Keywords

Introduction

Adolescent substance abuse is a growing global issue. Teenagers are not only taking conventional drugs such as tobacco and alcohol but also abusing other psychotropic substances (e.g., ketamine, cannabis, and ecstasy). Worse still, young people have a common myth that these substances are nonaddictive, harmless, and trendy (Shek et al. 2011). In Hong Kong, drinking and smoking were reported as the most frequent substance abuse behaviors among school adolescents (Shek 2007; Shek and Ma 2011a). Besides, adolescent substance abusers were getting younger (e.g., with the youngest at the age of 12); they tended to have longer substance abuse history, and substance abuse was hidden in nature (such as taking drugs at home). Psychotropic substances also become more popular (Narcotics Division 2009, 2014). These changing trends bring a warning call for parents, school educators, social workers, and policy makers who are concerned about the healthy development of adolescents. Hence, it becomes necessary to develop a more accurate understanding of adolescent substance abuse in the individual, familial, community, and socioeconomic contexts. Against this background, this longitudinal study focused on personal and familial predictors of substance addiction among adolescents in the schools in Hong Kong.

Shek (2007) proposed to understand adolescent substance abuse in Hong Kong from an ecological perspective. Adolescent substance abuse is not only self-driven but also impacted by multiple risk and protective factors (Hemphill et al. 2011; Shek and Ma 2011a). On the one hand, adolescent substance abuse is correlated with some risk factors, such as familial economic disadvantage (Shek 2002), parental conflict and lack of parental care (Chilcoat and Anthony 1996; Shek 2003; Wagner et al. 2010), negative peer and social modeling (Garham et al. 1991), unstable and non-supportive community environment (Hemphill et al. 2011). On the other hand, assets of positive youth development and well family functioning are protective factors which protect adolescents from drug addiction. Studies showed that positive youth development programs could assist adolescents in preventing problem behaviors (Catalano et al. 2002; Shek 2010). Improvement of intrapersonal and interpersonal competencies (e.g., positive and healthy identity, cognitive, emotional, behavioral, and social competencies) also lowered the risk of adolescent substance abuse (Shek and Ma 2011a; Shek and Yu 2011a). Family functioning was also important to prevent adolescent substance abuse (Shek 2003; Shek and Ma 2011a), with mutual respect, good communication, and harmony inside the family help to create a caring and supportive environment for adolescents to discuss the harm of abusing drugs with their parents.

In particular, family attributes make a fundamental impact on adolescent substance abuse. These attributes can be described in three levels: economic status of the family (e.g., families on welfare), parental marital quality (e.g., divorce and separation), and parent–child relationship (e.g., mutual respect and communication). Adolescents from poor family may lack parental bonding because their parents need to work for a long time. As a result, they may have a higher risk of joining street gangs and developing addictive behaviors. Family intactness was also associated with adolescent substance abuse, with adolescents experiencing parental divorce having higher risk of substance abuse than did those who grow up in intact families (Shek and Yu 2011a). Parental modeling and parent–child relationship were considered more important than peer influence on adolescent’s engagement in substance abuse (Feit and Wodarski 2014). However, few scholars have systematically examined the influence of multiple family attributes on adolescent substance abuse behavior in a single study.

A review of the scientific literature on adolescent substance abuse showed that most studies are cross-sectional studies, and there are few studies on the overtime effects of multiple predictors on substance abuse, especially in middle adolescence. Besides, research examining the influence of factors in the individual and family systems on adolescent substance abuse is urgently needed. Shek and his colleagues are filling this research gap through conducting a serious of systematic longitudinal studies based on the Project PATHS (Shek and Ma 2011a, c, 2012; Shek and & Yu 2011a, b, 2012). These studies utilized classical longitudinal data analyses methods (e.g., cross-sectional and longitudinal regression analyses) and multi-level linear modeling methods (e.g., individual growth curves). In this study, we continued to use such statistical analyses to explore the developmental trajectories as well as the effects of individual and family factors on adolescent substance abuse over a period of 4 years.

Method

Participants and Procedures

In 2009/2010 school year, 3328 students (Secondary 1 or Grade 7, mean age = 12.59, with 51.7 % males and 47.2 % females) from 28 schools were recruited to a 6-year longitudinal research project titled “A Longitudinal Study of Adolescent Development and Their Families in Hong Kong.” Further details of the project can be seen elsewhere (Shek et al. 2014). The demographic characters of the students are shown in Table 1.

Table 1 Data profile across four waves
Full size table

They were invited to complete a questionnaire on adolescent development and the related psychosocial correlates in a self-administrated and voluntary manner. Written consent was obtained from all schools, parents, and participants. Enough time was given to the students to complete the questionnaire with the assistance of a well-trained research team (one research assistant for one class). Data cleaning and data matching were executed by well-trained research associates.

The present study utilized the data collected from Wave 1 to Wave 4. The attrition rates were acceptable, with 12.7 % of Wave 2, 14.1 % of Wave 3, and 19.4 % of Wave 4. Among those 3328 participants who had completed the questionnaire in Wave 1, 2682 students completed the questionnaires in all 4 years. Only those students who had took Wave 1 (N = 3328) were included in the later longitudinal data analyses.

Instruments

Assessment of Substance Abuse

The substance abuse scale contains eight questions which examine the frequency of different adolescent substance behaviors (such as tobacco, alcohol, and illicit drugs), on a six-point Likert scale (0 = never, 1 = 1–2 times, 2 = 3–5 times, 3 = more than 5 times, 4 = several times a month, 5 = several times a week, 6 = everyday). Previous study showed that the scale had good psychometric properties (Shek and Ma 2011a). Table 2 shows eight items of adolescent substance abuse and the frequency of related behavior for each item.

Table 2 Frequencies of substance abuse in the past 12 months across four waves
Full size table

A composite score was calculated for each wave by combining the scores of all items, with a higher score indicating a higher frequency of substance abuse. In multi-level modeling, substance abuse (SA) was a new variable which converted the measures (the above composite score at each wave) at Wave 1 to Wave 4 into a stacked format. In view of the prevalent use of tobacco and alcohol, additional analyses were also carried out by using Item 1 (“smoking”) and Item 2 (“drinking”) as separate indicators.

Assessment of Positive Youth Development

Positive youth development was assessed by the Chinese Positive Youth Development Scale (CPYDS), which includes 15 key attributes assessing adolescent’s holistic development. The scale was validated with good reliability and validity (Shek et al. 2007). In this study, the variable “positive youth development” was the mean score of the 15 constructs. Except spirituality (SP) which was a seven-point Likert scale, all other constructs were assessed on a six-point Likert scale.

Economic Disadvantage as a Family Attribute

Economic disadvantage was operationalized in terms of whether the respondent’s family received Comprehensive Social Security Assistance (CSSA), with “yes” as indicating economic difficulty. Missing data were removed from the analyses.

Family Intactness as a Family Attribute

Different marital statuses of the parents were assessed in the questionnaire (1 = divorced but not remarried, 2 = separated but not remarried, 3 = married (first marriage), 4 = married (second or subsequent marriage), 5 = others). “Intact families” referred to those students whose parents were in their first marriage (with the response of answer 3), while “non-intact families” were defined by the rest of the responses.

Family Functioning as a Family Attribute

Family functioning was assessed by the Chinese Family Assessment Instrument (CFAI) which assesses three major attributes: mutuality (mutual support, love, and concerns among family members), communication (frequency and nature of interaction among family members), and conflicts and harmony (the presence of conflicts and harmonious behavior in family). The variable “family functioning” was a combined score of the relevant items, with a higher score indicating a higher level of family functioning. Previous research showed that CFAI was a reliable and valid scale in the Chinese context (Shek and Ma 2010).

Research Questions and Hypotheses

Four observations were revealed in the previous studies. First, gender made a difference, with more male than female adolescents had substance abuse. Second, positive youth development constructs were inversely associated with substance abuse. Third, adolescents in non-intact families had more substance abuse than did intact families. Fourth, economic disadvantage was a risk factor of substance abuse behavior (Shek and Ma 2011a, c, 2012 Shek and Yu 2011a, b, 2012). On top of these findings, this study further explored the developmental trajectories and psychosocial predictors of adolescent substance abuse. The following research questions and several hypotheses were addressed in this study.

Research Question 1: Are Economic Disadvantage and Family Intactness Associated with the Initial Level and Growth Rate of Adolescent Substance Abuse?

Based on the previous findings, the following hypotheses were proposed:

  • Hypothesis 1: Consistent with the global scientific findings, the shape of the growth curves of substance abuse would show a rising trend across 4 years.

  • Hypothesis 2a: Based on the literature on risk factors, adolescents from poor families would have a higher initial level and faster growth rate of substance abuse than did those from nonpoor families.

  • Hypothesis 2b: Based on the literature on risk factors, adolescents from non-intact families would have a higher initial level and faster growth rate of substance abuse than did those from intact families.

Research Question 2: Do Family Functioning and Positive Youth Development Predict Substance Abuse Across Four Waves?

  • Hypothesis 3a: Family functioning at Wave 4 would have a negative relationship with substance abuse at Wave 4.

  • Hypothesis 3b: Positive youth development at Wave 4 would have a negative relationship with substance abuse at Wave 4.

  • Hypothesis 4a: Family functioning at Wave 1 would predict decline in substance abuse across 4 years, particularly for drinking and smoking.

  • Hypothesis 4b: Positive youth development at Wave 1 would predict decline in substance abuse across 4 years, particularly for drinking and smoking.

Data Analyses Strategies

Descriptive statistical analyses were performed to portrait the sociodemographic statuses of the participants from Wave 1 to Wave 4, including percentages of economic disadvantage and family intactness (Table 1), frequencies of eight single substance abuse behaviors (Table 2), and means, standard deviations, as well as internal consistency of the scales (Table 3).

Table 3 Descriptive statistics of key variables and internal consistency coefficients of scales (Waves 1–4)
Full size table

Linear mixed method (LMM) analyses were performed to understand the developmental trajectories of adolescent substance abuse in Hong Kong. We focused on the analyses of individual growth curves (IGC) which were generated by LMM through using SPSS 21 (IBM SPSS Statistics, IBM Corp, Somers, NY). The advantage of IGC and details of the method fitting the project can be found in Shek and Ma’s paper (2011b).

Four waves of data were analyzed with maximum likelihood (ML) as the estimate method (Shek and Ma 2011b). A two-level hierarchical model that nested time (Level 1) within individuals (Level 2) was tested by using: (a) substance abuse (included intercepts and slopes) as the dependent variables (DVs), (b) time (four waves) as the Level 1 predictor, and (c) individual social-demographic characteristics (economic disadvantage and family intactness in Wave 1 as the main predictors, while initial age and gender as the control variables) as the Level 2 predictors. The first level of analyses were the repeated measures of within-person initial substance abuse behavior as well as the rate of change over time without taking account of the individual social-demographic characteristics. The second level of analyses examined the effects of the individual social-demographic characteristics on the initial level as well as the developmental patterns of substance abuse.

For the Level 1 predictor, “time” was recoded (0 = Wave 1, 1 = Wave 2, 2 = Wave 3, and 3 = Wave 4). Dummy variables were used for the Level 2 predictors (“gender”: female = −1, male = 1; “economic disadvantage”: not receiving CSSA = −1, receiving CSSA = 1; “family intactness”: non-intact family = −1, intact family = 1). “Initial age” was grand mean centered in order to simplify the interpretation of results (Kwok et al. 2008; Shek and Ma 2011b). Following the strategies suggested by other scholars (Singer and Willet 2003; Shek and Ma 2011b), there were three steps in the LMM analyses:

  • Step 1: Unconditional mean model (Model 1) was conducted to measure individual substance abuse (intercepts only) without adding any level of predictors. It helped examine the total variance of substance abuse predicted by intrapersonal differences, which could indicate the need for further multi-level analyses.

  • Step 2: Unconditional growth model (Model 2) was performed to examine the substance abuse trajectories of the participants by combining the individual trajectories into a collective one. This was a Level 1 model which examined the variation (both initial status and rate of change) among individual over time without adding the higher level of predictors.

  • Step 3: Conditional growth model (Model 3) was a Level 2 model which evaluated the effects of interpersonal differences on initial substance abuse behavior as well as the rate of change (by adding both Level 1 predictor and Level 2 predictors). In this study, it focused on examining how the risk factors (economic disadvantage and non-intact family) predicted the initial status (intercepts) and rate of change (slopes) of adolescent substance abuse with initial age and gender controlled.

To further examine the longitudinal influences of protective factors (family functioning and positive youth development) on adolescent substance abuse, two multiple regression analyses were conducted. First, for the concurrent relationships among the variables, we tested whether Wave 4 family functioning and positive youth development would predict Wave 4 substance abuse behavior with gender, economic disadvantage, and family intactness controlled. Second, for the overtime effects, we tested whether Wave 1 family functioning and positive youth development would predict Wave 4 substance abuse behavior with gender, economic disadvantage, and family intactness from Wave 1 controlled. For both concurrent and overtime effects, we examined the predictions of the above two protective factors as a single predictor (Model A and Model B) and their predictions as simultaneous predictors (Model C).

Results

Adolescent Substance Abuse Across Time

Table 1 shows that percentages of adolescents from impoverished families or non-intact families were low, with the rate of receiving CSSA slightly deceased (from 6.8 % in Wave 1 to 4.9 % in Wave 4) and the rate of living in a non-intact family slightly increased (from 15.5 % in Wave 1 to 17.4 % in Wave 4). Table 2 reveals the percentages of frequently substance abuse in the past 12 months (for those who took the substance equal or more than several times a month) across four waves. Although some behaviors increased across time, overall substance abuse behavior in adolescents was not high. For example, there was an increasing trend of frequent smoking from 0.96 % in Wave 1 to 1.71 % in Wave 4. Similarly, the percentage of frequently taking alcohol increased consistently across four waves (from 2.42 % to 8.38 %). Drinking and smoking were the two most frequent substance abuse behaviors in all 4 years.

In Table 3, the mean scores of substance abuse increased consistently from Wave 1 to Wave 4 (0.09, 0.11, 0.12, 0.13), which indicated an increasing trend of substance abuse behavior in general over 4 years.

Male adolescents had higher mean score of substance abuse than did female adolescents (0.11 vs. 0.08, respectively, in Wave 1; 0.15 vs. 0.11, respectively, in Wave 4). Adolescents from non-intact families had higher mean abuse behavior than did those from intact families (0.14 vs. 0.08, respectively, in Wave 1; 0.16 vs. 0.12, respectively, in Wave 4). Adolescents who received CSSA (had economic disadvantage) had higher mean abuse behavior (0.16) than did those who did not receive CSSA (0.12) in Wave 4.

Linear mixed method was conducted to examine the developmental trajectories of adolescent substance abuse as well as the effects of economic disadvantage and family intactness on the initial status and rate of change of substance abuse behavior, while initial age and gender of the adolescents were controlled. The results of the three models were shown in Table 4.

Table 4 Results of individual growth curve of substance abuse for all models
Full size table

Firstly, the unconditional mean model (Model 1) was tested to identify the total variance of adolescent substance abuse at different levels using intraclass correlation coefficient (ICC) as an indicator (Singer and Willet 2003). Results showed that about 43.02 % of total variance of adolescent substance abuse (ICC = 0.4302) was due to interindividual differences. Therefore, the multi-level modeling analyses would be meaningful (Shek and Ma 2011b) to identify how different parameters relate to the development of adolescent substance abuse behavior. As we only had four points (W1 to W4), a linear trend would be sufficient.

Secondly, results showed that Model 2 had a better fit than Model 1 (Δχ2 (3) = 281.48; ΔAIC = 275.48; ΔBIC = 253.42; p < .001). In the unconditional growth model (Model 2), the initial level of substance abuse was very low (γ00 = .765, p < .001). From the linear slope, the substance abuse behavior increased linearly by an average of 0.147 per year (γ10 = .147, p < .001), which supported Hypothesis 1. On average, adolescent substance abuse behavior increased over the years. In addition, significance of two between-subject random-effect variances (intercept, p < .001; linear slope, p < .01) indicated the potentials of some interpersonal factors (Level 2 predictors) to explain the developmental patterns of adolescent substance abuse.

Finally, when adding the Level 2 predictors, the conditional growth model (Model 3) showed a better model fit than the unconditional growth model (Model 2), (Δχ2 (8) = 7562.92; ΔAIC = 7546.92; ΔBIC = 7490.63; p < .001). For the predictors (of both Level 1 and Level 2) of the intercept of substance abuse, results showed that family intactness was negatively associated with the initial level of substance abuse (γ04 = −.214, p < .001) above and beyond the effects of initial age (γ01 = .301, p < .001) and gender (γ02 = .069, p < .05). In the initial assessment, adolescents of non-intact families reported higher frequency of substance abuse; adolescents with older age reported a higher level of substance abuse than did those in younger age; male adolescents reported a higher level of substance abuse than did female adolescents. At the same time, family economic status had no significant association with the initial status of adolescent substance abuse. Around 20 % (from 1.293 in Model 2 to 1.035 in Model 3) of the variance of intercepts could be explained by Level 2 predictors.

For the predictors of slope, results of the fixed effects showed that both family intactness (γ14 = .051, p < .05) and economic disadvantage (γ13 = .094, p < .01) were significantly associated with the growth rate of substance abuse above and beyond the effects of gender (γ12 = .033, p < .05). Adolescents from poor families showed faster increasing rate in substance abuse behavior than did those from nonpoor families. Although adolescents from non-intact families reported slower increasing rate of their substance abuse behavior than did those from intact families, adolescents from non-intact families still had a higher mean of substance abuse (0.16) than did those from intact families (0.12) in Wave 4. In terms of gender, male adolescents reported a faster growth of substance abuse than did female adolescents.

In short, the above findings partly supported Hypothesis 2a (e.g., adolescents from poorer families showed a faster growth rate of substance abuse) but fully support Hypothesis 2b (e.g., adolescents from non-intact families had a higher initial level and faster growth rate of substance abuse than did those growing up in intact families).

Concurrent and Longitudinal Effects of Family Functioning and Positive Youth Development on Prevention of Adolescent Substance Abuse

For the concurrent effects at Wave 4 (Table 5), multiple regression analyses showed that family functioning and positive youth development were inversely related to substance abuse behavior no matter they were examined separately or simultaneously.

Table 5 Multiple regression analyses on substance abuse at Wave 4
Full size table

Gender was positively related with substance abuse behavior, which echoed with the LMM results that male adolescents had a higher risk of substance abuse. Supporting Hypothesis 3a and 3b, these findings indicated that adolescents with better concurrent family functioning and positive youth development demonstrated lower risk of substance abuse. Family functioning and positive youth development had unique direct effects on reducing adolescent substance addiction.

For the longitudinal effects, when Wave 1 family functioning and positive youth development were examined separately, they did not have significant contribution to the substance abuse behavior at Wave 4, with their initial levels controlled. Similarly, when they were examined simultaneously, family functioning and positive youth development at Wave 1 did not significantly predict Wave 4 substance abuse behavior (see Table 6) as well as the drinking behavior (see Table 7). However, positive youth development at Wave 1 predicted smoking at Wave 4 (see Table 8). In short, there was partial support for Hypotheses 4a and 4b only.

Table 6 Wave 1 variables predict Wave 4 substance abuse
Full size table
Table 7 Wave 1 variables predict Wave 4 substance abuse in drinking
Full size table
Table 8 Wave 1 variables predict Wave 4 substance abuse in smoking
Full size table

Discussion

Compared with the previous longitudinal studies in the same research project (Shek and Ma 2011a; Shek and Yu 2011a), the study applied advanced multi-level analyses to further examine the trend and psychosocial correlates of adolescent substance abuse. Consistent with previous practice (Shek and Ma 2011b, c), individual growth curve modeling was applied to analyze the longitudinal data. By analyzing different models of individual trajectories and change of adolescent substance abuse, key constructs of risk factors (economic disadvantage and family intactness) and protective factors (positive youth development and family functioning) of adolescent substance abuse could be identified. The study provides rich information on the normative profile of adolescent substance abuse in Hong Kong over time, hence expanding our understanding of the objective picture of the developmental patterns of substance abuse from early adolescence to mid-adolescence.

Consistent with our expectation, adolescent substance abuse behavior increased in the adolescent years especially in the areas of smoking and drinking. This phenomenon may be partly associated with the addiction culture in young people and the relatively ease in getting tobacco and alcohol. Although the prevalence of adolescent substance abuse was not high, the rising rate of smoking and drinking suggests that there is a need to step up preventive education on adolescence substance abuse. Considering the risk factors of adolescent substance abuse, we examined how economic disadvantage and family intactness influenced the developmental patterns of adolescent substance abuse by controlling initial age and gender. We found that adolescents of non-intact families showed a higher risk of having substance abuse behavior initially. Although adolescent substance abuse of non-intact families increased slower than did the adolescents in intact families, they still had a higher mean abuse behavior at Wave 4 compared to those with parent staying in the first marriage. Consistent with our expectation, substance abuse behavior in adolescents from poor families increased faster than did those from nonpoor families. These findings echo the previous literature on the negative impacts of economic disadvantage and non-intact parental relationship on adolescent substance abuse (Feit and Wodarski 2014; Shek and Yu 2011a), and it enriches the related Chinese literature.

Furthermore, the positive attributes such as positive youth development and family function had positive effects on prevention of adolescent substance abuse. Both attributes had significant negative correlation with substance abuse behavior. However, the significant long-term effects of these two attributes were not established in this study. There are two possible explanations for these findings. First, as high-risk adolescents might dropout after Secondary 3, the sample at Wave 4 may be less at risk in terms of substance abuse, hence making the spread of scores in substance abuse less variable. Second, the influence of family functioning and positive youth development on substance abuse may weaken over time. Some research has shown that overconfidence and social maturity at an early age may in fact lead to more adolescent risk behavior (Lerner et al. 2005).

The study is not without limitations. First, the data were based on self-report format. Due to social desirability effect, they may not fully reflect the substance abuse situation in early adolescents. Therefore, replication studies with multiple methods of assessment (such as assessing the views of parents and teachers) are necessary to further examine the interrelationships among adolescent substance abuse and family attributes. Second, it will be helpful to explore other important potential determinants according to the ecological perspective (Shek 2002). Besides examination of the family environment, factors from the social environment, school environment, and peer environment should be examined in future analyses. In particular, as peer influence is important in early and middle adolescence, there is a need to take into account the related factors. Last but not least, family functioning as an important protective factor should be further explored. As parents are the live “models” of adolescents, it will be helpful to study the role of parenting style and parent–child relationship in the prevention of adolescent substance abuse.