The Role of the Cost-of-Crime Literature in Bridging the Gap

RESEARCH ARTICLE

ROLE OF THE COST-OF-CRIME

LITERATURE

The Role of the Cost-of-Crime Literature in

Bridging the Gap Between Social Science

Research and Policy Making

Potentials and Limitations

Patricio Dom

ınguez

Steven Raphael

University of California, Berkeley

Research Summary

In this article, we review the theoretical paradigm underlying cost–beneﬁt analysis and

address some of the critiques of this framework that have arisen within criminal justice

circles and other policy areas. We also review existing studies devoted to estimating the

costs of speciﬁc crimes. We offer a brief discussion categorizing the alternative costs of

crime and the various methodological approaches taken (hedonic analysis, contingent

valuation, and accounting methods), with an explicit discussion of the relative strengths

and weaknesses of each approach and debates within the economics profession pertaining

to these methodologies. We argue that cost considerations broadly deﬁned should be of

central importance in criminal justice policy debates. However, we also highlight the

potential for cost-consideration and important equity criteria to come into conﬂict.

Policy Implications

Policy makers should consider careful cost–beneﬁt analysis as an important criterion in

criminal justice policy. Given some features of criminal justice policy choices such as the

unequal distribution of the costs of criminal victimization, anti-crime enforcement,

We thank Daniel Baker, Daniel Nagin, Robert MacCoun, and Jesse Rothstein for helpful discussion and

feedback on this project. Direct correspondence to Patricio Dom

ınguez, Goldman School of Public Policy,

University of California, Berkeley, 2607 Hearst Avenue, Berkeley, CA 94720-7320 (e-mail: pdominguez@

berkeley.edu).

DOI:10.1111/1745-9133.12148

C

2015 American Society of Criminology 589

Criminology & Public Policy

Volume 14

Issue 4

Research Article Role of the Cost-of-Crime Literature

and the potential for perceived illegitimacy of the criminal justice system to undermine

various public institutions, we argue that equity considerations also deserve careful

attention. In practice, we place greater conﬁdence in the use of cost-of-crime estimates

to judge the relative effectiveness of alternative interventions, and we are cautious

regarding policy prescriptions emanating from beneﬁt–cost ratios that are marginally

greater than one.

uring the past three decades, considerable research effort has been devoted to

articulating and measuring the various pathways through which crime and our

public and private responses to crime impact overall the well-being of society. An

important substrata of this research has focused on placing a monetary value on the social

cost of criminal victimization. These efforts often have aimed to generate separate ﬁgures

for different crime types with the goal of generating a common metric against which the

relative severity of alternative criminal incidents can be judged.

Estimates of the costs of criminal victimization play an important role in bridging

the gap between basic social science research on criminal justice interventions and policy

analysis. By monetizing and rendering commensurable crimes of varying severity, these

estimates facilitate an analysis of the degree to which the beneﬁts of a given proposal exceed

the costs broadly deﬁned. Equally as important, these estimates facilitate a comparative

analysis of alternative policy tools, permitting estimates of the “bang-per-buck” across

policy inter ventions and assessments of the efﬁciency of a given pattern of deployment of

public resources.

Of course, there are limitations to this conceptual framework for public choice and, by

extension, to the utility of the costs-of-crime literature in informing and guiding criminal

justice policy. To start, cost–beneﬁt analysis and cost-effectiveness analysis provide a speciﬁc

weighting (or social accounting) of the relative welfare of alternative groups in society that

often conﬂicts with widely held beliefs regarding fairness and equity. This is a direct result

of the use of money as the common metric used to place various beneﬁts and costs on a

common footing. Responsible analysis requires cognizance of this fact and a careful parallel

analysis of the equity implications of policy alternatives.

There are also important qualiﬁcations regarding the measurement of the costs of

crime and suspicion that current methods overestimate these costs. For example, pain-and-

suffering estimates based on observed jury awards reﬂect judgments made by jurors that

have no consequences for the jurors themselves. Damage assessments are imposed on a third

party; the jury effectively spends someone else’s money. Similar concerns are often raised

regarding estimates of willingness to pay to avoid victimization via surveys querying about

hypothetical referenda. By design, nothing is at stake for the respondent in answering a

hypothetical question, although careful contingent valuation studies have made great efforts

to articulate and make salient the t ax consequences of the scenario under consideration.

590 Criminology & Public Policy

Dom

ınguez and Raphael

Both examples illustrate how cost-of-crime estimates are often based on assessments that

are divorced from personal budget considerations.

In what follows, we argue that the cost–beneﬁt framework and economic approach

to public choice should play an increasingly important role in criminal justice policy and

that the body of research estimating speciﬁc costs of crime should ﬁgure prominently in

policy analysis and public choice in the criminal justice domain. Such analysis is already

quite important and inﬂuential in other policy domains, in which risk of injury and lives

lost are inﬂuenced by policy choices, such as occupational safety and health, environmental

regulation, and transportation safety. The increasing emphasis on cost–beneﬁt analyses and

the search for policy tools with an “evidence base” brings criminal justice policy into this

fold.

However, we do not support the proposition that policy choices should strictly adhere

to and be determined solely by the results of cost–beneﬁt calculations. Cost–beneﬁt anal-

ysis alone often fails to consider adequately the equity implications of policy alternatives.

Equity is an especially important criterion in criminal justice policy, and perhaps it is a

more important criterion relative to other policy domains. The burdens of criminal justice

enforcement are disproportionately borne by the poor and the marginalized, and often they

can exacerbate existing societal divides, such as inter-racial economic disparities or the rela-

tive deprivation of individuals with severe mental illness. Moreover, an unequally distributed

burden of the impact of criminal justice enforcement may delegitimize law enforcement,

the criminal justice system more broadly, and government in general in speciﬁc commu-

nities. Inadequate consideration of equity, fairness, proportionality, procedural justice, and

disparate impact may ultimately serve to undermine and offset the value of crime-control

beneﬁts achieved through speciﬁc policy strategies.

Moreover, we have more conﬁdence in the ability of cost-of-crime estimates to gauge

the relative severity of alternative crime types than we do in the absolute magnitudes

of the individual cost point estimates. Having a gauge of relative severity that renders

alternative crime types commensurable greatly facilitates “cost-effectiveness” analysis of the

relative efﬁcacy of alternative uses of public funds. Irrespective of whether cost estimates for

speciﬁc crimes are too high or low, a solid and reliable relative ordering permits the ranking

of policy alternatives that have differential effects on different crime types. Conclusions

regarding whether a speciﬁc policy has a cost–beneﬁt ratio in excess of one, however,

requires conﬁdence in the precision and unbiasedness of the costs-of-crime estimates.

We begin by discussing the theoretical paradigm underlying cost–beneﬁt analysis and

address some of the critiques of this framework that have arisen within criminal justice

circles and other policy areas. We argue that cost considerations broadly deﬁned are, in the

language of Adler and Posner (1999), “morally relevant” and certainly an important criteria

1. Similarly, this also requires confidence in the precision and unbiasedness of the underlying estimates of

the intervention on crime.

Volume 14

Issue 4 591

Research Article Role of the Cost-of-Crime Literature

to be considered along with equity consequences and perhaps more intangible deontological

ethical concerns. We present the commonly offered defense of cost–beneﬁt analysis that

all policy choices with net positive beneﬁts are potentially win–win for all involved and

then discuss the practical impediments to redistribution in the criminal justice context that

render the equity criteria particularly important.

Next, we review existing studies devoted to estimating the cost of speciﬁc crimes. We

begin with a brief discussion categorizing the alternative costs of crime and t he various

methodological approaches taken (hedonic analysis, contingent valuation, and accounting

methods), with an explicit discussion of the relative strengths and weaknesses of each

approach and debates within the economics profession pertaining to these methodologies.

We close with a discussion of the use of these ﬁgures in policy analysis and their potential to

render social science research results more useful and perhaps actionable to policy makers

and public deliberation.

Cost–Benet Analysis, Public Choice, and Criminal Justice Policy

Anderson (1999, 2012) and Cohen (2005) presented comprehensive estimates of the dollar

value of the social cost of crime. The most recent analysis by Anderson (2012) placed the

total external cost of crime in 2012 at roughly $3.2 trillion. By external cost, we are referring

to the cost generated for crime victims, for potential victims, and for taxpayers. To arrive

at social cost, one would have to subtract from this total any cost for which there is an

offsetting gain (a pure transfer in the language of welfare economics). Losses from theft and

fraud clearly represent such a transfer as the gain to the offender offsets the loss to the victim.

Excluding the value of property loss to victims yields a social cost of $1.6 trillion (Anderson,

2012). Regardless of one’s position regarding whether criminal offenders have standing in

these cost calculations, both aggregate ﬁgures are large, amounting to 19.5% and 9.7%

(respectively) of the country’s gross national product in 2012. Behind these aggregate totals

are $646 billion in crime-control expenditures including $113 billion on police protection;

$81 billion in correctional expenditures; $76 billion in expenditures by various federal

agencies; $84 billion to combat drug t rafﬁcking; and more than $300 billion in private

expenditures on security devices and services, repairs, insurance, and private advocacy (for

example, Mothers Against Drunk Drivers). Victim costs in the form of lost life and injury

exceed $750 billion.

These ﬁgures illustrate the fact that much is at stake in criminal justice policy choices.

Interventions, whether they are increases in incarceration, expanding police forces, dropout

prevention efforts, or some other policy proposal that reduces crime rates stand to generate

substantial beneﬁts for society to the extent that on net they chip away at these colossal cost

totals. However, is positive net beneﬁts a sufﬁcient or even necessary criterion for choosing

a particular policy path?

Cost–beneﬁt analysis provides at least partial guidance on this question. First, cost–

beneﬁt analysis articulates and places dollar values on the consequences of a policy choice.

592 Criminology & Public Policy

Dom

ınguez and Raphael

Second, cost–beneﬁt analysis provides a clear set of criteria by which society and policy

makers can rank alternative policy choices and pass judgment on speciﬁc proposals. More

generally, this set of criteria also provides guidance regarding the appropriate level of gov-

ernment intervention, or alternatively stated—the optimal level of provision of a public

good.

In this section, we begin by providing a theoretically precise deﬁnition of the costs and

beneﬁts of policy changes and the microeconomic concepts that link speciﬁc harms and

beneﬁts to dollar values. This discussion illustrates the theoretical concepts that the cost-of-

crime literature has so far aimed to operationalize empirically. We devote particular attention

to how interpersonal variation in income levels gives rise to interpersonal variations in the

monetary valuation of victimization. Next, we discuss the ranking of policy alternatives

by cost–beneﬁt and cost-effectiveness analysis and the implicit social preferences associated

with these rankings. Finally, we discuss some of the general ethical objections to cost–

beneﬁt analysis and issues that have been raised speciﬁcally within the criminal justice

policy context.

Articulating Policy Outcomes and Monetizing Harms and Beneﬁts

A cost–beneﬁt analysis of a speciﬁc policy proposal involves identifying all members of

society that are harmed, that beneﬁt, and the key outcomes impacted by a proposal that

ultimately impact the welfare of individuals. For example, a proposal to enhance sentences

for drunk driving would negatively impact the welfare of drunk drivers who would be

incarcerated as a result, positively impact t he welfare of sober drivers and their passengers via

enhanced safety, and negatively impact taxpayers through higher correctional expenditures.

To be sure, this enumeration of consequences is limited to some degree by the imagination

and our ability to anticipate secondary impacts of and behavioral responses to the policy. The

policy world is rife with examples of the unintended, unanticipated, and often socially costly

consequences of policy initiatives. For example, if increased prison time for those convicted

of driving under the inﬂuence has an unintended criminogenic impact on postrelease

2. Criminal justice and safety services have the classic characteristics of goods that are underprovided by

the private market or are provided at suboptimal levels. For example, to some degree, the benefits

derived from police patrols are nonrival (our benefit from the general deterrent effect of more police on

the street does not detract from the comparable benefit that you would derive) and nonexcludable

(one cannot exclude a given city resident who refuses to pay his or her taxes from realizing the benefit

from greater general deterrence). As a result, individual citizens have a strong incentive to “free ride” on

the services provided by others, and only those with great demand for security services will purchase

them. As another example, some privately purchased security services can generate negative

externalities. One neighbor’s purchase of a home security system could increase the likelihood that the

next-door neighbor is robbed. An economic analysis of welfare is motivated in large part by an attempt

to identify policy choices that bring society closer to optimal allocations of public resources, taking into

account the underprovision of public goods and the suboptimal overprovision of the private market of

goods that generate negative externalities.

Volume 14

Issue 4 593

Research Article Role of the Cost-of-Crime Literature

behavior, then the higher postrelease offending levels would detract from the net beneﬁts

of this policy option.

Once the likely consequences of a policy change are articulated, cost–beneﬁt analysis

renders consequences commensurable by placing a dollar value on individual beneﬁts

and harms. Microeconomic theory offers two alternative yet related theoretical concepts

for assessing the dollar value of harms and beneﬁts experienced at the individual level.

Compensating variation is deﬁned as the income transfer needed to restore an individual’s

well-being to what it would be in the absence of a speciﬁc policy change. For policies that

provide a beneﬁt, the compensating variation measures how much one would be willing to

pay for the beneﬁt in question. For a policy that harms someone, compensating variation

is the minimum amount that one would be willing to accept in compensation. Equivalent

variation, in contrast, measures the amount of money that would have the equivalent impact

on one’s happiness level as the consequences of a policy change. Fo r a policy change that has

a negative effect, equivalent variation measures one’s maximum willingness to pay to avoid

the consequences (think ransom), whereas for a policy that generates a beneﬁt, equivalent

variation measures the minimum amount of cash one would be willing to accept to forgo

the beneﬁt.

Table 1 conceptually illustrates these ideas in two criminal justice contexts:

1. The calculation of appropriate compensation for an identiﬁed crime victim

2. The measurement of the individual beneﬁts from a crime-mitigating public investment

Regarding compensation for crime victims, compensating variation equals the mini-

mum income payment needed to restore the person’s original level of happiness (i.e., the

victim’s minimum willingness to accept). Equivalent variation equals the amount that the

individual would have been willing to pay to avoid being victimized (i.e., the maximum

willingness to pay). Clearly, compensating variation in this instance will exceed equivalent

variation, as the latter calculation is limited by the individual’s income whereas the former

is not. In addition, both measures of victim cost are likely to be higher for high-income

victims. This is most obvious for the equivalent variation measure, where victimization cost

3. In addition, the quality of the analysis is contingent, to some degree, on our understanding of the likely

behavioral responses to a policy change. For example, if people are deterred by tough punishments for

drunk drivers, then cost calculations would have to account for such a behavioral response.

4. The difference between compensating and equivalent variation is subtle and often is a source of

confusion. Because compensating variation measures the income adjustment needed to restore one’s

utility to what it was, the calculation relies on the individual’s happiness level in the absence of the

policy and the state of the world under the policy (i.e., the prices and incomes that result with the policy

change) to calculate the necessary compensation. Equivalent variation, in contrast, measures the

change in income that would have the equivalent effect on happiness of introducing the policy. In

other words, the reference happiness level is that which would result if the policy were enacted while

combined with prices and incomes in the absence of the policy to calculate the comparable income

change.

594 Criminology & Public Policy

Dom

ınguez and Raphael

TAB LE 1

Conceptual Examples of the Calculation and Denition of Compensating and

Equivalent Variation in Criminal Justice Contexts

Calculation Based on Calculation Based on

Harm or Benet Being Quantied Compensating Variation Equivalent Variation

Calculating appropriate compensation

for an identied crime victim.

The minimum amount of money that

the crime victim would be willing

to accept after the fact as

compensation for the damages,

pain, and suering caused by the

victimization.

The maximum amount of money that

an individual crime victim would

have been willing to pay before

the fact not to have experienced

the damages, pain, and suering

caused by the victimization.

Calculating individual benets from a

public investment in crime

prevention.

The maximum amount that an

individual would be willing to pay

for the policy in question.

The minimum amount of money that

an individual would be willing to

accept in lieu of the benets

generated by the

crime-prevention policy.

is limited from above by the income of the person in question. However, even the com-

pensating variation deﬁnition, which is conceptually not limited by an individual’s income

level, is likely to generate a higher assessment of victim damages for high-income victims

for reasons we will discuss shortly.

Both measures are t heoretically correct gauges of the damages done. However, one might

argue that for the purposes of calculating damages and attempting to make the victim whole

(likely an impossible goal for victims of serious crimes), compensating variation is the more

appropriate of the two. The compensating variation effectively endows victims with the

right to their level of well-being in the status quo ante. Such an allocation of property rights

is defensible on ethical grounds (and perhaps in terms of the deterrence value of damage

assessment) as crime victims by deﬁnition are unjustiﬁably harmed by a third party.

In the second example in Table 1, compensating and equivalent variation swap places

in terms of willingness to accept and willingness to pay. Speciﬁcally, compensating variation

measures the beneﬁts to an individual of a crime-mitigating public investment in terms of

the maximum amount that the individual is willing to pay for these beneﬁts. Equivalent

variation, in contrast, equals the minimum amount the individual would be willing to

accept to forego the beneﬁts of the policy change. Again, the willingness-to-pay valuation

of the policy change will be less than the minimum willingness to accept, as the former is

constrained by income levels and the latter is not.

For the purposes of deciding whether to spend public funds on a particular policy,

one might argue that willingness to pay provides the most appropriate of the two measures

for several reasons. First, willingness to pay is the more conservative of the two, and some

might argue for a generally high level of judiciousness and parsimony when it comes to the

Volume 14

Issue 4 595

Research Article Role of the Cost-of-Crime Literature

stewarding of taxpayer dollars. Second, the aggregate valuation of a policy option derived

from cumulating compensating variation over all individuals in society is perhaps better

aligned with the calculation that would ideally be made by a policy maker with a ﬁduciary

responsibility to constituents; namely, whether willingness to pay for the beneﬁts created by

a speciﬁc policy (essentially the summary measure regarding how society values the policy)

exceeds policy c osts.

These cost concepts measured at the individual level are key inputs to cost–beneﬁt

analysis, with the aggregation of costs and beneﬁts providing the key statistics for ranking

policy options. In essence, these individual valuations of costs and beneﬁts provide the

weights placed on the welfare of various individuals in assessing the merits of speciﬁc policy

choices. With this in mind, it is important to explore how these standard measures of

costs and beneﬁts vary with individual income levels, and by extension how cost–beneﬁt

analysis weighs the relative welfare of the poor and the wealthy. We have already noted that

willingness-to-pay valuations are limited by individual incomes. The fact that maximum

ability to pay is higher for higher income individuals is likely to generate higher monetary

valuations for similar beneﬁts among those with high incomes. However, it is also likely

that personal safety is what economists refer to as a normal good, that is, a good for which

demand increases with income. Hence, in addition to a greater ability to pay for safety, high

incomes should theoretically be associated with a higher willingness to pay for safety. This

positive relationship between income and beneﬁt and/or cost valuation ultimately results in

greater weight being placed on the welfare of the well-to-do in cost–beneﬁt calculations.

An alternative pathway through which these monetary gauges of welfare may favor the

wealthy operates through the relationship between individual well-being and income levels.

Speciﬁcally, if the marginal increment to well-being associated with an additional dollar of

income decreases as income increases, then the wealthy will place higher dollar valuations on

victimization than the poor, holding constant the characteristics of the criminal incident.

This point is best illustrated with an example. Suppose two people, one poor and one

wealthy, both of whom own the same model cell phone, are independently robbed of

their phones while leaving their respective workplaces. Suppose that both individuals suffer

similar levels of psychological distress as a result of the robbery.

Suppose further that the

marginal effect on well-being of an additional dollar of income is higher for the poor person

relative to the wealthy person.

5. In this example, we are implicitly making what is referred to in microeconomic theory as an

interpersonal utility comparison. By stating that both victims suffer comparable psychological harm, we

are implicitly assuming that our underlying gauge of well-being can be used to distinguish different

happiness levels cardinally and compare changes in well-being across individuals. This idea is

somewhat controversial in economics, and many are reluctant to draw such comparisons. In fact, most

of the results of basic microeconomic theory can be derived with a nonunique utility function so long as

the ordinal ranking of alternative states is preserved.

596 Criminology & Public Policy

Dom

ınguez and Raphael

The value of damages done in this example consists of two components: the cost of

replacing the phone and the minimum compensation for pain and suffering needed to

restore the victim’s preincident level of well-being. By construction, the monetary value of

the ﬁrst component is equal, and hence we can focus on putting a dollar value on pain

and suffering. Compensation for pain and suffering theoretically equals the amount of

income that would be sufﬁcient to offset the welfare loss associated with being traumatized.

Although both individuals suffer similar traumas, the minimum necessary payment to the

poor victim is smaller because his or her “marginal utility of money” is higher and each

dollar goes further in improving the poor victim’s welfare relative to that of the wealthy

victim.

In other words, given his or her relative income deprivation, it takes fewer dollars to

offset the utility loss from psychological distress. Hence, the dollar value of the victimization

under these assumptions is lower for the poor victim.

As these welfare measures are essentially the building blocks of cost–beneﬁt analysis,

the systematic tendency to place greater weight on the welfare of the wealthy is certainly of

concern. Of course, we still must discuss how theses valuations map into the absolute and

relative ranking of policy options and carefully ponder whether such rankings may conﬂict

with social preferences for equity and fairness.

Cost–Beneﬁt Analysis and the Absolute and Relative Assessment of Policy Options

Individual valuations of the net beneﬁts of a policy ultimately provide key inputs into

cost–beneﬁt and cost-effectiveness analyses, and by extension, they provide one manner of

prioritizing the uses of the powers of the state. Cost–beneﬁt analysis essentially sums the

monetary values of beneﬁts and costs across all individuals and normatively recommends

pursuing policies that generate net positive beneﬁts (or, equivalently, that have cost–beneﬁt

ratios greater than one). Cost-effectiveness analysis, in contrast, ranks policies according to

their cost–beneﬁt ratios. Although these two analytical t ools are similar and interdependent,

one might conceptualize cost–beneﬁt analysis as providing an incremental test for whether

more or less public resources should be devoted to a speciﬁc policy effort. Cost-effectiveness

analysis, however, provides a manner of assessing whether the resources spent within a given

policy domain are being used wisely (i.e., whether we could achieve the same outcomes

with fewer dollars or better outcomes with similar resources by simply reallocating resources

across alternative policy interventions).

To illustrate these concepts, Figure 1 reproduces cost–beneﬁt ratio estimates produced

by the Washington State Institute for Public Policy pertaining to interventions intended

to reduce juvenile offending.

The cost–beneﬁt ratios vary greatly across interventions

with high ratios for Functional Family Therapy and various cognitive behavioral therapy

6. See Friedman (2002) for a more formal discussion of this point.

7. See the explanation on cost–benefit results from the Washington State Institute for Public Policy (2015).

Volume 14

Issue 4 597

Research Article Role of the Cost-of-Crime Literature

FIGURE 1

Cost–Benet Ratios from Juvenile Justice Interventions Estimated by the Washington State Institute for Public Policy

Source:WSIPP,Beneﬁt-Cost Results. Retrieved July 10, 2015 from wsipp.wa.gov/BeneﬁtCost.

598 Criminology & Public Policy

Dom

ınguez and Raphael

interventions to ratios below one and even negative for programs such as Scared Straight.

Cost–beneﬁt analysis would prescribe investing in all of the programs in Figure 1 with a

cost–beneﬁt ratio in excess of one. Cost-effectiveness analysis, however, would prescribe

that for a given level of expenditures, we allocate resources to the interventions with the

highest cost–beneﬁt ratios, perhaps involving reallocating expenditures from interventions

with relatively low cost–beneﬁt ratios to interventions with high ratios.

In the context of the programmatic interventions illustrated in Figure 1, the recom-

mendations of cost–beneﬁt and cost-effectiveness analysis seem perfectly reasonable and

frankly uncontroversial. Why would society not discard programs that generate negative

net beneﬁts and seek to spend public dollars in a manner that maximizes the impact per dol-

lar spent? In many instances, we would argue that such analysis indeed should provide the

principal criterion for public choice. However, the peculiar relationship between individual

valuations of policy beneﬁts and costs and individual income levels, in conjunction with

strict adherence to the normative prescriptions of cost–beneﬁt analysis, may yield policy

recommendations that conﬂict with commonly held preferences for equity and fairness.

This issue is best illustrated with examples. Suppose a police department receives

funding for one additional police ofﬁcer and is trying to decide how to deploy this new

resource. For simplicity, suppose the city is socioeconomically segregated into a wealthy area

and a poor area and that the additional ofﬁcer, through general deterrence, would prevent

10 residential burglaries per month in whichever area he or she is deployed. With higher

incomes, lower average marginal utility of money, and more expensive possessions, high-

income households will place greater monetary value on preventing 10 residential burglaries

than will low-income households even if we assume that the psychological trauma associated

with being burglarized is the same for a poor and a wealthy household. Here, an analysis of

the cost-effectiveness of alternative deployment choices would yield the recommendation

that the additional ofﬁcer be deployed to the wealthy neighborhoods.

Alternatively, suppose that the material possessions of poor households are sufﬁciently

modest that the beneﬁts as measured by compensating variation associated with the 10

prevented burglaries per month falls short of the cost of an additional ofﬁcer. Suppose also

that the material possessions of the wealthy households are sufﬁciently opulent such that

the beneﬁts from preventing one burglary per month would exceed the cost of an additional

ofﬁcer. Now suppose that allocating the ofﬁcer to the wealthy neighborhood would prevent

only one burglary (because crime rates are already low in the wealthy neighborhoods)

but 10 burglaries in the poor neighborhood. A cost–beneﬁt test would not recommend

additional police for the poor neighborhood but would for the wealthy neighborhood.

8. Presumably, each of these interventions is likely to exhibit diminishing returns to scale, implying that

reallocating dollars from intervention A toward intervention B would increase the cost–benefit ratio of A

and diminish that of B. The optimal allocation of resources across interventions would occur where the

cost–benefit ratios are equalized across interventions.

Volume 14

Issue 4 599

Research Article Role of the Cost-of-Crime Literature

A cost-effectiveness analysis would recommend reallocating police ofﬁcers from the poor

neighborhood to the wealthy neighborhood.

As one ﬁnal example, consider the analysis of the ﬁnances of a drug-selling gang from

the south side of Chicago presented by Levitt and Venkatesh (2000). By analyzing variation

in compensation for the gang’s foot soldiers associated with periods of high and low mor tality

risk, the authors estimated the compensation required among these individuals to accept an

elevated risk of death on the job, and by extension, the value that these (for the most part)

young African American men were implicitly placing on lives lost. The authors estimated

that the additional annual compensation for one additional death ranged from $8,000 to

$120,000. We will discuss in detail the concept of the value of statistical life in the next

section. For the moment, however, we will note that observed risk-wage trade-offs among

the general working public yields an average estimate of $9 million. If we use this estimate in

conjunction with the upper bound estimate for foot soldiers in the drug-selling gang, then

cost–beneﬁt analysis would rank as equally attractive a policy that prevents one death among

the general public to a policy that prevents 75 deaths among young men who participate

in the drug trade. Cost-effectiveness analysis would rank a policy that prevents one death

among the general public as preferred to any policy that prevents 74 or fewer deaths among

young men involved in the drug trade.

Examples of policy proposals that pass the cost–beneﬁt test but have questionable

equity implications abound in many policy domains. For example, redistributive programs

that provide aid to the poor often fail to generate net positive beneﬁts because the value of

the transfer to the poor falls short of the tax revenue needed to ﬁnance the beneﬁts and the

administrative costs of the program. A strict adherence to the normative recommendations

of cost–beneﬁt analysis would preclude such redistributive efforts. As an alternative example,

consider the issue of generating tax revenue. The net social costs of raising tax revenue via

an excise tax are lower when levied on goods for which the price sensitivity of demand tends

to be low.

Hence, cost-effectiveness analysis would imply that we should always levy taxes

on goods for which demand exhibits very low price sensitivity, such as cigarettes or insulin.

The potential for cost–beneﬁt analysis to yield normative prescriptions that conﬂict

with a common sense of fairness is well known. In fact, we believe that most advocates of

cost–beneﬁt analysis would argue for parallel assessments of the equity and distributional

consequences of policy options and the consideration of equity as well as efﬁciency in policy

9. In a simple static market analysis, imposing an excise tax always generates negative net benefits

because of the reaction of the supply-and-demand sides of the market to the change in price. Generally

speaking, the more elastic the demand for a good, the greater the distortion to individual behavior

caused by an excise tax. An excise tax (a) increases the prices paid by consumers and reduces the prices

received by producers; (b) reduces production and consumption, relative to what it would be in the

absence of the tax; and (c) generates a “deadweight loss,” which is defined as a loss of value for which

there is no offsetting gain. Deadweight loss will be greater in markets with more elastic demand

holding all else constant.

600 Criminology & Public Policy

Dom

ınguez and Raphael

deliberation and public choice. Nonetheless, there is a strong case to make for cost–beneﬁt

analysis as a principal input for policy making, equity concerns notwithstanding, for the

following reasons.

First, any policy choice with net positive monetary beneﬁts provides what economists

call a potential Pareto improvement. A Pareto improvement is a situation in which a change

that makes someone better off does so without harming anyone else, inclusive of changes

where everyone beneﬁts. Pareto improvements represent movements toward efﬁciency in

that one generates more happiness within the resource constraints of the society without

generating a loss. A policy proposal that would generate net positive beneﬁts is one that

has the potential to be a Pareto improvement if a mechanism exists to redistribute some of

the net beneﬁts to those who would otherwise be the policy losers. Of course, this point

is qualiﬁed by the empirical reality that, for whatever reason, there are few examples of

cross-party compensation and many examples where policy changes create uncompensated

losers. Our hypothetical examples regarding the allocation of additional police resources

are instances in which redistributions converting potential Pareto improvements into actual

Pareto improvement are unlikely to occur. Nonetheless, with positive net beneﬁts, there is

always room for a deal, so to speak, between those who beneﬁt and those who are harmed

by a policy change that in the end may make all better off.

Second, the prescriptions of cost–beneﬁt analysis need not conﬂict with equity objec-

tives. For many interventions, the costs and beneﬁts of policy alternatives will be borne by

similar socioeconomic segments of society, and thus, the relationship between net-beneﬁt

valuation and income is less relevant (take for example, the earlier discussion of juvenile

justice interventions). Moreover, it is certainly possible that the prescriptions of cost–beneﬁt

analysis are aligned, rather than are in conﬂict, with key equity objectives. One example

is the Justice Reinvestment Initiative (JRI) of the Bureau of Justice Assistance within the

U.S. Department of Justice. The aim of JRI is to evaluate the costs and beneﬁts of various

criminal justice interventions broadly deﬁned, to include policing, prisons, jails, commu-

nity corrections, and violence-prevention efforts, and to foster allocation of public resources

toward more cost-effective policies. Suppose that it can be shown that the value of additional

crime resulting from reverse deterrence and incapacitation associated with the elimination

of a mandatory minimum prison sentence falls short of the value of crime that would be

prevented should these resources be reallocated toward hiring more police ofﬁcers. Making

such a reallocation would reduce the cost of crime to society holding constant public expen-

diture levels, while mitigating the collateral consequences of incarceration that are borne

disproportionally by poor communities.

10. Schelling (1984) provided many interesting thought experiments in which policy proposals that on the

surface appear to be unfair can be rendered mutually beneficial though appropriate compensatory

mechanisms and transferable p roperty rights assignments.

Volume 14

Issue 4 601

Research Article Role of the Cost-of-Crime Literature

Third, cost–beneﬁt analysis forces the policy process to struggle with the fact that we

operate in a world of scarcity and that resources devoted to one intervention cannot be

devoted to others. Spending on an expensive medical intervention may displace spending

on preventive activities. It may be the case that the life years saved per dollar spent may be

greater from the latter relative to the former. Criminal justice expenditures may displace

public expenditures on early childhood education and other investments in health and

well-being targeted toward at-risk families. A mature policy process must confront these

trade-offs responsibly, even if it requires unseemly tabulations such as the placing a dollar

value on a murder prevented.

Ethical Objections to Cost–Beneﬁt Analysis

Cost–beneﬁt and cost-effectiveness analyses are inﬂuential and widely employed in other

policy domains involving the regulation of risk. For example, in their review of research

on the “value of a statistical life,” a concept that we will discuss in detail in the next

section, Viscusi and Aldy (2003) cited several federal agencies employing cost–beneﬁt

analysis between 1985 and 2000, including the Federal Aviation Administration, the U.S.

Environmental Protection Agency, the Food and Nutrition Service division as well as the

Food Safety and Inspection Service division of the U.S. Department of Agriculture, the

U.S. Food and Drug Administration, and the Consumer Product Safety Commission. These

agencies employed the workhorse tools of welfare economics in deliberating over a variety

of topics ranging from the regulation of the sale and distribution of cigarettes, the use of

lead paint, and the regulation of fuel additives, to ozone air quality s tandards and portable

bed rails.

To be sure, the use of cost–beneﬁt analysis and the tools of welfare economics in these

alternative policy domains are controversial to some and have certainly been criticized on

moral and ethical grounds. In a widely cited paper, Kelman (1981) provided a general cri-

tique of the use of cost–beneﬁt analysis in policy making. Although many of the examples

cited pertain to regulatory activity in the domains of health and safety and the environment,

the relevance to criminal justice policy is apparent. Kelman’s (1981) critique can be summa-

rized as follows. First, in some situations, even though the beneﬁts may outweigh the costs,

the violation of a moral duty (i.e., the duty not to kill, lie, maim, or defame) may simply

preclude consideration of the policy in question. Second, even when duties and rights are

not absolute, they do have “prima facie moral validity,” and when they come into conﬂict

with a favorable cost–beneﬁt calculation, they should be given due consideration in the

deliberation process. Kelman further contended that some issues of such great importance

should not be governed by cost–beneﬁt analysis. To quote Kelman (1981: 36) directly,

“many environmentalists fear that subjecting decisions about clear air or water to the cost

beneﬁt tests that determine the general run of decisions removes those matters from the

realm of specially valued things.”

602 Criminology & Public Policy

Dom

ınguez and Raphael

It is important that we take these critiques seriously as these arguments do point to

limitations of welfare economic analysis as a guiding light for policy makers. In fact, we would

argue that most defenders of cost–beneﬁt analysis are fully cognizant of these limitations,

as is evident in the testy written responses to Kelman’s (1981) essay. For example, DeLong

(1981: 39) argued that the “decision to use cost-beneﬁt analysis by no means implies the

adoption of the reductionist utilitarianism described by Kelman.” Solow (1981: 41) wrote

that “Kelman hints broadly that ‘economists’ are so morally numb as to believe that a

routine cost-beneﬁt analysis could justify killing widows and orphans, or abridging freedom

of speech, or outlawing simple evidences of piety and friendship.” Both quotes reveal a

healthy appreciation of the fact that cost–beneﬁt calculations provide one of many criteria

that must be considered in public deliberations.

A recent salient example whereby net social cost is considered in tandem with, and per-

haps is constrained by, larger ethical concerns is provided by the recent National Academy

of Sciences (Travis and Western, 2014) report on the increase of mass incarceration in

the United States. The committee authors of the report laid out a set of principles that

should guide sentencing with an eye on minimizing the net social costs of our criminal

justice policy while emphasizing some of the rights and duties that Kelman argued have

prima facie moral value. Although Travis and Western recognized the value of sentencing in

preventing crime through incapacitation and deterrence, they explicitly called for consider-

ation of social justice (that the nation’s correctional apparatus should not exacerbate existing

salient inequalities and perhaps aim to narrow them), respect for citizenship (guaranteeing

minimum human rights, preserving dignity, and not damaging the long-term prospects of

prison inmates, most of whom will eventually be released), proportionality (that sentencing

disparities reﬂect heterogeneity in the severity of crime and culpability of the offender), and

parsimony (that sentences be no harsher than needed to achieve retributive, crime-control,

and rehabilitative goals) in sentencing reforms. The authors argued that these principles of

restraint provide “complements, not alternatives, to recent emphases on offender account-

ability and crime control” (Travis and Western, 2014: 323).

Clearly, deliberations over criminal justice policy must strike a balance between the net

beneﬁts of our policy options and many of the fundamental ethical concerns that pervade

all aspects of the criminal justice system. Thus, careful deliberations of the nature prescribed

by welfare economics should certainly play a prominent role. It is interesting to consider the

closing remarks of Kelman’s 1981 essay. Although not written speciﬁcally about criminal

justice policy, the sentiments eerily foreshadowed some of the excesses to come in criminal

sentencing:

My own judgment is that modest efforts to assess beneﬁts and c osts are justiﬁed,

although I do not believe that government agencies ought to sponsor efforts to

put dollar prices on non-market things. I also do not believe that the cry for

more cost beneﬁt analysis in regulation is, on the whole, justiﬁed. If regulatory

Volume 14

Issue 4 603

Research Article Role of the Cost-of-Crime Literature

ofﬁcials were so insensitive about regulatory costs that they did not provide

acceptable raw materials for deliberative judgment, my conclusions might be

different. (Kelman, 1981: 40)

Although in the counterfactual world more cost–beneﬁt analyses may not have pre-

vented the subsequent quadrupling of the nation’s incarceration rate, one might reasonably

argue that more deliberative and less emotional response to crime would have likely yielded

a different national path.

Placing a Dollar Value on Criminal Victimization

Cost–beneﬁt and cost-effectiveness analyses of criminal justice interventions require two

key sets of inputs. First, one needs estimates of the effects of a given policy intervention on

key consequences including such outcomes as crime rates, incarceration levels, and collateral

consequences of the policy for offenders, potential offenders, family of offenders, and so

on. Second, one needs monetary metrics that facilitate placing these consequences on a

common footing. In this section, we discuss the small but growing literature that has so far

attempted to place a monetary value on criminal victimization.

Several key cost concepts merit discussion before we review speciﬁc methodological

approaches and studies. First, research on the costs of crime usually draws a distinction

between ex ante estimates of crime costs (society’s willingness to pay to reduce criminal

victimization before the identity of the victim is known) and the ex post costs of crime

(the monetary value of the damage imposed on an identiﬁed victim). Ex post costs provide

the key concept for formulating and arriving at a remedy. However, to the extent that

cost–beneﬁt analysis is motivated by the quest to identify the optimal provision of public

goods, ex ante cost should guide public deliberations regarding preventative expenditures.

Second, most analysts draw a distinction between the direct costs of crime and the

indirect costs of crime. Direct costs of crime are those that are generated by speciﬁc criminal

victimizations, inclusive of the costs to victims and the costs to society of responding

to criminal victimization. Indirect costs include all of the resources devoted t o, and the

various ways that our lives are altered by, crime-prevention effor ts and efforts to avoid

becoming a victim. Expenditures on locks and security systems; residential choices that

increase commute times; and costly changes in behavior such as avoiding certain areas, not

leaving one’s home after dark, or not allowing one’s children to play outside would all be

subsumed under the indirect costs heading.

Finally, many researchers draw a distinction between the tangible and intangible costs

of crime. Tangible costs are those that can be measured through economic transactions,

for example, medical costs of victimization, processing costs for a criminal defendant, costs

associated with punishment, the value of lost property, and the value of earnings losses asso-

ciated with diminished capacity. Although tangible costs are conceptually straightforward,

measurement challenges abound especially when one considers loss of life or diminished

604 Criminology & Public Policy

Dom

ınguez and Raphael

earnings capacity. Perhaps the most controversy surrounds the articulation and measurement

of intangible costs. Such costs include pain and suffering to victims, lost companionships

to family members, and general diminished quality of life resulting from criminal victim-

ization. We will discuss key examples of cost–beneﬁt analyses next, where consideration of

intangible costs is the difference between net-positive and net-negative social beneﬁts.

Several alternative methodological approaches are available for measuring the cost

of crime, each with its own relative strengths and weaknesses. Many of the studies in this

literature have been juxtaposed and compared in the high-quality reviews by Chalﬁn (2013),

Heaton (2010), and in an earlier review by Nagin (2001). In the proceeding discussion, we

focus principally on the relative strengths of the alternative approaches and the degree to

which the cost estimates align with the willingness-to-pay metric that undergirds welfare

economics, and we selectively review key studies within each category.

Measuring the Cost of Crime Through Hedonic Analysis

The principal challenge faced by researchers striving to place a dollar value on criminal

victimization is that individuals do not directly buy and sell victimization. However, con-

sumers and producers buy and sell goods that bundle together the services of the good with

other attributes, some of which may involve risk of injury or risk of personal victimization.

For example, working the night shift at a convenience store may entail an elevated risk

of victimization relative to working during the day. Living in a high-crime neighborhood

may elevate the risk of being robbed or suffering a residential burglary relative to living in

a low-crime neighborhood. To the extent that the markets for the underlying goods being

traded are competitive (i.e., perfect information regarding good attributes, many buyers

and sellers, and low transaction costs), negative or positive attributes of the good will be

“capitalized” into the price. That is to say, a variation in price will reﬂect both the value of

the principal service of the good in question as well as monetary compensation (that can be

either positive or negative) for the hedonic, implicitly traded attribute. In our two examples,

the night shift, all else held equal, should pay more than the day shift to account for the

higher victimization rate. Similarly, a housing unit in the high-crime neighborhood, all else

held equal, should sell or rent at a discount. Dividing the price differential associated with

working the night shift by the increment in risk of a speciﬁc form of victimization yields an

estimate of the cost of that form of victimization in terms of the compensation demanded

by workers in the labor market.

The reader has likely already anticipated many of the critiques of hedonic analysis.

First, there may be confounding factors not observed by the researcher that are correlated

with risk and that explain observed price differentials for reasons that are independent of

the risk of victimization. For example, working at night while most people are asleep is

certainly a negative job attribute that likely requires additional compensation to induce

someone to ﬁll this position. Absent an estimate of the independent effect on compensation

Volume 14

Issue 4 605

Research Article Role of the Cost-of-Crime Literature

of an irregular schedule, one might falsely overattribute a pay differential between the day

shift and the night shift to the higher risk of victimization.

Second, the assumption of perfect information regarding secondary attributes of goods

bought and sold is unlikely to hold. Information regarding victimization risk in either the

night shift or housing example is likely to be imperfect and subjective perceptions inaccurate.

People are notoriously bad at conceptualizing the true risk posed by low-probability events

and may either systematically overestimate or underestimate the risks they face on a speciﬁc

job or from purchasing a given product type.

Third, choices may be constrained by noncompetitive forces that also distort the price

signal from which the researcher is trying to uncover the cost of a higher victimization

risk. For example, racial housing discrimination may drive up housing costs in poor Black

neighborhoods, reinforcing racial segregation in cities by constraining the residential choices

of Black households, and may increase prices in predominantly Black neighborhoods relative

to what they would be in the absence of such discrimination (see Massey and Denton,

1998; Yinger, 1997). If Black neighborhoods have higher crime rates on average than White

neighborhoods, then the effect of discrimination on housing price differentials will confound

the effect of higher victimization risk. As an alternative example, high unemployment rates

among certain groups may force them to accept whatever employment is available and

prevent the payment of compensating differentials for negative job attributes. Indeed, several

studies found negative associations between workplace risk and compensation, which is an

empirical ﬁnding consistent with noncompetitive forces driving the most vulnerable workers

into low-paying jobs with negative nonpecuniary job amenities (Black and Kniesner, 2003;

Hersch, 1998).

Fourth, hedonic analysis is often based on the trade-offs made by individuals whose

risk preferences are not representative of the general population. For example, the least risk-

averse workers are those who will select into risky jobs because those workers will accept

risky employment for the lowest risk compensation. Should we be using the value place

on victimization risk from such individuals for policy analysis in which loss of life is a key

outcome?

Finally, most of the research on the cost of crime has not generated crime-speciﬁc esti-

mates of willingness to pay to avoid victimization usually because of insufﬁcient independent

variation in the underlying crime rates associated with traded goods.

Thaler (1978) provided one of the earliest estimates that employed hedonic analysis to

explore the economic cost of crime. This study estimated the relationship between property

crime rates and property values in Rochester, New York. Thaler ran regressions of property

values on neighborhood crime rates, holding constant housing unit characteristics, and he

found a signiﬁcant negative effect of crime on property values. Of course, this early study

was subject to critiques regarding potentially omitted variables (e.g., other unfavorable

neighborhood characteristics that are correlated with crime such as local pollution or poor

606 Criminology & Public Policy

Dom

ınguez and Raphael

school quality). In addition, the study did not generate separate estimates of resident

willingness to pay to avoid distinct crimes such as burglary or robbery.

A more recent analysis by Linden and Rockoff (2008) employed a methodological

design that more adequately addressed the omitted variables bias problem; yet it is still

subject to many of the other concerns pertaining to risk perception and pricing. Linden

and Rockoff used sex offender registries to estimate the effect of the arrival of a new sex

offender in one’s neighborhood on values of nearby properties. They focused on a single

county in North Carolina and combined data on parcel sales, parcel characteristics, and the

presence and arrival date of sex offenders from the state’s sex offender registry program. By

using longitudinal and latitudinal coordinates, Linden and Rockoff identiﬁed homes within

0.3 miles of a sex offender’s residence and estimated (a) the distance price gradient once a

sex offender is present, (b) the distance-price gradient prior to the sex offender’s arrival, (c)

the change in home prices within 0.1 miles pre-post the arrival of someone on the registry,

(d) the comparable change between 0.1 and 0.3 miles of the residence, and (e) the relative

change in prices for homes closer to the residence of the sex offender. Linden and Rockoff

found sizable declines in home values within 0.1 mile of a registered sex offender’s residence,

on the order of 4.5%. For those living next door, the changes are larger, on the order of

12%. The price effect, however, declines rapidly with distance with little evidence of an

effect on homes values just beyond 0.1 miles.

Linden and Rockoff (2008) used these results to back out an estimate of willingness to

pay to prevent a sexual assault base on the decline in home values and a calculation based

on the national victimization survey of the likelihood of being assaulted by a sex offender in

one’s vicinity. Assuming that individuals understand the true risks of having such a neighbor,

the authors estimated a cost per sexual assault on the order of $1.2 million. They explored

the sensitivity of this result to various assumptions, with perhaps the most important being

a systematic tendency to overestimate the likelihood of rare events. Under one set of extreme

assumptions (namely, that the assault in the neighborhood would deﬁnitely happen to the

person in question), the cost estimate drops to $90,000. To be sure, if the price decline

associated with a sex offender’s moving in next door is driven in part by induced moral

repugnance, then these estimates of ex ante victimization cost would be upwardly biased.

The body of hedonic analysis studies that has had the largest impact on cost-of-crime

research involves studies devoted to estimating the value of a statistical life (VSL). Most

of this research has involved testing via the use of multivariate econometric models for

risk-wage trade-offs in labor market data for the United States and for other countries. This

risk-wage premium is then used to estimate society’s willingness to pay to reduce statistical

deaths by one by essentially tabulating the amount of additional wage compensation needed

to accept an increase in fatality risk that would generate one additional death. An alternative,

though smaller, body of studies has generated VSL estimates from consumer product price

differentials, where given products such as car seats or air bags are used to estimate individual

Volume 14

Issue 4 607

Research Article Role of the Cost-of-Crime Literature

willingness to pay to reduce the risk of injury or death. This research was reviewed in detail

by Viscusi (1993, 2008) and Viscusi and Aldy (2003).

The ﬁndings from the VSL literature often provide key inputs to cost-of-crime analyses

and serve as benchmarks against which many of the results using alternative methods

are compared (especially results for homicide). Hence, it is worthwhile to summarize the

general magnitudes of the values from VSL studies, as well as some of the conceptual and

measurement problems in this literature. Chalﬁn (2013) provided an exhaustive summary

of VSL estimates from various markets as well as the use of VSL estimates by federal

agencies, converting all values to 2010 dollars. The results from 31 separate labor market

studies yielded a range of $0.7 million to $26.4 million with a mean value of $9 million.

The 13 studies based on consumer products yielded a range of estimates of $1 million

to $11.1 million, with a much lower average relative to the labor market studies of $3.4

million. Finally, Chalﬁn enumerated 18 separate cost–beneﬁts analyses conducted by federal

agencies, employing VSLs ranging from $1.3 million to $8.6 million (average of $5.3

million). Notably, the cost–beneﬁt analyses using relatively low values are all fairly old,

occurring during the 1980s and early 1990s.

Most VSL estimates are subject to the same critiques that we raised in our more general

discussion of hedonic analysis. Namely, jobs with relatively high workplace risk are also

likely to be characterized by other negative job amenities that would require compensation

to attract workers (for example, the risk of nonfatal serious injury). In addition, individuals

who ﬁll employment oppor tunities with relatively high fatality risks are unlikely to have

preferences that are representative of the average person, or the median voter (take, for

example, the low VSL inferred from the behavior and compensation of foot soldiers in

the drug-selling gang analyzed in the 2000 study by Levitt and Venkatesh). Finally, how

individuals perceive fatality risk and cross-occupational variation in this risk is uncertain,

and there is no reason to believe that subjective assessments of cross-job differences in fatality

risk are accurate.

Two relatively recent VSL estimates addressed many of these concerns and yielded

estimates on the low side of the cost range reviewed in Chalﬁn (2013). The ﬁrst exploited

the 1987 change in federal policy permitting states to increase their speed limits on rural

interstate freeways from the federal mandatory maximum of 55 miles per hour to 65 miles

per hour. Ashenfelter and Greenstone (2004) assessed the effect of this change in states that

opted for the increase on (a) total travel time savings and (b) auto-related fatalities. They

found robust evidence of a 35% increase in auto fatality rates associated with the change in

speed limits and travel time savings of approximately 125,000 hours per additional fatality.

Ashenfelter and Greenstone (2004) argued that state policy makers were well aware of

the mortality consequences of increasing the speed limit as the imposition of the federal

speed limit in 1974 led to a notable and widely discussed decline in trafﬁc fatalities. In

addition, research on the average speed/fatality trade-off was well developed by 1987 and

actively discussed by state legislators across the country in deliberating about whether to

608 Criminology & Public Policy

Dom

ınguez and Raphael

increase speed limits. The change in average travel time was clearly exogenous and the

result of deliberate decision making by policy makers. Moreover, the beneﬁciaries of the

change in policy in terms of reduced travel time (the general driving public) and those who

bore the costs of this change in terms of higher fatality risk (the general driving public)

were essentially the same. Hence, the VSL revealed by this particular policy change comes

as close as possible to a deliberate public choice trading off higher fatality levels for an

economic beneﬁt in terms of lower travel costs. Pricing the time savings at the average wage

for workers in each state, the authors estimated a VSL of $1.54 million in 1997 dollars

($2.25 million in 2015 dollars).

The second study, by Greenstone, Ryan, and Yankovich (2015), modeled the likelihood

of reenlisting in the U.S. Army as a function of retention bonuses that vary over time and

by army occupation and the variation over time and by occupation in fatality risk for the

period 2002 to 2010. Greenstone, Ryan, and Yankovich estimated the effects of changes in

fatality risk as well as of changes in the retention bonus on the likelihood of reenlisting and

used these estimates to derive the VSL revealed by the c hoices of young men and women

deciding whether to continue with their military careers. This exercise yields VSL estimates

ranging from $165,000 to $769,000.

These two studies raised several issues that are pertinent to cost–beneﬁt analysis in

criminal justice policy. First, the one explicit example in which we observe policy makers

deliberately choosing higher fatality levels in exchange for an economic beneﬁt yields a

VSL much lower than what is commonly assumed in crime-related cost–beneﬁt analyses

that price the loss of life. Although we also noted cost–beneﬁt studies conducted by federal

agencies that assume higher VSLs, the deliberations of states and their ultimate choices

reveal the actual implicit valuations of lives from these state choices. Notably, none of these

states subsequently reversed the decision to increase speed limits, and the federal limit on

speed limits was subsequently abolished in 1995. Hence, to the extent that we are willing

to interpret the policy process as reﬂecting deliberate informed choice, these subsequent

actions (or inactions to be precise) suggest an implicit judgment that the value of the travel

time savings was not outweighed by the extra fatalities.

To be sure, it may be the case that the manner in which a life is lost bears substantively

on society’s willingness to pay to avoid such an outcome and that a homicide is more costly

in terms of willingness to pay than an auto-related fatality. Nonetheless, Ashenfelter and

Greenstone’s (2004) results, in conjunction with the general sensitivity of crime-related

cost–beneﬁt analysis to the assumed VSL (see Chalﬁn and McCrary, 2013), should inspire

caution in interpreting the cost–beneﬁt ratio calculations for criminal justice interventions in

which the homicide effect of a policy intervention is the principal contributor to net beneﬁts.

The second issue concerns the question of what is the appropriate value to place on

the lives of individuals who often willingly and for meager reward place themselves at great

risk. Consider the relatively low VSL revealed by the reenlistment behavior of soldiers. A

strict adherence t o cost–beneﬁt analysis would value the lives of these individuals ex ante at

Volume 14

Issue 4 609

Research Article Role of the Cost-of-Crime Literature

the VSL that is revealed by their choices. Hence, if the cost per life saved of providing body

armor or reinforcing the ﬂoors of transport vehicles exceeds $769,000, then ex ante net

beneﬁts from such an investment would be negative. Our best g uess is that most Americans

would oppose a policy tabulation and subsequent policy decision that placed a lower value

on the lives of soldiers than the value we place on members of the general public in other

risk-related deliberations. Interestingly, Greenstone et al. (2015) found lower VSLs among

veterans that have seen combat, which is suggestive of a complex relationship between past

experience and the willingness and motivation to take on lethal risk. For example, perhaps

having experienced combat engenders a greater sense of duty and a consequent increase in

the likelihood of reenlisting.

As a related example, consider a cost–beneﬁt analysis of a policy choice designed to

decrease auto fatalities among teenagers (for example, increasing the minimum age for a

driver’s license to 18). We k now that teenagers, on average, are more likely to engage in risky

behaviors than adults, often in exchange for relatively small rewards (for example, speeding

and texting while driving). If one were to back out the revealed value that teenagers place

on a statistical life from observed risk–reward trade-offs, one might reasonably propose that

such calculations would yield low valuations relative to the valuations that the behaviors

of the same individuals would reveal 10 years later. A cost–beneﬁt analysis that values lives

saved using the valuations revealed by the behavior of teenagers themselves may fail to

yield a cost–beneﬁt ratio greater than one. Of course, policy makers may paternally place

higher values on the lives of teenagers in their deliberations (for example, by applying

an average estimate for young adults) by taking into account neuroscientiﬁc research on

brain development, age, and impulsive and risky behavior. However, this would depart

(reasonably, we would add) from strict cost–beneﬁt analysis.

This brings us back to the low VSLs revealed by the choices of members of drug-

selling gangs. One can certainly argue that their choices reveal very low valuations of their

own lives and that perhaps economic valuations of homicides involving criminally active

individuals should reﬂect these low valuations. However, one may also argue that many

of these participants are young, that their preferences for and attitudes toward risk are

likely to evolve as they age, and that their willingness to accept risk may be inﬂuenced by

past experiences and current circumstances in manners that are complex and not necessarily

immutable. Interestingly, in assessing civilian mortality risk in an attempt to characterize the

alternative risk proﬁles of soldiers who do not reenlist, Greenstone et al. (2015) estimated

fatality risks for young African American men in many U.S. cities that are comparable with

many of the high-risk, army occupations on an active tour of duty. “Revealed preferences

for risk” as commonly conceptualized in relatively reductive models of human behavior may

fail to account adequately for the endogenous determination of these preferences, and the

role of such factors as prior traumatization, structural inequalities that impact aspirations,

and general perceptions regarding one’s future. To place a relatively low value on the lives

of young crime victims who engage in high-risk behavior, in our view, places too much

610 Criminology & Public Policy

Dom

ınguez and Raphael

stock in the underlying theoretical paradigm and, in particular, the implicit assumption of

exogenous risk preferences.

Contingent Valuation Studies

We noted in our general discussion of welfare economics that one main goal of cost–

beneﬁt analysis is to uncover the public’s preferences or demand for goods and services that

are generally suboptimally provided by private markets. We also noted that public safety

services have many of the key characteristics of a public good. Namely, once provided, it is

difﬁcult to exclude someone who refuses to pay taxes from enjoying these beneﬁts, and that

one person enjoying the beneﬁts of a safe environment does not detract from the beneﬁts

enjoyed by others. Although hedonic analysis provides an indirect path to uncovering these

valuations, an alternative strategy would be simply to ask people how much they value

a safe environment. This is essentially the methodological strategy underlying contingent

valuation surveys.

Contingent valuation surveys aim to estimate the distribution of willingness to pay for

a speciﬁc public good. For example, one might inquire via a contingent valuation survey

whether a household is willing to pay $0, $50, $100, $150, or $200 in the form of higher

taxes for an intervention that would decrease homicide by 10%, for example. Usually the

amount queried is randomly varied across respondents. With completed responses from a

random sample of households, one can estimate the proportion of households unwilling

to pay at least $50, the proportion willing to pay $50 but not $100, the proportion

willing to pay $100 but not $150, the proportion willing to pay $150 but not $200, and

the proportion willing to pay $200. Subject to speciﬁcation choices regarding the value

to assign to each category (for example, should those unwilling to pay $50 be assigned

values of zero), the researcher can then calculate the average willingness to pay among all

households. Multiplied by the number of households in the United States, one arrives at

the total willingness to pay for a 10% reduction in homicide. Dividing by the number

of homicides that would constitute a 10% reduction yields willingness to pay to reduce

statistical homicides by one.

Contingent valuation analyses have a fairly long and somewhat controversial history

in environmental economics. A historical ﬂashpoint arose in the litigation and damages

assessment after the March 1989 Exxon Valdez oil spill along the Alaskan coastline. A damage

assessment by Hausman, Leonard, and McFadden (1995) valued the direct economic

losses from diminished recreation and ﬁshing at approximately $4 million. An alternative

assessment by Carson et al. (1992), based on society’s stated willingness to pay to avoid such

environmental disasters, put the total losses at approximately $3 billion.

The difference

11. Exxon and the state of Alaska ultimately settled out of court for $3 billion in damages. See Carson (2012)

for a discussion of the Exxon Valdez event and the role of economic research in arriving at the

settlement.

Volume 14

Issue 4 611

Research Article Role of the Cost-of-Crime Literature

between these two ﬁgures is primarily a result of the inclusion in the latter estimate of what

environmental economists refer to as “existence” or “passive use” value, namely, the value

placed on the existence of something despite whether one derives direct use and enjoyment

from the object of the valuation exercise. The wedge between these two cost estimates is

analogous to the wedge that exists between realized out-of-pocket crime cost totals and

crime cost totals that account for indirect and intangible costs of victimization, such as

crime avoidance and victim pain and suffering.

This discrepancy in damage assessments sparked an intense debate within the economics

profession regarding the validity of contingent valuation methods. Much of the arguments

for and against the use of these methods are summarized in two symposia published in

the Journal of Economic Perspectives. Peter A. Diamond and Jerry A. Hausman (1994) laid

out the case against under the title “Contingent Valuation: Is Some Number Better than

No Number?” First, they argued that hypothetical statements of willingness to pay are

divorced from any true underlying economic transactions and often inquire about issues for

which taxpayers and consumers have little direct experience in terms of actual purchases. In

addition, respondents essentially have nothing at s take in answering a hypothetical interview.

Responses may be strategic, aimed at pleasing the interviewer, or express protest, but they

may not necessarily bear any resemblance to actual willingness to pay for a speciﬁc public

good. This critique often fuels suspicion that contingent valuation estimates are in general

biased upward because nothing is at stake for the respondent.

Diamond and Hausman (1994) raised additional concerns about the convenience

samples used in some contingent valuation surveys (for example, stopping shoppers in

malls), concerns that numerical answers may ultimately be anchored by the payment choices

speciﬁed by researchers beforehand in designing the contingent valuation survey, and that the

existence value ﬁgures are unrealistically large. Finally, several authors have raised concerns

about what some refer to as the embedding effect, namely, that willingness to pay tends

to be insensitive to variation in the ultimate impact of the intervention. For example,

willingness to pay to clean up 10 lakes may seem to be the same as the willingness to pay to

restore 100 or 1,000. In other words, respondents may simply be stating their willingness to

contribute to some vague environmental effort, with the magnitude of the ultimate effect

being poorly understood. There is reason to believe that this may be a particular problem

in the cost-of-crime estimates, a point that we expand on shortly.

On the pro-side, W. Michael Hanemann (1994) contended that much of what society

values is never traded in markets, and if we were to restrict public-sector interventions to

those domains for which there exists a close market substitute, the public sector would fail

to address and appropriately deliberate over many issues and problems that constitute the

fundamental responsibilities of the state. Although Hanemann was principally referring to

valuing environmental damage, the applicability of this argument to the criminal justice do-

main is self-evident. Hanemann noted that most serious contingent valuation efforts employ

sophisticated sampling designs, pretest their survey instruments, and word survey questions

612 Criminology & Public Policy

Dom

ınguez and Raphael

in a manner that reminds the respondent of the tax implications of the program. Hanemann

also cited several studies that cross-validated the results from contingent valuation analysis

with actual market transactions and noted that in many of these studies, willingness to pay

from contingent valuation is within the ballpark of observed expenditures.

The debates surrounding the valuation of the Exxon Valdez spill led to an unusually

high-proﬁle assessment of the merits of this relatively obscure research tool (Portnoy, 1994).

The National Oceanic and Atmospheric Administration (NOAA) convened a panel of

experts to assess whether contingent valuation methods can credibly provide estimates of

environmental damage and, in particular, lost passive use value. The ﬁnal report (Arrow

et al., 1993), which included the Nobel laureates Kenneth Arrow and Robert Solow as

co-authors and panel chairs, is often cited as a how-to guide and quality control document

in many of the contingent valuation studies that have followed.

The panel ultimately concluded that contingent valuation surveys can indeed feasibly

be employed to evaluate environmental damage. Most relevant to the topic of this essay, the

NOAA panel laid out a set of best practices as they pertain to contingent valuation methods,

and a set of guideposts against which to assess study quality (Portnoy, 1994). These include

the recommendations that surveys be administered in person rather than by phone, and

by phone rather than by mail, and that the surveys be based on probability samples rather

than on convenience samples. The report also recommended that contingent valuation

surveys measure costs based on willingness to pay rather than on willingness to accept

as the key measure of valuation (an explicit acknowledgment of concern that estimates

may be implausibly large). Additional guidelines include the recommendation that surveys

pose questions in a referendum format that emphasizes the tax and income consequences

of supporting a proposal, that respondents be reminded of the effect of the proposal on

personal income, and that surveys include follow-up questions that permit assessment of

whether the individual understood the question and whether willingness to pay varies with

observable characteristics such as income in a manner predicted by simple economic theory.

The cost-of-crime literature that has taken this approach has beneﬁtted greatly from

this intellectual history. Notably, the earliest applications of contingent valuation methods

(Cook and Ludwig, 2000; Ludwig and Cook, 2001; Zarkin, Cates, and Bala, 2000) post-

date this debate and cite the NOAA recommendations in their methods sections, although

not all are followed in at least one of these studies (Zarkin et al. [2000] relied on mall

intercept samples in two communities on the East Coast). Although this body of literature

is small, the existing studies to date have been high quality and generally have followed

most of the NOAA guidelines. In addition, the costs of crime, even the intangible costs,

12. The second

Journal of Economic Perspectives

symposium, published after the 2010 Deepwater Horizon

oil rig disaster in the Gulf of Mexico, reiterated many of the same arguments, although as noted by

Kling, Phaneuf, and Zhao (2012), this literature h as advanced and matured considerably in the

intervening 20 years ( see Hausman, 2012, and Carson, 2012, for their updated assessment of the merits

of contingent valuation methods).

Volume 14

Issue 4 613

Research Article Role of the Cost-of-Crime Literature

have more conceptual clarity than the somewhat vague concept of passive-use value in

environmental studies. Cert ainly, victim pain and suffering is a real phenomenon that we

believe most individuals can understand and conceptualize. Even passive use value in the

criminal justice realm (personal welfare associated with knowing that society is safer for all

even if one is personally not at a high risk of becoming a crime victim) is a more concrete

social objective over which respondents can express opinions relative to the valuation of a

particular ecosystem that respondents are unlikely to visit or may not even be cognizant of.

Another beneﬁt that the crime contingent valuation studies have over the environmental

studies is that one of the big-ticket crimes (in particular, homicide) has an external body

of research based on market transactions against which values can be compared, that is,

the extant VSL estimates. Here, we provide a selective review of key studies in contingent

valuation analysis applied to the costs of crime.

Cook and Ludwig (2000; Ludwig and Cook, 2001) provided perhaps the ﬁrst applica-

tions of contingent valuation methods to criminal justice research. The authors generated

estimates of the cost associated with gun violence in the United States. In their book, Cook

and Ludwig (2000) devoted great effort to drawing qualitative and quantitative contrasts

between gun violence cost estimates based on the cost-of-illness (COI) approach, which

effectively enumerates gun homicides and injuries and attempts to generate ex post estimates

of each, and measures of society’s willingness to pay to prevent gun violence. By design,

contingent valuation methods are intended to generate ex ante estimates of willingness to

pay for public goods that would decrease the risk of these particular outcomes. Given the

many preventative measures that individuals take to avoid becoming a victim, including

security service and device purchases and avoidance behavior, it is not clear a priori whether

COI estimates or contingent valuation estimates would yield larger social cost ﬁgures.

Cook and Ludwig (2000; Ludwig and Cook, 2001) developed a contingent valuation

questionnaire that was administered to a nationally representative sample of households

and inquired about willingness to pay for an intervention that would decrease gun violence

by 30%. Survey respondents were randomly assigned to be queried about their willingness

to pay $50, $100, and $200 for a program. Those who responded afﬁrmatively were

asked whether they would pay double the queried amount. Those who answered negatively

were asked a follow-up question regarding whether they would be willing to pay half the

original amount. The questions were worded to ensure that the proposed policy reduced gun

availability among convicted offender and juvenile delinquents but would not interfere with

the gun rights of law-abiding citizens. Cook and Ludwig estimated a national willingness to

pay of $24.5 billion 1998 dollars, amounting to $1.2 million per prevented gunshot injury.

They also used their estimates to generate alternative estimates of the value of a statistical

life assuming (a) all of the willingness to pay is to prevent gun-related deaths and (b) the

cost of gun-related deaths is twice that of nonfatal gun injuries. These yield estimates of a

VSL ranging from $5.4 million to $6.8 million in 1998 dollars ($7.8 to $9.8 million in

2015 dollars). These estimates lie well within the range of VSL estimates reviewed in an

614 Criminology & Public Policy

Dom

ınguez and Raphael

earlier literature review by Viscusi (1993) and the estimate summarized by Chalﬁn (2013),

but the numbers are higher than the implicit valuations of trafﬁc fatalities estimated by

Ashenfelter and Greenstone (2004).

Perhaps the most widely used set of cost-of-crime estimates from the contingent valu-

ation literature was presented by Cohen, Rust, Steen, and Tidd (2004), who presented the

results from a contingent valuation survey inquiring about household willingness to pay to

prevent a host of serious violent offenses. Based on a random sample of U.S. households, the

study inquired about willingness to pay for a program that would reduce speciﬁc crime rates

by 10%. Each household was asked about three of ﬁve serious crimes, with the crimes in-

quired about and the order randomly determined. The willingness-to-pay amount is drawn

from a payment schedule ranging from $25 to $200, with the amount randomly chosen

and the same amount asked for each crime category. From these responses, the authors

estimated the probability density function of the distribution of willingness to pay for each

crime and estimated an average per household using this distribution.

Next, by using contemporaneous crime rates and estimates of the number of households

in the United States, Cohen et al. (2004) generated an estimate of the value of a 10% crime

reduction and the amount that the public would be willing to pay per crime averted. The

per crime ﬁgures suggested $25,000 for burglary, $174,000 for armed robbery, $57,000 for

serious assault, $185,000 for rape/sexual assault, and $8.5 million for a murder. Again, it

is interesting to note that the estimate for murder lies within the range of VSL estimates

discussed earlier. Cohen et al. (2004) presented conﬁdence intervals derived from the

sampling variance of their estimates. These cost estimates exceeded the extant ex post

cost of crime estimates that took a more “bottom-up” approach that we review in the

next section. The authors speculated that the higher values from the willingness-to-pay

studies are explained by the implicit inclusion of all indirect costs of crime that accrue

to society beyond victimization cost. This inference is consistent with Anderson’s (1999,

2012) accounting of the total costs of crime as well as with Nagin’s (2001) critique of earlier

efforts to arrive at bottom-up estimates.

One might raise several concerns about both the Cook and Ludwig (2000; Ludwig and

Cook, 2001) studies as well as the work by Cohen et al. (2004). First, although Cook and

Ludwig were careful to specify in their survey that gun rights for law-abiding citizens would

not be impacted by the proposed intervention, Cohen et al.’s work left the actual nature

of the intervention vague. It is difﬁcult a priori to decide which approach is better; yet the

juxtaposition of these two alternative designs raises the question of whether it is possible to

value victimization in a manner that is divorced from the nature of the policy intervention

that yields the crime abatement. Second, both studies speciﬁed that the program would have

a given percentage effect on crime rates, and then they each backed out values per incident

based on the implicit number of crimes prevented implied by the hypothetical percentage

impact and contemporaneous crime rates. If we are to take these calculations as actual

measures of society’s willingness to pay to prevent a speciﬁc crime, we must believe that

Volume 14

Issue 4 615

Research Article Role of the Cost-of-Crime Literature

(a) survey respondents know what the actual crime rate is and (b) survey respondents can,

during the interview, calculate, digest, and cognitively process a given percentage decline

in crime and what it means to them. This raises concerns regarding the embedding effect

discussed previously. For example, do we ﬁnd lower willingness to pay for a 10% decrease

in crime when crime rates are low?

These questions can be explored in part by reviewing the contingent valuation results

presented by Nagin, Piquero, Scott, and Steinberg (2006) and Piquero and Steinberg (2010).

Nagin et al. (2006) employed a contingent valuation survey to gauge the relative value that

the public places on alternative crime-control strategies that have similar ends yet different

means. Nagin et al. assessed the degree to which the public was differentially willing to pay for

programs that focused on rehabilitation relative to programs that focused on incarceration

and crime control through enhanced deterrence and incapacitation. Nagin et al. surveyed

a random sample of Pennsylvania households, inquiring about their w illingness to pay

for a program that would reduce juvenile offending by 30%. Households were randomly

assigned to be asked about a program that would be focused on rehabilitation as opposed to

a program that would involve incarcerating a convicted youth for an additional year. The

impact was arbitrarily held constant across the two interventions, and individuals were asked

about their willingness to pay $100 for the randomly assigned intervention. For those who

answered yes, a follow-up question inquired whether they would pay $200. For those who

answered no to the ﬁrst inquiry, a follow-up question inquired whether they would pay $50.

The authors tabulated average willingness to pay based on the responses to these questions

and found a statistically signiﬁcant and substantively higher (roughly 25%) willingness to

pay for programs with a rehabilitation focus.

Piquero and Steinberg (2010) repeated this exercise by using the same survey in-

strument and research protocol but ﬁelding the survey 2 years later (2007) in four states

(Pennsylvania, Illinois, Louisiana, and Washington). The results were nearly identical in

this much larger survey; yet the authors found some substantial cross-state differences. The

overall difference in willingness to pay for rehabilitation over incarceration remained and

was similar in magnitude to that found in the earlier survey by Nagin et al. (2006). In ad-

dition, a willingness to pay for the rehabilitation intervention was comparable across states.

However, Piquero and Steinberg (2010) found greater disparities favoring rehabilitation

in Washington and Illinois, no difference in willingness to pay in Louisiana, and a higher

willingness to pay for additional incarceration in Louisiana relative t o the other states in the

study.

This four-state study permitted an exploration of w hether contingent valuation s tudies

of cost of crime exhibit the embedding effect problem highlighted in our discussion of

the environmental literature. As we have already noted, contingent valuation critics raise

concerns regarding whether respondents exhibit reasonable variation in willingness to pay

that accords with basic economic theory. To be speciﬁc, do the results of these studies suggest

that individuals are willing to pay more when the likely effect on crime is larger? The four

616 Criminology & Public Policy

Dom

ınguez and Raphael

states surveyed in 2007 by Piquero and Steinberg (2010) differed greatly in terms of their

violent and property crime rates. All interviewees were asked about their willingness to pay

for a 30% reduction in juvenile offending. Assuming that juvenile offending is higher in

states with relatively high crime rates, one would expect higher willingness to pay for a 30%

decline in relatively high crime states because a 30% decline translates into relatively larger

reduction in crime rates and individual victimization risks.

Figure 2 graphically depicts violent crime rates (incidents per 100,000) from the

Federal Bureau of Investigation’s Uniform Crime Reports and average willingness to pay

from Piquero and Steinberg (2010) for the included states. The four states are ordered

from lowest to highest in terms of the violent crime rate. Even though the violent crime

rate in the highest crime state (Louisiana) is more than double the violent crime rate in

the lowest (Washington), there is no detectable relationship between willingness to pay and

the crime rate. Willingness to pay for a rehabilitative intervention is basically ﬂat across

states. Willingness to pay for an additional year of incarceration is roughly comparable

across all states and then jumps by $20 for Louisiana. Figure 3 presents a comparable

graph for property crime rates. Here, we express property crimes as incidents per 10,000

to facilitate the visual comparison between crime rates and willingness to pay on the same

scale. Although the ordering of states changes some what, we still ﬁnd little evidence of

higher willingness to pay in states with higher crime rates.

Does this imply that residents of Louisiana value victimization less than residents of

Washington or Illinois? Not necessarily. An alternative interpretation would be that the

average respondent simply does not comprehend in a meaningful way what constitutes a

30% reduction in offending among juvenile offenders. If this is true, however, the relative

constancy across states suggests that the survey is picking up a general willingness to

contribute to an effort, which is similar to the critique regarding embedding effects that has

been raised in the environmental damages literature.

Interestingly, the ﬁnding of a preference for rehabilitation over incarceration foreshad-

ows some of the more high-proﬁle ballot referendums as well as other criminal justice

reforms that have occurred in recent years. For example, voters in California passed voter-

initiated propositions to divert drug offenders to treatment rather than to incarceration in

2000; to moderate sentences for less serious third-strike felonies in 2012; and to redeﬁne

many low-level drug offenses and larceny from felonies to misdemeanors, with a provision

to redistribute savings toward schools and reentry efforts in 2014. As another example, the

federal Justice Reinvestment Initiative is essentially consistent with the social ranking of

policy tools derived from these contingent valuation studies.

Bottom-Up Estimates of the Cost of Crime

An alternative strategy for estimating cost of crime was hinted at in our discussion of

Cook and Ludwig (2000). Speciﬁcally, in theory one can enumerate the various costs that

Volume 14

Issue 4 617

Research Article Role of the Cost-of-Crime Literature

FIGURE 2

Violent Crime Rates (per 100,000) and Willingness to Pay (WTP) for a 30% Reduction in Juvenile Oending from Piquero

and Steinberg (2010)

Sources: Piquero and Steinberg (2010) and the Uniform Crime Reports, 2007.

618 Criminology & Public Policy

Dom

ınguez and Raphael

FIGURE 3

Property Crime Rates (per 10,000) and Willingness to Pay (WTP) for a 30% Reduction in Juvenile Oending from Piquero

and Steinberg (2010)

Sources: Piquero and Steinberg (2010) and the Uniform Crime Reports 2007.

Volume 14

Issue 4 619

Research Article Role of the Cost-of-Crime Literature

are created by speciﬁc victimizations (medical expenses, lost work, cost of investigation,

criminal case processing, punishment, etc.), place dollar values on each, and then sum. In

practice, these bottom-up estimates have been based on a hybrid of ex post and ex ante cost

estimates, with loss of life being valued with ex ante VSL estimates and all other costs being

generated from information on typical injuries sustained, estimates of lost work time, and

estimates of average costs associated with each.

Perhaps the most controversial cost category in bottom-up estimates pertains to pain,

suffering, and diminished quality of life (i.e., the intangible costs to crime victims). Certainly,

such costs are real and deserve to be included in any cost–beneﬁt calculation. A particularly

salient example of where the inclusion of such costs generates a w idely different cost–beneﬁt

result is found in contrasting an analysis of an Illinois early release program by Austin

(1986) with the reanalysis of this policy by Cohen (1988). By relying on cost estimates by

the National Institute of Justice (NIJ) that were limited to out-of-pocket victim expenditures

such as medical costs and property damage, Austin (1986) concluded that the budgetary

savings from the early release program far exceeded the marginal increases in crime associated

with recidivism among early releases.

Cohen (1988), however, argued that this conclusion was based on the very low valuation

placed on serious offenses and on the failure of the NIJ ﬁgures to consider the intangible

costs to victims. In several publications, he singled out the low assumed victim costs for rape

($350 in 1986 dollars) as a key ﬂawed input in the Austin analysis (Cohen, 1988, 2005).

Cohen (1988) generated more inclusive estimates of victimization costs that accounted

for pain and suffering via an imputation based on data from jury awards from civil trials.

Cohen’s (1988) principal strategy was to estimate the relationship in civil cases between

medical costs of an injury and pain-and-suffering awards and then to use this relationship

in conjunction with medical cost estimates for speciﬁc forms of victimization to impute a

pain-and-suffering value. For example, pain and suffering for rape cases was estimated in

the following manner. Cohen took a weighted average of the medical costs for traumatic

neurosis, severe psychiatric distress, broken bones, and other injuries that occur in the

commission of rape, where the weights were the incident rates of the particular condition

associated with the victimization. Pain-and-suffering awards were then imputed from the

estimated empirical relationship between medical costs and jury awards. In generating the

total cost of speciﬁc victimizations, Cohen presented separate estimates that incorporated

the likelihood that an incident would result in a fatality. Multiplying the fatality incident

rate by a VSL estimate generated a contribution to cost associated with being exposed to

the fatality risk. By incorporating intangible costs, Cohen estimated the victimization cost

of rape at $57,000 rather than at $350, and he drew the opposite conclusion regarding the

sign of the net beneﬁts of the Illinois early release program. Cohen’s conclusions strongly

admonished against partial victimization cost estimates. To quote the author directly, “in

620 Criminology & Public Policy

Dom

ınguez and Raphael

my mind, to conclude that this early-release program was worthwhile was tantamount to

balancing the state budget on the backs of crime victims” (Cohen, 2005: xii-xiii).

Cohen’s early work, and in particular the jury-award methodology, ﬁgured prominently

in the work of Miller, Cohen, and Wiersema (1996). This study is perhaps the most widely

cited in this literature and is the source of victimization cost ﬁgures frequently used in

back-of-the-envelope calculations made by economists and others studying criminal justice

topics. The project makes great effor t to generate victimization counts that adjust for

underreporting for a wide range of violent and property crimes. For each crime type, Miller

et al. (1996) provided separate estimates of “tangible losses” (costs associated with proper ty

damage and loss, medical care, death, mental health care, policy and ﬁre services, victim

services, and lost productivity) as well as “intangible losses” in the form of pain and suffering

and diminished quality of life. Again, the victimization cost estimates were largely ex post,

although for fatal victimizations, the authors employed an estimate from the VSL literature.

In some tabulations, they estimated separate costs for each crime type that incorporated the

costs associated with the fatality risk.

McCollister, French, and Fang (2010) presented an updated bottom-up analysis of

victimization costs by using more recent jury-award data as well as more detailed in-

formation to generate cost-of-illness estimates of the tangible costs of victimization.

McCollister et al. (2010) also presented victimization cost estimates for a larger range

of offenses relative to previous studies and presented a detailed comparison of their

estimates with previous estimates from this literature. Table 2 reproduces a subset of

the results presented in Table 1 of McCollister et al.’s (2010) article. All estimates

have been converted to 2015 dollars. The ﬁrst column presents the ex ante costs from

Cohen et al.’s (2004) contingent valuation analysis, whereas the second column presents the

bottom-up estimates from Miller et al. (1996).

The broader set of cost estimates in McCollister et al. (2010) is presented in the ﬁnal

column of Table 2. Several interesting points can be made about these comparative esti-

mates. First, the most recent bottom-up estimates are fairly close to the results from the

contingent valuation study, with a generally consistent ranking of crime severity (despite

drastic differences in methodology). Second, in the two bottom-up studies, the estimated

13. Rajkumar and French (1997) provided another early contribution to the bottom-up cost estimates of

criminal victimization that generate estimates of intangible costs based on pain and suffering jury

awards. Although this publication generated the cost of criminal victimization, the principal aim was to

provide a cost–benefit analysis of drug-treatment programs. The authors used hospital emergency

room data to estimate the costs of t reating various injuries. In conjunction with information on the

incidence of various injuries associated with specific criminal victimizations, the authors generated

estimates of the direct costs of a specific crime. To generate intangible costs, they followed the jury

award methods proposed by Cohen (1988), estimating the relationship between medical expenses and

pain and suffering awards. They did not include information on psychological injuries and estimated

intangible costs for robbery and assault only. They then used this in conjunction with estimates of the

crime-mitigating effects of d rug treatment to perform a cost–benefit analysis of drug treatment taking

into account crime victimization costs.

Volume 14

Issue 4 621

Research Article Role of the Cost-of-Crime Literature

TAB LE 2

Comparison of Unit Cost of Crime from Two Key Bottom-Up Studies and One

Key Contingent Valuation Study (All Costs in 2015 Dollars)

Oense Cohen et al. (2004) Miller et al. (1996) McCollister et al. (2010)

Murder $12,372,429 $4,774,809 $9,791,369

Rape or sexual assault $312,042 $135,617 $262,446

Assault $92,165 $23,382 $116,652

Robbery $305,458 $20,264 $46,118

Arson — $58,546 $23,002

Larceny theft — $577 $3,850

Motor vehicle theft — $6,235 $11,741

Household burglary — $2,338 $7,044

Embezzlement — — $5,973

Fraud — — $5,485

Stolen property — — $8,692

Forgery or counterfeiting — — $5,739

Vandalism — — $5,297

See reference sections for study citations. All gures are converted to 2015 dollars based on the 2008 valuations reported in McCol-

lister et al. (2010) and the percentage change in the CPI-all urban consumers between 2008 and early 2015.

property crime costs were low relative to the costs of violent crime, homicide in particular.

Other scholars compared the sensitivity of cost–beneﬁt studies of criminal justice inter-

ventions to whether one can demonstrate an impact of the intervention on homicide (see

Chalﬁn, 2013; Chalﬁn and McCrary, 2013). The reason for this is perfectly clear when

one notes that the estimated cost of a homicide is 37 times the next most costly crime

(rape/sexual assault) and 833 times the most costly property offense (motor vehicle theft).

Clearly, the method for estimating intangible costs is the most controversial and de-

batable component of the bottom-up cost estimates. Few would argue that the true costs

of victimization should be limited to out-of-pocket expenditures. That is to say, it is hard

to debate against the proposition that crime victims’ suffering goes beyond the sum of lost

wages, medical expenses, and property damages. However, one may legitimately question

whether the decisions o f juries reﬂect anything akin to the compensation and/or equivalent

variation associated with the nontangible aspects of victimization. First, instructions given

to jurors heading into deliberation over general/pain-and-suffering damage evaluation vary

widely across states and tend to be vague (Abel, 2006). In addition, these awards tend to

exhibit great variation for similar injuries (Abel, 2006; Cooter and DePianto, 2011). Jurors

are being tasked with placing a value on unusual and often horriﬁc scenarios for which

their preferences are most likely incomplete and poorly deﬁned, and where any amount of

money is likely to be insufﬁcient compensation (Cooter and DePianto, 2011). There is also

experimental evidence of instability in pain-and-suffering awards in mock trials. Speciﬁ-

cally, Saks, Hollinger, Wissler, Evans, and Hart (1997) as well as Hinsz and Indahl (1995)

622 Criminology & Public Policy

Dom

ınguez and Raphael

provided evidence that caps on pain-and-suffering awards tended to present psychological

anchors around which jurors set pain-and-suffering award amounts. If set high enough,

the award caps have the counterintuitive effect of increasing both average general awards as

well speciﬁc damage awards. Robbennolt and Studebaker (1999) found similar results in

experimental manipulation of punitive damage awards in mock trials. Hence, in addition

to being highly variable within injury class, the average awards exhibit sensitivity to the

framing of deliberation instructions. Finally, some evidence (cited in MacCoun, 2006) of

systemic bias was found in pain-and-suffering awards; minor injuries are overcompensated,

and major injuries are undercompensated.

Many of these critiques of the jury-award methods were raised in the past (see the

2001 review by Nagin and the earlier 1995 discussion by Zimring and Hawkins). To the

list of issues raised earlier one might add the fact that jurors are making decisions that are

divorced from their own personal budget constraint as they are imposing a cost on a third

party. Zimring and Hawkins (1995) also raised concerns pertaining to the fact that the total

cost of crime once one takes into account pain-and-suffering estimates sums to an amount

that would greatly exceed what most people would consider reasonable expenditures. For

example, we noted that by some estimates, the costs of crime amounted to 9% to 19% of

the country’s gross national product. Does this then imply that society would be willing to

pay 9% to 19% of gross national product to eliminate crime completely?

One might raise additional concerns regarding the general high level of jury awards

in the United States and whether these awards are reﬂective of some structural bias in

our system for mediating civil disputes (torts in particular). Media reports tend to focus on

outsized awards for pain and suffering and punitive damages to an extent that they likely bias

common perceptions regarding average tort compensation (MacCoun, 2006). In fact, most

tort cases settle out of court for modest amounts, and there is little evidence of a trend toward

larger awards (Robbennolt, 2000). Nonetheless, on average, pain-and-suffering awards are

higher in the United States than in other Western developed countries (Sugarman, 2005).

In addition, awards in the United States exhibit greater variation for similar injuries (in the

language of public ﬁnance, they exhibit greater horizontal inequity) than in other nations,

which suggests greater overall capriciousness in our system for valuing pain and suffering.

Finally, awards in the United States do not accurately reﬂect the net receipts to plaintiffs, as

attorney fees are often charged as a percentage (commonly one third) of awards (Sugarman,

2005). Of course, higher awards in the United States might reﬂect international variation

in the value placed on intangibles.

Relative Comparison of These Three Approaches to Estimating the Cost of Crime

The three approaches to measuring victimization costs—hedonic analysis, contingent valu-

ation, and the bottom-up accounting method—can be compared along several dimensions.

First, both hedonic analysis and the results from contingent valuation surveys estimate

ex ante willingness to pay to avoid or prevent criminal victimization. From the viewpoint of

Volume 14

Issue 4 623

Research Article Role of the Cost-of-Crime Literature

a public choice theorist, this approach to deliberating prescriptively about possible criminal

justice interventions is correct. The bottom-up studies presented, for the most part, are

ex post cost estimates. A subset of these estimates presented victimization costs that were

adjusted for fatality risk by adding the product of the likelihood of a fatality multiplied

by a chosen VSL estimate. This particular adjustment seems logically inconsistent with the

general premise of a bottom-up approach. As these studies carefully and comprehensively

articulated the consequences of an actual offense, these estimates are best thought of as the

compensating variation associated with being a crime victim (i.e., the minimum payment

needed to restore utility to the previctimization level). Incorporating into these estimates

an ex ante cost component based on a risk of an outcome that does not materialize is

inconsistent with a damages-done approach.

A second dimension along w hich these methods can be compared is the degree of

inclusiveness of the cost estimates. The hedonic and contingent valuation estimates in

many ways provide a summary or omnibus assessment of the costs of crime that bypasses

the need to articulate all possible consequences of a criminal victimization. To the extent that

consumers or survey respondents are knowledgeable about the consequences of a speciﬁc

crime as well as the likelihood of becoming a victim, then their actions (in hedonic analysis)

or their stated preferences (in contingent valuation analysis) will reveal true valuations

inclusive of direct and indirect costs of crime and perhaps passive-use value derived from the

knowledge that others are victimized at a lower rate. Of course, one might question whether

individuals know what the consequences of a victimization are and whether the average

voter knows current crime rates and the objective likelihood of becoming a crime victim. If

they do not, then the actual point estimates of the cost of crime will be inaccurate. Bottom-

up estimates of course require careful articulation of all possible costs of victimization and

likely underestimate indirect costs associated with crime avoidance and prevention.

A third dimension along which these three sets of estimates can be compared concerns

the likely direction of bias (are they too high or too low?). There are few hedonic estimates

of speciﬁc crime costs. The one exception is the body of VSL estimates that aims to

price fatalities in general rather than fatalities resulting from homicides. Stipulating to the

applicability of VSL estimates as a homicide cost measure, it is important to note that few

of the VSL studies have identiﬁed plausibly exogenous variation in fatality risk, with the

exception of Ashenfelter and Greenstone (2004) and Greenstone et al. (2015). There is

reason to believe that confounding omitted variables will impart an upward bias on VSL

estimates; this conjecture is consistent with the lower estimates found in these two studies

relative to the average values from the wider VSL literature.

Bias in the contingent valuation studies likely depends on several factors, including the

extent to which the average person overestimates crime rates and whether the hypothetical

nature of the survey creates divergence between stated and actual willingness to pay. If

perceived crime rates are higher than actual crime rates, then contingent valuation studies

will overestimate willingness to pay to prevent speciﬁc crimes, whereas the reverse is true if

624 Criminology & Public Policy

Dom

ınguez and Raphael

perceived crime rates are below actual crime rates. Interestingly, public perception of crime

rates and change therein are notoriously inaccurate and tend to overestimate the likelihood

of becoming a crime victim. For example, despite continuously decreasing crime rates in

the United States since 1994, most Americans believe (and in some years the overwhelming

majority) that crime has increased each year (McCarthy, 2014). Aside from inaccurate

perceptions of risk, if people are more generous in terms of their willingness to pay in

hypothetical situations relative to actual referenda, then contingent valuation studies will

yield estimates of willingness to pay that are systematically higher than actual willingness to

pay.

Whether the bottom-up studies are biased in one direction or the other depends

crucially on the measure of intangible costs derived by pain-and-suffering awards, as this

cost component accounts for the lion’s share of cost estimates. Several factors suggest

that average awards are likely to be higher than the willingness-to-pay concept central to

cost–beneﬁt analysis. First, jurors make decisions with someone else’s money. Second, jur y

awards in the United States are inclusive of attorney fees; this factor can be taken into

account by juries when deciding on a pain-and-suffering total. Third, jury awards are high

in the United States relative to other developed nations, which likely reﬂects aspects of

our civil-conﬂict resolution practices. Thus, pricing intangibles is a difﬁcult and poorly

deﬁned task and may often involve putting a price on unimaginable harm. For the most

serious injuries and harms, awards by construction are likely to be insufﬁcient to “make one

whole.”

In the end, all three estimates tend to yield cost ﬁgures that are fairly consistent with one

another in terms of general magnitude and the ordering of crime severity. For homicide,

contingent valuation methods tend to align with VSL estimates from observed market

behavior when loss of life is involved. Given the concerns about actual crime-cost point

estimates raised in this article, we are cautious about analyses that take strong positions

based on whether net beneﬁts are positive. We are more comfortable, however, with policy

analysis that employs cost-of-crime estimates to rank policy alternatives according to their

cost effectiveness.

Conclusion

Mark Cohen (2005), perhaps the largest contributor to the economic cost of crime literature,

wrote about his experience during the 1980s as a young scholar tasked with helping the newly

formed federal sentencing commission assess the relative severity of felony offenses with an

eye on building proportionality and fairness into federal sentencing guidelines. He noted the

lack of guidance from commission members, the dearth of research on the relative severity of

crime, and the rudimentary and partial accounting for victimization costs that existed at the

time. Over the following three decades, a relatively modest number of researchers have set

about ﬁlling in this knowledge gap, employing clever yet controversial innovations (making

Volume 14

Issue 4 625

Research Article Role of the Cost-of-Crime Literature

use of jury awards

) and drawing from other important policy domains where one must

value what is very difﬁcult to value (contingent valuation methods from environmental

economics and VSL studies from analysis of labor market policy). This research effort

has made considerable progress and often has provided key inputs to back-of-the-envelope

calculations summarizing speciﬁc research projects and more systematic and serious efforts

at cost–beneﬁt and cost-effectiveness analysis of criminal justice policy interventions.

Despite this progress, there is certainly room for further development and more research

on this topic is needed. As we discussed, there are reasons to believe that the principal

methods used to generate cost-of-crime estimates may be biased upward. To the extent that

the average survey respondent tends to overestimate the likelihood of becoming a victim or

erroneously believes that crime as a social problem is getting worse, a hypothetical percentage

point decline in crime will be perceived as having an impact on crime that is exaggerated

relative to the true effect. When willingness to pay is normalized using objectively measured

crime rates, the per-incident cost estimates will be too high. In addition, evidence of

embedding effects raises questions regarding the accuracy of these estimates and their

dependence on context. More research is certainly needed on the application of contingent

valuation methods in criminal justice policy, with an eye on assessing whether crime cost

estimates are sensitive to alternative manners of framing the questions (and perhaps to

educating respondents about actual crime rates and victimization risks).

Regarding bottom-up estimates, awards for pain and suffering in the United States (a)

exhibit great variation holding injuries constant, (b) tend to be higher in the United States

than other countries, (c) reﬂect decision making that is not subject to a personal budget

constraint, and (d) may implicitly incorporate attorney fees to be drawn from the damages

award. Whether these factors result in upwardly biased cost estimates (and the magnitude

of this bias) is difﬁcult to assess. However, one could imagine exploring sensitivity of cost–

beneﬁt calculations to, for example, discounting pain-and-suffering awards by one third to

account for attorney fees.

One potential extension of this research might be to apply contingent valuation survey

methods to nationally representative samples of individuals to gauge what the public believes

is the appropriate value to place on criminal victimization in public deliberations. Such a

survey could emphasize the dual purpose of these numbers for victim compensation and

for deliberation over criminal justice policy options. For example, respondents may be

queried about their maximum willingness-to-compensate crime victims out of a publicly

ﬁnanced victims’ compensation fund. Respondents could be reminded that payments occur

at taxpayer expense and that policy interventions that prevent crime would lower the tax

burden accordingly. Consequently, respondents would be further prompted with examples

14. Cohen attributed this methodological innovation to Phillip Cook.

15. See, for example, the exhaustive and careful cost–benefit analysis effort of the Washington State

Institute for Public Policy (wsipp.wa.gov/).

626 Criminology & Public Policy

Dom

ınguez and Raphael

illustrating how these numbers might ﬁgure into criminal justice policy deliberations. For

example, if the established victim’s compensation for auto theft is $2,000 and a program

promises to reduce auto thefts at a cost o f $40,000 per theft, then policy makers would take a

pass. In contrast, if the standard victim’s compensation for a sexual assault were $50,000 and

a program promised to reduce sexual assaults at a cost of $4,000 per incident prevented, then

policy makers would invest in this program. Setting aside issues of behavioral responses to

a guaranteed compensation schedule, framing the question in this manner could effectively

prompt respondents to think as policy makers. Not acting to prevent crime imposes a cost

on society through victim compensation. Taking action may either increase or decrease these

costs, depending on the costs of crime prevented relative to the set compensation amount.

Framed in this manner, such a contingent valuation survey would prompta respondent to

consider how much is too much to pay to prevent a speciﬁc type of victimization.

Given our concerns regarding the potential upward bias to cost-of-crime estimates, we

feel more comfortable with the use of these estimates to rank policy alternatives rather than

to pass overall judgments regarding whether speciﬁc proposals have net-positive beneﬁts. To

be sure, in some cost–beneﬁt calculations, the evidence of the net-positive beneﬁts or the

potential for net-positive beneﬁts is overwhelming and modest upward bias is unlikely to

alter conclusions regarding w hether the program pays for itself. For example, many of the

programs reviewed in Welsh and Farrington (2000) as well as many of the programs evaluated

and ranked by the Washington State Institute for Public Policy yield cost–beneﬁt ratios much

greater than one. As another example, tabulations of victimization costs generated during

the life course by criminally active youth are suggestive of enormous social savings from

early and effective preventive measures (Cohen, 1998; Welsh et al., 2008). Moreover, we

have conﬁdence that cost–beneﬁt calculations that use the more recent, more inclusive

cost-of-crime ﬁgures reviewed here that yield net-negative beneﬁts are actually identifying

ineffective interventions (see, for example, Lofstrom and Raphael, 2013, and Raphael

and Stoll, 2013, for cost–beneﬁt calculations of this sort pertaining to marginal prison

expansions). Nonetheless, cost–beneﬁt tabulations with ratios near one or that are being

determined primarily by an underlying effect of the intervention on homicide deserve

careful scrutiny. For example, researchers could follow the lead of Chalﬁn and McCrary

(2013) and chart the relationship between estimated cost–beneﬁt ratios and the assumed

VSL.

Another practical issue concerns the manner in which uncertainty in cost–beneﬁt tab-

ulations is conveyed when summarizing research and conveying ﬁndings to policy makers.

Of course, there are two sources of variation in cost–beneﬁt ratio estimates: the sampling

variance of the intervention effect size estimates and the sampling variance of the cost-

of-crime estimates. Typical cost–beneﬁt calculations by economists writing about criminal

justice tend to appear in conclusions and usually neglect to characterize the uncertainty

of this estimate. However, a serious effort to incorporate effect size sampling variance

into cost–beneﬁt calculations was undertaken by the Washington State Institute for Public

Volume 14

Issue 4 627

Research Article Role of the Cost-of-Crime Literature

Policy. Calculations in their rankings were based on the effect-size sampling distribution

rather than on a unique point estimate. Thus, when evaluating a program, they reported

both the cost–beneﬁt ratio (using a point estimate effect of a policy) as well as an estimate of

the chance beneﬁts will exceed costs measure (Washington State Institute for Public Policy,

2014).

Aside from measurement issues, we have argued that the results from cost–beneﬁt

analysis should be one of many criteria considered by policy makers and that there certainly

are many public choice problems where net beneﬁts may take a backseat to other legitimate

social objectives. Given the unequal distribution of the cost of criminal victimization and

anti-crime enforcement efforts, as well as the potential for perceived illegitimacy of the

criminal justice system to undermine various public institutions, equity considerations

deserve careful attention in all criminal justice policy choices.

These caveats aside, careful cost–beneﬁt analysis akin to what occurs in other policy

domains throughout the countr y should be one of the key drivers of criminal justice policy.

The cost-of-crime literature has done a great deal to facilitate the incorporation of this

normative paradigm and set of analytical tools into the criminal justice domain.

References

Abel, Richard. 2006. General damages are incoherent, incalculable, incommensurable, and

inegalitarian (but otherwise a great idea). DePaul Law Review, 55: 253–329.

Adler, Matthew D. and Eric A. Posner. 1999. Rethinking Cost-Beneﬁt Analysis. John M.

Olin Law and Economics Working Paper No. 72. Chicago, IL: University of Chicago.

Anderson, David A. 1999. The aggregate burden of crime. Journal of Law & Economics, 42:

611–642.

Anderson, David A. 2012. The costs of crime. Foundations and Trends in Microeconomics,

7: 209–265.

Arrow, Kenneth, Robert Solow, Paul R. Portnoy, Edward Leamer, Roy Radner, and Howard

Schuman. 1993. Report of the NOAA panel on contingent valuation. Federal Register,

58: 4601–4614.

Ashenfelter, Orley and Michael Greenstone. 2004. Using mandated speed limits t o measure

the value of a statistical life. Journal of Political Economy, 112: S226–S267.

16. This calculation of the “chance benefits will exceed costs” has two basic steps (Washington State

Institute for Public Policy, 2014). First, they determine a program effect size through a meta-analysis that

considers several reports and peer-reviewed evaluations of the policy intervention of interest. They

weight studies according to their credibility, which basically depends on the research design of the

study, among other attributes. They use the meta-analyses to calculate average effect sizes as well as

effect size variance. The second step consists of calculating the “chance will exceeds costs.” They

implement Monte Carlo simulations in which each simulation calculates a cost–benefit ratio from

drawing a normally distributed policy effect. The parameters of these distributions are those found in

the previous meta-analysis: the effect size and its standard error. Finally, after running 10,000

simulations, they calculate the “chance will exceeds costs” as the proportion of times the cost–benefit

ratio was greater than one.

628 Criminology & Public Policy

Dom

ınguez and Raphael

Austin, James. 1986. Using early releases to relieve prison overcrowding: A dilemma in

public policy. Crime & Delinquency, 32: 404–502.

Black, Dan A. and Thomas J. Kniesner. 2003. On the measurement of job risk in hedonic

wage models. Journal of Risk and Uncertainty, 27: 205–220.

Carson, Richard T. 2012. Contingent valuation: A practical alternative when prices aren’t

available. Journal of Economic Perspectives, 26: 27–42.

Carson, Richard T., R. C. Mitchell, W. M. Hanemann, R. J. Kopp, S. Presser, and P. A.

Rudd. 1992. A contingent valuation study of lost passive use values resulting from the Exxon

Valdez oil spill (No. 6984). University Library of Munich, Germany.

Chalﬁn, Aaron. 2013. The economic costs of crime. Working paper. Cincinnati, OH: Uni-

versity of Cincinnati.

Chalﬁn, Aaron and Justin McCrary. 2013. Are U.S. Cities Under-policed? Theory and Evi-

dence. NBER Working Paper 18815. Cambridge, MA: National Bureau of Economic

Research.

Cohen, Mark A. 1988. Pain, suffering, and jury awards: A study of the cost of crime to

victims. Law & Society Review, 22: 537–556.

Cohen, Mark A. 1998. The monetary value of saving a high risk life. Journal of Quantitative

Criminology, 14: 5–33.

Cohen, Mark A. 2005. The Costs of Crime and Justice. New York: Routledge.

Cohen, Mark A., Roland T. Rust, Sara Steen, and Simon T. Tidd. 2004. Willingness to pay

for crime control programs. Criminology, 42: 89–109.

Cook, Philip J. and Jens Ludwig. 2000. Gun Violence: The Real Costs. New York: Oxford

University Press.

Cooter, Robert D. and David E. DePianto. 2011. Damages for Incompensable Harms.

Working paper. Berkeley: University of California, Berkeley, Boalt School of Law.

DeLong, James V. 1981. Reply to Steven Kelman’s “Cost-beneﬁt analysis: An ethical critique.

AEI Journal on Government and Society Regulation, January/February: 33–40.

Diamond, Peter A. and Jerry A. Hausman. 1994. Contingent valuation: Is some number

better than no number? Journal of Economic Perspectives, 8: 45–64.

Friedman, Lee S. 2002. The Microeconomics of Public Policy Analysis.Princeton,NJ:Prince-

ton University Press.

Greenstone, Michael, Stephen P. Ryan, and Michael Yankovich. 2015. The Value of a Statis-

tical Life: Evidence from Military Retention Incentives and Occupation-Speciﬁc Mortality.

Unpublished manuscript.

Hanemann, W. Michael. 1994. Valuing the environment through contingent valuation.

Journal of Economic Perspectives, 8: 19–43.

Hausman, Jerry. 2012. Contingent valuation: From dubious to hopeless. Journal of Economic

Perspectives, 26: 43–56.

Hausman, J. A., G. K. Leonard, and D. McFadden. 1995. A utility-consistent, combined

discrete choice and count data model assessing recreational use losses due to natural

resource damage. Journal of Public Economics, 56: 1–30.

Volume 14

Issue 4 629

Research Article Role of the Cost-of-Crime Literature

Heaton, Paul. 2010. Hidden in Plain Sight: What Cost-of-Crime Research Can Tell Us About

Investing in Policy. Santa Monica, CA: Center on Quality Policing, the RAND Corpo-

ration.

Hersch, Joni. 1998. Compensating differentials for gender-speciﬁc job injury risks. American

Economic Review, 88: 598–607.

Hinsz, Verlin B. and Kristin E. Indahl. 1995. Assimilation to anchors for damage awards

in a mock civil trial. Journal of Applied Social Psychology, 25: 991–1026.

Kelman, Steven. 1981. Cost-beneﬁt analysis: An ethical critique. AEI Journal on Government

and Society Regulation, January/February: 33–40.

Kling, Catherine L., Daniel J. Phaneuf, and Jinhua Zhao. 2012. From Exxon to BP: Has

some number become better than no number? Journal of Economic Perspectives, 26:

3–26.

Levitt, Steven D. and Sudhir Alladi Venkatesh. 2000. An economic analysis of a drug-selling

gang’s ﬁnances. Quarterly Journal of Economics, 115: 755–789.

Linden, Leigh and Jonah E. Rockoff. 2008. Estimates of the impact of crime risk on property

values from Megan’s Law. American Economic Review, 98: 1103–1127.

Lofstrom, Magnus and Steven Raphael. 2013. Incarceration and Crime: Evidence from Cali-

fornia’s Public Safety Realignment. IZA Working Paper #7838. Bonn, Germany: Institute

for the Study of Labor.

Ludwig, Jens and Philip J. Cook. 2001. The beneﬁts of reducing gun violence: Evidence

from contingent-valuation survey data. The Journal of Risk and Uncertainty, 22: 207–

226.

MacCoun, Robert J. 2006. Media reporting of jury verdicts: Is the tail (of the distribution)

wagging the dog. DePaul Law Review, 55: 539–562.

Massey, Douglas S. and Nancy A. Denton. 1998. American Apartheid: Segregation and the

Making of the Underclass. Cambridge, MA: Harvard University Press.

McCarthy, Justin. 2014. Most Americans still see crime up over last year.

Gallup. November 21. Retrieved April 22, 2015 from gallup.com/poll/179546/

americans-crime-last-year.aspx.

McCollister, Kathryn E., Michael T. French, and Hai Fang. 2010. The cost of crime to

society: New crime-speciﬁc estimates for policy and policy evaluation. Drug and Alcohol

Dependence, 108: 98–109.

Miller, Tod, Mark A. Cohen, and Brian Wiersema. 1996. Victim Costs and Consequences: A

New Look. Washington, DC: National Institute of Justice, Ofﬁce of Justice Programs,

U.S. Department of Justice.

Nagin, Daniel S. 2001. Measuring the economic beneﬁts of development prevention pro-

grams. In (Michael Tonr y, ed.), Crime and Justice: A Review of Research, vol. 28. Chicago,

IL: University of Chicago Press.

Nagin, Daniel S., Alex R. Piquero, Elizabeth S. Scott, and Laurence Steinberg. 2006. Public

preferences for rehabilitations versus incarceration of juvenile offenders: Evidence from

a contingent valuation survey. Criminology & Public Policy, 5: 627–652.

630 Criminology & Public Policy

Dom

ınguez and Raphael

Piquero, Alex R. and Laurence Steinberg. 2010. Public preferences for rehabilitation versus

incarceration of juvenile offenders. Journal of Criminal Justice, 38: 1–6.

Portnoy, Paul R. 1994. The contingent valuation debate: Why economics should care.

Journal of Economic Perspectives, 8: 3–17.

Rajkumar, Andrew S. and Michael T. French. 1997. Drug abuse, crime costs, and the

economic beneﬁts of treatment. Journal of Quantitative Criminology, 13: 291–323.

Raphael, Steven and Michael Stoll. 2013. Why Are So Many Americans in Prison?NewYork:

Russell Sage Foundation.

Robbennolt, Jennifer K. 2000. Determining punitive damages: Empirical insights and

implications for reform. Buffalo Law Review, 50: 110–203.

Robbennolt, Jennifer K. and Christine A. Studebaker. 1999. Anchoring in the courtroom:

The effects of caps on punitive damages. Law and Human Behavior, 23: 353–373.

Saks, Michael J., Lisa A. Hollinger, Roselle L. Wissler, David Lee Evans, and Allen J. Hart.

1997. Reducing variability in civil jury trials. Law and Human Behavior, 21: 243–256.

Schelling, Thomas. 1984. Economic reasoning and ethics in policy. National Affairs, 63:

37–61.

Solow, Robert M. 1981. Reply to Steven Kelman’s “Cost-beneﬁt analysis: An ethical critique.

AEI Journal on Government and Society Regulation, January/February: 33–40.

Sugarman, Stephen D. 2005. A comparative law look at pain and suffering awards. DePaul

Law Review, 55: 399–434.

Thaler, Richard. 1978. A note on the value of crime control: Evidence from the property

market. Journal of Urban Economics, 5: 137–145.

Travis, Jeremy and Bruce Western. 2014. The Growth of Incarceration in the United States:

Exploring Causes and Consequences. Washington, DC: National Academies Press.

Viscusi, W. Kip. 1993. The value of risks to life and health. Journal of Economic Literature,

31: 1912–1946.

Viscusi, W. Kip. 2008. Value of life. In (Steven N. Durlauf and Lawrence E. Blume,

eds.), The New Palgrave Dictionary of Economics, 2nd Edition. London, U.K.: Palgrave

Macmillan.

Viscusi, W. Kip and Joseph E. Aldy. 2003. The Value of a Statistical Life: A Critical Review

of Market Estimates Throughout the World. National Bureau of Economic Research

Working Paper #9487. Cambridge, MA: National Bureau of Economic Research.

Washington State Institute for Public Policy. 2014. Beneﬁt Cost Technical Documentation.

Olympia: Author.

Washington State Institute for Public Policy. 2015. Beneﬁt-Cost Results. Retrieved April 9,

2015 from wsipp.wa.gov/BeneﬁtCost.

Welsh, Brandon C. and David P. Farrington. 2000. Monetary costs and beneﬁts of crime

prevention programs. In (Michael Tonry, ed.), Crime and Justice: A Review of Research,

vol. 27. Chicago, IL: University of Chicago Press.

Welsh, Brandon C., Rolf Loeber, Bradley R. Stevens, Magda Stouthamer-Loeber, Mark

A. Cohen, and David P. Farrington. 2008. Costs of juvenile crime in urban areas: A

longitudinal perspective. Youth Violence and Juvenile Justice, 6: 3–27.

Volume 14

Issue 4 631

Research Article Role of the Cost-of-Crime Literature

Yinger, John. 1997. Closed Door, Opportunities Lost: The Continuing Costs of Housing Dis-

crimination. New York: Russell Sage Foundation.

Zarkin, Gary A., Sheryl C. Cates, and Mohan V. Bala. 2000. Estimating the willingness to

pay for drug abuse treatment: A pilot study. Journal of Substance Abuse Treatment, 18:

149–159.

Zimring, Franklin E. and Gordon J. Hawkins. 1995. Incapacitation: Penal Reﬁnement and

Constraint of Crime. New York: Oxford University Press.

Patricio Dom

ınguez is a Ph.D. student in public policy at the Goldman School of Public

Policy at the University of California, Berkeley.

Steven Raphael is a professor of public policy at the Goldman School of Public Policy at

the University of California, Berkeley.

632 Criminology & Public Policy