You are on page 1of 8

Drug and Alcohol Dependence 167 (2016) 67–74

Contents lists available at ScienceDirect

Drug and Alcohol Dependence


journal homepage: www.elsevier.com/locate/drugalcdep

Full length article

Do drinking-age laws have an impact on crime? Evidence from


Canada, 2009–2013
Russell C. Callaghan a,b,c,∗ , Jodi M. Gatley a,b,c , Marcos Sanches d , Claire Benny e
a
Northern Medical Program, University of Northern British Columbia (UNBC), 3333 University Way, Prince George, British Columbia, V2N 4Z9, Canada
b
Human Brain Laboratory, Centre for Addiction and Mental Health (CAMH), 33 Russell St., Toronto, Ontario, M5S 2S1, Canada
c
Dalla Lana School of Public Health, University of Toronto, 155 College St., Health Science Building, Toronto, Ontario, M5T 3M7, Canada
d
Biostatistical Consulting Unit, Centre for Addiction and Mental Health (CAMH), 33 Russell St., Toronto, Ontario, M5S 2S1, Canada
e
Department of Community Health Sciences, University of Northern British Columbia, 3333 University Way, Prince George, British Columbia, V2N 4Z9,
Canada

a r t i c l e i n f o a b s t r a c t

Article history: Background: International debate is ongoing about the effectiveness of minimum legal drinking age
Received 12 November 2015 (MLDA) laws. In Canada, the MLDA is 18 years of age in Alberta, Manitoba and Québec, and 19 in the rest
Received in revised form 3 June 2016 of the country. Surprisingly few prior studies have examined the potential impacts of MLDA legislation
Accepted 24 July 2016
on crime, and the current study addresses this gap.
Available online 30 August 2016
Methods: Regression-discontinuity analyses of police-reported criminal incidents from the 2009–2013
Uniform Crime Reporting (UCR) Survey, Canada’s crime database.
Keywords:
Results: Nationally, in comparison to males slightly younger than the MLDA, those just older than the
Minimum legal drinking age legislation
Alcohol
MLDA had sharp increases in: all crimes, (7.6%; 95% CI = 3.7%–11%, P < 0.001); violent crimes, (7.4%; 95%
Crime CI = 0.2%–14.6%, P = 0.043); property crimes, (4.8%; 95% CI = 0.02%–9.5%, P = 0.049); and disorderly conduct,
(29.4%; 95% CI = 15.6%–43.3%, P < 0.001). Among females, national criminal incidents increased sharply
following the MLDA in: all crimes, (10.4%; 95% CI = 3.8%–17.0%, P = 0.002), violent crimes, (14.9%; 95%
CI = 6.4–23.2, P = 0.001); and disorderly conduct, (35.3%; 95% CI = 11.6–58.9, P = 0.004). Among both males
and females, there was no evidence of significant changes in cannabis- or narcotics-related crimes (quasi-
control outcomes) vis-à-vis the MLDA (P > 0.05).
Conclusion: Release from drinking-age laws appears to be associated with immediate increases in
population-level violent and nonviolent crimes among young people in Canada.
© 2016 Elsevier Ireland Ltd. All rights reserved.

1. Introduction 2014; Kypri et al., 2014; Pitts et al., 2014). Recently, the Canadian
Public Health Association (2011) has recommended increasing the
Alcohol use is the leading contributor to the worldwide bur- MLDA to 19 years of age across all provinces, and a recent Canadian
den of injury, disability, and mortality among individuals aged expert panel on alcohol policy has identified a national MLDA of 21
10–24 years old (Gore et al., 2011). In efforts to reduce these years as the ideal (Giesbrecht et al., 2013). Calls for increasing age-
alcohol-related harms, many countries worldwide have imple- based restrictions on alcohol use among young people run counter
mented minimum legal drinking age (MLDA) laws (International to current, broader policy developments in some provinces towards
Center for Alcohol Policies, 2015), which usually impose age-based increased convenience of liquor access, greater privatization of
restrictions on the consumption, purchase, and possession of alco- alcohol sales (away from a government monopoly model), and
hol among young people. In Canada, the MLDA is 18 years of age in economic gains (and taxation revenue) through increased alcohol
Alberta, Manitoba and Québec, and 19 in the rest of the country. sales and profits (MADD Canada, 2014). In addition, some provin-
International debate is ongoing about the effectiveness of cial political parties have proposed recent platforms to lower the
drinking-age laws (Giesbrecht et al., 2013; DeJong and Blanchette, status-quo legal drinking age (CBC News, 2013) or intentionally
have maintained the MLDA-18-years position (Owen, 2014).
Given that a large and long-standing body of work has demon-
strated a consistent association between alcohol consumption and
∗ Corresponding author at: Northern Medical Program, University of Northern
a broad range of crime (Joksch and Jones, 1993; Graham and West,
British Columbia, 3333 University Way, Prince George, British Columbia, V2N 4Z9,
Canada. 2001), it is striking that relatively few studies have addressed the
E-mail address: russ.callaghan@unbc.ca (R.C. Callaghan). potential impact of drinking-age laws on criminal behavior among

http://dx.doi.org/10.1016/j.drugalcdep.2016.07.023
0376-8716/© 2016 Elsevier Ireland Ltd. All rights reserved.
68 R.C. Callaghan et al. / Drug and Alcohol Dependence 167 (2016) 67–74

young people, especially violence and other alcohol-related crimes database held and managed by the CCJS Statistics Canada. The UCR
(Wagenaar and Toomey, 2002). Almost all the literature in this is a comprehensive, population-based registry of police-reported
area has focused on the impacts of the MLDA on traffic-related crimes that have been substantiated through investigation from
crimes (e.g., alcohol-impaired driving; Wagenaar and Toomey, all federal, provincial and municipal police services in Canada. It
2002; McCartt et al., 2010; Carpenter and Dobkin, 2015). At least in is mandatory for all Canadian police services to provide standard-
our understanding, only four original studies (and one replication) ized reports of all criminal incidents in their administrative files to
have addressed directly the relation between MLDA legislation the UCR Survey, as the incident data collection is mandated by law
and non-traffic-related crimes (Smith and Burvill, 1987; Davis under the authority of the Canadian Statistics Act (Canadian Centre
and Reynolds, 1990; Joksch and Jones, 1993; Carpenter, 2005; for Justice Statistics and Statistics Canada, 2013b). As a result, the
Carpenter and Dobkin, 2015), and only one of these (Carpenter and UCR data captures virtually 100% of police-reported criminal inci-
Dobkin, 2015) can provide recent data. Using California arrest data dents in Canada (Canadian Centre for Justice Statistics and Statistics
from 1976 to 2006 and a regression-discontinuity research design, Canada, 2013b).
Carpenter and Dobkin (2015) found that in comparison to youth
slightly younger than the MLDA, those just older had sharp and sig- 2.2. Data collection and data processing: UCR survey
nificant increases in assault, robbery, driving-related crimes, and a
range of nuisance crimes. The CCJS provides initial and ongoing training and support for
The remaining studies are much older, drawing on nat- each police department in Canada to complete the UCR survey
ural experiments to compare rates of crime pre- and post- forms (Statistics Canada, 2015). Using a standardized submission
implementation of jurisdictional MLDA laws. Joksch and Jones process, trained police personnel send department UCR survey
(1993) examined population-based crime patterns in 18 US states data to the CCJS central data processing system, which uses more
increasing their MLDA to 21 years during the period from 1981 to than 300 intensive checks and edits of all responses from police
1986, and the findings demonstrated significant decreases in disor- department respondents. The CCJS makes contact with police
derly conduct and vandalism, but not in violent crimes in the newly departments to resolve all edit failures or data anomalies. Upon
alcohol-restricted age groups. Similarly, Australian researchers completion of the year-end verification process, and prior to final
found that legislated decreases in the MLDA in the early 1970s annual release, each police service is required to approve the annual
from 21 to 18 years in three Australian states were associated with department-specific UCR summary data tables as accurate and
significant increases in nonviolent crimes (e.g., burglary, larceny, complete. The final, annual master Statistics Canada UCR data file is
drunkenness) and, in one state, violent crimes among males, with considered the gold-standard crime-data resource in Canada, and it
significant increases also appearing in the all-crimes category is used in national governmental reports on crime in Canada (Boyce,
among females (Smith and Burvill, 1987). Still other researchers 2015).
found no evidence of increases in self-reported property-crime per-
petration in affected undergraduate students vis-à-vis New York 2.3. Identification of outcomes: most serious incident
State MLDA changes from 19 to 21 years in the mid-1980s (Davis
and Reynolds, 1990). Each incident represents a police substantiated report of a crim-
In light of current debates and the paucity of evidence in this inal act (or acts), or an arrest, and it is the term used in the
area, it is important to provide more comprehensive research to UCR reporting manual by Statistics Canada (CCJS, Statistics Canada,
inform the MLDA debates in Canada and other countries. A key 2013a). Incidents reported in the UCR are not equivalent to convic-
advantage of the current approach is that it draws upon 2009–2013 tions. The reported crimes may be cleared by charge (or otherwise)
data from Canada’s Uniform Crime Reporting (UCR) survey—a or may be unsolved.
national, population-based registry of all police-confirmed crimes Aggregated crime count data was comprised of only the most
in Canada. The current study focuses on those crime outcomes serious offense for each incident, following the Statistics Canada
prominently associated with alcohol in the prior alcohol-crime approach (Perreault, 2013). In the case of incidents with multi-
(Greenfeld, 1998; Graham and West, 2001; Pernanen et al., 2002) ple violations, the most serious incident was determined by the
or MLDA-crime literature (Smith and Burvill, 1987; Davis and reporting police department according to the Seriousness Index
Reynolds, 1990; Joksch and Jones, 1993; Carpenter, 2005; Carpenter (CCJS, Statistics Canada, 2013a), which uses an incident’s maximum
and Dobkin, 2015): violent crimes, property crimes, and disorderly penalty under law as the primary guide for determining the most
conduct. It was expected that in comparison to their counterparts serious offense.
slightly younger than the MLDA, individuals just older than the
MLDA would have significant and sharp increases in these alcohol- 2.4. Definition of outcomes
related crime outcomes. In addition, the research design employed
two crime-outcome quasi-controls: cannabis-related offenses; and The major crime outcomes were defined using standardized
narcotics-related offenses. Given that prior Canadian research has UCR codes to identify offenses, arranged by the subcategories
demonstrated only a very small alcohol-attributable fraction (0.02) nested within each major crime category (see Supplemental Table
for drug-related crimes (Pernanen et al., 2002), these crimes were 1). Coding categories of UCR outcomes (CCJS, Statistics Canada,
expected to be distributed smoothly across the MLDA transition, 2013a) were constructed based on the Canadian Criminal Code
showing no evidence of significant change associated with the (Government of Canada, 2015) and prior research (Boyce, 2015).
drinking age. Based on their prominence in the prior alcohol-crime (Parker,
1995; Graham and West, 2001) and MLDA-crime literature (Smith
and Burvill, 1987; Davis and Reynolds, 1990; Joksch and Jones,
2. Methods 1993; Carpenter, 2005; Carpenter and Dobkin, 2015), the follow-
ing crime outcomes were selected for the study: violent crimes,
2.1. Data source property crimes, and disorderly conduct. The UCR does not include
any fields indicating alcohol involvement in the record-level
Our analyses draw on data from the Uniform Crime Report- incident. Also, we selected two illicit-drug quasi-control out-
ing (UCR) Incident-based Survey, 2009–2013 (Canadian Centre come categories—cannabis-related offenses and narcotics-related
for Justice Statistics (CCJS) Statistics Canada, 2013a)—a national offenses. Given that most of the research on the MLDA-crime nexus
R.C. Callaghan et al. / Drug and Alcohol Dependence 167 (2016) 67–74 69

has focused on traffic-related outcomes [which we also explored strategy for identifying the optimum bandwidth in the local regres-
in a prior paper (Callaghan et al., 2016)], driving-related crime sion series (Calonico et al., 2014b). All RD analyses were stratified
outcomes are not included in the current work. The remaining by gender, provincial MLDA of 18 years of age (MLDA-18), 19 years
“other” crime categories [i.e., fraud, weapons charges, other sex- of age (MLDA-19), as well as at the aggregated national level.
ual crimes, and all remaining crimes uncategorized (e.g., breach of
probation, prostitution-related offenses, gambling charges, fabri-
cating evidence)] were excluded from the analyses because of their 2.7.3. Sensitivity analyses. To assess the reliability of our pri-
heterogeneity, relatively small sample sizes, and a lack of a strong mary local regression models with triangular-kernel weighting,
empirical link to alcohol consumption. we undertook a series of local regression models with varying
weighting approaches (i.e., triangular kernel versus uniform ker-
2.5. UCR count data nel), functional forms (i.e., linear versus quadratic), and bandwidth
selection strategies. In addition, to assess the reliability of our pri-
The primary analyses in the current study used counts of the mary results, we also used linear parametric regression models
most serious offense within each incident listed in the UCR system, with a restricted range of 26 points (age-in-weeks) before and after
pooled across the study period (2009–2013). We chose to analyze the MLDA to model the impact of the MLDA on the major crime out-
counts—rather than rates (which would be based on counts in the come categories. Given that estimates of the impacts of the MLDA
numerator and population statistics in the denominator)—for two may be confounded with the effects of hazardous birthday drink-
reasons: the denominator for our primary age-in-weeks indepen- ing at the MLDA (Callaghan et al., 2014c), the primary local linear
dent variable cannot be calculated (or imputed accurately) from regression models were re-run without the MLDA birthday-week
population statistics; and the use of census-based data for the cal- values.
culation of rates might bias the estimate of the “jump” at the MLDA
because the census-based population numbers can only be gener-
ated for age-in-years groupings (e.g., 17-year-olds) and, as a result, 3. Results
these denominators would change immediately at the potential
MLDA discontinuity. We limited the analysis to 208 weeks-of-age Of all non-traffic-related crimes committed by individuals aged
(approximately 4 years) before and after MLDA (14–22 years). 14–22 years during the period from 2009 to 2013, 57.3% were
comprised of our main outcome categories—violent crimes, prop-
2.6. Perpetrator’s age erty crimes, and disorderly conduct—and 14.3% were comprised
of cannabis-related and narcotics-related crimes. Of the remaining
We calculated the age of the perpetrator at the time of the crime data, motor vehicle-related offenses accounted for 4.7% and
offense in terms of age-in-weeks. “other” crimes accounted for another 23.7% of all crimes.
Given that the R rdrobust package (Calonico et al., 2014a; R Core
2.7. Analytic plan Team, 2014) does not allow for the inclusion of interaction terms
in local regression models [and given that we are not aware of any
2.7.1. Regression-discontinuity. We employed a regression- of the other standard statistical packages (e.g., Stata, SAS, SPSS)
discontinuity (RD) design (Thistlethwaite and Campbell, 1960; allowing for such], we employed 3 quadratic parametric regres-
Shadish et al., 2002)—a quasi-experimental approach which can sion models using provinces as a random effect variable to assess
provide credible estimates of the causal effect of an intervention potential gender-by-MLDA effects across each our three main
on a specified outcome (Lee and Lemieux, 2009). The RD design crime outcomes. The results showed highly significant gender-
takes advantage of the sharp discontinuity in the legality of by-MLDA effects vis-à-vis each of the three outcomes modeled.
alcohol purchasing and consumption appearing at the MLDA: As a result, these findings, along with our previous work demon-
our approach assigns individuals younger than the MLDA to the strating differential impacts of the MLDA across gender (Callaghan
“alcohol-restricted” group and young adults no longer subject to et al., 2013a, 2013b, 2014a, 2014b), supported the construction of
the MLDA to the “alcohol-accessible” group. The primary, intuitive gender-specific models. However, analyses were also presented for
idea of the RD approach is that individuals slightly older than the combined sample (males and females) to facilitate comparison
the MLDA and those slightly younger than the MLDA will be with prior work (Joksch and Jones, 1993; Carpenter and Dobkin,
similar on observed (and unobserved) characteristics—except 2015).
for the influence of the removal of the MLDA in the alcohol- Table 1 shows our main RD results for males and females.
accessible group. The RD design assumes that all observed and In comparison to males slightly younger than the MLDA,
unobserved variables (which might influence the crime outcomes) those just older than the MLDA had sharp increases in: all
are smoothly distributed across the age-cutoff (Hahn et al., 2001), crimes, 7.6% (95% CI = 3.7%-11.5%, P < 0.001); violent crimes,
and the effects of the MLDA can be inferred if the regression line 7.4% (95% CI = 0.2%–14.6%, P = 0.043); property crimes, 4.8% (95%
shows a discontinuity at the MLDA (Imbens and Lemieux, 2007). CI = 0.0%–9.5%, P = 0.049); and disorderly conduct, 29.4% (95%
In other words, since the observed and unobserved determinants CI = 15.6%–43.3%, P < 0.001). Among females, national criminal
of the crime outcomes (other than the legal granting of access to incidents increased sharply immediately following the MLDA
alcohol) are likely to be distributed smoothly across the MLDA for: all crimes 10.4% (95% CI = 3.8%–17.0%, P = 0.002); violent
threshold, any sharp increases in the number of criminal incidents crimes, 14.9% (95% CI = 6.4%–23.2%, P = 0.001); and disorderly
immediately following the MLDA can reasonably be attributed to conduct, 35.3% (95% CI = 11.6%–58.9%, P = 0.004). There was no
legal availability of alcohol. evidence of significant national increases immediately follow-
ing the MLDA for males or females in our quasi-control
2.7.2. Primary RD analyses. Non-parametric local regression: In cannabis/narcotics categories [cannabis: post-MLDA increase, 0.5%
our primary nonparametric local regression analyses, we used a (95% CI = −3.3%–4.4%, P = 0.785); narcotics: post-MLDA increase,
recent approach (Calonico et al., 2014b) implemented in the R 1.2% (95% CI = −5.7%–8.2%, P = 0.744).
package—rdrobust (Calonico et al., 2014a; R Core Team, 2014)—to Figs. 1 and 2 display scatterplots of female and male national
fit the robust RD models with bias correction. We also used the crime outcomes across age-in-weeks; “jumps” in crime incidents
triangular kernel weighting approach and a recently proposed among males and females can be seen immediately following the
70 R.C. Callaghan et al. / Drug and Alcohol Dependence 167 (2016) 67–74

Table 1
Regression-discontinuity results for local regression models with triangular kernel weighting across categories of crime in Canada, 2009–2013.

Absolute Effect (Jump in Number of Weekly Arrests) Relative Effect (Jump in% of Weekly Arrests)

Categories of Crime MLDA−18a MLDA−19b Total Canadac MLDA-18 MLDA-19 Total Canada
All Crimes
Female 47.0 [16.3;77.6]** 38.8 [8.0;69.5]* 93.0 [33.7;152.3]** 14.7% [5.1%;24.2%] 6.7% [1.4%;12.0%] 10.4% [3.8%;17.0%]
P-value P = 0.003 P = 0.014 P = 0.002
Male 55.2 [1.1;109.3]* 205.9 [110.8;301.0]*** 260.7 [126.6;394.9]*** 4.0% [0.1%;7.9%] 10.1% [5.4%;14.7%] 7.6% [3.7%;11.5%]
P-value P = 0.046 P < 0.001 P < 0.001
Total 86.4 [24.4;148.5]** 240.5 [131.1;349.9]*** 331.5 [168.7;494.3]*** 5.1% [1.4%;8.7%] 9.2% [5.0%;13.3%] 7.7% [3.9%;11.4%]
P-value P = 0.006 P < 0.001 P < 0.001

Violent Violations
Female 8.1 [−2.1;18.2] 15.9 [6.4;25.5]** 26.8 [11.4;42.1]** 10.7% [−2.8%;24.0%] 15.1% [6.1%;23.9%] 14.9% [6.4%;23.2%]
P-value P = 0.119 P = 0.001 P = 0.001
Male −1.2 [−22.4;19.9] 50.2 [17.5;82.9]** 46.1 [1.4;90.8]* −0.5% [−8.5%;7.5%] 14.1% [4.9%;23.3%] 7.4% [0.2%;14.6%]
P-value P = 0.909 P = 0.003 P = 0.043
Total 8.3 [−21.7;38.2] 65.3 [28.9;101.8]*** 70.2 [11.9;128.5]* 2.4% [−6.4%;11.2%] 14.2% [6.3%;22.0%] 8.8% [1.5%;16.0%]
P-value P = 0.589 P < 0.001 P = 0.018

Property Violations
Female 15.5 [−3.1;34.1] −4.1 [−21.6;13.5] 15.9 [−14.3;46.1] 13.0% [−2.7%;28.4%] −2.0% [−10.6%;6.5%] 4.9% [−4.5%;14.2%]
P-value P = 0.103 P = 0.650 P = 0.302
Male 16.3 [−12.5;45.1] 40.5 [2.2;78.8]* 51.3 [0.2;102.4]* 3.4% [−2.6%;9.5%] 6.7% [0.4%;12.9%] 4.8% [0.02%d ;9.5%]
P-value P = 0.267 P = 0.038 P = 0.049
Total 28.5 [−4.2;61.2] 32.4 [−9.2;73.9] 63.8 [9.0;118.7]* 4.8% [−0.7%;10.3%] 4.0% [−1.1%;9.0%] 4.6% [0.6%;8.4%]
P-value P = 0.088 P = 0.127 P = 0.023

Disorderly Conduct
Female 9.9 [4.1;15.8]** 10.1 [1.5;18.7]* 19.8 [6.5;33.0]** 46.3% [19.0%;73.8%] 29.6% [4.5%;54.4%] 35.3% [11.6%;58.9%]
P-value P = 0.001 P = 0.021 P = 0.004
Male 17.1 [8.0;26.1]*** 33.0 [16.5;49.5]*** 48.6 [25.9;71.3]*** 29.2% [13.6%;45.1%] 30.8% [15.3%;46.4%] 29.4% [15.6%;43.3%]
P-value P < 0.001 P < 0.001 P < 0.001
Total 27.1 [13.6;40.7]*** 40.7 [17.7;63.7]** 68.3 [33.3;103.2]*** 33.9% [16.9%;51.2%] 28.5% [12.4%;44.8%] 30.8% [15.0%;46.8%]
P-value P < 0.001 P = 0.001 P < 0.001

Cannabis-related Violations
Female 1.1 [−2.8;5.0] 1.5 [−4.7;7.6] 0.9 [−5.2;7.1] 5.9% [−15.1%;26.4%] 2.6% [−8.2%;13.5%] 1.3% [−6.9%;9.4%]
P-value P = 0.581 P = 0.636 P = 0.763
Male −3.7 [−13.2;5.7] −1.0 [−20.8;18.7] −1.8 [−23.6;20.1] −2.4% [−8.6%;3.8%] −0.3% [−6.3%;5.6%] −0.4% [−4.9%;4.2%]
P-value P = 0.442 P = 0.917 P = 0.874
Total −0.7 [−10.8;9.3] 2.8 [−17.4;22.9] 3.0 [−18.4;24.3] −0.4% [−6.4%;5.5%] 0.7% [−4.5%;5.9%] 0.5% [−3.3%;4.4%]
P-value P = 0.889 P = 0.787 P = 0.785

Narcotics-related Violationse
Female 0.8 [−1.6;3.2] 1.2 [−2.3;4.7] 3.3 [−1.7;8.2] 13.4% [−26.2%;52.7%] 8.9% [−16.9%;34.8%] 17.2% [−9.0%;43.3%]
P-value P = 0.507 P = 0.500 P = 0.198
Male 0.1 [−4.3;4.6] −0.1 [−6.3;6.0] 0.3 [−7.2;7.9] 0.4% [−12.1%;13.1%] −0.2% [−9.7%;9.5%] 0.4% [−7.2%;8.0%]
P-value P = 0.948 P = 0.967 P = 0.927
Total 0.4 [−4.6;5.5] 0.8 [−5.8;7.3] 1.4 [−6.9;9.7] 1.1% [−10.9%;13.2%] 1.0% [−7.4%;9.5%] 1.2% [−5.7%;8.2%]
P-value P = 0.864 P = 0.821 P = 0.744
***
P-value < 0.001.
**
P-value < 0.01.
*
P-value < 0.05.
a
Data were aggregated across provinces with a minimum legal drinking age (MLDA) of 18 years old (i.e. Quebec, Manitoba, and Alberta).
b
Data were aggregated across provinces with an MLDA of 19 years old (the remaining 7 provinces).
c
Data were aggregated across all 10 Canadian provinces.
d
The lower limit of the 95% confidence interval for male property crimes across Canada was reported to 2 decimal places to avoid a 0 value.
e
The Narcotics-related violations category contains all illicit drug-related crimes, except for those related to cannabis (e.g., heroin, ecstasy, cocaine, methamphetamine).
As a result, it is a heterogeneous crime outcome.

MLDA. Fig. 3 provides a bar chart detailing the relative increases Table 1) but without conventional statistical significance (at
(expressed in%) in crimes occurring immediately after the MLDA. P < 0.05). In addition, the sensitivity analyses replicating the
primary local regression results—but excluding the MLDA birthday-
week values—demonstrated somewhat attenuated effect sizes but
3.1. Sensitivity analyses
similar patterns of statistical significance as those indicated in our
primary local regression models.
In comparison to the primary local regression results, the esti-
To assess the possibility that the observed effects were due to
mated MLDA effects at the national level in the sensitivity analyses
age (i.e., impacts of age-18 effects in MLDA-19 provinces, and age-
for males and females were generally similar in terms of size
19 effects in MLDA-18 provinces) or age-of-majority (i.e., impacts
and statistical significance for: local regression models with lin-
of gaining adult status), we conducted 60 additional regression-
ear or quadratic forms and either uniform- or triangular-kernel
discontinuity analyses associated with these potential effects. The
weighting schemes; and linear parametric models using 26 data
results demonstrated very little evidence that crime outcomes were
points before and after the MLDA. There were some inconsis-
associated with age-18, age-19, or age-of-majority variables (see
tencies across models for male property-crime outcomes—which
Supplemental Tables 2–3).
had similarly-sized post-MLDA increases (as those indicated in
R.C. Callaghan et al. / Drug and Alcohol Dependence 167 (2016) 67–74 71

Fig. 1. Number of female crime incidents in Canada (2009–2013) by age-in-weeks leading up to and following the minimum legal drinking age (MLDA).

Fig. 2. Number of male crime incidents in Canada (2009–2013) by age-in-weeks leading up to and following the minimum legal drinking age (MLDA).

4. Discussion showed immediate and sharp post-MLDA increases of approxi-


mately 20–35% in national alcohol-impaired driving arrests among
The current study demonstrated that release from drinking- males and females (Callaghan et al., 2016). In light of our study
age restrictions is associated with significant and sharp increases results, it is important that current policy discussions consider the
not only in male and female perpetration of violent crimes, but neglected role of drinking-age legislation in preventing alcohol-
also nonviolent crimes, such as disorderly conduct and property related violent and nonviolent crimes in underage populations.
crimes. Given that our “control” crime categories—cannabis-related Our study demonstrated that violent crime increased by approx-
and narcotics-related crimes—showed no evidence of a signifi- imately 9% immediately after the MLDA, and this pattern is similar
cant relation with the MLDA, this pattern strengthens our primary to the approximate 7.2% increase in the post-MLDA violent crime
results. In addition, our prior work using the same UCR data system rate found in a recent regression-discontinuity study using Cali-
72 R.C. Callaghan et al. / Drug and Alcohol Dependence 167 (2016) 67–74

Fig. 3. Relative change (%) in the number of national criminal incidents immediately following the minimum legal drinking age (MLDA).

fornia arrest data (Carpenter and Dobkin, 2015). Nonetheless, the age restrictions. In particular, the national estimates for post-MLDA
scant prior research in this area has produced mixed results, with increases from our study for property crimes (4.6%) and disor-
most studies showing no evidence of a relation between MLDA derly conduct (30.8%) are higher than similar outcomes described
legislation and violent crime. In addition, some studies did not in an earlier RD approach, which found a small increase in
address gender effects (Joksch and Jones, 1993; Carpenter and property crimes (2.2%) and a negligible change in disorderly con-
Dobkin, 2015), or did not have adequate statistical power to assess duct/vagrancy (Carpenter and Dobkin, 2015). For the most part,
female violent crimes (Smith and Burvill, 1987; Davis and Reynolds, however, the prior evidence in this area has shown that MLDA leg-
1990). While the current study did not assess specific mecha- islation has an impact on nonviolent crime, with higher MLDAs
nisms of how drinking-age legislation might affect violent crime, associated with reduced patterns of nonviolent crimes in the
it is possible to speculate. First, transition across the MLDA may affected age groups and lower MLDAs associated with increases
be associated with increased alcohol consumption, especially haz- (Smith and Burvill, 1987; Davis and Reynolds, 1990; Joksch and
ardous drinking (Smart and Schmidt, 1975; Carpenter and Dobkin, Jones, 1993; Carpenter, 2005; Carpenter and Dobkin, 2015). This
2009). Increased levels of hazardous alcohol consumption leads to prior literature, however, has not been able to estimate the differ-
a greater propensity towards the escalation of violence not only ential impact of drinking-age legislation across gender. In our study,
because of the biological effects of alcohol on risk-taking, impaired both males and females show sharp and significant increases in
cognition, emotion regulation, and aggression (Boles and Miotto, nonviolent criminal incidents immediately following the legislated
2003; Giancola, 2004; Parrott and Giancola, 2006; Giancola et al., drinking-age.
2009), but also because hazardous drinking often takes place in The current study has a number of potential limitations. Post-
public settings in which violence may be more likely to occur—such MLDA increases in police-reported criminal incidents may be due
as bars, nightclubs, and other drinking establishments (Graham and to increases in alcohol consumption and impairment in the per-
West, 2001; Wells et al., 2005). Bar-and-nightclub settings may petrator, leading to a greater probability of arrest (i.e., alcohol use
have a number of factors associated with increased propensity for makes for sloppier criminals)—rather than actual population-level
violence, such as the presence of other intoxicated individuals (and increases in criminal behavior. However, this is likely a negligible
suitable victims), low levels of social control, and increased promi- issue, as it would be expected that such a pattern would lead to
nence of power concerns (e.g., “macho” posturing and male honor; increases in all specific crime categories—including the illicit-drug
Graham and Wells, 2003), as well as overarching setting-specific “control” offenses—rather than the variation seen in the current
(Graham et al., 2006) and cultural expectations which might con- study. In addition, this argument would have difficulty accounting
done violent offending (Graham et al., 2000; Graham and West, for the sharp and immediate increases in alcohol-impaired driv-
2001). Thus, drinking-age legislation might exert its influence on ing incidents in the UCR database occurring immediately after the
violent crime not only by increasing individual-level episodes of MLDA—a pattern we describe in an earlier paper (Callaghan et al.,
hazardous drinking, but also by shifting, to some extent, the loca- 2016). Also, the regression-discontinuity approach assumes that
tion of alcohol consumption to settings which may carry a greater potentially confounding variables are smoothly distributed across
propensity for violent offending (Graham and West, 2001; Graham the MLDA cutoff. Given that late adolescence and early adulthood
et al., 2006). is a time of significant life transitions, it is possible that changes
Our findings also add to the existing literature by indicating associated with such periods [e.g., in educational attainment, liv-
that transition across Canada’s legal drinking age is associated ing status (e.g., living at home with parents), socioeconomic status]
with significant and immediate population-level increases in non- may have affected our results. However, it seems unlikely that
violent crimes, such as property crimes and disorderly conduct, life-transition variables would have strongly influenced our esti-
among young men and women newly discharged from drinking- mates of MLDA impacts because young people near the MLDA
R.C. Callaghan et al. / Drug and Alcohol Dependence 167 (2016) 67–74 73

would be quite similar on these life-transition patterns, and these Conflict of interest
life-transition variables would likely have had an impact on both
alcohol-related crime and our quasi-control crime categories. There No conflict declared.
was no evidence of significant jumps, however, appearing imme-
diately after the MLDA in cannabis- or narcotics-related crimes Acknowledgements
in our study. Nonetheless, the possibility of confounding by these
lifestyle variables increases as the timespan in age-in-weeks from Kathy Aucoin and Jacob Greenland at the Canadian Centre for
the MLDA increases. It is not possible to untangle the potential Justice Studies (Statistics Canada) provided invaluable assistance
confound between the direct effects of release from drinking age during the data-request and data acquisition phases of the project,
laws from the concomitant effects of gaining of legal access to and they as well as Sara Dunn contributed to the interpretation of
bar and nightclub settings—locations which likely carry a greater results.
propensity (than usual pre-MLDA drinking contexts) for the com-
mission of crimes and police involvement. Crimes occurring at bars Appendix A. Supplementary data
and nightclubs (and “open-air” areas surrounding such establish-
ments) accounted for approximately 33% of all criminal incidents Supplementary data associated with this article can be found, in
(non-traffic and traffic violations) in the age group 14–22 years. In the online version, at http://dx.doi.org/10.1016/j.drugalcdep.2016.
criminal incidents occurring at such locations, the MLDA may have 07.023.
a more distal effect by not only allowing alcohol consumption, but
also legal access to these locations. As well, for incidents involv- References
ing multiple offenses with identical maximum penalties, police
department personnel must assign the most serious offense in Boles, S.M., Miotto, K., 2003. Substance abuse and violence. Aggress. Violent Behav.
the specific incident record based on individual discretion. The 8, 155–174.
Boyce, J., 2015. Police-reported Crime Statistics in Canada, 2014 Juristat. Catalogue
variance in recording of the most serious offense would only No. 85-002-X. Statistics Canada, Ottawa, ON.
bias our estimates if there were differential and systematic biases CBC News, 2013. Drinking Age Will Remain 19 in Saskatchewan. CBC News,
in the recording of the most serious offenses for the age group Available at:
http://www.cbc.ca/news/canada/saskatchewan/story/2013/03/04/sk-
slightly younger than the MLDA versus those just older. At this drinking-age-1303.html (accessed
time, we are not aware of any evidence to support this possi- 11.11.5).
bility. Also, the regression-discontinuity design cannot provide Callaghan, R.C., Sanches, M., Gatley, J.M., 2013a. Impacts of the minimum legal
drinking age legislation on in-patient morbidity in Canada, 1997–2007: a
specific estimates of potential crime reductions associated with
regression-discontinuity approach. Addiction 108, 1590–1600.
raising the MLDA to 19 or 21 years. However, study findings do Callaghan, R.C., Sanches, M., Gatley, J.M., Cunningham, J.K., 2013b. Effects of the
suggest that drinking-age laws limit crime patterns among youth minimum legal drinking age on alcohol-related health service use in hospital
under drinking-age restrictions and, as a result, it seems rea- settings in Ontario: a regression-discontinuity approach. Am. J. Public Health
103, 2284–2291.
sonable to expect that if the MLDA were raised, alcohol-related Callaghan, R.C., Sanches, M., Gatley, J.M., Stockwell, T., 2014a. Impacts of
crime patterns would be attenuated in the newly alcohol-restricted drinking-age laws on mortality in Canada, 1980–2009. Drugs Alcohol Depend.
groups. 138, 137–145.
Callaghan, R.C., Gatley, J.M., Sanches, M., Asbridge, M., 2014b. Impacts of the
Despite these potential limitations, study results provide impor- minimum legal drinking age on motor vehicle collisions in Québec,
tant population-based evidence that release from drinking-age 2000–2012. Am. J. Prev. Med. 47, 788–795.
restrictions is associated with significant and immediate increases Callaghan, R.C., Sanches, M., Gatley, J.M., Liu, L., Cunningham, J.K., 2014c.
Hazardous birthday drinking among young people: population-based impacts
in male and female criminal behavior, including a range of vio- on emergency department and inpatient hospital admissions. Addiction 109,
lent and nonviolent crimes. MLDA effects on non-traffic-related 1667–1675.
crime rarely have been examined, and the current findings Callaghan, R.C., Gatley, J.M., Sanches, M., Asbridge, M., Stockwell, T., 2016. Impacts
of drinking-age legislation on alcohol-impaired driving crimes among young
should be incorporated into the current debates and policy
people in Canada, 2009–2013. Addiction 111, 994–1003.
discussions. Calonico, S., Cattaneo, M.D., Titiunik, R., 2014a. Package ‘rdrobust’: Robust
Data-driven Statistical Inference in Regression-discontinuity Designs (Version
0.70) 2014, Software available at:
http://cran.r-project.org/web/packages/rdrobust/index.html (accessed
Role of funding source 11.11.15).
Calonico, S., Cattaneo, M.D., Titiunik, R., 2014b. Robust nonparametric confidence
The current study was funded by an Open Operating Grant intervals for regression-discontinuity designs. Econometrica 82, 2295–2326.
Canadian Centre for Justice Statistics, Statistics Canada, 2013a. Uniform Crime
from the Canadian Institutes of Health Research (MOP − 133699) Reporting Incident-Based Survey, Updated January 7th 2013. Canadian Centre
awarded to the lead author (RCC). The funding body had no role in for Justice Statistics, Statistics Canada, Ottawa, ON.
the data collection, data analysis, data interpretation, preparation Canadian Centre for Justice Statistics, Statistics Canada, 2013b. Uniform Crime
Reporting Incident-Based Survey: Research Data Centre User Manual.
or submission of the manuscript. Canadian Centre for Justice Statistics, Statistics Canada, Ottawa, ON.
Canadian Public Health Association, 2011. Canadian Public Health Association
Position Paper. Too High a Cost: A Public Health Approach to Alcohol Policy in
Contributors Canada. Canadian Public Health Association, Ottawa, ON.
Carpenter, C., Dobkin, C., 2009. The effect of alcohol consumption on mortality:
regression discontinuity evidence from the minimum drinking age. Am. Econ.
All authors have made substantial contributions warranting J. Appl. Econ. 1, 164–182.
authorship on the current manuscript. RCC conceived of the study Carpenter, C., Dobkin, C., 2015. The minimum legal drinking age and crime. Rev.
Econ. Stat. 97, 521–524.
and supervised the statistical analyses and manuscript preparation; Carpenter, C.S., 2005. Heavy alcohol use and the commission of nuisance crime:
JG provided a critical review of relevant literature and contributed evidence from underage drunk driving laws. Am. Econ. Rev. 95, 267–272.
to interpretation of the statistical analysis and manuscript prepa- Davis, J.E., Reynolds, N.C., 1990. Alcohol use among college students: responses to
raising the purchase age. J. Am. Coll. Health 38, 263–269.
ration; MS conducted the statistical analyses and contributed to DeJong, W., Blanchette, J., 2014. Case closed: research evidence on the positive
manuscript preparation; CB contributed to the review of relevant public health impact of the age 21 minimum legal drinking age in the United
literature and contributed to manuscript preparation. All authors States. J. Stud. Alcohol Drugs Suppl. 17, 108–115.
Giancola, P.R., Levinson, C.A., Corman, M.D., Godlaski, A.J., Morris, D.H., Phillips, J.P.,
have given their approval for the current version to be submitted
Holt, J.C., 2009. Men and women: alcohol and aggression. Exp. Clin.
for peer-review at the journal. Psychopharmacol. 17, 154–164.
74 R.C. Callaghan et al. / Drug and Alcohol Dependence 167 (2016) 67–74

Giancola, P.R., 2004. Executive functioning and alcohol-related aggression. J. McCartt, A.T., Hellinga, L.A., Kirley, B.B., 2010. The effects of minimum legal
Abnorm. Psychol. 113, 541–555. drinking age 21 laws on alcohol-related driving in the United States. J. Saf. Res.
Giesbrecht, N., Wettlaufer, A., April, N., Asbridge, M., Cukier, S., Mann, R., 41, 173–181.
McAllister, J., Murie, A., Plamondon, L., Stockwell, T., Thomas, G., Thompson, K., Owen, B., 2014. No Hike to Drinking Age Coming: Province Won’t Make It 19
Vallance, K., 2013. Strategies to Reduce Alcohol-Related Harms and Costs in Despite Looming Report. Winnipeg Free Press, Available at:
Canada: A Comparison of Provincial Policies. Centre for Addiction and Mental http://www.winnipegfreepress.com/local/no-hike-to-drinking-age-coming-
Health, Toronto, ON. 274994161.html (accessed
Gore, F.M., Bloem, P.J., Patton, G.C., Ferguson, J., Joseph, V., Coffey, C., Sawyer, S.M., 11.11.15).
Mathers, C.D., 2011. Global burden of disease in young people aged 10–24 Parker, R.N., 1995. Alcohol policy and crime control. In: Parker, R.N., Rebhun, L.
years: a systematic analysis. Lancet 377, 2093–2102. (Eds.), Alcohol and Homicide: A Deadly Combination to Two American
Government of Canada, 2015. Criminal Code (R.S.C. 1985, c. C-46). Government of Traditions. State University of New York Press, Albany, NY, pp. 102–117.
Canada, Ottawa ON, Available at: http://laws-lois.justice.gc.ca/eng/acts/C-46/ Parrott, D.J., Giancola, P.R., 2006. The effect of past-year heavy drinking on
(accessed 11.11.15). alcohol-related aggression. J. Stud. Alcohol 67, 122–130.
Graham, K., Wells, S., 2003. ‘Somebody’s gonna get their head kicked in tonight!’ Pernanen, K., Cousineau, M., Brochu, S., 2002. Proportions of Crimes Associated
Aggression among young males in bars—a question of values? Br. J. Criminol. with Alcohol and Other Drugs in Canada. Canadian Centre on Substance Abuse,
43, 546–566. Ottawa, ON.
Graham, K., West, P., 2001. Alcohol and crime: examining the link. In: Heather, N., Perreault, S., 2013. Impaired Driving in Canada, 2011 Juristat. Catalogue No.
Peters, T.J., Stockwell, T. (Eds.), International Handbook of Alcohol Dependence 85-002-X. Statistics Canada, Ottawa, ON.
and Problems. John Wiley and Sons Ltd, Toronto, ON, pp. 439–470. Pitts, J.R., Johnson, I.D., Eidson, J.L., 2014. Keeping the case open: responding to
Graham, K., West, P., Wells, S., 2000. Evaluating theories of alcohol-related DeJong and Blanchette’s case closed on the minimum legal drinking age in the
aggression using observations of young adults in bars. Addiction 95, 847–863. United States. J. Stud. Alcohol Drugs 75, 1047–1049.
Graham, K., Bernards, S., Osgood, D.W., Wells, S., 2006. Bad nights or bad bars? R Core Team, 2014. R: A Language and Environment for Statistical Computing
multi-level analysis of environmental predictors of aggression in late-night (Version 3 1.2). R Foundation for Statistical Computing, Vienna.
large-capacity bars and clubs. Addiction 101, 1569–1580. Shadish, W.R., Cook, T.D., Campbell, D.T., 2002. Experimental and
Greenfeld, L.A., 1998. Alcohol and Crime: An Analysis of National Data on the Quasi-Experimental Designs for Generalized Causal Inference. Houghton Miflin
Prevalence of Alcohol Involvement in Crime. U.S. Department of Justice, Office Company, Boston, NY.
of Justice Programs, Washington, DC. Smart, R.G., Schmidt, W., 1975. Drinking and problems from drinking after a
Hahn, J., Todd, P., Van der Klaauw, W., 2001. Identification and estimation of reduction in the minimum drinking age. Br. J. Addict. Alcohol Other Drugs. 70,
treatment effects with a regression-discontinuity design. Econometrica 69, 347–358.
201–209. Smith, D.I., Burvill, P.W., 1987. Effect on juvenile crime of lowering the drinking
Imbens, G.W., Lemieux, Thomas, 2007. Regression discontinuity designs: a guide to age in three Australian states. Brit. J. Addict. 82, 181–188.
practice. J. Econom. 142, 615–635. Statistics Canada, 2015. Uniform Crime Reporting Survey (UCR). Statistics Canada,
International Center for Alcohol Policies, 2015. Policy Table: Minimum Age Limits Ottawa ON, Available at:
Worldwide. International Center for Alcohol Policies, Washington, DC, http://www23.statcan.gc.ca/imdb/p2SV.pl?Function=getSurvey&SDDS=3302
Available at: (accessed 21.10.15).
http://icap.org/PolicyIssues/YoungPeoplesDrinking/PolicyTableMinimumAgeLimits Thistlethwaite, D.L., Campbell, D.T., 1960. Regression-discontinuity analysis: an
Worldwide/tabid/206/Default.aspx (accessed 11.1115). alternative to the ex post facto experiment. J. Educ. Psychol. 51, 309–317.
Joksch, H.C., Jones, R.K., 1993. Changes in the drinking age and crime. J. Crim. Just. Wagenaar, A.C., Toomey, T.L., 2002. Effects of minimum drinking age laws: review
21, 209–221. and analyses of the literature from 1960 to 2000. J. Stud. Alcohol. Suppl. 14,
Kypri, K., Davie, G., McElduff, P., Connor, J., Langley, J., 2014. Effects of lowering the 206–225.
minimum alcohol purchasing age on weekend assaults resulting in Wells, S., Graham, K., Speechley, M., Koval, J.J., 2005. Drinking patterns, drinking
hospitalization in New Zealand. Am. J. Public Health 104, 1396–1401. contexts and alcohol-related aggression among late adolescent and young
Lee, D.S., Lemieux, T., 2009. Regression discontinuity designs in economics. J. Econ. adult drinkers. Addiction 100, 933–944.
Lit. 48, 281–355.
MADD Canada, 2014. Provicial Liquor Boards: Meeting the Best Interests of
Canadians MADD Canada Policy Backgrounder. MADD Canada, Oakville, ON.

You might also like