Estimating a Crime Equation in the Presence of Organized Crime

ELSEVIER
Estimating a Crime Equation in the Presence
of Organized Crime: Evidence from Italy
RICCARDO MARSELLI
Istituto Universitario Navale, Napoli, Italy
E-mail:rmarselli @synapsis. it
M A R C O VANNINI
Universitgt di Sassari, Sassari, Italy and CRENoS
We estimate a crime equation using a panel dataset of Italian regions for the period
1980 to 1989. Four different crimes are considered: murder, theft, robbery, and fraud.
To take account of the presence of criminal organizations--a salient feature of the
Italian context and a relevant factor usually disregarded by the empirical literature--we
exploit the panel structure with unobservable individual components. Our results suggest that: (i) the probability of punishment is relatively more effective than both the
severity of punishment and the efficiency of police authority in deterring crime; (ii)
among the variables representing the opportunity costs of participating in illegal activities, the rate of unemployment, the value of public works started by government, and
the proportion of people employed in the service sector have a significant effect; (iii)
for three types of crime the regional unobservable component is correlated with the
regressors; (iv) spillovers from drug consumption to theft are substantial; and (v) with
the exception of fraud, the results are in contrast with the predictions of the standard
economic model of crime. © 1997 by Elsevier Science Inc.
I. Introduction
Despite its remarkable features, Italy's criminal activity has received little attention and
remains largely neglected by the economics of crime literature, which sprung from the
The paper is the result of a joint effort. For formal purposes, however, Sections I, III, a n d V may be attributed to
M. Vannini, while Sections II and IV to R Marselli. We would like to thank M. D'Antonio, A. Graziani, L. Guiso, R.
Cooter, E. Eide, P. Zweifel and seminar participants at CRENoS (University of Cagliari), GREAT (University of Napels),
IGIER (L. Bocconi University, Milan), Eastern Economic Association (New York), European Economic Association
(Prague), European Association of Law a n d Economics (Gerzensee), European University Institute (Florence) for their
comments and P.L. Capellino for providing assistance in data collection. We would also like to thank two anonymous
referees of this Journal for their constructive suggestions a n d comments. The usual disclaimer applies.
International Review of Law a n d Economics 17:89-113, 1997
© 1997 by Elsevier Science Inc.
655 Avenue of the Americas, New York, NY 10010
0144-8188/97/$17.00
PII S0144-8188(96)00060-9
90
Estimating a crime equation
seminal contributions by Becker (1968) and Ehrlich (1973). 1 Indeed, if one goes
through this copious literature one will hardly find any allusion to Italyz or to any other
European country with a civil law legal system. 3 The large majority of studies considers
common law countries, like the United States and Canada [Taylor (1978); Schmidt and
Witte (1984); Trumbull (1989) ], Australia [Whithers (1984); Lewis (1987) ], and United
Kingdom [Wolpin (1978,1980); Field (1990); Reilly and Witt (1992)]. This is rather
surprising for a variety of reasons, not least because the phenomenon in question seems
extraordinary and unique, both quantitatively and qualitatively.
In our opinion, this neglect needs remedying not only for the sake of the geographic
completeness of the empirical evaluation of the economic analysis of crime, but also for
a better understanding of how crime and punishment work in a meaningfully different
environment.
The salient features of Italy vis-~t-vis the most studied countries seem to be: (i) a large
variability of crime rates across time and space; (ii) a pervasive presence of organized
crime with strong regional roots; 4 (iii) a higher frequency of offenses involving material
gains relative to those related to passion or ethnic/racial conflicts; (iv) a criminal justice
system based on a codified criminal law, where the judge is not a lawmaker--as in many
common law countries--and in which the sentencing process is strictly predetermined
by the penal code. 5
Each of these distinguishing characteristics, in principle, may affect both the specification and estimation of a crime equation based on the economic model of crime
(EMC). As is well known, this model reflects the presumption 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," and "links formally
the theory of participation in illegitimate activities with the general theory of occupational choices by presenting the offender's decision problem as one of an optimal
allocation of resources under uncertainty to competing activities both inside and outside the market sector, rather than a choice between mutually exclusive activities"
[Ehrlich (1973), p. 522]. Within this framework "some individuals become criminals
because of the financial and other rewards from crime compared to legal work, taking
account of the likelihood of apprehension and conviction, and the severity of punishment. The amount of crime is determined not only by the rationality and preference of
would-be criminals but also by the economic and social environment created by public
policies, including expenditure on police, punishments for different crimes, and opportunities for employment, schooling, and training programs" [Becker (1993), p.
390].
Interesting though it may be, the estimation of crime rate determinants with Italian
data is by no means straightforward, because the main categories of reported offenses
reflect the operation of both individual and organized crime. Moreover, in the absence
of official data 6 on the spatial distribution and internal structure of these criminal
conglomerates, it is impossible to include a direct measure of this factor in the equations to be estimated. It becomes necessary to develop an estimation strategy capable of
resolving or reducing the problems associated with unobserved explanatory components.
Our objective here, therefore, is 2-fold: (i) to assess empirically the relevance of crime
determinants in a context marked with some invaluable features for this kind of exercise and (ii) to acknowledge, attempting this evaluation with Italian regional data,
the existence and pervasiveness of organized crime. We consider four categories of
crime--murder, robbery, theft, and fraud--which represent, on average, more than
MARSELLIAND VANNINI
91
TABLE1. Reported crimes*
First-degree
murder,
attempted
murder, child
murder
Manslaughter
Personal injuries
Personal offences
Crimes against
family and
public
morality
Theft
Robbery,
extortion,
and kidnapping
Fraud
1951-1960
(a)
1961-1970
(b)
1971-1980
(c)
1981-1987
(d)
3.8
12.1
304.6
82.7
2.6
11.9
282.6
56.9
3.3
11
205.1
40.3
4
8.6
169.2
36.8
52.1
540.8
55.1
748.7
36.2
2290.2
25.1
2287.6
6.1
81
5.7
68.5
24.8
60.7
70
84.9
(d)/(a)
(A%)
5.26
-28.92
-44.45
-55.50
(d)/(b)
(A%)
(d)/(c)
(A%)
53.84
-27.73
-40.12
-35.32
21.21
-21.81
-17.50
-3.50
-51.82
-54.45
3 2 3 . 0 0 205.54
-30.66
-0.10
1 0 4 7 . 5 0 1128.10 182.20
4.81
23.94
24.20
Source: La criminalit/tattraversole stadstiche,ISTAT- Note e relazionin.3. 1989.
*Mean values;ratios per hundred thousand residents.
75% o f all r e c o r d e d crimes a n d have recently r e a c h e d very alarming rates o f growth (see
Table 1).
T h e r e m a i n d e r o f the p a p e r p r o c e e d s as follows. Section II outlines the m a i n characteristics o f Italy's crime rates. Section III reviews the basic version o f the e c o n o m i c
m o d e l o f crime a n d its empirical implications a n d describes the data. Section IV illustrates the estimation p r o c e d u r e a n d presents the results. Section V summarizes a n d
c o n c l u d e s the paper.
H. A Few Stylized Facts About Italy's Crime Rates
O n e o f the striking characteristics o f Italy, relative to most countries investigated, is the
c o n s i d e r a b l e disparity o f crime rates t h r o u g h time a n d across regions. O n the basis of
the latest official records, the diffusion o f crime in the "Bel Paese" can be d e p i c t e d as
in Figure 1. Two aspects are worth noting. First, the intensity o f the p h e n o m e n o n across
regions varies dramatically a c c o r d i n g to the offense considered: A l t h o u g h the total
crime rate identifies very few regions with a low p e r capita n u m b e r o f offenses, when
specific offenses are e x a m i n e d , the picture gets m o r e complex. F u r t h e r m o r e , only in
the case o f r o b b e r y a n d crimes against p r o p e r t y d o two or m o r e c o n t i g u o u s regions
seem to display similar high rates. Second, the r a n k i n g o f regions d e p e n d s on the type
o f crime selected. A l t h o u g h U m b r i a (Central Italy), for instance, ranks very high with
respect to crimes against p e r s o n a n d very low for b o t h r o b b e r y a n d crimes against
property, the opposite h a p p e n s to Campania.
A f u r t h e r insight into the p r o p e r t i e s o f these crime rates is p r o v i d e d in Figures 2a a n d
2b, which show that cross-sectional variation d o m i n a t e s i n t e r t e m p o r a l variation in the
92
Estimating a crime equation
Figure 1. Crime Rates per 100,000 Inhabitants, Italy 1991
All offences
r"--I<2,000
2,000 - 3,000
3,000-4,000
i i 4,0005,000
I >5,000
-
IrMB Um~rib
I
I
<105
I05-2i0
210-315
315-420
>420
Robbely
~
<32
~
32-64
I
64-96
I
g6-120
I
~
~
Aga,mstproperty
< 1,500
1,500- 3,000
3,000-3,500
IIIIII3,500-4,000
~ >4,000
~
I
~
FIG. 1. Crime rates per 100,000 inhabitants, Italy 1991.
> 120
Theft
['-'-'7< 1,230
~ 1,2002,400
I 2,400-3,000
III 3,600-4,800
i >4,800
MARSELLI AND VANNINI
93
d e p e n d e n t variable. Figure 3 contrasts the crime rates at the beginning and at the end
of the sample period: The regions located in the u p p e r right (lower left) quadrant have
experienced higher (lower) than average crime rates during the whole period, whereas
the position of those located in the u p p e r left (lower right) quadrant has deteriorated
(improved). Excluding fraud, the regions are clearly clustered, 7 and their relative position in the crime rate hierarchy did not change m u c h during the sample period.
Furthermore, the Southern regions (with the exception o f Abruzzo, Molise, and Basilicata) are generally located in the u p p e r right quadrants. W h e n we consider murder,
four regions (Sicilia, Campania, Calabria, and Puglia) clearly dominate: What they have
in c o m m o n is certainly a massive presence of criminal organizations, which operate
nationwide but exert their strongest influence within those areas where they historically
emerged. 8 Table 2 presents some recent data on the relevance of organized crime
m u r d e r in relation to total murder.
In the absence of official data on the n u m b e r of people involved in criminal organizations, we can only speculate on the impact and the influence of such organizations
from observed crime rates. As d o c u m e n t e d in Figure 4, the average crime rates by
region change quite substantially as we move from fraud and theft to robbery and
murder; that is, as we move towards offenses increasingly associated with organized
crime. In the case of m u r d e r we find four peaks that dramatically differentiate Southern
regions from the rest of the country.
III. The Economics of Crime: Theory and Empirical Counterparts
In the literature on the EMC originating from Becker (1968) and Ehrlich (1973), a
potential offender acts as an expected-utility maximizer who allocates his time a m o n g
competing activities, legal and illegal, with uncertain consequences. It is worth recalling
that people are n o t assumed to have the same "taste for crime" and that their ethical
attitudes matter: Everybody has his own vulnerability threshold defined positively by his
morality or negatively by his proclivity to commit a crime. However, individuals respond
to incentives, and if the rewards structure associated with an action changes, their
choices are also likely to change. Therefore, unlike most criminological studies that try
to explain deviant behavior on the basis of social privation or deviant factors (psycho-
TABLE2. Organized crime murder
Regions
January-July
1992
Percentage of
total murder
January-July
1991
Percentage of
total murder
A%
163
147
108
18
18
454
52.2
63.1
58.4
15.3
5.4
38.5
59
105
70
4
33
271
25.8
58.7
38.9
5.7
11.8
28.9
176.3
40.0
54.3
350.0
-45.4
67.5
Sicilia
Campania
Calabria
Puglia
Other regions
Italy
Source: Home Office.
94
Estimating a crime equation
Figure 2a Reported
Crimes-coefficientsof variation of regional rates by year
Murder
t,40
1,20t
1,001
1112
-o'g7
-
0.94
i
1,14
0"7
__
0.94
0'97
--
I
tg68
lgeg
O,40
o,2O
°1111111
O,00
lg~O
t981
1982
lg83
1984
1985
1g~8
1987
Robbery
1,60
]
1,40
13~
1,35
J
1,25
1,20
1.24
1,17
1,16
1,17
18G8
1980
1.03
1,00
O,eOi
0,80
0,411
0,20
1980
1981
lge2
lg83
1964
1g~15
1~
1987
FIG. 2. (a) Reported crimes; coefficients of variation of regional rates by year.
logical a n d / o r physiological) unique to criminals, ~ the economic approach takes personal tastes as given and emphasizes the role of competing legal and illegal opportunities in determining the offender's choice.
The theory predicts that a rational agent will embark on some illegal activity as long
as the marginal return from crime exceeds the marginal return from legal occupation
by more than the expected value o f the penalty, and that the a m o u n t o f time devoted
to illegal activities decreases unambiguously as the probability of apprehension increases. As to the effect on individuals' decisions of a change in initial wealth or of the
stiffening o f sanctions, it depends on the individuals' attitude toward risk.
The plausible conclusions of the elementary model of participation in illegitimate
activities do not carry over to more sophisticated and reasonable versions of the model.
More precisely, as shown by Heineke (1978), as soon as non-monetary aspects of both
the time allocation problem and the penalty are incorporated into the analysis, "it is not
possible to establish the sign of any one of these comparative static derivatives unless
one is willing to make m u c h stronger assumptions about the preference of offenders" (p. 25).
Despite the unsettled micro-theoretic underpinnings, the economic approach to
crime has provided guidance for a host of studies intended to explain the observed
MARSELLI
AND VANNINI
95
0,70
0,80
0,52 0,55
O'iiHili
0,50
0,47
1
0,30
0,20
0,10
0,00
:
1880
1gel
t982
,
lge3
1084
lgG5
:
1866
1987
lEG8
1989
Fnlud
0,~0
0.50
0,47
0,40 0,3~
0,40
O,.oIl,oi o,i °1.li°lli
0,00
.
lg80
1961
1982
.
tgQ3
.
.
1084
,
lgQ5
lgG8
1987
1988
Ig~
FIG. 2a. Continued.
(aggregate or individual) frequency of offenses in terms of factors such as the probability and severity of punishment, legal and illegal opportunities, and tastes. The
empirical counterparts o f these theoretical constructs, mostly used in applied works, 1°
are as follows: an objective measure of the probability of punishment, like the fraction
of all offenders convicted, to approximate the subjective probability of detection; the
average time served in prison to reflect the cost of punishment; the level of consumption and some indicator of labor market conditions to measure legal and illegal opportunities; and characteristics like race, age, education, geographic location, and sex to
control for attitudes toward crime.
To analyze illegal behavior in Italian regions we selected four categories of crimes
reported to the judicial authorities: murder, theft, robbery, and fraud. The first category, which refers to crimes against the person, has been restricted to those types
of offenses (first-degree murder and attempted murder) that are cause for great social concern and that often culminate in extremely violent episodes. The last three
are forms of property crime, although robbery cannot easily be labeled as purely property or purely personal. The focus, in any case, is on so-called crimes of profit, i.e.,
offenses committed to get hold of somebody's goods either directly, as in the case of
Estimating a crime equation
96
Figure 2b Reported Crimes - coefficients of variation of annual crime rates by region
Mun4er
0,6
o,51
0,5
0,46
[
0,4~
0,42
0,4
0,3
~2g
O,2 0,16
0.16
OA
ill l i
....
0
nob~-y
0,6
/
0,44
0,44
0,4
0,5 [
O,3 ]
0,33
0,31
0,30
02"/
0 36
'
034
•
0,28
0
' 34
0,27
0,23
0"2 0,18
0,18
0,19
0,14
0,20
0,14
0,1
0
FIG. 2. (b) Reported crimes; coefficients of variation of annual crime rates by region.
housebreaking and robbery, or indirectly, as in the case of blackmailing and kidnapping.
For each of the selected categories of crime we constructed three deterrence variables
(UNKNOWN, PROBABILITY, and SEVERJTY) intended to capture the effect of the
criminal justice factors potentially affecting the regional level of crime. UNKNOWN is
the ratio of crimes c o m m i t t e d by persons unknown to all recorded crimes in a given
category and takes care of the deterrence effect s t e m m i n g from the criminal investigation efficiency of the local police force and from their knowledge of the local underworld; PROBABILITY is the ratio of the number of offenders convicted to the total
number of offenders recorded in a given category of crime]I; and SEVER/TY is the
average time to be served in prison according to the final j u d g m e n t in a given category
of crime. The coefficients of variation of these variables are shown in Figure 5. Not
surprisingly, due to the characteristics of the sentencing process, a m o n g the deterrence
factors severity seems to vary the least across regions and time.
As for the e c o n o m i c factors affecting crime, we considered both the level of regional
real consumption per capita (CONSUMUITON) and the regional rate of u n e m p l o y m e n t
(UNEMPLOYMENT). The level of consumption is usually included in this kind of study
to capture the effect of the business cycle, whereas u n e m p l o y m e n t enters as an indirect
MARSELLI
AND
97
VANNINI
T~'t
0,5
0,45
0,40
0,4
0,35
0,3
0,29
0,26
0.25
0.22
0,22
0,18
0,2
0,15
0,14
0,18
0,15
O.lO
0,15 0,14
o.]1
O.lO
:
0.1
0,O5
0
Fmud
0,7
0,64
0,0
0,5
11,4
0,35
0,2
'
0.17
0,12
o,,
o
oilOio
0,49
•
...
MMM
•
FIG. 2b. Continued.
measure of the opportunity cost from crime. It has been noted, however, that if the rate
of unemployment is the only economic indicator among the explanatory variables, it
ends up channeling mainly the effect of the business cycle [Freeman (1983)]. 12 In
principle the relation between crime and the business cycle/consumption is indeterminate and depends on the specific type of crime considered. With regard to property
crime, for instance, Field (1990) points out three sorts of effects: (i) an opportunity effect
(with business booming, and consumption growing, opportunities and returns for
crime increase simultaneously with both the number and the value of goods); (ii) a
lifestyle effect (consumption growth induces a change in routine activities in a direction
that favors potential criminals); and (iii) a motivation effect (during booms there is more
room for the lawful acquisition of goods, and vice versa). Opportunity and lifestyle
effects have a positive impact on crime rates, whereas the influence of the motivation
effect is negative; when the cycle is proxied by the rate of unemployment, the same
effects are operative with reversed signs.
In addition to these variables that regularly feature in the empirical literature on the
EMC, we also considered a set of socioeconomic and demographic variables. Public
works (PBWORKS) has been introduced as an indicator of social privation, because it
represents government efforts to invest in regional social fixed capital. Although we
would expect a pr/or/a negative relationship between this variable and the regional
Estimating a crimeequation
98
F,gure 3 Part 1.
60
THEFT
I~z
1989
5O
4O
• pug
epie
3O
20
10
•
• Iom
°t~
~
•emr • ~lm
yen
TA
•taa
I~
ea~ • alx
0
5
10
15
20
25
30
35
40
45
50
1980
0,25 ........................................................................................................................
MURDER
.ca,
1989
O,2
*sic
0,15
0,1
0sllr
~
0,05
Ocllm
,,beslae i o n 6 ~ A
0
0,02
0,04
0,06
/
+
0,08
0,1
0,12
0,14
0,16
0,10
1980
FIG. 3. Changes in crime rates, 1980-1989.
crime rate, we should emphasize that it may be proxying several influences, the most
important of which, we believe, is the growth of rent- (and profit)-seeking activities
associated with public spending programs. The number of employed in the service
sector relative to the total number of employed (SERV/CES) in this context acts as a
measure of the opportunities linked to white collar crimes, but it should be stressed that
it may capture other important factors affecting crime that are positively correlated
with the growth of the service sector, like income inequality. 13 Social security benefits
(BENEFITS) and average monthly salary (WAGE) represent further indicators of opportunities associated with participation in legitimate activities. Finally, to take account of
the widespread opinion that the younger and the less educated the population the
greater the crime rate, we defined two more variables, representing, respectively, the
percentage of all males in the age group 14 to 29 (YOUTH) and the number of students
who have completed secondary and high schools relative to the population (EDUCATION). The coefficients of variation of all these variables over the sample period are
presented in Figure 6.
M~kRSELL1 AND VANNINI
99
[~jum 3 PartZ
4RO68ERY
i INJ
cam
3,5
3
2,S
•
m~c
2
1,5
~z
*
pug
1
.
tom
TA
*~
0,2
1,2
.
.
FRAUD
.
••
0.4
.
.
p~
0,8
.
.
.
.
0,8
.
.
I
.
.
1,2
.
.
. j¢
.
.
1,4
lf~0
1,6
.
1.1
cam
I
0,9
0,B
•
taz
0.7
• pug
Ig
* nw
o.e
0,5
0,4
*t01
* o~
• ~"
*~
.
•
0,3
0,15
* unlb
;TA
0,2
0,25
tim
0,3
0,35
FIG. 3.
0,4
0,45
0.5
0,55
0,6
1~0
0,~
Continued.
IV. The Empirical Procedure and the Estimation Results
We based our estimation on pooled data concerning 19 Italian administrative regions, a4
Because we were working with only 10 annual observations for each region, the timeseries analysis was necessarily limited, and we dealt primarily with the question of
whether there existed a homogeneous structure capable of adequately representing the
phenomenon of illegal behavior across different regions.
As already mentioned, the most striking feature of the Italian context is the dominant
role played by organized crime, ]5 whose ramifications extend throughout the country
but are particularly thick and deep rooted in certain regions of the South, where
criminal organizations like the Mafia, Camorra, and 'Ndrangheta emerged far back
in history. This phenomenon, which certainly differentiates regions, has evolved gradually through time. In modern econometric language [Hsiao (1986), p. 25] it can be
regarded as an attribute that is the same for a given cross-sectional unit through time
but varies across cross-sectional units. When this is the case, and the specific attribute is
not observed, one can take advantage of the panel structure and estimate a model
100
E s t i m a t i n g a crime equation
Figure4Repo~tedCrimes-
mean values,
1980 - 1989
Murder
0,10
0,16
0,16
0,14
012
0,12
0,10
0,08 0,08
0,(~
0,06
O,O4
0,06
0,03
0O4
0,03
•
0,02 0,02
0,03
0,02
0,02
O,O2 0,(~
0,02
0,C
3,00
2,91
2,00
1,65
!
LSO
0,99
o,]io o olo| | ~o•,1
o..
......
.° . . . . . .
a:l:;;:.,
I °i Oio
o:;
o.,g:
o.,,
,,
,=,,
,
:
F1G. 4. Reported crimes; mean values, 1980-1989.
treating these specific effects either as fixed constant or as random variables [Hsiao
(1986)]. 16
In the first case the m o d e l takes the following form:
N
y . -- e~ + ~ i + f3'xit + uit = e~* + f3'x. + uit i = 1 . . . . .
N ; t = 1. . . . .
T; Z
I~i = 0; cx* = 0L+ IXi
(1)
where ca is the population mean, lai is the variable capturing the effects o f the regional
time-invariant c o m p o n e n t s measured as deviations from the c o m m o n mean ~, and u~t
is an i n d e p e n d e n t identically distributed random variable•
Operationally, o n e can either estimate m o d e l (1) as it stands by ordinary least squares
(OLS) or, to reduce the n u m b e r o f parameters, apply the OLS procedure to the
transformed model, which can be obtained by measuring regional observations as deviations from regional means (over time).
Alternatively, the region-specific effects that are not observed, instead o f being
M A R S E L L I AND V A N N I N I
101
~,11o
o0,o0
40.00
38,93
32,41
1
21 51
~
23,28
Frm~l
1,20
I,O0
O,96
0,80
0,49
0 43
0.40
0,37
0,~
0 42
"
0,52
0 42
'
0,44
0,45
0,39
0.20
0,o0,
:
:
. . . . . . .
.
:
:
:
FIc. 4. Continued.
treated as fixed (though different for different cross-sectional units), can be treated as
r a n d o m variables. Consequently, the m o d e l can be written as:
Yit = <x + [3 xit + uit = ~x + [3'xa + eit
(2)
where eit = [li + uit a n d [lit a n d uit are white noises orthogonal to the regressors.
T h e similarity between the two formulations is apparent: In e q u a t i o n 1 the regionspecific effects omitted are s u b s u m e d u n d e r the intercept term, whereas in equation (2)
they are treated as a sample of r a n d o m drawings from a p o p u l a t i o n a n d e n t e r the
disturbance term. Now, however, the least square estimator is n o longer best linear
unbiased, i n a s m u c h as the covariance matrix is n o t diagonal, a n d a general least squares
(GLS) estimator is called for. 17
Despite the terminology, "fixed" versus " r a n d o m " regional effects, the difference
between the models (1) a n d (2) does n o t d e p e n d so m u c h o n the n a t u r e of the
u n o b s e r v e d c o m p o n e n t , b u t rather o n whether or n o t this c o m p o n e n t can be considered o r t h o g o n a l to the explanatory variables i n c l u d e d in the model. TM For, as shown by
M u n d l a k (1978), to obtain efficient estimates it would be advisable to apply the GLS
Estimating a crime equation
102
Figure
5
Coefficients of varistion across time and across regions, 1980 - 1989
1,4
1,20
1,2
1,13
1
0,97
0,97
°8
7
073
~!/i~
~
0,67
0,6
0,4
0~
0
i
~
~, ~
t
Theft
v
~
~.~.~T
Murder
Io R ~ c ~
r 1 1 ~
~ x l
I
Robbery
I4"f"
~
Fraud
[]Unkn0wn ~Probability []Sev¢l~'l
FI6. 5. Coefficients of variation across time and across regions, 1980-1989.
p r o c e d u r e to m o d e l (2) ff the answer is positive, an d the OLS estimator to m o d e l (1)
otherwise.
Within this framework, for the m o d e l selection p r o c e d u r e we can resort to several
tests. To check w h e t h e r the fixed effects or the r a n d o m effects m o d e l conveys m o r e
in f o rm at i o n than a classical regression model, o n e can use an F-test for the j o i n t
significance o f the regional fixed c o m p o n e n t s in the f o r m e r case, and a Breusch-Pagan
(1979) LM-test on the null ~ -- 0 in the latter case. F u r t h e r m o r e , to decide w h e t h e r the
regional u n o b s e r v e d features are better m o d e l e d as " f i x e d " or " r a n d o m " , o n e can
p e r f o r m a H a u s m a n (1978) test on the orthogonality o f the o m i t t e d regional c o m p o -
MARSELL[ AND V A N N I N I
103
Figure 6 Coefficient# of vxrixtion across time and across regions, 1980 - 1989
o,6[ ~°'56
o,ss
0,5
0,3
0,19
0,15
0,1
Unempl
PbWo~4~ Coawmp
l~m~'~,
Scrvic~
Y o u t h Education W a ~
FIG. 6. Coefficients of variation across time and across regions, 1980-1989.
nents with respect to the explanatory variables explicitly included in the model. Finally,
to allow for c o m p o n e n t s that are the same for all cross-sectional units at a given point
in time but that vary t h r o u g h time, model (1) can be slightly modified and written as:
Yit = OL + Pi + ~kt + ~ ' X i t + Uit
(3)
where )kt is the period regional-invariant variable. An LR-test on the joint significance of
these t c o m p o n e n t s can be used to assess the legitimacy of model (1) or (3).
For each category of crime considered we estimated a general formulation of both
models (1) and (2), including, as regressors o f each crime equation, all empirical
counterparts of the i n d e p e n d e n t variables suggested by the EMC plus the set of backg r o u n d socio-demographic variables. We then simplified the estimated equations, dropping those variables that turned out highly insignificant and testing the imposed restrictions at each step. 19
T h e results are set out in Table 3, where for each category of crime both the "fixed
effects" and the " r a n d o m effects" model are reported along with the Breusch-Pagan
LM-test and the H a u s m a n test. 2° Leaving aside r a n d o m and fixed effects to begin
Estimating a crime equation
104
TABLE 3. E s t i m a t i o n results
Murder
Fixed
effects
Random
effects
UNKNOWN
PROBABILITY
-0.136"**
(0.024)
SEVER/TY
0.193"**
(0.074)
UNEMPLOYMENT 7.977*
(4.115)
-0.141"**
(0.024)
0.190"**
(0.073)
8.902**
(3.973)
PBWORKS
Robbery
Fixed
effects
Random
effects
0.850*** 1.052***
(0.166)
(0.159)
-0.194"** -0.201"**
(0.035)
(0.035)
0.154"*
0.189"*
(0.077)
(0.076)
7.375**
4.642
(3.326)
(2.976)
0.094*
0.151"**
(0.056)
(0.054)
Theft
Fixed
effects
2.364***
(0.563)
Fraud
Random
effects
SERVICES
CONSTANT
R2
LR(10)
Hausman
LM
LRg~,,
No. Resu-.
Steps
1.275"** 0.713"**
(0.388)
(0.365)
5.340**
2.365
(2.402)
(1.863)
-7.449*** -6.044***
(0.995)
(0.814)
89.68
36.20***
14.85"*
369.38***
1.032
6
3
2.063**
(0.893)
2.747***
(0.788)
-3.543***
(0.518)
93.55
13.85
26.47***
294.21"**
0.59
5
1
Random
effects
2.784***
(0.544)
-0.191"** -0.080
(0.073)
(0.065)
-6.082** -10.559"**
(2.353)
(1.672)
CONSUMPTION
BENEF/TS
Fixed
effects
0.306*** -0.305***
(0.030)
(0.030)
-0.107
-0.106'
(0.067)
(0.056)
0.820**
(0.349)
-0.509*
(0.301)
2.048**
(0.919)
3.628***
3.558***
(0.140)
(0.142)
90.56
44.239***
15.24"**
312.8"**
8.5
8
1
0.514"
(0.271)
-0.378*
(0.228)
2.353***
(0.675)
-6.232***
(0.455)
74.53
17.09
8.617
251.45"**
2.28
6
1
Note: Standard errors in brackets; *, **, *** reject the null at the 10%, 5%, and 1% level, respectively. LR~10) is a
test for the significance of the trend component; Hausman is a test for the orthogonality of the stochastic c o m p o n e n t
with respect to the set of regressors in model (2); LM is the Breusch-Pagan test for the r a n d o m effect model versus the
linear classical model with an overall constant; LRgen tests the validity of the restrictions imposed on the general model
to achieve the chosen specification; No. Restr. is the number of restrictions tested; a n d Steps is the n u m b e r of
estimation runs necessary to reduce the general model and achieve the chosen specification.
with the Breusch-Pagan LM-test always rejects the classical model, 21 and in two cases
(murder and theft) the time trend turns out to be significant. Furthermore, the results
indicate that u n d e r all the specifications the estimated coefficients of average monthly
salary (WAGE), percentage of all males in the age g r o u p 14 to 29 (YOUTH), and
n u m b e r of students who have completed secondary and high schools relative to the
population (EDUCATION) are statistically insignificant. These variables may be very
weak proxies of the influences that we wished to capture. This seems particularly
true for the variable EDUCATION, which ideally should represent the average cultural
profile of the population but which in practice, as pointed out by Tabellini and Piras
(1992), fails to accomplish this task because it varies very little across regions and, in
many instances, it simply reflects the choice of the only available alternative to unemployment. 22 The same reasoning, however, does not apply to a variable like YOUTH,
which accurately identifies an age and sex group that, both in theory and practice,
is considered to be m o r e p r o n e to crime. 93 As suggested by Figure 6, however,
the variability of YOUTH across regions over the sample period is very small, so that it
MARSELLI AND VANNINI
105
fails to discriminate the offenders' participation in illegitimate activities across the
regions.
On the whole, we observe a considerable uniformity of estimated parameters and
their significance across different crimes under both the fixed effect and the random
effect versions of the model. Moreover, in all instances but one, the Hausman test
suggests the adoption of a "fixed effects" representation: This means that the unobserved regional component is highly correlated with the set of regressors that explain
crime. This result is quite reasonable, given that the Hausman statistic is significant for
those offenses (murder, theft, and robbery) that, unlike fraud, are more closely related
to organized crime. 24 Estimated fixed effects are represented in Figure 7. It may be
noted that although the graph concerning fraud is relatively smooth, the graphs associated with the remaining offenses are more spiky and have some distinctive peaks
corresponding to those regions involved with organized crime.
Looking at the estimated equations, primafade they would seem to contradict what
Figure 7 Resional Fixed EfFects
0.35
I
0.3-
0,25
0,2-
0,15
0,1 I
0,05 -
i
~l~oft
'!I
30.
IO-
FIG. 7. Regional fixed effects.
106
Estimating a crime equation
would be expected on the basis of the standard version of the EMC [Becker (1968)],
which predicts a negative relationship between crime and the probability and severity of
punishment. 25 We mentioned, however, that more realistic versions of the basic EMC
have ambiguous implications, so the role of empirical investigation is to help in assessing the direction and weight of the partial effects suggested by the theory. We now
examine each equation separately.
The results for murder show first of all that the percentage of all murders cleared by
the judicial authority seems not to have a statistically significant association with the
level of this offense, whereas both the probability of conviction and the severity of
punishment are significant.
These findings have a number of possible explanations. One is that because murder
is one of the most serious crimes, the judicial apparatus spends considerable efforts
identifying those who commit this offense: Knowing this, 26 murderers are more discouraged by the probability of arrest and conviction than by the likelihood of being
identified. It is certainly not unusual to hear of a "mafioso" guilty of murder, whose
presence is evident but whom nobody manages to convict. As for the positive sign
attached to the severity variable, it may be noted that harsh sentences that put dangerous criminals out of circulation (think of Mafia bosses and the like) may result in
increasing crime simply because they trigger a fight among criminal rivals over the
market share left unattended or because "incapacitation effects benefit potential entrants as they are now safe from the threat of entry repelling activity by the former
incumbents" [Cameron (1988), p. 305].
Turning to the socioeconomic variables, the unemployment rate has a positive and
significant effect. One can interpret this result either in terms of falling opportunity
costs of crime, encouraging more offenses, or, taking unemployment as a proxy for the
business cycle, by saying that of the three effects linking business cycles to crime the
motivation effect dominates. On this point it should be noted that contrary to the findings
of other researchers [Trumbull (1989) ; Field (1990) ], here the unemployment rate has
a significant effect in all crime equations, despite the presence among the regressors of
a variable like consumption, which in theory is more closely related to the business
cycle. A plausible explanation of this result is that public transfers to Italian regions that
traditionally have tended to supplement household incomes have weakened the link
between consumption and business cycles.
The number of people employed in the service sector relative to the total number of
employed people also has a positive and significant effect, and it provides positive
evidence of the importance of ex7Plicitly including some measures of income distribution in the explanatory factors. ~ As for the variable BENEFITS, an obvious way to
rationalize its positive and significant impact is to think of this variable as proxying
changes in initial wealth (increased welfare payments), in which case the standard EMC
predicts that time devoted to illegal activities will increase with wealth.
The equation for robbery shows that all three of the criminal justice variables have a
significant impact on this type of offense but the severity of punishment has the wrong
sign, whereas the percentage of all murders cleared by the judicial authority and the
probability of conviction have the expected signs. To explain this finding one can resort
to the same considerations previously made concerning murder, plus two further qualifications. The first is that stiffer sentences in a region may shift criminal activity to other
regions, and if the importing regions retaliate by the same token, one may observe
crime rising in tandem with severity (spiUover effects). The second is that because rational
potential victims may reduce their level of self-protection if deterrence is increased
MARSELLIANDVANNINI
107
(self-protection effect), the equilibrium between demand and supply might entail a greater
number of robberies. 2s
Of the remaining variables in the robbery equation, in addition to UNEMPLOYMENT
and SERVICE, the a m o u n t of public works started by Government intervention
(PBWORK) has a significant positive association with crime.
The theft crime rate is found to vary in the expected direction with the clear-up rate
and the severity of punishment; in contrast, it shows no significant relation with the
probability of punishment. This finding calls for an interpretation akin to the one given
for murder. For it can be argued that thieves are relatively more deterred by the
probability of being identified than of being arrested and convicted, because the judicial apparatus pursues and punishes this type of offender less vigorously than more
serious offenders. Among the socioeconomic variables, only UNEMPLOYMENT seems
to affect theft significantly, although with a negative sign, which means that for this
specific crime the "lifestyle" and "opportunity" effects dominate the "motivation"
effect. It should be stressed that the results concerning theft may be strongly affected by
the huge increase in offenses related to the use of hard drugs, which in many countries
constitutes a crime of its own, and by the strategies adopted to fight this phenomenon.
A policy shift from harassing buyers to harassing sellers, for instance, is likely to push up
the price of illegal drugs and to increase total expenditure on them, which in turn may
increase the number of property crimes.
We made an attempt to establish the spillover from drug consumption to property
crime by including an additional regressor in the final equation for theft, namely, the
number of reported acquired immunodeficiency syndrome (AIDS) cases, as a proxy for
the number of drug consumers. This choice was motivated by the strong evidence that
in Italy the human immunodeficiency syndrome is acquired mostly by needle sharing
and, hence, it is highly correlated with the number of hard drug addicts. The estimation
results shown in Table 4 confirm the existence of significant spillovers. But even more
interesting is the finding that, once we control for drug consumption, the regional fixed
effect is greatly reduced (Figure 8): The inclusion of this new variable has allowed us to
unravel a substantial part of the regional unobservable component previously left unexplained.
The results examined so far stand in contrast to the basic EMC, whereas those
concerning fraud support it. M1 three of the deterrence variables have the expected
signs and are statistically significant, and the same applies to the opportunity costs
factors. Moreover, this is the only case in which consumption, rather than unemployment, turns out to be (marginally) significant. To some extent, this result holds no
surprise, inasmuch as the core of this specific crime, represented by an offender deceiving a potential victim, is such that considerations concerning time and psychic costs
are relatively unimportant. Thus, the basic hypotheses of the model seem appropriate.
VI. Conclusions
There are a number of reasons why estimating and testing a crime equation based on
the ECM with Italian regional data can be a fruitful exercise: the substantial variability
of crime rates across time and space; the prevalence of offenses involving material gains
relative to crimes of passion or related to ethnic/racial conflicts; the long-standing
presence of organized crime in many areas of the country; a codified criminal law that
108
Estimating a crim~e equation
TABLE 4. Estimation results: Theft
With AIDS
UNKNOWN
PROBABILITY
SEVER/TY
AIDS
UNEMPLOYMENT
PBWORKS
CONSUMPTION
BENEFITS
SERVICES
CONSTANT
R2
LR¢10~
Hausman
LM
LRge,
No. Restr.
Steps
Without AIDS
Fixed
effects
Random
effects
Fixed
effects
Random
effects
2.225***
(0.558)
2.769***
(0.533)
2.364***
(0.563)
2.784***
(0.544)
-0.179**
(0.072)
0.040**
(0.016)
-4.649*
(2.392)
-0.075
(0.062)
-0.022***
(0.008)
-8.858***
(1.729)
-0.191 ** *
(0.073)
-0.080
(0.065)
-6.082**
(2.353)
-10.559"**
(1.672)
3.805***
(0.156)
3.373***
(0.144)
3.628***
(0.140)
3.558***
(0.142)
90.84
45.21"**
36.37***
317.47***
0.84
8
1
90.56
44.239***
15.24"**
312.8***
8.5
8
1
Note: Standard errors in brackets; *, **, *** reject the null at the 10, 5, and 1% level, respectively.LR~10~is a test for
the significance of the trend component; Hausman is a test for the orthogonality of the stochastic component with
respect to the set of regressors in model (2); LM is the Breusch-Pagan test for the random effect model versus the linear
classical model with an overall constant; LRg~, tests the validity of the restrictions imposed on the general model in
order to achieve the chosen specification; No. Restr. is the number of restrictions tested; Steps is the number of
estimation runs necessary to reduce the general model and achieve the chosen specification.
severely limits t h e e x e r c i s e o f d i s c r e t i o n by t h e j u d g e . I n this c o n t e x t , g u i d e d by t h e
e c o n o m i c m o d e l o f c r i m e , we s t u d i e d t h e d e t e r m i n a n t s o f f o u r c a t e g o r i e s o f c r i m e , i.e.,
m u r d e r , r o b b e r y , t h e f t , a n d f r a u d . T h e r e s u l t s relative to t h e first t h r e e c a t e g o r i e s
c o n t r a s t w i t h t h e p r e d i c t i o n s o f t h e s t a n d a r d v e r s i o n o f t h e E M C [ B e c k e r (1968); E r l i c h
(1973) ], w h e r e a s t h o s e c o n c e r n i n g t h e last o n e l e n d s o m e s u p p o r t to its h y p o t h e s e s .
B e c a u s e m o r e s o p h i s t i c a t e d v e r s i o n s o f t h e m o d e l [ H e i n e k e (1978) ] h a v e a m b i g u o u s
i m p l i c a t i o n s , t h e e x e r c i s e was m e a n t to assess e m p i r i c a l l y t h e p a r t i a l effects o n c r i m e o f
t h e m a i n f a c t o r s s u g g e s t e d by t h e t h e o r y . A m o n g t h e s e factors, we c o n s i d e r e d o r g a n i z e d
c r i m e , a p h e n o m e n o n usually n e g l e c t e d in t h e e m p i r i c a l l i t e r a t u r e o n t h e EMC. T h e
m a j o r f i n d i n g s are: (i) t h e d e g r e e o f c e r t a i n t y o f p u n i s h m e n t s e e m s m o r e effective t h a n
t h e severity o f p u n i s h m e n t a n d t h e i n v e s t i g a t o r e f f i c i e n c y o f t h e c r i m i n a l j u s t i c e system,
a m o n g t h e d e t e r r e n c e v a r i a b l e s ; (ii) p r o x i e s f o r t h e r e l a t i v e r e t u r n s f r o m l e g i t i m a t e a n d
i l l e g i t i m a t e activities, s u c h as t h e r a t e o f u n e m p l o y m e n t , t h e n u m b e r o f p u b l i c w o r k s
s t a r t e d , a n d t h e relative size o f t h e service sector, a r e f o u n d to h a v e a s i g n i f i c a n t i m p a c t
o n c r i m e , w h e r e a s t h e a v e r a g e m o n t h l y w a g e h a s n o effect; (iii) t h e a v a i l a b l e m e a s u r e
o f t h e level o f e d u c a t i o n a n d t h e p e r c e n t a g e o f all m a l e s in t h e a g e g r o u p 14 to 29 d o
MARSELLI
AND VANNINI
109
Fillure II l ~ , g i o u l F t x ~ l Effects
THEIrr wl~um A~6
7O
3O
10
THEFr wldtAEDS
2,5
2
1,5
1
O.5
0
FIG. 8. Regional fixed effects.
not seem significant in any of the equations estimated. Finally, some law-enforcement
variables present signs that might be counterintuitive in the light of the standard econometric model, in which crime is supplied by individuals under exogenously given
opportunities, but are less surprising in estimated equations that refer to a situation
marked by the presence of organized crime.
Though the overall results, and their interpretation with reference to the peculiarities
of the Italian context, disclose many interesting aspects, the conclusions must be mitigated by some cautionary remarks on the caveats of this kind of aggregate level work.
First of all, as stressed repeatedly in the literature [Taylor (1978)], the set of observations is affected by very serious measurement problems: We want to explain crime, but
we can only use recorded crime, and, as is well known, many crimes are not reported
at all. Perhaps, on this point, the consistency of qualitative results between murder,
which is measured correctly, and robbery, which is not, may be considered reassuring.
Second, there is the issue of simultaneity, stemming from the fact that in general the
criminal justice system does respond to observed crime levels and to social concern
about them. According to the distinction recently made by Cornwell and Trumbull
(1994, p. 361), in this study we left aside (basically for lack of valid instruments)
endogeneity arising from "conventional simultaneity" and concentrated on endogeneity due to "neglected heterogeneity." Here, however, the former source of endogeneity should be less important than elsewhere, especially in those applications regarding
110
Estimating a crime equation
c o m m o n law c o u n t r i e s , in view o f t h e g r e a t e r v a r i a b i l i t y o f t h e r e g i o n a l c r i m i n a l j u s t i c e
e q u a t i o n relative to t h e r e g i o n a l c r i m e e q u a t i o n a n d t h e s c a n t y m a r g i n s o f d i s c r e t i o n
left to j u d g e s i n o u r j u d i c i a l system. I n o t h e r w o r d s , w i t h r e f e r e n c e to o u r s a m p l e , we
b e l i e v e t h a t a r e a s o n a b l e a r g u m e n t f o r i d e n t i f i c a t i o n [see T a y l o r ( 1 9 7 8 ) , p p . 5 9 - 6 2 ]
t h a t c a n b e m a d e o n c e o n e t a k e s a b r o a d e r view o f t h e p r o b l e m is likely to b e m e t .
Notes
1. Only recently in Italy are we witnessing a surge of papers based on the economic approach and its
methods both inside and outside the field of economics [Gambetta (1989-1990, 1993); Campiglio
(1990) ; Sacconi (1991); Della Porta (1992); Zamagni (1992); Commissione Parlamentare Antimafia
(1993) ].
2. We are ignoring passing references to Beccaria's Dei delitti e dellepene (1764) and Lombroso's L'uomo
delinquente (1876).
3. We must mention, however, some notable exceptions, like Sandelin and Skogh (1986) and Wahlroos (1981).
4. It is well known that Italy is the motherland of such archetypal criminal organizations as Mafia,
Camorra, 'Ndrangheta, and the newborn Sacra Corona Unita.
5. A very useful overview of the legal systems of the United Kingdom, the major Commonwealth
countries, the United States, and the countries of Western Europe can be found in the compendium
by Walker (1980). On the principles and policies of the English criminal law and penal system see,
respectively, Ashwort (1991) and Vinciguerra (1992).
6. The entry "delitti per Associazione di tipo mafioso" features in the official recorded crime categories only since 1988 (see ISTAT, Statistiche giudiziarie).
7. This is much more evident for murder and robbery than for theft.
8. Mafia, Camorra, and Ndrangheta come from Sicilia, Campania, and Calabria respectively, whereas
the newborn Sacra Corona Unita emerged in Puglia. It goes without saying that these criminal
groups affect their habitat not only directly, committing offenses, but also indirectly, controlling
legal and illegal businesses and shaping in a variety of ways norms and wants that govern individual
attitudes toward crime. An extended framework of rational choice that explicitly acknowledges a
role for norms in this context has been recently developed by Eide (1995).
9. The article by Opp (1989), directed both to economists and sociologists, illustrates the fundamental
theory of differential association and develops a theoretical confrontation of the economic and
sociological perspectives on crime.
10. See Taylor (1978) and Schmidt and Witte (1984).
11. This variable represents our empirical counterpart of the probability of punishment in the standard
EMC.
12. In the case of Italy, though, it is more likely that consumption and unemployment embody structural differences among regions.
13. On the general relationship between income inequality (as a measure of the level of transferable
goods and assets in the economy) and crime see Ehrlich (1973, pp. 538--539).
14. There are twenty of them, but we aggregated Valle d'Aosta and Piemonte.
15. Here, by organized crime we mean primarily crime organized by such criminal enterprises as Mafia,
Camorra, and 'Ndrangheta. It goes without saying that there is not an objective definition of
organized crime, and that most crime is rarely disorganized. Leaving aside historical consideration
as to the origin of these criminal formations, we agree with Schelling's (1984, p. 182) interpretation
that organized crime "seeks not only influence, but exclusive influence. In the overworld its counterpart would be not just organised business, but monopoly. And we can apply to it some of the
adjectives that are often associated with monopoly--ruthless, unscrupulous, greedy, exploitative,
unprincipled".
16. To our knowledge, this is the first attempt at estimating the economic model of crime in the
presence of organized crime. The same technique, however, has been used recently by Cornwell and
MARSELLI AND VANNINI
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
111
Trumbull (1994) for the purpose of determining the statistical consequences of ignoring unobserved heterogeneity, like jurisdictional differences in crime reporting without data on actual
crimes.
The literature provides many procedures to obtain such GLS estimator. See for instance Greene
(1990).
This claim was made by Mundlak (1978) and is discussed in Hsiao (1986, p. 45).
During the simplification process we checked that the choices we made were not affected by high
standard errors due to multicollinearity. In particular, whenever we excluded a block of variables,
we also verified that each o f them, taken alone, was not significant. The results seem robust on the
grounds that (i) the cell values of the correlation matrix for the regressors are never greater than
0.7 and (ii) for each estimated model, the condition n u m b e r of the X' X matrix takes values around
19.5. According to Belsley, Kuh, and Welsch (1980) only for values of this ratio greater than 30 a
linear d e p e n d e n c e a m o n g the columns of X exists that may seriously affect the standard errors o f
the estimated coefficients.
All variables are in logs, except for UNEMPLOYMENT and SERVICES.
This rejection is corroborated by the joint significance o f the regional dummies according to a
standard F-test, whose results are not reported for space constraints.
Future applications should be able to exploit the recently published 1990 Census informafions to
estimate, using inter-censal linear interpolation, a more reliable proxy for completed schooling.
That young m e n are more crime p r o n e is an opinion largely supported by the fact that this group
is over-represented in reported crime. But it can be justified on sociological grounds, as in Field
(1990, pp. 41-42) who shares Easterlin's view that "for a given opportunity structure and means of
social c o n t r o l . . , persons in large cohort will face relatively fewer opportunities for social advancement, and relatively fewer social controls, than persons in small age cohorts."
It goes without saying that this claim does not apply to criminal enterprises in general, but seems
plausible within the Italian context, where organized crime is still "strongly anchored to its predatory stage" [Becchi (1993), p. 89].
Following a c o m m o n practice, law-enforcement variables have been entered in current levels, and,
therefore, we are assuming that individuals' perceptions of the certainty and severity o f p u n i s h m e n t
are related to actual levels in the regions in which they commit their crimes. This choice leaves room
for a simultaneity bias. However, (i) the tight margin of discretion left to judges in the Italian
judicial system and (ii) the greater regional variability in law enforcement relative to the regional
crime rates are likely to render this bias less severe than in other applications concerning c o m m o n
law countries. In a subsequent unpublished work [Marselli and Vannini (1995)] using data on the
n u m b e r of police by region recently released by the Anti-Mafia Commission, we explicitly address
the identification issue and check that the basic results are unaltered.
As the saying goes: "Murder will out!"
This result can also be interpreted from a structuralist perspective: The growth of the service sector,
especially of services oriented toward the local market and non-market services, captures many
features of the local socioeconomic context (like the existence of protected sectors easily penetrable by organized crime subsidiaries) particularly suited for the expansion of illegitimate activities [D'Antonio and Scarlato (1993)].
A full account o f the economic explanations as to why p u n i s h m e n t need not deter can be found in
Cameron (1988).
References
ASHWORT, A. (1992). Principles of Criminal Law, Oxford: Clarendon Press.
BECCHI, A. (1993). La criminalitd organizzata come impresa economica in Italia: Paradigini incerti. In
Commissione Parlamentare Antimafia, Economia e criminalita, Camera dei Deputati, Roma, pp. 80-91.
BECKER, G.S. ( 1968 ). Crime and punishmen t: An economic approach. Journal of PoliticalEconomy 78:169217.
112
Estimating a crime equation
BECKER,G.S. (1993). Nobel lecture: The economic way of looking at behavior. Journal of Political Economy
101:385-409.
BELSLEY, D.A., KUH, E., AND WELSGI-I,R.E. (1980). Regression Diagnostics: Identifying Influential Data and
Sources of Collinearity. New York: John Wiley.
BREUSCH, T., AND PAGAN,A. (1980). The Lagrange multiplier test and its application to model specification in econometrics. Rev/ew of Economic Studies 47:239-253.
CAMERON, S. (1988). The economics of crime deterrence: A survey of theory and evidence. Kyklos
41:301-323.
CAMPIGLIO,L. (1990). L'illecito. In IRER-PROGETTO MILANO, Tensioni e nuovi bisogni della cittd in trasformazione. Milano: F. Angeli.
COMMISSIONEP~IAMENTAREANTIMAVlA.(1993). Economia e Criminalitd. Roma: Camera dei Deputati.
CORNWELL,C., ANDTRUMBULL,W. (1994). Estimating the economic model of crime with panel data. The
Review of Economics and Statistics 76:360-366.
D'ANTONIO, M., ANDSCARLATO,M. (1993). L'economia del crimine. Parte I: il ruolo economico. Economia
e Lavoro 3:107-122
DE LEO, G., AND PATRIZI, P. (1992). La Spiegazione del Crimine. Bilancio Critico e Nuove Prospettive Teoriche.
Bologna: I1 Mulino.
DELI~ PORTA, D. (1992). Lo scambio occulto. Casi di corruzione politica in Italia. Bologna: I1 Mulino.
EHRLICH, I. (1973). Participation in illegitimate activities: A theoretical analysis and empirical investigation. Journal of Political Economy 81:521-565.
EIDE, E. (1994). Economics of Crime. Amsterdam: North Holland.
FIELD, S. (1990). Trend in Crime and Their Interpretation: A Study of Recorded Crime in Post-War England and
Wales. London: Home Office Research Study.
FREEMAN,R.B. (1983). Crime and Unemployment. InJ.Q. W1LSON (Ed.), Crime and Public Policy. ICS: San
Francisco.
G~BETTA, D. (1989-1990). La mafia elimina la concorrenza. Ma la concorrenza pu6 eliminare la mafia?
Meridiana 7-8:319-336.
GAMBETTA, D. (1993). The Sicilian Mafia: The Business of Private Protection. Cambridge, Massachusetts:
Harvard University Press.
GREENE, W.H. (1991). Econometric Analysis. New York: MacMillan.
HAUSMAN,J. (1978). Specification tests in econometrics. Econometrica 46:69-85.
HEINEI~, J.M. (1978). Economic Models of Crime Behavior. Amsterdam: North Holland.
HStAO, C. (1986). Analysis of Panel Data. Cambridge: Cambridge University Press.
ISTAT (1989) La criminalitd attraverso le statistiche. Note e relazioni, Roma.
JUDGE, G.G., CARTERHILL, R., GRIFFITHS,W., LUTKEPOHL,H., AND LEE, T. (1980). The Theory and Practice
of Econometrics, New York: Wiley.
LEwls, D.E. (1987). The economics of crime: A survey. Economic Analysis & Policy 17:195-219.
MARSELLI,R., ANDVANNI~I, M. (1995). General deterrence, unobserved heterogeneity, and conventional simultaneity in a panel ofltalian regtons, revised version of the paper presented at the IEA Congress, Tunis.
MUNDIAK,Y. (1978). On the pooling of time series and cross section data. Econometrica 46:69-85
OPP, K.D. (1989). The economics of crime and the sociology of deviant behavior: A theoretical confrontation of basic proposition. Kyklos 42:405-430.
P~ILL¥, B., ANDWITr, R. (1992). Crime and unemployment in Scotland: An econometric analysis using
regional data. Scottish Journal of Political Economy. 39:213-228.
SACCOM, L. (1991). Etica degli affari. Milano: I1 Saggiatore.
SANDELIN, B., AND SKOGH, G. (1986). Property crimes and the police: An empirical analysis of Swedish
data. Scandinavian Jonrnal of Economics 88:547-561.
SCHELL1NC,,T. (1984). Choice and Consequences. Cambridge, Massachusetts: Harvard University Press.
SCHMIDT, P., AND WITTE, D. (1984). An Economic Analysis of Crime and Justice. New York: Academic Press.
TAgELLINI, G., AND PIRAS,G. (1992). A1 sud menD aiuti e pifa legge, II Sole 24 Ore, 14 maggio.
TAYLOR,J.B. (1978). Econometric models of criminal behavior: A review. In J.M. HEINE~ (Ed.), Economic
Models of Crime Behavior. Amsterdam: North Holland.
I~ELLI
AND VANNINI
113
T~UMBULL,W. (1989). Estimation of the economic model of crime using aggregate and individual data.
Southern EconomicJournal 56:423-439.
VINCIGUEm~A,S. (1992). Introduzione allo Studio del Diritto Penale Inglese. Padova: Cedam.
WAHLROOS, B. (1981). On Finnish property criminality: An empirical analysis of the postwar era using
Ehrlich model. Scandinavian Journal of Economics 83:553-562.
WAL~R, D. (1980). The Oxford Companion to Law. Oxford: Clarendon Press.
WHITHERS, G. (1984). Crime punishment and deterrence in Australia: An empirical investigation. The
Economic Record 60-'176-185.
WOLPIN, K. (1978). An economic analysis of crime and punishment in England and Wales. Journal of
Political Economy 86:815-840.
WOLPlN, K. (1980). A time series-cross section analysis of international variation in crime and punishment. Rev/ew of Economics and Statistics 62:417--423.
ZhMA6N1, S. (1992). Mercati lllegali e Marie. Bologna: I1 Mulino.