Multi-Level Effects on Behaviour Outcomes in Canadian Children

2.3 Results

This section addresses the research questions outlined earlier by incorporating alternative operationalizations of neighbourhood and family effects across models. The results are presented by the type of aggression assessed. The two and three level hierarchical linear models are first tested using the national sample. The enumeration area types are used as a form of neighbourhood measurement not yet incorporated into the study of aggression. These results are then are compared to analyses with the highly clustered set of families in ninety-six census tract units. Census tract level neighbourhood measures and PMK perceptions of the neighbourhood environment as deviated from the census tract means are incorporated in the highly clustered subsample analyses.

The bivariate correlations at the child level of analyses among the risk factors and indirect and physical aggression are presented in Appendix Tables B.1 and B.2. The correlations are generally in the hypothesized directions for the set of family and individual risk factors. For example, being female is positively associated with indirect aggression (r=0.07, p<0.001) but is negatively associated with physical aggression (r=-0.12, p<0.001). There is a small association among neighbourhood types and features with each form of aggression. Using the enumeration area types indices, the results indicate a consistent small to modest association between living in very high socioeconomic status environments compared to all other types of neighbourhoods and decreases in both forms of aggression (r=-0.05, p<0.001). However, for indirect aggression, Type 5 or lower middle class environments were positively associated with childhood aggression (r=0.04, p<0.001).

The PMK's level of perceived neighbourhood problems shows a stronger correlation with aggression than did the objective neighbourhood characteristics. An increased level of perceived problems was associated with higher levels of indirect aggression (r=0.09, p<0.001) and physical aggression (r=0.11, p<0.001). The parenting variables and exposure to violence in the home were generally the most strongly associated with childhood aggression; however, the results are suggestive of a potential contribution of neighbourhood influences to models including more established risk factors.

2.3.1 Three-Level Models: Structural and Mediational National Models of Children Nested in Families in Census Tracts

The two level hierarchical linear models with children nested in families were first assessed (results not shown). These analyses are similar to the two level models by Tremblay et al. (1996) but include additional structural risk factors, in particular the neighbourhood variables. The three level models were then assessed. The model building procedure in the first set of tables with the national sample adds variables in steps one through six listed in the left hand column of the tables with a deviance statistic reported in the far right column per stage. The columns "objective" and "subjective" refer to the final models following distinctions made in the work of Upchurch et al. (1999), where the subjective model adds the PMK's perceptions of the neighbourhood to the model. The fit statistics are presented separately in accompanying tables. The set of models with the ninety-six census tracts was built from an unconditional model, through the addition of controls, with the neighbourhood effects added last.

The results of three level random intercept ANOVA models partition the variance in indirect and physical aggression into three components as shown in Table 2.1. Fifty percent of the variability in indirect aggression is between-individuals. However, 47% of the variability in indirect aggression is between-families. Finally, there is a statistically significant variance component indicating that three percent of the variance in indirect aggression is between-census tracts. As this intraclass correlation is significant at p=0.048 with a large sample size, these analyses are considered exploratory and require further replication. However, the patterns indicate a component of the variability in indirect aggression is due to between-neighbourhood variance.

Table 2.1
Model Fit and Variance Decomposition
Random Intercepts ANOVA Models	Three-Level Model	Deviance	Parameters	Variance Levels
				One	Two	Three
Indirect Aggression (4-11)	10,971 Children in 7452 Families/ in 2294 Census Tracts	39,586.95	4	1.31	1.23 p=0.000	0.08 p=0.048
Physical Aggression (4-11)	11,216 Children/ in 7585 Families/ in 2315 Census Tracts	43,062.50	4	1.71	1.38 p=0.000	0.17 p=0.000
Physical Aggression (2-11)	14,287 Children/ in 9117 Families/ in 2493 Census Tracts	42,996.92	4	0.77	0.57 p=0.000	0.08 p=0.000
Note: These sample sizes for the national analyses fall within the range of 4,500 to 17,500 children recommended by Statistics Canada to yield "marginal" estimates for analyses with 4-11 year olds (HRDC/STC, 1997: 168).

The results for physical aggression of 4 to 11 year olds indicate that 52% of the variability in physical aggression is between-individuals, and 42% is between-families (p=0.000). Five percent of the variability in physical aggression is due to between-neighbourhood differences (p=0.000). Similarly, approximately six percent of the variance in 2-11 year old physical aggression is explained by between-tract variability (p=0.000). Forty percent of the variability in 2-11 year old physical aggression is due to family level differences (p=0.000), and 54% is between-individual variance. The between-neighbourhood intraclass correlations are within the range of those found for other behavioural outcomes of children and youth, where neighbourhoods account for up to 10% of the variance (see Leventhal & Brooks-Gunn, 2000).

Covariates are introduced into subsequent models presented across the columns in the tables to examine whether the neighbourhood effects, if any, are mediated by home environment factors (see Ross, 2000). The results are presented in Tables 2.2 and 2.3. Tables 2.4 and 2.5 present the model variance components and the deviance statistics for Tables 2.2 and 2.3, respectively.

The results of the "objective" model with the enumeration area types variables of neighbourhood effects are presented in the first column of Table 2.2. These results show a trend for Type Three environments consisting of a higher percentage of single parents and immigrant population, along with low income levels and low adult employment levels, to be associated with elevated levels of indirect aggression (b=0.27, p<0.01), net of children's age, gender, and the gender by age interaction effect. The direction of this effect is consistent with the bivariate correlations between Type Three neighbourhoods and indirect aggression indicated in Appendix Tables B.1 and B.2 While the multivariate hypothesis test between the deviance for the models in columns one and two indicates a trend toward statistical significance in the improvement in model fit (for a change of 20.8 and 7 df), this result is viewed with caution as the full set of controls are not yet included in the model. The effects of the objective environment becomes non-significant when the subjective neighbourhood environment is included in Column Two of Table 2.2 (b=0.11, p<0.001). This addition to the model provides a highly significant drop in the model deviance (82.57, p<0.001 for 1 df).

Table 2.2
Three Level Multilevel Models for Indirect Aggression of 4-11 Year Old Children Nested in Families
(N=10,971 Children in 7,452 Families in 2,294 Census Tracts)
	Objective b	Subjective b	Mediating b	Final b
Random Intercept Model
Intercept	1.06^***	1.07^***	1.02^***	1.06^***
Individual Level Control Variables
Female^a	0.23^***	0.23^***	0.32^***	0.31^***
Child's Age	0.08^***	0.08^***	0.07^***	0.08^***
Individual Level Interaction Effect
Female^* Child's Age	0.03^***	0.03^***	0.03^***	0.03^***
Enumeration Area Types (Census)^b:
Low Adult Employment, Low Income,
Low Education	-0.02	-0.04	0.01	-0.03
Low Youth Employment, Low Education	0.06	0.04	0.10	0.07
High % Single Parents and Immigrants,
Low Income, Low Employment	0.27^**	0.12	0.11	-0.00
High % Single Parents and Immigrants	0.11	0.03	0.02	-0.08
Lower Middle Class	0.10	0.09	0.10	0.08
Upper Middle Class	-0.03	-0.02	0.01	0.03
Very High socioeconomic Status	-0.21	-0.19	-0.08	-0.01
PMK Perceived Problems in the Neighbourhood		0.11^***	0.08^***	0.06^***
Family Mediating Processes
Hostile/ Punitive Parenting			0.09^***	0.08^***
Positive Interaction			-0.02^**	-0.02^*
Consistent Parenting			-0.03^***	-0.02^***
Exposure to Violence in Home			0.24^***	0.20^***
Other Family Level Controls
Household socioeconomic Status				-0.04
Recent Immigrant (last 4 years)^c				0.18
Immigrant (last 5-9 years)				0.21
Blended family--Two Parents^d				0.26^**
Single Parent Family				0.20^**
Year Lived at Current Address				-0.00
Home Ownership				-0.09
Residential Crowding				-0.15^**
Number of Siblings				0.02
Biological Mother-Years of Age at First Birth				-0.01^*
PMK Level of Depression				0.02^***
=0.10 (two-tailed tests of statistical significance), ^=p<0.05, ^=p<0.01, ^**=p<0.001, a=Males, b=Type six or middle class neighbourhoods, c =Non-immigrants or immigration to Canada over 10 years ago, d=Two biological parents family structure.

Table 2.3
Three Level Multilevel Models for Physical Aggression of 2-11 Year Old Children Nested in Families
(N=14,287 Children in 9,117 Families in 2,493 Census Tracts)(ML-F)
	Objective b	Subjective b	Mediating b	Final b
Random Intercept Model
Intercept	1.50^***	1.51^***	1.39^***	1.18^***
Individual Level Control Variables
Female^a	-0.29^***	-0.29^***	-0.19^***	-0.20^***
Child's Age	-0.05^***	-0.05^***	-0.04^***	-0.03^***
Enumeration Area Types (Census)^b:
Low Adult Employment, Low Income,
Low Education	-0.05	-0.07	-0.03	-0.02
Low Youth Employment, Low Education	-0.16^**	-0.18^**	-0.13^*	-0.12^*
High % Single Parents and Immigrants,
Low Income, Low Employment	0.05	-0.07	-0.07	-0.10
High % Single Parents and Immigrants	0.08	0.01	0.02	0.01
Lower Middle Class	-0.00	-0.01	-0.01	-0.01
Upper Middle Class	-0.06	-0.05	-0.02	-0.00
Very High socioeconomic Status	-0.22^**	-0.21^**	-0.13	-0.09
PMK Perceived Problems in the Neighbourhood		0.09^***	0.06^***	0.05^***
Family Mediating Processes
Hostile/ Punitive Parenting			0.09^***	0.09^***
Positive Interaction			0.01^**	0.01
Consistent Parenting			-0.01^***	-0.01^***
Exposure to Violence in Home			0.22^***	0.17^***
Other Family Level Controls
Household socioeconomic Status				-0.01
Recent Immigrant (last 4 years)^c				-0.13
Immigrant (last 5-9 years)				-0.12
Blended family--Two Parents^d				0.07
Single Parent Family				0.17^*
Year Lived at Current Address				-0.01^*
Home Ownership				-0.05
Residential Crowding				-0.07^*
Number of Siblings				0.12^***
Biological Mother-Years of Age at First Birth				-0.01^***
PMK Level of Depression				0.01^***
=0.10 (two-tailed tests of statistical significance), ^=p<0.05, ^=p<0.01, ^**=p<0.001, a=males, b=Type Six or Middle Class neighbourhoods, c =Non-immigrants or immigration to Canada over 10 years ago, d=Two biological parents family structure.

Table 2.4
Fit Statistics and Variance Components for Three Level Multilevel Models for Indirect Aggression of 4-11 Year Old Children Nested in Families
(N=10,971 Children in 7,452 Families in 2,294 Census Tracts)
	Deviance	Parameters	Variance Levels
			One	Two	Three
Random Intercept Model Intercept	39,586.95	4	1.31	1.23 p=0.000	0.08 p=0.048
Individual Level Control Variables	39,317.57	6	1.27	1.21 p=0.000	0.08 p=0.045
Individual Level Interaction Effect	39,310.55	7	1.27	1.21 p=0.000	0.08 p=0.049
Enumeration Area Types (Census)	39,289.75	14	1.27	1.21 p=0.000	0.08 p=0.074
PMK Perceived Problems in the Neighbourhood	39,207.18	15	1.27	1.20 p=0.000	0.07 p=0.172
Family Mediating Processes	38,260.76	19	1.18	1.07 p=0.000	0.05 p=0.261
Other Family Level Controls	38,142.72	30	1.18	1.03 p=0.000	0.04 p=0.483

Table 2.5
Fit Statistics and Variance Components for Three Level Multilevel Models for Physical Aggression of 2-11 Year Old Children Nested in Families
(N=14,287 Children in 9,117 Families in 2,493 Census Tracts)(ML-F)
	Deviance	Parameters	Variance Levels
			One	Two	Three
Random Intercept Model Intercept	42,996.92	4	0.77	0.57 p=0.000	0.08 p=0.000
Individual Level Control Variables	42,566.93	6	0.75	0.56 p=0.000	0.08 p=0.045
Enumeration Area Types (Census)	42,542.33	13	0.74	0.56 p=0.000	0.07 p=0.000
PMK Perceived Problems in the Neighbourhood	42,413.49	14	0.74	0.55 p=0.000	0.07 p=0.000
Family Mediating Processes	40,159.82	18	0.64	0.46 p=0.000	0.05 p=0.000
Other Family Level Controls	40,000.81	29	0.64	0.44 p=0.000	0.05 p=0.000

The results of a mediational model testing whether the subjective neighbourhood effect on indirect aggression would hold net of parenting variables in the three level model are presented in Column Three of Table 2.2. The results indicate that the PMK perceptions of neighbourhood problems coefficient is reduced by 27%, but it remains statistically significant (b=0.08, p<0.001). In comparing Columns Two and Three, there is a notable suppression effect for gender. In controlling these processual home environmental variables, the effect of being female is increased (0.32, p<0.001). Being female is negatively associated with hostile and punitive parenting, and is less associated with the other aspects of the home environment, including exposure to violence. This pattern suggests that because male and female children are differentially exposed to hostile and punitive parenting, the gender gap in indirect aggression is muted. Taking levels of home environment factors into account, the gender gap in aggression actually increases. Furthermore, as the effect of being female depends upon age, it should be interpreted along with the interaction effect component, indicating that the effect of being female, particularly at older ages, is in part suppressed by exposure to hostile and punitive parenting.

The final model presented in the fourth column of Table 2.2 indicates that the perceptual effect of neighbourhoods on indirect aggression holds net of family controls (b=0.06, p<0.001). Furthermore, the family structure dummy variables indicate a salient role of family structure in addition to family process on children's indirect aggression. A further model investigated whether the effect of single parent households and blended family households on indirect aggression was equal. The hypothesis of equality of these effects was rejected (c² change of 17.11 for 2 df). As these results are in the same metric, it may be useful for future research to consider how family structure impacts indirect aggression, indicating a slightly greater impact of blended family status than single parent family status compared to two biological parent families. These effects require further investigation, but may be signalling that something about the family environment in blended families is putting children more at risk for indirect aggression.

The variance components for the models presented in Table 2.2 are detailed in Table 2.4. These results generally show a decrease in the model variance components as the successive models were fit. Although the deviance test indicates a contribution of the subjective environment to the explanation of children's levels of indirect aggression, the subjective environment and the objective environment together explain only approximately 13% of the available between-neighbourhood variability in aggression. However, these results may best be interpreted with the overall results from the final model that indicate an additive accumulation of risk factors including the individual, family, and additionally the neighbourhood levels of analysis combine to increased levels of indirect aggression.

Table 2.3 presents the three level model results for physical aggression among two to eleven year olds. The results of the "objective" model are presented in Column One. The enumeration area types are used to represent the neighbourhood level of analysis, which is assumed equivalent to the family level of analysis in these models. The results in Table 2.3 indicate a protective effect of Type Two environments (those with low youth employment and low education) compared to Type Six or middle class environments (b=-0.16, p<0.01). Furthermore, the effect of very high socioeconomic environments is protective against physical aggression (b=-0.22, p<0.01). While these levels of significance should be viewed with caution in a sample as large as the NLSCY, some corroborating evidence was found in examining the same model among 4-11 year old children with the broader measure of physical aggression (results not shown). The results with 4-11 year olds indicated that very high socioeconomic environments exerted a protective effect (b=-0.40, p<0.001), and that Type Two environments were also protective (b=-0.25, p<0.05). In those results, there was additionally a protective effect of upper middle class environments compared to middle class environments (-0.13, p<0.05). As indicated in Table 2.5, the objective characteristics decrease the deviance of the model with only children's gender and age added (by a change of 24.67 and 7 df), which is at the critical value on the chi-square distribution for a statistically significant reduction at p<0.001. Again, these results should be viewed with caution as not all of the theoretical control variables are yet in the model.

These results may be suggesting that the opportunity structure in middle class environments free children from some of the more protective influences of certain combinations of neighbourhood factors. Willms (1999) draws on work by Raffe and Willms (1989) on Scottish pupils' academic attainment to link the "local opportunity structure" in communities and educational attainment (p. 88). Their findings indicated that "in communities where there were few opportunities for employment, pupils achieved better grades on national examinations and were more likely to stay in school beyond the compulsory period" (Willms, 1999, p. 88).

Other models examined in this research use the "neighbours score" and "neighbourhood safety score" constructed by Statistics Canada. Those variables were found to have no discernible impact when added at the family level of analyses net of the other variables in the equation. Finally, given the theoretical rationale behind the categories of "social" and "physical" disorder in the neighbourhood effects research, an alternative set of models with a latent class version of the perceived problems subjective environment was used. These results were generally consistent with the continuous version of this scale.

2.3.2 Three-Level Structural and Mediational Models with Census Tract Level Characteristics at the Census Tract Geographical Level of Analysis (96 Units)

A subsample of ninety-six census tract units from the NLSCY are used in this section to operationalize the neighbourhood context. These units generally contain between ten and twenty families, when missing data on the full set of variables used in these analyses is taken into account. The more highly clustered sample facilitates the aggregation of the PMK subjective neighbourhood assessments across families to the tract level. Further, the more highly clustered sample permits better reliability of the tract mean level of aggression.

As the coding is central to the interpretation of the results that follow, the scales used in these analyses are briefly reviewed. These scales were constructed specifically for these ninety-six units and are slightly different than the Statistics Canada scales used in the national analyses. Three items were combined together as the new scale of perceived tract problems, where a score of zero indicates high problems, and a score of six indicates low problems. The collective efficacy score combines five items with a score of zero indicating high collective efficacy, and a score of 15 indicating low collective efficacy.

The zero order correlations from the ninety six census tract units using the tract level file from the 2-11 physical aggression analyses are presented in Appendix Tables B.1 and B.2. As the percentage of low income families in the neighbourhood increases, neighbourhood problems increase, as indicated (given the coding) by the negative relationship (-0.24, p<0.05). Lower levels of collective efficacy are indicated by higher levels on that scale, therefore, a positive correlation is observed between higher scores and a higher percentage of low income families in the neighbourhood (0.34, p<0.001). This indicates that children living in disadvantaged families face higher levels of problems in their neighbourhoods and have fewer resources available in those environments to offset the additional stressors. Finally, higher levels of neighbourhood problems at the tract level, as indicated by lower scores on this scale, are associated with higher scores on the collective efficacy score, which is indicative of lower levels of collective efficacy (-0.27, p<0.01). Therefore, tract level collective efficacy and neighbourhood problems are inversely related.

Given the salience of the subjective environment in the national analyses for both types of aggression, it was considered central to these analyses to include the person most knowledgeable (PMK) about the child's scores in the models in addition to the tract level scores. However, in order to derive unique measures, it was necessary to deviate the PMK measure from the tract level variable. A PMK score was calculated at the family level in parallel to the tract level score (a=0.70), as the three item scale was constructed for these tract level units. The PMK score was then deviated from the tract level score. This deviation in neighbourhood problems was used as an indicator of the differences between the tract level problems and the PMK's perception of problems in the neighbourhood. Similarly, a deviation score from the tract level mean on collective efficacy was calculated by subtracting the parallel measure at the PMK level.

Given the complex coding, examples of the deviation score are considered to guide interpretation. A census tract level of high problems with a score of 0 minus a PMK score of 2 would equal -2. A census tract level measure of low problems, indicated by a score of six, subtracting a PMK score of 2 (indicating higher perceived problems by the PMK) would yield a score of 4. Therefore as deviations increase there is a larger gap between the tract score and the PMK perception. Positive deviations indicate the PMK perceived problems are greater than the CT means, while negative deviations indicate a lower level of PMK perceived problems than the CT means. As this scale uses fewer items than the original PMK perceived problems score, it allows slightly more cases into the analyses with listwise deletion across all variables. The sample sizes are reported per table.

Table 2.6 was reported from a set of analyses for indirect aggression that showed a distribution as follows: 11.5% of families had less than ten families per tract but greater than three families, 6.2% had more than twenty families per tract, and 82% had 10-20 families per tract. In turn, 59% of those families had one child per home, 33% had two children per home, 7% had three children per home, and less than 1% had four children per home. These analyses show a small to modest intraclass correlation (ICC=2.7%), indicating statistically significant variability in indirect aggression across census tracts. These results are similar to those found nationally with the census tract units. The controls were added to the model in stages from the child level controls to the neighbourhood effects. The perceptual environment deviation scores were added last to the model to assess their contribution net of all other factors on childhood aggression. While the tract level objective census tract variables did not provide a significant reduction in the deviance between models, the addition of the perceptual deviations scores decreased the deviance by 16.98, (p<0.001). These results indicate a modest (2%) additional decrement to the level of between-family variability explained, however, the results show that the deviation in the PMK's score has an influence on children's indirect aggression. This effect was estimated at b=0.19 (p<0.001). As the positive deviation increases (that is, the PMK score is lower than the CT score, indicating greater levels of PMK perceived problems than CT problems), children's indirect aggression increases. In an additional set of analyses, the PMK's perceptions of social support were included as a potential source of spuriousness for the deviations in perceptions of neighbourhood problems results. Social support did not have a statistically significant effect on children's indirect aggression with all the other variables in the model, however, the deviations in perceived problems variable remained significant.

The results presented in Table 2.7 indicate that for 2-11 year old physical aggression, there is statistically significant between-neighbourhood variability (ICC=4%, p<0.001). There are significant declines in the model deviance at steps two and three through the addition of child and family level controls. The consistent single parent family structure effect found nationally on physical aggression is no longer significant in this smaller group. However, an increased number of siblings remains a risk factor for physical aggression (b=0.15, p<0.001). Similar patterns for the other fixed effects in the model are observed in these results compared to the national analyses. When the census tract level measures are added to the model, they do not provide an improvement in the model deviance. However, a fixed effect for the subjective tract level environment was found to be statistically significant (at b=-0.23, p<0.05). This result indicates that as neighbourhood problems increase (indicated by a lower score), childhood physical aggression increases. Finally, adding the PMK perceptual deviations provides a statistically significant drop in the deviance (14.57 for 3 df). As found for indirect aggression, the deviation of the PMK's score from the CT score is positive at (0.11, p<0.01). Again, given the coding, this result indicates that as the deviation positively increases (that is the PMK's score is lower than the CT score, indicating greater levels of PMK perceived problems than CT problems), children's indirect aggression increases. The addition of the PMK deviation scores contributes about two percent to the variance explained in physical aggression, controlling for all other factors.

However, the consistency of the subjective PMK neighbourhood results is suggestive of robust effects on aggression.

Table 2.6
Three Level Structural Models with Tract Level Effects and PMK
Deviations Effects on 4-11 Year Old Indirect Aggression
(N=2,011 Children in 1,350 Families in 96 Census Tract Units)
	b	Deviance (ML-F)	Estimated Parameters	Variance Levels
				One	Two	Three
Random Intercept Model		7,299.41	4	1.44	1.04	0.07
					p=0.000	p=0.027
Intercept	1.14^***
Child Level Control Variables		7,039.40	11	1.26	0.94	0.05
Female^a	0.21^**
Child's Age	0.13^***
Hostile/ Punitive Parenting	0.09^***
Positive Interaction	- 0.02
Consistent Parenting	-0.00
Exposure to Violence in Home	0.26^***
Female^* Child's Age	-0.03
Family Level Control Variables		7,013.50	22	1.25	0.90	0.05
Household SES	-0.07
Recent Immigrant^b (last four years)	0.63
Immigrant (last 5-9 years)	0.18
	0.50^*
Two Parents-Blended Family^c
Single Parent Family	-0.14
Years Lived at Current Address	-0.00
Home Ownership	-0.12
Residential Crowding	-0.34^*
Number of Siblings	-0.02
Biological Mother's Years of Age at First Birth	-0.01
PMK Level of Depression	0.00
Tract Level Variables		7,007.77	26	1.25	0.90	0.04
						p=0.086
Tract Incidence Low Income Families	1.39
Census Tract Population	0.00
Tract Collective Efficacy	-0.06
Tract Neighbourhood Problems	0.03
Deviations of PMK		6,990.79	28	1.25	0.88	0.04
						p=0.080
Neighbourhood Variables from the CT Means:
Perceptions of Problems in the Neighbourhood	0.19^***
Collective Efficacy	0.03
=0.10 (two-tailed tests of statistical significance),^=p<0.05, =0.10 (two-tailed tests of statistical significance), ^=p<0.01, ^**=p<0.001, a=males; b=Type Six or Middle Class neighbourhoods; c =non-immigrants or immigration to Canada over 10 years ago; d=two biological parents family structure.

Table 2.7
Three Level Structural Models with Tract Level Effects and PMK
Deviations Effects on 2-11 Year Old Physical Aggression
(N=2,579 Children in 1,625 Families in 96 CT Units)
	b	Deviance (ML-F)	Estimated Parameters	Variance Levels
				One	Two	Three
Random Intercept Model		7,912.21	4	0.82	0.61	0.05 p=0.000
Intercept	1.21^***
Child Level Control Variables		7,452.31	10	0.71	0.47	0.03
Female^a	-0.11^**
Child's Age	-0.03^**
Hostile/ Punitive Parenting	0.09^***
Positive Interaction	0.01
Consistent Parenting	-0.01
Exposure to Violence in Home	0.27^***
Family Level Control Variables		7,401.14	21	0.71	0.44	0.03 p=0.008
Household socioeconomic Status	-0.01
Recent Immigrant (last four years)^b	-0.18
Immigrant (last 5-9 years)	0.07
Two Parents-Blended Family^c	-0.09
Single Parent Family	-0.08
Years Lived at Current Address	-0.01
Home Ownership	-0.17^*
Residential Crowding	-0.22^*
Number of Siblings	0.15^***
Biological Mother's Years of Age at First Birth	-0.02^***
PMK Level of Depression	0.01^*
Tract Level Variables		7,396.10	25	0.70	0.44	0.02
Tract Incidence Low Income Families	-0.45
Census Tract Population	0.00
Tract Collective Efficacy	0.01
Tract Neighbourhood Problems	-0.23^*
Deviations of PMK		7,381.53	27	0.70	0.43	0.02 p=0.010
Neighbourhood Variables from the CT Means:
Perceptions of Problems in the Neighbourhood	0.11^***
Collective Efficacy	0.02
=0.10 (two-tailed tests of statistical significance), ^=p<0.05, ^=0.10 =p<0.01, ^**=p<0.001, a=males; b=Type Six or Middle Class neighbourhoods; c =non-immigrants or immigration to Canada over 10 years ago; d=two biological parents family structure.

Contents

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

2.3 Results

2.3.1 Three-Level Models: Structural and Mediational National Models of Children Nested in Families in Census Tracts

2.3.2 Three-Level Structural and Mediational Models with Census Tract Level Characteristics at the Census Tract Geographical Level of Analysis (96 Units)