Journal of Abnormal Child Psychology

, Volume 36, Issue 7, pp 1029–1045

Rumination in Response to Stress as a Common Vulnerability Factor to Depression and Substance Misuse in Adolescence

Authors

    • Department of PsychologyMcGill University
  • John R. Z. Abela
    • Department of PsychologyMcGill University
Article

DOI: 10.1007/s10802-008-9233-9

Cite this article as:
Skitch, S.A. & Abela, J.R.Z. J Abnorm Child Psychol (2008) 36: 1029. doi:10.1007/s10802-008-9233-9

Abstract

The current study examined rumination in response to stress as a common vulnerability factor to both depression and substance use problems in adolescence. Specifically, we used a multi-wave longitudinal design to examine whether adolescents who tend to ruminate in response to stress exhibit increases in depressive symptoms and substance misuse following the occurrence of negative events. At time 1, adolescents (n = 161) completed measures assessing depressive symptoms, substance misuse, and the tendency to ruminate in response to stress. Every 6 weeks for the next 18 weeks participants completed measures assessing the occurrence of negative events, depressive symptoms, and substance misuse. Hierarchical linear modeling analyses indicated that adolescents who tend to ruminate in response to stress report greater elevations in depressive symptoms and substance misuse following elevations in negative events than other adolescents. The relationship between rumination, negative events, and substance misuse was moderated by age. Support was not obtained for fluctuations in depressive symptoms as a mediator of the relationship between negative events and substance misuse. Fluctuations in negative affect, however, were found to mediate this relationship.

Keywords

DepressionSubstance useAdolescenceComorbidityRuminationDiathesis-stress

Introduction

Adolescence is a critical period for understanding the etiology of both depressive disorders and substance use problems.1 Rates of depressive disorders increase six-fold between 13 and 18 years of age (Hankin et al. 1998) with community studies indicating that approximately 20% of adolescents will have experienced an episode of major depression by age 18 (Lewinsohn et al. 1998). Similarly, levels of substance usage increase markedly throughout adolescence (Johnston et al. 2006) and by age 18 between 8% to 20% of adolescents will meet lifetime criteria for a substance use disorder (Lewinsohn et al. 1993; Young et al. 2002). Not only do prevalence rates of both depressive and substance use disorders increase considerably during adolescence, but there is also substantial comorbidity between these two disorders. Community studies indicate that adolescent substance users are over three times more likely to have suffered a major depressive episode than adolescents who are not substance users (Armstrong and Costello 2002). Comorbidity is of particular concern as adolescents with both disorders have poorer global functioning and are at increased risk for making a suicide attempt compared to adolescents with either disorder alone (Lewinsohn et al. 1995). As a result of such alarming statistics, there is growing recognition of the need for research examining the mechanisms contributing to co-occurring depression and substance use problems in adolescence (Volkow 2004).

A variety of models have been proposed to account for the high comorbidity between depressive disorders and substance use disorders (for review see Swendsen and Merikangas 2000). These models can be broadly delineated into casual models and common factors models. Causal models postulate that the development of one disorder increases risk for the development of the other disorder. For example, self-medication models propose that substance use may be employed as means of self-medicating depressive symptoms, a pattern of consumption which is posited to lead to the development of substance use problems (Cooper et al. 1995). Alternatively, secondary depression models propose that negative consequences of excessive substance use such as academic failure and disruptions in interpersonal relationships lead to the development of depression. Common factor models postulate that the comorbidity between depressive and substance use disorders arises as a result of shared etiological risk factors that confer vulnerability to both disorders. For example, Brady and Sinha (2005) propose that repeated occurrence of negative life events increases vulnerability to the development of both depressive and substance use disorders. It is important to note that multiple models of comorbidity are not necessarily mutually exclusive. Rather, both causal and common factor models may operate in concert to explain the association between depression and substance use problems (Swendsen and Merikangas 2000).

Individual differences in self-focused attention have been identified as a common vulnerability factor in a variety of psychopathological conditions including depression and substance use disorders. Self-focused attention is a cognitive variable defined as “an awareness of self-referent, internally generated information that stands in contrast to an awareness of externally generated information derived through sensory receptors” (Ingram 1990, p. 156). Ingram (1990) proposes that individuals differ in their tendency towards self-focusing and that self-focused attention that is excessive, sustained, and inflexible confers vulnerability to both depressive and substance use disorders. Specifically, he posits that maladaptive self-focused attention interacts with the occurrence of negative events to predict initiation and intensification of negative affect. As negative affect is a central characteristic of depression, the interaction between self-focused attention and stress is expected to predict increased depressive symptoms. Ingram (1990) also notes that individuals high in maladaptive dispositional self-awareness may employ substance use as a means of ameliorating the psychological distress that accompanies negative events. As such, depressive symptoms can be conceptualized as mediating the relationship between negative events and substance misuse among individuals prone to maladaptive self-focused attention. The current study will focus on individual differences in self-focused attention as a vulnerability factor to both depression and substance use problems in adolescence.

Self-Focused Attention and Depression

A variety of theorists have proposed models emphasizing self-focused attention as a predictor of depressive symptoms (for a review see Ingram et al. 1998). Duval and Wicklund (1972) theorize that heightened self-focused attention increases levels of negative self-evaluations which in turn leads to heightened negative affect. Other theorists have emphasized that maladaptive self-focused attention must be activated by negative events to predict depressive symptoms. According to Pyszczynski and Greenberg (1987), individuals who are vulnerable to depression have a tendency to self-focus following the occurrence of negative events but not positive events. In turn, self-focused attention is posited to lead to increased salience of negative self-aspects and intensification of negative affect which ultimately leads to depression. Similarly, Lewinsohn et al. (1985) propose that in vulnerable individuals negative events lead to a state of heightened self-awareness which mediates the relationship between stress and the development of depressive symptoms. Consistent with theoretical models that link excessive self-focused attention to depression, research conducted with both community and clinical samples has found that individuals who are high in dispositional self-awareness tend to report higher levels of depressive symptoms (Ingram et al. 1987; Ingram and Smith 1984; Larsen and Cowan 1988; Sloan 2005). In addition, empirical work has indicated that heightened self-focused attention is associated with a variety of corollaries of depression including intensification of negative affect, lowered self-esteem, and increased self-criticism (for review see Wells and Matthews 1994).

A large body of research has focused on a particular form of maladaptive self-focused attention, rumination, which has been found to be an especially potent predictor of depressive symptoms. Rumination is conceptualized as a style of responding to depressed mood defined as “repetitively focusing on the fact that one is depressed; on one’s symptoms of depression; and on the causes, meanings, and consequences of depressive symptoms” (Nolen-Hoeksema 1991, p. 569). Nolen-Hoeksema (1987, 1991) proposes that individuals who tend to ruminate in response to their depressed mood will experience longer lasting and more severe depressive symptoms than those who distract themselves from their depressed mood. Rumination is proposed to increase depression by increasing the impact of negative affect on cognitive processing and by interfering with the implementation of effective, problem-solving behaviors that could alleviate depressive symptoms. Although Nolen-Hoeksema (1991) emphasized rumination as a style of responding to depressive symptoms she also posited that individuals can ruminate in response to negative life events. Irrespective of whether rumination is induced by negative events or depressed mood, “the key characteristic of a ruminative response style is that people are focusing on their negative emotional state” (Nolen-Hoeksema 1991, p. 569). Rumination is distinguished from more general forms of self-focused attention by its specific focus on emotional content and its repetitive nature. Meta-analytic studies indicate that rumination is more strongly predictive of depressive symptoms that more general forms of self-focused attention (Mor and Winquist 2002).

Studies conducted with adult samples have consistently indicated that rumination predicts increases in depressive symptoms and the onset of depressive episodes even after controlling for initial levels of depressive symptoms (for reviews see Lyubomirsky and Tkach 2003; Nolen-Hoeksema et al. 2008). Although there have been fewer studies conducted examining rumination in youth samples, there is evidence that rumination prospectively predicts increases in depressive symptoms among children and adolescents (Abela and Hankin 2007). Recently researchers have posited that rumination is a multi-dimensional construct that includes both efforts to reflect on the causes of problems and passively brooding about the symptoms and consequences of depression (Treynor et al. 2003; Watkins 2004). Factor analytic studies have found support for two different components of rumination with only the brooding component prospectively predicting depressive symptoms (Treynor et al. 2003). Similarly, research conducted with adolescents has found that the brooding component of rumination uniquely predicts elevations in depressive symptoms while controlling for the reflection component of rumination (Burwell and Shirk 2007).

There have only been a few studies conducted examining the interaction between rumination and stress in predicting levels of depressive symptoms and results have been mixed. Providing support for a diathesis-stress conceptualization, Kraaij et al. (2003) reported that rumination moderated the concurrent association between negative life events and depressive symptoms in a sample of older adolescents with the association being strongest among adolescents reporting high levels of rumination. In contrast, a study examining reactions to a college midterm exam found that rumination and exam results did not interact to predict depressive symptoms over time (Sarin et al. 2005). Although Abela et al. (2008) found that that the main effect of rumination prospectively predicted increases in depressive symptoms in a sample of ninth graders, they did not find a significant interaction between rumination and negative events. Also failing to provide support for a diathesis-stress conceptualization, Schwartz and Koenig (1996) reported that the interaction between rumination and negative events failed to predict increases in depressive symptoms at a 6-week follow-up in a sample of adolescents.

These prospective studies may have failed to find support for an interaction between rumination and negative events because they all measured rumination as a response to depressed mood rather than specifically as a reaction to stress. Theorists have proposed that rumination in response to depression and rumination in response to stress may represent distinct processes that operate differently in predicting depressive symptoms (Sakamoto 2000; Spasojevic et al. 2003). In support of this view, Robinson and Alloy (2003) found that in a sample of non-depressed college students rumination in response to stress prospectively predicted the onset of depressive episodes whereas rumination in response to depression did not. To our knowledge no prospective studies have examined the interaction between rumination when measured as a response to stress and the occurrence of negative events as a predictor of depressive symptoms.

Self-Focused Attention and Substance Use Problems

Various theorists have proposed that a propensity towards maladaptive self-focused attention is associated with substance consumption to escape from aversive self-awareness and negative affect following the occurrence of negative events (e.g. Hull 1981; Baumeister 1992). Hull’s (1981) model specifically focuses on alcohol consumption and predicts that individuals who are high in private self-consciousness are likely to consume alcohol heavily following the occurrence of negative events. Private self-consciousness is defined as a trait tendency to focus on and analyze one’s inner thoughts and feelings and generally be more self-aware (Fenigstein et al. 1975). According to Hull’s model, individuals who are high in dispositional self-focused attention tend to experience disproportionately elevated levels of negative self-evaluations and negative affect following events that indicate failure. Alcohol consumption is posited to interfere with cognitive processes necessary to maintain self-focused attention and as a result reduce levels of self-awareness. As such, individuals who are high in self-consciousness are predicted to consume alcohol as a means of reducing negative affect following the occurrence of stress. Although Hull’s model focuses exclusively on alcohol consumption, other types of drugs produce comparable impairments in cognitive functioning (Heishman et al. 1997) and may also be used as a means of reducing aversive self-focused attention.

Research conducted with adult samples has provided support for a relationship between individual differences in levels of dispositional self-awareness and levels of substance consumption. In an experimental study, Hull and Young (1983) found that those individuals who reported high levels of private self-consciousness drank more heavily following a manipulated failure experience than following a manipulated success experience. Alcohol consumption did not vary between success and failure conditions for participants low in private self-consciousness. As well, the interaction between private self-consciousness and negative events predicted alcohol relapse in a sample of patients followed for 3 months following completion of treatment for alcohol dependence (Hull et al. 1986).

Studies conducted with adolescents have also provided some support for an association between heightened dispositional self-awareness and greater substance consumption. Steinhausen and Winkler Metzke (2003) reported that adolescent problem drinkers reported higher levels of self-awareness related cognitions than social drinkers and non-drinkers. Providing partial support, Plueddemann et al. (1999) reported that in a sample of older adolescents, only male problem drinkers reported higher levels of self-awareness. Hull et al. (1986) reported that academic failure was correlated with levels of alcohol consumption but only for adolescents reporting high levels of private self-consciousness. Similarly, psychological distress was associated with levels of marijuana consumption but only for adolescent reporting high levels of self-consciousness (Zablocki et al. 1991). Chassin et al. (1988) provided the most comprehensive published examination of the self-awareness model of alcohol consumption using a two-time point longitudinal design and multiple community adolescent samples. In line with the self-awareness model, negative life events predicted levels of problematic drinking only among adolescents possessing high levels of private self-consciousness. Overall, support for the model was mixed, however, as this result was obtained in only one of the two samples studied.

The majority of studies examining the relationship between self-focused attention and substance use problems have focused on individual differences in private self-consciousness, a general measure of dispositional self-focused attention. It is important to note, however, that self-focused attention model of substance use problems emphasize that substance consumption is employed to reduce negative affective states. Rumination is a stronger predictor of depressive symptoms than general forms of self-focused attention and should therefore be more strongly related to substance use problems. There have been only a few studies examining the relationship between rumination and substance use, however. In a study conducted with a college student sample, Nolen-Hoeksema and Morrow (1991) reported that higher levels of rumination were correlated with a tendency to engage in reckless behaviors, including substance use, in response to a depressed mood. Similarly, in community adult sample, rumination was associated with the occurrence of more alcohol problems and a greater tendency to use substances to cope with stress (Nolen-Hoeksema and Harrell 2002). Furthermore, these authors found that rumination prospectively predicted alcohol problems at a 1 year follow-up among female participants. The prospective relationship between rumination and alcohol problems among males was of similar magnitude but only approached significance. In a sample of adolescent females, rumination was found to prospectively predict onset of substance abuse disorders and increases in substance abuse symptoms over a 4-year follow-up (Nolen-Hoeksema et al. 2007). On the other hand, Goldstein (2006) did not find that rumination was associated with alcohol consumption in a college student sample. Overall, these results provide partial support for rumination as a vulnerability factor for substance use problems.

To date there have been no published studies examining the association between rumination, the occurrence of negative events, and substance use problems. Such research is needed given that self-focused attention is posited to function within a diathesis-stress framework to predict excessive substance consumption (Ingram 1990). Furthermore, to our knowledge, no studies have examined depressive symptoms as a mediator of the relationship between negative events and substance use problems among individual who tend to ruminate in response to stress.

Current Study

The goal of the current study is to provide a stringent examination of individual differences in the tendency to ruminate in response to stress as a common vulnerability factor to both depressive symptoms and substance use problems. The procedure involved an initial assessment in which adolescents completed measures assessing the tendency to ruminate in response to stress, depressive symptoms, and substance misuse. The procedure also involved a series of follow-up assessments, every 6 weeks for the subsequent 18 weeks, in which adolescents completed measures assessing depressive symptoms, substance misuse, and the occurrence of negative events. The use of such a multi-wave longitudinal design allowed us to take an idiographic approach towards examining the diathesis-stress component of our model of rumination as a vulnerability factor to both depressive symptoms and substance misuse. More specifically, we examined whether the slope of the relationship between negative life events and maladaptive outcomes (either depressive symptoms or substance misuse) within participants varied across participants as a function of levels of rumination. Higher levels of rumination were hypothesized to be associated with greater increases in depressive symptoms following the occurrence of negative events. Similarly, we hypothesized that higher levels of rumination would be associated with greater increases in substance use problems following the occurrence of negative events. Furthermore, analyses were conducted to examine depressive symptoms as a mediator of the relationship between rumination and elevated levels of substance misuse following high levels of stress. Specifically we examined two alternative models: (1) a self-medication model in which fluctuations in depressive symptoms mediates the relationship between fluctuations in negative events and substance misuse and (2) a secondary depression model in which fluctuations in levels of substance misuse mediates the relationship between fluctuations in negative events and depressive symptoms. Consistent with self-awareness model of substance consumption, we hypothesized that support would be obtained for the self-medication model but not the secondary depression model.

Methods

Participants

Participants in the current study were recruited from secondary schools in the greater Montreal area. Letters describing the project were sent to the principals of all English-language schools. Seven schools agreed to participate. The schools were located in both suburban (four schools) and urban (three schools) areas and students were drawn from neighbourhoods representing a diversity of socioeconomic groups. Although the project was open to students in all grades (i.e. grades 7–11) at each participating school, only select classes were available to participate. Consent forms were distributed to the parents of each adolescent who was eligible to participate in the project as well as to the adolescents themselves. Both parental and personal consent were required to participate. Consent rates varied between schools from 6% to 38% with a median rate of 21%. There was no systematic difference between consent rates obtained at school situated in rural neighbourhoods as compared to schools situated in urban neighbourhoods. The final sample included 161 students (46% male and 54% female) whose ages ranged from 12 to 18 (\(\overline X \) = 15.17; Mdn = 15; SD = 1.22). The sample was 78.9% Caucasian, 5.6% Asian, 4.3% Black, 3.1% East-Indian, 2.5% Native American, 1.2% Hispanic, and 3.1% reported other as their ethnicity. The ethnic composition of the current sample was found to be representative of the demographic profile of the community from which it was drawn.2 Participant’s mother tongue included English (83.2%), French (13.7%), and other (3.1%).

Procedure

Following the collection of consent forms, researchers went to each school to complete the initial assessment with the adolescents. At the start of each assessment, students were reminded of their right to withdraw from the study at any time. Questionnaires were completed by each participant with research assistants available at all times to answer any questions. During the initial assessment, adolescents completed a demographics form and the following questionnaires: (1) Center for Epidemiologic Studies Depression Scale (CES-D; Radloff 1977), (2) Substance Misuse Severity Measure (SMSM), and the (3) Responses to Stress Questionnaire (RSS; Connor-Smith et al. 2000). Follow-up assessments occurred at 6-week intervals over the subsequent 18 weeks (Times 2–4). At each follow-up assessment participants completed the following questionnaires: (1) Adolescent Life Events Questionnaire (ALEQ; Hankin and Abramson 2002), (2) CES-D, and the (3) SMSM. Following the final assessment, all adolescents were debriefed and invited to participate in a workshop modeled after the Penn Depression Prevention Program (Jaycox et al. 1994).

Measures

Adolescent Life Events Questionnaire (ALEQ; Hankin and Abramson 2002)

The ALEQ is a self-report questionnaire that was developed to assess a broad range of negative life events (e.g., school/achievement problems, friendship and romantic problems, and family problems) that typically occur in the lives of adolescents. Examples of items include, “you fought with your parents over your personal goals, desires, or choice of friends,” “you did poorly on or failed a test or class project,” and “you had an argument with a close friend.” Participants were asked to indicate how often such events occurred in the past month on a Likert scale ranging from 1 (never) to 5 (always) with higher scores reflecting a greater number of negative life events. Past research has found that the ALEQ is both reliable and valid (Hankin and Abramson 2002). In the current study, Cronbach’s alphas ranged from 0.92 to 0.93 across administrations indicating high internal consistency.

Center for Epidemiologic Studies Depression Scale (CES-D; Radloff 1977)

The CES-D is a 20-item self-report measure that assesses level of depressive symptoms in the past week. Examples of questions include: “I felt sad,” “I felt hopeless about the future,” and “I felt lonely.” Items on the scale range from 0 to 3 and higher scores reflect greater depressive symptomology. Symptoms are assessed in several domains including mood items, interpersonal items, and somatic items. Adolescents reported how they felt during the past week using the following scale: rarely (<1 day), some or a little of the time (1–2 days), occasionally or a moderate amount of time (3–4 days), and most or all of the time (5–7 days). The CES-D has been shown across numerous studies to have strong test–retest reliability and high correlations with other measures of depressive symptoms (Roberts et al. 1990; Phillips et al. 2006). In the current study, Cronbach’s alphas ranged from 0.89 to 0.94 across administrations indicating high internal consistency.

Substance Misuse Severity Measure (SMSM)

This measure was created the by the authors to measure adolescent’s level of substance misuse and is divided into two sections. The first portion of the measure assessed past month consumption of alcohol, marijuana, and other types of illicit drugs. Alcohol items on the measure asked participants to indicate their past month frequency of drinking, typical number of drinks consumed, number of binge episodes (five or more drinks), and frequency of drinking “to get drunk.” Additionally, participants reported their frequency of marijuana consumption and use of other illicit drugs. Participants responded to items using a five-point Likert scale to indicate their consumption level (e.g. Never, once, 2–3 times, 4–7 times, 8 or more times over the past month). Previous research has indicated that adolescents are able to provide valid self-reports of their substance use using a questionnaire assessment when they are assured of confidentiality (Winters et al. 1990). Cronbach’s alphas for the consumption items ranged from 0.77 to 0.82 across administrations indicating moderate internal consistency.

The second portion of the substance misuse measure assessed the occurrence of various negative consequences as a result of drug and alcohol use. Items were derived from the Rutgers Alcohol Problems Index (RAPI; White and Labouvie 1989), which was developed to provide a brief, unidimensional measure of the severity of adolescent drinking problems. In the current study, however, we were interested in the occurrence of negative consequences as a result of all types of substance use. As a result, participants were instructed to indicate negative consequences that had occurred as a result of alcohol consumption and/or drug use (for an example of a similar approach see Measelle et al. 2006). At the initial assessment adolescents were asked to report negative consequences that occurred in the past year, whereas at each follow-up assessment adolescents reported negative consequences that had occurred in the past month. Examples of items include: “Not able to do your homework or study for a test,” “Caused shame or embarrassment to someone,” and “Passed out or fainted suddenly.” Participants indicated the frequency that they had experienced each negative consequence using the following five-point Likert scale: never (0 times), rarely (1–2 times), sometimes (3–5 times), frequently (6–10 times), or more than ten times. Previous research has indicated that the RAPI items have strong validity, internal consistency, and test–retest reliability (Miller et al. 2002; White and Labouvie 1989). In the current study, Cronbach’s alphas for the subscale ranged from 0.89 to 0.95 across administrations indicating high internal consistency.

Responses on the two sections of the substance abuse measure were summed to provide a composite measure of adolescent’s level of hazardous substance use.3 Higher scores on the measure indicate heavier levels of substance misuse. In the current study, Cronbach’s alphas ranged from 0.89 to 0.94 across administrations indicating high internal consistency.

Responses to Stress Scale (RSS, Connor-Smith et al. 2000)

The RSS is a 57-item self-report measure that assesses a broad variety of voluntary coping responses and involuntary responses that adolescents employ when confronted with stressful situations. Adolescents use a four-point Likert scale to indicate how often they use a particular response when confronted with a stressor (from not at all to a lot). Previous research, conducted in multiple samples of adolescents, has indicated that the RSS demonstrates good reliability and validity (Connor-Smith et al. 2000; Wadsworth et al. 2004).

There are 19-factor analytically derived scales on the RSS; however, given the current study hypotheses we focused exclusively on the Rumination subscale as a measure of the tendency to ruminate in response to negative life events. The Rumination subscale consists of three items (e.g. When problems come up, I can’t stop thinking about how I am feeling.) and higher scores on the Rumination subscale indicate a tendency to respond to negative life events by focusing uncontrollably on thoughts and emotions following the occurrence of a stressor. Previous research has indicated that Rumination subscale is moderately correlated with other measures of rumination in adolescent samples (Abela and Skitch, in preparation; Burwell and Shirk 2007). In addition, the Rumination subscale has been found to be more strongly related to the depressogenic brooding component of rumination rather than the reflection component of rumination (Abela and Skitch, in preparation). In the current study, Cronbach’s alpha for the Rumination subscale was 0.77 indicating moderate internal consistency.

Results

Descriptive Statistics

Means, standard deviations, and Pearson correlations between Time 1 measures and adolescents’ age are presented in Table 1. Although higher levels of initial depressive symptoms were associated with higher rumination scores, the association between rumination and substance use problem scores was not significant. Levels of depressive symptoms were positively related to reports of substance use problems. Age was positively associated with reports of substance misuse at the initial assessment. Males and females were compared using t tests to examine whether there were sex differences on any measures. These analyses indicated that females (\(\overline X \) = 7.02) reported significantly greater rumination scores than males These analyses indicated that females (\(\overline X \) = 6.15, t(159) = 2.18, p < 0.05) Further, in the current sample female participants (\(\overline X \) = 15.43) were significantly older than male participants (\(\overline X \) = 14.86, t(159) = 2.18, p < 0.01). There were no significant sex differences in initial levels of depressive symptoms or substance use problems.
Table 1

Means, Standard Deviations, and Intercorrelations Between All Time 1 Measures

 

1

2

3

4

CESD

   

SUBABUSE

0.20**

  

RUMINATION

0.35**

0.03

 

AGE

0.03

0.30**

0.04

Mean

13.84

10.91

6.63

15.17

SD

10.44

14.89

2.51

1.22

CES-D Center for Epidemiologic Studies Depression Scale, SUBABUSE Substance Misuse Severity Measure, RUMINATION Responses to Stress Scale, Rumination Subscale, GENDER Coded variable (GENDER = 0 for males and 1 for females)

**p < 0.01

Adolescents who completed the initial assessment were invited to participate in all subsequent follow-up assessments. The majority of adolescents (73%) completed all three follow-up assessments with an additional 20% completing two of the three follow-ups. The number of follow-up assessments completed was not significantly correlated with scores on any Time 1 measures or demographic variables. Additional t tests were conducted to determine whether participants with incomplete follow-up data differed from participants with complete follow-up data on any of the follow-up measures. None of these analyses were significant. Means and standard deviations for the ALEQ, CES-D, and SMSM scores across the three follow-up assessments are reported in Table 2. As each adolescent completed these measures at multiple follow-up assessments, each adolescent has his/her own mean level on the measure (e.g. his/her average CES-D score during the follow-up interval) as well as his/her own degree of variation in scores on each measure during the follow-up interval (e.g. his/her SD on the CES-D across administrations). Descriptive statistics for within-subject means and standard deviations on each follow-up measure are reported in Table 3.
Table 2

Means and Standard Deviations of Stress Scores, Depressive Symptoms, and Substance Use Variables over Three Follow-up Assessments

 

Follow-up number

1

2

3

ALEQ

Mean

100.05

93.57

88.89

SD

24.06

23.76

22.25

CES-D

Mean

12.95

11.37

9.96

SD

10.27

10.99

10.57

SUBABUSE

Mean

8.56

7.21

6.84

SD

10.40

9.97

9.07

ALEQ Adolescent Life Events Questionnaire, CES-D Center for Epidemiologic Studies Depression Scale, SUBABUSE Substance Misuse Severity Measure

Table 3

Mean and Standard Deviations of Within-Subject Means and Standard Deviations for Follow-up Stress, Depressive Symptoms, and Substance Use Problems

 

Within-subject

Mean (μ)

SD

ALEQ

Within-subject mean

95.33

9.28

Within-subject standard deviation

22.32

6.44

CES-D

Within-subject mean

11.60

4.02

Within-subject standard deviation

9.88

3.54

SUBABUSE

Within-subject mean

7.80

3.38

Within-subject standard deviation

9.35

4.07

ALEQ Adolescent Life Events Questionnaire, CES-D Center for Epidemiologic Studies Depression Scale, SUBABUSE Substance Misuse Severity Measure

Multilevel Regression

Overview of Analysis

To examine the associations between stress, depressive symptoms, and substance misuse we utilized multilevel modeling analysis as our primary data analytic strategy. This data analytic approach is becoming increasingly applied to social and clinical research involving multiwave longitudinal data. Some of the advantages of multilevel modeling for longitudinal analysis are well-known (Raudenbush and Byrk 2002; Snijders and Bosker 1999), including the relaxation of the requirement of balanced (non-missing) data and the capacity to model a variety of error covariance structures. Such structures provide alternative methods of specifying the pattern of covariance between observations taken from the same subject at different time points. As well, it is important to note that multilevel modeling allows the examination of the relationship between within-subject and between-subject factors which are inherent to multiwave longitudinal data sets in which observations are nested within participants. Analyses were carried out using the SAS (version 8.1) MIXED procedure and maximum likelihood estimation.

In multilevel modeling analysis, hierarchically organized sets of regression equations are simultaneously estimated. In our analyses, the “Level I” equation represents the within-person relationship between the predictor variable and the dependent variable. For example, Eq. 1 shows the equation regressing depressive symptoms on stress.
$${\text{CES - D}}_{ti} = \beta _{0i} + \beta _{1i} \left( {{\text{FU}}\_{\text{STRESS}}} \right) + e_{ti} $$
(1)
In this equation, CES-D is participant i’s depressive symptoms at time t, β0i is the predicted value of depressive symptoms for participant i when Stress equals zero, β1i is the within-person slope of the stress depressive symptoms relationship for person i, and eti is a within person residual component. For all analyses, within-subject (Level I) predictors were centered at each participant’s mean prior to analyses, such that fluctuations in FU_STRESS reflects upwards or downwards fluctuations in a adolescent’s level of stress compared to his or her mean level of stress.
The “Level II” portion of the model consists of a set of two equations that represent how the within-subject relationship between the predictor and dependent variable varies as a result of between-subject factors. The first portion of the Level II model, also know as the intercept-as-outcome model, predicts variation in each participant’s average level of the dependent variable as a function of between-person differences. For example, Eq. 2 shows each participant’s Level I intercept (β0i or average depressive symptoms) regressed on the between person characteristics of rumination subscale scores (RUMINATION), average levels of stress across the follow-up assessments (MN_STRESS), initial depressive symptoms (T1_CES-D), age, gender, and a random person effect (μ0i).
$$\beta _{0i} = \gamma _{00} + \gamma _{01} \left( {{\text{RUMINATION}}} \right) + \gamma _{02} \left( {{\text{MN}}\_{\text{STRESS}}} \right) + \gamma _{03} \left( {{\text{T1}}\_{\text{CES - D}}} \right) + \gamma _{04} \left( {{\text{AGE}}} \right) + \gamma _{05} \left( {{\text{GENDER}}} \right) + \mu _{0i} $$
(2)
Participant’s average level follow-up of stress is included in the intercept-as-outcome model to reintroduce variability lost by person centering of the predictor in Level 1 as recommended by Raudenbush and Byrk (2002). Initial depressive symptoms are included in this model to control for individual differences in baseline depressive symptoms. Demographic variables were controlled for given previous findings of significant age and gender effects in depression and substance use among adolescents (Hankin et al. 1998; Johnston et al. 2006).
The second portion of the Level II model, also know as the slope-as-outcome model, predicts variation in the within-subject relationship between the Level 1 independent and dependent variable as a function of between-subject differences. For example, Eq. 3 shows each participant’s Level 1 slope (β1i or within subject relationship between fluctuation in stress and depressive symptoms) regressed on the between-person characteristic of rumination subscale scores (RUMINATION) and a random person effect (μ0i).
$$\beta _{1i} = \gamma _{10} + \gamma _{11} \left( {{\text{RUMINATION}}} \right) + \mu _{1i} {\text{ }}$$
(3)
For all analyses, between-subject (Level II) predictors were standardized prior to analyses.

Model specification was achieved using a sequential strategy, which involved first examining random effect components and then examining fixed effect components (for further discussion see Snijders and Bosker 1999). As mentioned previously, models initially included random effects for intercept (RE_PARTICIPANT) and slope (RE_SLOPE).4 Additionally, each model included an autoregressive covariance parameter (AR) to account for the correlation between within-subject residuals over time, which tends to occur in studies in which multiple responses are longitudinally obtained from the same individual. Non-significant random-effect parameters were deleted from the model prior to examining the fixed-effects component. For all analyses reported, the random effect for intercept (RE_PARTICIPANT) and the AR(1) parameter were significant and were retained in the model when examining the fixed effects. Fixed effects components of the models were specified through a process of backwards deletion. Specifically, higher order terms were examined and removed if non-significant prior to estimating the significance of lower order terms.

In order to examine possible gender differences, preliminary analyses were conducted examining if gender moderated the Level I relationship between predictor and dependent variables. If any significant interaction effects were found then gender interactions were included in the final model. Preliminary analysis also examined if age significantly moderated any Level I relationships. If any significant interaction effects were found then age interactions were included in the final model. In cases where no significant age or gender interactions were found then these interaction effects were excluded from the final model.

Predicting Depressive Symptoms

In our first set of analyses, we examined our hypothesis that adolescents who possess a ruminative response style would report greater elevations in depressive symptoms following elevations in negative events than adolescents who do not possess a ruminative response style. When examining the random effects component of the model, the random effect for slope was not significant and was subsequently dropped from the model. Preliminary analyses indicated that the relationship between RUMINATION, FU_STRESS, and depressive symptoms during the follow-up interval was not moderated by either age (AGE × RUMINATION × FU_STRESS; β = −0.02, SE = 0.05, F(1, 255) = 0.12, ns) or gender (GENDER × RUMINATION × FU_STRESS; β = −0.05, SE = 0.04, F(1, 255) = 1.09, ns).

The final results with respect to the fixed-effects component of the model are presented in Table 4. Of primary importance, a significant two-way, cross-level interaction emerged between RUMINATION and FU_STRESS. In order to examine the form of this interaction, the model summarized in Table 4 was used to calculate predicted CES-D scores for adolescents possessing either low or high levels of RUMINATION (plus or minus 1.5 SD) and who are experiencing either low or high levels of stress in comparison to their own average level of stress (plus or minus 1.5 × mean within-subject SD). The results of such calculations are presented in Fig. 1. As FU_STRESS is a within-subject variable centered at each participant’s mean, slopes are interpreted as the increase in a adolescent’s CES-D score that would be expected given that he or she scored one point higher on the ALEQ.
Table 4

Rumination and Stress Predicting Depressive Symptoms during the Follow-up Interval

Predictor

β

SE

F

df

Time 1 CES-D

3.27

0.65

25.03***

1,147

AGE

0.01

0.51

0.01

1,147

GENDER

1.83

1.05

3.01

1,147

RUMINATION

1.39

0.53

6.78**

1,147

MN_STRESS

4.30

0.61

49.42***

1,147

FU_STRESS

0.24

0.02

111.54***

1,255

RUMINATION × FU_STRESS

0.05

0.02

4.57*

1,255

Time 1 CES-D Time 1 Center for Epidemiologic Studies Depression Scale, GENDER Coded variable (GENDER = 0 for males and 1 for females), RUMINATION Responses to Stress Scale, Rumination Subscale, MN_STRESS Average Level on Adolescent Life Events Questionnaire, FU_STRESS Fluctuations in Adolescent Life Events Questionnaire

*p < 0.05; **p < 0.01, ***p < 0.001

https://static-content.springer.com/image/art%3A10.1007%2Fs10802-008-9233-9/MediaObjects/10802_2008_9233_Fig1_HTML.gif
Fig. 1

Predicted slope of the relationship between negative life events and depressive symptoms for adolescents reporting high and low levels of rumination

Adolescents reported higher levels of depressive symptoms when reporting high STRESS scores than when reporting low STRESS scores regardless of whether they reported high rumination (slope = 0.31; t(255) = 7.71, p < 0.001) or low rumination (slope = 0.17; t(255) = 4.38, ns). Planned comparisons of the slopes of the relationships between FU_STRESS and depressive symptoms, however, revealed that the slope was significantly greater in adolescents reporting high RUMINATION scores than in adolescents reporting low RUMINATION scores (t(255) = 2.15, p < 0.05).

Predicting Level of Substance Misuse

In our second set of analyses, we examined our hypothesis that adolescents who possess a ruminative response style would report greater elevations in substance misuse scores following elevations in stress than adolescents who do not possess a ruminative response style. As with the previous analysis, the random effect for slope was not significant and was dropped from the model. Preliminary analysis indicated that the relationship between RUMINATION, FU_STRESS, and substance misuse during the follow-up interval was not moderated by gender (GENDER × RUMINATION × FU_STRESS; β = −0.05, SE = 0.05, F(1, 255) = 1.16, ns), however, the relationship was significantly moderated by age (AGE × RUMINATION × FU_STRESS; β = 0.05, SE = 0.02, F(1, 255) = 4.93, p < 0.05). Therefore, age interactions were retained when examining the fixed-effects component of the model.

The final results with respect to the fixed-effects component of the model are presented in Table 5. Of primary importance, a significant three-way, cross-level interaction emerged between AGE, RUMINATION, and FU_STRESS. In order to examine the form of this interaction, the model summarized in Table 5 was used to calculate predicted SUBMISUSE scores for younger (13.97 years) and older adolescents (16.39 years; plus or minus 1.5 SD), possessing either low or high levels of RUMINATION (plus or minus 1.5 SD), and who are experiencing either low or high levels of stress in comparison to their own average level of stress (plus or minus 1.5 × mean within-subject SD). The results of such calculations are presented in Fig. 2.
Table 5

Rumination and Stress Predicting Substance Misuse during the Follow-Up Interval

Predictor

β

SE

F

df

Time 1 SUBMISUSE

5.35

0.54

97.53***

1,147

AGE

1.22

0.53

5.33*

1,147

GENDER

−0.83

1.05

0.62

1,147

RUMINATION

−0.72

0.50

2.04

1,147

MN_STRESS

3.36

0.52

41.70***

1,147

FU_STRESS

0.07

0.03

6.55*

1,255

RUMINATION × FU_STRESS

0.05

0.03

2.94

1,255

AGE × FU_HASSLES

−0.01

0.03

0.04

1,255

AGE × RUMINATION

0.47

0.47

1.00

1,147

AGE × RUMINATION × FU_STRESS

0.05

0.02

4.93*

1,255

Time 1 CES-D Time 1 Substance Misuse Severity Measure, GENDER Coded variable (GENDER = 0 for males and 1 for females), RUMINATION Responses to Stress Scale, Rumination Subscale. MN_STRESS Average Level on Adolescent Life Events Questionnaire, FU_STRESS Fluctuations in Adolescent Life Events Questionnaire

*p < 0.05; ***p < 0.001

https://static-content.springer.com/image/art%3A10.1007%2Fs10802-008-9233-9/MediaObjects/10802_2008_9233_Fig2_HTML.gif
Fig. 2

Predicted slope of the relationship between negative life events and substance misuse for younger and older adolescents reporting high and low levels of rumination

Simple slope analyses were conducted for each AGE × RUMINATION condition examining whether the slope of the relationship between stress and substance misuse scores significantly differed from zero. Older adolescents who reported higher RUMINATION scores reported higher levels of substance misuse when reporting high STRESS scores than when reporting low STRESS scores (slope = 0.25; t(255) = 2.67, p < 0.01). At the same time, older adolescents who reported lower RUMINATION scores reported lower levels of substance misuse when reporting high STRESS scores than when reporting low STRESS scores (slope = −0.13; t(255) = −2.30, p < 0.05). Level of substance misuse did not vary as a function of level of stress for (1) younger adolescents with high RUMINATION scores (slope = 0.02; t(255) = 0.30, ns), or (2) younger adolescents with low RUMINATION scores (slope = 0.12; t(255) = 1.40, ns).

Mediation Analysis—Self Medication Model

We next conducted analysis to examine the hypothesis that fluctuations in depressive symptoms would mediates the relationship between fluctuations in negative events and substance misuse. Mediation was examined using the causal step approach proposed by Baron and Kenny (1986), which involves estimating a series of regression models to examine if three conditions for mediation are fulfilled. This approach to mediation analysis has recently been extended to models involving hierarchically nested data and moderated mediation hypotheses (Muller et al. 2005; Kenny et al. 2003). The first condition requires that there be a significant relationship between the predictor variable, stress, and the dependent variable, substance misuse. The second condition requires that there be a significant relationship between the predictor variable and the mediator variable, depressive symptoms. These conditions were examined in our first two sets of analyses. Finally, the third condition requires that there be a significant association between the mediator variable and the dependent variable. The association between the mediator and dependent variables should continue to be significant while controlling for the predictor variable. In addition, the magnitude of the association between the predictor variable and the dependent variable should decrease when the mediator variable is included in the model. Full mediation occurs when the association between the predictor and dependent variable is no longer significant when the mediator variable is included in the model.

To evaluate the third condition required for mediation we examined fluctuations in depressive symptoms (FU_DEPRESSION) as a predictor of fluctuations in levels of substance misuse. Additional fixed effects were included as control variables, specifically: initial levels of substance misuse, age, gender, and average levels of follow-up depression (MN_DEPRESSION). As depressive symptoms is a Level I predictor, a random effect for slope for depressive symptoms was included in the model. This random effect for slope was significant and was retained in the final model. Preliminary analysis indicated that the relationship between fluctuations in depressive symptoms and substance misuse was not moderated by either age (AGE × FU_DEPRESSION; β = 0.09, SE = 0.08, F(1, 255) = 1.31, ns) or gender (GENDER × FU_DEPRESSION; β = 0.01, SE = 0.16, F(1, 255) = 0.01, ns). Significant fixed effects components were obtained for the following variables: T1_SUBMISUSE (β = 6.11, SE = 0.54, F(1, 147) = 126.74, p < 0.001) and MN_DEPRESSION (β = 2.05, SE = 0.52, F(1, 147) = 15.78, p < 0.001). Of primary importance, fluctuations in depressive symptoms did not predict substance misuse scores so no further mediation analyses were conducted with depressive symptoms.

We conducted exploratory analysis to examine a possible reason for the lack of support for our self-medication model. Self-awareness models of substance consumption emphasize the reduction of negative mood rather than other symptoms such as somatic and interpersonal symptoms that are included in depression measures such as the CES-D. As such, we examined the relationship between substance misuse and the mood symptom items from the CES-D while excluding the interpersonal and somatic items. We examined fluctuations in negative mood (FU_NEGMOOD) as a predictor of fluctuations in levels of substance misuse. As with the previous analysis, control variables were included in this model. As fluctuations in negative mood symptoms is a Level I predictor, a random effect for slope for negative mood was included in the model. This random effect for slope was significant and was retained in the final model. Preliminary analysis indicated that the relationship between fluctuations in negative mood and substance misuse was not moderated by either age (AGE × FU_NEGMOOD; β = 0.06, SE = 0.11, F(1, 255) = 0.28, ns) or gender (GENDER × FU_NEGMOOD; β = −0.15, SE = 0.22, F(1, 255) = 0.49, ns). Significant fixed effects components were obtained for the following variables: T1_SUBMISUSE (β = 5.94, SE = 0.52, F(1, 147) = 131.78, p < 0.001), MN_NEGMOOD (β = 2.05, SE = 0.49, F(1, 147) = 17.46, p < 0.001), and FU_NEGMOOD (β = 0.37, SE = 0.11, F(1, 255) = 12.27, p < 0.001). Of primary importance, fluctuations in negative mood were a significant predictor of substance misuse scores.

We next entered negative mood (MN_NEGMOOD and FU_NEGMOOD) into the model predicting levels of substance misuse from the interaction between RUMINATION, AGE, and FU_STRESS summarized in Table 4. Fluctuations in negative mood continued to be a significant predictor of substance misuse while controlling for the three way interaction (FU_NEGMOOD: β = 0.32, SE = 0.12, F(1, 255) = 7.16, p < 0.01). At the same time, the three-way cross level interaction between AGE, RUMINATION, and FU_STRESS was no longer a significant predictor of substance use problems after negative mood was entered into the model (β = 0.04, SE = 0.03, F(1, 255) = 3.06, ns). These results provide support for fluctuations in negative mood symptoms as a mediator of the relationship between fluctuations in negative events and substance misuse among older adolescents reporting a tendency to ruminate in response to stress.

Alternate Mediation Analysis—Secondary Depression Model

Similar analyses were conducted to examine fluctuations in substance misuse as a mediator of the relationship between fluctuations in negative events and depressive symptoms. Analyses were similar to those described above with the exception of (a) our dependent variable being depressive symptoms during the follow-up interval, and (b) our mediator variable being fluctuations in substance misuse severity scores during the follow-up interval. We examined fluctuations in substance misuse (FU_SUBMISUSE) as a predictor of fluctuations in levels of depressive symptoms. Additional fixed effects were included as control variables, specifically: initial levels of depressive symptoms, age, gender, and average levels of follow-up substance misuse (MN_SUBMISUSE). As substance misuse is a Level I predictor, a random effect for slope for substance misuse was included in the model. This random effect for slope was significant and was retained in the final model. Preliminary analysis indicated that the relationship between fluctuations in substance misuse and depressive symptoms was not moderated by either age (AGE × FU_SUBMISUSE; β = 0.04, SE = 0.08, F(1, 255) = 0.23, ns) or gender (GENDER × FU_SUBMISUSE; β = −0.02, SE = 0.17, F(1, 255) = 0.02, ns). Significant fixed effects components were obtained for the following variables: T1_DEPRESSION (β = 6.33, SE = 0.60, F(1, 147) = 111.45, p < 0.001), GENDER (β = 2.48, SE = 1.19, F(1, 147) = 4.32, p < 0.05), MN_SUBMIUSE (β = 1.37, SE = 0.62, F(1, 147) = 4.93, p < 0.05), and FU_SUBMISUSE (β = 0.22, SE = 0.09, F(1, 255) = 7.14, p < 0.01). Of primary importance, fluctuations in substance misuse were a significant predictor of depressive symptoms.

We next entered substance misuse (MN_SUBMISUSE and FU_SUBMIUSE) into the model predicting levels of depressive symptoms from the interaction between RUMINATION and FU_STRESS summarized in Table 3. Fluctuations in substance misuse were not a significant predictor of depressive symptoms after controlling for the two-way, cross-level interaction between rumination and fluctuations in stress (FU_SUBMISUSE, β = 0.10, SE = 0.08, F(1, 255) = 1.83, ns). Overall, these analyses did not provide support for the alternative mediation model.5

Discussion

The results of the current study provide support for the hypothesis that a tendency to ruminate in response to stress serves as a vulnerability factor for both depressive symptoms and substance use problems in adolescents. Specifically, the association between negative life events and both depressive symptoms and substance misuse varied as a function of adolescent’s level of rumination. Neither of these predictive relationship varied as a function of gender, however, female adolescents did report a greater tendency to ruminate in response to stress than males. This is line with previous studies that have found a significant gender difference in levels of rumination among adolescents (Abela et al. 2008; Schwartz and Koenig 1996). Support was not obtained for fluctuations in general depressive symptoms as a mediator of the relationship negative events and substance misuse. Fluctuations in negative mood, however, were found to mediate this relationship. Several findings warrant additional attention.

Rumination and Depression

Both adolescents possessing low and high levels of rumination reported elevations in depressive symptoms in response to elevations in negative events; however, the magnitude of elevations was significantly greater among adolescents who reported high levels of rumination. In addition, these groups differed markedly in their severity levels of depressive symptoms. When adolescents possessing high levels of rumination experienced low levels of negative events they reported CES-D score similar to the mean score typically reported by youth in their age range (Rushton et al. 2002). When these youth experienced high levels of negative events they reported elevations in depressive symptoms such that their predicted CES-D scores were in the mild/moderate severity level of depressive symptoms (Roberts et al. 1991). In contrast, adolescents possessing low levels of rumination reported very few depressive symptoms when experiencing low levels of negative events. When experiencing high levels of negative events these youth reported elevations in depressive symptoms, however, their predicted CES-D scores during such periods were still below the mean scores typically reported by youth in their age range (Rushton et al. 2002) and were in the minimal severity level of depressive symptoms (Roberts et al. 1991). Although all individuals respond to stressful experiences with increases in negative affect, theorists have posited that differences in responses to stress determine whether these experiences lead to significant depressive symptoms (Compas et al. 1993). These results indicate that a tendency to ruminate in response to stress is associated with both greater stress reactivity and higher levels of depressive symptoms.

The current study extends previous research by providing evidence that the interaction between rumination and negative events prospectively predicts elevations in depressive symptoms. Although research conducted with adult and adolescent samples has indicated that rumination is associated with higher levels of depressive symptoms over time (Abela and Hankin 2007; Lyubomirsky and Tkach 2003), previous studies have not found that the interaction between rumination and negative events predicts depression (Abela et al. 2008; Sarin et al. 2005; Schwartz and Koenig 1996). This discrepancy may be attributable to differences in the methodology employed in the current study. First, previous prospective studies examining the interaction between rumination and negative events have utilized a two-time point design whereas the current study employed a multi-wave longitudinal design. By obtaining repeated assessments of levels of negative life events, depressive symptoms, and substance misuse within individuals over an extended period of time, we are able to gather a relatively reliable estimate of each adolescent’s degree of stress reactivity (e.g., his or her slope of the relationship between negative events and depressive symptoms). In addition, by using this approach high levels of negative events can be operationalized in reference to each youth’s own mean level of negative life events rather than the sample’s mean level of negative life events. Such an approach towards operationalizing stress is likely to minimize the impact of individual differences in the reporting of negative life events on findings. Secondly, previous prospective studies have measured rumination as a response to depressed mood whereas the current study specifically measured rumination as a response to stress. Theorists have posited that rumination is a multifaceted construct with rumination in response to stress and rumination in response to depression operating at distinct points in the cycle of depressogenic thinking. Specifically, the interaction between negative events and rumination in response to stress is proposed to be related to the onset of depressive symptoms, whereas depressive rumination is proposed to be related to the maintenance of depression (Robinson and Alloy 2003; Sakamoto 2000). The results of the current study indicate that which type of rumination is measured may have important empirical consequences when examining the interaction between negative events and rumination.

Rumination and Substance Misuse

Partial support was obtained for rumination as a vulnerability factor to substance misuse. Specifically, older adolescents with a tendency to ruminate in response to stress reported elevations in substance misuse following elevations in negative life events. This finding is in line with hypothesis and extends previous research (Nolen-Hoeksema et al. 2007; Nolen-Hoeksema and Harrell 2002) by providing support for a diathesis-stress conceptualization of rumination as vulnerability factor to substance misuse. At the same time, older adolescents who reported low rumination scores evidenced lower levels of substance misuse following elevations in negative life events. This is an unanticipated finding that requires replication before being interpreted further. At the same time, one possible explanation is that adolescents who are able to reduce their substance use during periods of heightened stress posses more adaptive coping skills than adolescents who increase their substance use in response to stress. In support of this interpretation, researchers have found that adolescents who employ more problem-focused coping responses report lower levels of substance use problems (Windle and Windle 1996) and adaptive coping can buffer the deleterious impact of heightened stress on adolescent substance use (Wills et al. 2001). Differences in coping styles may be important as rumination has been found to interfere with the employment of adaptive coping responses to stress such as distraction and problem-solving (Lyubomirsky and Tkach 2003). Recent research indicates that the relationship between rumination and poor coping skills may be unique to specific subtypes of rumination such as brooding rumination (Burwell and Shirk 2007). Older adolescents who do not tend to ruminate may be able to reduce their substance misuse during period of high stress because they are able employ more adaptive coping behaviors to deal with their difficulties. On the other, hand older adolescents who tend to ruminate may be placed in a double bind because they both experience higher levels of depressive symptoms following elevations in stress and have difficult employing more adaptive coping behaviors. Future research is needed to explore this interpretation of the findings by examining the longitudinal relationship between negative events, different subtypes of rumination, and specific coping responses in predicting substance misuse.

Contrary to hypotheses, younger adolescents’ levels of substance misuse did not vary as a function of fluctuations in negative events irrespective of differences in the tendency to ruminate in response to stress. Lack of support for rumination as a vulnerability factor to substance misuse among younger adolescents may be attributable to developmental differences in the propensity to employ substance use as a strategy for coping with distress. Prevalence rates and levels of substance use increase steadily throughout adolescence (Johnston et al. 2006). As a result, older adolescents have more experience with and exposure to substance use and they may be more likely than younger adolescents to have incorporated substance into their repertoire of coping behaviors. In line with this interpretation, previous studies have indicated that older adolescents endorse higher levels of coping motives for substance use than younger adolescents (Cooper 1994; Kuntsche et al. 2006). Further, the magnitude of the association between coping motives for drinking and alcohol misuse is greater among older adolescents than among younger adolescents (Bradizza et al. 1999). Future research should examine differences in coping motives for substance use as a mediator of age-related differences in the relationship between stress and substance misuse.

Relationship Between Depression and Substance Misuse

Research has consistently indicated that there is a significant concurrent association between levels of depression and substance use among adolescents (e.g. Brook et al. 1998; Chinet et al. 2006; Henry et al. 1993). This is in line with the results of the current study which found a positive correlation between initial levels of depressive symptoms and substance misuse. On the other hand, elevations in depressive symptoms during the follow-up period did not significantly predict elevations in levels of substance misuse. Elevations in follow-up levels of substance misuse did predict elevations in depressive symptoms. This relationship, however, was no longer significant when the interaction between rumination and negative events was controlled for. Previous studies examining the prospective reciprocal relationships between depression and substance use problems in adolescents have provided a mixed pattern of results. Several studies have provided support for depressive symptoms as a prospective risk factor for the development of substance use problems (e.g. Costello et al. 1999; Newcomb et al. 1986; Henry et al. 1993), whereas other studies have indicated that substance use problems in adolescents predict the development of subsequent depressive disorders (Brook et al. 1998; Hansell and Raskin White 1991). At the same time, several studies that have found no prospective relationship, in either direction, between depression and substance use problems (Chinet et al. 2006; McGee et al. 2000). All of these prospective studies have examined predictive relationships between depression and substance use problems over a longer follow-up period (i.e. years) than the current study. The results of the current study provide support for substance use problems as a predictor of depressive symptoms, although, this relationship may only be a result of the fact that both outcomes are associated with fluctuations in stress and further replication is needed.

Analysis did not support our self-medication hypothesis that fluctuations in depressive symptoms would mediate the relationship between fluctuations in negative events and substance misuse. However, exploratory analyses indicated that fluctuations in negative mood symptoms of depression mediated the relationship between negative events and substance misuse among adolescents who tended to ruminate in response to stress. These results are consistent with self-focused attention models of substance consumption (Hull 1981; Ingram 1990) that emphasize substance use as a means of ameliorating negative mood states rather than general depressive symptoms. Similarly, self-medication models of substance use postulate that the desire to escape from negative emotional states provides a particular powerful, maladaptive motivation for substance consumption (Cooper et al. 1995). Substance use may provide temporary relief from depressive mood but not have an impact upon other components of depression such as somatic and interpersonal symptoms. There have been few studies examining the relationship between substance misuse and specific symptoms of depression and future research is needed to replicate the results of the current study.

General Comments

Several limitations of the current study should be noted. First, depressive symptoms and substance use problems were both assessed using self-report measures. Although the CES-D and the SMSM both possess adequate reliability and validity, it is difficult to make conclusion regarding clinically significant disorders based upon self-report questionnaires. Future research should employ additional assessment techniques, such as clinical interviews, to examine if the current results generalize to the onset of clinically diagnosed depressive and substance use disorders. Second, self-report measures were also used to measure the occurrence of stress. Although measures of life events that require participants only to indicate whether or not an event occurred are probably less likely to be influenced by informant bias than those that ask subjects to rate the subjective impact of each event, more sophisticated methods of analysis such as interviewing procedures may provide more comprehensive assessments of stress (Grant et al. 2003). In addition, the reliance on self-report measures in the current study may have affected results because of common method variance between the measures. This is an additional reason for future research to incorporate supplementary assessment methods such as interviews and parental or peer reports. Third, participants in the current study were primarily Caucasian adolescents drawn from the high school population. Future research is needed to determine if the current results generalize to youth from other ethnic backgrounds and populations such as high-school dropouts and clinically referred youth. Fourth, future studies should utilize larger sample sizes. Such an approach would allow more powerful evaluation of complex interactive models of the relationship between rumination, depressive symptoms, and substance misuse in adolescents. Fifth, it is important to note that the effect size for the interaction between rumination and negative events in predicting depression and substance use problems was in the small to medium range. Although effects sizes of this magnitude are common in the social sciences (Rosenthal 1984), they also indicate that factors other than rumination are important in the etiology of depression and substance use disorders. Future studies are needed to examine the interaction between rumination and these other vulnerability factors in predicting depressive symptoms and substance misuse in adolescents. Last, the current study only examined the relationship between rumination, stress, and adolescent depressive symptoms and substance misuse. Thus, we were unable to identify whether the interaction of rumination with negative events is specific to depressive symptoms and substance misuse rather than broadly applicable to other disorders (anxiety disorders, conduct disorder, etc.).

In conclusion, the current study provides support for a common factor model of the relationship between depression and substance misuse in adolescence. Specifically, the results provided support for a diathesis-stress model of rumination as a common vulnerability factor to the development of both depression and substance use problems. Rumination was found to interact with the occurrence of negative events to predict elevations in both depressive symptoms and substance misuse. In addition to advancing research into adolescent depression and substance use disorders the current results have important clinical implications. The identification of common vulnerability factors is of importance for clinicians seeking mechanisms to target when treating comorbid depressive and substance use disorders (Conrod and Stewart 2005).

Footnotes
1

In the current manuscript the terms substance use problems and substance misuse will be used interchangeably to refer to both clinical symptoms of DSM-IV substance use disorders and subthreshold symptoms that arise as a consequences of hazardous substance use. Research indicates that a large proportion of adolescents experience negative consequences as a result of excessive substance use that do not meet criteria for substance abuse or dependence, but nonetheless cause significant distress and impairment (Lewinsohn et al. 2004; Zoccolillo et al. 1999).

 
2

Comparison demographic data was obtained by from census data obtained by Statistics Canada for the greater Montreal area. Retrieved December 16th, 2007 from http://www12.statcan.ca/english/Profil01/CP01/Details/Page.cfm?Lang=E&Geo1=CMA&Code1=462__&Geo2=PR&Code2=24&Data=Count&SearchText=montreal&SearchType=Begins&SearchPR=01&B1=All&Custom=.

 
3

The consumption and negative consequences subscale of the SMSM were examined separately in supplementary analyses. In all cases, similar patterns of significant findings were obtained as compared to those obtained when the composite substance misuse severity measure was employed as the dependent variable. For the sake of parsimony and to provide a comprehensive assessment of a spectrum of substance use problems, results are reported for the composite measure. Additional details regarding these analyses are available by contacting the lead author.

 
4

We also conducted supplementary analyses to account for the possible correlation in response variables between adolescents from the same school. Specifically, we specified models which included the following random effects components: random effects for adolescents (RE_PARTICIPANT; random intercept) nested within schools (RE_SCHOOL; random intercept), a random effect for slope (RE_SLOPE), and an autoregressive covariance parameter. In all analyses, the RE_SCHOOL parameter was not significant. In addition, inclusion of the RE_SCHOOL parameter did not change the pattern of significant findings in any analysis. For the sake of parsimony models are presented without the inclusion of RE_SCHOOL.

 
5

Exploratory analyses were conducted to examine fluctuations in substance misuse as a mediator of the relationship between fluctuations in negative events and negative mood symptoms. Of primary importance, results were identical to those obtained when examining substance misuse as a mediator of the relationship between fluctuations in negative events and total depressive symptoms. Additional details regarding these analyses are available by contacting the lead author.

 

Copyright information

© Springer Science+Business Media, LLC 2008