* * *
The Puzzle of Prisons?
The moral so far is that the whole empirical literature on public and private prisons is inconclusive. As I noted in Monday’s post, this should be somewhat of a puzzle for activists on both sides who claim that privatization should turn prisons into either humanitarian disaster zones or models of quality and efficiency.
Of course, that the empirical literature is inconclusive doesn’t mean the sectors are equivalent; it means that current methods haven’t been good enough to detect the difference. A methodologically deficient literature could hide evidence of either good or bad quality. But if the differences are great enough, you’d think they might show through even with bad methods.
The tentative conclusion I draw from the literature, though, is that there may be modest, but not huge, quality differences between the sectors; the public sector is better on some dimensions and worse on others, and there’s no good evidence that either sector does better at reducing recidivism. And while the private sector is probably cheaper, it remains to be seen whether the cost savings is on the order of 15% (respectable) or on the order of 3% (somewhat negligible).
But this puzzle largely disappears when we consider the institutional environment of private prisons. In many areas, the private sector has been good at delivering better results at a lower cost. This is because private producers are accountable to customers who care about the quality of the end product, and because they have the flexibility to change how they do things in response to problems they may encounter. Neither of these conditions is true for private prisons—not even slightly, not even as a first approximation.
I have noted above that there is limited evidence of private firm innovation. But this is because private prisons are highly constrained in how they operate. Private prison contracts essentially “‘governmentalize’ the private sector,” reproducing public prison regulations in the private contract. Privatization can come to resemble an exercise in who can better pretend to be a public prison.
For instance, back in 1985, Robert Levinson complained of a contract with the Eckerd Foundation for the management of the Okeechobee School for Boys in which “[v]irtually every” contract item
concerned input activities and pertained to administrative/operational functions. Thus, Eckerd could have been in total compliance with all contractual provisions even if every released client committed a new offense on the first day in the community. Moreover, at no point in the contract were the criteria for noncompliance stated nor its consequences specified.
More recently, in Arizona, an auditor general report stated,
The Department requires that private prisons mirror state-operated facilities, and performs extensive oversight activities to ensure that its contractors meet its requirements. In order to maintain uniform standards for state and private prisons, the Department requires contractors to follow Department Orders, Director’s Instructions, Technical Manuals, Institution Orders, and Post Orders. These requirements extend to specific details, such as following the same daily menus as state-operated facilities. Contractors may request waivers from the Department for policies that are not applicable to private prisons, such as state fiscal management practices, employee evaluations, and employee benefits.
The same daily menus! In Tennessee, “it even appears that private sector innovation was deliberately thwarted by making the private sector provider . . . abide by [state Department of Corrections] policy” in running the facility.
Subjecting private contractors to public regulations is actually quite common; one exception to this trend is Florida, where public and private prisons are controlled by different agencies, and the agency that regulates private prisons tries to balance “setting policy and encouraging innovation.” More generally, input specification in private-prison contracts is routine, though of course the level of inputs specified can (and should) be “output-driven” in the sense that it’s “related to output objectives.” For instance, one can find liquidated damages provisions for certain input-based breaches like not complying with the state’s policies or not filling certain required positions.
If inputs and procedures are highly regulated, it’s not surprising that the evidence for private-sector improvements isn’t overwhelming. The market is a discovery process; one shouldn’t expect different methods to emerge unless innovation is permitted.
And not only permitted: one shouldn’t expect different methods to emerge unless the incentives favor it. If the premise of privatization is that incentives work, particularly given the greater flexibility of private industry, micromanaging inputs and failing to incorporate the full range of desirable outcomes into the contract price means giving up on much of the possible benefit of privatization.
But the efforts to measure performance in various areas of government from the Job Training Partnership Act of 1982 and the Government Performance and Results Act of 1993—and the limited efforts to make funding contingent on those performance measures—have largely passed prisons by.
Outcome measures aren’t totally absent. Contracts do include a limited range of outcome measures—for instance, limited penalties for escapes. But by and large, outcome-based compensation is rare. And to the extent there are outcome-based rewards or penalties, Charles Thomas argues, “the amounts involved commonly have little or no correlation with the true magnitude of what independent contractors accomplished or failed to accomplish,” and “the dollar value of the reward or sanction is often too trivial to encourage superior performance or to deter defective performance.” (Of course this isn’t always true: the state of Ohio recently fined CCA nearly $500,000 for contract violations found during audits, and many of these violations were performance-relevant.) Even developing outcome measures hasn’t been a high priority.
In 1998—not that long ago—Douglas McDonald and his coauthors identified two exceptional cases of performance-based compensation: the “Bureau of Prisons’ contract with Wackenhut for the operation of the Taft Correctional Institution in California,” which allowed for “an award-fee incentive worth up to 5 percent of paid invoices,” and a District of Columbia contract with CCA for the Correctional Treatment Facility, “which permit[ted] financial rewards for meeting targets based on performance indicators.”
Florida recently would have taken a good step in this direction, if the bill in question hadn’t been defeated. The bill would have required that private prison contracts make provision for measuring a number of dimensions of performance (though note that some of these are output measures): number of batteries, number of major disciplinary reports, percentage of negative random drug tests, number of escapes, percentage of inmates in “a facility that provides at least one of the inmate’s primary program needs,” and so on. The number of escapes also showed up in a more specific way: the contractor would have been required to reimburse the state for the costs of escapes. The Florida bill also listed required various performance measures for work release centers. (I discuss various other performance measures later.)
The following sections develop these themes and discuss two distinct benefits of using performance measures. The first set of advantages of using performance measures, discussed in the next section, is a pure accountability advantage: we, as citizens and policymakers, would know how well our prisons are doing; we’d be better informed in deciding which sector to choose, either systemwide or on discrete projects; and we could think more clearly about what prisons should be doing. The second type of advantage, discussed in the section after that, goes more to harnessing incentives to improve the system over time: incorporating performance measures into contracts, and tying providers’ compensation to how well they do, would give providers a reason to care about quality and simultaneously let us grant them greater flexibility. And tomorrow’s post will discuss the normative issues involved in choosing the actual measures.
Accountability, Neutrality, and Goal Setting
1. To Know What Works
We all want to improve prisons. But forget about that for a moment. Even before any of these improvements were possible, performance measures would have the obvious effect of allowing us to measure performance. This would be a great step forward in researchers’ ability to conduct quality studies. We would have a better sense of which sector provides better quality; combine that with better cost studies that take into account the pitfalls described above, and we’d be better able to decide whether to be one of the nineteen states that (as of 2011) don’t have private prisons. If we do decide not to use private prisons, performance measures would help us determine which public prisons performed badly and where to look for improvement.
2. To Implement Competitive Neutrality
Suppose we decide not to use private prisons. Should we then contract out the entire prison system? Probably not: someone has to be able to run a facility if the current contractor has fallen down on the job or gone bankrupt, and given how concentrated the private prison industry currently is, it may not always be realistic to count on being able to easily bring in a competitor when this happens.
How much of the system, then, should we privatize? The standard way to proceed is to choose particular prisons to privatize and put them up to bid to private firms, or to contract with private firms to use their own prisons. A more beneficial approach, though, would be to have a regime of “competitive neutrality,” where the public and private sector compete on the same projects. The best system may be one of mixed public and private management, where private programs “complement existing public programs rather than replace them.” (Health care reformers’ advocacy of the “public option” in health insurance was premised on a similar idea: that public participation can make competition more fair by disciplining private providers more than they would discipline each other.)
For instance, Gary Mohr, director of the Ohio Department of Rehabilitation and Correction, has talked about creating a “culture of competition” in corrections. Ohio has pursued a combination of outsourcing and insourcing: some public prisons have been sold or their management has been contracted out to the private sector, while one private prison has been taken in-house. The result, according to Mohr, is that one can “ratchet up the best practices that can be created from both the public sector and multiple private vendors.”
But for this sort of system to work, we have to be able to fairly compare private-sector and public-sector bids before the fact. The cross-fertilization that’s supposed to result from competitive neutrality depends on flexibility, otherwise both sectors will try to do the same thing. But, without performance measures, flexibility undermines the ability to do the comparative analysis of bids that’s necessary to successfully implement cross-fertilization; the most straightforward way of making efficiency comparisons without performance measures is to mandate that the private sector replicate every public-sector procedure, down to the tiniest detail. And indeed, this is what Mohr did when contracting out the management of the North Central Correctional Complex facility to the private sector or when selling the Lake Erie Correctional Institution.
But with performance measures—and with an understanding of how proposed programs and methods translate into performance—he would have been able to take different proposals, translate them into expected performance, and thus have a basis for comparison, even if the proposals were radically dissimilar. (The beliefs about expected performance would then have to be verified by evaluating the winning contractor’s performance after the fact.)
In particular, recall the problems involved in figuring out the public sector’s true costs: the same problems can make for unfair competitions if public providers’ bids don’t include the costs they bear that are paid for by other departments, their different tax treatment, and the like. So it’s not surprising that such a regime is rare in the United States.
One of the advantages of competitive neutrality is that—as in Ohio—prisons can be both outsourced and insourced at different times, depending on who wins the contract, so particular prisons can “churn” between the public and private sectors. The result, according to Richard Harding, would be a “process of positive cross-fertilization,” where best practices migrate from one sector to another. “[T]he opening up of the private sector,” Harding writes, “may heighten awareness of how sloppy public accountability has often been in the past, leading to the creation of innovative mechanisms applicable to both the private and the public sectors.” In fact, Harding argues, systemic improvement has been one of the best consequences of privatization, so narrowly focusing on which sector is better in a static sense is almost beside the point.
3. To Express What We Want
Measuring performance would do more than just let us know which sector is better and promote cross-fertilization by facilitating a competitive neutrality regime. On an even higher level, it would encourage governments to better conceptualize what makes for a good prison—an exercise that’s long overdue.
Jon Vagg, for instance, argues that, in the U.K., private prisons “were a key factor in persuading the administration that standards were necessary, if only for the purpose of monitoring contractual compliance.” And that example isn’t just a fluke. Prisons have been operating for centuries, and yet it was the experience of privatization that spurred the development of performance measures, as private-prison critics made arguments that privatization harmed quality and private-prison advocates made arguments to the contrary. Now that performance measures exist, one can use them to evaluate both the private and the public sectors, to the benefit of both.
For Performance-Based Contracting
With performance measures, we can go further than just knowing how good public and private prisons are, implementing competitive neutrality, and formulating the proper goals of the prison system—important as all that is. We can also incorporate the performance measures into contracts and make compensation contingent on performance, finally giving prison providers strong incentives to deliver high quality.
1. Limited Current Efforts
Performance-based compensation is being implemented in the United States to a very limited extent. As noted above, 5% of the contract price at the Bureau of Prisons’ Taft facility was performance-based. Taft was a demonstration project, which should give one a sense of how new this enterprise is.
The U.K. is now on the forefront of performance-based compensation, which it calls “[p]ayment-by-outcome” or “payment-by-results.” The idea was floated in a 2008 Conservative Party Green Paper and, once the Conservative Party came into power, it was developed in a 2010 Green Paper from the Ministry of Justice. Payment-by-results is being introduced in three prisons: two private prisons, Peterborough and Doncaster, and a public prison, Leeds, though the plan is to extend the model to all prisons by 2015. The measure is the twelve-month reconviction rate, compared to a matched comparison group. At Peterborough, performance-based “[p]ayments start when the reconviction rate of the intervention group is 7.5% less than that of the matched comparison group, with increasing returns up to a maximum rate of 13%.” “The Peterborough pilot is the first in the world where private investors have assumed financial risk for reducing re-offending.” In addition to having access to a range of prison programs to prevent recidivism, offenders at Doncaster are assigned case managers to support them during their sentence and after release, offering advice and help on employment, housing, and benefits issues. (Earlier experience with payment-by-results was “primarily limited to the welfare to work market[,] where success [was] varied and limited.”)
A parallel program focused on finding jobs for offenders, called Job Deal, compensates providers based on employment rates. Compensation is 70% fixed and 30% conditional—a third of the conditional payment is for an output measure, “successfully enrolling offenders” in the program; another third is for “a combination of outputs and processes” such as “helping clients open bank accounts”; and another third is “for achieving ‘hard outcomes.’” Note, though, that even these “hard outcomes” are softer than they might seem, because they include finding a job but also include “enrolling in further learning.” Some additional payment-by-results programs have also been proposed by the government or by the Social Market Foundation, focusing either on reoffending rates or on other outcomes or outputs like “drug use cessation or employment.”
2. The Range of Possible Contracts
a. General Considerations
These examples suggest how performance-based contracts could be structured. The contract could provide that the contract price is not just the usual flat per-diem per prisoner, but an incentive payment that—as a simple example—could vary (positively) with how many inmates find jobs or (negatively) with how many inmates are rearrested within two years.
Outcome measurements may not always be available for all dimensions of quality, so some measurement of inputs may continue to be necessary. But as far as possible, the ideal should be to make compensation contingent not on inputs like guard training, or even on outputs like the number of GEDs granted or the number of rehabilitative programs offered or ACA accreditation, but primarily on actual outcomes like the extent of unconstitutional conditions or how well prisoners are actually rehabilitated or how many prisoners get jobs.
The amount of the bonus can be a flat fee, or it could be more complicated—in the case of recidivism bonuses, the bonus could be inmate-specific, depending on “the probability and social cost of recidivism for each inmate”—or it could even be determined by competitive bidding. It’s often charged that private prisons have little incentive to invest in rehabilitation, and in fact have an incentive to try to increase recidivism, so that they can get (at least some of) the same inmates back later; if this is so, the bonuses should be at least high enough to counteract this incentive so rehabilitating inmates is affirmatively attractive to prison firms.
Though I focus here on monetary rewards and penalties, there are other possibilities. High performance could, instead of increasing a firm’s compensation in the individual contract, merely confer a reputational benefit, increasing its probability of winning future bids. One could give out certificates or “even simply publiciz[e] league tables of recidivism performance.” Or one could reward good performers by giving them more flexibility in future contracts.
b. Rewards or Penalties
Going back to monetary incentives, one can choose between penalties for bad performance and rewards for good performance—or one could have both—though the difference needn’t be that important.
Consider a “rewards” contract that offers a $1 per diem reward for each unit of quality on a hypothetical 0-to-10 scale, so the potential reward is $0 to $10. Suppose Acme Corrections Corp. expects to achieve a quality level of 5 at a total cost of $35 per diem. Then it would be willing to submit a bid of $30 or above for the project; it would just cover its costs with the $30 payment plus the $5 reward. (Recall that prison bids are bids on how much money the contractor will get from the government; a $30 per diem winning bid means that the contractor will be paid $30 per inmate-day.) Suppose bidding is competitive, other firms have similar technology, and Acme is the most efficient firm; then Acme wins the auction with its $30 bid. (A less efficient firm, say one that would require $36 per diem to achieve quality level 5, wouldn’t bid below $31, so Acme, as a more efficient firm, would be automatically rewarded up front for its higher quality by having a better chance of winning the auction. The bids don’t tell us the true social cost, the true cost to the government, or the true quality—that requires waiting for the actual realized level of quality, which determines the level of the reward—but they do signal which firm is (or believes that it is) more efficient.)
Now consider an alternative “penalties” contract that offers a $1 penalty for each unit of quality below 10 (i.e., 7 units of quality lead to a $3 penalty). This contract has equivalent incentive effects to the previous one: a provider will invest in a unit of quality as long as its cost of doing so is under $1. Therefore, these incentives, as before, make Acme expect to achieve the same quality level of 5, which we have seen carries a total cost of $35 per diem. Now Acme is willing to submit a bid of $40 or above for the project; it would just cover its cost with the $40 payment minus the $5 penalty. Again, with the competitive bidding assumptions listed above, Acme wins the auction with its $40 bid.
So even though the contracts look different, they have essentially identical incentives, and any superficial differences between them are, roughly speaking, ironed out in the bidding process. The provider’s degree of risk aversion doesn’t change the result. The government can offer contracts with penalties, but then it will pay more to the winning bidder; or it can offer contracts with rewards, and the winning bidder will be satisfied with less. (One difference might be in the timing of the payments: if the base price is paid up front while rewards or penalties are processed some time later, the first contract is somewhat less valuable than the second because its payments are more delayed.)
c. Controlling for Baselines
In the same way, it probably doesn’t make a huge difference whether the compensation takes into account the baseline level of quality.
Controlling for baselines is a huge issue in the literature on performance measures. For instance, an early paper on performance measures, by Gloria Grizzle and coauthors, discussed methodological issues regarding what makes for a good performance measure. A large part of the discussion focused on doing the proper econometric modeling to figure out the causal factors behind a performance measure. Figuring out these causal factors is important for at least two reasons (beyond merely understanding the process). One is to have a sense of what input or output measures to use if the outcome measures aren’t available in a given case. Another is to be able to properly assign credit, so providers who get a bad (or good) population of inmates aren’t blamed (or praised) for bad (or good) results.
Similarly, Gerald Gaes and his coauthors argue that “social scientists should push ultimate outcomes as far as they can be pushed,” but that, in light of the other factors that affect recidivism, “[i]t is also desirable to have more direct measures of intermediate changes to human behavior that precede desistance, and that may be influenced by criminal justice interventions.” They don’t directly list desirable performance measures—they give an example of performance measures for the specific element of “Prison Security Performance,” though they stress that one should do a similar exercise for other elements of prison performance such as health care. The main characteristic of their approach is its emphasis on adequately modeling prison performance in terms of individual-level and institutional-level independent variables so that one can properly attribute credit where credit is due, avoid blaming prisons for factors beyond their control like the characteristics of the inmates, and figure out what inputs are actually important in producing prison performance. For instance, for health care, rather than measure (or in addition to measuring) the prevalence of a disease in the prison, which indicates the potential for transmission, it would be useful to use the number of cases in the incoming population as a baseline, and measure the number of new cases.
Is all this necessary? Let’s do our numerical example again: Consider the rewards contract discussed above, with a $1 per diem reward for every unit of quality on a 0-to-10 scale; the winning bidder, who expected to deliver quality level 5 at a cost of $35, would have won the contract with a bid of $30. Now consider a rewards contract that controls for the baseline level of quality; suppose the expected level of quality for this prison is 4, so a quality level of 5 would yield a reward of $1.
The only effect of the quality adjustment is to reduce reward payments by $4. A bidder who was willing to bid $30 on the unadjusted contract would be willing to bid $34 on the adjusted contract, to take into account the $4 reduction in the expected reward. Either way, the payoff is the same to the contractor—and the price is the same to the government. The government saves $4 on reward payments but pays it all out again in the base contract price that emerges from the auction. Jeremy Bentham argued against controlling for baselines two centuries ago:
I would make [the contractor] pay so much for every one that died, without troubling myself whether any care of his could have kept the man alive. To be sure he would make me pay for this in the contract; but as I should receive it from him afterwards, what it cost me in the long run would be no great matter. . . .. . . [Under this system,] you need not doubt of his fondness of these his adopted children; of whom whosoever may chance while under his wing to depart this vale of tears, will be sure to leave one sincere mourner at least . . . .
To be sure, the bidder has to have a way to figure out that the expected level of quality is 4. This requires two things. First, the bidder should have a belief about the proper model to predict the baseline quality level; different bidders can have competing beliefs about reality that lead them to different predictions. Second, it needs to have enough information about the population of inmates to plug into its model. Where either of these is absent, the contractor won’t know how much to bid—this might lead to excessive payments from the taxpayer’s point of view or insufficient payments from the contractor’s point of view—but the incentive effects will remain the same.
So while adjusting for the baseline is relevant for various reasons—it allows one to more accurately assign praise or blame, rank different facilities, and so on—it doesn’t seem absolutely necessary for a compensation scheme to provide the proper incentives for improvement.
Moreover, risk aversion makes a difference here, but not in the way one would expect. Controlling for baselines might even increase risk, depending on the uncertainty in the calculation of the baseline.
If the contractor gets too little, there is the concern that it might not be able to fund the project and might go bankrupt within the contractual term. But this is the same concern that happens with all bidding. Whether or not we adjust the payment for the baseline, the winning bid under a low-bid system will be subject to the “winner’s curse.” As a simple example, consider many firms with identical technology. They each have slightly different models for predicting how profitable a prison will be, and firms with higher predictions will submit lower bids. At most one of these models is correct; everyone else’s model is incorrect to some degree. The lowest bid will thus come from the bidder who makes the most wildly incorrect overestimate of his profits. Sophisticated bidders adjust their bids to take the winner’s curse into account, but the winning bidder might either be unsophisticated or end up not having adjusted his bid enough. So the threat of contractors who go bankrupt—or of contractors who bid low and then try and hold the government up for more money—is real. But, again, this happens regardless of whether we adjust for baselines. The solution is instead to require performance bonds, to rely on a track record of past performance (and restrict complete newcomers to small projects until they’ve proven themselves), or otherwise to try to weed out financially unsophisticated or untrustworthy parties.
d. Discrete vs. Continuous Measures
Note that, in the preceding example, the contract price varied continuously with the level of quality. Another possibility would have been to use a binary compensation scheme, where the reward or penalty is contingent on whether one reaches a particular target. This could look like “Get a fixed reward only if you achieve less than 50% recidivism.”
These binary schemes, while easier to implement, are problematic in several ways. Providers who don’t expect to be able to reach anywhere near the target have little incentive to try to achieve anything at all. Providers who do expect to be able to reach the target quite comfortably have little incentive to try to achieve anything additional. Providers who may or may not be able to reach the target are subjected to more risk than they would bear under a continuous scheme. Perhaps a large corporation might act somewhat risk-neutrally, so risk won’t matter; but smaller firms or nonprofits may refrain from bidding, or may require more money to take the project, or may be reluctant to try high-expected-value but risky strategies.
(Of course, one could also imagine intermediate reward schemes: for example, the reward could be almost flat for any level of recidivism above 50% and increase rapidly at or below 50%, for instance, “Get a reward of $0.01 for every percentage-point reduction of recidivism below 100% and down to 50%, and then a reward of $1.00 for every percentage-point reduction beyond 50%.” British performance contracts, where payments don’t start until the decrease in recidivism is 5% or 7.5%, and where payments are capped once the decrease is high enough, fit this mold. At this point I won’t do anything more than signal the existence of such contracts, though the optimal slope of the compensation scheme is something I’ll return to below when I discuss risk allocation.)
The same is true of penalties that may occur during the contractual term. Governments can terminate their contracts—this is a form of binary scheme—though this is a rare remedy that tends to be reserved for the most extreme abuses. Providing for graduated financial penalties for abuses of different severity is probably a better solution than merely providing for contract rescission, because draconian penalties are less likely to be used. Not that termination isn’t appropriate in extreme cases—governments should always retain the ability to take over a prison if a contract is terminated. The need to retain a credible threat of termination is one reason to prefer that governments, not prison firms, own the prisons, since government ownership of the physical facility reduces termination costs.
3. The Feasibility of Merit Pay in the Public Sector
Note, also, that while I’ve been primarily concentrating on incentives for private firms, there’s no inherent reason why performance-based compensation can’t also be considered for public prison wardens—consider the example of Leeds noted above—especially if we simultaneously pursue competitive neutrality. As John Donahue says, “the fundamental distinction is between competitive output-based relationships and noncompetitive input-based relationships rather than between profit-seekers and civil servants per se.” Proposals to reward public servants for high performance aren’t rare, and merit-based compensation in the public sector has increased in recent years, but it’s still hard to find in corrections.
Researchers differ on how feasible merit pay is in the public sector; I won’t resolve the argument here, except to note that the Government Performance and Results Act of 1993 has a procedure by which agencies can make “proposals to waive administrative procedural requirements and controls, including specification of personnel staffing levels, limitations on compensation or remuneration, and prohibitions or restrictions on funding transfers . . . in return for specific individual or organization accountability to achieve a performance goal.” Any such proposal, according to the statute, must “describe the anticipated effects on performance resulting from greater managerial or organizational flexibility, discretion, and authority, and . . . quantify the expected improvements in performance resulting from any waiver,” “precisely express the monetary change in compensation or remuneration amounts, such as bonuses or awards, that shall result from meeting, exceeding, or failing to meet performance goals,” and be “endorsed by the agency that established the requirement.” Just reading the statutory language—and this is a statute that purports to encourage flexibility—doesn’t exactly give one confidence that public-sector flexibility is easy to come by, at least in the federal system.
At the very least, though, to the extent performance-based compensation is a good idea in the private sector, it may well also be a good idea in the public sector. How feasible that is is a question of the relevant state or federal law.
In tomorrow’s post, I’ll discuss what measures to choose, and introduce some of the concerns and critiques of the performance measures concept.