FINDINGS FROM A RANDOMIZED CONTROL TRIAL OF THE HEALTHY FAMILIES NEW YORK PROGRAM (HFNY). Kimberly Dumont1, Chris Spera2, Susan Mitchell-Herzfeld1, Eunju Lee2, Rose Greene2, Ann Lowenfels2, Dorothy Baum2, 1Office of Child and Family Services, New York State, Rensselaer, NY United States; 2University at Albany, SUNY, Albany, NY United States
Goal: The purpose of the proposed individual paper presentation is to report on early findings from a randomized trial of Healthy Family New York (HFNY) in the areas of maltreatment and parent attitudes. The study design involves nearly 1200 women in three sites and thus facilitates a rigorous evaluation of the program´s effectiveness.
Program Description: HFNY is a community based prevention initiative designed to improve the well being of disadvantaged families living in New York State. Modeled after Healthy Families America (HFA), HFNY is a paraprofessional home visiting program that aims to 1) reduce child abuse and neglect, 2) enhance parent-child interactions, 3), optimize child development, health, and school readiness, and 4) promote parental life course development. HFNY is operating in 28 sites, including nine sites in New York City.
Methodology: The evaluation consists of a randomized controlled trial in which families were assigned to either an intervention group that was offered HFNY services or a control group that was given information or referrals to other appropriate services. Baseline interviews were conducted with 1,157 study participants shortly after random assignment. Follow-up interviews were conducted around the time of the target child´s first, second and third birthdays. Child maltreatment outcomes were measured using parent self-reports on the CTS-PC, and official records of reports and substantiated cases of abuse and neglect. Parenting attitudes were assessed using the AAPI. This papers focuses on results of data collected at the baseline and first two follow-ups (retention rates of 92% and 87%, respectively). The analyses assume “an intent to treat” approach and therefore provide conservative estimates of the program´s effectiveness.
Results: Findings from the first year of the evaluation suggest that HFNY had a significant impact on parenting outcomes. HFNY parents were less likely to report neglecting their children and reported committing fewer acts of severe physical abuse, minor physical aggression, and psychological aggression against their children than control parents. HFNY parents also reported less favorable attitudes toward the use of corporal punishment compared to control parents. The presentation will also report on program impacts as of the target child´s second birthday.
Discussion: We will discuss findings, best practices, and lessons learned in light of the national investment in HFA, the current debate about the programs´ effectiveness, and the needs of policy-makers and practitioners for credible, relevant evaluation results.
EFFECTS OF PARENTAL RELIGIOUSNESS, RELIGIOUS FUNDAMENTALISM, AND WARMTH ON CHILD EMOTIONAL PROBLEMS. Erin St. John1, James Paulson1, Natalie Jacobowski1, 1Eastern Virginia Medical School, Norfolk, VA United States
Parents´ choice of childrearing strategies can have a significant impact on children´s development and mental health. Although past research is limited, parental religiousness/ spirituality has been reported as predictive of parenting style (Mahoney 2001). Better understanding this influence on parenting style would improve identification of risk and resiliency factors, faith-based targeting of preventive interventions for parenting, and facilitate more religiously-sensitive parenting services. This study explores how parents´ religiousness and religious fundamentalism predict child emotional mental health and how parental warmth might mediate that effect. An anonymous survey was given to parents of children age 16 or younger at four Department of Motor Vehicles sites. Survey items included scales from the Fetzer Institute religiousness/spirituality instrument, Strayhorn and Weidman´s Parent Practices Scale, and the Emotional Problems subscale of the Strengths and Difficulties Questionnaire (SDQ-E). Of 235 returned surveys, 181 completed surveys were analyzed with multiple regression. This approach followed a multi-step process for assessing mediation (as per Baron & Kenny, 1986). Child age and a socially desirably response index were used as covariates.Regression of SDQ-E on Total Religiousness and Fundamentalism in separate models revealed no significant relationships. However, including Religiousness and Fundamentalism in the same model revealed significant prediction of SDQ-E, but with effects in opposite directions, suggesting a suppression effect. In this model, greater Fundamentalism was positively associated with elevated SDQ-E scores (F (1,108) = 6.533, p = .012, b = .25, R2partial = .047). Increased Total Religiousness, however, was negatively associated with elevated SDQ-E scores (F(1,108) = 5.09, p = .026, b = -.220, R2partial = .04). Although Parental Warmth was negatively associated with SDQ-E scores (F(1,102) = 4.52, p = .036, b = -.198, R2partial= .031), and positively associated with Total Religiousness (F(1,102) = 11.93, p = .001, b = .306, R2partial = .071), there was insufficient evidence to support a hypothesis that Parental Warmth mediates the relationship between Total Religiousness and Child Emotional Problems.These results indicate that different aspects of religion act as protective and risk factors for child mental health and parental warmth. This has potential implications for identification of risk, with overall religiousness acting as a resiliency factor and fundamentalism increasing risk for both parental warmth and child emotional problems. The differing outcomes of Total Religiousness and Fundamentalism indicate the need to further improve techniques of measurement to better understand the influence of religiosity.
CONCURRENT 5, EMERGING OPPORTUNITIES FOR PREVENTION RESEARCH, Organized symposia
SYMPOSIUM TITLE: COMPUTER BASED INCREDIBLE YEARS PARENT TRAINING FOR PARENTS OF HEAD START CHILDREN: CHALLENGES AND OUTCOMES
SYMPOSIUM TITLE: COMPUTER BASED INCREDIBLE YEARS PARENT TRAINING FOR PARENTS OF HEAD START CHILDREN: CHALLENGES AND OUTCOMES. Edward Feil1, Herb Severson2, Ted Taylor2, Shawn Boles2, Berry Broadbent2, Mike Eldridge2, Chris Widdop2, Ron Prinz3, 1University of Oregon, Eugene, OR United States; 2Oregon Research Institute, Eugene, OR United States; 3University of South Carolina, Columbia, SC United States
Researchers now distinguish two distinct forms of antisocial behavior patterns referred to as “early” versus “late” starters, or “life-course persistent” versus “adolescent limited” antisocial behavior. Evidence indicates that children who show antisocial forms of behavior early in their lives have severely elevated levels of risk for a host of negative developmental outcomes. Children from low SES homes have increased risk of developing antisocial behavior patterns. Mental health research has shown that behavioral parent training (BPT) is an empirically validated treatment for Oppositional Defiant and Conduct Disorders, but major obstacles to delivering such parent training have been both the lack of access to validated programs and barriers to participation, including the high cost of professional facilitators and travel by parents to attend group meetings.
This symposium will discuss the development and results from a randomized trial of an innovative Internet delivery system to provide BPT to parents of Head Start children. The interactive Internet-base project is based on an effective behavioral parenting program (Incredible Years, Webster-Stratton, 1984) that is designed to prevent the development of behavioral risk factors associated with conduct problems. Building on previous research showing the Incredible Years program is effective when offered in a self-administered format (Webster-Stratton, Kolpakoff, & Hollinsworth, 1988; Webster-Stratton, 1990, Webster-Stratton, 1992), we adapted the Incredible Years videos and self-administered manual to be available on computer. Participants were loaned computers for the duration of the project, and could view the program via the internet around the clock, as well as join in discussions online with other parents also taking the program. Additionally, a parent “coach” made five home visits throughout the program, as well as regular phone contact to encourage parents to continue with the program. The project evaluates the impact of the program with parents who have identified their child as exhibiting high risk behaviors.
Paper presentations will discuss the (a) development of the Internet-based intervention, including the digital translation of the empirically-validated program from analogue videotape to computer, (b) clinical implications for the use of the Internet, including supervision and tracking of participants and therapist activities and (c) design and results from the randomized trial. The discussant will review the presentations and implications for dissemination.
DEVELOPMENT OF USER-FRIENDLY TECHNOLOGIES FOR HOME-BASED INTERNET INTERVENTIONS. Edward Feil1, Mike Eldridge2, Ted Taylor2, Herb Severson1, 1University of Oregon, Eugene, OR United States; 2Oregon Research Institute, Eugene, OR United States
There are currently 162 million U.S. adults that have access to the Internet, with 123 million of those active Internet users (US Census, 2003). The continuing expansion of Internet use from the upper middle class and well-educated has moved toward more "mainstream" America until the typical user resembles an average American. Women continue to be the fastest growing segment on the Internet, now constituting 51% of the U.S. online population. Internet use is rapidly crossing key demographic thresholds. The US census bureau found that 40% of Internet users have not attended college as well as 40% having an annual household income below $50,000. In keeping with this trend, we found in our survey of Head Start families that six out of eight families already accessed the Internet through either home or community sites, such as community centers and libraries.
Computers and the Internet are being put to a variety of behavioral uses, including providing social support and information to breast cancer and AIDS patients, supporting home care givers of patients with Alzheimer's disease, providing expert systems for smoking cessation and supporting parenting skills. Currently on the Internet, people are participating in conferences, chat rooms, and related Internet services for parenting.
From our perspective, there are considerable barriers for parents to attend parent training classes (e.g., transportation, harsh weather, stigma because parents are required to attend “special” parenting classes, long distances or schedule conflicts, time commitments, etc). Limited reach is one of the biggest failings of behavioral parent training. We have found that computer and Internet-based communication can be a powerful adjunct to help increase participation and meeting goals through the flexibility that computer technology can afford. In the “Parent-Net” project, we developed dynamically-generated tabular and graphic computer displays as components in feedback loops that address intervention process control for clinical support staff and clinical supervisors. The program tracking feature allows the clinical staff to track where the user is in the program, as well as how often they have logged in or completed assessments that were requested of them.
This paper will present the development of an Internet-based intervention, including the digital translation of an empirically-validated program from analogue videotape to computer. Issues such as usability and video playback design considerations will be discussed.
ROLE OF REMOTE CLINICAL SUPERVISION WITH INTERNET-BASED INTERVENTIONS. Ted Taylor1, Herb Severson1, Edward Feil1, Berry Broadbent1, 1Oregon Research Institute, Eugene, OR United States
This presentation will review the important role played by these coaches in motivating and encouraging parents to watch the videos. Coaches also played an important role by meeting with families on five (or more) occasions to clarify any questions parents had, problem-solve difficulties, role play the use of the skills, and assist parents to commit to practice the skills learned with their children. In this presentation, the strategies used by coaches to motivate parents to participate are highlighted. Data involving the date and time coaches implemented strategies such as phone contacts, home visits, and letters to parents, will be used to demonstrate the temporal impact of specific strategies on parents use of the program. The vital role played by coaches in getting parents to complete the program will be highlighted. Similarly, the role of the supervisor in problem-solving difficulties as well as monitoring parent progress and coach behavior to motivate coaches to continue to engage parents will be highlighted. In this project, over 77% of the 90 Head Start families assigned to the intervention condition completed at least 5 of 9 topics (the equivalent of attending 7 of 12 group sessions), and over 63% of parents watched 100% of the program. When only the second cohort data is considered, the participation rates are even more impressive, with over 75% of parents watched 100% of the program, and over 82% watching at least 5 of 9 topics. These participation rates were achieved because of the dedicated work of the “parent coaches”. This illustrates that large numbers of low income, multi-stressed families can be engaged in a self-administered behavioral parent training program, but only with continued support and encouragement of a flexible parent coach.
INITIAL RESULTS FROM A RANDOMIZED TRIAL OF AN INTERACTIVE INTERNET INTERVENTION. Herb Severson1, Shawn Boles1, Ted Taylor1, Edward Feil2, Chris Widdop1, 1Oregon Research Institute, Eugene, OR United States; 2University of Oregon, Eugene, OR United States
While there have been a number of advances in the development of the Internet-based therapies, there has not been the same level of rigor in the evaluation of clinical results. A randomized control trial is the gold standard for effectiveness and is difficult to perform in an open Internet environment. In this project, we were fortunate to be able to provide computers and computer access to a low-income sample. We present results from a 2-arm 2-cohort treatment/control randomized controlled trial intended to test the effects of an Internet-based parenting intervention combined with home visits and phone consultation on the behavior of children. Four-year-old children attending Head Start classrooms in Oregon were screened via parent rating to identify those exhibiting the early signs of antisocial behavior. One-hundred and seventy-seven families (divided between 2 cohorts) of identified children consenting to participate were randomized to either (a) computer control or (b) computer/Internet parenting program.
Families randomized to the Control Condition/Usual Care received a computer system which includes the computer, monitor, modem, and Internet connection for 6 months. Some families received phone service. This condition was intended to specifically control for the potential confounding effect of the possible introduction of computer and Internet technology into the home of Head Start families. In addition, this condition allowed for the parallel collection of all computerized, Parent Weekly Report of Child Behavior.
Families randomized to the intervention condition received the computer, plus a computerized adaptation of the empirically supported self-administered Incredible Years parenting program. The computerized version of this program included all of the 10 self-administered sessions for the program. To supplement the parenting program, we provided for a parenting “coach” to visit parents in their homes, intended to engage parents in the process, train on computer skills, explain the procedures, answer any questions about the program. In addition, an electronic bulletin board was created to provide social support for parents facilitated by project staff.
This paper will present results from parent report on the Child and Adolescent Disruptive Behavior Inventory, Eyberg Child Behavior Inventory, Parent Weekly Report of Child Behavior, Walker-McConnel Social Skills inventory, Parenting Scale and Parenting Practices, Parental involvement in School, and depression. In addition, we will present results from independent observations in the home using the Dyadic Parent-Child Interactive Coding System Revised.
CONCURRENT 9, METHODS, Grouped papers
EVALUATION METHODOLOGIES: SAMPLE CHARACTERISTICS AND PROGRAM EFFECTIVENESS, SAMPLE SIZE AND POWER, AND MEDIATION ANALYSIS WITH BINARY OUTCOMES
Chair: George Howe 509
METHOD OF PROBING SIMPLE MEDIATION EFFECTS FOR MODELS WITH COMBINED MEDIATED AND MODERATED EFFECTS. Jenn-Yun Tein1, David Mackinnon1, Oi-Man Kwok1, 1Arizona State University, Tempe, AZ United States
The importance of examining mediating mechanisms has been noted by many prevention researchers (Botvin et al., 1992; Chen, 1990; Sandler, Wolchik et al., 1997). There have also been considerable developments in the appropriate analysis of mediating mechanisms (Baron & Kenny, 1986; MacKinnon & Dwyer, 1993; MacKinnon et al., 2002). Despite conceptual and methodological progress in the study of mediating processes, evaluations of mediation have not sufficiently dealt with the finding that prevention program effects are often moderated by participant characteristics, such as age, gender, or initial level of functioning (see Brown & Liao, 1999). The identification of subgroups who would benefit most from the intervention has important implications for the cost effectiveness of the program, recruitment of individuals into the intervention programs, and program redesign. Given the common findings of the mediation in the context of moderated prevention effects, it is important to develop temporal models and methods to examine the combined mediated and moderated effects in evaluations of prevention programs, specifically, to identify for which subgroup or under what condition the program has an effect. Thus, statistical probing of the simple effect becomes important. Under the mediation model framework, simple mediation effects represent the theoretical causal relations of the program, mediator, and outcome for each level of the moderator variable. Group comparison such as using multisample structural equation modeling is commonly applied to examine simple mediation effects for categorical moderator variables (e.g., ethnicity, gender). However, the methods for testing simple mediation effects for continuous moderator variables are underdeveloped. Tein, Sandler, MacKinnon, & Wolchik (2004) proposed a method of probing simple effects for the model where the strength or the direction of the relation between the program and the mediator is significantly affected by the moderator variable. However, it is likely that certain participant characteristics simultaneously affect the relations between the program and the mediator as well as the mediator to the outcome. In this study, we will demonstrate the method of examining the mediation model when a continuous moderator variable not only affects the relation between the program and the mediator but the relation between the mediator and the outcomes. Specifically, we will demonstrate the method of probing simple mediation effects.
SAMPLE SIZE AND STATISTICAL POWER IN TESTS OF MEDIATION. Matthew Fritz1, David Mackinnon1, 1Arizona State University, Tempe, AZ United States
Mediation models are widely used in prevention research and there are many ways to test for the mediated effect, including single sample (MacKinnon, Lockwood, Hoffman, West, & Sheets, 2002) and resampling tests (MacKinnon, Lockwood, and Williams, 2004). The most common method of testing for mediation is the causal steps method (Baron & Kenny, 1986). However, concern has been raised about the statistical power of the causal steps test. MacKinnon et al. (2002) found that the Baron and Kenny test of causal steps has low overall power compared to other mediation tests. Given the low sample sizes that can occur in prevention studies, this is of great concern. Although studies have investigated Type I error rates and statistical power of mediation tests, prior research has not produced sample size requirements useful for applied researchers. The current study investigates the sample size necessary to achieve 0.8 power for four single sample tests of mediation: Baron and Kenny (1986) causal steps test, MacKinnon and Lockwood (2001) asymmetric distribution of the product test, Sobel (1982) first order test, and Aroian (1947) second order test. Also investigated are two resampling tests of mediation, the percentile bootstrap and the bias corrected bootstrap. Sample size requirements were investigated empirically using computer simulations in which data of a specific sample size was generated using predetermined parameter sizes and then tested for mediation using each of the tests. This was repeated a large number of times and then power was defined as the percentage of times a mediated effect was found. All combinations of small, medium, and large effects were used, including an additional condition testing complete mediation. Results from the study show that the bias corrected bootstrap had the most statistical power, followed by the asymmetric distribution of the product test, the percentile bootstrap, the first order test, the second order test, and the Baron and Kenny (1986) causal steps test, respectively. For example, in the small-small completely mediated condition, the Baron and Kenny test required a sample size of 20,886 for 0.8 power, compared to a sample size of 509 for the asymmetric distribution test. The results also show that models where the effect of the mediator on the dependent variable is smaller than the effect of the independent variable on the mediator have lower power than models where the size of the effects are reversed or where the effects are equal. Implications and future research is discussed.
MEDIATION ANALYSIS WITH BINARY OUTCOMES. Chondra Lockwood1, David Mackinnon1, Hendricks Brown2, Jeanne Hoffman3, 1Arizona State University, Tempe, AZ United States; 2University of South Florida, Tampa, FL United States; 3University of Washington, Seattle, WA United States
Mediation analysis is gaining importance in the evaluation of prevention programs, providing an exploration of the causal process of program effectiveness. Mediation analysis is the statistical testing of models in which a program effect is transmitted through one or more intermediate variables, the mediators. It is common for the dependent variable of interest in prevention studies to be binary (e.g., drug use/non-use, presence/absence of disease), in which case researchers will use logistic or probit regression in the analysis. The properties of logistic and probit regression have problematic consequences for the most widely used methods in mediation analysis, leading to distorted estimates of the mediated effect and potentially to incorrect conclusions.
Using the well-known relationship between latent variable modeling and logistic or probit regression, we provide a solution to the problem with conducting mediation analysis with a binary outcome. We provide a simple method to adjust logistic regression estimates and eliminate the distortion. The problem and solution are demonstrated with simulation results and data from the Midwestern Prevention Project, a school- and community-based drug prevention program.