Multi-Level Effects on Behaviour Outcomes in Canadian Children

1.2 Method

1.2.1 Sample and Procedure

Between 1994-1995, the NLSCY conducted its first data collection cycle on Canadian households with children aged 0 to 11 years, using a stratified, multistage probability sample design based on information collected by Statistics Canada's Labour Force Survey. One child from each household was randomly selected for the study. If there were other children in the family who fell within the specified age range, they were also randomly selected, with a maximum of four children per household. In each household, the person most knowledgeable about the child (the child's mother in 89.4% of cases) participated in a face-to-face interview during which she provided basic demographic and socioeconomic data about the family as well as extensive information (e.g., physical health, education, behaviour, social development, parenting practices) about each selected child (HRDC/STC, 1995).

Complete interview data were obtained from 13,439 households across Canada, representing an overall response rate of 86.3% and resulting in a sample of 22,831 newborn to 11-year-old children. The sample did not include children living in the two Canadian territories, in institutional facilities, and on Aboriginal reserves. We chose Statistics Canada 1996 census tracts as the spatial definition of neighbourhood. In order to maintain adequate reliability for the neighbourhood problems and collective efficacy measures (described in detail below), it was necessary to eliminate census tracts with less than 15 households (i.e., 2,671 census tracts). Our sample consisted of 96 census tracts including 1,982 families and 2,745 children aged 2 to 11 years. The sample was weighted according to NLSCY procedures.

There was an approximately equal number of girls (49.5%) and boys (50.5%) in the present study, and the average age of children was 6.53 years (SD = 2.87). Mother's average age was 33.7 years (SD = 5.7), and average level of education was 12.07 years (SD = 2.23). The majority of children (81.3%) lived with both biological parents while 17.3% lived with only one biological parent. The remaining 1.4% were in living arrangements which included neither biological parent. Approximately 12% of families had incomes below the poverty level. The average number of families in each of the 96 census tracts was 19.67 (SD = 5.95).

1.2.2 Measures

Neighbourhood (Census Tract) Predictor Variables

We used a combination of techniques to gather neighbourhood information. Statistics Canada provided the objective census tract information on the variables of area size and poverty. Mother reports related to perceptions of their neighbourhood (e.g., problems, collective efficacy) were obtained through personal interviews.

Based on Statistics Canada data, neighbourhoods were classified into one of the six following area size categories: urban with a population of 500,000 or over; urban with a population between 100,000-499,999; urban with a population between 30,000-99,999; urban with a population between 15,000-29,999; urban with a population less than 15,000; and rural area.

A neighbourhood poverty score was obtained from Statistics Canada census survey information on low-income cut-offs, a widely used though unofficial poverty measure. Scores ranged from 0 to 1, with higher scores indicating a greater proportion of families in the neighbourhood living below the poverty level.

Four items selected from the revised Simcha-Fagan Neighbourhood Questionnaire (Barnes McGuire, 1997) were used in the interviews to measure the extent of neighbourhood problems. Mothers rated each item along a 3-point scale from 0 (a big problem) to 2 (no problem). Negatively oriented items were reverse scored so that higher scores represented more neighbourhood problems. We evaluated the reliability of the neighbourhood problems scale through a hierarchical statistical model that assessed within- and between-neighbourhood variation. Given the categorical nature of item responses, we formulated the model in terms of the logit of the cumulative response probability, shown as

where represented the cumulative response probability for the item of the individual in the neighbourhood. The coefficient was the loading of the neighbourhood score and the coefficient was the loading of the specific individual score on the item. We incorporated different difficulty thresholds into the model by way of the coefficient . We then extended the model by including respondent characteristics as covariates, allowing for the control of possible bias. The expanded model is

The analysis revealed that of the five original neighbourhood problem items (i.e., "Litter, broken glass, or garbage in the street or road, on the sidewalk, or in yards," "Selling or using drugs," "Alcoholics and excessive drinking in public," "Groups of young people who cause trouble," and "Burglary of homes and apartments"), the burglary item did not lie on the same dimension as the other items. Frequency of burglaries showed a positive association with the level of neighbourhood income. Thus, we excluded the burglary item from the neighbourhood problems scale. Inter-subject agreement on the four remaining neighbourhood items was 0.13, yielding a reliability coefficient that ranged from 0.69 to 0.79 for neighbourhoods with 15 to 25 survey respondents, respectively (as measured by the formula . Therefore, in order to maintain an acceptable reliability level for the neighbourhood problems scale, only neighbourhoods with a minimum of 15 households were retained for analysis, resulting in 96 census tracts comprising 1,982 families.

Five items from Barnes McGuire's (1997) revised Simcha-Fagan Neighbourhood Questionnaire were used in the interviews to measure neighbourhood collective efficacy, which combines social cohesion (i.e., mutual trust among neighbours, sharing common values) and informal social control (i.e., neighbours can rely on one another to monitor and supervise youth). Mothers rated each item along a 3-point scale from 0 (strongly agree) to 2 (strongly disagree). Negatively oriented items were reverse scored so that higher scores represented greater collective efficacy. We conducted the same hierarchical statistical analyses for the collective efficacy scale as for the measure of neighbourhood problems. Results indicated that all five original items fell along the same dimension (i.e., "If there is a problem around here, the neighbours get together to deal with it," "There are adults in the neighbourhood that children can look up to," "People around here are willing to help their neighbours," "You can count on adults in this neighbourhood to watch out that children are safe and don't get in trouble," and "When I'm away from home, I know that my neighbours will keep their eyes open for possible trouble"). Inter-subject agreement was 0.15, yielding a reliability coefficient that ranged from 0.72 to 0.81 for neighbourhoods consisting of 15 to 25 respondents, respectively. Again, only census tracts with a minimum of 15 households were retained for subsequent analyses.

Family Predictor Variables

Socio-demographic data were gathered on mother's age and educational level as well as on household income and family status (two-parent or one-parent led).

socioeconomic status (SES) was calculated using the method proposed by Willms and Shields (1996), which takes into account the five following variables: mother's level of education; level of education of mother's spouse; mother's occupational prestige; occupational prestige of mother's spouse; and household income. Higher scores reflected higher SES levels.

Twelve items derived from the shortened version of the Center for Epidemiological Studies Depression scale (Radloff, 1977) were used in the interviews to assess feelings of depressed mood over the past week, and the reliability of the measure was reported to be 0.82. Mothers rated each item (e.g., "I felt that I could not shake off the blues even with help from my family and friends") along a 4-point scale from 0 (rarely or none of the time; less than one day) to 3 (most or all of the time; 5-7 days). Higher scores indicated greater levels of depressed mood.

Twelve items based on a subscale of the McMaster Family Assessment Device (Byles, Byrne, Boyle, & Offord, 1988) were used in the interviews to gather information on various aspects of family functioning, namely problem solving, communication, roles, affective responsiveness, affective involvement, and behaviour control. Mothers rated each item (e.g., "We avoid discussing our fears or concerns") along a 4-point scale from 0 (strongly agree) to 3 (strongly disagree). Negatively oriented items were reverse scored so that higher scores represented greater family dysfunction. The measure had a reliability rate of 0.88.

Six items representing a shortened version of the Social Provisions Scale (Cutrona & Russell, 1987) were used in the interviews to measure various social support characteristics, namely guidance, reliable alliance (i.e., feeling assured that others would be available to offer practical help), and attachment. Mothers rated each item (e.g., "I have family and friends who help me feel safe, secure, and happy") along a 4-point scale from 0 (strongly disagree) to 3 (strongly agree). Higher scores represented greater social support, and the reliability of the measure was reported to be 0.82.

Twenty-one questions adapted from the Parent Practices Scale (Strayhorn & Weidman, 1988) were used in the interviews to measure the four following parenting behaviours: positive interaction (n = 5) had a reliability of 0.81; hostility (n = 7) had a reliability of 0.71; consistency (n = 5) had a reliability of 0.66; and punitive parent management techniques (n = 4) had a reliability of 0.57. Mothers rated each item (e.g., "Do something special with your child that he/she enjoys") in terms of frequency from 0 (never) to 4 (many times each day). Higher scores indicated greater frequencies for each type of parenting behaviour. We calculated average scores across all children for whom mothers responded to interview questions. For example, if there were three children in the family whose ages ranged from 0-11 years, the mother answered parenting questions for each child, and the scores (coded as mother positive interaction, mother hostility, mother consistency, mother punitive parenting) represent the average level of these parenting behaviours across the three children. This measure allowed us to examine parenting behaviours primarily as a function of the mother and her interactions with all the children in the family.

Child Predictor Variables

Information was gathered on each child's sex and age in years.

Using the information provided by mothers on the four types of parenting behaviour, we calculated another set of scores to examine exposure to parenting behaviours (coded as positive interaction, hostility, consistency, punitive parenting). These scores represented the deviation in individual scores from the average across all children for whom mothers responded to interview questions. Thus, a higher score on hostile parenting, for example, indicates that a child's exposure to this type of parenting behaviour is greater than the average for all children in the family. This measure allowed us to examine parenting behaviours primarily as a function of the child and his or her particular interactions with the mother.

Child Behaviour Outcome Variables

There were slight differences in the child behaviour measures for children aged 2-3 years and those aged 4 years and older. The hierarchical linear models were adjusted for the differences in the number of items between the 2-3 and 4-11 year old children. The four behavioural measures were developed for the NLSCY, based on measures from the Montreal Longitudinal and Experimental Study (Tremblay et al., 1991; Tremblay, Vitaro, Gagnon, Royer, & Piché, 1992) and the Ontario Child Health Study (Boyle et al., 1987). All the child behaviour outcome variables were standardized.

Six items were used in the interviews to derive physical aggression scores for children. Mothers rated each item (e.g., "Gets into many fights") along a 3-point scale from 0 (never or not true) to 2 (often or very true), with higher scores reflecting more physically aggressive behaviour. The measure had a reported reliability of 0.78.

Eight items were used in the interviews to measure hyperactive-inattentive behaviours among children. Mothers rated each item (e.g., "Cannot settle to anything for more than a few moments") along a 3-point scale from 0 (never or not true) to 2 (often or very true). Higher scores indicated higher levels of hyperactivity-inattention, and the measure had a reliability rate of 0.84.

Eight items were used in the interviews to assess children's anxiety-emotional problems, and the reliability of the measure was reported to be 0.79. Each item (e.g., "Seems to be unhappy, sad, or depressed") was rated along a 3-point scale from 0 (never or not true) to 2 (often or very true), with higher scores reflecting greater anxiety-emotional problems.

Ten items were used in the interviews to derive prosocial behaviour scores. Mothers rated each item (e.g., "Will try to help someone who has been hurt") along a 3-point scale from 0 (never or not true) to 2 (often or very true), with higher scores indicating greater prosocial behaviours. The measure had a reliability rate of 0.82.

1.2.3 Data Analyses

We used hierarchical linear modelling to statistically account for the clustering of the sampled children within families and neighbourhoods and to explicitly model the relationship between individual, family, and neighbourhood characteristics and the dependent variables (Bryk & Raudenbush, 1992; Goldstein, 1995). As such, we conducted three-level hierarchical (individual children nested within families which are nested within neighbourhoods) linear models for each of the four standardized dependent variables, namely physical aggression, hyperactivity-inattention, anxiety-emotional problems, and prosocial behaviour. There were 6 child-level predictor variables (sex, age, positive interaction, hostility, consistency, punitive parenting), 11 family-level predictor variables (age, educational level, mother depressed mood, family status, SES, family dysfunction, social support, mother positive interaction, mother hostility, mother consistency, mother punitive parenting), and 4 neighbourhood-level predictor variables (area size, poverty, problems, collective efficacy). We chose variables that seemed important based on past research findings and that appeared to fit well together at a theoretical level. As such, the variables were not completely independent of one another. We did, however, test for multicollinearity and found no problems. All statistical analyses were conducted using MLnWin (Rasbash & Woodhouse, 1996).

Contents

Multi-Level Effects on Behaviour Outcomes in Canadian Children - May 2001