Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
J Cogn Behav Psychother. Author manuscript; available in PMC 2010 May 24.
Published in final edited form as:
J Cogn Behav Psychother. 2008 September 1; 8(2): 147–168.
PMCID: PMC2874909

Clusters of Behaviors and Beliefs Predicting Adolescent Depression: Implications for Prevention



Risk factors for various disorders are known to cluster. However, the factor structure for behaviors and beliefs predicting depressive disorder in adolescents is not known. Knowledge of this structure can facilitate prevention planning.


We used the National Longitudinal Study of Adolescent Health (AddHealth) data set to conduct an exploratory factor analysis to identify clusters of behaviors/experiences predicting the onset of major depressive disorder (MDD) at 1-year follow-up (N=4,791).


Four factors were identified: family/interpersonal relations, self-emancipation, avoidant problem solving/low self-worth, and religious activity. Strong family/interpersonal relations were the most significantly protective against depression at one year follow-up. Avoidant problem solving/low self-worth was not predictive of MDD on its own, but significantly amplified the risks associated with delinquency.


Depression prevention interventions should consider giving family relationships a more central role in their efforts. Programs teaching problem solving skills may be most appropriate for reducing MDD risk in delinquent youth.

Keywords: adolescence, depression, prediction, prevention

Adolescence is a critical period for the onset of major depressive disorder (MDD), with as many as 24% of adolescents experiencing an episode by age 24 (Kessler & Walters, 1998; Klerman, 1988; Klerman & Weissman, 1989). It is also an important developmental period during which critical processes of socialization and educational development occur. Adolescents who experience a depressive episode during this period often have reduced educational attainment, greater relationship dysfunction, more job absenteeism, increased risk of substance abuse and tobacco use, and increased risk for MDD recurrence within 5 years (Binder & Angst, 1981; Breslau, Kilbey, & Andreski, 1991; Breslau, Kilbey, & Andreski, 1994; Christie, Burke, Regier, Rae, & et al., 1988; Ernst, Foldenyi, & Angst, 1993; Hallowell, Bemporad, & Ratey, 1989; Horwitz & White, 1991; Kessler, Avenevoli, & Ries Merikangas, 2001; Kessler & Walters, 1998; Reinherz, Giaconia, Hauf, Wasserman, & Silverman, 1999; Runeson, 1989; Skodol, Schwartz, Dohrenwend, Levav, & Shrout, 1994). Furthermore, adolescent depression is a major contributor to suicide and is the third leading cause of death among older adolescents (Fombonne, Wostear, Cooper, Harrington, & Rutter, 2001b; Harris & Ammerman, 1986). Despite the public health impact of depressive disorders, little is known about how characteristics, experiences and behaviors related to onset of disorder may naturally cluster in community settings.

Multiple vulnerability characteristics and behaviors are associated with the onset of adolescent depression. This includes genetic (short allele of the serotonin transporter gene promoter region), personality (neuroticism), biological stress response (hypo-pituitary-adrenal axis changes), problem solving/attribution (negative inferential styles, dysfunctional attitudes, rumination, self-criticism), and family/interpersonal relations (low social support from peers and family) (Hankin, 2006; Reinecke, 2005). These characteristics and behaviors interact with adverse events to increase the risk of a depressive episode. Prevention interventions targeting primarily the problem solving/attribution domain have produced variable results with regard to efficacy, and these benefits often attenuate after six months (Merry, McDowell, Hetrick, Bir, & Muller, 2004). Inconsistent results from prevention studies suggest that our understanding of the organization of behaviors and characteristics relevant to the onset of depression in adolescence is insufficient.

Several reports advocate research on the development and evaluation of robust, practical public health strategies to reduce the burden of depressive disorders in youth (Bramesfeld, Platt, & Schwartz, 2006; Saxena, Jane-Llopis, & Hosman, 2006). The Institute of Medicine Report “Prevention of Mental Disorders” asserts that a sound understanding of vulnerability and protective factors and their relationship to one another is essential for developing and fielding efficacious interventions (Munoz, Mrazek, & Haggerty, 1996). Furthermore, a recent review by Garber suggested that as multiple risk factors accumulate and interact with one another, they should be targeted simultaneously (Garber, 2006). An empirically-based model for organizing a large number of vulnerability characteristics and behaviors into broader schema or factors would be very useful for developing appropriately targeted depression-prevention interventions.

To address this need for an organizational schema, we sought to develop a factor model of vulnerability and protective factors within a representative, community-based sample of US adolescents. We selected the National Longitudinal Study of Adolescent Health because it is the only current, prospective, representative sample of US adolescents for which a broad array of vulnerability behaviors are available at baseline and depression outcome at follow-up. We conducted an exploratory factor analysis of all variables representing depression risk-factors identified in previous empirical and theoretical models and subsequently developed a logistic regression model using these factors to predict the new onset of MDD.


Survey Design and Data Collection

The National Longitudinal Study of Adolescent Health was a representative sample of US adolescents in grades 7–12. A baseline survey (wave 1) was conducted in 1995, with follow-up (wave 2) completed in 1996 (Resnick et al., 1997). The survey systematically identified schools to represent urban, regional, and ethnic strata. Seventeen adolescents were chosen randomly from each age-gender group within each of the 200 schools. There was an additional over-sampling of African American youth with highly-educated parents. The overall response rate for the wave 1 survey was 76.8% with a total sample of N=6,504 (public-use data set, Sociometrics, Inc) (National Longitudinal Study of Adolescent Health (Add Health), Wave I and II, 1994–1996, 1998). Interviewers conducted face-to-face interviews with adolescents and their parents in home (wave 1 and 2) and in school (wave 1 only, not completed by all participants). For wave 2, nearly three quarters (73.5%) of the wave 1 participants completed the survey (N=4,834). Previous work has described this survey’s methods (Resnick et al., 1997). The University of Chicago Institutional Review Board approved this secondary data analysis.

Independent Variables

Two researchers experienced with adolescent and adult depression (B.W.V.V., & J.G.) reviewed the 5,800 variables included in the Sociometrics public use dataset to identify those believed to confer protective and vulnerability effects based on theoretical models developed by Reinecke (Reinecke, 2005) and Hankin (Hankin, 2006). We chose these models because they reflected a consensus understanding of the major categories of proximal vulnerability factors. We considered models and perspectives from a developmental psychopathology model with a more longitudinal understanding of risk (Cicchetti & Toth, 1998; Compas, Hinden, & Gerhardt, 1995). Discrete categories of proximal vulnerability factors were identified in support of our aims to predict disorder within the next year. Each investigator independently reviewed the candidate variables that were developed based on the face validity of the items in association with the theoretical models and previous research. The final list of potential protective and vulnerability factors was identified and compiled by consensus of both investigators. The variables used in the exploratory factor analysis are broken up into logical categories and listed in Table 1.

Table 1
Items Considered for Factor Models


Variables used to adjust logistic regression results included sociodemographics (age, gender, race/ethnicity), socioeconomic status (household income), and baseline depressive symptoms [Center for Epidemiologic Studies Depression (CES-D) Scale (20 items, score 0–60)] (Radloff, 1977,, 1991). Parental education level was assessed for mother and father, distinguishing between those with no high school degree and those with high school and college degrees.

Outcome Variable

Major depressive disorder, the primary outcome variable, was developed from depression-specific items from the Center for Epidemiologic Studies Depression (CES-D) Scale. We constructed the MDD outcome variable using a method developed by Schoenbach (V. J. Schoenbach, Kaplan, Grimson, & Wagner, 1982) and validated by Radloff (Radloff, 1991) in school and college samples.

MDD was defined as having at least one core symptom (depressed mood or anhedonia) and three additional symptoms from the MDD description in the Diagnostic and Statistical Manual of Mental Disorders (DSM)-III. We chose a cut-off of four symptoms versus five in the DSM-III because the ADD Health study employed a shortened modified version of the CES-D that omitted two items of potential importance, “I had crying spells” and “I felt just as good as other people” (Rushton, Forcier, & Schectman, 2002). We assessed the validity of this outcome variable by comparing the mean CES-D score of those identified as having MDD and the percentage represented by the study population with those found in other samples. We chose to employ this method to construct the outcome variable rather than using standard CES-D cutoff scores because of the latter’s high sensitivity (86–100%), but moderate specificity (53%–84%), areas of potential concern with regard to the high mood liability seen among adolescents (Attkisson & Zich, 1990; Haringsma, Engels, Beekman, & Spinhoven, 2004; Houston et al., 2001; Lin & Parikh, 1999; Radloff, 1977,, 1991; Smit et al., 2006). We are aware that there are formal differences for MDD classification between DSM-III, DSM-IIIR, and DSM-IV. However, we are unaware of any validated method for converting the popular CES-D scale into DSM-IV classification. MDD classification based upon DSM III is still considered important and is extensively reviewed in one of the leading textbooks on psychiatric epidemiology (Horwath, Cohen, & Weissman, 2002).

Exploratory Factor Analysis

The principal factor (pf) command in the Stata/SE 9.0 (StataCorp, 2005) statistics package was used. As missing the value of a single variable eliminates all the variables from a study participant from the analysis, variables with data missing from many participants were eliminated. Starting with the original set of 71 variables and 6,504 observations in the first wave, all variables containing fewer than 5,000 observations were eliminated. This left 54 of the original 71 candidate variables, and 4,010 observations remained if all observations missing any variables of the wave 2 outcome measures were removed. The 17 dropped variables consisted of variables asked in the course of the in-school questionnaire, which had a lower response rate. Those variables included questions on mood, family function, and school attitudes. The excluded mood and family function variables did not meaningful differ from those that were included. The school attitudes questions, such as feeling “like a part of the school”, “happy at school”, “like teachers are fair”, etc. could unfortunately not be included in this analysis because of low response rates.

Principal factor analysis was performed with varimax rotation, and the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was conducted. KMO suggests if data will factor appropriately and is calculated using correlation for its numerator and partial correlation for its denominator. It has values ranging from 0 to 1. The overall KMO should be at least 0.5 in order to continue with the factor analysis. Individual items with low KMO values should be dropped and then the overall KMO recalculated to determine if there is now an appropriate KMO. In our sample, four variables had individual KMO below 0.5 and were dropped from the analysis. A principal factor analysis and varimax rotation was then performed on the 50 remaining variables, and a satisfactory overall KMO (0.85) was obtained. At this stage, 50 of the original 71 candidate variables and 4,021 observations remained; using fewer variables led to the inclusion of 11 more observations because the four dropped variables were missing observations. We then chose variables with factor loadings above 0.3 to maximize internal consistency and standardized them to create a z-score with mean=0 and standard deviation=1 to compensate for differences in variable measurement (i.e., some variables were binary while other variables were 5 or 10 point Likert scales), and summed them to create the total factor scores.

Logistic Regression Factor Prediction Model

We evaluated the relationships between the factors and MDD at wave 2 using logistic regression models. We excluded 113 participants who had MDD at baseline because we were interested in the new onset of depression. First, each factor was analyzed in a univariate logistic regression model predicting MDD at wave 2. We then conducted two other adjusted logistic regression models. One included the factor and adjusted for age, gender, race/ethnicity, and household income. The other included the factor, adjusted for the sociodemographic and socioeconomic covariates above and now also adjusted for baseline depressive symptoms. Next, we included all four factors simultaneously into one multivariate logistic regression model. We then repeated the other two adjusted models in addition to the four factors, and adjusted for the covariates mentioned above. In all regression models, we adjusted for the complex survey design using the Stata “svy: logisitic” command and the recommended samping weights and strata to make our sample representative of the US adolescent population.

Each model was tested for Hosmer-Lemeshow goodness-of-fit (Hosmer & Lemeshow, 1989). Finally, we added dummy terms to the multivariate logistic regression model to test for multiplicative interactions between factors and also performed post hoc exploratory Sobel-Goodman mediation tests to test for the presence of mediation.


Study Sample

This sample was 48% male, 58% white, 23% African American, 12% Hispanic, 1% American Indian, 3% Asian, and 5% multiracial (over sampled for racial/ethnic minorites). Mean age was 16.1 years (SD=1.8 years). Mean annual household income was $46,735 (SD= $47,529). In terms of parental education, 11.3% had less than a high school degree, 50.9% had a high school degree, and 37.9% were college graduates or higher. Participants who did not respond to the follow-up survey were more likely to be African American (p<0.001), more likely to be older (p<0.001), and less likely to have a parent with a college degree (p<0.001). Those who did not follow-up were not more likely to have MDD at baseline or greater levels of depressive symptoms than those who did follow up.

Validation of the Outcome Variable

In this sample the point prevelance of MDD was 2.49% in the baseline survey and 2.80% in the follow-up. This point prevelance is similar to the 2.90% obtained by Schoenbach in his original study where he compared the construction of a major depression variable from the CES-D to actual structured psychiatric interviews (V. J. Schoenbach et al., 1982). Baseline prevalence was significantly higher in persons over 18 (prevalence=2.30% for ages 13–17, and 3.51% for age>18, p=0.028). Mean CES-D scores for persons with MDD were comparable to the mean score of psychiatric MDD inpatients (M=38.1, SD=9.0) (Radloff, 1991) in both the baseline sample (M=34.0, SD=7.1) and follow-up sample (M=32.6, SD=7.1).

Exploratory Factor Analysis

The initial principal factor model retained 21 factors; five of them had eigenvalues greater than one. Using Kaiser’s criterion, these five factors were retained, and the resulting five-factor model was varimax rotated (Sherin, 1966). The fifth factor had only three component variables and a Cronbach alpha value of 0.37; it was dropped because of its low explanatory power and poor cohesion. The remaining four factors had Cronbach alpha values between 0.69 and 0.83. Each factor’s component variables, eigenvalue, and Cronbach alpha are shown in Table 2. The overall KMO of the model was 0.85 and the eigenvalues of its factors ranged from 1.23 – 5.13 before rotation. Each of the factors is described below, and the specific question wording can be found in Table 2.

Table 2
Factor Composition

Family/Interpersonal Relations

This factor included 9 items and represents closeness to parents, family function, and general feelings of social acceptance. Component variables included questions about how much the respondent feels that his or her family “pays attention to [respondent]”, “understands [respondent],” and “has fun together.” It also includes responses to questions about “how close” the respondent feels to the residential mother and how much the respondent feels that the residential mother and adults in general “care about” him or her; observations with no residential mother were dropped from the analysis (6% of sample). Other variables included the respondents’ desires to leave home, the degree to which the respondents feel socially accepted, as well as the extent to which they responded positive to the statement: “I like myself just the way I am”. Higher scores on these factors reflect more family/interpersonal relations, connectedness, and acceptance.


This factor included 9 items and represents the development of relationships, behaviors, and actions outside of (and in the case of the conduct-related items, directly opposed to) established authorities such as the adolescent’s family, religion, or school. Because many variables for this factor were set up as binary flags (0=”no,” 1=”yes”), a higher score on this scale represents greater independence or conflict with established authority in the adolescent’s life. Component variables included the frequency of running away from home over the previous year, the frequency of “just hanging out,” “drinking alcohol,” lying to parents or guardians, smoking marijuana, having had a “special romantic relationship,” having “had sexual intercourse,” and the number of different kinds of violent and non-violent crimes committed over the previous year. Higher scores on this factor reflect higher self-emancipation.

Avoidant Problem Solving/Low Self-Worth

This factor included 5 items and represents problem solving orientation, defined as avoidant or approach-oriented, and self-worth. Approach-oriented problem solving was defined by positive responses to questions concerning the degree to which the respondent “analyzes what went right and what went wrong” after a solution to a problem is attempted, uses a “systematic method for judging and comparing alternatives” to problems, tries to “get as many facts about the problem as possible,” and “tries to think of as many different ways to approach the problem.” Self-worth was assessed with responses to the question “I have a lot to be proud of.” Higher scores on this factor reflect more avoidant problem solving style and lower self-worth.

Low Religious Activity Factor

This factor included 3 items and represents the degree of religious participation. Component variables included how often the respondent prays, attends religious services, and participates in a religious youth group. Due to the format of the individual religious-activity questions, lower scores on this factor reflect more frequent religious participation.

Logistic Regression Models

Univariate Models

We summarize the results from the unadjusted and adjusted bivariate logistic regression models relating each factor to MDD at follow-up in Table 3. Low family/interpersonal relations functioning, high self-emancipation, and low religious activity (reflected by higher scores on the low religious activity factor) predicted greater risk for MDD at follow-up. These significant relationships were maintained in the model adjusting for the sociodemographic/socioeconomic variables. Also, in this model, avoidant problem solving/low self-worth significantly predicted MDD as well. In the model adjusting for the sociodemographic/socioeconomic variables and baseline depressive symptoms, only low family/interpersonal relations and low religious activity were significantly related to MDD at follow up. The self-emancipation factor was no longer significant.

Table 3
Univariate Factor Model Analyses

Multivariate Models

The four factors were combined in a single unadjusted multivariate model, as summarized in Table 4. Low family/interpersonal relations significantly predicted MDD in all of the multivariate models, even in the two additional models adjusting for the sociodemographic/socioeconomic variables and baseline depressive symptoms. None of the other factors were significant in any of the models. Hosmer-Lemeshow goodness-of-fit tests for all models were satisfactory.

Table 4
Multivariate Factor Model Analyses

Additional Analyses

Multiplicative interactions were tested with dummy variables created by multiplying every possible pair of variables and testing their significance in a predictive model including the four original factors. Dummy variables were used for ease of interpretation. The only significant interaction was between the avoidant problem solving/low self-worth and self-emancipation factors; Figure 1 shows the association of avoidant problem solving/low self-worth (categorized by “intermediate,” or middle 50%, and “proactive” and “avoidant,” or upper and lower quartile, respectively) on MDD incidence by the self-emancipation factor score (also broken down by middle 50% score (med) and upper (high) and lower (low) quartiles).

Figure 1
Incidence of Major Depression at Follow-up by Cognitive Style, Stratified by Delinquency.

We conducted Sobel-Goodman tests of mediation to determine if the associations of any of the factors with MDD at follow-up were expressed indirectly through the family/interpersonal relations factor. We limited our analyses to those mediated through the family/interpersonal relations factor since it was the only variable significant in the multivariate model. The results of these tests suggest that the relationships of self-emancipation and religious activity to MDD are largely expressed (61.5% and 54.6% respectively) through the influence of these factors on the family/interpersonal relations factor, rather than directly on MDD.

Evaluation of Multiple Statistical Tests

There are different opinions with regard to many multiple comparisons and the potential for increasing the Type I error (identifying false positive results). The Bonferroni p-value adjustment method is a strict way to adjust the p-value for significance level due to the multiple comparisons. The formula is alpha level divided by the number of comparisons. The Bonferroni adjusted p-value was calculated using two different approaches. The more strict approach considered each of the indicator values for race/ethnicity as separate comparisons. The more liberal approach considered race/ethnicity as one comparison.

In our analyses we included the variables of sex, age, race/ethnicity (6 indicator variables), income, the 4 factors, and depressive symptoms. Below we calculate the Bonferroni adjusted p-values for the multivariate analyses shown in Table 4. Also, in general for “survey adjusted data,” this approach is more rigorous and usually results in higher p-values than analyses that do not take into account the survey sampling. The model that included everything of “survey, sociodemographics/socioeconomics, and baseline depressive symptoms adjusted” has either 14 comparisons (strict) or 9 comparisons (liberal). This results in a Bonferroni adjusted p-value of either 0.05/14= 0.003571 (strict) or 0.05/9=0.005556 (liberal). Family/Interpersonal relations remain significant even with these new lower p-value levels as we obtained values of <0.001, <0.001, and 0.003. This Bonferroni adjusted p-value approach indicates that our findings for family/interpersonal relations are not likely to be a chance outcome that was only significant due to multiple comparisons.


We consolidated variables from multiple domains of depression risk into family/interpersonal relations, self-emancipation, avoidant problem solving/low self-worth, and low religious activity factors. The strongest factor identified in relation to MDD was the family/interpersonal relations factor; it persisted though all models and multivariate adjustment for the other factors. Our findings suggest that risk and protection are derived from both family and community context, as demonstrated in cross-sectional surveys (Youngblade et al., 2007).

Family/interpersonal relations was highly protective in this study, a finding consistent with previous work (Goodman et al., 1997; Lewinsohn et al., 1994; Muris, Schmidt, Lambrichs, & Meesters, 2001). Sheeber found that family support and conflict predicted adolescent depression at 1-year follow-up when controlling for baseline depression (L. Sheeber, Hops, Alpert, Davis, & Andrews, 1997). Others have found that greater levels of parent conflict and lack of closeness to either parent is associated with both major depression and sub-threshold depressive symptoms in adolescents (L. B. Sheeber, Davis, Leve, Hops, & Tildesley, 2007). Depression was also far more common among community adolescents who described their parents as unaffectionate and controlling on the Parental Bonding Instrument (Martin & Waite, 1994). The central role of family structure and interactions in mediating the effects of other vulnerability factors or influencing development has also been postulated (Palosaari, Aro, & Laippala, 1996; L. Sheeber, Hops, & Davis, 2001; Yap, Allen, & Sheeber, 2007). Our study is the first to demonstrate that these results persist even after adjustment for baseline depressive symptoms in a large community study. The composition of the factor further suggests that both closeness to parents and overall family function are important (perhaps including siblings). This finding is consistent with previous work (Harlow et al., 2002). McCauley and colleagues (1993) found that the overall quality of the family environment may predict the course of depressive episodes in clinically depressed children (Levin et al., 2005). Similarly, our findings suggest that adolescent perceptions that their parents care, understand them, pay attention to them, and have a close relationship with them is inversely related to the development of depressive disorder.

High self-emancipation has previously been associated with depressive disorder (Beyers & Loeber, 2003). Several items within our study’s self-emancipation factor, including greater levels of sexual activity (Kosunen, Kaltiala-Heino, Rimpela, & Laippala, 2003), substance abuse (Riggs, Baker, Mikulich, Young, & Crowley, 1995), and the commission of violent and non-violent crime (Pliszka, Sherman, Barrow, & Irick, 2000) have been associated with the onset of depressive disorders in youth. Most longitudinal studies of adolescent mental health have either not reported self-emancipation as a risk factor for depression or did not clearly establish the direction of the causation (Beyers & Loeber, 2003; Frost, Reinherz, Pakiz Camras, Giaconia, & Lefkowitz, 1999; Lewinsohn et al., 1994).

In our study, self-emancipation significantly increased the risk for future depression, and this risk was amplified by avoidant problem solving/low self-worth. This may be attributable to an interplay between the increased risk for high-stress events that accompanies self-emancipation such as parental conflict and possible legal problems as the result of extreme misbehavior. Simply put, highly self-emancipated teens are more likely to find themselves in depressogenic situations, and a proactive problem solving style mitigates the risk of depression presented by such situations. This interpretation is consistent with the reported effectiveness of teaching problem solving approaches toward reducing symptoms of mental disorders in highly self-emancipated youth (Biggam & Power, 2002).

The negative relationship between religious participation and depression risk is a new finding in the adolescent context. Several observations from the adult and adolescent literature conflict with each other. Some show protection with increased religious activity (Cotton, Larkin, Hoopes, Cromer, & Rosenthal, 2005; Sinha, Cnaan, & Gelles, 2006) while others do not (Baetz, Griffin, Bowen, Koenig, & Marcoux, 2004). The cross-sectional nature of many studies has left open the question of whether low religious involvement is merely an association or an actual predictor of MDD. The persistence of the association in our univariate longitudinal model even after adjustment for baseline depressive symptoms suggests a protective role for religious experience, but its mediation through the family/interpersonal factor may suggest that religious involvement only lowers depression risk because it serves as an indicator for a healthy, well-connected family life.

The lack of significance regarding avoidant problem solving/low self-worth in the univariate model contradicts previous work showing poor coping skills to be associated with increased future risk of depressive disorder (Lewinsohn et al., 1994). However, problem solving based prevention interventions have demonstrated minimal benefits in universal interventions and have been deemed ineffective in some racial/ethnic groups (Cardemil & Reivich, 2005; Horowitz & Garber, 2006). This suggests that problem solving skills have variable effects on depression risk depending on individual, family, and cultural factors, and the interaction between avoidant problem solving and self-emancipation in our data suggests that these problem solving skills are most important to those who are most likely to face serious problems.

Large probability samples of community-dwelling adolescents that permit the longitudinal examination of depression and depression risk variables have been rare. We identified an underlying factor structure for vulnerability and protective factors not previously evaluated or reported in the National Longitudinal Study of Adolescent Health (Resnick et al., 1997; Rushton et al., 2002). The findings complement work performed on smaller, more homogenous US and European samples (Allen et al., 2006; Cole, Martin, Peeke, Seroczynski, & Hoffman, 1998; Fergusson, Horwood, Ridder, & Beautrais, 2005; Fombonne, Wostear, Cooper, Harrington, & Rutter, 2001a; Fombonne et al., 2001b; Hankin et al., 1998; Knapp, McCrone, Fombonne, Beecham, & Wostear, 2002; Lee & Murray, 1988; McCrone, Knapp, & Fombonne, 2005; Reinherz et al., 1999). Although an important limitation of this study was the construction of the outcome variable from the CES-D questionnaire, this method has been validated in other samples. Additionally, the depression severity levels are similar to those in clinical populations, and the incidence of major depression is consistent, albeit on the low end, with other community samples (Radloff, 1991; V. Schoenbach, Kaplan, BH, Wagner, EH, Grimson, RC, Miller, FT, 1983). It would have also been helpful if items assessing family history of depression were included in the survey.


This model has the advantage of defining social relationships in broad terms. It constructs known risk factors for adolescent depression into discrete targets for analysis, explanation, and perhaps prevention interventions. This work has implications for research, practice, and social policy. In terms of research, this analysis suggests that families and their contribution to depressive disorder cannot be ignored in the development of interventions. Specifically, integrated universal or targeted prevention strategies that seek to change adolescent behaviors in a favorable direction (e.g., enhanced communication and social problem solving skills) should simultaneously address parental behaviors (e.g., encourage parents to receive these behavior changes favorably). Such approaches may help the adolescent to feel understood and attended to and reduce internalizing symptoms. Conversely, problem solving and cognitive approaches may deserve elevated prominence in interventions targeted at self-emancipated youth compared to universal interventions. In terms of practice, clinicians may find it helpful to consider with their adolescent patients how self-emancipation behaviors, which often bring short-term release from adult control, increase the future risk of depressive illness while more community and religious engagement reduce risk. From a social policy standpoint, encouraging community engagement in religious and civic activities may provide new behavioral targets for structural interventions. Most importantly, there is an overall importance of families and communities in fostering depression resiliency and we recommend that more resources should be devoted to family and community based intervention approaches.


Financial support

Dr. Van Voorhees is supported by a NARSAD Young Investigator Award, Robert Wood Johnson Foundation Depression in Primary Care Value Grant, and a career development award from the National Institutes of Mental Health (NIMH K-08 MH 072918-01A2).


Conflict of Interest

The authors report no conflict of interest.


  • Allen JP, Insabella G, Porter MR, Smith FD, Land D, Phillips N. A social-interactional model of the development of depressive symptoms in adolescence. J Consult Clin Psychol. 2006;74(1):55–65. [PMC free article] [PubMed]
  • Attkisson CC, Zich JM, editors. Depression in primary care: Screening and detection. Florence, KY, US: Taylor and Francis/Routledge; 1990.
  • Baetz M, Griffin R, Bowen R, Koenig HG, Marcoux E. The association between spiritual and religious involvement and depressive symptoms in a Canadian population. J Nerv Ment Dis. 2004;192(12):818–822. [PubMed]
  • Beyers JM, Loeber R. Untangling developmental relations between depressed mood and delinquency in male adolescents. J Abnorm Child Psychol. 2003;31(3):247–266. [PubMed]
  • Biggam FH, Power KG. A controlled, problem-solving, group-based intervention with vulnerable incarcerated young offenders. Int J Offender Ther Comp Criminol. 2002;46(6):678–698. [PubMed]
  • Binder J, Angst J. [Social consequences of psychic disturbances in the population: a field study on young adults (author’s transl)] Arch Psychiatr Nervenkr. 1981;229(4):355–370. [PubMed]
  • Bramesfeld A, Platt L, Schwartz FW. Possibilities for intervention in adolescents’ and young adults’ depression from a public health perspective. Health Policy. 2006;79(2–3):121–131. [PubMed]
  • Breslau N, Kilbey M, Andreski P. Nicotine dependence, major depression, and anxiety in young adults. Arch Gen Psychiatry. 1991;48(12):1069–1074. [PubMed]
  • Breslau N, Kilbey MM, Andreski P. DSM-III-R nicotine dependence in young adults: prevalence, correlates and associated psychiatric disorders. Addiction. 1994;89(6):743–754. [PubMed]
  • Cardemil E, Reivich K. Preventing Depressive Symptoms in Low-Income Minority Children. Clark University University of Pennsylvania; 2005.
  • Christie KA, Burke JD, Regier DA, Rae DS, et al. Epidemiologic evidence for early onset of mental disorders and higher risk of drug abuse in young adults. Am J Psychiatry. 1988;145(8):971–975. [PubMed]
  • Cicchetti D, Toth SL. The development of depression in children and adolescents. Am Psychol. 1998;53(2):221–241. [PubMed]
  • Cole DA, Martin JM, Peeke LG, Seroczynski AD, Hoffman K. Are cognitive errors of underestimation predictive or reflective of depressive symptoms in children: a longitudinal study. J Abnorm Psychol. 1998;107(3):481–496. [PubMed]
  • Compas BE, Hinden BR, Gerhardt CA. Adolescent development: pathways and processes of risk and resilience. Annu Rev Psychol. 1995;46:265–293. [PubMed]
  • Cotton S, Larkin E, Hoopes A, Cromer BA, Rosenthal SL. The impact of adolescent spirituality on depressive symptoms and health risk behaviors. J Adolesc Health. 2005;36(6):529. [PubMed]
  • Ernst C, Foldenyi M, Angst J. The Zurich Study: XXI. Sexual dysfunctions and disturbances in young adults. Data of a longitudinal epidemiological study. Eur Arch Psychiatry Clin Neurosci. 1993;243(3–4):179–188. [PubMed]
  • Fergusson DM, Horwood LJ, Ridder EM, Beautrais AL. Subthreshold depression in adolescence and mental health outcomes in adulthood. Arch Gen Psychiatry. 2005;62(1):66–72. [PubMed]
  • Fombonne E, Wostear G, Cooper V, Harrington R, Rutter M. The Maudsley long-term follow-up of child and adolescent depression. 1. Psychiatric outcomes in adulthood. Br J Psychiatry. 2001a;179:210–217. [PubMed]
  • Fombonne E, Wostear G, Cooper V, Harrington R, Rutter M. The Maudsley long-term follow-up of child and adolescent depression. 2. Suicidality, criminality and social dysfunction in adulthood. Br J Psychiatry. 2001b;179:218–223. [PubMed]
  • Frost AK, Reinherz HZ, Pakiz Camras B, Giaconia RM, Lefkowitz ES. Risk factors for depressive symptoms in late adolescence: A longitudinal community study. American Journal of Orthopsychiatry. 1999;69(3):370–381. [PubMed]
  • Garber J. Depression in children and adolescents: linking risk research and prevention. Am J Prev Med. 2006;31(6 Suppl 1):S104–125. [PubMed]
  • Goodman SH, Lahey BB, Fielding B, Dulcan M, Narrow W, Regier D. Representativeness of clinical samples of youths with mental disorders: a preliminary population-based study. J Abnorm Psychol. 1997;106(1):3–14. [PubMed]
  • Hallowell EM, Bemporad J, Ratey JJ. Depression in the transition to adult life. In: Feinstein SC, Esman AH, editors. Adolescent psychiatry: Developmental and clinical studies, Vol. 16. Annals of the American Society for Adolescent Psychiatry. Chicago, IL, US: The University of Chicago Press; 1989. pp. 175–188.
  • Hankin BL. Adolescent depression: description, causes, and interventions. Epilepsy Behav. 2006;8(1):102–114. [PubMed]
  • Hankin BL, Abramson LY, Moffitt TE, Silva PA, McGee R, Angell KE. Development of depression from preadolescence to young adulthood: emerging gender differences in a 10-year longitudinal study. J Abnorm Psychol. 1998;107(1):128–140. [PubMed]
  • Haringsma R, Engels GI, Beekman AT, Spinhoven P. The criterion validity of the Center for Epidemiological Studies Depression Scale (CES-D) in a sample of self-referred elders with depressive symptomatology. Int J Geriatr Psychiatry. 2004;19(6):558–563. [PubMed]
  • Harlow BL, Cohen LS, Otto MW, Liberman RF, Spiegelman D, Cramer DW. Demographic, family, and occupational characteristics associated with major depression: the Harvard study of moods and cycles. Acta Psychiatrica Scandinavica. 2002;105(3):209–217. [PubMed]
  • Harris FC, Ammerman RT. Depression and suicide in children and adolescents. Education and Treatment of Children. 1986;9(4):334–343.
  • Horowitz JL, Garber J. The prevention of depressive symptoms in children and adolescents: A meta-analytic review. J Consult Clin Psychol. 2006;74(3):401–415. [PubMed]
  • Horwitz AV, White HR. Becoming married, depression, and alcohol problems among young adults. J Health Soc Behav. 1991;32(3):221–237. [PubMed]
  • Houston TK, Cooper LA, Vu HT, Kahn J, Toser J, Ford DE. Screening the public for depression through the Internet. Psychiatric Services. 2001;52(3):362–367. [PubMed]
  • Kessler RC, Avenevoli S, Ries Merikangas K. Mood disorders in children and adolescents: an epidemiologic perspective. Biol Psychiatry. 2001;49(12):1002–1014. [PubMed]
  • Kessler RC, Walters EE. Epidemiology of DSM-III-R major depression and minor depression among adolescents and young adults in the National Comorbidity Survey. Depress Anxiety. 1998;7(1):3–14. [PubMed]
  • Klerman GL. The current age of youthful melancholia. Evidence for increase in depression among adolescents and young adults. Br J Psychiatry. 1988;152:4–14. [PubMed]
  • Klerman GL, Weissman MM. Increasing rates of depression. Jama. 1989;261(15):2229–2235. [PubMed]
  • Knapp M, McCrone P, Fombonne E, Beecham J, Wostear G. The Maudsley long-term follow-up of child and adolescent depression: 3. Impact of comorbid conduct disorder on service use and costs in adulthood. Br J Psychiatry. 2002;180:19–23. [PubMed]
  • Kosunen E, Kaltiala-Heino R, Rimpela M, Laippala P. Risk-taking sexual behaviour and self-reported depression in middle adolescence--a school-based survey. Child Care Health Dev. 2003;29(5):337–344. [PubMed]
  • Lee AS, Murray RM. The long-term outcome of Maudsley depressives. Br J Psychiatry. 1988;153:741–751. [PubMed]
  • Levin HS, McCauley SR, Josic CP, Boake C, Brown SA, Goodman HS, et al. Predicting depression following mild traumatic brain injury. Arch Gen Psychiatry. 2005;62(5):523–528. [PubMed]
  • Lewinsohn PM, Roberts RE, Seeley JR, Rohde P, Gotlib IH, Hops H. Adolescent psychopathology: II. Psychosocial risk factors for depression. J Abnorm Psychol. 1994;103(2):302–315. [PubMed]
  • Lin E, Parikh SV. Sociodemographic, clinical, and attitudinal characteristics of the untreated depressed in Ontario. J Affect Disord. 1999;53(2):153–162. [PubMed]
  • Martin G, Waite S. Parental bonding and vulnerability to adolescent suicide. Acta Psychiatrica Scandinavica. 1994;89(4):246–254. [PubMed]
  • McCrone P, Knapp M, Fombonne E. The Maudsley long-term follow-up of child and adolescent depression. Predicting costs in adulthood. Eur Child Adolesc Psychiatry. 2005;14(7):407–413. [PubMed]
  • Merry S, McDowell H, Hetrick S, Bir J, Muller N. Psychological and/or educational interventions for the prevention of depression in children and adolescents. Cochrane Database Syst Rev. 2004;(1):CD003380. [PubMed]
  • Munoz RF, Mrazek PJ, Haggerty RJ. Institute of Medicine report on prevention of mental disorders. Summary and commentary. Am Psychol. 1996;51(11):1116–1122. [PubMed]
  • Muris P, Schmidt H, Lambrichs R, Meesters C. Protective and vulnerability factors of depression in normal adolescents. Behav Res Ther. 2001;39(5):555–565. [PubMed]
  • National Longitudinal Study of Adolescent Health (Add Health), Wave I and II, 1994–1996. Los Altos, CA: Sociometrics Corporation; 1998.
  • Palosaari U, Aro H, Laippala P. Parental divorce and depression in young adulthood: adolescents’ closeness to parents and self-esteem as mediating factor. Acta Psychiatrica Scandinavica. 1996;93(1):20–26. [PubMed]
  • Pliszka SR, Sherman JO, Barrow MV, Irick S. Affective disorder in juvenile offenders: A preliminary study. Am J Psychiatry. 2000;157(1):130–132. [PubMed]
  • Radloff LS. The CES-D scale: a self-report depression scale for research in the general population. Applied Psychological Measures. 1977;1:385–401.
  • Radloff LS. The use of the Center for Epidemiologic Studies Depression Scale in adolescents and young adults. J Youth Adolesc. 1991;20(2):149–166. [PubMed]
  • Reinecke MaSA. Vulnerability to DepressionAmong Adolescents: Implications for Cogntive Treatment. Cognitive and Behavioral Practice. 2005;12:166–176.
  • Reinherz HZ, Giaconia RM, Hauf AMC, Wasserman MS, Silverman AB. Major depression in the transition to adulthood: Risks and impairments. Journal of Abnormal Psychology. 1999;108(3):500–510. [PubMed]
  • Resnick MD, Bearman PS, Blum RW, Bauman KE, Harris KM, Jones J, et al. Protecting adolescents from harm. Findings from the National Longitudinal Study on Adolescent Health. Jama. 1997;278(10):823–832. [PubMed]
  • Riggs PD, Baker S, Mikulich SK, Young SE, Crowley TJ. Depression in substance-dependent delinquents. J Am Acad Child Adolesc Psychiatry. 1995;34(6):764–771. [PubMed]
  • Runeson B. Mental disorder in youth suicide. DSM-III-R Axes I and II. Acta Psychiatr Scand. 1989;79(5):490–497. [PubMed]
  • Rushton JL, Forcier M, Schectman RM. Epidemiology of depressive symptoms in the National Longitudinal Study of Adolescent Health. J Am Acad Child Adolesc Psychiatry. 2002;41(2):199–205. [PubMed]
  • Saxena S, Jane-Llopis E, Hosman C. Prevention of mental and behavioural disorders: implications for policy and practice. World Psychiatry. 2006;5(1):5–14. [PubMed]
  • Schoenbach V, Kaplan BH, Wagner EH, Grimson RC, Miller FT. Prevalence of self-reported depressive symptoms in young adolescents. Am J Publ Health. 1983;73:1281–1287. [PubMed]
  • Schoenbach VJ, Kaplan BH, Grimson RC, Wagner EH. Use of a symptom scale to study the prevalence of a depressive syndrome in young adolescents. Am J Epidemiol. 1982;116(5):791–800. [PubMed]
  • Sheeber L, Hops H, Alpert A, Davis B, Andrews J. Family support and conflict: prospective relations to adolescent depression. J Abnorm Child Psychol. 1997;25(4):333–344. [PubMed]
  • Sheeber L, Hops H, Davis B. Family processes in adolescent depression. Clin Child Fam Psychol Rev. 2001;4(1):19–35. [PubMed]
  • Sheeber LB, Davis B, Leve C, Hops H, Tildesley E. Adolescents’ relationships with their mothers and fathers: associations with depressive disorder and subdiagnostic symptomatology. J Abnorm Psychol. 2007;116(1):144–154. [PMC free article] [PubMed]
  • Sherin RJ. A matrix formulation of Kaiser’s varimax criterion. Psychometrika. 1966;31(4):535–538. [PubMed]
  • Sinha JW, Cnaan RA, Gelles RJ. Adolescent risk behaviors and religion: Findings from a national study. J Adolesc 2006 [PubMed]
  • Skodol AE, Schwartz S, Dohrenwend BP, Levav I, Shrout PE. Minor depression in a cohort of young adults in Israel. Archives of General Psychiatry. 1994;51(7):542–551. [PubMed]
  • Smit F, Willemse G, Koopmanschap M, Onrust S, Cuijpers P, Beekman A. Cost-effectiveness of preventing depression in primary care patients: randomised trial. Br J Psychiatry. 2006;188:330–336. [PubMed]
  • Yap MB, Allen NB, Sheeber L. Using an Emotion Regulation Framework to Understand the Role of Temperament and Family Processes in Risk for Adolescent Depressive Disorders. Clin Child Fam Psychol Rev 2007 [PubMed]
  • Youngblade LM, Theokas C, Schulenberg J, Curry L, Huang IC, Novak M. Risk and promotive factors in families, schools, and communities: a contextual model of positive youth development in adolescence. Pediatrics. 2007;119(Suppl 1):S47–53. [PubMed]