, police corruption and police violence often go hand in hand. Comparing a sample of 234 problem officers that NYPD selected based on corruption allegations and comments from field commanders to a random sample of 234 officers from similar commands, the Commission found that the officers alleged to be corrupt were over five times as likely to have five or more unnecessary force allegations against them than the officers from the random sample group (Mollen Commission, 1994, 46).

Lack of impartiality in the form of racial bias, explicit or implicit, is a major problem in many police departments. Over the last two decades, law enforcement officials with alleged connections to white supremacists have been exposed in many states (German, Reference German2020). Implicit bias is harder to show, but experimental studies (e.g., Correll et al., Reference Correll, Park and Judd2007; Sadler et al., Reference Sadler, Correll and Park2012) find evidence of bias against Blacks among police stemming from implicit attitudes and stereotypes. Compounding those problems is that police are rarely held accountable for their actions. Although approximately 1,000 police killings are reported each year in America, as of 2021, only 139 police officers had been arrested since 2005 for murder and manslaughter – a 1 percent arrest rate. Of those 139 officers, only forty-four (with forty-two cases still pending in 2021) were convicted, with most convictions coming on lesser charges, and some serving no prison time (Lopez, Reference Lopez2021). With little or no accountability and impartiality, the police can become a powerful force not only in themselves but for themselves, fostering an us-versus-them mentality. Based on data from two million 911 calls in two cities, Hoekstra and Sloan (Reference Hoekstra and Sloan2022) find that while White and Black officers dispatched to White and racially mixed neighborhoods fire their guns at similar rates, White officers are five times more likely to fire their guns when they are dispatched to predominantly Black neighborhoods.

Bad policies led by bad governance only exacerbate the problem. Since the 1990s, police departments in many cities have implemented variants of the policy known as “broken windows” policing, encouraging aggressive enforcement against minor offenses. That strategy has eventually morphed into “quota-based” policing. Arrest and ticket quotas are informal, and even illegal in several states, but they exist. A 2015 lawsuit filed by a group of police officers claimed that members of the NYPD were coerced into targeting Black men to fulfill their arrest quotas (Goldstein and Southall, Reference Goldstein and Southall2020). In 2019, three officers, including a police chief, in Florida were sentenced to prison for falsely arresting Black men in order to keep their department’s burglary clearance rate at 100 percent. Such policies have caused the number of contacts – often physical – between police and Black Americans to increase, increasing the likelihood of more deaths of Blacks. George Floyd, for example, was accused of passing a bad $20 bill, while Eric Garner was suspected of selling cigarettes illegally on a street corner.

5.3.2 Political Culture

We would expect social values to influence the laws, the social order police are expected to uphold, and the powers and limits – both intended and actual – shaping their conduct. To incorporate those values into our analysis, we draw upon Elazar’s (Reference Elazar1984) typology of American political subcultures. Elazar, as noted in Chapter 3, classified the states by three political subcultures or combinations among them: Moralistic (M), Individualistic (I), and Traditionalistic (T).

Elazar’s view of states as civil societies is important here. Political culture as an attribute of a civil society is not merely an overall balance of public opinion, much less a harmonious consensus. Rather, it is a dynamic but lasting pattern of relationships among segments of the population and the wider society. Internal diversity is a fact of life in most states but its significance, and how a state deals with it, is part of what Elazar’s theory compares and contrasts. A state’s political culture is not a matter of formal agreement; indeed, it is unlikely to be fully spelled out anywhere nor be consciously articulated by anyone. It is more a set of longstanding, overarching beliefs about what the civil society is; what defines and justifies relationships among government, public interest, and self-interest; acceptable tactics and limits of political contention and social competition; and expectations regarding the ways people will deal with each other. Larger and more diverse states can reflect a mix of political cultures, and social change may gradually reshape the political culture over the long term. Moreover, segments of the population may well dissent from the dominant settlement – sometimes, sharply. Thus, a state’s law enforcement and policing, involving as they do the implementation and defense of laws, policies, underlying values, and visions of the social order, will be the focus of strongly held – at times, clashing – values, aspirations, and fears. States embodying contrasting political cultures deal with those internal issues in distinctive ways, as we shall see.

A focus on state political cultures in a discussion of killings by police may strike many as misguided, particularly when the scheme dates back to the 1960s and characterizes states based on migration and settlement patterns in the late 1800s and early 1900s. Elazar makes it clear, however, that American states as civil societies have longstanding and distinctive histories, value systems, and responses to change. Moreover, as local police forces exercise powers delegated by their respective states, they are at least theoretically accountable to state laws, and routinely deal with people and expectations originating well beyond the city limits. Those expectations can vary significantly: according to a recent report by Amnesty International (2015), only eight states require that a warning be given (where feasible) before lethal force is used, and no state meets the requirement for a warning under international standards. Moreover, none of the states’ “use of lethal force” laws include accountability such as obligatory reporting for the use of force and firearms by law enforcement officers.

In the absence of mandated standards governing the use of force, we might well expect local values, traditions, and political dynamics to play a critical role in shaping police behavior. As for the time dimension, a strength of Elazar’s scheme is that it spells out influences that are deeply rooted, slow to change, and reflect long-term influences upon matters of current concern – a key issue when considering systemic racism and corruption. Political cultures are integral to the systems of social order police are expected to maintain, and it seems more likely they would be widely expected to moderate major social change (hence, the “thin blue line”) than to facilitate it.

5.3.3 Political Ideology

Political culture is not the same as political ideology. As Fisher (Reference Fisher2016) argues, states exhibiting any of the three political cultures can be either liberal or conservative or some mixture of both. Utah, for example, is a moralistic state, as is Minnesota, but while Utah is one of the most conservative and Republican states, Minnesota is liberal and Democratic.

According to social dominance theory (Sidanius and Pratto, Reference Sidanius, Pratto, Sniderman, Tetlock and Carmines1993), conservatism and racism are correlated because both seek to uphold the superiority of one group over others. Social dominance theory views all societies as group-based hierarchies in which at least one dominant group enjoys a disproportionate share of positive social value (e.g., wealth, health), and at least one subordinate group endures a disproportionate share of negative social value (e.g., incarceration). Politics becomes an exercise in intergroup competition over scarce resources, with political ideology supporting claims upon these resources (Sidanius et al., Reference Sidanius, Pratto and Bobo1996). Liberals, by contrast, tend to be more sensitive to, and upset by, inequality than conservatives (Napier and Jost, Reference Napier and Jost2008; Jost et al., Reference Jost, Federico and Napier2009), which helps explain why liberals are more likely than conservatives to perceive racism as a problem (Cooley et al., Reference Cooley, Brown-Iannuzzi and Cottrell2019). Not surprisingly, there is also a big divide between liberals and conservatives in how the police are perceived. According to a Cato Institute survey, Democrats (40 percent) are about half as likely as Republicans (78 percent) to believe the police are impartial. Some 80 percent of Republicans believe that police only use lethal force when necessary, while 63 percent of Democrats believe police are too quick to use it (Ekins, Reference Ekins2017). Quite apart from the attitudes of police officers themselves, ideology seems likely to influence the ways the police and their conduct are perceived, as well as community expectations and responses integral to encounters with the public.

5.3.4 Special Interest Groups: The Influence of Police Unions

Police cannot police themselves, and ordinarily will not be policed by civilian elected officials, for familiar reasons including their effective monopoly on legitimate coercion in most situations and the us-versus-them outlooks noted already. But another major factor contributing to weak police accountability has only come in for public debate relatively recently. That is the way police unions push back in almost every state against laws intended to hold police accountable and the influence they exercise in prosecutorial and other local and state elections via contributions and endorsements. Officers who see themselves as beyond the control of civilian officials, and who view prosecutors as reliable allies, may be less constrained by restrictions on the use of force. While police are theoretically accountable to elected officials in their jurisdictions, if they are not just a political force in themselves but a force for themselves such accountability may be weak. When district attorneys run for office, they often get donations and public backing from powerful police unions representing the officers they are supposed to prosecute in the event of misconduct. District Attorneys are not the only ones receiving donations from police unions. According to a report published by Campaign Zero, a police reform group launched in 2015, over a ten-year period Governor Jerry Brown and 118 of California’s 120 state legislators received contributions from police unions (Campaign Zero, 2018). In 2016, a bill that would have allowed public access to police misconduct records was killed in that state’s Senate Appropriations Committee (Thompson, Reference Thompson2016). In New York, it took almost a decade and the killing of George Floyd to repeal Section 50A of New York’s Civil Rights Law, which exempted police misconduct records from disclosure under the state’s open records law. In recent years, police unions spent more than $1 million supporting vulnerable incumbent legislators who opposed the repeal. In more than two-thirds of the states, a police officer’s disciplinary history is mostly unavailable through public records requests (Barkan, Reference Barkan2020). Indeed, fourteen states have “police bill of rights” laws giving special protections to police officers under investigation for misconduct (Campaign Zero, 2018).

5.3.5 Political Party Competition

Political parties offer special interest groups access to policymaking in exchange for their support, typically in the form of campaign contributions and endorsements. While competing with each other in some respects, parties have issues they “own” in the sense that voters consistently consider one better than others at dealing with those issues (Otjes and Green-Pedersen, Reference Otjes and Green-Pedersen2019). While the Republican Party is known as the “party of law enforcement,” Democratic Party is known as the “party of labor.” Republicans oppose unions in the public or private sector unless they represent the police – because police tend to vote Republican and the union is an effective way to mobilize them. Thus, police unions differ from the rest of the labor movement; they rely on the conflict symbolized by the “thin blue line” rhetoric and view themselves as police first, with other issues being secondary. Even their terminology differs: they are generally called “associations,” not “unions,” and their chapters are usually called “lodges,” not “locals” (Williams, Reference Williams2015). Most are not affiliated with the nation’s major labor federations: the International Union of Police Associations (IUPA), for example, is the only police union affiliated with the AFL-CIO. Nevertheless, IUPA is very influential. For example, while one AFL-CIO union passed a resolution urging that IUPA be disaffiliated after the killing of George Floyd, the move was rejected by the larger Federation: because of the continuing decline of unionization in America, the AFL-CIO could not afford to lose any members (Kelly, Reference Kelly2020). Similarly, the Democratic Party could not afford to lose police union support; while Democrats want police reform, as the “party of labor,” they also support public sector unions, including police unions. Moreover, neither party can afford to be viewed as “soft on crime.” In these ways, police unions have accumulated significant influence in both political parties.

Via public endorsements, campaign contributions, and personal connections, police unions can build lasting political relationships in either party, weakening pressures for accountability. From the unions’ point of view, such relationships reflect important long-term political investments – relationships expected to withstand short-term controversies. In more competitive states, both parties will compete for police unions’ support. In Illinois, for example, Rahm Emanuel, then the Democratic Mayor of Chicago, did not mind picking a fight with the Chicago Teachers Union but took care to maintain a friendly relationship with the Chicago Fraternal Order of Police (Emmanuel, Reference Emmanuel2020).

5.4.1 Data

Our dependent variable is the percentage of Blacks among people killed by the police (% Black Civilians Killed) in each state from 2013 to 2019. The data are from mappingpoliceviolence.org, an advocacy group collecting data using sources such as fatalencounters.org as well as social media, obituaries, and police reports. The data from both mappingpoliceviolence.org and fatalencounters.org have been used by Nicholson-Crotty et al. (Reference Nicholson-Crotty, Nicholson-Crotty and Fernandez2017) and Johnson et al. (Reference Johnson, Vil and Gilbert2019). Police killings are defined as a person killed as a result of being shot, beaten, restrained, intentionally hit by a police vehicle, pepper sprayed, tasered, or otherwise harmed by police, be they on-duty or off-duty. Over the years covered in our study, more than 7,500 people were killed by the police in America; 23 percent of them were Black. A 2019 Census Bureau estimate put the Black share of the United States population at 13.5 percent.

Our main variable of interest explaining variations in police killings is corruption. We measure corruption using CCI. Because it likely takes time for corruption to affect police behavior, we use moving averages of CCI over five years. The relationship between % Black Civilians Killed and CCI is shown in Figure 5.1.

Figure 5.1 CCI and % Black Civilians Killed

The other political variables we are interested in are political culture and ideology of the states, the influence of special interest groups – particularly police unions – in state politics, and political party competition. To investigate how political culture affects corruption, an operational measure of Elazar’s (Reference Elazar1984) classification is needed. Following Dincer and Johnston (Reference Dincer and Johnston2017), we create a dummy variable equal to 1 if the dominant political culture in a state is Moralistic and 0 otherwise. To measure how liberal/conservative the voters are in each state, we use the Ideology Index constructed by Berry et al. (Reference Berry, Ringquist and Fording1998, Reference Berry, Fording and Ringquist2013). The index runs from 0 to 100, with 0 representing extremely conservative and 100 representing extremely liberal states. To measure police union influence, we use data from the National Institute on Money in Politics showing the number of special interest groups registered to lobby in each state (Lobby). The influence of police unions in state capitols is positively related to the influence of special interest groups in general. Business associations and labor unions are the two most influential special interest groups in each state, frequently opposing each other. It is interesting to note, however, that no matter who the winner is, police unions never lose. Union-busting bills backed by business associations in Wisconsin in 2011 and Iowa in 2017, for example, both exempted police and firefighters. Finally, political party competition is measured by the Folded Ranney Index, which runs from 0.5 to 1, with 0.5 representing no competition and 1 representing perfect competition in each state (Holbrook and La Raja, Reference Holbrook, La Raja, Gray, Hanson and Kousser2017).

Our demographic and economic control variables follow the literature and reflect our arguments earlier about the importance of policing in maintaining an established social order and, by implication, in dealing with inter-group tensions. We first control for population density (Density) and share of Black population (% Black Population) in each state. We expect more killings in densely populated states with large Black populations. Next, we control for how segregated Blacks and Whites are in the largest city in each state, as segregation might be expected to be associated with more killings. As Enos and Celaya (Reference Enos and Celaya2018) argue, segregation facilitates categorization, making social categories, such as race, more cognitively salient and, thus, leading to stereotyping and discrimination. The most commonly used measure of segregation between two groups is the Dissimilarity Index, measuring the relative distributions of each group across neighborhoods within the same city. The index runs from 0 to 100, with 0 representing total integration – that is, that both groups are distributed in the same proportions across all neighborhoods. An index of 100 represents total segregation, indicating that the members of one group reside in completely different neighborhoods from those of the second group (data from CensusScope). Economic controls are Median Income and Racial Income Gap, measured as the White/Black median income ratio (data from the Bureau of the Census). We expect the Racial Income Gap to be positively related to police killings, but we have no explicit hypothesis regarding Median Income.

Our final three controls relate to police in each state: share of full-time sworn police officers who are Black (% Black Police), share of full-time officers represented by a union (% Collective Bargaining), and share of feloniously killed police officers (% Police Killed). The data for % Black Police and % Collective Bargaining are from the Bureau of Justice Statistics’ Law Enforcement Management and Administrative Statistics (LEMAS), while the data for % Police Killed are from the FBI’s Law Enforcement Officers Killed and Assaulted (LEOKA) Program. Previous studies did not find statistically significant effects of % Black Police on police killings (Hickman and Piquero, Reference Hickman and Piquero2009; Johnson et al., Reference Johnson, Vil and Gilbert2019), but none of them investigated how share of Black police officers interacts with share of Black population. We expect fewer police killings in states with large Black populations if the police departments in these states employ more Black police officers. With respect to collective bargaining by police, several studies have found that collective bargaining agreements protect police from accountability for misconduct including killings (Dharmapala et al., Reference Dharmapala, McAdams and Rappaport2019; Johnson et al., Reference Johnson, Vil and Gilbert2019). Jason Van Dyke, for example, convicted for shooting Laquan McDonald in Chicago in 2014, had had twenty citizen complaints, at least ten of them for excessive force; none resulted in disciplinary action. Finally, we expect violence against police officers to increase police shootings in a state. There is anecdotal evidence showing a positive relationship between violence against police and police shootings (BondGraham, Reference BondGraham2019).

5.4.2 Results

We estimate a random effects (RE) model with feasible generalized least squares (FGLS) because some of our variables of interest and some of our control variables are time invariant:

FGLS estimation, which uses both between- and within-groups variation in the data, is efficient, and the Hausman test does not reject the null hypothesis that the RE model is consistent. The results of FGLS estimation are presented in Table 5.1. In all estimations, we control for region and year dummies.

We start with estimating a parsimonious model with only a few demographic control variables and end with a comprehensive model with all political variables. The estimated coefficient of CCI is positive and statistically significant in all estimations, indicating that in more corrupt states Blacks are killed by the police disproportionately. The magnitude of the effect is significant as well. A one standard deviation increase in CCI causes % Black Civilians Killed to increase by 0.10 standard deviations, or 5 percentage points. The median value of % Black Civilians Killed is approximately equal to 0.20. The standardized coefficients of % Collective Bargaining and % Black Police (in states with large Black minority populations such as Alabama and South Carolina) are the same.

The signs of the political variables follow our expectations. In Moralistic states, the share of Black civilians killed is significantly lower (more than 5 percentage points). The influence of police unions in state politics and the share of police covered by a collective bargaining agreement negotiated by unions are both positively related to % Black Civilians Killed. Political party competition too increases the share of Black civilians killed by the police.

The signs of the estimated coefficients of our control variables are mostly in line with the previous literature. In densely populated states in which Blacks and Whites are segregated, Blacks are killed disproportionately. The effect of the size of the Black minority in a state’s population on police killings varies with the Black percentage of officers employed in police departments: a larger Black percentage in the population is related to more police killings of Blacks, but as police departments employ more Black officers that effect decreases. The effects of Racial Income Gap on % Black Civilians Killed are also conditional on how wealthy a state is: in states with higher median incomes, the effect of Racial Income Gap upon police killings of Blacks remains positive but is smaller. Figures 5.2 and 5.3 show the marginal effects of % Black Population (conditional on % Black Police) and Racial Income Gap (conditional on Log Median Income) on % Black Civilians Killed are within the observed range of data. The downward sloping solid line represents the marginal effects, while the dotted lines represent the 95 percent confidence intervals for those estimates. Finally, as expected, there is a positive relationship between violence against the police and the police violence against Black civilians. In several respects, our data are consistent with the view that policing tends to uphold the dominant social order and values rather than impartially applying the law, and that corruption – signaling as it does problems with accountability to legal and professional standards – adds to those problems and their consequences.Footnote ³

Figure 5.2 Marginal effects of % Black Population (conditional on % Black Police)

Figure 5.3 Marginal effects of Racial Income Gap (conditional on Log Median Income)

5.4.3 Robustness of Results

To investigate whether our results are robust with respect to different measures of corruption, we also estimate our regressions using CRI. As with CCI, we use the five-year moving averages. The results of the RE estimation are presented in the first column of Table 5.2. The estimated coefficient of CRI is positive and statistically significant.

The second robustness issue is the estimation method. We also estimate our comprehensive model controlling for state fixed effects, excluding the time invariant control variables. The estimated coefficient of CCI is equal to 0.008 – only slightly lower than the FGLS estimate of 0.009, but only marginally significant statistically, since within-state variation in CCI is not high enough in the period covered in our sample.Footnote ⁴

7.2.1 Corruption and Compliance with Social Distancing and Vaccination

Starting with California in mid-March of 2020, the majority of American states instituted shelter-in-place/stay-at-home orders as part of their social distancing policies. Schools, restaurants, and bars were closed; nonessential businesses were ordered to keep workers home and let them work remotely. People were asked not to leave their homes unless necessary.

Enforcement and penalties, however, varied significantly across the states. While in some there were no penalties, in most violations were considered a misdemeanor punishable with a small fine, usually not enforced. In Arizona, for example, violators could be charged with a class 1 misdemeanor, which has a fine up to $2,500. In Tennessee, on the other hand, no penalties were specified (see Mazziotta, Reference Mazziotta2020, for state-by-state fines and penalties). In states such as Kentucky, Maryland, Michigan, and Pennsylvania protesters, sometimes armed with assault rifles, packed state capitols and streets in defiance of state orders. In London, by contrast, police arrested people in Hyde Park protesting social distancing orders (Estes, Reference Estes2020; Parveen, Reference Parveen2020). Thus, although social distancing was mandated in theory, in practice states often depended heavily on voluntary compliance. This resulted in significant variations in social distancing across the states.

While policies such as shelter in place/stay at home were somewhat effective, as a response to widespread resistance to complying with social distancing, the federal government introduced Operation Warp Speed (OWS), supporting the development of vaccines. Once vaccines were granted emergency use authorizations by the Food and Drug Administration (Fisk, Reference Fisk2021) in late 2020, OWS worked with pharmacies across the US to make them available at no cost to individuals. To achieve herd immunity, a large majority of the population needed to be vaccinated. Although the herd immunity threshold is not known for COVID-19, according to the World Health Organization, the figure for measles is 95 percent and for polio, 80 percent. As of early 2023, only 65 percent of the US population had been vaccinated against COVID-19, and there was significant variation across states ranging from 50 percent in Alabama and Mississippi to 80 percent in Rhode Island and Vermont.

Corruption might not seem a likely influence upon compliance with social distancing and vaccination, but in many ways, it embodies the ways people sort out their own preferences versus civic responsibility. Our data suggest that it did influence compliance through its effects on trust in government, social capital, and government legitimacy, which in turn are important determinants of compliance with shelter-in-place/stay-at-home orders and vaccination.

7.2.2 Trust in Government and Social Capital

Trust in government is an important determinant of social capital, which is defined as networks and norms shared among people and facilitating cooperation to help solve collective action problems (Fukuyama, Reference Fukuyama1995; Putnam, Reference Putnam2000). Social capital manifests itself in communities as reciprocal relationships between civic participation and interpersonal trust. The more people participate in their communities, the more they trust others, all else being equal; the greater trust people hold for others, the more likely they are to participate (Fukuyama, Reference Fukuyama1995; Brehm and Rahn, Reference Brehm and Rahn1997). Interpersonal trust depends heavily on trust in government (Levi, Reference Levi, Braithwaite and Levi1998; Levi and Stoker, Reference Levi and Stoker2000; Rothstein, Reference Rothstein2000, Reference Rothstein2005). According to Rothstein and Eek (Reference Rothstein and Eek2009), when forming their beliefs about others in a community, people draw inferences from the behavior of government officials: “[I]f it proves that I cannot trust the local policemen and judges, then whom in the society can I trust?” (Rothstein and Eek, Reference Rothstein and Eek2009, 90). Several political science studies find negative effects of corruption on trust in government (Anderson and Tverdova, Reference Anderson and Tverdova2003; Chang and Chu, Reference Chang and Chu2006; Rothstein and Eek, Reference Rothstein and Eek2009; see also Ch. 5 of this book). Because trust in government and interpersonal trust are positively related, corruption affects interpersonal trust negatively. Lower interpersonal trust in turn means lower civic participation, and lower civic participation means lower social capital. Since both social distancing and vaccination are collective action problems, we expect less compliance with shelter-in-place/stay-at-home orders and less vaccination in more corrupt states with low social capital.

Collective action problems arise when short-term interests of individuals conflict with long-term collective interests. According to Olson (Reference Olson1971: 2), if the members of a large group rationally seek to maximize their personal welfare, they will not act to advance their group objectives unless there is coercion to force them to do so, or unless some separate incentive, distinct from the achievement of the group interest, is offered to the members of the group individually on the condition that they help bear the costs or burdens involved in the achievement of the group objectives.

Sønderskov (Reference Sønderskov2009) argues that social capital facilitates collective action in large groups. Individuals are more likely to cooperate in collective action when they expect others to do likewise. Given a large number of individuals, it is costly and difficult, even impossible, to get specific information on other individuals’ trustworthiness. In such situations, social capital serves as a short-cut to information on trustworthiness (Sønderskov, Reference Sønderskov2009: 53–54; see also Ostrom, Reference Ostrom1998).

Gilson (Reference Gilson2003) argues that the production of health and health care requires cooperation within health systems, which in turn requires trust. Trust facilitates the use of the health system and leads to better self-rated outcomes (Radin, Reference Radin2013). Several empirical studies present persuasive evidence regarding the effects of trust in government in particular, and social capital in general, on compliance with public health policies (Buckman et al., Reference Buckman, Liu, Cortright, Tumin and Syed2020). Blair et al. (Reference Blair, Morse and Tsai2017), for example, investigate the behavior of Liberians during the 2014–2015 Ebola outbreak, finding that people with lower trust in government took fewer steps to protect themselves and were less likely to comply with social distancing orders. Vinck et al. (Reference Vinck, Pham and Bindu2019), in the context of a later outbreak in the Democratic Republic of Congo, find similar results: low trust in government explains lower willingness to adopt preventative behavior and accept a vaccine. The relationship among trust, social capital, and compliance with public health policies is not specific to COVID-19, or to Sub-Saharan Africa: Nawa and Fujiwara (Reference Nawa and Fujiwara2019) using data from Japan, Ronnerstrand (Reference Berry, Fording and Ringquist2013) from Sweden, Chuang et al. (Reference Chuang and Ya-Li Huang2015) from Taiwan, and Jung et al. (Reference Berry, Fording and Ringquist2013) and Ronnerstrand (Reference Rönnerstrand2014) from America all find a positive relationship between trust/social capital and vaccination. Yaqub et al. (Reference Yaqub, Castle-Clarke and Sevdalis2014) identify low trust/social capital as a main cause of vaccine hesitancy in Europe. Trust and social capital are also found, in some contexts, to be determinants of mental and physical health (Lochner et al., Reference Lochner, Kawachi and Brennan2003; Yip et al., Reference Yip, Subramanian and Mitchell2007; Kim et al., Reference Kim, Subramanian, Kawachi, Kawachi, Subramanian and Kim2008; Almedom and Glandon, Reference Almedom, Glandon, Kawachi, Subramanian and Kim2008; Ahnquist et al., Reference Ahnquist, Wamala and Lindstrom2012; Rodgers et al., Reference Rodgers, Valuev and Hswen2019). According to Kawachi and Berkman (Reference Kawachi, Berkman, Berkman and Kawachi2000), trust and social capital affect health-related behaviors via diffusion of health information and adoption of healthy norms of behavior (see also Lindstrom, Reference Lindstrom, Kawachi, Subramanian and Kim2008).

7.2.3 Trust in Government and Legitimacy of Government

The second channel through which corruption affects social distancing is the legitimacy of government. Legitimacy is linked to trust in government and is defined as the belief that government does what is appropriate and fair most of the time. As such, we expect legitimacy to affect how people behave toward government in crises (Easton, Reference Easton1965; Tyler, Reference Tyler2006; Christensen and Lægreid, Reference Christensen and Lægreid2020). Legitimacy increases compliance with government policies (Tyler, Reference Tyler2006). Christensen and Lægreid (Reference Christensen and Lægreid2020) argue that the comparative success of the Norwegian government in fighting COVID-19 (Cohen, Reference Cohen2022) was partially due to its legitimacy. It is crucial that policies implemented to fight the pandemic be seen by the people as appropriate and fair (Christensen and Lægreid, Reference Christensen and Lægreid2020). Government legitimacy is important for compliance with laws and regulations. Paternoster et al. (Reference Paternoster, Brame and Bachman1997) find that domestic assault suspects arrested in a procedurally fair manner are less likely to commit further acts than those arrested in procedurally unfair ways. Kuperan and Sutinen (Reference Kuperan and Sutinen1998) find that legitimacy affects compliance by Malaysian fishermen with a regulation banning them from fishing in a zone along the coast. Levi and Sacks (Reference Levi and Sacks2009), using survey data from Sub-Saharan Africa, find that compliance with tax laws is positively related to government legitimacy. Seligson (Reference Seligson2002) and Boly et al. (Reference Boly, Gillanders and Miettinen2019) find negative relationships between corruption and government legitimacy, and Ali et al. (Reference Ali, Fjeldstad and Sjursen2014) find that corruption weakens tax compliance in South Africa and Uganda. In the COVID-19 context, therefore, to the extent that corruption weakens the legitimacy of government, we expect it will also reduce compliance with shelter-in-place/stay-at-home orders.

7.2.4 Empirical Analysis: Data

Investigating the relationship between corruption and social distancing presents several challenges, perhaps the most important being the measurement of compliance with social distancing. We use the Shelter-in-Place Index constructed by SafeGraph,Footnote ³ measuring stay-at-home – the percentage of people in a state staying home on a particular day – compared to population movement data representing 45 million smartphone devices (CDC, 2021). The index ranges from −100 to 100, where 0 indicates no change from a baseline defined as the average percentage of people staying home all day and every day across the seven days ending February 12, 2020. Home is defined as the most common night-time location of the smartphone device in the previous months, identified to a precision of about 100 square meters. The Shelter-in-Place Index for any particular day is constructed as follows:

$\begin{array}{ll}Shelter-in-PlaceIndex=\hfill & \text{Actual}\text{\% Staying Home}\hfill \\ \hfill & -\text{Baseline}\text{\% Staying Home}\hfill \end{array}$

On May 2, 2020 (the last day of our sample), for example, the actual percentage of people staying home in Illinois was 40.5 while the baseline was 25. Hence, the index for Illinois on May 2 is equal to 15.5 (see data and documentation at CDC, 2021 for details). Our sample covers four consecutive Saturdays starting from April 11. Over the last three weeks of April and the first week of May, shelter-in-place/stay-at-home orders were in place in all but five states. While these data are drawn from the early days of what became a multi-year pandemic, we believe they are appropriate: the pandemic was a new and threatening public concern, people’s attention regarding government countermeasures ran high, and lines of partisan contention over COVID-19 responses had yet to harden into place. We measure corruption using CCI. We also control for several economic and demographic variables to minimize omitted variable bias.

We expect corruption to affect social distancing through two channels: social capital/interpersonal trust and trust in government/legitimacy of government. To control for the mediating effects of social capital, we first control for how generous people are, and how engaged they are civically, in each state in terms of charitable giving and volunteering. These are the most relevant components of social capital in the context of this study. We use the State Generosity Index (SGI) constructed by WalletHub, which ranges from 0 to 100 (see WalletHub, 2022 for details). We construct a dummy variable, SGI, which is equal to 1 if a state falls into the highest quartile of the SGI.

In addition, we control for the percentage of people tested positive for COVID-19 in each state on four consecutive Thursdays starting from April 9, 2020 (COVID-19 Positive). Since higher positivity rates indicate a higher infection risk, they should affect compliance with social distancing orders. The third control variable is Trump – the percentage of votes Donald Trump received in each state in the 2016 presidential election (US Elections Project, 2023). Trump showed his support loudly and repeatedly for people protesting (and violating) shelter-in-place/stay-at-home orders issued by the governors of several states. In addition, he called into question the legitimacy of public institutions and leaders in numerous ways. The variable Trump also controls for political ideology, which affects risk perception: according to a poll conducted by Axios/Ipsos, while 60 percent of Republicans believed that the real number of COVID-19-related deaths was lower than the official count, more than 60 percent of Democrats believed the real number was higher (Axios, 2020). Using state-level data, Hsiehchen, Espinoza, and Slovik (Reference Hsiehchen, Espinoza and Slovic2020) find that Republicans were less likely to comply with social distancing orders.

The next two control variables are per capita personal income (Income) and unionization (Union). Social distancing is costlier for some than others: people living paycheck to paycheck, and those who cannot work from home, may be less likely to comply with stay-at-home orders. Regarding unionization, the pandemic resulted in job losses across the country, but many union workers had various protections due to their collective bargaining agreements. NBC News reported that approximately 150,000 United Auto Workers members at Ford, General Motors, and Fiat Chrysler lost their jobs but continued to receive supplemental unemployment benefits. Their contracts gave members at least six months of extra pay on top of unemployment insurance, adding up to 85 percent of their hourly wages. Unionization also affects social capital via political and civic engagement (Nissen, Reference Nissen2010; Kerrisey and Schofer, Reference Kerrisey and Schofer2013). Our Income data are from the Bureau of Economic Analysis (2023) and Union data are from the Union Membership and Coverage Database (Unionstats, 2023). Finally, we control for population density, as the risk of infection is higher in densely populated urban communities. The data are from the Census Bureau (United States, Department of Commerce, 2023).

In our vaccination regressions, we use the share of population who are fully vaccinated as of January 2023 as our dependent variable (Kaiser Family Foundation, State Health Facts – KFF, 2023). Our control variables are chosen based on several nationally representative surveys conducted during the pandemic (Daly and Robinson, Reference Daly and Robinson2021; Sziglayi et al., Reference Sziglayi, Kyla and Shah2021; Ruiz and Bell, Reference Ruiz and Bell2021) and systematic reviews such as Wang and Liu (Reference Wang and Liu2022). Demographic and economic factors such as education, race, and income play a significant role in people’s decision to get vaccinated. While politically conservative people are also less likely to be vaccinated, vaccination rates are higher in more densely populated states. We measure education as the share of population age 25 and above who have less than a high school degree, race as the population share of Blacks and income as per capita personal income. The data are from the Census Bureau and the Bureau of Economic Analysis as before. Finally, to measure political ideology, we construct a dummy variable, which is equal to 1 if the governor of a state is Republican.

7.2.5 Empirical Analysis: Results

To investigate the effects of corruption on social distancing, we estimate a system of four equations with seemingly unrelated regressions (SUR) in which the dependent variables are the Shelter-in-Place Indices for April 11, April 18, April 25, and May 2, 2020. Each Saturday forms one fourth of the system. Estimating a system with SUR has several advantages: first, SUR is more efficient because it allows errors to be correlated across the equations. Second, because each equation has the same set of independent variables, it allows us to conduct joint tests to investigate whether quarantine fatigue worsens the collective action problem. Quarantine fatigue manifests itself as boredom, anxiety, and stress (Marcus, Reference Marcus2020) and can make it more difficult to comply with stay-at-home/shelter-in-place orders when people are feeling the psychological effects of social distancing. Where trust/social capital is low, this becomes an even greater problem.

The maximum likelihood estimates with robust standard errors clustered at the state level are presented in Table 7.1. To show the mediating effects of social capital, we first estimate our regression without SGI, the generosity index. The estimated coefficient of CCI is negative and statistically significant in all four equations, indicating that in corrupt states people are less likely to comply with shelter-in-place/stay-at-home orders. The absolute value of the estimated coefficient decreases when we estimate the regression with SGI, indicating that one of the channels through which corruption affects social distancing is indeed via deterioration in social capital. The magnitude of the effect is significant as well. Based on the estimated coefficients in Equation 4, a one-standard deviation increase in CCI causes the Shelter-in-place Index to decrease by approximately 10 percent. The standardized effect of COVID-19 Positive is only slightly greater than 10 percent, suggesting that corruption was roughly as strong a determinant of compliance as were trends in COVID-19 itself. Perhaps more interesting, the magnitude of the corruption effect increases each week. The estimated coefficient of CCI in Equation 4 (Shelter-in-Place Index on May 2) is 2.5 times greater than the one in Equation 1 (Shelter-in-Place Index on April 11).Footnote ⁴ As “quarantine fatigue” developed, people stayed home less and less during the time period that our sample covers. Across all states, on average, the Shelter-in-Place Index decreased by 30 percent from April 11 to May 2. On the other hand, in states such as Mississippi, Montana, Oklahoma, and South Dakota that fall into the highest quartile of CCI, compliance decreased more than 50 percent, even though the risk of infection remained high and the percentage of people testing positive for COVID-19 stayed the same or increased. Our results show that “quarantine fatigue” makes the collective action problem even more difficult where corruption is high.

The estimated coefficients of the control variables are also statistically significant and their signs are consistent with our expectations.Footnote ⁵ People responded to shelter-in-place/stay-at-home orders if more people tested positive for COVID-19. In densely populated and more charitable states, they stayed at home more as well. The estimated coefficient of Trump is not only negative, as expected, but its magnitude is also greater than that of CCI, consistent with the increasing political polarization regarding COVID-19 issues. Based on the estimated coefficients in Equation 4, a one-standard deviation increase in Trump reduces the Shelter-in-Place Index by 20 percent. Finally, in richer states and states where more workers are unionized, more people complied with social distancing policies.Footnote ⁶

We use the share of population who were fully vaccinated as of January 2023 as our dependent variable in our vaccination regressions and estimate the effect of corruption using ordinary least squares (OLS). The results are given in Table 7.2. The estimated coefficient of CCI is again negative and statistically significant in all estimations, indicating that in more corrupt states people are less likely to be vaccinated. The absolute value of the coefficient decreases when we estimate the regression with SGI, again indicating that one of the channels through which corruption affects vaccination is via deterioration in social capital. The magnitude of the effect is significant as well. Based on the estimated coefficients in Equation 1, a one-standard deviation increase in CCI causes the percentage of people fully vaccinated to decrease by approximately 3.5 percent. Again, the estimated coefficients of the control variables are statistically significant and their signs are consistent with our expectations.Footnote ⁷

Unfortunately, we cannot control for some important variables in our regressions because state-level data are not available. The first two are interpersonal trust and trust in government, both of which are likely to mediate the effects of corruption on social distancing. Low interpersonal trust may make people more likely to self-quarantine because they question others’ social distancing behavior. It may also weaken social networks. Without networks, staying home is less of a sacrifice for an individual and less costly. Trust in government is a crucial determinant of government legitimacy. Unfortunately, two frequently used surveys that ask questions regarding trust, the American National Election Study (ANES) and the General Social Survey (GSS), are both nationally representative surveys, and sampling is not done at the state level.

The second variable is social and traditional media misinformation, which may affect people’s behavior towards social distancing and vaccination by, among other things, undermining government legitimacy. Bursztyn et al. (Reference Bursztyn, Rao, Roth and Yanagizawa-Drott2020) investigate the effects of misinformation by comparing two major cable news shows, Hannity and Tucker Carlson Tonight. These shows aired back-to-back on Fox News and had relatively similar content but differed significantly in their coverage of the COVID-19 pandemic. While Carlson warned his viewers about the dangers of COVID-19, Hannity dismissed the risks as less threatening than the common flu and contended that Democrats were using it against Donald Trump. Using survey data on more than 1,000 Fox News viewers aged 55 or older, Bursztyn et al. (Reference Bursztyn, Rao, Roth and Yanagizawa-Drott2020) found that viewership of Hannity was associated with changing behavior (washing hands, social distancing, etc.) four days later than other Fox News viewers, while viewership of Tucker Carlson Tonight was associated with changing behavior three days earlier. Bridgman et al. (Reference Bridgman, Merkley and Loewen2020) conducted a nationally representative survey of more than 2,000 Canadians 18 or older that included questions about common misperceptions regarding COVID-19, social distancing compliance, and exposure to traditional news and social media. They found that exposure to traditional news media was associated with fewer misperceptions and more social distancing compliance, while conversely, social media exposure was associated with more misperceptions and less compliance. Pierri et al. (Reference Pierri, Perry and DeVerna2022) find similar results regarding vaccination using county and state-level Twitter data from the first quarter of 2021 in the US.

7.2.6 A Matter of Trust?

Studies from across the social sciences have pointed to corruption as corrosive to trust/social capital, which are vital to compliance with public health orders. Using data from American states, we have shown that more corrupt states are likely to have lower compliance with shelter-in-place/stay-at-home orders and lower rates of vaccination.

While it would be facile to expect our results to hold in all other countries, the fact that many of the channels motivating our study are salient in a variety of different contexts does suggest that the links between corruption and compliance with public health orders could be fruitfully investigated in other societies. Even with the external validity caveats that would require, we believe that our results have practical implications. Fang et al. (Reference Fang, Wang and Yang2020) found that lockdown was an effective policy in China, while Dave et al. (Reference Dave, Friedson and Matsuzawa2021) found that stay-at-home/shelter-in-place orders at the outset of the pandemic reduced the number of COVID-19 infections by more than 50 percent over a span of three weeks in America. In that light, our findings suggest corruption has significant implications for the spread of COVID-19. Corruption also reduces the quality of healthcare (Mostert et al., Reference Mostert, Njuguna and Olbara2015; Friedman, Reference Friedman2018), making the problem even worse. Hence, more corrupt states might need additional resources to fight future pandemics, even though such states’ institutions might struggle to deal effectively with both a pandemic and a rapid influx of funds.

Finally, evidence shows that corruption reduces the willingness to contribute to pure public goods (Beekman et al., Reference Beekman, Bulte and Nillesen2014), and to the provision of quasi-public goods such as infrastructure (Gillanders, Reference Gillanders2014) – though perhaps only once it passes a threshold level (Bose et al., Reference Bose, Capasso and Murshid2008). Compliance with public health policies generates a benefit to society that is both non-rival and non-excludable – the two characteristics of a pure public good. Therefore, one interpretation of our results is that corruption reduces individuals’ willingness to contribute to pure public goods in the health domain, a finding with implications reaching well beyond the COVID-19 crisis. Organizations such as the World Bank view public health in general, and pandemic preparedness in particular, as global public goods (Stein and Sridhar, Reference Stein and Sridhar2017). Our results, subject to the external validity caveats noted earlier, identify corruption as a significant barrier to the provision of such global public goods.

7.3.1 Empirical Analysis: Data

Evidence from the 171 large metropolitan counties in thirty-five American statesFootnote ⁸ for which race-related data are available, as classified by the National Center for Health Statistics (NCHS), shows that corruption increases the gap between Black and White COVID-19-related deaths. We first estimate a linear model using both state and county-level variables with OLS regression, in which the dependent variable is the gap in COVID-19-related deaths between Blacks and Whites. Death Gap is measured as the population share of Blacks died from COVID-19 divided by the population share of Whites died from COVID-19.

The data for COVID-19-related deaths by race are from the Centers for Disease Control and Prevention (CDC) and cover the 15-month period from January 1, 2020 to March 31, 2021. Next, the same model is estimated with quantile regression to analyze whether the effects of corruption on Death Gap differ in counties with high health inequities. Corruption is measured using CCI data as described in Chapter 2.

Several county- and state-level politico-economic and demographic control variables are used in the regressions to minimize omitted-variable bias. If corruption is correlated with any of these control variables, omitting them causes the coefficient of CCI to be estimated with a bias. The first control variable is the Poverty Gap between Blacks and Whites, measured as the population share of Blacks who are poor divided by the population share of Whites who are poor. The data are from the Census Bureau’s American Community Survey (ACS). The second is political support for Donald Trump during the 2020 presidential election (Trump). Several opinion polls show that COVID-19 has strong partisan dimensions: Gollwitzer et al. (Reference Gollwitzer, Martel and Brady2020), for example, find that conservative counties are less likely to follow the CDC’s social distancing/facemask guidelines, leading to higher infections and deaths (see also the results earlier in this chapter). Indeed, two decades’ worth of mortality data from a variety of causes (Denworth, Reference Denworth2022) show that both Blacks and Whites in Democratic-voting counties have lower mortality figures than their counterparts in Republican-voting counties, although it must be noted that mortality rates are higher for Blacks than Whites in both cases. Perhaps a surprise is that differences in vaccine uptake accounted for only 10 percent of the county-level contrasts in COVID-19 deaths; the differences likely had more to do with behavioral contrasts such as the frequency of un-masked social activities. In any event,

The third control variable is per capita spending (in $1,000s) on public welfare in each state (Public Welfare Spending). It includes cash assistance through Temporary Assistance for Needy Families, Supplemental Security Income, and other payments made directly to individuals as well as payments to health care providers under programs like Medicaid. The data are from the Urban Institute. Demographic control variables are population density (Density) and Age Gap – measured as the population share of Blacks who are older than 64 divided by the population share of Whites who are older than 64.

Several studies, such as Wong and Li (Reference Wong and Li2020), find that densely populated counties with older populations have higher COVID-19-related death rates. The data are from ACS. All control variables are from 2020.

7.3.2 Empirical Analysis: Results

The results of the OLS estimation are presented in Table 7.3. The estimated coefficient of CCI is positive and statistically significant, indicating that corruption widens the gap in COVID-19-related deaths between Blacks and Whites. A one-standard deviation increase in CCI causes Death Gap to increase by approximately 10 percentage points. That result is not trivial: Poverty Gap has the same standardized positive effect on Death Gap. Then, the full model is estimated by adding Public Welfare Spending. The estimated coefficient of CCI stays positive and statistically significant, but its magnitude decreases, indicating that corruption not only affects health inequities directly but also indirectly through spending on public welfare.

Estimated coefficients of all control variables are statistically significant and have the expected signs: for example, in densely populated counties in which Age Gap is wider, Death Gap is wider as well. While increase in Distrust in Health Care causes Death Gap to widen, increased Public Welfare Spending narrows it. Moreover, in conservative counties supporting Donald Trump, the gap in COVID-19-related deaths between Blacks and Whites is wider.

Finally, the effects of corruption on Death Gap are estimated with quantile regression. OLS regression models the relationship between CCI and the conditional mean of Death Gap across all counties in the data set, thus providing only a partial view of the relationship. Quantile regression, by contrast, allows us to analyze this relationship at different points in the conditional distribution of Death Gap. This is important because the effects of corruption are likely to differ significantly, with a stronger effect on Death Gap at higher quantiles. In other words, corruption widens the Black/White gap in COVID-19-related deaths even more in counties with high health inequities. Figure 7.1 shows how the effects of CCI on Death Gap differ over quantiles and how the magnitude of the effects at various quantiles differs significantly from the overall OLS coefficient estimates, even in terms of the confidence intervals around each estimated coefficient.

Figure 7.1 Differential effects of corruption on the gap in COVID-19-related deaths between Blacks and Whites across quantiles

Black solid line represents the coefficients of CCI estimated by quantile regression while the gray solid line represents the one estimated by ordinary least squares. Dashed black and gray lines represent the respective 95 percent confidence intervals.

These results have important implications regarding vaccine equity; for example, data show that as of mid-2022, Black Americans were still getting vaccinated at a lower rate than Whites (Ndugga et al., Reference Ndugga, Hill and Artiga2022). Some might argue that vaccine hesitancy plays a significant role in that gap, but several nationwide surveys show that Black Americans are not hesitant to get vaccinated (Summers, Reference Summers2021). Political factors are important as well: while federal guidelines regulate the distribution of vaccines both across and within states, state governors decide who receives the first doses of the vaccine and who administers them. The responsibility to distribute and administer vaccines fairly and effectively falls first on governors and only then on health care providers.

Structural corruption affects vaccine equity as well. Corruption in that sense is not merely a matter of misconduct by individuals, but rather a manifestation of bad governance and a lack of impartiality on the part of institutions. There is ample anecdotal evidence of lack of impartiality in vaccination in the form of racial discrimination. In Louisiana and Georgia, two states that score high on the CCI, an NPR study (McMinn et al., Reference McMinn2021) found few vaccine sites in predominantly Black neighborhoods in urban counties. In Florida, another high-scoring state, priority was given to predominantly White gated communities whose developers were major contributors to political action committees supporting the governor (Shaw, Reference Shaw2021). Not surprisingly, the 2021 White-to-Black vaccination ratio was greater than 2 in Florida (Pham and Alam, Reference Pham and Alam2021).