neil donnelly, patricia menéndez & nicole mahoney nsw bureau of crime statistics and research...
TRANSCRIPT
Neil Donnelly, Patricia Menéndez & Nicole Mahoney
NSW Bureau of Crime Statistics and Research
February, 2015
BackgroundEvidence of relationship between total liquor
licence concentrations & some harms (e.g. assaults, motor vehicle accidents)
Local areas with a higher no. of liquor outlets have more of these problems (Gruenewald et al., 2006; Chikritzhs et al., 2007)
However some variability about the most important licensed premises type for these harms (e.g. hotels/on-premises or packaged liquor)
Current outlet density study
Investigate the relationship between liquor licence concentrations and assault rates in New South Wales LGAscross-sectional design using 2011 data
Research questions
Is there an association between liquor licence concentrations in LGAs in NSW and:
1. DV related assault rates?
2. Non-DV related assault rates?
after controlling for important covariates
Are concentrations of particular licence types
associated with higher assault rates?
a) Hotel licences
b) Packaged liquor licences
c) On-Premises licences
d) Club licences
Is there a linear or a non-linear relationship
between liquor licence concentration and assault?
Does this differ by liquor licence type?
Spatial autocorrelation between LGAs and assault rates measured & taken account of
Data sourcesRecorded crime
• DV and non-DV assault incidents in 2011 (COPS data)
• DV & non-DV assault rates (per 1,ooo pop in LGAs)
Liquor licensing • Licence types operating in 2011 (OLGR, NSW)
• Hotel rates (per 1,ooo pop in LGAs)
• Packaged liquor rates • On-Premises rates• Club rates
Other LGA data LGA population size (ERP)
LGA population density
% males aged 15-34 yrs
% Indigenous (ATSI)
Socio-economic disadvantage (SEIFA IRSD)
location category (ARIA)
% born in non-English speaking country
LGAs included 147 of 152 LGAs used in final analyses (97% of
LGAs) Exclusions
City of Sydney Snowy River Broken Hill Urana Conargo
One LGA excluded during final analyses diagnostics as an outlier (n=146; 96% of LGAs)
Warren
AnalysesLog transformation of each assault rate
Linear regression (OLS)Moran’s I - spatial autocorrelation present?
Simultaneous Autoregression (SAR)Lambda (λ) – spatial autocorrelation taken account of?
SAR weighted
Diagnostics – model selection
Mean Median25th
percentile75th
percentile
DV related assault rate
(per 1,000 population)
5.13 3.67 2.55 5.68
Non-DV related assault rate
(per 1,000 population)
5.88 4.83 2.88 7.30
SAR weighted model
(n=146)
Estimate SE p value
Constant 7.107 1.328 < .001 *
Hotels linear -0.400 0.175 = .023 *
Hotels non-linear squared -0.324 0.194 = .096
Hotels non-linear cubed 0.236 0.053 < .001 *
Packaged linear -1.647 0.255 < .001 *
Packaged non-linear squared -3.043 0.709 < .001 *
Packaged non-linear cubed 10.974 1.564 < .001 *
On-Premises linear 0.229 0.049 < .001 *
Clubs linear 0.291 0.125 = .020 *
Population density# 0.000 0.000 = .320
Indigenous (%) 0.029 0.006 < .001 *
Males 15-34 years (%) 0.055 0.024 = .023 *
Socio-economic disadvantage -0.007 0.001 < .001 *
Born NES country (%) -0.002 0.013 = .848
City 0.077 0.333 = .817
Outer regional/remote 0.167 0.094 = .075
λ (lambda) = .237
LR test = 0.60, p = .439
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.0 1.0 2.0 3.0DV a
ssau
lt ra
te p
er 1
,000
pop
ulati
on (
log)
Hotel concentration per 1,000 population
Figure 1. Hotel concentration and DV assault rate
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.0 0.5 1.0DV
ass
ault
rat
e pe
r 1,0
00 p
opul
ation
(lo
g)
Packaged liquor concentration per 1,000 population
Figure 2. Packaged liquor concentration and DV assault rate
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.0 1.0 2.0 3.0
DV a
ssau
lt ra
te p
er 1
,000
pop
ulati
on (
log)
On-Premises concentration per 1,000 population
Figure 3. On-Premises concentration and DV assault rate
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.0 0.5 1.0 1.5DV
ass
ault
rat
e p
er 1
,000
pop
ulati
on (
log)
Club concentration per 1,000 population
Figure 4. Club concentration and DV assault rate
DV assault rate – Elasticity effectsLog-Linear
On-Premises10% increase from mean concentration level produced a
2.2% increase in DV assault rate (log)
Clubs10% increase from mean concentration level produced a
1.3% increase in DV assault rate (log)
SAR model
(n = 146)
Estimate SE p value
Constant 6.638 0.681 < .001 *
Hotels linear -0.119 0.108 = .270
Hotels non-linear squared -0.258 0.128 = .045 *
Hotels non-linear cubed 0.146 0.044 = .001 *
Packaged linear -0.852 0.237 < .001 *
Packaged non-linear squared -1.330 0.713 = .062
Packaged non-linear cubed 5.177 1.569 = .001 *
On-Premises linear 0.314 0.051 < .001 *
Clubs linear -0.463 0.200 = .021 *
Clubs non-linear squared 0.494 0.199 = .013 *
Population density# 0.000 0.000 = .057
Indigenous (%) 0.029 0.005 < .001 *
Males 15-34 years (%) 0.090 0.016 < .001 *
Socio-economic disadvantage -0.006 0.001 < .001 *
Born NES country (%) -0.014 0.004 = .001 *
λ (lambda) = .109
LR test = 0.90, p = .342
# Population density estimate is -0.0000586
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.0 0.5 1.0
Non
-DV
assa
ult
rate
per
1,0
00 p
opul
ation
(lo
g)
Packaged liquor concentration per 1,000 population
Figure 6. Packaged liquor concentration and non-DV assault rate
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.0 1.0 2.0 3.0
Non
-DV
ass
ault
rat
e p
er 1
,000
pop
ulati
on (
log)
On-Premises concentration per 1,000 population
Figure 7. On-Premises concentration and non-DV assault rate
Non-DV assault rate – Elasticity effectLog-Linear
On-Premises10% increase from mean concentration level produced a
3.0% increase in non-DV assault rate (log)
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0.0 0.5 1.0 1.5Non
-DV
assa
ult
rate
per
1,0
00 p
opul
ation
(lo
g)
Club concentration per 1,000 population
Figure 8. Club concentration and non-DV assault rate
SummaryDifferent concentration effects found by licence type
and assault type adjusted for important covariates & spatial autocorrelation
Hotels, very strong non-linear predictor of DV & non-DV assault rates
Packaged liquor also a non-linear predictor but not as strong as hotels
On-Premises, strong linear predictor of both assault rates
Clubs strong linear predictor of DV assault rates non-linear predictor of non-DV assault but smaller effect size
LimitationsHotel licences can also supply packaged alcohol
Does not apply to LGAs with a very high transient population
Lack of alcohol sales data
Cross-sectional study, not longitudinal
General conclusionsConsistent with other cross-sectional outlet density
studies strong relationship between high concentrations of licensed
premises and assault rates
Non-linear effects for hotels of particular policy importance
Longitudinal studies also very important to assess effects of changes in the concentration of licence types