The Deterrent Effect of Criminal Law Enforcement

Author(s): Isaac Ehrlich

Source: The Journal of Legal Studies , Jun., 1972, Vol. 1, No. 2 (Jun., 1972), pp. 259-276
Published by: The University of Chicago Press for The University of Chicago Law
School

Stable URL: https://www.jstor.org/stable/724093

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide
range of content in a trusted digital archive. We use information technology and tools to increase productivity and
facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected].

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
https://about.jstor.org/terms

The University of Chicago Law School and The University of Chicago Press are collaborating
with JSTOR to digitize, preserve and extend access to The Journal of Legal Studies

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 THE DETERRENT EFFECT OF CRIMINAL
LAW ENFORCEMENT

ISAAC EHRLICH*

I. INTRODUCTION

THE specter of crime has been haunting the U.S. and other western coun
throughout the last decade. The President's Commission on Law Enforcement
and Administration of Justice estimated the economic costs of reported crim
in the U.S. in 1965 at $21 billion (about four per cent of the reported nation
income),' which exceeded the economic costs of unemployment in that year.
A survey sponsored by the Commission revealed that crime is seen by t
public as the second most important domestic issue after race.2 Crime, broad
defined as all violations of law, has probably been on the rise throughout th
century if only because legislation has expanded beyond providing protection
for personal safety and property rights to cover virtually every aspect of bus
ness life.3 Crime control is far from costless. In 1965 public expenditure
police, criminal courts, defense counsel and "corrections" at the federal, stat
and local levels amounted to $4.6 billions. But while considerable resourc
are spent on enforcement of existing laws, and while new laws are being con
templated, very little is known in fact about the deterrent effect of the vario
measures of combatting crime or about the extent to which crimes of variou
kinds can be deterred.5
The idea that law enforcement-the apprehension and punishment of la
breakers-serves partly as a means of deterring future crimes by those a

* Assistant Professor of Economics, Graduate School of Business, University of Chicag

and Research Associate, National Bureau of Economic Research. This article is based o
my studies of participation in illegitimate activities which have been sponsored by t
National Bureau of Economic Research, under a grant from the National Science Founda
tion for research in law and economics. See Isaac Ehrlich, Participation in Illegitima
Activities: An Economic Analysis, 1970 (unpublished Ph.D. dissertation, Columbia Uni
Participation in Illegitimate Activities: A Theoretical and Empirical Investigation, Fe
1972 (unpublished paper at Center for Math. Studies in Bus. & Econ., Univ. of Chica
1 U.S. President's Comm'n on Law Enforcement & Admin. of Justice, Crime and
Impact-An Assessment 44 (Task Force Report, 1967).
2 U.S. President's Comm'n on Law Enforcement & Admin. of Justice, The Challen
of Crime in a Free Society 49 (1967).
3 Id. at 47.

4 U.S. President's Comm'n on Law Enforcement & Admin. of Justice, supra note 1, at 55.
5 U.S. President's Comm'n on Law Enforcement & Admin. of Justice, supra note 2, at
96, 141.

259

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
260 THE JOURNAL OF LEGAL STUDIES

prehended and by others is basic to crime control legislation, a

modern, but has been seriously questioned in the criminological lit
the past hundred years or so. An extreme view is that law enforce
no deterrent effect on offenders (at least those who commit seriou
essentially because offenders are very different from other hu
Indeed, both "subjective" and "objective" schools of criminology hav
ally linked criminal behavior to the offender's unique motivatio
turn, has been attributed to his unique "inner structure," to th
exceptional social and family circumstances, or to both.6 This view
by reference to some psychological studies of punishment,7 and so
cal investigations of the effectiveness of capital punishment.8 In co
article presents extensive empirical evidence that is consistent with
that law enforcement does deter the commission of crimes.
Even if law enforcement does not have any deterrent effect, it must surely
have some preventive effect since those who are apprehended and punished
by some form of detention are prevented, at least temporarily, from commit-
ting further crimes. Thus, the extreme view discussed above implicitly at-
tributes to law enforcement a preventive effect only. As will be demonstrated
later, however, that effect may be quite small in practice.
An alternative view, and one not necessarily incompatible with some tradi-
tional approaches in criminology, is that even if those who violate certain
laws differ systematically in various respects from those who abide by the
same laws, the former, like the latter, do respond to incentives: the opportun-
ities (costs and gains) available to them in legitimate and illegitimate pursuits.
This approach, which is in the spirit of the general framework used by econo-
mists to analyze individuals' occupational choices, does not ignore the role of
"respect for the law," penchant for violence, preference for risk and, in general,
individual propensities and preferences as determinants of participation in

6For an overview of the literature, see Donald R. Taft & Ralph W. England, Jr.,
Criminology (4th ed. 1964).
7 Following their experimental investigation of the effect of punishment on rats, Skinner
and Estes concluded that punishment is ineffective in eliminating undesired behavior, and
Skinner suggested that it be eliminated as a social instrument. B. Skinner, The Behavior
of Organisms (1938); Estes, An Experimental Study of Punishment, 57 Psychological
Monographs 263 (1944), cited in Barry F. Singer, Psychological Studies of Punishment,
58 Calif. L. Rev. 405, 413 (1970). This conclusion stood unchallenged for 20 years and
undoubtedly influenced the thinking of criminologists. However, recent studies in ex-
perimental psychology have firmly rejected Skinner and Estes' earlier conclusions. Barry
F. Singer, supra, at 413-414.
8 A well-known study concluded on the basis of a simple correlation of homicide rates
in contiguous retentionist and abolitionist states that "the presence of death penalty-in
law or in practice-does not influence homicide death rates." Thorsten Sellin, Homicides
in Retentionist and Abolitionist States, in Capital Punishment 138 (Thorsten Sellin ed.
1967). The findings presented in this paper constitute an indirect challenge to Sellin's
conclusion, although I do not discuss the specific effect of punishment by execution.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 CRIMINAL LAW ENFORCEMENT 261

criminal activities. But rather than rest on assumpti

minants of preference for crime, the economic appr
tempts to separate preferences from objective opp
what extent criminal behavior can be explained by t
given preferences.
It is a basic theorem of economics that an increa
jectively desirable) activity relative to other comp
preferences and real opportunities held constant-
away from that activity and toward activities that a
This theorem provides the analytical justification fo
theory of law enforcement to have general validity. B
thus indicates the direction of offenders' response to
an equally important question is the magnitude of
is perhaps the extent of the deterrent effect of law e
existence, that is principally at issue. This article
and empirical examination of this general quest
exposition of a model of participation in illegitim
few propositions regarding the relative effect of
activities on offenders' behavior. I then examine the
deterrent and the preventive aspects of law enforce
empirical evidence from crime statistics that bea
between law enforcement and crime.

II. THE ANALYTICAL FRAMEWORK

In spite of the diversity of activities defined as illegal, all such activitie

share common properties that form the subject matter of our general investig
tion. Any violation of the law can be conceived of as yielding an increase in
the offender's pecuniary wealth, in the pecuniary equivalent of his psych
wealth, or in both. But in committing the violation one also risks a reduction
in one's wealth, for conviction entails "paying" a penalty, acquiring a crimina
record, and other disadvantages. As an alternative to violating the law one can
engage in a legal wealth-generating activity, which may also be subject t
risks. The net gain in both activities is thus subject to uncertainty.

9 This approach is largely due to recent efforts by Belton M. Fleisher, The Economics
of Delinquency (1966); A. Smigel-Leibowitz, Does Crime Pay? An Economic Analys
1965 (unpublished M.A. Thesis, Columbia Univ.); Gary S. Becker, Crime and Punishment:
An Economic Approach, 76 J. Pol. Econ. 169 (1968); Isaac Ehrlich, The Supply of Il-
legitimate Activities, 1967 (unpublished paper, Columbia Univ.); Isaac Ehrlich, Particip
tion in Illegitimate Activities: An Economic Analysis, 1970 (unpublished Ph.D. disserta
tion, Columbia Univ.) [hereinafter cited as An Economic Analysis]; Isaac Ehrlich
Participation in Illegitimate Activities: A Theoretical and Empirical Investigation, Feb
1972 (unpublished paper at Center for Math. Studies in Bus. & Econ., Univ. of Chicag
[hereinafter cited as A Theoretical and Empirical Investigation].

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
262 THE JOURNAL OF LEGAL STUDIES

A simple model of choice between legal and illegal activities ca

lated within the framework of the usual economic theory of be
uncertainty.1' A central hypothesis of this theory is that if in
two activities were mutually exclusive, one would choose betwee
comparing the expected utility associated with each alone. The f
cussion elaborates on the determinants of such choice.
The gain associated with illegal behavior is a function of gross returns and
various costs. By gross returns is meant the value of the "output" of an
offender's activity: the direct monetary and psychic benefits he reaps from
committing offenses of a specific crime category (i). It is in general, and par-
ticularly in the case of crimes involving material gains, a function of the
offender's skill and ability to commit offenses, plus the inputs of his own time,
accomplices' services and other resources used in his illegitimate activity. In
addition, payoffs on most crimes against property and on some crimes against
the person depend in large measure on the amount of transferable assets and
other wealth available to potential victims of crime, as well as on the extent of
the self-protection the latter provide against victimization. We denote the re-
turns from illegitimate activity, net of the direct costs of purchased inputs, by
Yi.
The monetary and psychic costs associated with illegitimate behavior in-
clude both immediate and delayed costs. Immediate costs include those of
purchased inputs and the opportunity costs of the offender's own time, that
is, his returns from the alternative legitimate activity. Delayed costs are
typically those costs the offender incurs if he is apprehended and convicted of
crime (and include the prospect of losing his loot). One such cost is the penalty
that society imposes on convicted offenders in the form of a monetary fine,
a prison term, parole, or a combination of these. Unlike a monetary fine, which
is a unique quantity, the cost incurred in the case of, say, imprisonment is
indirect and specific to the individual. It is equal to the properly discounted
value of his opportunity costs of time spent in prison and psychic costs of
detention, net of any benefits obtained during the period of incarceration (in
extreme cases, which we shall henceforth disregard, the latter may even
dominate the former). An additional cost element that may be incurred by
the convicted offender is a reduction in his future stream of income in legiti-
mate activities due to the effect of his "criminal record" on job opportunities.

to Note that under this approach crime always "pays" if the variety of monetary and
psychic costs and returns offenders derive from crime are taken into account, for if the
offender is free to choose, and always acts to maximize his utility given his opportunities,
the fact of his engaging in crime indicates that utility is thereby maximized. This positive
approach may constitute the main difference between our analysis and some traditional
theories in criminology.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 CRIMINAL LAW ENFORCEMENT 263

We denote the discounted value of the future costs o

other losses by Fi.
Since only apprehended and convicted offenders
Fi, while other offenders are not, there is an uncer
net gain. If the offender is assumed to have a subjec
caught and punished for crime, Pi,1 then accord
analysis, his expected utility from engaging in crim
the utility he obtains in two general (and mutual
world": one in which his wealth increases by the net
ful" crime, Yi, and another in which his wealth afte
the loss of Fi following his apprehension and pun
the probabilities associated with each of these st
tively.12
The alternative (legitimate) prospect open to the individual, analogously
to the criminal prospect described above, comprises the legitimate return one
obtains when gainfully employed, Y1, and possible losses, D1, associated with
hazards typical of legitimate activities, such as unemployment. The expected
utility from the alternative prospect is the weighted average of the utilities
associated with the various levels of wealth contingent upon states such as
"employment" and "unemployment," with the weights being the subjectively
evaluated probabilities of these states: ti and 1 - p .
We have thus identified the basic set of opportunities affecting the decision
to participate in illegitimate activities: an individual's legitimate and criminal
returns Y1 and Y1, the probability and severity of punishment for crime Pi
and Fi, and the probability of and losses from legitimate hazards, such as un-
employment, 1i and D1 respectively. However, our discussion of the offender's
choice has proceeded on the strong assumption that legal and criminal activ-
ities are mutually exclusive. The decision to engage in crime is not, in fact,
inherently an either/or choice and offenders may in practice combine a num-
ber of legitimate and illegitimate activities or switch from one to another
throughout their lifetime. In addition, the (objective) probability of being
apprehended and convicted is not determined by society's actions alone; it
can be modified by deliberate actions of offenders. For example, an offender
can reduce the chances of his being apprehended or punished by spending

11 Other things being equal, the subjective probability Pi of an "average individual" is.
assumed throughout our empirical analysis to be a positive monotonic function of the
actual extent of law-enforcement activity by police and courts, and linearly related to the
fraction of offenders actually apprehended and punished for crime.
12 In practice, the criminal's prospects include more contingencies than the two just
mentioned, depending upon the form and extent of the punishment imposed or the re-
ward obtained. The analysis can easily be generalized to cover any finite number of
states.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
264 THE JOURNAL OF LEGAL STUDIES

resources on "covering" his illegal activity, by "fixing" policemen

nesses, by employing legal counsel, or, in general, by providing "sel
tion."

An attempt to attack this more comprehensive decision problem is

presented elsewhere.13 One of the results derived from the formal
that essentially the same set of variables identified in our preceding
as underlying an offender's decision to enter a specific criminal activ
defined in terms of marginal rather than total quantities, also deter
extent of participation. In particular, when earnings in both legi
criminal activities are not subject to strong time dependencies such
resulting from specific training or "learning by doing," many o
especially those who are risk avoiders-have an incentive to part
both activities, partly to hedge against the relatively greater risk in
a full-time pursuit of a criminal activity. Others, whose opport
criminal activities are relatively large or who are risk preferrers, h
centive to specialize in crime, or to participate in criminal activ
relatively full-time basis. The extent of offenders' specialization in
shown to be particularly influenced by the marginal penalty: the ad
punishment imposed for the "last" offense committed by an off
example, if the marginal penalty is zero, as is the case when the cou
concurrent imprisonment terms for a number of offenses, offenders
an incentive to participate in criminal activity on a relatively
basis.13a The analysis thus shows that for many offenders the optim
of participation in crime may be sufficiently large to involve the co
of several offenses in any given period. It also shows why many offe
to repeat their crimes even after being apprehended and punish
there were no systematic variations in preferences for crime o
toward risk from one period to another (these may, in fact, int
offender is likely to make the same choice of an optimal partic
crime if the opportunities available to him remain unchanged. In
plausible to assume that legitimate opportunities become scarcer
criminal opportunities in periods following conviction for crime, be
the criminal-record effect on legitimate job opportunities. Recidivis
has frequently been considered as evidence of offenders' irrationalit

13 See An Economic Analysis.

13a The analysis shows that what is required to avoid creating this incentive
ization in crime is that the marginal punishment be made to exceed the adva
offense over an alternative legitimate activity (see A Theoretical and Empiric
tion 36). This rephrases and modifies a well known argument that "the evil of
must be made to exceed the advantage of the offense." Jeremy Bentham, Theory of
Legislation 325 (Harcourt Brace Co. ed. 1931).

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 CRIMINAL LAW ENFORCEMENT 265

behavior or lack of self-control, is thus not necessarily a res

these: it may rather be the result of choice dictated by opportun

III. THE DETERRENT EFFECT OF LAW ENFORCEMENT

Two of the determinants of participation in crime that were identified in

the preceding discussion express the direct effect of law enforcement activity:
the probability and severity of punishment per offense (Pi and Fi respec-
tively). The effect of changes in law enforcement activity can be predicted
through consideration of the effects of shifts in each of these variables on the
incentive to enter a criminal activity and the extent of participation in such
activity.
An increase in either Pi or Fi with no change in the other determinants of
criminal activity may generally be expected to reduce the incentive to partici-
pate in crime because it increases the expected cost of punishment (PiFi).
If an offender had a neutral attitude toward risk, and was only interested in
the expected value of his wealth prospect, the magnitude of his response to a
one per cent increase in either Pi or Fi would be the same, for equal percentage
changes in each of these variables have the same effect on PiFi. But while equal
percentage changes in Pi and Fi have the same impact on expected wealth, they
may have quite different effects on the expected utility from crime if one has
nonneutral attitudes toward risk. The reason is partly related to the fact that
a one per cent increase in Fi always increases the variance of wealth associated
with the criminal prospect more than does a one per cent increase in Pi.14 It
follows that the incentive to enter a criminal activity would be reduced more
by a one per cent increase in Fi than the same increase in Pi if one had an
aversion for risk, and the converse would be true if one had a preference for
risk. Moreover, if an offender were a risk preferrer and yet partly engaged in
legitimate activity, then an increase in the average punishment per offense
might not deter his participation in crime: the latter might even increase.15
This result is not inconsistent with an assertion often made by writers on
criminal behavior regarding the low (and positive) effect of punishment on
the criminal propensities of some offenders. Such behavior is here seen to be

14 The variance of income accruing from an offense is V = P(1 - P)F2, and the differ-
ence between the effect on V of a one per cent increase in F relative to P is PF2[2(1 - P)
--(1 - 2P)], which is necessarily greater than zero.
15 The reason for this somewhat surprising result is that an increase in the potential
punishment facing those actually engaged in criminal activity does not isolate a pure
substitution effect in favor of alternative activities, but generates a wealth effect as well.
If an offender is a risk preferrer, the reduction in his prospective wealth in the event of
punishment can be shown to enhance his willingness to venture additional criminal activity,
which may partly or fully offset the opposite substitution effect.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
266 THE JOURNAL OF LEGAL STUDIES

consistent with preference for risk, and need not be interprete

of an offender's lack of response to incentives.
Since punishment always follows the commission of crime (an
payoff collected thereupon), and is often extended over a long p
(as in the case of imprisonment) following an offender's convic
counted (present-equivalent) monetary value of the delayed
is necessarily lower than, and less than proportionally related t
value if the discount rate is positive. Consequently, a one pe
in "severity of punishment" measured, say, as the actual i
period implies a less than one per cent increase in the discounte
earnings foregone while in prison. The relative increase in the
particularly small the larger the initial incarceration period or
offender's discount rate. And the lower the increase in the relevant cost of
punishment the lower its deterrent effect. Little response by offenders to
changes in the prescribed or enforced penalties (especially by offenders from
low-income brackets whose rates of discounting future income may be rela-
tively high) therefore need not indicate preference for risk or lack of response
to incentives, but may simply result from the fact that the costs of these
penalties are deferred.
As mentioned previously, an offender's probability of being apprehended and
punished is not entirely the product of law enforcement activity, but is modi-
fiable through his self-protective efforts. Indeed, it can be shown that an in-
crease in Fi-police and court activity held at a constant level-will generally
increase an offender's incentive to spend resources on self-protection (largely
independently of his attitudes toward risk). This may decrease the probability
of his being apprehended and punished which in turn may at least partly offset
the deterrent effect of Fi. In contrast, an increase in direct law enforcement
efforts to increase Pi (holding severity of punishment, Fi, constant) is not
expected to generate a similar effect on the incentive to provide self-protection
unless the offender is a risk preferrer.6" The observation that efforts to appre-
hend and convict offenders generally have a greater effect on the crime rate
than the adjustment of the severity of punishments meted out to convicted
offenders may therefore be explained, in part, by the interaction between self-
protection and crime, and may be consistent with neutral or negative attitudes
toward risk on the part of offenders.
We have so far discussed the deterrent effects of probability and severity of
punishment relative to each other. The general model outlined in the preceding
sections can also be used to derive predictions concerning the absolute extent
of these effects (as well as the effect of other opportunities), based on objective
considerations. To illustrate, let the total amount of time an offender spends

16 Mathematical proofs of these statements are given in An Economic Analysis, app. II.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 CRIMINAL LAW ENFORCEMENT 267

in both legitimate and criminal market activities ("work

constant level. If an offender specializes (is exclusively e
activity because, say, the opportunities available in such
dominate his alternative legitimate employment opportunit
a risk preferrer, then small variations in Pi and Fi might n
tive to specialize in crime and hence the number of offe
any given period. In contrast, if an offender allocates his wo
legitimate and criminal activities, then an increase in Pi
optimal allocation of time in favor of legitimate activities.
be shown that the responsiveness of criminal activity
portunities is greater the lower the amount of working tim
It follows that full-time or hard-core offenders may be less
magnitude by an increase in law enforcement activity than
casional offenders, not because of the former's unique mot
self-control, but simply because of their greater involvemen
Even though our predictions concerning the deterrent eff
ment activity are somewhat ambiguous when applied to spec
can be expected to hold with fewer qualifications when app
The aggregate supply curve of offenses can be conceived of
distribution of a function showing variations across person
the minimum expected monetary net gain that is sufficien
enter an illegal activity (their "entry payoffs"), as well as
extent of response of active offenders to changes in net ga
entry payoffs across persons reflect different attitudes to
different psychic net benefits. People with strong pref
penchant for violence may enter crime even when their
gains are negative. Others, risk avoiders or law abiders, ma
when the expected net gains are very high. A positive elast
ness) of the aggregate supply of offenses with respect t
gains due to a reduction in probability and severity of pun
be expected even if all individual supply curves were in
is, if all active offenders specialized in crime and did n
reduction at all-because the higher net gains would induce
offenders into criminal activity.

IV. DETERRENCE VERSUS PREVENTION

Our prediction of a negative correlation between law enforcement activity

and the aggregate amount of crime in the community (or crime rate) so far
rests on the assumed deterrent effect of the former. However, an increase in
the probability and severity of punishment by imprisonment (and to som
extent also by probation) may reduce the rate of crime in any given perio

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
268 THE JOURNAL OF LEGAL STUDIES

even if it does not generate any deterrent effect, simply because t

hended and convicted are, at least temporarily, prevented from
in) committing further crimes.
The preventive effect is limited in scope since it only affects
actually apprehended and punished, whereas the deterrent effect, i
affects all actual and potential offenders. The preventive effect is
another sense: it is generated only by certain types of punishment
volving the actual separation of offenders from potential victims-t
an economic cost on society as well as on the offender. Not only ar
costly to operate but the "payment" of a penalty in the form of an
incarceration is largely at the expense of his contribution to useful p
Fines, and to a large extent probation, spare society such costs. It i
to establish the independent deterrent effect of law enforcement bot
the validity of the approach of this paper and to determine the eff
of penal modes that have a deterrent effect only.
A systematic analysis of the preventive effect of punishment by
ment may be found elsewhere;17 to spare the reader the technical
analysis we present here only its implications. It can be shown that
ventive effect on the crime rate of a one per cent increase in (1) th
bility of apprehension and punishment for crime (defined as the fr
offenders apprehended and punished in a given period out of the t
of offenders at large), P, or (2) the number of periods convicted
actually spend in prison, T, is approximately the same after a pe
justment to the dynamic processes generated by the changes in P a
is equal to the ratio a - PT/(1+PT).18 Since a is positively relat
the expected length of imprisonment for particular crimes-the
effect of law enforcement may be relatively small in the case of le
crimes. This in turn establishes the important point that the preven
of law enforcement are in principle distinguishable from the deterre
Since as mentioned the preventive effects of P and T are expected to
equal in magnitude, significant differences between empirical estim
responsiveness of crime rates to changes in each of these variab
indicate the existence of an independent deterrent effect of law en

17 See A Theoretical and Empirical Investigation.

18 represents the "steady state" preventive effect of P and T when the rat
of the total population is negligible, or if Ti is not very large. Notice, howev
increase in P may have a temporarily greater preventive effect than an equal
increase in T because an increase in T does not have an immediate impact on t
of offenders at large whereas an increase in P does.
19 We derived such estimates from a cross-state regression analysis. Var
and Ti across states are likely to exhibit persistent and relatively steady diff
therefore the estimated elasticities of specific crime rates with respect to the
are likely to capture the "steady state" preventive effects referred to in the t

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 CRIMINAL LAW ENFORCEMENT 269

activity over and above its preventive effect. Sinc

PT/(1+PT) is bound to be lower than unity even if
empirical estimates of Pi for specific crimes are consid
even in the case of serious felonies), the existence o
law enforcement may be inferred, indirectly, if em
responsiveness of crime rates to changes in Pi and Ti a
than, unity.

V. EMPIRICAL ESTIMATES OF THE EFFECTS OF PROBABILITY AND

SEVERITY OF PUNISHMENT ON CRIME

Our basic theoretical predictions regarding the effect of probability and

severity of punishment for crime have been tested, so far, mainly via a cross
state regression analysis of variations in the rate of seven "index crimes
punishable by imprisonment (and listed in Table 1, infra) in the census years
1940, 1950 and 1960.20 The choice of this set of crimes as well as the cross
state context of our investigation were dictated by the availability of th
pertinent data in the Uniform Crime Statistics, the National Prisoner Statistic
and other publications of the U.S. Department of Justice. The basic statistical
model used in this investigation is a multiple regression equation of the form

In = ai + bi In Pi + b2i In T

-- ci In X ? c_2 In W + el In NW + [i21
where

(Qi) - rate
N per ofpopulation
state the ith crime category;
in year t. the number of offenses known

P! ( ) ratio of the number of commitments to state and federal

I prisons to the number of offenses known to have occurred in
the same state and year.22

20 The samples from 1960 include 47 state observations and those from 1950 include
46. The 1940 sample sizes vary between 36 and 43. Data are unavailable for auto theft
and rape in 1940 and 1950, and are incomplete in the case of murder and nonnegligent
manslaughter in 1940.
21 To obtain efficient estimates of the regression coefficients all the regressions under-
lying the results reported in Table 1 have been weighted by the square root of the popula-
tion sizes in each state, since a residual analysis of the corresponding unweighted
regressions showed a negative correlation between the absolute values of estimated residuals
and the population size.
22 The desired denominator of this ratio is the total number of offenders, rather than
the number of offenses known, in the same state and year-for which there is no informa-

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
270 THE JOURNAL OF LEGAL STUDIES

Ti = average time served in state prisons by offenders first re-

leased in year t.23
W = median family income in a state in year t.
X = percentage of families below one-half of W.
NW - percentage of nonwhites in the state population in year t.
?i -- a disturbance term.
In - natural logarithm.
This regression model is a partial attempt (limited by data exigencies) to
implement our general theoretical model in the explanation of crime variations.
The explanatory variables introduced in this equation represent (or approxi-
mate) the empirical counterparts of some of the major theoretical determinants
of criminal activity. We shall here consider a few issues involved in estimating
the effects of Pi and Ti.24
Our measures of crime rate, and of probability and severity of punishment

as well- (), P1i - and Ti, respectively--do not represent unbiased

estimates of the true levels of these variables across states, because most
crimes, including felonies, are heavily underreported, and because the felonies
investigated in this analysis are also punishable by means other than imprison-
ment. It has been our general assumption, however, that the relative variations
in these variables-random errors of measurement notwithstanding--do
represent an unbiased estimate of the relative variation in the true variables.
The general effect of errors of measurement in the explanatory variables of
a regression model, when using ordinary-least-squares (OLS) estimation tech-
niques, is to bias the absolute magnitudes of their associated regression co-
efficients toward zero. Thus, the estimated absolute values of the elasticities
of crime rates with respect to both Pi and Ti-bli and b2i respectively-may
generally be understated. In our special case it is possible, however, that the
effect of errors of measurement in the number of reported crimes would bias
the OLS estimates of the regression coefficient associated with Pi (bi) in a
negative direction. Recall that the number of offenses, Qi, appears both in the
numerator of the dependent variable and in the denominator of our estimate

tion available. However, the number of offenders and the number of offenses may be
linearly related across states.
23 A more efficient estimator of the offenders' expectation in year t of their prospective
length of imprisonment if apprehended and convicted of crime might be some weighted
average of the average time spent in prison by offenders released in year t as well as in
previous years (for which no information is available, in general). However, some scattered
evidence suggests that there has been very little variation over time in specific Ti's in
each state.

24 A full discussion may be found in An Economic Analysis or A Theoretical and

Empirical Investigation.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 CRIMINAL LAW ENFORCEMENT 271

of Pi. It can be shown that errors in Qi would bias the O

a negative direction if and only if the value of the true
(blj) were lower than unity in absolute magnitude.25 How
a spurious positive correlation between our estimates
probability of imprisonment: if the recovery of stolen m
geance played an important role in determining the
crime, as some studies show,26 a low probability of a
victing offenders would reduce the rate of reported crim
estimates of bli toward a positive value. A similar ar

regarding the correlation between - and Ti.

Another statistical problem is identifying the causal re

(-
N iand Pi. With a fixed level of resources devoted to
activity, the rate of crime and the probability of apprehension
ment for crime might be negatively related but the causalit
posite direction from the one predicted by our theoretical
crime rate might cause the probability of apprehension and
decline. (For example, in a riot, the probability of apprehen
vidual rioters, as well as for offenders committing other crim
considerably below their levels in normal times.) To cope wi
(as well as, indirectly, with the problem caused by errors of m
the reported crime rates) we used a two-stage least-squares
tion technique. Since in the second stage of the estimation proc
rates were regressed on combinations of variables that did
current estimates of crime rates or expenditures on law enforc
mates of bli and b2i derived via this method may be relatively
simultaneous-equation biases (i.e., those associated with the
problem referred to above) and the spurious negative corre
Q C
(
N )i
Q and Pi ( )i.27
The 2SLS estimates are
least-squares procedures.
ever, in the sense that
the rate of burglary ma

25 See An Economic Analys

26 See, e.g., U.S. President
note 1, at 17-18.
27 Due to data exigencies
1940 and the 1950 samples.
sion coefficients derived from

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 272 THE JOURNAL OF LEGAL STUDIES

rate of robbery. To derive efficient estimates of our regression coeffic

have also employed a simultaneous-equation method of estimating seem
unrelated regressions (SUR).28 Such estimates have been derived se
for the sets of crimes against property (robbery, burglary, larceny an
theft) and crimes against the person (murder, rape, and aggravated as
The results are presented in Table 1 along with our OLS and 2SLS e
The results of the regression analysis, despite the shortcomings of t
lend credibility to our basic hypothesis regarding the effect of law enf
on crime. The rates in all crime categories are found to vary inversely
estimates of the probability of apprehension and punishment by im
ment (Pi) and with the length of time served in state prisons (Ti).
estimates of the elasticities of offenses with respect to Pi and Ti (bli a
generally exceed twice the absolute values of their respective standard
tions, indicating that these coefficients are significantly different fr
The absolute magnitudes of the elasticities associated with Pi (bl1)
those associated with Ti (b2i) in regressions dealing with murder,
robbery, while the converse holds in the case of burglary in 1960. To
because of considerations noted earlier the OLS estimates of bli are lik
be biased in a negative direction relative to those of b21, especiall
regressions from 1940 and 1950, since the data on the distribution of
across states in these years may be somewhat less reliable than the
1960. In addition, the OLS estimates of bli and b2i may be subject t
taneous-equation biases. However, the results of the 2SLS and SUR estim
procedures strongly support the qualitative results achieved through t
procedure. Estimates of bli and b2i exceed twice the absolute values of
respective standard deviations and appear to be even higher in absolute
nitude than the corresponding OLS estimates. Again, the absolute v
bli, estimated via the 2SLS and SUR methods, exceed those of b2i in th
of murder, rape and robbery and fall short of b2i in the case of burg
larceny-the absolute differences exceeding twice the values of their re
standard deviations.29

28 See Arnold Zellner, An Efficient Method of Estimating Seemingly Unrelated Re-

gressions and Tests for Aggregation Bias, 57 J. Amer. Statistical Ass'n 348 (1962). Another
problem is that specific crimes may be complements (or substitutes) in the sense that an
increase in opportunities available in one crime would have a similar (or opposite) effect
on the rate of related crimes. For example, offenders charged with the crime of robbery
are often convicted of burglary; an increase in the penalty for burglary might then deter
participation in both crimes. To test such interdependencies we have introduced in the
regression equation the estimated probability and severity of imprisonment relating to
other crimes, in addition to Pi and Ti. The results indicate that robbery and burglary are
"complements" and that burglary and theft are "substitutes" in the sense defined above.
However, the effects of the additional variables were statistically insignificant and the
estimated coefficients reported in Table 1 were virtually unaffected by the introduction of
these other variables. See An Economic Analysis 86-89.
29 It should be recalled here that variations in Ti overstate the variations in the dis-

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 TABLE 1

REGRESSION ESTIMATES OF THE PARTIAL ELASTICITIES OF RATES OF SPECIFIC F

AND LENGTH OF IMPRISONMENT IN THE U.S.A. IN 194

Crime Category (i) 1940 1950 1960

Estimation Method OLSa OLSa OLSa 2

Variable Pi Ti Pi Ti Pi Ti

Robbery -0.9473 -0.

Burglary -0.4607 -0.2698 -0.4102 -0.4689 -0.5339 -0.9001 -0.72
Larceny -0.4131 -0.1680 -0.3477 -0.4301 -0.1331 -0.2630 -0.37
Auto-theft -0.2474 -0.1743 -0.407 -0.246
Murder and
non-negligent
manslaughter -0.5903 -0.2878 -0.3407 -0.1396 -0.852 -
Rape -0.5783 -0.1880 -0.896 -0.399 -
Aggravated
Assault -0.4239 -0.6036 -0.4791 -0.3839 -0.2750 -0.1797 -0.72

Source: See note 17 supra.

a (OLS) =Ordinary least-squares estimates.
b (2SLS) = Estimates derived via a two-stage-least-squares procedure.
c (SUR) = Estimates derived via a simultaneous-equation method of estimating seemingly unrelated regressions.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
274 THE JOURNAL OF LEGAL STUDIES

The fact that the absolute values of the estimated responsiven

certain or severe punishment of crimes against the person are n
the average than those associated with crimes against property
law-enforcement activity may not be less effective in combatting
and passion than crimes involving material gains. However, sinc
tive effect of law enforcement activity is expected to be greater
the relatively more serious crimes (involving higher value
crimes against the person generally belong in this category, the d
of law enforcement activity against these crimes may be lower
effect indicated by estimates of the relevant bli and b2i.
It is difficult to determine accurately, on the basis of the ava
what extent the estimated values of bli and b2i are attributable
effect of imprisonment, ai = PiTi/(1 + PiT1), because, as has
pointed out, the absolute values of our estimates of Pi may n
estimates of the true fractions of offenders apprehended an
However, since our 2SLS and SUR estimates of bli (correspon
significantly different in the cases of murder, rape, robbery, b
larceny from the estimates of b21 (corresponding to Ti)-some o
mates approaching or even exceeding unity in absolute value-the
deterrent effect of law enforcement activity appears to be confi
tioned earlier, the preventive effects of Pi and Ti are expected
identical and-in view of the available information regarding
of Ti and reasonable estimates of Pi in the U.S. in 1960-consi
than unity.

VI. CONCLUSION

The approach one takes in analyzing the deterrent effect of law enforcement
on crime is inseparable from the approach one takes in analyzing the determi-
nants of criminal behavior itself. The basic thesis underlying this article is that
offenders, as a group, respond to opportunities (costs and gains) available to
them in legitimate and criminal activities in much the same way that those
engaged in strictly legitimate activities do as a group. This does not necessarily
imply that offenders are similar to "other people" in all other respects or that

counted cost of imprisonment (Fi) and therefore our estimates of b2i necessarily under-
state the true elasticities of the various crimes with respect to the appropriately measured
severity of punishment (Fi), especially in the case of crimes punished by long imprison-
ment such as robbery and murder. Thus, given the validity of our 2SLS and SUR estimates
of bli and b2i, we may conclude unambiguously that burglars and thieves are risk avoiders.
Whether other offenders are risk preferrers cannot be determined unambiguously, how-
ever, without specific knowledge of offenders' discount rates: the higher the latter, the
larger would be the absolute values of our revised estimates of the elasticities of specific
crimes with respect to Fi.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
 CRIMINAL LAW ENFORCEMENT 275

the extent of their response to incentives is the same. I

gests that the extent of individual offenders' response t
negatively with the extent of their specialization in ille
so may not be uniformly high or low. We do emphasize,
opportunities, including the level of law-enforcement ac
the extent of an offender's participation in illegitimate
directly, also the extent of his response to incentives.
The results of our regression analysis of variations
index crimes across states in the U.S. are consistent with this basic thesis. In
spite of the shortcomings of the crime statistics used, the imperfect estimates
of probability and severity of punishment (and other theoretical constructs)
used in the regression analysis, and the somewhat stringent econometric speci-
fication of functional relationships, the signs and estimates of the responsive-
ness of offenses to changes in Pi and Ti-and in other variables-exhibit a
remarkable consistency with the theoretical predictions, as well as with one
another, across independent samples. In particular, and contrary to some
popular arguments, the absolute magnitudes of the elasticities of specific
crimes with respect to probability and length of imprisonment are consistent
with the hypothesis that law-enforcement activity has a deterrent effect on
offenders that is independent of the preventive effect of imprisonment. More-
over, the elasticities associated with crimes against the person are not found
to be lower, on the average, than those associated with crimes against prop-
erty.
The evidence reported in this article is also highly consistent with evidence
on the response of white and nonwhite offenders to changes in probability and
length of imprisonment that was derived from a regression analysis of varia-
tions in whites' and nonwhites' imprisonment rates for all felonies across
states in the U.S. in 1960.30 Still, the limitations of the crime statistics used in
our empirical investigation, and the absence of sufficiently accurate statistics
on specific determinants of criminal activity, cast some doubt on the validity

30 Two-stage least-squares (indirect) regression estimates of the elasticities of the ag-

gregate crime rates of white felons with respect to P and T (measured by the ratio of all
commitments to state prisons for felonies to the total number of reported felonies, and
by the length of time served in state prisons by felons first released across states) are
found to be - 1.2009 and - 1.236, respectively. The corresponding estimated elasticities
of the aggregate crime rates of nonwhite felons with respect to these variables are - 0.4860
and - 1.0015, respectively (all of the estimated elasticities generally exceed twice their
standard deviations in absolute value). These results are particularly significant in view
of the fact that the dependent variables in the regressions used for calculating the relevant
elasticities are commitment rates of white and nonwhite offenders in the respective white
and nonwhite population in each state rather than crime rates. Thus, even OLS estimates
of the elasticities associated with P and T are free from a potential spurious negative
correlation between P and the dependent variables due to errors of measurement in the
number of reported crimes. The statistical model employed in these regressions and alterna-
tive OLS regression estimates are discussed in An Economic Analysis.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms
276 THE JOURNAL OF LEGAL STUDIES

of our regression estimates of the elasticities of crime rates wit

probability and severity of punishment. A more definitive
these elasticities must await further data and research.
I should emphasize that even though the results presented in this article
may be used to derive indirect estimates of the effectiveness of law enforce-
ment in reducing crime and its net economic returns (some partial estimates
indicate that at least in 1960 law enforcement activity against felonies did
"pay"31) they do not have direct implications regarding the relative useful-
ness of law-enforcement activity as a social instrument. Since the rates of
all felonies, especially crimes against property, are found to be positively
related to the degree of a community's income inequality,32 equalization of
training and earning opportunities across persons may be an effective method
of crime control also, independently of ethical considerations. Whether it
would pay society to spend more resources in order to enforce existing laws
would then depend not only on the effectiveness of such expenditures in
deterring and preventing crime, but also on the payoff from alternative
methods of combatting crime.

31 An Economic Analysis or A Theoretical and Empirical Investigation.

32 Ibid.

This content downloaded from

152.118.24.31 on Sat, 30 Nov 2024 10:01:36 UTC
All use subject to https://about.jstor.org/terms