Explaining NSW long term trends in property and violent crime Steve Moffatt and Lucy Snowball NSW Bureau of Crime Statistics and Research.

Explaining NSW long term trends in property and violent crime

Steve Moffatt and Lucy Snowball

NSW Bureau of Crime Statistics and Research

Purpose of research

• Determine the general structure of trends and seasonality

• Explain some exogenous influences on crime trends, particularly those useful for forecasting

• Forecasts for state and regions• Test scenarios

Background ~ property crime

Long term rise (1990s) followed by fall in property crime recorded incidents since 2000– Motor vehicle theft, steal from motor vehicle, dwelling,

retail store, person

– Robbery

– Break and enter

– Receiving/handling stolen goods

– Fraud (stabilised after rise)

Property crime (theft + robbery) NSW 95-07

Cubic trend

R2 = 0.84

0

5000

10000

15000

20000

25000

30000

35000

40000

45000

50000

1995

/Jan

1995

/Jun

1995

/Nov

1996

/Apr

1996

/Sep

1997

/Feb

1997

/Jul

1997

/Dec

1998

/May

1998

/Oct

1999

/Mar

1999

/Aug

2000

/Jan

2000

/Jun

2000

/Nov

2001

/Apr

2001

/Sep

2002

/Feb

2002

/Jul

2002

/Dec

2003

/May

2003

/Oct

2004

/Mar

2004

/Aug

2005

/Jan

2005

/Jun

2005

/Nov

2006

/Apr

2006

/Sep

2007

/Feb

2007

/Jul

2007

/Dec

Re

cord

ed i

nci

den

ts

Property offences

Poly. (Property offences)

Background ~ violent crime

Steep rise (1990s) followed by flattening rise since 2001 in violent crime recorded incidents– Assault– Sexual assault– Harassment– Other offences against the person

[Stable or falling murder, attempted murder, manslaughter,

blackmail, extortion]

Violent recorded crime NSW 95-07

Quadratic trend

R2 = 0.85

0

2000

4000

6000

8000

10000

12000

1995

/Jan

1995

/Jun

1995

/Nov

1996

/Apr

1996

/Sep

1997

/Feb

1997

/Jul

1997

/Dec

1998

/May

1998

/Oct

1999

/Mar

1999

/Aug

2000

/Jan

2000

/Jun

2000

/Nov

2001

/Apr

2001

/Sep

2002

/Feb

2002

/Jul

2002

/Dec

2003

/May

2003

/Oct

2004

/Mar

2004

/Aug

2005

/Jan

2005

/Jun

2005

/Nov

2006

/Apr

2006

/Sep

2007

/Feb

2007

/Jul

2007

/Dec

Rec

ord

ed i

nci

den

ts

Violent offences

Poly. (Violent offences)

Background ~ Summary

• Fall in property crime incidents• Coincided with continuation of upward trend in

violent crime incidents• Demand for short term forecasting at state and

local area level• Previous trend research has focused more on

property crime• Few clues on why violent crime trend persisting,

recent focus on alcohol related assaults

Predictors

• Seasonality and month characteristics

• Police and Justice – Police activity, incapacitation, deterrence

• Alcohol and drug use

• Economic cycles

General Models

Trends (quadratic, cubic)

Seasonality (months, weekends)

Police and Justice (POIs by status)

Exogenous influences (economy, drugs)

First equation:...... 11

12

2321

2210

tt

iiitt eeMyWtty

Second equation:

...... 114

,

3

11,1

12

2

2210

tt

l

ktkk

ktkkt

iiit eeXPWMtty

Model characteristics

• Violent offences model in levels (ARMA)

– Quadratic trend

• Property offences in differences (ARIMA)

– Cubic trend

• Lagged dependent variable or POI variables by status

• MA(1) error term

Property crime – POI trends

0

2000

4000

6000

8000

10000

12000

Jan-

95

Jul-9

5

Jan-

96

Jul-9

6

Jan-

97

Jul-9

7

Jan-

98

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Jan-

03

Jul-0

3

Jan-

04

Jul-0

4

Jan-

05

Jul-0

5

Jan-

06

Jul-0

6

Jan-

07

Jul-0

7

proppois

proppoipatc

propoipaottc

Violent crime – POI trends

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

Jan-

95

Jul-9

5

Jan-

96

Jul-9

6

Jan-

97

Jul-9

7

Jan-

98

Jul-9

8

Jan-

99

Jul-9

9

Jan-

00

Jul-0

0

Jan-

01

Jul-0

1

Jan-

02

Jul-0

2

Jan-

03

Jul-0

3

Jan-

04

Jul-0

4

Jan-

05

Jul-0

5

Jan-

06

Jul-0

6

Jan-

07

Jul-0

7

Nu

mb

er o

f P

OIs

rec

ord

ed

violpois

violpoispatc

violpoispaottc

Model results (Violent offences)Results: Violent offences (levels)

Significance Sign problemsTrend Linear coef **

Quadratic coef **Months ** (7 mths)

* (2 mths)Weekends *POIs (lag1) Charged ** *

Attend Court *No action taken **

Incapacitation (lag 5)

* Weakly significant .05<p<.10** Significant p<.05

This model underestimated total violent offences for 2007 by 2.9%

Forecasts – Violent offencesViolent offences

0

2000

4000

6000

8000

10000

12000

1995

/Jan

1995

/Jun

1995

/Nov

1996

/Apr

1996

/Sep

1997

/Feb

1997

/Jul

1997

/Dec

1998

/May

1998

/Oct

1999

/Mar

1999

/Aug

2000

/Jan

2000

/Jun

2000

/Nov

2001

/Apr

2001

/Sep

2002

/Feb

2002

/Jul

2002

/Dec

2003

/May

2003

/Oct

2004

/Mar

2004

/Aug

2005

/Jan

2005

/Jun

2005

/Nov

2006

/Apr

2006

/Sep

2007

/Feb

2007

/Jul

2007

/Dec

Rec

ord

ed i

nci

den

ts o

f vi

ole

nt

off

ence

s

Violent offences

Forecasts – Violent offences

Violent offences

0

2000

4000

6000

8000

10000

12000

1995

/Jan

1995

/Jun

1995

/Nov

1996

/Apr

1996

/Sep

1997

/Feb

1997

/Jul

1997

/Dec

1998

/May

1998

/Oct

1999

/Mar

1999

/Aug

2000

/Jan

2000

/Jun

2000

/Nov

2001

/Apr

2001

/Sep

2002

/Feb

2002

/Jul

2002

/Dec

2003

/May

2003

/Oct

2004

/Mar

2004

/Aug

2005

/Jan

2005

/Jun

2005

/Nov

2006

/Apr

2006

/Sep

2007

/Feb

2007

/Jul

2007

/Dec

Rec

ord

ed i

nci

den

ts o

f vi

ole

nt

off

ence

s

Violent offences

violentoff fitted

Forecasts – Violent offences

Violent offences

0

2000

4000

6000

8000

10000

12000

1995

/Jan

1995

/Jun

1995

/Nov

1996

/Apr

1996

/Sep

1997

/Feb

1997

/Jul

1997

/Dec

1998

/May

1998

/Oct

1999

/Mar

1999

/Aug

2000

/Jan

2000

/Jun

2000

/Nov

2001

/Apr

2001

/Sep

2002

/Feb

2002

/Jul

2002

/Dec

2003

/May

2003

/Oct

2004

/Mar

2004

/Aug

2005

/Jan

2005

/Jun

2005

/Nov

2006

/Apr

2006

/Sep

2007

/Feb

2007

/Jul

2007

/Dec

Rec

ord

ed i

nci

den

ts o

f vi

ole

nt

off

ence

s

Violent offences

violentoff fitted

Model results (Property offences)

Results: Property offences (differences)

Significance Sign problemsTrend Quadratic coef **

Cubic coefMonths ** (9 mths)

POIs (lag1) ChargedAttend CourtNo action taken *

Incapacitation (lag 5) *Drug activity (lag1) *Economic activity (lag1) *

* Weakly significant .05<p<.10** Significant p<.05

This model underestimated total property offences for 2007 by 2.4%

Forecasts – Property offences

Property offences (theft + robbery)

15000

20000

25000

30000

35000

40000

45000

50000

1995

/Jan

1995

/Jun

1995

/Nov

1996

/Apr

1996

/Sep

1997

/Feb

1997

/Jul

1997

/Dec

1998

/May

1998

/Oct

1999

/Mar

1999

/Aug

2000

/Jan

2000

/Jun

2000

/Nov

2001

/Apr

2001

/Sep

2002

/Feb

2002

/Jul

2002

/Dec

2003

/May

2003

/Oct

2004

/Mar

2004

/Aug

2005

/Jan

2005

/Jun

2005

/Nov

2006

/Apr

2006

/Sep

2007

/Feb

2007

/Jul

2007

/Dec

Rec

ord

ed i

nci

den

ts o

f vi

ole

nt

off

ence

s

Property offences

Forecasts – Property offences

Property offences (theft + robbery)

15000

20000

25000

30000

35000

40000

45000

50000

1995

/Jan

1995

/Jun

1995

/Nov

1996

/Apr

1996

/Sep

1997

/Feb

1997

/Jul

1997

/Dec

1998

/May

1998

/Oct

1999

/Mar

1999

/Aug

2000

/Jan

2000

/Jun

2000

/Nov

2001

/Apr

2001

/Sep

2002

/Feb

2002

/Jul

2002

/Dec

2003

/May

2003

/Oct

2004

/Mar

2004

/Aug

2005

/Jan

2005

/Jun

2005

/Nov

2006

/Apr

2006

/Sep

2007

/Feb

2007

/Jul

2007

/Dec

Rec

ord

ed i

nci

den

ts o

f vi

ole

nt

off

ence

s

Property offences

property off. fitted

Forecasts – Property offences

Property offences (theft +robbery)

15000

20000

25000

30000

35000

40000

45000

50000

1995

/Jan

1995

/Jun

1995

/Nov

1996

/Apr

1996

/Sep

1997

/Feb

1997

/Jul

1997

/Dec

1998

/May

1998

/Oct

1999

/Mar

1999

/Aug

2000

/Jan

2000

/Jun

2000

/Nov

2001

/Apr

2001

/Sep

2002

/Feb

2002

/Jul

2002

/Dec

2003

/May

2003

/Oct

2004

/Mar

2004

/Aug

2005

/Jan

2005

/Jun

2005

/Nov

2006

/Apr

2006

/Sep

2007

/Feb

2007

/Jul

2007

/Dec

Rec

ord

ed i

nci

den

ts o

f vi

ole

nt

off

ence

s

Property offences

property off. fitted

Model selection and forecast accuracy

• Stationarity of dependent variable

• Most appropriate trend

• MLE ARMA/ARIMA

• Log likelihood and Wald Chi Sq

• Error tests and RMSE for forecast

Accuracy vs. Parsimony

• Over fitting (including non significant variables) improves forecast accuracy

• However reduction in significance of model

• Fit for purpose:– Overfitted models useful for forecasting– Parsimonious models useful for determining

which factors influence long term trends

Conclusions

• Can achieve well fitting models for violent and property crime with good forecasting power

• Majority of trend explained using structure (quadratic or cubic), seasonal (month) terms

• Weekend dummy and summer months a good proxy for alcohol consumption

• POIs (clear-up variables) act as a control for autocorrelation

Next steps

• Report state level trends, seasonal components and influences to NSW Police

• Project models from state level to regional level– Demand at local area command level

• Panel data sets for regions• Develop models for other crimes, particularly high

volume offences that are resilient to police activity– Malicious damage– Assault (domestic violence related and non-domestic

violence)– Harassment