The NCS-A provides an opportunity to obtain nationally representative data on a broad range of risk and protective factors for adolescent mental disorders. A multi-construct, multi-informant (adolescent, parent), multi-method (interview, questionnaire, biological samples) battery was developed for this purpose. The research teams from Harvard, Yale, and McMaster worked collaboratively to develop this battery using a four-step process that included: review of the literature on risk and protective factors; selection of existing measures and assignment of the informant (i.e. adolescent, parent) needed for the assessment; preliminary development, pilot testing, and field testing of the modules; and final modifications prior to production interviewing.
A central list of constructs was developed to organize our work in developing a risk-protection factor battery. The primary goal in selecting the constructs from this list was to develop a broad-based inventory that could be assessed feasibly in a large face-to-face national survey that featured self-report and parent-informant reports as the main sources of data. A comprehensive literature review was conducted to develop the central list. Advice was also sought from numerous experts in child and adolescent developmental psychopathology to refine the central list and to select the subset of constructs to be used in the survey. Constructs were considered if they had either been identified in previous clinical or community studies or hypothesized to be important predictors of child-adolescent mental disorders.
The constructs we considered were divided into three levels: individual
level (e.g., socio-demographics, developmental factors, cognitive and academic abilities-achievements, physical health, stressful life events), family
level (e.g., family structure, stability and adaptability, parenting behavior, parental psychopathology, family stress), and environmental/contextual
level (e.g., school and neighborhood characteristics). In selecting the final subset of these constructs, priority was placed on constructs identified in prospective research that appeared to have causal influences on the development of psychopathology in adolescents.5, 16, 58–62
Evidence for the specificity of the constructs was also considered,63
with lower priority given to constructs that were not reliably assessed in an interview or questionnaire format. When multiple scales for a given construct were available, priority was given to scales that were widely used and accepted in the field, had sound psychometric properties, and were available in abbreviated form.
The specific measures used to operationalize each construct were selected from reviews of the epidemiological, clinical, and developmental literatures and evaluations of existing instruments from epidemiological studies. The methodological work carried out in preparation for the UNOCCAP study was of great value in this regard, as was the methodological work in a number of other studies, including the Methods for the Epidemiology of Child and Adolescent Mental Disorders (MECA) study,10, 64
the National Longitudinal Study of Adolescent Health,65
the National Health Interview Survey,66
the National Health and Nutrition Examination Survey (NHANES),24, 67
the Ontario Child Heath Study and Follow-up Study,68
the National Longitudinal Survey of Children and Youth (NLSY),69
the Great Smoky Mountain Study,6
and the Yale High Risk Study of Comorbidity of Anxiety and Substance Use Disorders.70
We also consulted with numerous experts on the selection of measures. The battery of risk and protective factors was evaluated in cognitive interviews that assessed the ability of respondents to comprehend the questions and response categories. Questions that were unclear in the cognitive interviews were modified and re-tested in subsequent pilot interviews. presents a summary of the measures that were included for the three levels of assessment by source of information. A brief overview of some of the main measures is presented in the remainder of this section.
Risk factors and Correlates in National Comorbidity Survey-Adolescent Extension
A. Individual-level constructs
The NCS-A assesses all the well-established socio-demographic correlates of mental disorders (e.g., age, sex, race-ethnicity, parental socio-economic status, religious affiliation). A number of these measures are markers of environmental adversity that have been associated with child, adolescent, and adult disorders.9, 71, 72
More explicit information is also collected on lifetime exposure to traumatic stressful events, ongoing childhood adversities, and past-year life events and difficulties based on evidence that such stressors are significantly associated with child-adolescent mental disorders.73, 74
Along with these more general stress measures, extensive questions are included on chronic physical conditions, accidents, and injuries based on evidence of strong associations between mental and physical disorders.75–77
A wide range of health behaviors are also assessed, including smoking, sleep patterns and problems, diet, and exercise.
Dimensional measures of psychological traits that have been shown to be associated with specific mental disorders in prospective research are also included in the NCS-A battery. These measures include: sensation seeking, neuroticism, aggression-hostility, sociability, self esteem, behavioral inhibition and emotionality as assessed in a number of standard scales of personality, temperament, and worldviews.78–82
The core personality questions are included both as self-reports in the adolescent survey and as informant reports in the parent survey in order to provide two independent assessments.
The battery also assesses a range of individual competencies that may protect against the development of psychopathology. Included here are measures of cognitive, academic, and social competence, as well as measures of coping, talents, and activities. In addition, the matrices scale of the Kaufman Brief Intelligence Test (KBIT),83
a nonverbal measure of fluid thinking and problem-solving ability, is included as an index of cognitive function. This scale, which takes approximately 10–15 minutes to complete, is the only KBIT scale included in the NCS-A because of time constraints. It was chosen because it is not as highly correlated with language skills and schooling experiences as other scales of cognitive functioning84
and because, unlike most other brief intelligence tests, it can be administered with good reliability and validity by a lay interviewer who has no formal experience in intelligence testing.
An extensive series of questions about prenatal, perinatal, and early childhood development was selected for use in the NCS-A from the larger batteries developed for the National Collaborative Perinatal Project85
and the Nurses Health Study.86
Included here are questions asked of the parent (typically the biological mother) about pregnancy factors and complications, developmental milestones across infancy and early childhood, and adolescent development. This information will enable us to investigate links between early development and subsequent mental disorders. Questions about pubertal development are also included in the NCS-A with the Tanner stage illustrations from the Adolescent Development and Behavior project87
administered to our adolescent respondents in addition to retrospective reports about the age when they first entered each Tanner stage up to their current stage.
Separate saliva samples were collected at the beginning and end of each NCS-A adolescent interview to assess stress hormones, including concentration of cortisol, DHEA-S (dehydroepiandrosterone) sulfate and subacute hypothalamic-pituitary-adrenal (HPA) axis and autonomic nervous system activation. With the large sample and complexity in scheduling interviews it was not feasible to collect the saliva sample at a specific time or times of day. However, collection of saliva samples is still valuable because this study provided a unique opportunity to collect these measures in such a large and well-characterized sample. We considered the interview a mild stressor so we could test hypotheses from prior studies regarding associations of pre- and post-interview levels and before-after changes in neuroendocrine activation during stress with a number of mental disorders.88–90
Saliva measures of reproductive hormones can also be assessed to expand our understanding of the role of sex hormones on sex and age differences in the development of psychopathology in the age range of the sample.91, 92
We did not collect DNA in the adolescent sample because we had not established consent procedures for DNA collection in the adult sample. However, we intend to collect saliva samples for DNA in the first follow-up survey of the cohort when the youngest respondents reach adulthood and can provide informed consent for genetic studies.
B. Family-level constructs
The NCS-A includes extensive questions about family factors that have been linked to child-adolescent psychopathology in previous research.72, 93
One set of these measures assesses aspects of family structure (e.g., parental, death, divorce, single parenthood), birth order, and sibship size that have been related to child development and psychopathology in previous research.4, 94, 95
Because of the well-established links between parental and child psychopathology,96–98
additional questions are included in both the adolescent and parent surveys about parental history of psychopathology using items from the Family History Research Diagnostic Criteria Interview99
and its extensions.100
A large body of research exists regarding the associations of various parenting styles (e.g., parental monitoring, neglect, harsh discipline, intrusiveness) with specific mental disorders among children and adolescents.101, 102
The NCS-A consequently includes questions about parenting styles. The Parental Bonding Instrument103
is used for this purpose. Based on uncertainties about the differential effects of parenting styles of fathers and mothers, these assessments are carried out separately for fathers and mothers. As the associations of parenting styles with child outcomes are part of a more complex web of family relationships and stresses,63
though, the adolescent and parent surveys also assess parent and sib relationships with the adolescents along with a number of dimensions of parent-child and familial functioning (e.g., communication, protection, relationship quality).
C. Environmental constructs
In addition to the more immediate family factors assessed, the NCS-A battery includes assessments of the three other environmental domains of risk and protection most relevant to adolescents: peers, schools, and neighborhoods. Peer factors are known to be strongly related to the psychological outcomes of youth as potential risk, protective, and mediating factors in the development of psychopathology.104, 105
Schools and neighborhoods are the primary environmental contexts in which adolescents develop.106
Schools have been a focus of study as both complex contexts in which disorders develop and as service provision settings. The neighborhoods in which adolescents live have also been studied in relation to health and mental health outcomes.107, 108
Most notably, the Project on Human Development in Chicago Neighborhoods found substantial variation in internalizing disorders across different neighborhoods,109
affirming the potential importance of contextual factors.
The NCS-A measures of peer factors include assessments of the extent to which each adolescent is embedded in a peer network, duration of network membership, size and density of the network, the age and sex composition of the network, the behavior patterns (e.g., substance use, delinquency, sexual behaviors, risk-taking behaviors) and normative expectations (e.g., plans for attending college) of network members, and the extent to which the adolescent has access to various types of social support from network members along with various types of negative network interactions (e.g., demands, conflicts). Parallel questions are asked about relationships with romantic partners. As we have a special interest in the associations of early physical maturity in conjunction with participation in social networks of older youth with subsequent externalizing disorders, a number of questions are included in this section of the instrument about timing and duration of network membership.
NCS-A respondents were sampled from a nationally representative sample of 320 schools. Information about school context was obtained in two ways.31
First, aggregated small area Census data for the catchment area of each school were collected to provide information on the socio-demographic composition of the population served by the school. Second, we administered separate surveys to the Principal and the Mental Health Coordinator in each participating school to collect information about school structures and processes that could be used to study the determinants of between-school variation in the probability that an adolescent with a mental disorder was detected and treated. Principals were asked questions about school resources, curriculum, policies, and services for adolescents with emotional and behavioral problems. Mental Health Coordinators were identified by the Principal and asked additional questions about off-site resources and outreach to parents. In case where the school did not have a Mental Health Coordinator, the data in the Mental Health Coordinator SAQ was provided by a school nurse or guidance counselor or the Principal. These school-level surveys were considered important in light of previous evidence that a substantial proportion of the treatment of youth with mental disorders in the US occurs either at school or at the behest of school officials106
and evidence that substantial between-school variation exists in the probability that youth with mental disorders are detected and treated.110
Neighborhood effects have only seldom been studied in previous research of youth mental disorders. The scant research in this area has generally, although not always,109
failed to find powerful neighborhood-level effects after controlling for individual and family factors.111, 112
However, there is reason to believe that this failure might have been due to these studies focusing on aggregate assessments of small area Census data rather than on more nuanced dimensions of neighborhood context. Specific characteristics of communities, particularly “social capital,”113
or the extent to which adolescents living in a given community are connected to family, friends, neighbors, and civic institutions and have relationships characterized by trust, mutual aid, and norms of reciprocity, have been shown to be powerful determinants of adolescent health and well-being.114
We have assessed many of these characteristics of communities in the NCS-A, and plan to combine these measures with other aggregated data bases that include community-level indicators of civic involvement that can be used to characterize community social capital (e.g., number of churches and average weekly attendance in every community in America, number of boy scouts in each community in America, etc.).
D. Interview length
Because of the large amount of material we needed to include in the assessment, the adolescent interview in the NCS-A was quite long, with an average length of two and a half hours (range 69 minutes to 347 minutes). The long duration can be attributed to inclusion of extensive information that supplemented the actual diagnostic sections There was considerable variability in response times, though, with adolescents in the household sample taking, on average, slightly longer than those in the school sample. This variation was due to interview length depending on the number of disorder sections completed by the adolescent, as these sections have a stem-branch structure. This structure has been found to create problems in the past, as respondents who endorse multiple disordersrather quickly catch on to the fact that endorsement of a diagnostic stem question will result in more questions, leading some respondents to deny stem questions in what appears to be a conscious effort to shorten the interview.115
In order to address this problem, the CIDI 3.0 begins with a screening section in which respondents are asked the diagnostic stem questions for all the lifetime disorders assessed in the survey. The responses to these screener items determine which supplementary sections are subsequently administered. Respondents who did not endorse any screening items for disorders took an average of a little under two hours to complete the survey, while individuals with several complex disorders spent an average of nearly three hours on the interview. The average duration of the NCS-A CIDI is comparable to that of the K-SADS,116
which has an average length of 180 minutes for the diagnostic sections, but considerably longer than that of the highly structured Diagnostic Interview Schedule for Children and Adolescents (DISC)17
which takes an average of 70 minutes in the community and up to 120 minutes in clinical samples.
To address the problem of respondent burden in what we knew during the design phase would be a long interview, we carefully evaluated each skip instruction to make sure we were skipping respondents out of sections as soon as we had the information needed to evaluate the issues under consideration. This was especially important in the diagnostic sections, where it was possible to skip respondents once it became clear that they either met criteria or failed to meet any symptom required for a diagnosis. However, given our interest in sub-threshold diagnoses in adolescents, we balanced the desire to use skips with our interest in obtaining sub-threshold information. During the data collection phase, administration of especially long interviews was broken up into multiple interview sessions.