PUNISHMENT AND DETERRENCE: EVIDENCE FROM DRUNK
DRIVING
- By Benjamin Hansen
Nicholas & Nikolaj
Introduction
Motivation:
From 1975-2012 drunk driving was a factor in 585136 fatalities compared to 725347 murders in USA
Research question:
Do BAC limits as currently administered reduce future drunk driving?
Vocabulary
BAC: Blood Alcohol Concentration
DUI: Driving under the influence of alcohol or other drugs
Recidivism: The tendency of a previously convicted individual to reoffend
Existing Literature
Becker 1968
Suggests that criminals commit
crimes rationally when the
expected benefits of the crime
outweigh the expected costs.
Enhanced punishment
--> Lower recidivism rates
No effect
Enhanced punishment
--> Higher recidivism rates
Abrams 2011
Lee and McCrary 2009
Chen and Shapiro 2007
Finds lower crime rates in states
which enhanced punishments
for committing crimes with a firearm
Finding that youth
respond little to the
large change in
penalties which occurs
upon reaching 18
years of age
Greater punishments in
the form of harsher
imprisonment condition
s lead to increases in
recidivism
Why this variety?
• Time in prison may have criminogenic effects either through peer effects or the depreciation rate of human
capital.
• Higher prison sentences change the age of an individual at the time of release. As such, the age of an
individual may directly affect the trade-offs an individual faces when choosing between crime and traditional
labour supply
Why drunk driving?
o Punishments and sanctions for drunk driving
lack many of the challenges normally
present in testing the deterrent effect of
punishments
o 1: Being stopped for drunk driving is a purely
reactive process
o
In other circumstances, police often choose
who to investigate based on previous
offenses or convictions
o 2: Measuring recidivism in terms of DUI (or
refusal of test) avoids problems/bias with
when to record recidivism (arrest, charge or
conviction)
o 3: Milder prison (or driving restriction)
sentences remove potential demographic
effects of longer sentences
Empirical Strategy
The specific cut-offs for DUI and aggravated DUI allow the sharp RD to test the effect of punishment severity on recidivism.
● How much does the punishment increase
Identifying Assumptions
● For identification, the following assumptions need to be met:
○
○
Continuity of other variables
■ Only treatment status can have an expected change across the threshold
No perfect manipulation!
■ Meaning that some drivers are randomly lucky, and some are not
Continuity of other variables at the threshold
●
Fail to reject the null that the
predetermined characteristics are
unrelated to the BAC cut-off.
Manipulation at the threshold?
Why does this assumption hold?
●
●
Speed of alcohol consumption, food intake, hydration, activity, and
metabolism are difficult to measure, making any BAC calculation
based on consumption and physical characteristics a rough
approximation.
Personal breathalysers is available for individuals to purchase, But
they are far more volatile than the official breathalysers used.
Estimation Procedure
● Regression model:
● DUI → Indicator of the threshold
● y → 0, if they are not pulled over by police in next 4 years. --> 1, if there are subjected
to a test or refuse a test
● Slopes can change at the threshold
● Weighted with a rectangular Kernel - (Sensitivity are checked with other kernels)
● Bandwidth of 0.05 and 0.025 - (Sensitivity are checked with other bandwidths)
● Local linear specification
● All observations are clustered at finest bin BAC (0.001). This captures autocorrelation
between individuals. which have similar BAC levels.
Data
● Administrative records on 512,964
DUI BAC tests in the state of
Washington from 1995 to 2011.
● In 1999, Washington applied a 0.08
threshold for determining DUI, and
0.15 threshold for an aggravated
DUI.
● An example: After a positive test
in 1/1/1999, the paper investigates
recidivism for that specific individual
from 2/1/1999 to 1/1/2003
Results DUI Threshold
● Notable drop at 0.08 and 0.15 threshold
● Having a BAC above the .08 threshold decreases recidivism by 2 %points, and is statistically significant
● The results suggest having a BAC over the DUI threshold results in
either less drunk driving or more attentive drunk driving later.
● Importantly, those with prior tests reduce their recidivism by a larger
margin. This could be because the expected penalties are much higher
if they are caught drunk driving again
Results DUI Aggravated Threshold
●
●
Having a BAC above the aggravated DUI
threshold reduces recidivism for all potential
offenders by 1.1 % points, for those with no prior
tests by 0.9 percentage points, and those with
prior tests by 1.9 percentage points.
They are all highly significant
Why is the linear specification a
reasonable choice
●
●
The stability of the estimates across various
bandwidths suggests that the linear specification
is a reasonable choice
Smaller bandwidths decrease precision →
Decrease in sample size
Disaggregated recidivism levels Heterogeneity of recidivism
● For the entire population of suspected drunk drivers,
having a BAC above the 0.08 threshold results in a
decreased likelihood of being stopped and having a BAC in
any of the recidivism categories.
● Those with at least 1 prior test have a large and significant
reduction in the probability of being in a subsequent
accident involving alcohol
● The estimates suggest that having a BAC above the aggravated DUI
threshold decreases the likelihood of recidivating in all the BAC
content categories.
First stage
●
Having a BAC over the 0.08 threshold is associated with an increase in essentially all the
potential sanctions an individual could receive. --> Good first stage!
Incapacitation
●
●
●
Is the previous estimates driven by long run changes in behaviour or short run changes due to incapacitation
Typical jail time for drunk driving : 24 hours - few months
We see both a short-term and long-term reduction in recidivism
Rehabilitation
●
Victims panel, taking an alcohol abuse assessment, and receiving alcohol treatment.
●
Reduce drunk driving only because it reduces future alcohol consumption
●
So let us test if other alcohol related incidents decreases at the 0.08 and 0.15 threshold
●
No significant values
Conclusion
●
●
●
This paper finds evidence that having a BAC above either .08 DUI threshold or the .15
aggravated DUI is associated with reduced recidivism both in the short and long
term.
This would suggest that the sanctions and ones imposed at current BAC thresholds
are effective
Notice, in research question it is difficult/not possible to separate the ”direct effect”
of receiving a punishment, or the effect of the “raised punishment” for recidivism