CLASS ENUMERATION FOR LATENT CLASS MODELS: RESULTS OF A SIMULATION STUDY CONSIDERING THE LO-MENDELL-RUBIN TEST. Karen Nylund1, Bengt Muthen1, Tihomir Asparouhov2, 1University of California, Los Angeles, Los Angeles, CA United States; 2Mplus, Los Angeles, CA United States
Applications of mixture analysis have become more widely used in social science research as evident by the increasing number of manuscripts and applications using Latent Class Analysis (LCA) and Growth Mixture Modeling (GMM). While these are only two types of mixture models, the rich information provided by modeling unobserved heterogeneity in a study population has far reaching appeal. Despite their usefulness, one unresolved issue in the application mixture modeling has to do with determining the number of mixtures that exists in a study population. To date, there is not common acceptance of a statistical test for determining the number of mixtures (or latent classes). Previous applied researchers using Latent Class Analysis, for example, have used a combination of indicators to guide the decision on the number of classes, including congruence with substantive theory and the combination of statistical guidelines like AIC and BIC. The commonly used likelihood ratio test difference test (or chi-squared difference test) is not applicable in testing nested mixture models due to regularity conditions not being meet. Lo, Mendell, and Rubin (LMR; 2001) created an exact parametric likelihood ratio test that can be used to determine the number of classes in a normal mixture model that uses the accurate distribution of the difference between nested mixture models. The acceptance and use of this test has been limited due to limited studies to fully understand its usefulness in practice.
As a result, this paper is focused on exploring the performance of the newly proposed Lo-Mendell-Rubin (LMR) test as a possible tool for class enumeration in a specific type of mixture model, namely the Latent Class Analysis model with binary outcomes. We present the results of a series of simulation studies carried out in Mplus Version 3 (Muthén & Muthén, 2004) Monte Carlo facilities where the Lo, Mendell and Rubin test statistic is available as an output option (Tech 11). Model specification varied over models considered including varying sample size (N=200, 500, 1000), item probabilities, and structure (simple/complex). Results of the newly proposed Lo-Mendell-Rubin test are compared to the traditional Likelihood ratio test. Further, summary information on the performance of AIC, BIC and Adjusted BIC are included for comparison. Results indicate that the LMR outperforms the wrong LRT, across al models. Further, the Adjusted BIC seems to be the best indicator of class for LCA models. This simulation study is the first formal look at the performance of the LMR, building its case as a useful tool for class enumeration.
GENDER AND ADOLESCENT RISK TAKING: A PERSON-ORIENTED APPROACH. Keri K O'Neal1, Nancy J Bell2, 1California State University, Hayward, Hayward, CA United States; 2Texas Tech University, Lubbock, TX United States
Traditional research techniques often fail to provide an accurate picture of the underlying processes affecting adolescent risk behaviors. Consequently, prevention and intervention programs are being created that may not provide the best chance for reducing risk behaviors. Utilizing a developmental contextual, person-oriented, approach has specific implications for conducting research and highlights the importance of diversity and context. Further, it renders unlikely the possibility for "one size fits all" explanations, so common in much of the literature. Many studies on juvenile delinquency and existing delinquency theories are based on samples of young men. Further, these findings are assumed to be accurate for both genders. Assuming that norms for males translates into norms for females is antithetical to a feminist research approach which views gender as a basic organizing principle in the study of human development. Thus, the possibility for explaining female behavior based on male-dominated research is called into question. Although there has been an increased focus on gender and risk behavior, many questions remain about the way in which young women´s experiences may be different than those of young men. Issues surrounding person-oriented approaches, gender and risk behaviors served as the impetus for this research. Cluster analysis is one technique that allows for the identification and empirical classification of individuals based upon a specified set of factors. This study explored how different clustering procedures alter conclusions about gender differences or similarities in risk behavior patterns. Males and females were included in the cluster analysis as a whole sample compared with males and females entered as separate samples in the cluster analyses. One wave of the National Youth Survey was used resulting in a sample of 717 males and 674 females included in the cluster analyses. Ward's minimum-variance method yielded five clusters based upon the pseudo-T test and group sample sizes. Gender differences emerged; suggesting differential patterns associated with adolescent risk behaviors and the advantage of clustering separately by gender. The findings suggest that prevention and intervention efforts, designed to reduce risk behaviors, must have components that are gender specific and address the particular concerns of those sought to serve. However, it is insufficient to draw conclusions about “all” women or “all” men when asking gender questions, instead it is important to examine the nature of the variability within each gender as well.
CELEBRATORY RIOTS: CAMPUS POLICE AND COLLEGE STUDENTS DISCUSS WHY THEY OCCUR AND HOW TO PREVENT THEM. Dexter Taylor1, Robert Voas1, 1Pacific Institute for Research and Evaluation, Calverton, MD United States
Celebratory riots on college campuses have become commonplace, but there is no prevailing research or theory that explains this recent phenomenon. Psychological theories used to explain spectator violence are quite informative but still incomplete; some of the most useful theories include the frustration-aggression hypothesis, social learning theory, and deindividuation. Research and anecdotal evidence also suggest that a reduction of alcohol consumption at sports events can lead to a reduction in violence. This research examines the opinions of students and campus police at a mid-Atlantic university that has experienced several celebratory riots after sports events. Three groups of students and three groups of police officers were presented with a set of five questions regarding the causes of celebratory riots, the best measures to prevent future riots, and the relationship between riotous behavior and alcohol consumption among students. The participants included a few students who were bystanders during celebratory riots, but mostly students who simply attended the university during the riots. All of the participants in police groups had direct experience policing celebratory riots. The Nominal Group Technique (NGT) was used to ensure participation of each group member and to create a record of the discussion that could be easily and objectively analyzed. NGT, however, was modified to allow opportunity for open discussion. Both student and police cite alcohol consumption as a contributing factor in celebratory riots. Further, police and students recommend a designated, safe, alcohol-free venue for students to express celebration or frustration after the campus participates in national sports events. Points of divergence between police and student opinions were varied. Police nearly uniformly favored increased punishment for rioters and increased enforcement of alcohol policies. Students consistently criticized the police response as being too aggressive or ill-prepared for crowd control after sports events. Some of the student and police responses were consistent with psychological theory regarding aggression and riots. As in other surveys with students after celebratory riots, students involved in this research described the behaviors of rioters as embarrassing to the university community. Survey data from another campus that experienced riots show a majority of students were embarrassed by riots; consequently, we can assume students are predisposed to do more to prevent campus riots. An effective approach to riot prevention might include a campaign that attempts to reduce alcohol abuse during sports events and persuade potential riot bystanders to leave the gatherings.
MODELING RECIPROCAL RELATIONS IN LONGITUDINAL DEVELOPMENTAL DATA. Zhiqun Tang1, Linda Collins2, 1The Pennsylvania State University, State College, PA United States; 2Pennsylvania State University, University Park, PA United States
Quantifying or modeling reciprocal relations has been challenging to social scientists due to the complexity of social behavioral processes and because of measurement and other methodological difficulties. Although auto-regressive cross-lagged path analysis or structural equation modeling and cross-domain growth curve modeling have been popularly used approaches to modeling bi-directional relations, they are not satisfying to model reciprocal relations. The present study demonstrates methodological developments representing a new approach to modeling reciprocal relations exhibited in longitudinal developmental data. This technique is called associative latent transition analysis (ALTA), an extension of latent transition analysis (LTA; Collins & Wugalter, 1992; Collins, Hyatt, & Graham, 2000). In this study, using adolescent substance use as an illustrative example, both adolescent drinking behavior and association with alcohol using peers are conceptualized as latent sequential stages. Adolescent drinking behavior contains four stages from no use of alcohol to regular use of alcohol and being drunk, and association with peers contains three stages from no association with alcohol using peers to frequent association with alcohol using peers. Models under different hypotheses are tested by fitting these models to a multi-wave large sample of longitudinal data, featuring developmental phases from early to middle adolescence. These models reflect intra-individual changes over time in each of the two behavioral processes and how change in one process relates to change in another process over time. Particularly, two models with unidirectional influences are tested against a model where reciprocal relations are allowed. The results show that ALTA technique provides a very unique way to examine longitudinal data, helping to address substantive questions interesting to researchers in prevention and intervention areas.
Collins, L. M. & Wugalter, S. E. (1992). Latent class models for stage-sequential dynamic latent variables. Multivariate Behavioral Research, 27 (1), 131-157.
Collins, L. M., Hyatt, S. L. & Graham, J. W. (2000). Latent transition analysis as a way of testing models of stage-sequential change in longitudinal data. In T. D. Little, K. U. Schnabel, & J. Baumert (Eds.), Modeling Longitudinal and Multiple-Group Data: Practical Issues, Applied Approaches, and Specific Examples (pp147-161). Hillsdale, NJ: Erlbaum.
PREVALENCE AND PREVENTION FOR EMOTIONAL AND BEHAVIORAL DISORDERS AMONG DIVERSE PRESCHOOLERS. Cheryl Boyce1, 1National Institutes of Health, Bethesda, MD United States
Along with pediatric primary health care providers, Head Start, as a comprehensive service delivery program, serves as one of the earliest mechanisms for identification and intervention with a vulnerable population of low-income children and their families. Mental health within Head Start can incorporate a balanced emphasis that includes prevention as a key cornerstone of early intervention efforts, but also allows for the appropriate early identification and treatment of children at-risk for, or manifesting emotional and/or behavioral difficulties. This session highlights research from the Head Start Mental Health Research Consortium which has develop and tested applications of theory-based research and state-of-the-art techniques for the prevention, identification and/or treatment of children's mental health disorders within a Head Start context. This consortium of researchers has focused on advancing our current level of understanding and improving the provision of high quality, comprehensive, developmentally appropriate prevention and intervention services to young low-income children, families and staff, served by Head Start programs across the country. It is clear that early childhood offers an opportunity for prevention. In this session multi site data on symptom, impairment and language among diverse Head Start preschool populations will be presented. An evidenced based behavioral intervention will also highlight the opportunities for the prevention of emotional and behavioral problems among preschool populations. The presentations will be informed by a discussion of the cultural sensitivity issues of assessment and intervention with diverse young children and their families.
SYMPTOM AND IMPAIRMENT MEASURES AND THEIR RELATIONSHIP TO GENDER AND ETHNICITY FOR A MULTI-SITE LOW-INCOME SAMPLE. Edward Feil1, Jason W. Small2, Steve Forness3, Cheryl Boyce4, Michael Lopez5, 1University of Oregon, Eugene, OR United States; 2Oregon Research Institute, Eugene, OR United States; 3University of California, Los Angeles, Los Angeles, CA United States; 4National Institutes of Health, Bethesda, MD United States; 5United States Department of Health & Human Services, Washington, DC United States
Research on the prevalence, severity, and topography of mental health needs among low income preschoolers and their families has not been well documented. While the effects of poverty, gender and race/ethnicity has been shown to be profound, these effects in tandem with differing measurement types has not been well researched among young children in poverty. In this study, four symptom and three functional impairment measures completed by teachers and/or parents were obtained on a sample of 1,781 Head Start children from diverse ethnic backgrounds from the Head Start Mental Health Research Consortium. Participants were selected from four of the five sites of the Head Start Mental Health Research Consortium, a cross-site multi-year study designed to examine mental health issues in children enrolled in Head Start preschool settings (Boyce, Hoagwood, Lopez & Tarullo, 2000). The current sample includes data for children, ages 3 to 5 years old, from six major ethnic categories (African American, Hispanic, Caucasian, Native American, and Asian). Clinical cut off scores were used to identify children who could be considered at relatively serious risk for emotional or behavioral disorders. At risk classifications using clinical cut-offs at both 1.0 and 1.5 standard deviations for each measure were examined singly and in combination and then compared to the overall sample for gender and ethnicity. Identification of children considered at risk ranged from a low of 1% to a high of 38%, with considerable evidence of differential effects on gender or ethnicity for some measures and combinations of measures. The combination of a teacher symptom measure with a teacher rating of functional impairment also resulted in a notable reduction in differential effects on age, gender, and ethnicity. We explore the assignment of “caseness” through the combinations of six parent and teacher measures. Rates ranged from 1.3% to 5.2%. It should be noted that statistically significant differential effects are no longer evident at either cutoff. Implications for choosing instruments to establish eligibility for emotional or behavioral disorders in preschoolers are discussed.
LANGUAGE AND BEHAVIOR AS RISK INDICATORS AMONG CHILDREN ENROLLED IN HEAD START. Ann Kaiser1, Terry Hancock1, Stephanie Milan2, 1Vanderbilt University, Nashville, TN United States; 2University of Conneticut, Hartford, CT United States
Although early emergent language delays and behavior problems are known to be persistent over time, relatively little is known about the relationship between these two domains of development during the preschool years. While some studies have indicated a relatively low correlation between language and behavior, other studies suggest that early behavior problems are predictive of later language related outcomes including reading and school achievement. This paper describes longitudinal data collected for 256 children enrolled in Head Start. Children were first screened for language status and behavior problems during the fall of the three year old year in Head Start. Follow up assessments occurred during the spring of their four year old year and kindergarten year and included teacher reports of behavior and social skills, parent reports of behavior and social skills, and language assessments. Four subgroups of children were identified based on their language and behavior status at the age 3 screening: high language (PLS-3>90); high language and high behavior problems (CTRF> 63); low language (PLS-<70); and low language and high behavior. Growth curves for language were plotted for these groups using data from the three assessments. As a group, children showed improvements in language over time, particularly on measures of vocabulary. Language progress differed depending on children´s initial language and behavior status. Among the children with low language, children with behavior problems showed significantly less growth in language than their peers with no behavior problems. There was a trend toward less growth by children with high language and behavior problems compared to children with high language and no behavior problems. A second analysis examined “recovery” rates in low language children with and without behavior problems. Sixty percent of children with low language and no behavior problems tested within 1SD of the normative mean at age 5, compared to less than 30% of the low language children with behavior problems. Age 5 school achievement and reading measures showed low language and behavior children to be performing lower than low language children with out behavior problems. Data from the current study suggests that children with low language and high behavior problems during the early preschool years may be uniquely at risk for subsequent school failure. Early identification and intervention to address both behavior and language skills in this group is recommended.
THE PRESCHOOL BEHAVIOR PROGRAM: EFFECTIVENESS OF A UNIVERSAL PREVENTIVE INTERVENTION FOR USE WITH HEAD START CHILDREN. Janis Kupersmidt1, Donna Bryant1, Mary Ellen Voegler-Lee1, 1University of North Carolina at Chapel Hill, Chapel Hill, NC United States
There are few evidence-based interventions designed to prevent aggression in preschool children and even fewer involving children in Head Start settings (Bryant, Vizzard, Willoughby, & Kupersmidt, 1999). The present intervention was conducted with 23 intervention and 15 control group Head Start teachers of primarily four-year-old children. Classroom quality was assessed in the middle of the intervention using the Early Childcare Environment Rating Scale and most classrooms were average or below average in quality. Data from multiple informants were collected from a subset of children and parents including 98 children in the intervention classrooms out of a possible 444 children, and 90 in the control classrooms out of a possible 270 children. The three major components of the Preschool Behavior Program included strategies to improve teacher-child and parent-child interactions and teacher's and parent's behavior management strategies (Webster-Stratton, 1989); children´s emotional intelligence and emotional regulation skills as well as social problem solving skills (Second Step Preschool Curriculum); and dialogic reading at home and in the classroom to improve children´s communication skills (Whitehurst, Arnold, Epstein, Angell, Smith, & Fischel, 1994). Interventionists spent one day per week with each intervention teacher in the classroom to mentor and assist her in the transfer of training across the day. A lending library was created with annotated children´s books containing sample questions designed to stimulate discussions between parents and their children. Interventionists offered home visits to the parent(s) of the most aggressive children in the classroom. Pre- and post-measures using multiple informants of aggressive and disruptive behavior (Kupersmidt, Bryant & Willoughby, 2000) were collected as well as child interview data to assess emotional intelligence and social problem solving skills. Also, teachers provided reports of social behaviors (Feil, Walker, & Severson, 1995), problem behaviors (Gresham & Elliot, 1990), and inattention/oppositional behaviors (Pelham, Milich, Murphy, & Murphy, 1989). Analyses suggest that children in the intervention classrooms were more appropriate in social interactions, exhibited fewer problem behaviors, and were less inattentive and oppositional than children in the control classrooms. Additional analyses will examine the moderating effects of child and teacher characteristics on the effectiveness of the intervention. Results will be discussed in terms of the importance of early intervention for the prevention and treatment of aggressive and disruptive behavior problems in preschool, and the importance of social and emotional skills development in all preschoolers for kindergarten readiness.
CONCURRENT 5, ETIOLOGY, Organized symposium
USING MENTAL HEALTH INTERVENTION OUTCOMES TO INFORM SUBSTANCE ABUSE ETIOLOGY AND PREVENTION
Chair: Naimah Weinberg
USING MENTAL HEALTH INTERVENTION OUTCOMES TO INFORM SUBSTANCE ABUSE ETIOLOGY AND PREVENTION. Naimah Weinberg1, George Howe2, 1National Institutes of Health, Bethesda, MD United States; 2George Washington University, Washington, DC United States
Cross-sectional studies of adolescents and adults in both clinical and general populations have found high rates of co-occurrence between substance use disorders (SUDs) and psychiatric disorders, particularly the conduct/antisocial disorders and the mood disorders. However, far fewer longitudinal studies have examined the temporal order or causal relationships for specific psychiatric disorders and SUDs, and the nature of the relationships among psychiatric and drug use disorders is unclear. Similarly, we have little data on whether interventions for childhood psychiatric disorders can alter initiation of drug use or SUD trajectories. Thus, understanding the relationships between precursor disorders, interventions, and SUD outcomes has important prevention and treatment implications. Moreover, reciprocal research is also needed: that is, findings from preventive interventions should be used to validate or question etiologic models and help distinguish risk markers from causal risk factors. For example, if an effective intervention for a known precursor disorder is delivered and SUD outcomes are not reduced, this may suggest that the childhood psychiatric disorder is not part of the causal chain for SUD or that the aspect of the disorder addressed by the intervention is not critically linked to SUD.
The three studies presented in this symposium are part of an ongoing NIDA-NIMH initiative to understand the impact of child psychopathology and childhood interventions on subsequent drug abuse. Each focuses on a different disorder commonly comorbid with SUD and each takes a different approach to sampling and intervention. The first presents a longitudinal study of a community-based sample identified in childhood with and without attention deficit hyperactivity disorder (ADHD), examining the relationships between childhood diagnoses, treatment service utilization, and late adolescent SUD outcomes. The second draws on data from the Fast Track Prevention Trial, which randomized children at high risk for conduct disorder to receive or not receive a long-term intervention; these subjects are now being followed to assess the impact of the intervention on later substance use, and mediating variables. The third follows adolescents who participated in the Treatment for Adolescents with Depression Study, randomized to receive fluoxetine, cognitive behavior therapy, a combination, or clinical management and placebo pills; these subjects are being followed to assess the impact of treatment on substance use. It is hoped that these findings will advance preventive interventions for SUDs both directly and through informing etiologic theory on the developmental course of SUDs.