Data source and methodology

Data source
Analytical techniques
Variables in the analysis

Data source

The primary data source for this study is the National Longitudinal Survey of Children and Youth (NLSCY), developed jointly by Human Resources Development Canada and Statistics Canada. The NLSCY is a comprehensive survey that follows the development of children in Canada over time. The survey monitors child development and measures the prevalence of various factors that influence development, both positively and negatively.

The first cycle of the NLSCY, conducted in 1994-1995, interviewed parents of approximately 23,000 children up to and including age 11. They reported information not only about their children, but also about themselves and the children's immediate families, schools and neighbourhoods. In the second and third cycles, the parents of these same children were interviewed. The NLSCY will continue to collect information on these same children every two years as they move into youth and adulthood.

This study is based on cross-sectional files from the third cycle of the survey conducted in 1998-99. The cycle 3 sample contains 31,963 children aged newborn to fifteen years living in one of the ten provinces in 1998-99.

This study focuses on the subgroup of 4,296 youth aged 12-15 years in the cycle 3 cross-sectional file. These youth comprise the oldest age cohort in cycle 3, and are the only respondents to complete the self-report delinquency questions of interest in this study. The cross-sectional data are weighted to represent about 1,661,000 youth from 10 provinces aged 12-15 years.

Analytical techniques

Logistic regression is employed to examine the odds of committing at least one violent or property-related delinquent act in the past year after controlling for the other variables in this study (see What is an odds ratio?).

The results of the analyses presented in this paper assess the extent to which different factors modify the risk of delinquency among males and females while controlling for a number of additional socio-demographic factors. Using logistic regression models, the study tests for the interaction between sex on the one hand, and levels of school commitment and self-reported victimization on the other hand. Interaction is used to describe a situation in which two factors modify the effect of each other with respect to the occurrence of a given outcome.

For example, if an interaction between sex and school commitment is present with respect to delinquency then the association between school commitment and delinquency will be different for males and females. In order to visualize the presence of significant interactions, this paper presents a number of plots showing differences in the male and female likelihood for committing a delinquent act at different levels of the risk and protective factors. Each line on the graph represents the estimated logit, or log odds, for sex by either school commitment or victimization.

Some imputation was undertaken in order to address the problem of partial missing data, or cases where respondents answered some, but not all of the questions related to the concepts of interest in this study. This was the case only when the variable being measured was a scale, or a group of questions that characterize a single concept when added together. For example, the school commitment scale is composed of seven questions. Respondents receive a score based on the sum of their responses for these questions. To avoid losing partial respondents from the analysis, scores were calculated based on the mean for the answers that were provided, but only if at least 50% of the questions in a scale were answered. Examination of partial non-respondents for each scale in the study revealed that this type of response pattern is not related to the scale. For example, based on the answers that were provided, partial non-responders were not more or less likely to be delinquent. Consequently, imputing scores for those who answered at least 50% of the items in a scale provides a reasonable estimate of the score that would have been obtained if all items were answered. This imputation method served to decrease the non-response rate, without altering the results for models fitted in this study. This method of imputation was applied to the delinquency variables, as well as to the school commitment and victimization variables.

Cross-sectional weights are applied in all analyses in this paper in order to account for unequal probabilities of sample selection. The complex sample design of the NLSCY necessitates the use of the bootstrap technique to estimate coefficients of variation, confidence intervals and to test for statistical significance of differences (Rao et al. 1992; Rust and Rao 1996).

Variables in the analysis

Property-related and violent delinquency
School commitment
Self-reported victimization
Socio-economic and demographic factors

Property-related and violent delinquency

In cycle 3, youth aged 12-15 were asked a series of questions about their involvement in violent and property-related delinquency. These concepts are analysed separately in this paper. The four-category response scales for each delinquency item ranged from never, to five times or more in the past year. Due to the relatively serious nature of the six violent and six non-violent or property-related delinquency items used in this analysis, each dependent variable is broken into two categories derived from the sum of the six questions, such that 0 = never committing a delinquent act and 1 = committing at least one of the delinquent acts, one or more times in the past year.

There are six items included in the property-related delinquency variable. A cross-sectional weighted estimate of about 24% of 12-15 year olds indicated that they had committed one or more property-related delinquency acts in the previous year.

The following items are included:

During the past 12 months, about how many times have you:

• Stolen something from a school or store?
• Broken into, or snuck into, a house or building with the idea of stealing something?
• Used or bought or tried to sell something you knew was stolen?
• Damaged or destroyed anything that didn’t belong to you (for example, damaged a bicycle, car, school furniture, broken windows or written graffiti)?
• Taken a car, motorbike or motorboat without permission?
• Set fire on purpose to a building, a car, or something else not belonging to you?

The six items comprising the violent delinquency variable include acts of physical violence as well as acts that are associated with potential physical violence (e.g., carrying weapons). About 20% (weighted) of 12-15 year olds indicated that they had committed one or more violent acts in the past year.

The following items are included:

During the past 12 months, about how many times have you:

• Fought with someone to the point where they needed care for their injuries (for example, because they were bleeding, or had broken bones)?
• Been in a fight where you hit someone with something other than your hands (for example, a stick, club, knife, or rock)?
• Carried a knife for the purpose of defending yourself or using in a fight?
• Carried a gun other than for hunting or target shooting?
• Carried any other weapon such as a stick or a club?
• Threatened someone in order to get his or her money or things?

School commitment

School commitment provides a measure of a youth’s orientation toward the school environment. The school commitment score is based on seven items describing attitudes such as the level of importance placed on doing well in school, making new friends at school, participating in school activities, showing up for class on time, learning new things, expressing one’s own opinion at school and participating in student council. Response categories ranged from 0 to 3, where 0 corresponds to ‘very important’ and 3 corresponds to ‘not important at all’. The score resulting from the combined items ranges from 0 (a high level of school commitment) to 21 (a low level of school commitment).

Self-reported victimization

The derived victimization variable used in this study includes both threats and actual physical harm, since previous research has indicated that the threat of physical violence can have serious consequences equal to actual physical harm (Selner-O'Hagan et al. 1998). Youth aged 12-15 in the NLSCY sample are asked four questions concerning their own victimization.

The following items are included:

During the past 12 months, how many times did someone:

• physically attack or assault you while at school or on a school bus?
• physically attack or assault you elsewhere including at home?
• threaten to hurt you without actually hurting you while at school or on a school bus?
• threaten to hurt you without actually hurting you elsewhere including at home?

Response categories ranged from 0 ‘never’ to 3 ‘five times or more’. The victimization score resulting from the combined items ranged from 0 (never victimized) to 12 (victimized 5 times or more for each of the four items). It should be noted that while the items included in the victimization variable do not specify sexual assault, youth may have included these incidents if they occurred within the context of a physical attack or assault. Therefore, the victimization measure used in this study can be viewed as a general measure of the youth’s perception of multiple forms of victimization.

Socio-economic and demographic factors

Models in this study control for the child’s sex, age in years, family structure, and level of family income adequacy. Although the link to family structure is by no means clear, some research indicates that the type of family that a child lives in affects his or her behavioural outcomes. For example, Lipman et al. (2002, 229) suggest that, on average, single parent families have greater levels of stress related to a variety of social and economic factors that may contribute to the development of problem behaviours in children.

Other studies have found that being in a stepparent family rather than residing with both biological parents is associated with an increased risk of juvenile delinquent behaviour, and that this is particularly the case for early onset delinquency initiated before the age of 15 (Coughlin and Vuchinich 1996). Finally, poverty during childhood has been linked to subsequent problem behaviour regardless of family structure (Sampson and Laub 1993).

Family structure is a three-category dummy-coded variable that contrasts families headed by two biological or two adoptive parents (reference category), with two other categories, single parent families and families in which a stepparent is present.

Income adequacy was derived from household income and household size . The three-category dummy-coded variable contrasts families with middle income (reference category) to families with lower levels of income adequacy and families at the upper-middle and highest income adequacy.