1. Neighbourhood and School Level
Variance Partitioning of Adolescent
Offending and Socialization Variables
Prof. Dr. Lieven J.R. Pauwels (Ghent University, Belgium)
Prof. Dr. Gerben J.N. Bruinsma (NSCR, The Netherlands)
Dr. Frank M. Weerman (NSCR, The Netherlands)
Dr. Wim Bernasco (NSCR, The Netherlands)
1
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
2. 2
Introduction
• Research problem: ecological concentration of
offenders in schools and neighbourhoods
• In general: Two types of explanations
• (1) Differential selection causes ecological
segregation of offenders in areas and schools
• (2) Segregation itself causes offending,
independent of background
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
3. Multilevel studies of offending
School contextual effect
• Contextual effects: Boxford,
2006; Pauwels, 2011; 2009;
Lindström, 1995;
Gottfredson, 2001;
Sapouna, 2010; Parcel,
Dufur, & Cornell Zito, 2010
• Beware: sometimes class-
level sometimes school
level
Neighbourhood contextual effect
• Area of residence studies!
• No contextual effects:
Rovers, 1997; Pauwels,
2007, Wikström, 2006
• Small contextual effects:
Oberwittler, 2004, Weijters,
2007 (city-level differences)
• Beware: different measures
of offending and units of
analyses
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
3
4. Are criminologists looking for the
phantom of the neighbourhood?
• Neighbourhoods are dangerous, especially
for researchers!
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
4
5. 5
The problem of theoretical
bifurcation
• MOST contextual delinquency studies have
focused on residential areas as contextual units .
• SOME contextual delinquency studies have
focused on schools as contextual units.
• ALMOST NO delinquency study has focused on
both contextual units simultaneously (some exc.
Oberwittler, 2009; Pauwels, 2011).
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
6. 6
The problem of theoretical
bifurcation
• Social structures are not always hierarchical. People may
belong to more than one grouping at a given hierarchical
level.
• Neighbourhood and school may both have effects on
offending.
– A school may contain adolescents from several neighbourhoods;
adolescents from one neighbourhood may attend different schools.
– Adolescents are nested in a cross-classification of neighbourhood and
school.
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
7. 7
The problem of theoretical
bifurcation
School S1 S2 S3 S4
Adolescents A1 A2 A3 A4 A5 A6 A7 A8 A9 A10
Neighbourhood N1 N2 N3
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
Conclusion: Adolescents may be exposed to criminogenic moral settings
before and after the bell
8. 8
The present study
• (1) Is there significant neighbourhood level variation
in adolescent offending and socialization variables?
• (2) Is there significant neighbourhood and school
level variation in adolescent offending when both
settings are taken into account simulaneously?
• (3) Is there significant neighbourhood and school
level variation in adolescent offending above and
beyond demographic background variables, when
both settings are taken into account simulaneously?
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
9. The present study
• SPAN project NSCR
• sample of 10 schools in the city of The Hague (486,000)
• field work: October 2008-May 2009
• PADS+ questionnaire: 1 interviewer ―› 4 respondents
• 843 adolescents WAVE 1
• age of sample: 12 & 13 and 15 & 16 years old
• sex composition of sample: 55% boys; 45% girls
• ethnic composition of sample: 47% (Moroccans, Turkish,
Antilleans)
• 2010: collection WAVE 2
• 2011-now: analysis of 2 wave panel data
9
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
10. The present study
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
10
Descriptive Statistics
637 5 24 12,10 5,604
637 3 146 96,11 45,279
637
Nhood resp
Schoolresp
Valid N(listw ise)
N Minimum Maximum Mean Std.Deviation
•Nett sample: 65 neighbourhoods and 10
schools (living in the Hague and attending
classes in the Hague)
11. Measurement of Concepts
• Likert-type of additive scales
• PADS-questionnaire (cf. Peterborough study
by P.O. Wikström)
• Adapted to SPAN-questionnaire
• All scale constructs are sufficiently reliable
(see SPAN-publications: Pauwels et al, 2011;
Svensson et al, EJC in press)
11
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
12. Neighbourhood ICC’s empty model
0
1
2
3
4
5
6
7
NHOOD level variance
NHOOD level variance
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
12
14. ICC’s CCM controlled for demographics
0
1
2
3
4
5
6
7
8
9
Unique NHOOD
Unique SCHOOL
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
14
15. Results
• Neighbourhood differences disappear when
control is held for individual level
demographics and school level variance
• Compositional explanation for
neighbourhood differences in offending
• School level differences remain significant
when adjusting for demographics
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
15
16. Conclusion
• There exist good kids from bad
neighbouroods
• And bad kids from good neighbourhoods
Pauwels, L. et al, 2012. Variance
partitioning of offending, ESC Presentation
16
In The Hague schools are much more
segregated with regard to correlates
of offending than neighbourhoods!
Neighbourhoods are much more
segregated with regard to SES and
other demographic make-up