|PARTICIPATION IN THE CHILD-PARENT CENTERS AND ADULT WELL BEING: A 20-YEAR FOLLOW UP OF PRESCHOOLERS. Arthur Reynolds1, Judy Temple2, Suh-Ruu Ou1, Dylan Robertson1, 1University of Wisconsin-Madison, Madison, WI United States; 2Northern Illinois University, DeKalb, IL United States
We investigated the effects on adult well being of participation in the Chicago Child-Parent Center (CPC) Program. Using data from the Chicago Longitudinal Study (CLS, 1999), three questions were addressed: (1) Is participation in the CPC preschool program associated with well-being at age 24, including educational attainment, income, health status, substance use, criminal behavior, and receipt of TANF/Food Stamps, (2) Is program participation associated with better family outcomes including parents´ educational attainment, employment status, receipt of AFDC/TANF and Food Stamps, and participation in children´s schooling, and (3) Do the estimated effects of participation vary by the age of entry and program length?
Most of the evidence for the link between early childhood intervention and adult well being comes from research demonstrations rather than large-scale public programs. The most consistent findings from the few studies that have followed children into adulthood are that participation is associated with greater educational attainment. Broader measures of health & well being, such as economic independence and health behavior, have rarely been investigated, especially for larger scale programs.
The CLS follows a complete cohort of 989 low-income children who attended the CPC program beginning at age 3. The comparison group included 550 children who enrolled in the usual kindergarten intervention. The program provides comprehensive services and is located in 24 public schools. At the age 24 follow up, outcomes measures were available for 1,390 study participants and 1,440 parents. The outcome measures came from CLS administrative data and survey questionnaires over two decades. The program and comparison groups had recovery rates of greater than 90%, and they were equivalent on most demographic attributes.
Results indicated, based on regression analysis with family risk status, gender, race, and program sites as covariates, that preschool participation was associated with significantly higher rates of school completion, more years of completed schooling, higher income, higher rates of health insurance coverage, and with lower rates of incarceration. For family outcomes, program participation was associated only with higher levels of parent educational attainment and with greater involvement in children´s education. Analysis of differential effects indicated that children with 4-6 years of intervention had significantly higher levels of educational attainment and higher income than children with less intervention. Starting the intervention at age 3 instead 4 did not lead to advantages in well being. Findings indicate that larger scale preventive interventions can have positive effects on the health and development into adulthood.
CHILD MALTREATMENT AND FOSTER CARE: EXAMINING THE INFLUENCE OF DIFFERENT CHILD WELFARE SERVICE INDICATORS ON YOUNG ADULT OUTCOMES. Dylan Robertson1, Joshua Mersky1, James Topitzes1, Arthur Reynolds1, 1University of Wisconsin-Madison, Madison, WI United States
Few longitudinal studies of low-income families have investigated rates of child welfare service utilization over time. The connections between child maltreatment, child welfare services and young adult outcomes are complex. This paper (1) describes rates of substantiated child maltreatment and other indicators of child protection services including foster care experiences during early childhood, middle childhood, and adolescence, and (2) investigates the relations among early childhood and family factors, child protection service indicators, and young adult outcomes by age 23 (highest grade completed, college attendance, incarceration and arrest, public aid, and substance use).
The participants in the Chicago Longitudinal Study are 1,539 low-income minority children born in 1980 who enrolled in the Chicago Child-Parent Center (CPC) program and other government-funded early childhood programs in 1985-86. Child welfare service indicators include substantiated child maltreatment records, foster care, and family case records from the Department of Children and Family Services (DCFS) and the juvenile court. Measures assessing the relations among early childhood experiences, family factors and child welfare service participation were collected from parent interviews, teacher and participant reports, and administrative records. Data for young adult outcomes were obtained from participant interviews and administrative records.
Overall, 325 children (21.1%) had child welfare family case histories. 197 children (12.8%) had substantiated reports of abuse and neglect. 98 children (6.4%) received foster care services. The mean number of months spent in foster care between ages 0 to 17 was 47.7. This compared to 45.3 months reported for children in care in Illinois on September 30, 1998 (Committee on Ways and Means, 2000). Preliminary probit regression analyses revealed that factors associated with child protection indicators (substantiated maltreatment, foster care placement, any case records) were CPC preschool participation, maternal education and age, two parent-status, and low-birth weight. Preliminary linear and probit regression analyses indicated that child protection indicators were associated with highest grade completed, college attendance, incarceration and arrest, public aid and substance use. Other significant predictors that were less robust across different young adult outcomes were CPC program participation, low-birth weight, maternal education and age, and two-parent status.
Findings suggest that child welfare service indicators are robust and predictive of young adult outcomes across different domains. Equally robust are the impacts of early childhood experiences, parent and family factors on child welfare indicators.
CONCURRENT 8, EMERGING OPPORTUNITIES FOR PREVENTION RESEARCH, Grouped Papers
FACTORS INFLUENCING THE DEVELOPMENT OF JUVENILE DELIQUENCY, PREVENTING SCHOOL FAILURE AMONG YOUTH, TELEPHONE PARENTING SKILLS INTERVENTION
Chair: Celene Domitrovich
FACTORS INFLUENCING THE DEVELOPMENT OF JUVENILE DELINQUENCY: DIFFERENCES BETWEEN EARLY AND LATE STARTERS. Kevin Alltucker1, 1University of Oregon, Eugene, OR United States
Juvenile delinquency and the resulting negative effects on youth, families, and society, have been identified as significant problems in the United States. The purpose of this study is to examine the contextual factors representing family disruption and coercive family processes that influence the age-related development of juvenile delinquency, and community reintegration outcomes for previously adjudicated youth. Data were used from the Transition Research on Adjudicated Youth in Community Settings data set, a follow-along prospective research project funded through the Offices of Special Education Program, Field Initiated Research Studies from October 1993 through September 1998 in the Institute on Violence and Destructive Behavior at the University of Oregon. Data were analyzed using logistic regression to determine odds ratios for the outcome variables of early start juvenile delinquency, and return to a correctional facility. Youth with previous foster care experience were four times more likely to be early start juvenile delinquents, than youth with no foster care experience. Youth with a parent or sibling convicted of a felony were nearly twice as likely to become early start juvenile delinquents than youth with no history of family felony. Youth with a special education disability were twice as likely to return to a correctional facility within 12 months after release, than youth with no special education disability. Therefore, this study suggests that low levels of family management skills negatively influence youth development. Family disruption, including interfamilial violence that leads to youth being placed in foster care, and family felony, were strong predictors for early start juvenile offending. Additionally, this study adds to the large body of literature that documents the adverse effects that special education disabilities have on youth development. Implications for practice and policy are discussed.
THE ROLE OF CHILD WELFARE SYSTEM IN PREVENTING SCHOOL FAILURE AMONG YOUTH AND CHILDREN. Sunny Hyucksun Shin1, 1Boston University, Boston, MA United States
Prevention research of behavioral problems among children who are involved with public child welfare systems has primarily focused on children placed in foster care. Behavioral problems and school performance of children who have contacted social service systems, but never placed in out-of-home care receive little attention. The purpose of this study was to compare behavioral problems at school, relationship with peers and teachers, and educational performance (e.g., grade progression) between the following two groups of children: (a) those who have been placed in out-of-home care and (b) those who have had contact with the child protective service, but never placed in foster care.
This study used the National Survey of Child and Adolescent Well-Being (NSCAW) data. The NSCAW is a longitudinal survey using two different samples: the Child Protective Service (CPS) sample and One Year in Foster Care (OYFC). The CPS sample includes 5,404 children, ages 0-14, who had an investigation closed and not been placed in foster care. The OYFC sample consists of 727 children, who were in out-of-home care for about 12 months at the time of sampling. This study examined and compared the following variables in both CPS and OYFC samples: Child Behavior Checklist (reported by both caregiver and teacher), relationship with teacher (reported by teacher), school achievement (measured by Mini Battery of Achievement), grade progression(reported by teacher), school engagement (measured by Drug Free Schools Outcome Study Questions), relationship with peer (measured by Loneliness and Social Dissatisfaction Questionnaire), and future expectation about education (measured by Adolescent Health Survey). A series of bivariate analyses and logistic regressions were run to model the relationship between a child's involvement with the child welfare systems and grade progression.
Findings indicate that children in both CPS and OYFC samples are at great risk of school failure. The results show very high rates of grade repetition and behavioral problems in both CPS and OYFC groups. Furthermore, children in both groups are less likely to perform at or above grade level than their peers in the general population. Sample differences are examined, but found to explain little of school engagement and educational performance. This study suggested that the child welfare systems have great potentials to provide educational support to children in contact with social services. The results suggest providing more concrete academic assistance and building academic achievement monitoring systems for children in contact with social services. Findings also suggest the need to explore more fully the role of child welfare services on educational achievement of foster children.
POPULATION TRIAL OF A PARENTING SKILLS INTERVENTION VIA TELEPHONE COUNSELING TO PREVENT ADOLESCENT SUBSTANCE USE. John Pierce1, Lisa James1, Mark Myers1, 1University of California, San Diego, La Jolla, CA United States
We used a large nationally representative survey to enroll and randomize over 1000 parents from families with an oldest child aged 11 through 13 years. The intensive phase of the intervention is a 14-session modification of the Oregon Adolescent Transitions Program to be delivered by telephone. The intensive portion is followed by quarterly maintenance calls, with booster sessions where needed. This program has three major parenting modules: building positive behaviors, setting effective limits, and relationship building. The intervention is delivered by facilitators chosen and extensively trained (60 hours) to use motivational-interviewing techniques to implement the study protocol, by focusing participants on setting goals for the next session and helping them with performance review. Each session in the intensive phase of the intervention is approximately one-half hour in duration and is scheduled at the convenience of the participant. This approach is modeled on successful interventions the study team has conducted in smoking cessation and dietary change. An intensive quality-assurance program includes regular-taped sessions with peer review and weekly case-management discussions supervised by a clinical psychologist. Participant evaluation of the intensive phase of the intervention has been very enthusiastic. An adolescent survey is scheduled twice a year and a $15 incentive is provided for completion. A parenting survey is undertaken at baseline and one year. This paper will describe the comparability of groups in this study on variables that might impact parenting performance. It will also describe the uptake continuum for the youth population at baseline for cigarette smoking, alcohol consumption, and marijuana use.
CONCURRENT 9, METHODS, Organized Symposia
BEYOND LATENT GROWTH CURVE MODELS SYMPOSIUM
Chair: George Howe
BEYOND LATENT GROWTH CURVE MODELS SYMPOSIUM. George Howe1, 1George Washington University, Washington, DC United States
Latent growth curve modeling, including the more advanced growth mixture modeling, has become a familiar site in the prevention science literature. The rich results that can be obtained in applying these models have affirmed the benefits of longitudinal observation and data collection. With more and more emergent longitudinal studies, the limitations as well as the advantages of the “standard” latent growth curve models have become increasingly apparent.
This symposium presents a collection of papers that propose distinct approaches to modeling longitudinal data for which the regular assumptions of latent growth curve models are not appropriate. The first paper grapples with the challenges of modeling mediation in a growth model with the mediator itself changing over time. A comparison is made between a mediation model specification in the latent growth curve framework and an alternate cross-lag model specification. The second paper addresses the challenges of measurement error in outcome when the outcome is episodic, calling for survival analysis techniques. Although time-specific variation and error variance maybe be incorporated in a straightforward way for continuous outcomes in a growth model, a more complex model is required to account for error in the assessment of event occurrence in a time-to-event model. The third paper examines longitudinal data that represents a record of changing individual states or conditions over time. Such data is best characterized by the transitions from time point to time point rather than mean trends in outcome over time. Investigating individual differences, particularly unobserved heterogeneity, in these transition patterns requires a mixture model specification that differs in important ways from the growth mixture models for continuous outcomes.
LATENT MIXED MARKOV CHAIN MODELS FOR ANALYZING CHANGES IN SELF-REPORTED VICTIMIZATION IN MIDDLE SCHOOL STUDENTS. Karen Nylund1, Bengt Muthen1, Amy Bellmore1, Adrienne Nishina1, Janna Juvonen1, Sandra Graham1, 1University of California, Los Angeles, Los Angeles, CA United States
This paper explores a type of longitudinal analysis dealing with change in discrete transitions from time point to time point. This analytic strategy, Latent Mixed Markov Chain Modeling, builds on two mixture modeling ideas: 1)Latent Transition Markov Modeling and 2)Latent Class Analysis (LCA). The combination of these modeling ideas in a general framework allows for more sophisticated research questions regarding discrete change in social states. To illustrate the utility of this framework, data from the UCLA Peer Project (Graham & Juvonen) explores the change in self-reported peer victimization over the course of three years (6th, 7th, and 8th grade) in a sample of approximately 2,000 urban public middle school students. Different than standard growth curve modeling where the focus is on estimating one growth mean representing an overall rate of change in outcome over time, Latent Transition Markov Modeling is focused on the change in discrete states from time period to time period. In Latent Transition Modeling, a person´s state is thought to be unobservable but evidenced by one or more observable indicators. LCA is used to identify an individuals´ “true” state at each time period, e.g., victim status during each semester. Accordingly, the transition modeling is focused on change in each person´s unobserved latent state from time period to time period. The addition of a second-order latent class variable to the transition model allows for the exploration of unobserved heterogeneity in transition patterns across time, allowing for subgroups to have characteristically different transition patterns. In the data illustration, a LCA of six survey items measuring self-reported peer victimization identified 3 latent subgroups at each time period: frequently, sometimes, and seldom victimized students. Latent Transition Analysis modeled the change in student´s victimization states across 6 time points (fall and spring) semesters from 6Th through 8th grades. The inclusion of background variables, such as gender and ethnicity, indicated differences in transitional patterns dependent on the variable values. The addition of the second-order Latent Class variable identified key transitional patterns, such as a subgroup of students who remained in the frequent victimization state throughout the study with low probabilities of transitioning out of a victimization state between any two consecutive semesters. The identification of such latent subgroups allows for the exploration of predictors of group membership as well as consequences of group membership, such as high school dropout. The identification of these subgroups can also predict stability of group membership such that school personnel can focus on intervening with the most at-risk students early on.
MODELING MEASUREMENT ERROR IN EVENT OCCURRENCE FOR DISCRETE-TIME SURVIVAL ANALYSIS OF THE TIME FROM ALCOHOL CONSUMPTION ONSET TO AN ALCOHOL USE DISORDER (AUD). Katherine Masyn1, 1University of California, Davis, Davis, CA United States
In previous work by Muthén and Masyn (2004) and Masyn (2003), it has been shown that discrete-time survival models may be estimated using a Latent Class Regression (LCR) framework (see also Vermunt, 1997, 2002). This framework conveniently accommodates many interesting model extensions including, for example, nonparametric specification of unobserved heterogeneity in survival processes across individuals. What has not yet been fully incorporated into this framework is the possibility of measurement error and bias in the determination of event occurrence.
Typically, the focus of measurement issues in survival analyses is on the reported timing of the event for those who are known to have experienced the event. For example, in retrospective studies, subjects may report event times as more recent than the actual occurrences. It could be important, however, to also consider that there may be measurement error and bias in the event occurrence determination itself. Take, for example, a situation in which subjects who have not experienced the event have a non-zero likelihood of incorrectly reporting having experienced the event. As another example, consider a clinical outcome, such as depression, as the event of interest. Although there may be extensive measurement models developed to utilize symptom-level information to obtain a proper diagnosis, such diagnoses are often treated as directly observable, terminating events in a survival analysis. Ignoring this type of measurement error and uncertainty in event status can lead to biased estimates of baseline hazard probabilities as well as biased estimates of covariate effects on the hazard of the event over time.
This paper presents a way to incorporate error in event determination motivated by a data example involving the time from alcohol consumption onset to an Alcohol Use Disorder. The data for this example comes from a study of 499 adolescents conducted by the Pittsburgh Adolescent Alcohol Research Center (PAARC). Subjects were drawn from both clinical and community sources. The subjects were assessed at multiple time points using a structured clinical interview. The ages of onset and offset were recorded to the nearest month for each DSM-IV symptom corresponding to Alcohol Use Disorders. This paper demonstrates how symptom-level data can be used in an event history model for the timing of specific disorders. The proposed model further allows for the investigation of covariates effects, e.g., parent history of AUDs, on event timing.
EVALUATION OF LONGITUDINAL MEDIATING MECHANISMS OF A DRUG PREVENTION PROGRAM FOR ADOLESCENT ATHLETES. Jeewon Cheong1, David Mackinnon2, Jason Williams2, Matt Fritz2, 1University at Albany, SUNY, Albany, NY United States; 2Arizona State University, Tempe, AZ United States
This paper investigates the longitudinal mediating mechanisms of a drug prevention program for adolescent athletes, entitled "ATLAS (Adolescents Training and Learning to Avoid Steroids)." When both the mediator and the outcome variables are measured repeatedly over time, the mediator and the outcome can be modeled as two distinctive processes that are influenced by the treatment program. In this presentation, we describe different modeling approaches to testing long term mediation and examine the mediating mechanisms of the ATLAS prevention program. The ATLAS program was designed to reduce high school football players; anabolic steroid use and improve healthy behaviors such as eating a nutritional diet and strength training. To change the outcome behaviors, the treatment program targeted several mediators, such as knowledge about drug effects, attitudes, peer and media influences, and resistance skills. Thirty one high school football teams from Oregon and Washington states participated in the program for four years. In this paper, the long term mediation process is modeled in two different modeling frameworks: Cross lagged model and latent growth curve model. In the cross lagged model, the treatment group membership is modeled to influence the repeated measures of the mediator and the outcome, while the effect of the mediator (M) on the outcome (Y) is specified with lag one effect (Mt 1®Yt). In the latent growth curve model, the group membership is modeled to influence both the growth of the outcome and the growth of the mediator, while the effect of the mediator on the outcome is modeled by relating the growth of the mediator to the growth of the outcome. In our preliminary analyses with the first cohort athletes, it was found that the treatment program successfully increased the knowledge of the effects of anabolic steroids after the implementation of the program, which, in turn decreased the athletes intention to use the drug at the one-year post-test (Mediated effect=-.086; SE=.018). In addition, it was found that the program changed the growth trajectory of the knowledge, which, in turn, was related to the trajectory of intent to use anabolic steroids (Mediated effect=.355; SE=.079). The presentation will focus on the different information on the mediating mechanisms provided by the different modeling approaches and report the findings on the mediating mechanisms for outcome variables.