In the late 1980s, Texas was one of the first states to contract with a private company to operate and manage correctional facilities. In the next decade, over twenty private prisons opened in the state, holding inmates under Texas, county, federal, or other states’ jurisdictions. Despite Texas’s growing experience with private prisons, the law continued to adapt to this new policy. Two Oregon inmates held in a Texas private prison escaped in 1996 and traveled nearly 200 miles before being apprehended (Associated Press 1996). Though this appears like a regular prison break, the state soon discovered standard practice in public prisons did not easily translate to private prisons, as the two men could not technically be charged with any crime because escaping from a private prison was not yet illegal. Texas soon corrected this oversight, though this example is typical of states’ ad hoc policymaking surrounding private prisons as they have become more commonplace across the country. Thirty-five states housed at least some of their inmates in private facilities at some point between 1986 and 2016, a policy choice that has attracted significant public scrutiny and controversy.
Modern prison privatization began in the 1980s as many state, local, and federal government services were also outsourced to the private sector. In 1986, at least 1,600 inmates were held in privately operated state, local, or federal prisons and jails. By 2016, that number had reached more than 163,000, about a hundred-fold increase in only thirty years. Though the share of inmates in these private facilities still remains low—18% of federal prisoners and 9% of state prisoners as of 2016 (Carson Reference Carson2018)—the policy has garnered significant controversy since its inception. State and federal policymakers and citizens alike have struggled to come to terms with the idea of government contracting with private companies to incarcerate the accused and convicted (Cody and Bennett Reference Cody and Bennett1987; Thompson and Elling Reference Thompson and Elling2002; U.S. Congress House Committee on the Judiciary and the Administration of Justice 1986). Research in the last few decades has sought to understand questions like whether private prisons are worse in terms of quality than public prisons, defining what is meant by a private prison, and chronicling their rise in the United States (e.g., Burkhardt Reference Burkhardt2019; Harding Reference Harding1997). Though these are all vital questions, this project takes a different tack and seeks to understand why states in particular choose to privatize their correctional facilities. Here I consider prison privatization to include only the private management and operation of state prisons, though many prisons also privatize individual services like healthcare, food service, or laundry (Dolovich Reference Dolovich2005; Selman and Leighton Reference Selman and Leighton2010).
The traditional political economy approach to framing and explaining prison privatization has been the focus of past studies (Kim and Price Reference Kim and Price2014; Nicholson-Crotty Reference Nicholson-Crotty2004; Page Reference Page2011). By this logic, Republican states and states with suffering economies will be more likely to privatize their prisons, whereas states with healthy unions will be less likely to privatize. I argue that none of these explanations alone are sufficient to explain states’ adoption of private prisons, highlighting the distinction between more common forms of privatization, like water and sewer services, and prison privatization.
Instead, I argue that an important overlooked variable in the decision to privatize prisons is increasing inmate litigation. States facing growing prisoner litigation are incentivized to privatize to avoid legal and political accountability for poor prison conditions (Raher Reference Raher2010; Tartaglia Reference Tartaglia2014). This unique set of circumstances is akin to other explanations of carceral state expansion that suggest additional institutional factors beyond partisanship explain prison policy (Gottschalk Reference Gottschalk2006; Hinton Reference Hinton2016; Murakawa Reference Murakawa2014). I contend that this paper has implications outside just prison policy, however. For one, it emphasizes the incentives of accountability and blame shifting in privatization. This is similar to dynamics described in the privatization of security forces or education, as governments are incentivized to privatize these essential services as well to shift accountability for the potential failure of security or education policy away from the state to private companies (Leander Reference Leander2010; Lipman and Haines Reference Lipman and Haines2007). But perhaps more importantly, my theory speaks to the literature on the utility of using courts for social change (e.g., Epp Reference Epp1998; Jacobs Reference Jacobs1980), and suggests that the rights revolution has potentially adverse consequences for those who stand to benefit from it. If the prisoners’ rights movement, and the broader rights revolution, encouraged inmates to be more litigious and one response to that growing litigiousness prompted privatization which may lead to adverse consequences for those inmates, how do we evaluate whether that movement was a success or not? To be clear, enshrining vulnerable populations like prisoners with legal rights is an essential development in our democracy, but I suggest that normatively appealing policy changes can have far-reaching downstream consequences.
This paper makes two main contributions. First, I argue and find empirical evidence for the core theoretical argument that growing inmate lawsuits is an important overlooked variable in the decision to privatize prisons, a strategy that helps states avoid legal and political accountability. Second, I introduce a novel dataset on publicly operated private prisons in the United States. I collected this dataset from thousands of pages of Securities and Exchange Commission reports on the location and capacity of private prisons from 1986 to the present, a significant data innovation in the study of private prisons. Taken together, I suggest that the character of states’ corrections systems, and in particular their policies regarding prison privatization, are significantly shaped by pressure exerted by the judicial branch and the legal challenges prisoners bring to bear on state carceral institutions.
Common Explanations for Privatization
At its most extreme, privatization strives for full private ownership of formerly public resources and organizations, that private businesses can operate these services better than the government can (Daley Reference Daley1996; Quinlan, Thomas, and Gautreaux Reference Quinlan, Thomas and Gautreaux2004). This political philosophy is associated with the Republican party specifically, as politicians under the conservative banner rallied against government operation of public services and privatized dozens of industries, including corrections, beginning in the 1980s (Daley Reference Daley1996; Nicholson-Crotty Reference Nicholson-Crotty2004; Price and Riccucci Reference Price and Riccucci2005; Schneider Reference Schneider1999).
The second expectation concerns an issue that private prison companies claim to ameliorate: government financial stress. States attempted to pass bonds to construct new prisons to stem prison overcrowding, but citizens repeatedly voted down these bonds or set controls on spending. Moreover, 57% of prison managers cited operational and construction cost savings as a reason why the facility privatized (McDonald et al. Reference McDonald, Fournier, Russell-Einhourn and Crawford1998). Despite these proclamations, however, private prisons are not consistently cheaper for the government to operate (Selman and Leighton Reference Selman and Leighton2010). Regardless of the financial reality, states often use financial stress as an express justification for privatizing prisons (even if those savings are unrealized or may not be the sole driver of this decision).
The final strand of thought from the literature explores the relationship between unions and privatization. Unions often oppose privatization on the grounds that it increases both costs and the potential for corruption and decreases accountability and job opportunities for union workers (Brudney et al. Reference Brudney, Fernandez, Ryu and Wright2005; Naff Reference Naff1991). Page (Reference Page2011) documents how the California Correctional Peace Officers Association (CCPOA), for instance, fought against private prison companies in California. Nationwide, bailiffs, correctional officers, and jailers have one of the highest rates of union membership, 47.9% in 2015, compared to an average public sector union membership rate of 35.2% (Hirsch and Macpherson Reference Hirsch and Macpherson2003). Corrections workers have a rate of union membership that ranks in the top twenty of nearly 500 occupations, after teachers, police officers, firefighters, and others (Hirsch and Macpherson Reference Hirsch and Macpherson2003). Therefore, the literature expects that states with lower rates of union membership among their corrections officers are more likely to privatize.
Though these three explanations dominate the existing literature on private prisons (e.g., Kim and Price Reference Kim and Price2014; Nicholson-Crotty Reference Nicholson-Crotty2004; Price and Riccucci Reference Price and Riccucci2005; Selman and Leighton Reference Selman and Leighton2010), there is fair reason to doubt that these mechanisms alone explain the choice to privatize. First, the partisanship explanation falls short, as Republican and Democratic states alike across the country currently have contracts with for-profit correctional companies, and platforms for all parties praised privatization at some point in the last three decades (see figure 3; Culp Reference Culp2005). More generally, this reflects the neoliberalism movement, a political shift that occurred in the United States in the twentieth century to prefer private operation of state responsibilities (Wacquant Reference Wacquant2009). Republicans and Democrats alike championed neoliberalism, thus exhibiting broad bipartisan support for not only prison privatization specifically, but also privatization of government services more generally (Gottschalk Reference Gottschalk2016). Furthermore, public support for private prisons does not neatly fall onto partisan lines—Republicans and Democrats are inconsistent in their approval of private prisons (Enns and Ramirez Reference Enns and Ramirez2018).
Second, though rhetoric surrounding private prisons frames the debate as one of saving money, there is inconsistent evidence on whether that claim is actually true (Selman and Leighton Reference Selman and Leighton2010). Moreover, though the 1980s in particular were a time of government austerity, public support for incarceration helped politicians acquire the necessary capital to build or renovate prisons (Enns Reference Enns2016; Harding Reference Harding2001). States could also circumvent public support for prison construction altogether via lease revenue bonds for that construction, a method that does not require voter approval (Gilmore Reference Gilmore2007). It is unlikely that fiscal stress is exclusively driving the growth of prison privatization, though it is often cited by policymakers as a primary motivation.
Third, it is unclear to what extent corrections officers’ unions have power over policy, save for the prototypical example of CCPOA in California (Page Reference Page2011). And even in that case, arguably the most powerful corrections officers’ union in the country did not stop private prisons from opening in California. Based on these cursory examinations, what, if at all, can we say about the common characteristics of states that adopt this policy? Next, I introduce a political explanation that deemphasizes the role of partisanship, fiscal stress, and unions, and instead emphasizes the growing political power of inmates and states’ desire to avoid accountability for prison conditions.
Inmate Political Activity & Prison Privatization
I argue that mounting pressure placed on the state bureaucracy from prisoners’ lawsuits convinced states to privatize their prisons. States seek to avoid legal and political accountability for poor prison conditions and transfer that responsibility to private companies, an important incentive currently overlooked in the literature. This emphasizes the importance of prisoner lawsuits, the frequency of which is a relatively modern phenomenon. The plight of prisoners was ignored for the first half of the twentieth century, as courts largely deferred to state governments in the administration of correctional facilities in an approach termed the “hands-off” doctrine (Feeley and Rubin Reference Feeley and Rubin2000). The federal courts soon stepped away from this doctrine in the wake of horrific reports of prison conditions and as activists linked the fate of prisoners to a broader struggle for rights in the United States (Brill Reference Brill2008). The Civil Rights Movement and the rights revolution had prisoners and other disadvantaged groups, like women and people of color, utilizing the judiciary to acquire rights previously denied to them (Epp Reference Epp1998; Rosenberg Reference Rosenberg2008). Activists from organizations like the National Association for the Advancement of Colored People (NAACP) Legal Defense Fund and the American Civil Liberties Union (ACLU) were crucial to the success of this litigation campaign as they linked the prisoners’ cause to that of other powerless groups, ensuring that inmates were part of a larger movement (Jacobs Reference Jacobs1980; Rosenberg Reference Rosenberg2008). Black activists in particular championed litigation as a political strategy, through the Nation of Islam and Black Panthers, often to win religious accommodations denied to them (Berger Reference Berger2014).
A wave of litigation hit the federal courts as prisoner lawsuits once dismissed by judges now received a fair hearing. In 1960, prisoners filed only 872 claims in federal court, just 2% of the total docket (Feeley and Rubin Reference Feeley and Rubin2000). That number soon exploded: by 1971, they filed 18% of all filings, more than 12,000 individual complaints (see figure 1; Feeley and Rubin Reference Feeley and Rubin2000). Though the majority of these claims were often dismissed quickly and legislation like the Prison Litigation Reform Act (PLRA) in 1996 significantly curtailed opportunities for inmates to sue (Schlanger Reference Schlanger2003), the federal judiciary now faced a mountain of litigation that they previously did not. This problem was confounded even further by a new and growing challenge: mass incarceration.
Prisoners’ lawsuits, filed and terminated, in each year from 1986 to 2016
Source: Data from the Federal Judicial Center.
The 1980s heralded a monumental shift in criminal justice in America. Prior to this, states largely relied on the rehabilitative approach to corrections, using indeterminate sentencing to personalize offenders’ sentences based on capacity for and evidence of rehabilitation (Gottschalk Reference Gottschalk2006). That policy was soon abolished and replaced with determinate sentencing and incarceration rates skyrocketed (see figure 2). Punitive laws at the state, federal, and local level criminalized drug crimes and increased mandatory minimum sentencing (Murakawa Reference Murakawa2014). This shift vastly expanded the reach and scope of the criminal justice system, as thousands of people, the majority of whom were Black or Latinx, were swept into prisons and jails (Alexander Reference Alexander2010). This remarkable shift in criminal justice found broad support across political and social lines: Republicans, Democrats, whites, Blacks, and the broader public supported the expansion of the carceral state, at least at the beginning of the 1980s (Beckett Reference Beckett1999; Enns Reference Enns2016; Fortner Reference Fortner2015; Murakawa Reference Murakawa2014; Smith Reference Smith2004).
Incarceration rate of prisoners in federal, state, and private facilities, 1980 to 2016
Notes: The ‘Private’ figure includes inmates in federal, state, and local facilities incarcerated in private correctional facilities.
Source: Data from the Bureau of Justice Statistics and the dataset on private prisons in the online appendix.
The challenge facing states in this decade was a complicated one: how to balance changing attitudes toward prisons with the constraint of outdated facilities too small to hold a burgeoning prison population. States could simply make do, house two to three inmates in a cell meant for one; release existing prisoners onto parole or probation; or construct new facilities (Enns Reference Enns2016; Feeley and Rubin Reference Feeley and Rubin2000; Taggart Reference Taggart1989; Vaughn Reference Vaughn1993). Beginning in the 1980s, however, states had an alluring new solution to the problem of overcrowded facilities: private prisons.
Prisoner Lawsuits and Private Prisons
Most states experienced at least one successful lawsuit that challenged prison conditions or prison overcrowding. By 1995, entire corrections systems in fifteen states were placed under court order, beginning with the first comprehensive court order of its kind in ArkansasFootnote 1 in 1970 (Feeley and Rubin Reference Feeley and Rubin2000). In other states, individual facilities were the only ones under court order. Eventually, forty-eight out of the fifty-three jurisdictions in the United States had at least one facility declared unconstitutional (Feeley and Rubin Reference Feeley and Rubin2000). Despite these successes, however, the vast majority of inmates lose their lawsuitsFootnote 2—estimates range from 88% to 98.4% (Ostrom, Hanson and Cheesman Reference Ostrom, Hanson and Cheesman2003; Schlanger Reference Schlanger2015).
I argue that states privatized in response to the unique pressures of inmate lawsuits, a choice undertaken by politicians of all parties subscribing to the overall philosophy of neoliberalism (Gottschalk Reference Gottschalk2016). States privatize to limit two types of accountability: political and legal. Whether it be overcrowding concerns, inadequacies in medical care, or other complaints, these lawsuits highlight the inadequacy of the existing prison system to accommodate the current prison population. From a theoretical perspective, prison privatization allows the state to shift accountability and blame away from the government to the private sector (Kay Reference Kay1987; White Reference White2001). This mechanism is similar to the debate about the use of private military contractors abroad and education privatization (Leander Reference Leander2010; Lipman and Haines Reference Lipman and Haines2007).
First, states are incentivized to privatize to evade political accountability. A growing number of inmate lawsuits brings public scorn and attention to poor conditions within prisons (Jacobs Reference Jacobs1980). Privatizing prisons, then, allows states to shift political accountability (and the negative media attention that comes with horrific prison conditions) to these private companies. Similarly, the appointment of a contract monitor or other government official, whose responsibility it is to oversee private prison operation and keep a close eye on any problems happening within the facilities (Selman and Leighton Reference Selman and Leighton2010), in fact helps lessen governmental accountability (Raher Reference Raher2010). Adding a layer of bureaucracy diffuses the blame for poor conditions within prisons. The state is even further removed from the day-to-day responsibility of facility operations with this additional layer of bureaucracy, placing the responsibility squarely on the private company to maintain appropriate prison conditions. It is thus more difficult, if not impossible, for voters to hold the government politically accountable for poor prison conditions, as there are multiple and complex layers of bureaucracy to contend with, and no clear attribution of responsibility for institutional failures.
Do citizens actually care about the problem of prison overcrowding enough to prompt state action? That is, public officials should only care that privatization diffuses blame if voters have seen, heard, or read about prison overcrowding in their state, and expect politicians to lead the effort to address that problem. Though public opinion on this issue is few and far between, one poll conducted by the National Center for State Courts Inter-Branch Relations Survey in 2009 is suggestive: 39% of respondents thought state legislators should respond to the problem of prison overcrowding; with 27% responding the governor should respond; and 22% placing responsibility for prison overcrowding problems on judges. About 84% of respondents also had “personally seen, heard or read about the problem of prison overcrowding in your state,” suggesting the wide scope of public knowledge about the issue (National Center for State Courts 2009).Footnote 3 Though comprehensive data does not exist on public opinion on prison overcrowding, this poll implies first, citizens demand accountability from multiple institutions and political actors for prison conditions and overcrowding and second, that they are knowledgeable about this issue.
Not only do people care about the issue of prison overcrowding, but in practice policymakers firmly attach blame for poor prison conditions within private facilities to the private companies. In the late 1990s, violent events occurred in Ohio at facilities of Corrections Corporation of America (CCA; now CoreCivic) and in New Mexico at the GEO Group. After these events, politicians went public with their concerns about private facilities, conducted investigations into their activity, and made it clear the private companies were the ones responsible for those violent altercations (McDonald and Patten Reference McDonald and Patten2003). In New Mexico, after GEO Group-operated facilities there experienced riots and murders of guards and inmates, the New Mexico Department of Corrections Commissioner was “reported in the press as saying that hard-core inmates take ‘special management,’ something that private prisons ‘are not really designed to do’” (McDonald and Patten Reference McDonald and Patten2003, xxiii), placing the blame squarely on GEO Group. Moreover, the New Mexico governor publicly warned GEO Group that if any more inmates were killed, he would order all private inmates transferred to public facilities, a warning repeated in the press under the headline “Private Prisons Warned” (Useem and Goldstone Reference Useem and Goldstone2002). By using the media to publicly name and shame these private companies, it all but guarantees the public attributes blame for these incidents to private companies. Moreover, privatizing allows policymakers to gain political capital from appearing as the “heroes,” that they are doing something about the failed criminal justice system and the evils of private prison companies (Schneider Reference Schneider1999). Indeed, modern demands to divest from private prison companies illustrates the blame placed on private companies for the operation of these facilities, and an absence of state responsibility for continuing contracts with them (Eisen Reference Eisen2018).
Second, there is the complex question of legal accountability. In public prisons, inmates can bring claims against corrections officers, wardens, or the state itself for unconstitutional conditions of confinement. When a state holds some of its inmates in private facilities, the question of who the inmate can sue is a broader question; a private corrections officerFootnote 4, the private company, a government monitor, or the government itself—or all four (Tartaglia Reference Tartaglia2014). The law surrounding who exactly is responsible for events within private prisons (and in other forms of government privatization) is far from settled, making an already opaque litigation system even more inaccessible to inmates (Gilmour and Jensen Reference Gilmour and Jensen1998; Raher Reference Raher2010). Though there is little evidence that private and public prisons experience judicial court orders at different rates (Burkhardt and Jones Reference Burkhardt and Jones2016; Makarios and Maahs Reference Makarios and Maahs2012), the question of whether private prisons experience less litigation overall (regardless of outcome) is unknown. Similar to questions about prison safety and quality (e.g., Burkhardt Reference Burkhardt2019; Perrone and Pratt Reference Perrone and Pratt2003), there is inconsistent evidence on inmates’ access to the courts across public and private institutions. Though I cannot provide an answer to whether inmates sue private prisons at lower rates, I point to incentives that might make it more difficult to sue in that context. Private prisons are not subject to public records laws, and it is thus difficult to gain access to information about what facilities to sue (Raher Reference Raher2010; Tartaglia Reference Tartaglia2014). Theoretically, private prison operators may have an incentive to provide lower quality services (including inmate legal access) in an effort to cut costs (Hart, Shleifer, and Vishny Reference Hart, Shleifer and Vishny1997). Therefore, I expect inmates to sue private prisons relatively less than public prisons.
Limiting state liability for privatization is evident in the construction of private prison contracts that contain indemnification clauses. As one example, a 2009 contract between CoreCivic and Nashville-Davidson County reads
The Contractor shall protect, defend, indemnify, save and hold harmless Metro, all Metro Departments, agencies, boards and commissions, its officers, agents, servants and employees, including volunteers, from and against any and all claims, demands, expenses and liability arising out of acts or omissions of the Contractor, its agents, servants, subcontractors and employees and any and all costs, expenses and attorney’s fees incurred as a result of any such claim, demand or cause of action. (In the Public Interest 2013, emphasis added)
These clauses are commonplace in both contracts and enabling legislation of privatization, specifically codifying that states are indemnified from legal action and only the private companies are responsible instead (Burkhardt and Jones Reference Burkhardt and Jones2016). Indeed, this is a strategy expressly mentioned by private prison firms: a 1986 annual report from CCA states that “the company assumes the primary exposure of legal liabilities, leaving the contracting agency in a secondary position. CCA reduces potential liability” (Corrections Corporation of America 1986).
Firm behavior is thus an important consideration here, as private prison companies market their services strategically to certain states, those states with rising legal liability for inmate lawsuits. This was a conscious marketing decision by these companies: a 1987 annual report from CCA reads, “the market segment with the most potential for the private sector is that portion which has the greatest need to relieve overcrowding, comply with court orders and operate with greater efficiency” (Corrections Corporation of America 1987). From CCA’s annual report in 1988: “in short, the additional contracts that have been awarded to CCA in the past year represent, in part, a lack of viable alternatives for government in a ‘must do’ environment” (Corrections Corporation of America 1988). CCA and other private prison firms recognized the need to limit legal liability for prison lawsuits as at least one of the reasons why states would want or need to privatize and targeted their business accordingly.
The limited liability also has financial and personnel benefitsFootnote 5 (Burkhardt and Jones Reference Burkhardt and Jones2016). Though there is no concrete source on the precise costs of inmate litigation, individual state estimates are significant: California, for example, spent over $200 million over fifteen years on legal fees and the costs of providing inmates with attorneys to sue the government (Associated Press 2013). Inmates at New York Rikers Island won $111.1 million in lawsuits over only a five-year period, including those dollars spent settling so-called “frivolous” lawsuits that would be more expensive to take to trial (UPI Top News 2013). One inmate in Wisconsin alone filed 117 lawsuits in the 1990s, costing the state $1.7 million (Wisconsin State Journal Reference Journal1998). Although this is not a significant proportion of state budgets, it nevertheless represents an unnecessary cost.
By privatizing, states therefore receive two potential significant benefits: it is more difficult to hold them responsible politically and legally for actions that happen within private prisons, and it makes the litigation process even more difficult for inmates to access, thus stemming the flow of litigation overall. This dynamic is driven by all lawsuits, not just lawsuits filed to protest overcrowding or those lawsuits that inmates win, because any lawsuit filed has the potential of revealing the poor state of the prison system. This argument is similar in flavor to others who argued that successful court orders promoted prison expansion and increases in spending on prison capacity, as state correctional agencies used successful lawsuits to demand higher budgets from legislators (Boylan and Mocan Reference Boylan and Mocan2014; Feeley and Rubin Reference Feeley and Rubin2000; Guetzkow and Schoon Reference Guetzkow and Schoon2015; Schoenfeld Reference Schoenfeld2010). This study, however, emphasizes the role of all lawsuits in this process, and not just those that inmates win. That is, both large and small cases, those that are won by prisoners and those that are not, are important drivers of the choice to privatize. This explanation does not suggest that prison privatization is the only avenue states took to minimize effects of lawsuits—namely, they also passed restrictive statutes to limit prisoner access to the courts (Brill Reference Brill2008)—nor is it the sole factor in the decision to privatize prisons, but suggests that it is an important overlooked variable and one that needs to be explicitly considered in studies of prison privatization. Of course, much like other explanations of the carceral state, this decision is heavily tinged by its racial implications, as historic for-profit prison labor targeted Black Americans, and most of those incarcerated in private prisons today are people of color (Hallett Reference Hallett2006). Finally, the argument here is about broad patterns of prison privatization across states; I acknowledge that state corrections systems are fragmented (e.g., Barker Reference Barker2009) and that states may react to inmate lawsuits differently. However, I expect the following hypothesis to be true in the aggregate.
Hypothesis: States in which more prisoner lawsuits are terminated, regardless of outcome, are more likely to privatize their prisons.
Data
I emphasize the important role inmate lawsuits play in a state's decision to privatize prisons, but to test this important overlooked incentive, I sought data on private prisons. This is an easier proposition in theory than in practice. For one, the federal government only began collecting data on private prisons in 1999 and no state keeps a comprehensive record of this information. Private prison companies were also not subject to Freedom of Information Act (FOIA) requests for most of the last three decades (Eisen Reference Eisen2018). Finally, though it might be desirable to measure the extent of privatization in prisons (i.e., how many services like laundry, food, or healthcare are operated by private companies), there is no source of this information for all states over my time period and, to my knowledge, is not something that states keep strict records of over time.
My dataset bridges this gap, providing the first opportunity for scholars to assess the growth of prison privatization at all levels—state, local, and federal—for the last three decades. I painstakingly read dozens of Securities and Exchange Commission (SEC) 10-K reports, the annual reports publicly traded companies are required to file. These reports contain information on the location of companies’ privately operated facilities and, for the most part, contain data on customers, design capacity, and contract length, which I then use to construct a longitudinal dataset on the growth of prison privatization (refer to the online appendix for more detail). For the purposes of this paper, I only include private prisons that house inmates under a state’s jurisdiction. I do not include federal private prisons (and immigration detention facilities) or local, city, or county private jails.
My sample includes facilities operated by four companies: CoreCivic, GEO Group, Correctional Services Corporation (CSC), and Cornell Companies.Footnote 6 The entire sample encompasses publicly operated private prisons from 1986 to the present, but the coverage differs across different firms. CoreCivic is included in the data from 1986 to the present, GEO Group from 1989 to the present, CSC from 1997 to 2004, and Cornell from 1996 to 2009. The GEO Group acquired CSC in 2005 and Cornell Companies in 2010, and both CCA and the GEO Group have acquired smaller companies over the last three decades.
An important caveat to this data source is that it only includes private prison companies publicly traded on the stock market. This is likely not a significant concern, however. The businesses that are included represent the vast majority of the private prison market in the United States. While there were once more than a dozen firms operating private correctional facilities (McDonald et al. Reference McDonald, Fournier, Russell-Einhourn and Crawford1998), that number has dropped dramatically. In 1998, for example, CoreCivic, GEO Group, CSC, and Cornell Companies together comprised more than 85% of the private prison market and as of 2014, GEO and CoreCivic alone comprised 85% of the market (Austin and Coventry Reference Austin and Coventry2001; Mumford, Schanzenbach, and Nunn Reference Mumford, Schanzenbach and Nunn2016). The third largest competitor, a privately-owned company called the Management and Training Corporation (MTC), comes in a distant third, with approximately 11% of the market (Mumford, Schanzenbach, and Nunn Reference Mumford, Schanzenbach and Nunn2016). This data substantially improves on the information currently available and helps us examine these facilities in finer-grained detail than before, across multiple decades and multiple states. Table 1 highlights the companies included in the dataset, how many unique facilities they operated from 1986 to 2016, and the sum of the capacities of those facilities. CoreCivic has operated the most, with 158 unique facilities and over 1.5 million prisoners housed from 1986 to 2016, with GEO Group second with approximately 900,000 prisoners housed over the same time period. Both Cornell and CSC operated fewer facilities for lower aggregate design capacities.
Original dataset of private prisons by company, 1986 to 2016
Note: This table includes all federal, state, and local correctional facilities operated by these companies.
These companies operate by entering into a contract with a governmental entity, with a variety of guarantees written in. For example, it is a source of much controversy that many private prison contracts contain occupancy guarantees, requiring facilities to remain 80% to 100% full or the state must pay for a certain number of prison beds, whether or not they are occupied (Eisen Reference Eisen2018; In the Public Interest 2013). These controversies aside, the states that contract with private prison companies allow these companies to manage a certain number of the state’s inmates. In some cases, as in Hawaii and Alaska, the state contracts with a private company to house inmates in private facilities out-of-state. On the other hand, states could require the private company to renovate or build a new facility within the state to house prisoners.
The following analysis considers one main dependent variable, Private Design Capacity, which is the capacity of private prisons that house inmates under the state’s jurisdiction each year. The geographic distribution of this variable is fairly diverse. Table 2 shows the first date of private prison adoption by state, from 1986 to 2016. Whereas some states opened a private prison within their borders and housed some of their inmates there (i.e., Arizona, California, or South Carolina), other states only housed inmates in private facilities outside their borders (i.e. Alabama held some of its inmates in private facilities in 2003, but only in Tallahatchie County Correctional Facility in Tutwiler, Mississippi). This highlights the variation in precisely how states privatized, whether by signing contracts with private prisons within their own state or signing contracts to house their inmates in private prisons out-of-state.
Original dataset of private prisons by state, 1986 to 2016
Note: First Private Prison refers to the first year the state opened its first private prison inside its borders. First Private Inmates refers to the first year the state housed any inmates in private facilities, inside or outside their borders.
Figure 3 displays the logged capacity of private prisons that house inmates under states’ jurisdictions over the last three decades. For the most part, once a state decides to house their inmates in private facilities, the government continues that policy. This is most obviously the case in states such as California, Arizona, Georgia, and Texas, all of which contract with private prison companies at least partially (and increasingly) throughout this time period. Other states like Wisconsin, Arkansas, and Nevada house inmates in private facilities at some point throughout this time period, but only do so temporarily, likely to alleviate short-term pressure in prisons. Finally, there are some states that never utilize private prisons: most of the Northeast and states like Nebraska and Missouri. Texas had the largest population of inmates in these private facilities between 1986 and 2016, at over 17,000 in any single year, while Hawaii had the highest proportion of inmates in private institutions relative to publicly-run ones, at over 70% in any given year (refer to the online appendix for more detail). The average state between 1986 and 2016 housed just over 4% of their inmates in private facilities.
Logged number of inmates held in private facilities, 1986 to 2016
Note: This variable reflects design capacity, or the capacity of private prisons that house inmates under the state’s jurisdiction over the last three decades.
Source: Data from the dataset on private prisons in the online appendix.
Armed with a new dataset on private prisons, I next construct a dataset of all the “Prisoner Petition” casesFootnote 7 filed in the federal courts from 1986 to 2016 from the Federal Judicial Center (FJC). The result is a dataset of 866,755 court cases filed by prisoners across all states from 1986 through 2016. Importantly, this dataset includes all lawsuits, whether victorious for prisoners or for the state.Footnote 8 I look at the federal courts since approximately two-thirds of all inmate litigation is filed there (Piehl and Schlanger Reference Piehl and Schlanger2004).
What Prompts a State to Privatize?
Comparing Litigation Theory and Privatization Theories
As a first cut, I analyze an ordinary least squares (OLS) model testing the relationship between prisoner lawsuits and private prisons, taking into account the other theories preeminent in the privatization literature.
$$ \begin{array}{c}PrivateDesignCapacit{y}_{i,t}={\alpha}_i+{\delta}_t+{\beta}_1SumLawsuit{s}_{i_{t-1}}+\\ {}{t}_{t-1}+{\beta}_xOtherTheorie{s}_i{}_{{}_{t-1},{t}_{t-1}}+{X}_{i_{t-1},{t}_{t-1}}+{\varepsilon}_{i,t}\end{array} $$
(1)
The outcome in equation 1 is private design capacity, which measures the capacity of private prisons that house state inmates. Companies only report the design capacity of their facilities and not the actual number of inmates located there. Though this may mean, for instance, a state private prison with a 1,200-inmate capacity may in reality house fewer inmates there, it is likely that states used the total capacity available to them since public prisons were massively overcrowded in this time period. This value does not include those in privately operated local jails or federal facilities, as I am expressly considering state decision making in privatizing prisons, and states do not need to consent to the private operation of local jails or federal facilities within their borders. Though this analysis (and those that follow) use private design capacity as the dependent variable, I test two additional dependent variables in tables A1 and A2 in the online appendix, the proportion of a state’s inmates in private facilities and the sum of private facilities with state inmates. I also investigate differential weighting, if a facility had multiple government customers.
The coefficient of interest is β 1, which identifies how the sum of all inmate litigation terminated in each state-year affects private design capacity. To assuage concerns about the potential of omitted variable bias regarding the most common explanations in the literature regarding privatization—partisanship, fiscal stress, and unionization—I include these variables in the equation as Other Theories. I include a dummy variable for Republican governor, a dummy variable for the presence of a Republican-controlled legislature (i.e., both chambers), and a final dummy variable—unified Republican government—for the interaction of these two. These values come from the National Conference on State Legislatures (NCSL) and the Book of the States. I also include budget gap per capita, from the Census Bureau, which represents the per capita difference between revenue and expenditures in any given state-year. Finally, I include a proxy for the number of unionized corrections officers. First, I use Page’s (2011) classification of which states had a corrections officers’ union as of 2011 (thirty-six states). Second, I use Hirsch and Macpherson’s Reference Hirsch and Macpherson2003 data on the nationwide percentage of unionized corrections officers measured annually from 1986 through 2016. I then multiply the national percentage of corrections officers who are unionized by the total number of corrections employees in each state-yearFootnote 9 before finally multiplying that number by the dummy variable of whether or not the state had a union in 2011. I divide the final measure by one thousand. Though this is a proxy, it provides a rough estimate of the number of unionized corrections officers, data that is not readily collected by the states.
The model also contains two control variables in X it−1,tt−1, violent crime rate, the number of violent crimes per 100,000 population from the Federal Bureau of Investigation, and incarceration rate, the number of prisoners in each state per 100,000 state population from the Bureau of Justice Statistics (BJS). These control variables help to mitigate concerns about additional omitted variable bias. Finally, αi and δt represent state and year fixed effects, and the errors are clustered by state.
Table 3 shows the results of equation 1. Column 1 estimates equation 1 without prisoner lawsuits, while column 2 includes all variables in the specification.
OLS model of level of prison privatization
*p < .1; **p < .05; ***p < .01
SE’s clustered by state.
The results highlight how broadly inconsequential the literature’s theories are at explaining the number of inmates privately incarcerated. Neither partisanship nor the budget gap is significantly related to the number of private inmates, and unionization is either barely positively significant or not significant, a result contra to the one expected by the literature. Though it is difficult to say why this is so, perhaps the reason is the potential weakness of these unions. Comprehensive studies of corrections officers’ unions have not been undertaken to my knowledge, and while the prototypical example is the CCPOA, the strength of that union may be an outlier in the context of the other state-level organizations.
The explanatory variable of interest, the sum of all prisoner lawsuits, is associated with a significantly positive effect on the number of private prison inmates (and the proportion in private facilities and the sum of state facilities; refer to tables A1 and A2 in the online appendix). Importantly, this result is significant at the 0.05 level, whereas none of the common explanationsFootnote 10 from the literature reach statistical significance. An increase of one additional inmate lawsuit in a state-year results in an increase of more than one inmate in a private facility, a magnitude that is consequential when considering that the average state faces around 500 of these lawsuits annually. The average state, then, would house more than 500 additional inmates in private facilities. Though the size of the significance of the incarceration rate is larger, that comports with the overall positive association between private prisons and inmate population (this variable loses its significance when the proportion of inmates in private facilities is the dependent variable). That the sum of prisoner lawsuits remains significant once the incarceration rate is accounted forFootnote 11 helps to bolster the theoretical perspective put forth in this paper.
Table 3 provides initial evidence for prisoner lawsuits spurring states to privatize. However, there is one potential issue with this analysis: endogeneity.
Instrumental Variables Estimation
Evaluating whether prisoners’ rights lawsuits caused a state to privatize part of their corrections systems is a difficult methodological task. Endogeneity likely exists, as prisoners’ lawsuits could lead to a higher degree of privatization within the state, or higher prison privatization could lower the number of prisoner lawsuits. Because the causal arrow is potentially bidirectional, estimating this relationship using OLS as in table 3 could lead to biased regression results.
To overcome this problem, I estimate an instrumental variables model, which uses an instrumental variable in place of the independent variable that only influences the dependent variable via the independent variable (Sovey and Green Reference Sovey and Green2011). A valid instrument is independent of other preexisting determinants of the dependent variable, prison privatization, and is a source of exogenous variation. The intuition is that the instrumental variable is plausibly exogenous and in theory randomly assigned, thus providing a less biased estimate of the effect of lawsuits on prison privatization than an OLS estimation would. The instrumental variables regression essentially replaces the independent variable, Sum Lawsuits, with a proxy variable that is plausibly exogenous and uncontaminated by error or unobserved factors that affect prison privatization (Sovey and Green Reference Sovey and Green2011).
I look to the district court caseload for this exogenous variation. Scholars often assume that district court judges are randomly given cases.Footnote 12 I use that exogeneity to my advantage in an instrumental variables framework. My instrumental variable for the independent variable, Sum of Lawsuits Terminated, is Weighted Cases Per Judge Serving, the weighted number of cases both active and senior judges hear in each state-yearFootnote 13 (Habel and Scott Reference Habel and Scott2014). This is a plausible instrument because one may expect judges who hear more cases each year to terminate more cases and vice versa, as an overburdened judge has an incentive to terminate cases quickly to clear her docket. This intuition expects judges to terminate more prisoner lawsuits as they hear more cases overall. However, it is important to ensure that the instrumental variable does not have any independent effect on prison privatization other than through the independent variable, the sum of all prisoners’ lawsuits (referred to as the exclusion restriction; Sovey and Green Reference Sovey and Green2011). This is likely satisfied, as it is unlikely varying numbers of cases prompt judges to alter a state’s corrections policy, as judges do not possess this policymaking power and judges cannot easily modify the number of cases they hear. Additionally, it is highly unlikely that states will modify the character of their prison systems due to the number of cases judges terminate in each year.
I estimate the following two-state least squares (TSLS; the instrumental variables method) equations to identify the effect of a higher number of lawsuits on prison privatization accounting for endogeneity:
$$ SumLawsuit{s}_{i_{t-1}}={\alpha}_c+{\delta}_t+{\beta}_3WeightedCasesPerJudgeServin{g}_{i_{t-1},{t}_{t-1}}+{\varepsilon}_{c_1,{t}_1} $$
(2)
$$ PrivateDesignCapacit{y}_{i,t}={\alpha}_c+{\delta}_t+{\beta}_4SumL\widehat{a}wsuit{s}_{i_{t-1},{t}_{t-1}}+{\varepsilon}_{c_2,{t}_2} $$
(3)
In equations 2 and 3, Private Design Capacityi,t reflects the sum of private prison inmates a state contracted with a company to manage. The instrumental variable is Weighted Cases Per Judge Serving
it−1,t
t−1 and the independent variable, as in equation 1, is Sum Lawsuits
it−1,t
t−1, the number of prisoners’ rights lawsuits terminated in each state-year. TSLS first regresses the sum of prisoner lawsuits on the weighted cases that each district court judge heard (equation 2) to get a fitted value of the sum of all lawsuits,
$ SumL\widehat{a}wsuit{s}_{i_{t-1},{t}_{t-1}} $
, which I then plug into the main equation of interest (equation 3) to estimate the effect of lawsuits on prison privatization. All explanatory variables, including the instrumental variable and the endogenous variable, are lagged by one year. So, Sum Lawsuits, and thus Weighted Cases Per Judge Serving, in one state-year are matched with Private Design Capacity in the following year, reflecting the time lag of the effect of court decisions.
I cluster by circuit to reflect the systematic differences between various circuits. I use circuit fixed effects and clustered standard errorsFootnote 14 rather than state, as in the earlier analysis, since nearly one-third of the states have no variation in the dependent variable. The control variables in equation 1 help to mitigate those concerns, but none are exogenousFootnote 15 and can be included in the TSLS estimation. A smaller number of circuits do not vary over time and while that is not ideal, it at least allows me to control for some geographic heterogeneity. Finally, αc is a vector of circuit intercepts, δ t is a vector of year intercepts, and ε c 1, t 1 and ε c 2, t 2 represent error terms.
The results from both the instrumental variables (IV) and OLS models for the dependent variable of the lagged number of inmates in private facilities are in table 4. The results corroborate the intuition behind Hypothesis 1—as the sum of all prisoners’ lawsuits increases, so too does the lagged number of inmates in private facilities. Encouragingly, the results from the OLS and IV estimations are fundamentally identical, highlighting that whether or not the estimation accounts for endogeneity does not substantively alter the findings presented here. Of note also is the F-statistic presented in the first stage IV results in column 2: Weight Per Judge Serving is a significant predictor of Sum Lawsuits, suggesting it is a strong instrument for the key independent variable of interest (Sovey and Green Reference Sovey and Green2011). Within a circuit, each additional lawsuit terminated in a year increases the number of private inmates by approximately 2, similar in magnitude to the OLS estimates in table 3. Moreover, table A17 in the online appendix uses the logged number of prisoner lawsuits as a robustness check and the positive and significant relationship remains, so this relationship does not appear to be driven by any outliers. And, indeed, since the vast majority of inmate lawsuits are not successful for inmates (Ostrom, Hanson, and Cheesman Reference Ostrom, Hanson and Cheesman2003; Schlanger Reference Schlanger2015), it is likely most of this effect is driven by the volume of these unsuccessful lawsuits rather than the few successful ones.
The effect of terminated prisoner lawsuits on lagged private design capacity
Notes: *p < .1; **p < .05; ***p < .01;
SEs clustered by state.
Substantively, what do these results mean? It is useful to consider two states’ experiences with privatization. Florida was one of the first states to privatize and began experimenting with private operation of correctional facilities in the 1980s. As the legislature and corrections department debated about prison privatization, the state mandated private vendors to be liable for the care and custody of inmates, and to indemnify the state against all legal liability (McDonald and Patten Reference McDonald and Patten2003). It was only after these provisions were in place that Florida privatized and indeed, this legal indemnification remains an important component of private prison contracts today (In the Public Interest 2013). This provides some qualitative evidence that Florida only privatized once the question of legal liability was settled.
Idaho contracted with CoreCivic to operate Idaho Correctional Center in 2000. For the next decade, the state knew that the facility was violating the contract and misrepresenting staff hours, but it was not until extensive litigation occurred that the state took action and took back state ownership of the facility (Tartaglia Reference Tartaglia2014). This is an example of the political accountability mechanism working in reverse: the state privatized initially to avoid the problems associated with prisoner litigation, but Idaho soon took back control of the facility, as privatization no longer helped the state avoid accountability for these lawsuits. Indeed, this suggests that state officials were fairly willing to ignore the problems litigation exposed in private prisons (one of the normatively troubling consequences of privatization) until conditions became so severe that the media and the public noticed.
Though concern about accountability was certainly not the only determining factor in either Florida or Idaho’s decision to privatize, it highlights the relevance of this accountability mechanism, and it is one explicitly considered by states in the decision to adopt or eliminate private prisons. These examples also help to provide context for the results in the tables, but they cannot tease out which mechanism is at play, whether states are avoiding legal or political accountability (or both). Future study of these mechanisms and case study analysis in particular may help to illuminate these considerations.
Discussion and Conclusion
I make an argument about a unique set of circumstances that contributed to the growth of prison privatization: a higher number of prisoner lawsuits results in higher numbers of private inmates. The oft-cited dynamics of partisanship, fiscal stress, or unionization do not solely explain carceral privatization, and I argue that the significant pressure of rising prisoners' lawsuits is an important contributor to prison privatization that needs to be explicitly considered.
These results are in line with other theoretical and quantitative work on the effects of prisoner litigation. Privatization is not the only potential response to growing litigation (though, for those incarcerated it may be one of the most consequential). For one, as Schoenfeld (Reference Schoenfeld2018) argues, successful prison litigation helped spur the growth of mass incarceration in Florida. That mechanism is similar to the one theorized here, in which prisoner lawsuits are able to influence state decision making about prison policy. Second, as Levitt (Reference Levitt1996) finds, states respond to prisoner litigation when the lawsuit is filed, and not only when the final decision is handed down. This intuition supports my argument, that states are responding with policy action to any pressure from the judicial branch, and even when faced with uncertainty over whether or not the state will be found at fault for poor prison conditions. Though I examine only state private prisons, my novel dataset could also be used to analyze the growth of federal private prisons and immigration detention facilities, and the degree to which federal or local privatization was motivated by similar incentives to reduce accountability.
We may be concerned that larger states with larger prison populations privatize at a higher rate, a decision driven by inmate population size and not litigation. There are a few reasons to cast doubt on this explanation, however. For one, states that use prison privatization at the highest rates (like Hawaii or New Mexico) are not the largest states, nor do they have the largest prison populations. This provides at least prima facie evidence that civilian or prison population size does not fully explain states' use of private prisons. Second and more importantly, the companies that are operating these prisons are actively considering the judiciary in their decisions to market—the 1986 annual report from CCA, for example, lists prison overcrowding as the major problem facing correctional facilities nationwide. This provides anecdotal reason to believe these companies are at least targeting the states facing the most litigation the most aggressively - and not simply those with the highest prisoner or civilian population. Similarly, states have similar patterns of prisoner litigiousness across my data: figure A4 in the online appendix shows that the number of lawsuits filed per inmate is fairly consistent across states. It is not the case, therefore, that some states facilitate or dampen prisoner lawsuits, but rather that this dynamic plays out similarly across states.
These findings have a few implications for both prison lawsuits overall and private prisons. First, the results cast doubt on the utility of prisoners filing as many lawsuits as possible to prompt procedural change within prisons. Even if prisons on the whole improved from successful litigation due to bureaucratization and fewer instances of physical brutality (Feeley and Rubin Reference Feeley and Rubin2000; Jacobs Reference Jacobs1980), there could still exist outcomes from litigation that are undesirable, like prison privatization. I point to the importance of organizations like the ACLU in helping to bring successful lawsuits against the state government. It is ironic that though the ACLU is heavily opposed to prison privatization and brings suits against the government for violations occurring within private correctional facilities, their activity could have inspired inmates to file lawsuits, thus making it more likely for a state to privatize. While courts can be an avenue for social change in prison policy, it is only by successful lawsuits and not necessarily by the thousands of other court cases those victories inspire. To what degree could these activists have foreseen this policy change and how, if at all, could they have altered their litigation strategy to reflect it?
The decision to privatize prisons is a vital one to understand because of the policy's significant and troubling normative concerns. A typical conceptualization of the state gives the government a monopoly on the use of force to keep citizens safe (Weber Reference Weber1965). It is unclear, then, how private prisons fit into this theoretical conceptualization. This difficulty is illustrated by variation across state governments—some passed statutes expressly forbidding privatization, while others found justification for this policy within existing laws (Quinlan, Thomas, and Gautreaux Reference Quinlan, Thomas and Gautreaux2004). Either way, normative and legal questions did and continue to swirl around the operation of these facilities. Some critics of private prisons are also concerned with quality differences across facility types, and it is theorized, though difficult to prove, that private prison operators cut corners to make more money and sacrifice inmate care for profit (Dolovich Reference Dolovich2005; Hart, Shleifer, and Vishny Reference Hart, Shleifer and Vishny1997). The evidence on this is mixed, with some studies finding that public correctional facilities are safer, more cost effective, and better managed, with others finding the opposite, that private facilities perform better on these metrics (Burkhardt Reference Burkhardt2019; Perrone and Pratt Reference Perrone and Pratt2003). Despite this inconsistency, state-sponsored reports of similar flavor find private facilities had a higher level of safety and security incidents (U.S. Department of Justice 2016). These troubling normative considerations make it vital for policymakers and scholars alike to understand how and why states turn to prison privatization.
Prisoners are often cited in the rights revolution as primary benefactors of the movement to imbue vulnerable populations with individual rights (Epp Reference Epp1998). Namely, as the early leaders of the prison litigation movement were Black activists involved in the Nation of Islam or Black Panthers, it begs the question of how Black political activism may have led to more Black Americans incarcerated in private prisons (Hallett Reference Hallett2006). If it is the case that efforts to bring about the rights revolution also brought forth policies like privatization that may be antithetical to that mission, how does this change the scholarly evaluation of that movement and its successes? When we analyze the outcomes of this revolution, should we consider downstream effects, like private prisons, that the founders of this movement not only did not intend, but did not want? To be clear, I do not endorse efforts to limit the flow and character of prison litigation, but I suggest that these normatively positive efforts may lead to a variety of undesirable and unanticipated outcomes.
Finally, this paper has implications beyond prison privatization. The desire to privatize any kind of government service in an effort to displace blame and accountability for government operations is a characteristic of other privatization efforts, including private military corporations and the move to charter schools in education (Leander Reference Leander2010; Lipman and Haines Reference Lipman and Haines2007). While accountability in core government functions like prisons, the military, or education may seem more normatively troubling than the privatization of water and sewer services or toll roads, the incentives to do so travel across policy type: privatization as a means to evade political and legal accountability for policy failure.
It is safe to say the ACLU, one of the organizations at the forefront of fighting private immigration facilities, could not have predicted how their involvement skewed the carceral landscape so much. Rather, if opponents of this policy seek to prevent future prison privatization, I point to the importance of inter-institutional dynamics between the executive and judicial branches, along with the alteration of state governments’ incentives to privatize in response to temporary problems within prisons and jails. Without addressing these dynamics, it is likely that these companies will enjoy the favorable position they currently hold under the Trump administration and within states that have grown to depend on private prisons. This is certainly the case in Hawaii, which housed about one-third of their inmates in private prisons in 2016. Hawaii’s auditor summarized their experiences in a report on the status of their contracts with private prison companies: “What started as a temporary solution to relieve prison overcrowding is today a matter of state policy” (Hawaii State Auditor Reference Auditor2010).
Supplemental Materials
Appendix to OLS Model
Data Appendix
Instrumental Variables Appendix
To view supplementary material for this article, please visit http://doi.org/10.1017/S1537592720003485.
