Part G-I
Some Examples of Empirical
Work on The Economics of
Crime and Punishment
20/11/2008 Crime_G 1
Objectives
- understand the importance of empirical
work in crime and punishment
- understand the nature of empirical work in
crime and punishment
- understand why the „crime and deterrence‟
debate will continue for some time
20/11/2008 Crime_G 2
Implications of measurement problems:
Example: questions such as how to decrease youth crime
become difficult to answer:
More police, more convictions, longer sentences – yes our
model of rational crime would conclude that this approach
should have a negative impact on crime.
More youth community programs, better education, better
employment opportunities - yes our model of rational crime
would conclude that this approach should have a negative
impact on crime.
But from the perspective of efficient policy the question is:
which approach yields the greatest deterrence per dollar ?
20/11/2008 Crime_G 3
Which approach yields the greatest deterrence per dollar ?
Recall there are two components to the Social Cost of Crime:
1. Direct harm to victim and harm to
society
2. Cost of deterrence
We need to measure both in comparing alternative
approaches to decreasing crime.
20/11/2008 Crime_G 4
Measurement problems
Measuring the social harm of crime:
1. Direct harm to victim and harm to society (perceived
safety) (security/insecurity) real vs. imagined
- difficult to quantify and aggregate (much of this harm is
subjective in nature)
- the number of reported crimes provides only a rough
gauge to whether or not harm costs are rising or falling
- very difficult to determine the marginal benefits
associated with changes in policy
20/11/2008 Crime_G 5
Measuring the cost of deterrence
2. Cost of deterrence
- not too difficult to quantify what we currently spend
(police, courts, prisons, private security expenditures)
- more difficult to determine costs such as „erosion of the
rights of all citizens‟
- very difficult to determine the marginal deterrence
associated with changes in specific expenditures (how much
deterrence per additional dollar spent?)
20/11/2008 Crime_G 6
What is the deterrence effect of an increase in the policing
budget (more apprehensions)? Court budget (more
convictions)? Prison budgets (more convictions/longer
sentences)?
What is the deterrence effect of an increase in preventative
measures (more community programs, more street
lighting, anti-theft devices, etc)?
What is the deterrence effect of a decrease in personal
freedoms (right to privacy, search and seizure, gun
ownership, etc)?
20/11/2008 Crime_G 7
We often discuss crime on a very general (aggregate level) but
there are many different types of crime: Homicide,
Assault, Robbery, Theft, Drunk driving, Prostitution, Gambling,
Pot smoking, Use of hard drugs, Non-Criminal Code violations
Note that the Motor Vehicle Act, Income Tax Act, by-laws, etc.
share prevention and enforcement resources with crime
prevention and enforcement).
In order to really discuss efficient deterrence, we would need to
consider the problem at a disaggregate level.
Consider some specific crime and the cost/benefit of alternative
deterrence strategies for that specific type of crime.
The „efficient‟ composition and amount of deterrence for homicide
is likely to be different than for shoplifting or tax evasion.
20/11/2008 Crime_G 8
Much (most?) empirical work focuses on relatively narrow
questions.
What impact did the change in a given law have on a given
level of crime activity?
What is the cost of imprisoning a person?
What determines the clearance rate for „break and enters‟?
Such narrow question are generally more tractable.
20/11/2008 Crime_G 9
Testing for „Rational Cheating‟
The Economics of Illegitimate Activities: Further Evidence –
Mixon and Mixon, Journal of Socio-Economics, Vol. 25, No.
3, PP 373-381, 1996
Question:
Can the benefits and costs of cheating (to the individual) be
identified and do these factors affect the likelihood of an
individual cheating as predicted by the model of rational
crime?
20/11/2008 Crime_G 10
The Survey
How to get data on cheating?
157 students economics and accounting students were
surveyed (at a larger Southern university).
This is really the only way that you could generate a data set
suitable for standard statistical analysis of this type of
phenomenon.
20/11/2008 Crime_G 11
The Results of the Survey
1. Have you ever observed another student cheating on an exam or
written assignment at SLU? OBCHEAT +
a. Yes (98) 62%
b. No (59)
NR (0)
2. Have you ever seen another student get caught cheating at SLU?
a. Yes (15) 9% CCAUGHT -
b. No (142)
NR (0)
3. Based on your experience in the classroom at SLU, what
percentage of students do you think cheat on a typical exam?
a. No more than 1% (37) 24% PERCHT +
d. Between 30% and 30% (16)
b. Between 1% and 10% (71)
e. More than 30% (4)
c. Between 10% and 20% (26) NR (3)
20/11/2008 Crime_G 12
4. Which response accurately describes your behavior at SLU?
a. Have never cheated on a test or written work (99) 63%
b. Have cheated once on a test or written work (22)
c. Have cheated more than once, but less than five times on a
test or written work (30)
d. Have cheated five times or more on a test or written work (6)
NR (0) CHEHAB dependant variable
5. If you answered “b,” ” c,” or “d” on question 4, have you ever
been caught?
a. Yes (3) 2%
b. No (55)
NR (0)
6. If you answered “c” or “d” on question 4, and you answered “a” on
question 5, did you cheat again after being caught?
a. Yes (0)
b. No (3)
NR (0)
20/11/2008 Crime_G 13
7. Do you know anyone who routinely cheats on exams?
a. Yes (39) 25% KNOCHT +
b. No (117)
NR (1)
8. If you were caught copying another student‟s answers on an
exam, what would you expect to happen to you?
a. Nothing more than a reprimand (2)
b. Be forced to retake the exam (33)
c. Have my course grade lowered by a letter or more (31)
d. Receive an F for the course (70) 45%
e. Be suspended from SLU for at least one semester (20)
NR (1) PENAL -
9. In your opinion, cheating at Southeastern Louisiana University is:
a. Not a problem (52) 33%
b. A trivial problem (65)
c. A problem deserving some concern (35)
d. A serious problem (4)
NR (1)
20/11/2008 Crime_G 14
10. My current classification is:
a. Freshman (8)
d. Senior (65)
b. Sophomore (40)
e. Grad. Student (7)
c. Junior (37) NR (0)
11. My current grade point average is: GPA -
a. 3.50-4.00 (15)
b. 3.00-3.49 (44)
c. 2.50-2.99 (54)
d. 2.00-2.49 (36)
e. less than 2.00 (8)
NR (0)
20/11/2008 Crime_G 15
A regression model
Y is the dependant variable (what you want to explain or
understand)
The X‟s are the explanatory variables
ε is the error term (what cannot be explained by the model)
A Linear regression model looks like this:
Y = b0 + b1*X1 + b2*X2 + ... + bk*Xk + ε
20/11/2008 Crime_G 16
A regression model
How to interpret the coefficients (b‟s)?
Y = b0 + b1*X1 + b2*X2 + ... + bk*Xk + ε
b1 measures the change in Y caused by a one unit change in
X1 holding all other X‟s constant
A linear regression model is a statistical technique for
implementing the ceteris paribus assumption
In some cases the b‟s can be interpreted as an elasticity or
they can be used to calculate the elasticity of the X. This
allows use to answer questions such as: How responsive is
the crime rate to „more police on the street‟?
20/11/2008 Crime_G 17
A regression model
Generally we are interested in two aspects of the regression
results:
Y = b0 + b1*X1 + b2*X2 + ... + bk*Xk + ε
1. Do the individual coefficients (b‟s) contribute to the explanation
of the variation in the dependent variable?
For example if b1 is found to be „statistically significant‟ then we
can concluded that variation in the value of the explanatory
variable X1 have a „statistically significant‟ effect on the
dependent variable Y, holding all other X‟s constant
2. How well does the overall model explain the variation in Y?
To answer this question we consider the error term (ε) and
functions of the error term (i.e. R2).
20/11/2008 Crime_G 18
Mixon and Mixon regression results
They used a LOGIT regression model since the dependent variable
was an index variable.
This complicates the interpretation of the coefficients (b‟s)
somewhat?
Ideally, Mixon and Mixon could have expressed the coefficients as
probabilities but they did not and we do not have enough
information to do it.
Unfortunately LOGIT does not allow use to easily assess the
explanatory power of the model
20/11/2008 Crime_G 19
Table 2. Ordered Logit Results
(1) (2) (3) (4)
CPA 0.3650* 0.422* 0.4181* 0.4125*
(2.28) (2.40) (2.38) (2.32)
OBCHEAT 1.3467* 1.4040* 1.2968*
(3.20) (3.23) (2.94)
CCAUCHT 0.9261 1.0586
(0.75) (0.86)
SEECC 0.1659 -0.1976
(0.12) (0.14)
PERCHT 0.2064 0.3353* 0.2195
(1.15) (1.94) (1.21)
20/11/2008 Crime_G 20
Table 2. Ordered Logit Results
(1) (2) (3) (4)
KNOCHT 1.0305* 0.9304*
(2.61) (2.31)
PENAL -0.0409 -0.1323 -0.1155 -0.1449
(0.25) (0.76) (0.67) (0.82)
INTERCEPT -1.4701 -2.9948* -3.1889* -.9428*
(1.80) (2.97) (3.25) (2.96)
CHI-SQUARE 11.80* 29.50* 80.46‟ 81.53*
20/11/2008 Crime_G 21
Deterring Drunk Driving
Deterring Drunk Driving Fatalities: An Economics of
Crime Perspective, Benson, Rasmussen and Mast,
International Review of Law and Economics, Vol.
19, pp 205-225, 1999
Question
What is the effectiveness of alternative policy tools
used in the control of DUI (driving under the
influence of alcohol)?
20/11/2008 Crime_G 22
They consider a sample of 48 US state over 9 years
The differ with respect to drink laws/alcohol laws, definition of
drunk driving, penalties, enforcement and many other non-
criminological factors.
The authors are interested in the deterrence effect of specific
laws BUT there experiment must control for all other factors
that might affect the dependent variable (the number of
„drunk‟ drivers in a state involved in a fatal car accident
divided by the total number of drivers in the state).
The number of explanatory variables in such regression
models grows quickly.
20/11/2008 Crime_G 23
TABLE 3. Driver involvement Equation (2) (BAC > 0.01)
Independent Variables Coefficient t statistic
Attitudes towards alcohol and driving
Legal drinking age 0.0901867** 2.208
Dram-shop laws -0.07977** -2.31
(tort liability against bars)
Enforcement rules (Pr. of being caught)
Open-container laws -0.10299** -2.51
(in cars)
Anti-consumption laws -199e-02 -0.48
(in cars)
Police per capita -0.00154 -1.50
20/11/2008 Crime_G 24
Independent Variables Coefficient t statistic
Enforcement rules (Pr. of being caught)
Preliminary breath-test laws -0.00343 -0.10
Illegal per se laws -0.01612 -0.21
Implied-consent laws -0.00015 -0.77
(for breadth tests)
No-plea-bargaining laws 0.008445 0.153
Administrative per se laws -0.00015 -0.51
(automatic suspension at time of arrest)
20/11/2008 Crime_G 25
Independent Variables Coefficient t statistic
Punishment
Jail for 1st conviction -0.04167 -1.53
Jail for 2nd conviction 0.000617 0.264
Fines for 1st conviction 0.000055 0.349
Fines for 2nd conviction -0.00002 -0.52
Suspension for 1st conviction 0.000551 0.871
Suspension for 2nd conviction -0.00002 -0.52
20/11/2008 Crime_G 26
Independent Variables Coefficient t statistic
Various control variables
Seat-belt laws -0.03042 -1.17
Vehicle miles per driver 6.52e-05* 5.962
Ethanol per capita 0.38952* 3.338
(alcohol consumption)
Metropolitan population -0.00874 -1.60
Males 16–44 per capita 6.6694 1.593
Per capita disposable income 0.00004** 1.961
Unemployment rate -0.0198** -2.38
20/11/2008 Crime_G 27
Independent Variables Coefficient t statistic
Control variables – Taste (community values)
Dry-county population 0.014147 0.101
Catholics 0.41875 0.32
Mormons -5.6694 -1.23
Southern Baptists 9.0812* 3.306
Other Protestants 2.8491* 2.893
20/11/2008 Crime_G 28
Summary Statistics
Adjusted R2 0.906
F-statistic 50.7
Note: Dependent variable = ln[R/(1 2 R)]. N = 432.
Intercepts, year, and state dummy variables not shown
*Significant at the 0.01 level.
**Significant at the 0.05 level.
***Significant at the 0.10 level (in two-tailed tests).
20/11/2008 Crime_G 29
Some additional tests in the presence of
multicollinearity
TABLE 4. F tests for deterrence variables Equation 2
Deterrence Variables Tested
All deterrence variables F[16,348] = 2.96*
Alcohol control
(legal drinking age, dram-shop laws) F[2,348] = 5.36*
Probability of arrest (police per capital, open-container laws, anti-
consumption laws, illegal per se laws, preliminary breath-test
laws, implied-consent laws) F[3,648] = 2.08***
Probability of being stopped (police per capita, open-container
laws, anti-consumption laws) F[3,348] = 3.20**
20/11/2008 Crime_G 30
Some additional tests in the presence of
multicollinearity
Probability of arrest given being stopped (illegal per se laws,
preliminary-breath-test laws, implied-consent laws)
F[3,348] = 0.27
Expected punishment for 1st and 2nd offenses (administrative
per se laws, no-plea-bargaining laws, fines, jail, suspensions)
F[8,348] = 1.21
Expected punishment given conviction for 1st and 2nd offenses
(jail, fines, suspensions) F[6,348] = 1.59
*Significant at the 0.01 level.
**Significant at the 0.05 level.
***Significant at the 0.10 level.
20/11/2008 Crime_G 31
Capital Punishment as a Deterrent
Does Capital Punishment Have a Deterrent Effect?
New Evidence from Postmoratorium Panel Data –
Dezhbahsh, Rubin and Sheperd, Americam Law
and Economics Review, Vol. 5, No. 2, 2003
Question
Does capital punishment deter murder?
20/11/2008 Crime_G 32
Table 1. Executions and Executing States
Year No. of Executions No. of States with Death Penalty
1977 1 31
1978 0 32
1979 2 34
1980 0 34
1981 1 34
1982 2 35
1983 5 35
1984 21 35
1985 18 35
1986 18 35
1987 25 35
1988 11 35
1989 16 35
1990 23 35
1991 14 36
1992 31 36
1993 38 36
1994 31 34
1995 56 38
1996 45 38
1997 74 38
1998 68 38
1999 98 38
2000 85 38
Source: Snell, Tracy L. 2001. Capital Punishment 2000. Washington, D.C.: U.S. Bureau
of Justice Statistics (NCJ 190598).
20/11/2008 Crime_G 33
Table 2. Status of the Death Penalty
Jurisdictions without a Death Jurisdictions with a Death Penalty on
Penalty on December 31, 2000 (No. of Executions 1977-2000)
Alaska Texas (239) Virginia (81)
District of Columbia Florida (50) Missouri (46)
Hawaii Oklahoma (30) Louisiana (26)
Iowa South Carolina (25) Alabama (23)
Minnesota Arkansas (23) Georgia (23)
Maine Arizona (22) North Carolina (16)
Michigan Illinois (12) Delaware (11)
Massachusetts California (8) Nevada (8)
North Dakota Indiana (7) Utah (6)
Rhode Island Mississippi (4) Maryland (3)
Vermont Pennsylvania (3) Washington (3)
Wisconsin Nebraska (3) Oregon (2)
West Virginia Kentucky (2) Montana (2)
Colorado (1) Wyoming (1)
Idaho (1) Ohio (1)
Tennessee (1) South Dakota (0)
Connecticut (0) Kansas (0)
New Hampshire (0) New Jersey (0)
New Mexico (0) New York (0)
Source: Snell, Tracy L. 2001. Capital Punishment 2000. Washington, D.C.:
20/11/2008 Crime_G 34
U.S. Bureau of Justice Statistics (NCJ 190598).
20/11/2008 Crime_G 35
Data and Estimation
Panel data for 3,054 counties from 1977 to 1996
(61,080 observations)
Three equation simultaneous system model
Must estimate subjective probabilities - the
econometrics is a bit tricky but we can interpret
the results in a more or less straightforward
manner.
20/11/2008 Crime_G 36
Table 3. Two-Stage Least Squares Regression Results for
Murder Rate
Estimated Coefficients
Regressor Model 1 Model 2 Model 3
Deterrent Variable
Probability of arrest -4.037 -10.096 -3.334
(6.941)** (17.331)** (6.418)**
Conditional probability
of death sentence -21.841 -42.411 -32.115
(1.167) (3.022)** (1.974)**
Conditional probability
of execution -5.170 -2.888 -7.396
(6.324)** (6.094)** (10.285)**
20/11/2008 Crime_G 37
Control Variables
Regressor Model 1 Model 2 Model 3
Other Crimes
Aggravated assault
rate 0.0040 0.0059 0.0049
(18.038)** (23.665)** (22.571)**
Robbery rate 0.0170 0.0202 0.0188
(39.099)** (51.712)** (49.506)**
20/11/2008 Crime_G 38
Control Variables
Regressor Model 1 Model 2 Model 3
Economic Variables
Real per capita
personal income 0.0005 0.0007 0.0006
(14.686)** (17.134)** (16.276)**
Real per capita
unemployment
insurance payments -0.0064 -0.0077 -0.0033
(6.798)** (8.513)** (3.736)**
Real per capita
Income maintenance
Payments 0.0011 -0.0020 0.0024
(1.042) (1.689)* (2.330)**
20/11/2008 Crime_G 39
Control Variables
Regressor Model 1 Model 2 Model 3
Demographic Variable
African American (%) 0.0854 0.1114 0.1852
(2.996)** (4.085)** (6.081)**
Minority other than
African American (%) -0.0382 0.0255 -0.0224
(7.356)** (0.7627) (4.609)**
Male (%) 0.3929 0.2971 0.2934
(7.195)** (3.463)** (5.328)**
20/11/2008 Crime_G 40
Control Variables
Regressor Model 1 Model 2 Model 3
Age 10±19 (%) -0.2717 -0.4849 0.0259
(4.841)** (8.021)** (0.4451)
Age 20±29 (%) -0.1549 -0.6045 -0.0489
(3.280)** (12.315)** (0.9958)
Population density -0.0048 -0.0066 -0.0036
(22.036)** (24.382)** (17.543)**
NRA membership rate,
(% state pop. in NRA) 0.0003 0.0004 -0.0002
(1.052) (1.326) (0.6955)
20/11/2008 Crime_G 41
Intercept 6.393 23.639 -12.564
(0.4919) (6.933)** (0.9944)
F-statistic 217.90 496.29 276.46
Adjusted R2 0.8476 0.8428 0.8624
Notes: Dependent variable is the murder rate (murders/100,000
population). In Model 1 the execution probability is (number of
executions at t)/(number of death row sentences at t-6). In
Model 2 the execution probability is (number of executions at
t+6)/(number of death row sentences at t). In Model 3 the
execution probability is (sum of executions at t+2+t+1+t+t-
1+t-2+t-3)/(sum of death row sentences at t-4+t-5+t-6+t-
7+t-8+t-9). Sentencing probabilities are computed accordingly,
but with a two-year displacement lag and a two-year averaging
rule.
The estimated coefficients for year and county dummies are not
shown.
*Significant at the 90% confidence level, two-tailed test.
**Significant at the 95% confidence level, two-tailed test.
20/11/2008 Crime_G 42
Does Crime Pay?
Wilson and Abrahamse, Does Crime Pay? 9 JUSTICE
QUARTERLY 359, 367 (1992).
Question
Why do some criminals become „career criminals‟?
Can the expected financial gain explain their
behaviour?
20/11/2008 Crime_G 43
Wilson and Abrahamse compared the gains from crime and
from legitimate work for a group of career criminals in state
prisons
Prisoners were divided prisoners into two groups: mid-rate
offenders and high-rate offenders
Income from crime: data from the National Crime Survey‟s
report of the average losses by victims in different sorts of
crimes were used to estimate the annual income for
criminals
Income from legitimate sources: the prisoners‟ estimates of
their income from legitimate sources
Two-thirds of the prisoners had reasonably stable jobs when
they were not in prison and, on average, the prisoners
believed that they made $5.78 per hour at those legitimate
jobs
20/11/2008 Crime_G 44
Table 12.1 Cooter and Ulen
Criminal and Legitimate Earnings per Year (1988 Dollars)
HIGH-RATE MID-RATE
Crime Crime Work Crime Work
type
Burglary/theft $ 5,711 $5,540 $ 2,368 $7,931
Robbery $ 6,541 $3,766 $ 2,814 $5,816
Swindling $14,801 $6,245 $ 6,816 $8,113
Auto theft $26,043 $2,308 $15,008 $5,457
Mixed $ 6,915 $5,086 $ 5,626 $6,956
Source: Wilson and Abrahamse, Does Crime Pay? 9 JUSTICE
QUARTERLY 359, 367 (1992).
20/11/2008 Crime_G 45
Findings:
For mid-rate criminals, working pays more than crime for every
type of crime except auto theft
For high-rate offenders, crime paid more than legitimate work
for all crimes except burglary
The major cost of crime to the criminals, time in prison is not
accounted for in Table 12.1 (specialist in crime)
When the cost of time in prison is included the net income
from crime fell below the income from legitimate work for
both mid-rate and high-rate offenders
20/11/2008 Crime_G 46
Why, then, do career criminals commit crime?
Wilson and Abrahamse reject two explanations.
First, prisoners felt they had no meaningful opportunity for
legitimate work BUT two-thirds of the prisoners were
employed for some length of time during the period
examined
Second, the prisoners had such serious problems with alcohol
and drugs that they could not hold legitimate jobs - two-
thirds of the offenders had drinking or drug problems BUT
evidence from other studies indicates that these problems
do not normally preclude legitimate employment
Wilson and Abrahamse conclude that career criminals are
“temperamentally disposed to overvalue the benefits of
crime and to undervalue its costs” because they are
“inordinately impulsive or present-oriented.” (they
discount punishments for uncertainty and futurity more
highly than other people)
20/11/2008 Crime_G 47