1 PREDICTORS OF PATTERNS OF CHANGE IN CHILD DISRUPTIVE BEHAVIOR AND PARENTING STRESS DURING PARENT CH ILD INTERACTION THERAPY AND ITS RELATION TO TREATMENT OUTCOME By JAIMEE CHRISTINA PEREZ A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2008
2 2008 Jaimee Christina Perez
3 To my daughter, Sophia, who inspires me to be au thentic and true in all my pursuits. I love you beyond all measure and hope you one day take pride in the legacy I now leave you with the completion of this dissertation.
4 ACKNOWLEDGMENTS I first want to acknowled ge my parents, whos e unfailing faith in me pushed me to finish this race. I also want to thank my dissert ation committee: Drs. Eyberg, Johnson, Robinson, and Graber. I am ever thankful for their patience, support, and encouragement. I want to thank my fellow labmates of the Child Study Lab. Their co mpanionship and peer support were invaluable to me. I feel fortunate to have worked with each of them. I also thank my friends. Their unswerving support as I pushed to br ing closure to this chapter in my life was my lifeline. I thank my supervisors and colleagues at the Treasure Coast Early Steps Program for understanding the importance of the dissertation a nd allowing me to use my vacation time to see this dissertation to fruition. God set me on this pa th many years ago to obtain this degree so that I may serve His will and purpose for my life. I praise His Holy Name. May this work glorify Him.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4 LIST OF TABLES ...........................................................................................................................8 LIST OF FIGURES .......................................................................................................................10 ABSTRACT ...................................................................................................................... .............11 1 INTRODUCTION .................................................................................................................. 13 Brief Overview .......................................................................................................................13 Disruptive Behavior Disorders ............................................................................................... 19 Risk Factors for the Development of Disruptive Behaviors in Children ................................ 21 Child Risk Factors ...........................................................................................................23 Parent Risk Factors .......................................................................................................... 25 Family Risk Factors .........................................................................................................28 Cumulative Risk Factor Models ...................................................................................... 29 Predictors of Therapeutic Change and Premature Termination from Treatment ................... 32 Child Factors ...................................................................................................................33 Parent Factors ................................................................................................................ ..35 Family Factors .................................................................................................................39 Summary ....................................................................................................................... ...41 Parent Child Interaction Therapy ............................................................................................ 42 Hypotheses .................................................................................................................... ..........53 Patterns of Change in Child Disrupti ve Behavior and Parenting Stress ......................... 56 Predicting Treatment Outcome from Patte rns of Change in Child Disruptive Behavior and Parenting Stress ..................................................................................... 57 Predicting Early versus Late Dropout in PCIT from Patterns of Change ....................... 57 2 METHOD ........................................................................................................................ .......59 Participants .................................................................................................................. ...........59 Measures ...................................................................................................................... ...........60 Demographic Questionnaires ..........................................................................................60 Child Cognitive Status .....................................................................................................60 Parent Cognitive Status ................................................................................................... 60 Outcome Variables .................................................................................................................61 Patterns of Change ...........................................................................................................61 Change in Parent and Child Outcomes ............................................................................ 62 Predictors of Patterns of Change ............................................................................................65 Child Factors ...................................................................................................................65 Parent Factors ................................................................................................................ ..66 Family Factors .................................................................................................................66 Procedures .................................................................................................................... ...........67
6 Analyses ...................................................................................................................... ............69 Descriptive Statistics ....................................................................................................... 69 Reliability ................................................................................................................... .....70 Analysis of Change ..........................................................................................................70 Dealing with the Unbalanced Design ..............................................................................70 Choosing a Model ............................................................................................................72 The Random Regression Coefficient Model ...................................................................73 The level 1 model ..................................................................................................... 74 Modeling predictors .................................................................................................75 Modeling maternal depression as a tim e varying covariate ..................................... 76 Predicting Change in Treatment Outcome and Dropout from Patterns of Change in Child Disruptive Behavior and Parenting Stress .........................................................77 3 RESULTS ....................................................................................................................... ........78 Descriptive Analyses .......................................................................................................... ....78 Patterns of Change Analyses ..................................................................................................80 Predictor Models .............................................................................................................83 Treatment Phase Model ................................................................................................... 84 Hypothesis 1 ............................................................................................................. 85 Hypothesis 2 ............................................................................................................. 87 Hypothesis 4. ............................................................................................................ 88 Hypothesis 5. ............................................................................................................ 88 Hypothesis 6. ............................................................................................................ 90 Hypothesis 7. ............................................................................................................ 91 Hypothesis 3. ............................................................................................................ 93 Predicting Treatment Outcome from Patterns of Change ....................................................... 96 Hypothesis 1. ............................................................................................................ 98 Hypothesis 3 ............................................................................................................. 99 Traditional Linear Regression Analyses ....................................................................... 100 Predicting Dropout from Patterns of Change ................................................................ 103 4 DISCUSSION .................................................................................................................... ...139 Predictors of Patterns of Change ..........................................................................................139 Maternal Depression ......................................................................................................141 Family Predictors ...........................................................................................................144 Gender ........................................................................................................................ ...146 Predicting Treatment Outcomes from Patterns of Change in Child Disruptive Behavior and Parenting Stress ..........................................................................................................147 Limitations and Challenges of the Current Study ................................................................. 150 Future Directions ..................................................................................................................155 TABLES OF NON-SIGNI FICANT FINDINGS ......................................................................... 158 LIST OF REFERENCES .............................................................................................................164
7 BIOGRAPHICAL SKETCH .......................................................................................................174
8 LIST OF TABLES Table page 1-1 DSM-IV-TR Criteria for ADHD .......................................................................................58 3-1 Sample Characteristics .................................................................................................... .105 3-2 Correlations for Predicto r and Outcom e Variables .......................................................... 106 3-3 Unconditional Means Models for the ECBI .................................................................... 107 3-4 Unconditional Means Models for the PSI Total and Subscales ....................................... 107 3-5 Unconditional Growth Model for the ECBI and PSI Scales ............................................ 108 3-6 Phase of Treatment Predictor Model ............................................................................... 109 3-7 Effect of Diagnosis of ADHD and Tr eatm ent Phase on ECBI and PSI DC .................... 111 3-8 Effect of Diagnosis of ADHD on Total, PD and PCDI PSI Scales .................................112 3-9 Effect of Attachment Q sort and Phase of Treatm ent on Child Disruptive Behavior ...... 114 3-10 Effects of Gender on Change in PCIT .............................................................................115 3-11 Barriers to Participation in Treatm ent Change Model .....................................................117 3-12 Effect of SES on Change in Childrens Disruptive Behavior ..........................................119 3-13 Effect of Perceived Barriers to Partic ipation in Treatm ent on Change in Parenting Stress ........................................................................................................................ ........121 3-14 Effect of Maternal Depression in Childrens Disruptive Behavior ................................. 126 3-15 Effect of Maternal Depression on Parenting Stress .........................................................129 3-16 Unconditional Growth Models with Reverse Time Code ................................................ 133 3-17 Effect of Change in Qsort on Patterns of Change in Parent Child Dysfunctional Interaction ................................................................................................................... .....134 3-18 Relation between Patterns of Change in Child D isruptive Behavior and Change in Attachment Security from Preto Post-treatment ............................................................ 136 3-19 Predicting Change in Child Disruptive Behavior f rom Change in Attachment ............... 138 3-20 Predicting Change in Parent Child Dysf unctional Interaction from Change in Child Disruptive Behavior ......................................................................................................... 138
9 3-21 Predicting Change in Maternal Perceived Social S upport from Change in Parent Distress ...................................................................................................................... .......138 3-22 Logistic Regression Model Predicting Early vs. Late Dr opout from Change in Child Disruptive Behavior and Parent ing Stress in CDI and PDI ............................................. 138 A-1 Patterns of Change in Child Disruptive Behavior and Parent ing Stress by Treatm ent Phase ......................................................................................................................... .......158 A-2 Relation between Patterns of Change in Child D isruptive Behavior and Change in Parenting Stress (PDH) from Preto Post-Treatment ...................................................... 159 A-3 Relation between Patterns of Change in Parent-Child Dysfunctional Interaction and Change in Attachm ent Security from Preto Post-treatment .......................................... 160 A-4 Relation between Patterns of Change in Parenting Stress and Change in Child Disruptive Behavior from Pr eto Post-treatm ent ............................................................161 A-5 Relation between Pattern of Change in Difficult Child and Change in Perceived Social Support from Preto Posts-treatment ................................................................... 162 A-6 Patterns of Change in Child Disruptive Behavior and Parent ing Stress by Treatm ent Phase ......................................................................................................................... .......163
10 LIST OF FIGURES Figure page 3-1 Patterns of Change in CDI and PDI ................................................................................. 110 3-2 Patterns of Change by ADHD Diagnosis .........................................................................113 3-3 Patterns of Change by Gender. ........................................................................................116 3-4 Predicting Change in Child Disruptive Behavior f rom Mothers Ratings of Barriers to Participation in Treatment ...........................................................................................118 3-5 Initial Status and Rate of Change in Child Disruptive Behavior by Socioeconom ic Status. ...............................................................................................................................120 3-6 Patterns of Change in Parenting Stress by Levels of Barriers to Participation in Treatm ent.. ................................................................................................................... ....122 3-7 Patterns of Change in Child Disruptive Behavior by Change in Maternal Depression during PCIT .....................................................................................................................127 3-8 Patterns of Change in Parenting Stress Related to Difficult Child by Changes in Maternal Depression during PCIT ...................................................................................131 3-9 Effect of Change in Attachment Security from Preto Post-treatment on Patterns of Change in Child Disruptive Behavior during PCIT ......................................................... 135 3-10 Effect of Treatment Phase and Change in Attachm ent Security from Preto Posttreatment on Patterns of Change in Pa rent Child Dysfuncti onal Interaction ................... 137
11 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy PREDICTORS OF PATTERNS OF CHANGE IN CHILD DISRUPTIVE BEHAVIOR AND PARENTING STRESS DURING PCIT AND IT S RELATION TO TREATMENT OUTCOME By Jaimee Christina Perez August 2008 Chair: Sheila Eyberg Major: Psychology Our main aims were to examine predictors of patterns of change in child disruptive behavior and parenting stress dur ing PCIT and to examine patterns of change as predictors of specific parent and child treatment outcomes. Participants were 100 boys and girls who met diagnostic criteria for Oppositional Defiant Disorder and their parents. Fa milies received Parent Child Interaction Therapy, an intervention specifi cally designed to target disruptive behavior problems in preschoolers, ages 3 to 6 years. Session-by-session data were collected on child disruptive behavior and parenting stress. These data were examined using a multilevel modeling approach to examine treatment, child, parent, and family predictors of patterns of change during PCIT. Patterns of change found in the multileve l model analyses were then examined by two different statistical approaches, as predicto rs of specific parent and child outcomes. Results from the first set multilevel modeling analyses found that treatment phase (CDI versus PDI), maternal depression, and perceived barriers to partic ipation in treatment predicted different patterns of change in child disruptive behavior and parenting stress. Families exhibited faster change in child disruptive behavior and parenting stress in CDI than PDI. Patterns of change in child disruptive behavior and parent ing stress were associated with changes in maternal depression from preto midto post-tr eatment, with mothers with moderate to severe
12 levels of depression during PCIT exhibiting slower change in levels of parenting stress during PCIT and children of mothers with moderate to severe levels of depression demonstrating less change in disruptive behavior during PCIT. For treatment completers, retrospective ratings of perceived barriers to participati on in treatment predicted a slower pattern of change in child disruptive behavior a nd parenting stress. Patterns of change were not found to predict to specific parent and child outcomes. Multilevel analyses examining change in outcomes as predictors of patterns change found a trend between patterns of change in child disruptive beha vior and change in attachment from preto post-treatment. Traditional regression analyses found a significant relationship between change in child disruptive behavior during CDI and change in attachment from pr eto post-treatment, with declines in child disruptive behavior predicting greater change in attachment from preto post-treatment. Changes in parent child dysf unctional interaction duri ng CDI and PDI predicted change in child disruptive behavior from preto post-treatment. Inspec tion of mean change in CDI revealed minimal improvement in parent child dysfunctional intera ction, which predicted minimal change in child disruptive behavior duri ng CDI. Change in parent child dysfunctional interaction in PDI, however, predicted improvement in child disruptive be havior from preto post-treatment. Patterns of change in child disruptive behavior during CDI and PDI did not predict to early versus late dropou t. Traditional logistic regressi on analyses examining change in child disruptive behavior in CDI and PDI also fa iled to predict early versus late dropout from PCIT.
13 CHAPTER 1 INTRODUCTION Brief Overview In recent years, research ers have provided evidence for the stability of conduct problems in young children (Hinshaw, Lahey, & Hart, 1993; Loeber, 1988, 1991; Moffitt, 1993). For example, in a review of longitudinal studies of hard-to-manage preschoolers, Campbell (1991) found that at least 50% of preschool-age chil dren with moderate-to-severe externalizing problems continued to show some degree of di sturbance at school age, even though they had been in treatment as preschoolers. Sixty-seven pe rcent of those children met diagnostic criteria for Attention Deficit Hyperactivity Disorder (ADHD), Oppositional Defiant Disorder (ODD) or Conduct Disorder (CD) by age 9. Developmental studies of the emergence of aggression in young children (Moffitt, Caspi, Dickson, Silva, & Stanton, 1996; Nagin & Tremblay, 1999; Shaw, Owens, Vondra, Keenan, & Winslow, 1996) ha ve also provided evidence for the stability of disruptive behaviors from early childhood into adolescence and adulthood. For example, Moffitt et al. (1996) reported that 7% of thei r community sample of boys from New Zealand, assessed from age 3 years until age 18 years, fo llowed a chronic and persistent pathway, in which they exhibited high levels aggressive and oppositional behaviors from preschool age up through late adolescence. This fi gure is similar to results from other longitudinal studies of the onset and development of disruptive behavior in children. For example, Nagin and Tremblay (1999) found that 5% of their French Canadian school sample of boys, ages 6 to 15 years, followed a similar course of high levels of aggressive and oppositional behaviors. As noted by Campbell, Shaw, and Gilliom (2000), these figures s uggest that, despite differences in the nature (clinical, high risk, or populat ion based) and characteristics of these samples (country of origin,
14 informant), a certain proportion of preschoolers a nd early school-age children is at risk for progressing along a persistent path way that leads to significant c onduct problems later in life. Developmental psychopathology researchers have worked to uncover the developmental pathways that lead to long-term antisocial conduct disordered be havior (Broidy, Nagin, Tremblay, Bates, Brame, Dodge, Fergusson, Horwood, Loeber, Laird, Lynman, Moffitt, & Pettit, 2003; Hinshaw, Lahey, & Hart, 1993; Loeber, 1988, 1991; Moffitt, 1993). Across various studies, similar developmental pathways have been reported. A certain percentage of children from these studies follow an early starter (Patterson, Capaldi, & Bank, 1991) or life-course persistent pathway (Moffitt et al., 1996), which is characterized by the early onset of aggression and other disruptive behaviors in the preschool or early school age years that persists into adolescence and adulthood. Another percentage of children in these samples follow a high level desister pathway, which is also characterized by early onset of aggression and other disruptive behaviors that decrease to normative levels as th e child ages into school age and adolescence. The remaining children in these samples follow a low level desister pa thway or a no problem pathway, characterized by low levels of aggression and other problem behavi ors that decrease as the child ages. Recent developmental trajectory studies (C t, Tremblay, Nagin, Zoccolillo, & Vitaro, 2002; Shaw, Gilliom, Ingoldsby, & Nagin, 2003; Spie ker, Larson, Lewis, Keller & Gilchirst, 1999) have attempted to address the question of which children will follow an early starter, persistent course and which children will eventually desist and follow a more normative pathway. The evidence from longitudinal studies of childrens disruptive be havior indicates that, for certain preschool and early school age childr en, disruptive behaviors are stable phenomena that can be detected at an early age (Campbell et al., 2000; Spieker et al., 1999; Owens and
15 Shaw, 2003). Researchers have attempted to elucidate which child, parent, and family factors put children at risk for the early starter pathway and which factor s may protect children from the early starter persistent pathway. Considering that these child ren persist and progress toward worse outcomes, such as poor academic func tioning, further conduct problems, increased likelihood for school dropout, and s ubstance abuse (Coie & Dodge, 1998), early intervention that specifically targets disruptive behavior in pr eschool-age children is clearly warranted. In response to findings from the child deve lopment literature on disruptive behavior, clinical researchers have deve loped various interventions to treat these problems and have extensively examined the effectiveness and effic acy of these interventions. They have also examined predictors that negatively affect tr eatment outcome in interventions specifically designed to target oppositional and aggressive behaviors in young ch ildren (Eyberg, 1992; Kazdin, 1996; Patterson, 1982; Webster-Stratton & Hammond, 1997). Similar categories of risk factors, which will be outlined shortly, have been shown to predict both less response to treatment (Dumas & Wahler 1983; Kazdin & Wassell, 2000; Webster-Stratton & Hammond, 1990) and attrition (Armbrus ter & Kazdin, 1994; Gould, Shaffer, & Kaplan, 1985). Though treatments in the child disruptive behavi or intervention literature have varied with respect to format and structure, their main goal is to alter the behavior such that the child engages in more prosocial, acceptable behaviors. Earl y research investigating the correlates and predictors of oppositional and aggressive behaviors in children revealed strong relations between parenting style and negative child outcomes (Azar & Wolfe, 1989; Baumrind, 1967, 1991; Franz, McClelland, & Weinberger, 1991; Olson, Bates, & Bayles, 1990). For example, Baumrinds (1967, 1991) research on parenting styles has sh own that parents who do not adequately meet young childrens dual needs for nurturance and lim its are less likely to have successful and
16 healthy adolescents. Additionally, researchers f ound that quality of the interactions between parent and child play an influential role in child behavior outcomes. Coercive interactions, in which the parent and child engage in increasing ly manipulative behaviors to gain power in the relationship, have been specifi cally identified as problematic and contributory to the later development of aggressive, oppositional and conduc t-disordered behavior in children (Bates, Bayles, Bennett, Ridge, & Brown, 1991; Campbe ll, 1991; Patterson, 1982; Patterson, Reid, & Dishion, 1992). These findings highlight the need to target th e interactions between parent and child to effect change in a childs maladaptive behaviors. The treatment format and structure of interest for this study is a parent-child interaction approach that targets the interaction patterns between parent and child (Foote, Eyberg, & Schuhma nn, 1998) and focuses on teaching the parent effective strategies to enhance the parent-child relationship and improve parent limit setting and consistency with discipline. A more detailed description of the intervention (Parent Child Interaction Therapy) is provided below. An extensive literature now exists on the mu ltiple factors associated with worse outcomes in children with disruptive be havior in both the developmental and clinical literature. Three domains of risk have been implicated as predicto rs that increase the lik elihood that a child will progress along an early starter pa thway: (a) child risk factors (, difficult temperament, high rates of impulsive, inattentive, hyperactive or aggressive behaviors, child sex); (b) parent risk factors (maternal depression, maternal stress, ineffective parenting strategies and negative attitudes); and (c) family risk factors (marital factors, socio economic factors, and ot her stressors; WebsterStratton, 1996). Though no definitive model has emerge d that is able to account adequately for how each of these factors works toward predic ting long-term developmental and clinical
17 outcomes, results of studies from both literatures have consistently linked risk factors in these three domains to risk for the development of disruptive behaviors in children and to less therapeutic change in childrens disruptive behaviors during treatm ent (Campbell, Shaw & Gilliom, 2000; Kazdin & Wassell, 1998; Owens & Shaw, 2003). Recently, researchers in the de velopmental and clinical lite rature have turned their attention toward more advanced statistical methods to measuring change in child disruptive behavior over time (Broidy et al., 2003; Hartman, Stage, & Webster-Stratton, 2003; Owens & Shaw, 2003). Studies examining change in individu als or groups across time employ longitudinal designs with repeated, time-ordered observati ons (Wu, Clopper, & Wooldridge, 1999). Pre/post designs are an example, in which the dependent va riable or outcome of interest is measured at some initial time point and then measured again either after the completion of treatment or at some other specified time point. Traditional appro aches to analyzing these types of data include univariate and multivariate analysis of variance as well as traditional regression approaches. Difference scores have also been employe d as a measure of change over time. In an article comparing tradi tional approaches to analyzing longitudinal data to multilevel modeling, Wu and colleagues (199 9) outlined the following id eal goals for analysis of longitudinal data: (a) direct st udy of (intra-) individual change (b) direct id entification of interindividual differences in intraindividual ch ange, (c) analysis of the relationship between intraand interindividu al changes, and (d) study of the va riables that influence intraand interindividual change. Methodological resear chers such as Rogosa (1995) and Willet (1988, 1994) have argued that traditiona l statistical approaches do not adequately address these goals, particularly in understanding intraindividual ch ange over time. Wu and colleagues also noted that ANOVA with repeated measures almost ex clusively focuses on between-participants or
18 interindividual effects. Instead, Rogosa and others (Raudenbush & Bryk, 2002; Willet, 1988, 1994) encourage researchers to use multileve l modeling to analyze change over time. Multilevel model analysis is also referred to as hierarchical linear modeling (HLM). These terms refer to a set of statistical models that an alyze change at multiple levels (, individual level and group or organizational level) and are able to handle repeated measures. The basic level in any multilevel model is the level 1 model which specifies change over time within the individual. The level 2 model specifies change over time at the group level and can include independent variables hypothesized to predict differences at both the betweenand withinparticipants levels. In the c ontext of analyzing individual change over time, observations are nested within individuals. Therefore, individu als are considered to be the level-2 grouping variable. The level 1 and level 2 models can be combined to specify change at both the individual and group levels simultaneously. Unlike trad itional ANOVA approaches, HLM does not assume that individuals are changing at the same level or the same rate over time. Therefore, it is possible to examine individual differences in the level and rate of cha nge at both the withinparticipant and between-participant levels. Additionally, HLM procedur es do not require that the time between assessment points be equally spaced apart or that each participant have an equal number of assessment points. Hence, all data points can be included in the analyses. Simulated data models have demonstrated that in clusion of all cases regardless of missing data points actually results in estimates closer to the expected values had all pa rticipants had complete data (Zeitman-Zait and Zumbo, 2005). The main aim of this study is to examine pred ictors of individual change in child disruptive behaviors and parenting stress during treatment in families referred for Parent Child Interaction
19 Therapy (PCIT; Brinkmeyer & Eyberg, 2003). Children in this study were diagnosed with Oppositional Defiant Disorder (ODD; American Psychiatric Associ ation (APA), 2000), a common childhood disruptive behavior diso rder. This study proposes to apply multilevel (or hierarchical linear) modeling to measure patterns of change in childrens disruptive behaviors and parenting stress during the course of treatment. In addition to modeling patterns of change in child disruptive behavior and pa renting stress, three domains of predictors (child, parent, and family) that have previously b een shown to negatively affect change in treatment will be examined to determine whether they predict diff erent patterns of change in child disruptive behavior and parenting stress dur ing PCIT. The patterns of cha nge in child disruptive behavior and parenting stress will then be examined as predictors of change in specific treatment outcomes, including dropout from PCIT. In the following sections, a more in-depth review is provided on the effects of child, parent, and family predictors on the developmen t of disruptive behavior disorders in children, and on the patterns of change in child disruptive behavior and parenting stress during treatment. Criteria for diagnosing a disruptiv e behavior disorder will also be reviewed, as well as the theoretical underpinnings and outcome research on the effect iveness of PCIT in treating disruptive behavior in children. We examine methodological considerations in multilevel modeling before discussing speci fic hypotheses for this study. Disruptive Behavior Disorders The catego ry of disruptive behavior disorders includes Oppositional Defiant Disorder (ODD), Conduct Disorder (CD), and Attenti on Deficit Hyperactivity Disorder (ADHD). According to the Diagnostic and Statistical Ma nual of Mental Disorders Fourth Edition, Text Revision (DSM-IV-TR, APA, 2000), the essentia l feature of Oppositiona l Defiant Disorder (ODD) is a recurrent pattern of negativistic, defiant, disobedien t, and hostile behavior toward
20 authority figures that [has] persisted for at l east 6 months (p. 100). Child ren who meet criteria for ODD present with at least four of the following behaviors: loses temper, argues with adults, actively defies or refuses to comply with adult s requests or rules, delib erately does things to annoy others, blames others for his or her own mistakes or misbehavior, is touchy or easily annoyed by others, is angry or resentful, or is spiteful or vindictive. Children with ODD engage in these behaviors more frequently than is typical of same-aged peer s, and these behaviors contribute to significant impairment in child f unctioning in social and academic settings. The DSM-IV-TR reports prevalence rates between 2% to 16% for ODD. Children diagnosed with Conduct Disorder (CD) engage in a repetitive and persistent pattern of behavior in which the basic rights of others or major age-appropriate norms or rules are violated (p. 93; APA, 2000). These behaviors are classified by the following four groups: aggressive conduct that causes or threatens physical ha rm to people or animals, non-aggressive conduct that causes property loss or damage, deceitf ulness and theft, and serious violations of rules. To meet criteria for a dia gnosis of CD, three or more of the characteristic behaviors must have occurred during the past 12 months, with one behavior present in the past 6 months. Children with CD often initiate aggressive behavi or and react aggressively to others, and they may display bullying or threatening behavior. Th ey may also use a weapon, such as a brick, bottle, knife, or gun, to cause harm and may be physically cruel to anim als. Deliberate fire setting with the intention of destr oying property may also be present. This study will focus on childhood-onset CD, in which at least one criterion characteristic of CD begins before the age of 10 (APA, 2000). The prevalence of CD ranges widely, from 1% to more than 10%, with higher prevalence rates among males (6% to 16%) than females (2% to 9%; APA, 1994).
21 Children with ODD or CD often present with co-morbid Attention Deficit Hyperactivity Disorder (ADHD). Estimates of co-occurrence range widely depending on the sample and assessment methods. In a review of the literatu re, Jensen, Martin, and Cantwell (1997) reported a range from 42.7% to 93%. Though these estimates were derived mainly from clinical samples, Jensen et al. (1997) noted that the available data suggest a rela tively high rate of comorbidity between ADHD and ODD/CD. Alone, prevalence rates for ADHD also vary considerably, depending on the age and nature of the sample. The DSM-IV reports estimates between 3% and 5% for school-age children (APA, 2000). For pr eschool children, prevalence rates range from 2% (Lavigne et al., 1996) to as high as 59% (Conners, 2002). Children meeting diagnostic criteria for ADHD present with a persistent pattern of inattention and/or hyperactivityimpulsivity that is more frequently displayed and more severe than is typically observed in sameage peers (APA, 2000). According to the DSM, a child must meet six or more of the inattentive and/or hyperactive-impulsive symptoms, and these symptoms must have persisted for at least 6 months at a maladaptive and inappropriate deve lopmental level to warrant a diagnosis of ADHD (see Table 1-1). Depending on his or her presentation, a child can be predom inantly inattentive, predominantly hyperactive-impulsive, or combined type. Risk Factors for the Development of Disruptive Behaviors in Children A review of the literatu re revealed that most applications of multilevel modeling have been in the developmental psychopathology literature. Specifically, researchers interested in the development of disruptive behavior have used multilevel modeling to map the developmental trajectories of these behaviors in children (Broidy et al., 200 3; Owens & Shaw, 2003; Spieker, Larson, Lewis, Keller, & Gilchrist, 1999). As mentioned above, similar developmental trajectories have been found acro ss multiple samples (Broidy et al., 2003), particularly an early starter pathway first described by Patterson (Patterson, Capaldi, & Bank, 1991) and Moffitt
22 (1993), in which a child exhibits high levels of di sruptive behaviors from an early age that persist into adolescence and adulthood. Recently, Broidy and colleagues conducted a six-site, crossnational study to examine the de velopmental course of physical aggression in childhood and its association with violent offending in adolesce nce. They found remarkable similarity in developmental trajectories across countries, with a sma ll percentage of school-age boys and girls from each sample (4% to 11%) engaging in consistently high levels of physical aggression over time. The description of this persistent, high-leve l trajectory appeared consistent with the early starter pathway (Patterson, Capa ldi, & Bank, 1991) described above Essentially, the persistent, high-level trajectory in the st udy was characterized by the early onset of disruptive behavior, specifically aggressive behavior (approximately around age 6), wh ich persisted in to adolescence and carried the most negative long-term prognosis in terms of predicting later violent offending. Although Broidy et al. (2003) demonstrated the presence of an early starter pathway in their samples, the ages at which most of the sites began assessing problem behaviors was age 6. Findings from studies with preschool child ren (Campbell, 1995) suggest that disruptive behaviors that begin before school entry are predictive of disruptiv e behavior in the school years. Therefore, researchers have also examined the trajectories of preschool age children, citing this age range as critical for the development of later conduct problems. Evidence from studies examining the developmental trajectories of di sruptive behaviors in preschool-age children and predictors that effect the deve lopment and maintenance of disr uptive behavior problems into school age are discussed below. Similar trajectories have been found in presc hool-age samples as those found in school-age samples. Shaw, Gilliom, Ingoldsby, and Nagin (2003) examined the trajectories of 284 lowincome boys ages 2 to 8 and identified four de velopmental pathways similar to those found in
23 research with older children: a persistent prob lem pathway; a high-level desister pathway; a moderate-level desister pathway; and a low-level problem pathway. The persistent problem pathway closely relates to Pattersons early starte r pathway. Six percent of the sample from the Shaw et al. (2003) study followed the persiste nt problem pathway. These boys exhibited relatively severe and persistent behavior problems from age 2 to age 8. Thirty-eight percent of the sample fell within the highlevel desister group. This path way was characterized by a higher level of disruptive behavior compared to the firs t two groups that showed a steady decline to the end of the study. The rest of the sample followe d the low and moderate le vel desister pathways, in which problem behaviors resolved on their own. What factors differentiate the toddler who w ill outgrow the terrible twos, as the children who follow the high-level desister pathway do, and the child who will persist with development, as the early starters do? Shaw et al. (2003) and others (Owens and Shaw, 2003; Spieker, Larson, Lewis, Keller, and Gilchrist, 1999; Munson, Mc Mahon, and Spieker, 2001; DeVito and Hopkins, 2001; Moffit and Caspi, 2001) have examined predic tors from three domains and their relation to the onset and development of disruptive behavior in preschool children. The first domain includes characteristics of the child, such as ch ild temperament, child gender, and hyperactivity. The second domain includes parent characteristics, such as pa rent stress and psychopathology. The third domain includes family characteristic s, such as socioeconomic status and family structure (single versus two-parent family). Child Risk Factors Child characteristics, such as tem perament a nd hyperactivity, have been linked to the onset of disruptive behavior and are predictive of poor long-term outcomes (McMahon, 1994). Difficult temperament in infancy, in combination with maternal perception of child difficulty,
24 male gender, prematurity, and low socioeconomic status, best predicted disruptive behavior during the preschool period (Sanson, Ob erklaid, Pedlow, & Prior, 1991). Hyperactivity has been found to predict futu re development of disruptive behaviors. Hinshaw, Lahey, and Hart (1993) found that children who display bot h hyperactivity and disruptive behavior exhibit more severe leve ls of disruptive behavi or and have a poorer prognosis than children with ei ther problem alone. In a follo w-up study of hyperactive children ages 4 to 12 years, Barkley, Fischer, Edelbroc k, and Smallish (1990) found that children with cooccurring conduct problems had more severe and persistent problems in adolescence and adulthood. Other researchers (Loeber, 1988; Mo ffitt, 1993b) have also found associations between hyperactivity and poor impulse control with more severe and chronic forms of disruptive behavior. Gender has also been examined as a risk f actor for development of disruptive behavior. Webster-Stratton (1996) noted that despite findings regarding the stability of behavior problems in preschoolers, little is known about developmental pathways for disruptive behaviors in terms of gender. The body of literature on gender differences has shown that boys are at higher risk than girls for developing disruptive behavior; ho wever, few studies exist have examined girls developmental pathways and whether similar risk factors are involved in the development of disruptive behaviors in girls as in boys. In a community sample, Moffitt and Caspi (2001) found the male-to-female ratio of childhood onset antisocial behavior was 10:1, indicating that fewer females in their sample followed the childhood-onset pathway; however, gi rls with childhood-onset antisocial behavior had similar high-risk backgrounds as childhoodonset boys. In a serv ice utilizing sample, McKabe and colleagues (2004) also found that fe wer girls than boys met criteria for childhood
25 onset conduct disorder, but their re sults revealed similar risk fact ors, including family history of mental illness, family history of antisocial beha vior, and history of maltreatment. Broidy et al. (2003) also examined developmental trajectorie s by gender and found that girls exhibited lower mean levels of physical aggressi on than did boys across the four si tes that included girls in their samples; however, girls were found to follow similar trajectories as boys. Gender has been examined in trajectory studies (Munson, McMahon, & Spieker, 2001; Spieker, Larson, Lewis, Keller & Gilchrist, 1999) focusing on how certain child, parent and family variables influence the level and rate of change in disruptive behaviors in preschoolers over time. In both studies, child sex was related to the level of behavior problems, but results differed in terms of which gender was found to have higher levels of behavi or problems. Spieker et al. (1999) found that boys had hi gher levels of behavior proble ms compared to girls. Munson et al. (2001) found the opposite, with girls in their sample havi ng higher levels of disruptive behavior. Despite differences in level of behavior problems between boys and girls, child sex was not related to the rate of change in behavi or problems in either st udy, suggesting that both boys and girls were changing in disruptive behavi ors at the same rate over the course of each study (Spieker et al., 1999; Munson et al., 2001). Parent Risk Factors A num ber of parent factors have been studied as risk factor s for the onset and development of disruptive behavior. Parenting stress and maternal depression are of partic ular interest to this study. Parents of children with disruptive behavior report higher le vels of parenting stress than parents of non-disruptive ch ildren (Morgan, Robinson, & Aldridge, 2002), and levels of parenting stress influence disciplin ary practices that directly prom ote and escalate aggressive and oppositional child behavior (Patterson, Reid, & Dishion, 1992). Moreover, parent stress appears to increase parent attention to deviant behavi or and to increase the likelihood of a parent
26 initiating or maintaining aversive interchanges with their child (Patterson, 1988; Patterson & Forgatch, 1990; Wahler & Dumas, 1989). Campbell, Shaw, and Gilliom ( 2000) also noted that parents who are stressed are more likely to engage in harsh or inconsistent parenting, which in turn exacerbates early parent-child struggles over autonomy and control. Williford, Calkins, and Keane (2007) examined patterns of change maternal parenting stress in a sample of boys and girls, from ages 2 to 5, at risk for externalizing behavior problems. They found that higher levels of disruptive behavior, anger proneness and emotional dysregulation in children predicted higher parenting stress in toddlerhood. They also found that mothers of children who displayed high levels of disruptive behaviors across all time points of the study (ages 2, 4 and 5) displaye d high levels of parenting stress across time. In other words, patterns of change for mothers of children with stable, significant levels of disruptive behaviors during toddlerhood, also remained relatively st able (demonstrated less change over time) in levels of parenting stress. Developmental trajectory studi es have examined other factors that are influenced by parenting stress, such as ineffective parenting practices. Shaw et al. (2003) examined whether rejecting parenting differentiated between boys who followed an early starter, persistent trajectory and boys who followed a high-level, de sister trajectory. Rejecting parenting was found to reliably distinguish these two trajectories. Spieker et al (1999) examined the effect of negative maternal control on development of child disruptive behavior and found negative maternal control to significantly relate to the rate of change in child disruptive behavior in preschoolers, with children of mothers who reported frequent yelling, threatening and spanking during conflict maintaining high levels of di sruptive behavior over time. Owens and Shaw (2003) examined effects of pare nt conflict and matern al acceptance on devel opment of disruptive
27 behaviors in preschoolers and found that preschoole rs of families with high parent conflict and low maternal acceptance maintained higher levels of externalizing behaviors across time. Findings from these studies provide support for the relation between ineffective parenting practices, parenting stress and the developmen t and maintenance of disruptive behaviors in children. Parenting stress has been shown to negatively affect parenting practices, which in turn predict the development and maintenance of disr uptive behaviors in pres choolers. In addition, Williford and colleagues (2007) demonstrated an a ssociation between change in parenting stress and changes in child disruptive behavior, with minimal change in sign ificant child disruptive behaviors predicting stable, high le vels of parenting stress during toddlerhood. It is anticipated that levels of parenting stress will also affect change in disr uptive behaviors during treatment, which will be explored further in the following section that discusses predictors of therapeutic change. Regarding parent psychopathology, paternal an tisocial behavior and maternal depression has been two of the best predictors of disr uptive behavior in boys (Frick, Lahey, Loeber, Stouthamer-Loeber, Christ, & Hanson, 1992). Studi es of children of affectively distressed parents suggest that these ch ildren are prone to detrimental outcomes (Cummings & Davies, 1994; Dodge, 1990; Gelfand & Teti, 1990). Maternal depression has been associated with disruptions in parenting behavi or, including parental uninvolvment, lack of responsivity, and lack of emotional support (Downey & Coyne, 1990) as well as increased hostility and criticism (Webster-Stratton & Hammond, 1988). Recent developmental trajectory studies have examined the relation between levels of and changes in maternal depression over time and the development of disruptive behaviors in children. Findings from these studies provide compelling evidence for the relation between
28 maternal depression and development of child di sruptive behaviors. Sh aw et al. (2003) found that high levels of maternal depression during the toddler period diffe rentiated children who followed an early starter or hi gh-level desister trajectory from children who followed moderate and low-level desister trajectories. Other developmental tr ajectory studies have found high levels of maternal depression to predict high levels of disruptiv e behaviors from preschool to school age (Munson et al., 2001; Owens & Shaw, 2003; Spieker et al., 199 9). Munson et al. (2001) examined the relation betw een maternal depression and th e rate of change in child disruptive behavior and found that children of mothers who report greater levels of depression developed disruptive behaviors at a faster rate over time and had higher levels of disruptive behavior at age 9 years. Muns on and colleagues also found that changes in maternal depression were related to changes in matern al ratings of disruptive behavior with increases in ratings of maternal depression associated with increases in ratings of child disruptive behaviors. (Author notes that this finding was specific to childre n in their study who were assessed as having avoidant, insecure attach ments to their mothers). Findings from these studies provide evidence th at maternal depression is a significant risk factor for both the development of disruptive be haviors and maintenance of these behaviors into school age. As will be discussed in the next se ction, maternal depression has also been found to adversely affect change in treatment. Changes in maternal depression during treatment may also affect change in disruptive be havior during treatment. This study aims to examine this possibility. Family Risk Factors W ith regard to family risk factors, socio economic status (SES) has been implicated as a risk factor in the onset and prognosis in disr uptive behavior. Moffit (1990) found that boys in the life-course persistent group, which is sim ilar to Pattersons (1982) early starter pathway,
29 came from families marked by chronic family adversity, as measured by low parental education and occupational status and low income. Aguilar, Sroufe, Egel and, and Carlson (2000) reported similar findings, with socioeconomic status differentiating between persistently disruptive children from children who were never disruptive or had adolescen t-onset behavior problems. DeVito and Hopkins (2001) found socioeconomic status to significantl y predict disruptive behaviors in children. Some researchers have examined the relati onship between SES and parenting stress in parents of children with disruptive behavior (Mash & Johnston, 1990; Baker, 1994). Findings are not conclusive. Mash and Johnston (1990) f ound that low SES predicted higher levels of parenting stress, while Baker (1994) found higher SE S to be associated with higher levels of parenting stress. Baker attempted to explain th is conflict in findings by suggesting that both high and low SES groups face a specific set of stressors that can he ighten parent-child conflict, thereby increasing parent stress. This study aims to investigate the association between SES and changes in parenting stress during the course of treatment. Evidence supporting a link between socioeconomic disadvantage and therapeutic change will be discussed in detail below. Cumulative Risk Factor Models Child, parent, and fam ily variables appear to play a significant role in the onset and persistence of disruptive behavior in children. In isola tion, child, parent and family risk factors appear to have consistent associations with level of child disruptive behavior; however, recent studies in the developmental psychopathology literature have shifted away from studying isolated risk factors and move d toward analyzing the cumulative models of risk (DeVito & Hopkins, 2001; Munson et al., 2001; Owens & Shaw 2003; Shaw et al., 2003; Spieker et al., 1999).
30 DeVito and Hopkins (2001) examined the cumulative effects of child attachment, permissive parenting, and marital dissatisfaction on disruptive beha vior in preschoolers. Sixty mother-child dyads attended a single 90 minut e appointment, during which mothers completed an interview and rating forms and dyads participat ed in a strange situa tion paradigm from which attachment style was assessed. The predictors of interest were entered into a hierarchical regression model and findings indi cated that children in coercively attached dyads whose mothers experienced less marital satisfaction a nd used more permissive parenting practices exhibited higher levels of disruptive behavior. Cumulative predictor models have also been ex amined in developmental trajectory studies. Munson, McMahon and Spieker (2001) examined th e effects of infant attachment, maternal depression, and child sex on the development of di sruptive behaviors. They found that children with avoidant insecure attachment styles at age 1whose mothers repo rted higher levels of maternal depression across time had higher levels of externalizing problems at age 9. In addition, they found that increases in mothers ratings of depressi on at a yearly assessment were associated with increases in mothers ratings of disruptive behavior at that same time point. Owens and Shaw (2003) examined the cumulative effects of infant negative emotionality, maternal depression, maternal acceptance, and pare ntal conflict on the development of disruptive behaviors in boys, ages 2 to 6, from low-income fa milies. Certain combinations of the four risk factors predicted higher levels of externalizing behaviors. The combination of high infant negative emotionality and high maternal depression resulted in less improvement in disruptive behavior over time and higher rates of extern alizing behaviors between ages 2 and 6. The combination of high parent conflict and low acceptance also resulted in less improvement of disruptive behavior over time and higher rates of externalizing behaviors between 2 and 6 years.
31 The combination of maternal acceptance and maternal depression was not statistically significant; however, high maternal depression w ith either high or low maternal acceptance resulted in a trend of higher levels of externalizing behavior at age 6. Research on trajectories of de velopment of disruptive behavior disorders has consistently linked child, parent, and family ri sk factors in the onset and deve lopment of disruptive behaviors. Recent studies provide evidence for an early starter trajectory across various samples of preschool age children and have el ucidated specific child, parent, and family risk factors that differentiate children who follow an early starte r trajectory and whose be haviors persist into school age (Spieker et al., 1999; Shaw et al., 200 3). These studies suggest that children who display conduct problems during the preschool years appear to be at increased risk for developing subsequent conduct problems as they get older. Findings from these studies also highlight the benefit of early identification and intervention to prevent these children from progressing toward subsequent conduct problems as they age and offset potentially worse outcomes. Shaw et al. (2003) notes that the mo re serious forms of condu ct problems in school age children and adolescence have been found to be more resistant to change, with few interventions proven consistently effective (Kazdin, 1995 as cited by Shaw et al., 2003). Early intervention, meaning interventi on during the preschool years, pr ior to school entry, captures families during a critical period of development, during which interventions have been shown to have higher probabilities of success (Dishion and Patternson, 1992). By targeting these behaviors early, children are prevented from progr essing along a trajectory that leads to more costly consequences (such as j uvenile delinquency) down the road. Developmental trajectory studies have linked specific child, pa rent, and family risk factors to the development of disruptive behaviors in chil dren. Findings from these studies have shown
32 that a sub-set of preschool age children develop significant levels of disruptive behaviors that persist into school age and adoles cence. This sub-set of presc hoolers can be differentiated to follow a high level, persistent trajectory accordin g to specific risk factors. Specifically, child factors such as hyperactivity and gender, parent factors such as maternal depression, and family factors such as socioeconomic disadvantage appe ar to contribute to onset and persistence of disruptive behaviors over time. Would these same predictors have a similar adverse effect on therapeutic change over the c ourse of an intervention? Predictors of Therapeutic Change and Premature Termination fro m Treatment An extensive literature exists on the predic tors of onset and development of disruptive behavior disorders. The treatment outcome and therapeutic change litera ture in child disruptive behavior has examined the same domains of risk factors in relation to treatment outcome, and associations have emerged between certain child, parent, and fam ily predictors and therapeutic change. A number of treatments have been deve loped to address disruptive behavior disorders in young children, and studies have shown positive th erapeutic changes in childrens disruptive behavior as a result of thes e treatments (Brestan & Eyberg, 1998; Hartman, Stage, & WebsterStratton, 2003; Kazdin, 1995; Nixon, 2001). Premature termination from treatment, however, is a common and significant challenge in research an d practice. Rates for psychotherapy dropout have been reported to range from 40% to 60% (Wierzbicki & Perkari k, 1993), indicating that around half of families who enter treatment leav e prematurely. Specific child factors (initial severity of disruptive behavior, comorbid di agnoses), parent factor s (depression, parenting stress), and family factors (socioeconomic di sadvantage, perceived barriers to treatment participation) have been found to predict premature termination (Dumas & Whaler, 1983; Kazdin, Holland, & Crowley, 1997; Webster-Stratton & Hammond, 1997); however, predictors of therapeutic change are less well understood and have been studied less extensively (Kazdin
33 and Wassell, 1998). A more extensiv e review of predictors of th erapeutic change and premature termination is discussed below. Child Factors Child factors, such as severity of child disruptive behavior, has been linked to poor outcom es in treatment, with children who disp lay more severe forms of conduct problems leaving treatment prematurely and showing less change in treatment (McMahon, 1994). Kazdin and colleagues have conducted a number of studies examining predictors of therapeutic change and premature termination from treatment and fo und that greater severity of child dysfunction -a general domain defined by tota l number of disruptive and co morbid symptoms -predicted both less improvement in child disruptive beha viors for those who remained in treatment (Kazdin, 1995; Kazdin & Wassell, 1999, 2000), as well as dropout from treatment (Kazdin & Mazurick, 1994; Kazdin, Mazurick, & Bass, 1993; Kazdin & Wasse ll, 1998), with children who drop out from treatment exhibiting more severe forms of disruptive behavior at start of treatment. Comorbidity has also been examined as a predictor of therapeutic change, though less frequently than other child f actors and with inconclusive findings (Abikoff & Klein, 1992). A comorbid diagnosis of ADHD is common among ch ildren with ODD, with estimates of cooccurrence in ADHD clinical samples ranging fr om 20% (Barkley, 1990) to 60% (Biederman, Munir, & Knee, 1987). Studies have shown that parents of children with comorbid ADHD and ODD have high rates of psychopathology, poor pare nting skills, and marital discord (Lahey, Piacentini, McBurnett, Stone, Hartdagen, & Hynd, 1988; Schachar & Wachsmuth, 1990). These findings suggest that a comorbid diagnosis of ADHD may have a negative impact on therapeutic change and potentially increase the risk for drop out from treatment. Results from the multisite, multimodal study of children with ADHD (MTA; Jens en et al., 2001) demonstr ated that children with comorbid ADHD and ODD/CD responded poorly to behavioral interventions compared to
34 children with only ADHD. Kazdin, Mazurick, and Bass (1993) found that multiple diagnoses were associated with premature termination from treatment. Together, these findings suggest that comorbid ADHD may have a negative infl uence on therapeutic change and may predict dropout. Child gender has not been widely investigated in treatment outcome studies, despite the inclusion of girls in samples discussed here (Kazdin & Wassell, 1998, 2000; Nixon, 2001). In a review of 82 psychosocial interventions for conduct problems for children and adolescents, Brestan and Eyberg (1998) noted that there was no information on gender differences in therapeutic change at the time, despite girls representing a signif icant minority of referrals for mental health services. Webster-Stratton (1996) attempted to address th is problem in the literature by exploring the effect of gender on therapeu tic change. She did not find a gender by time effect in her study, suggesting that boys and girls improved similarly in disruptive behaviors during treatment. She did find, however, a main effect for gender at pre-treatment, with mothers perceiving boys as having more disruptive beha viors than girls. One study was found (Hartman, Stage, & Webs ter-Stratton, 2003) that used multilevel modeling to examine the influence of child fact ors, specifically inatte ntion, impulsivity, and hyperactivity, on patterns of change in treatment in children with disruptive behavior. Hartman et al. (2003) found that children with attention pr oblems changed their disruptive behavior at a faster rate over the course of treatment and sh owed better improvement in disruptive behavior at one-year follow-up compared to children without attention problems. Though this finding contradicts earlier research findings it is important to note that th e children in this study did not carry a DSM diagnosis of ADHD; therefore, it is possible that th e findings for attention problems
35 may reflect a maternal perception bias, in which the mothers perceived their children as less attentive because they were so disruptive. Howe ver, the opposite could also be true, in which mothers perceive their children as more disrupt ive because they are less attentive and more active. Overall findings from this one study of patterns of change in ch ild disruptive behavior suggest that children with co-m orbid inattention symptoms may not have different patterns of change. Children with co-morbid attention symp toms or ADHD may improve at faster rates than children with only disruptive behaviors, especial ly if the behaviors are secondary to their difficulty with paying attention a nd sitting still. The current study proposes to examine the effect of child ADHD on patterns of change to determine if differences exist in the rates of change between children with and without ADHD. Parent Factors The effects of parent factors such as m aternal depression an d parenting stress have been examined as predictors of therapeutic change in treatment and dropout. Maternal depressive symptoms have predicted reduced responsiveness to treatment in children with disruptive behaviors (Kazdin, 1995; Kazdin & Wassell, 1999; Webster-Stratton & Hammond, 1990). Webster-Stratton and Hammond (1990) found that high levels of maternal depression predicted more negative parental perceptions of child adjustment at post-tr eatment. Kazdin and colleagues (Kazdin, 1995; Kazdin & Wassell, 1999, 2000) found that greater parent psychopathology predicted less therapeutic change in child disruptive behavior from pre to post-treatment. In the another study examining the relation between treatment completion and therapeutic change, Kazdin and Wassell(1998) found that children w ho responded favorably to treatment had parents who were less depressed compared to children who did not respond favorably to treatment. Studies have also examined the effect of child therapy on parent outcomes. Kazdin and Wassel (2000) examined whether parent outcomes improved during treatment of children with
36 disruptive behaviors. They predicted that pa rent functioning, includ ing maternal depression, would improve over the course of treatment, given the bidirectional, reciprocal and interdependent relationship between parent factors and child disr uptive behaviors (Munson et al, 2003). They found that levels of maternal depr ession significantly decreased from pre to posttreatment and that the magnitude of therap eutic change was small to medium. These improvements in maternal depression were not a result of an enhanced component. Families received standard forms of treat ment, yet improvements in child symptoms were correlated with improvements in maternal depression, providing support for the relati on between changes in maternal depression and changes in child di sruptive behavior during treatment. Sanders and McFarland (2000) did include an enhanced com ponent specifically designed to treat symptoms of depression in mothers, in addition to treating child disruptive behaviors. Mothers and children in both forms of behavioral family intervention (BFI and cognitive BFI) demonstrated both statistically and clinically significant improvements in depression and disruptive behavior at post-treatment. Though there were no differences between treatment conditions, Sanders and McFarland found that a larger percentage of mothers and children who participated in the enhanced component dem onstrated clinically significant and reliable concurrent change in maternal depression and child disruptive behavior s, again supporting an association between changes in maternal depres sion and changes in child disruptive behavior. In an earlier PCIT study, mothers who were experiencing severe parenting stress were more likely to drop out of PCIT (Werba, Eybe rg, Boggs, & Algina, in press). Kazdin and colleagues also examined the effects of paren ting stress on therapeutic change and found that high levels of parenting stress predicted increas ed risk for premature dropout from treatment (Kazdin, Mazurick, and Bass, 1993), as well as predicted less therapeutic change for families
37 who completed treatment (Kazdin and Wassell, 1999, 2000). In a study examining predictors of early versus late dropout from treatment, Kazdin and Mazurick (1994) found that parenting stress differentiated between families who dropped out of treatment early (completed fewer than six sessions) and families who completed treatment, w ith early dropouts reporting greater levels of stress compared to completers. In 2003, Kazdin and Whitley published results of a study examining the effect of an enhanced component designed to tr eat parenting stress while treati ng child disruptive behaviors. Though they found no significant differences in cha nge in parenting stress or child disruptive behavior between families who received the enhancement and those who did not, they did find that families who received the enhancement experienced greater reductions in parenting stress over the course of treatment. Families who recei ved the enhancement also demonstrated greater change in child disruptive behaviors compared to families who did not receive the enhancement. These findings match results from a previous study that demonstrated small to medium effects in changes in parenting stress during the course of treatment without enhancements (Kazdin and Wassell, 2000) and provide support for associatio ns between changes in parenting stress and changes in child disruptive behavior. Lack of maternal social support has also been associated with less change in treatment. Dumas and Wahler (1983) examined maternal insularity, defined as number of community contacts, and found a steady increase in the probab ility of treatment failure under conditions of maternal insularity. However, recent studies have demonstrated associations between improvements in social support during treatm ent and improvements in child disruptive behaviors. Harwood and Eyberg (2004) examined predictors of change in mother-child functioning during the Child Direct ed Interaction phase of PCIT. They found that pre-treatment
38 ratings of adequate maternal social support pr edicted greater improvements in mother-child functioning from start to end of CDI in PCIT. Kazdin and co lleagues (Kazdin & Wassell, 2000; Kazdin & Whitley, 2003) found that mothers rati ngs of perceived social support significantly improved from preto post-treatment in their studies, with improvements in social support (and other variables of parent and family functio ning) associated with improvements in child disruptive behavior during treatment. Overall, findings from these st udies suggest a relation between changes in child disruptive behavior and changes in maternal social support. In a study examining patterns of change in ch ild disruptive behavior, Hartman et al. (2003) did not find pre-treatment levels of maternal depression or paren ting stress to predict different patterns of change in child disr uptive behavior during treatment. They noted that the lack of significant findings for depression and stress coul d have been due to the minimal levels of depression and stress reported by mothers at pre-treatment in their sample. Fourteen percent of their sample of mothers reported mild depressi ve symptoms and 4% fe ll within the moderate range. The remaining 72% of mothers rated mini mal symptoms of depression, and only 1% of the sample reported negative life events within the clinical range. Hence th eir lack of significant findings was likely related to the fact that mothers in their sample were not severely depressed or stressed. Taken together, findings from studies review ed here suggest a re lation between parent factors and therapeutic change, as well as dropo ut from therapy. Greater levels of maternal depression and parenting stress as well as low levels of maternal support, are associated with less change in therapy and greater risk for dropout, particularly early on in treatment. However, evidence from studies of therapeutic change revi ewed here suggests that treatment also has a positive effect on parent factors, with changes in maternal depression and stress associated with
39 changes in child disruptive behavior for familie s who remain in treatm ent ( Kazdin & Wassell, 2000). This study aims to further investigate the nature of the associations between changes in maternal depression and changes in parenti ng stress and child disruptive behavior during treatment and to determine whether differences in patterns of change in parenting stress can predict change in specific child and parent outcomes, including ch anges in attachment, perceived social support, and dropout from PCIT. Family Factors Fa mily variables, such as socioeconomic disa dvantage and perceived barriers to treatment, have also emerged as signifi cant predictors of negative treatment progress and premature termination. Dumas and Wahler ( 1983) found that low socioeconomic status (SES) increased the probability of treatment failure in a sample of mothers seeking help with their childrens oppositional behavior. Webster-Stratton and Hamm ond (1990) found similar results in their sample of mothers and fathers involved in parent training for their childrens behavior problems, with low SES predicting poor er long-term outcome. Recent studies have continued to link socioeconomic disadvantage to dropout and lower rate of change in treatment (Kazdin, 1995a; Webster-Stratton & Hammond, 1990). Kazdin and Wassell (1998, 1999) examined the association between treatment comple tion and therapeutic change in the context of child, parenting, and family risk fact ors known to predict dropout in treatment. In their sample of 3to 13-year-o lds referred for outpatient treatment of disruptive behavior, they found that grea ter socioeconomic disadvantag e predicted less therapeutic improvement both in child disrupt ive behavior and in parent f unctioning (depression and stress) from preto post-treatment. Hartman et al. (2003) examined the effect soci oeconomic status on patterns of change in child disruptive behavior during treatment and found that SES did not predict different rates of
40 change in child disruptive behavior during treatmen t. They did find SES to predict differences in severity of disruptive behaviors at pre-treatment, with children from more disadvantaged families exhibiting greater levels of di sruptive behaviors; however chil dren changed at similar rates during the course of treatment. Socioeconomic disadvantage has been linked to a greater number of perceived barriers to treatment participation (Kazdin & Wassell, 1998). Studies have shown that families who report a high number of barriers to treatment participati on show less therapeutic change in both child disruptive behaviors and parent functioning (depression and stre ss; Kazdin & Wassell, 1999). Families who report high number of barriers are also at higher risk for dropout (Kazdin, Holland, & Crowley, 1997) and have poorer treatment attenda nce (more canceled, no-show, or late arrival appointments; Kazdin & Wassell, 1998). Families who drop out of treatment tend to perceive treatment as more demanding and less relevant compared to families who complete, and they tend to report less alliance and bonding to the therapis t (Kazdin & Wassell, 1998). Given results from studies reviewed here, fa mily factors are significant predictors of therapeutic change and dropout from treatment Though Hartman et al. (2003) returned nonsignificant findings for SES in their sample, th e evidence from other studies of therapeutic change implicate SES as an important predicto r of change; hence, it will be included as a predictor of patterns of change in child disruptive behavi or and parenting stre ss in this study to determine if differences exist in this sample. Treatment protocol for the larger study from which the data for this study comes included compone nts within each treatm ent session to address potential barriers to participat ion in treatment, which have been shown to adversely affect therapeutic change and treatment completion in pa st studies reviewed above. Hence, ratings of
41 perceived barriers to participation in treatment will be examined as a predictor of patterns of change in child disruptive behavior and parenting stress as well. Summary The literature on child, parent, and fam ily fact ors that affect change in treatment and dropout is extensive; however, a li mited number of studies exist that have examined the effect of these factors on patterns of change in child di sruptive behavior and parenting stress during treatment and whether patterns of change in child disruptive beha vior and parenting stress can predict specific child, parent, and family out comes, including dropout. This study aims to examine patterns of change in child disruptive be havior and parenting st ress, through application of multilevel modeling statistical procedures, and i nvestigate the effects of certain child, parent, and family predictors on patterns of change during PCIT. Child pr edictors of patterns of change to be investigated include gender and diagnosis of ADHD. Maternal depression will also be examined as a predictor of patterns of change, specifically changes in maternal depression from preto midto post-treatment predicting cha nges in child disruptive behavior and parenting stress. Family predictors of patterns of change include SES and barriers to participation in treatment. A secondary aim of this study is to determ ine whether patterns of change in child disruptive behavior and parenting stress predict change in specific child and parent outcomes. Pattern of change in child disr uptive behavior will be examined as a predictor in change in attachment and change in perceived maternal social support from preto post-treatment. Patterns of change in parenting stress will be examined as a predictor in change in attachment and change in child disruptive behavior. Finally, patterns of change in child disruptive behavior will be examined as a predictor of dropout from PCIT. Prior to discussion of specific hypotheses, a review of Parent Child Interaction Therapy is first discussed below.
42 Parent Child Interaction Therapy Parent Child Interaction Therapy (P CIT) is an evidence-based treatment specifically designed to address behavior pr oblems commonly seen in child ren with disruptive behavior disorders (ODD, CD) and ADHD. The behaviors a ssociated with these disorders are the ones most commonly referred to child mental health services (Kazdin, Siegel, & Bass, 1990). PCIT is based in developmental theory and draws from both attachment theory and social learning theory in its principles of change. PCIT aims to change parent-child interac tions by promoting more optimal styles of parenting. Baumrinds (1967, 1991) developmental theo ry proposed three styles of parenting that resulted in better or worse outcomes for childre n. She later reformulated her theory along two dimensions of parent responsiveness and parent demandingness. The authoritative parenting style, under this new typology, was one in wh ich parents are both hi ghly demanding and highly responsive. Baumrinds research and subsequent studies have consistently found an association between specific parenting styles and child be havior problems (Azar & Wolfe, 1989; Calzada & Eyberg, in press; Olson, Bates, & Bayles, 1990). Based on these findings, PCIT focuses on changing maladaptive parent-child interactions into ones reflecting an authoritative parenting style. It draws on both attachment theory and social learning theory in its principles of change and utilizes a behavioral approach to change parent child interactions. Attachment theory asserts that children whos e parents respond to th em reciprocally and with nurturance are more likely to develop a secu re attachment to thei r parents, leading to positive social, emotional, and behavioral outco mes. Conversely, children whose parents are intolerant and unresponsive to their childs need and distress are more likely to develop a maladaptive, or insecure, attachment to their parents. Maladaptive attachments have been associated with the development of disruptive behavior in children (, DeVito & Hopkins, 2001;
43 Fagot & Pears, 1996; Greenberg, Speltz, DeKlyen, & Endriga, 1992). The combination of maladaptive parent-child attachment and poor pare nting practices has been consistently linked to the severity of disruptive beha vior (Patterson, 1982; Loeber & Schmaling, 1985). Therefore, the first phase of PCIT, the child-directed interactio n phase (CDI), focuses on developing a secure parent-child relationship. Parents are coached to use skills that re structure the play interaction in ways designed to create a secure attachme nt. Social learning theory emphasizes the contingencies that shape the dysf unctional parent-child interacti ons seen in disruptive children and their parents. Pattersons (1982) coercion theory provides a transactional a ccount of early conduct problems (Eyberg, Schumann, & Rey, 1988) in which childrens behavior problems are inadvertently established or maintained by malada ptive parent child interactions. These coercive interactions between disruptive children and their parents ar e characterized by mutual and escalating aversive behaviors resulting from the a ttempts of both the parent and child to control each others behaviors. Unhealthy patterns result when the parent eith er withdraws from the conflict, thereby reinforcing the childs disruptive behavior ( cryi ng, whining, yelling, or hitting), or engages in parenting practices that could harm the child. The second phase of PCIT, the parent directed interaction (PDI), aims to interrupt th ese coercive cycles by teaching the parents to be more consistent, firm, and clear in their limit setting. In PDI, the parents are taught specific behavior management techniques that work to establish consistent contingencies for child misbehavior in the context of a positive parent child attachment. The goals of PCIT are (a) to improve the parent child attachment and (b) to improve parental behavior management skills. Each phase of PCIT begins with a teaching session, in which the skills are introduced and modeled fo r the parents. The teaching sessions are then followed by coaching sessions, in which the pa rents are coached on th ese skills over a bug-in-
44 the-ear-device as the therapist observes from an observation room behind a one-way mirror. In the CDI phase, parents are coached on a specific se t of skills until they reach mastery criteria. Once they reach mastery criteria, they move on to the PDI phase of PCIT. Parents remain in PDI until the childs behavior has decreased to within the normal range and the parents demonstrate mastery of the PDI skills. In the CDI phase, parents learn to follow th eir childs lead in play and incorporate the PRIDE skills. Parents are coached to give positive attention to their children by praising their behavior, reflecting their statements, imitating an d describing their play, and using enthusiasm. In the event their child misbehaves during the play, they are coached to ignore the negative behavior until it ceases. Through the application of differential social attention, parents teach their child that prosocia l behavior is rewarded with their praise and attention while negative behaviors result in the removal of their attention. Parents are al so directed to avoid behaviors that attempt to lead the play, such as aski ng questions, giving commands and criticizing the child. These behaviors are conceptu alized as potentially introduci ng negativity into the play, and the goal of CDI is to create or enhance a warm, secure parent-child rela tionship. In addition to practicing these skills in sessions, parents are asked to practice the PRIDE skills each day at home, which consists of a 5-minute special time during which the parent follows the childs lead in the play and uses only the PRIDE skills to communicate with the child. Parents are provided a handout to record how each home practice session goes. Parents remain in CDI until they demonstr ate mastery of the CDI skills. Mastery is defined by the number of PRIDE skills (, the numb er of labeled and unlabeled praises, reflective and descriptive statements) observed within a 5-mi nute coding period at the start of each session. To reach mastery criteria in CDI, parents must demonstrate (a) 10 behavioral descriptions, (b) 10
45 reflective statements, (c) 10 labe led praises, and (d) no more th an 3 total questions, commands, or criticisms within the 5-minute coding period. Once the parents have met CDI criteria, they move on to the PDI phase of treatment. However, parents continue to pr actice CDI at home and the therapist continues to code and coach CDI prio r to PDI in the latter sessions of PCIT should the parents CDI skills fall below criteria. The primary goals of PDI are to increa se compliance and decrease inappropriate behaviors that do respond to ignoring or are t oo severe to ignore (, hitting, biting, being destructive with toys). In PDI, parents continue to give their child positive attention, but they now learn to give their child specific directions and to follow through consistently with either praise for compliance or time out for noncomplian ce. The therapist coaches the parents to give clear, direct commands (Please hand me the truc k in your hand.) rather than criticisms (Dont you dare throw that truck.) or indirect comman ds that suggest complia nce is optional (Do you want to hand that truck to me? ). Parents are provided specific steps to follow once they have given a command that help them to avoid delay tactics and remain consistent until the child has obeyed the command. In PDI, parents are taught a specific time-out procedure th at provides them a standard, concrete set of steps to follow after they have given a command. The procedure has three levels: warning, chair, and time-out room. At each level, the child has the opportunity to obey the parent and end the time-out; however, the time-out doe s not end until the child obeys the parents command and the parent reinforces the childs co mpliance with labeled praise. Once the parent gives the child a command, the PDI procedure begi ns. If the child obeys immediately, the parent is directed to reward the child with a warm and enthusiastic labeled praise for complying to the parents command immediately. If the child does not obey immediatel y (within a 5-second
46 count), the parent is directed to issue a specific warning: If you dont [original command], then you will have to sit on the time-out chair. If the child obeys the warning, then parent gives the child a labeled praise an d the play continues. If the child does not obey th e warning (within the 5-second time limit), the parent is directed to take the child to the chair while calmly explaining, You didnt [original command], so you have to sit on the chair. This statement serves to remi nd the child of the reason for punishment and reiterates the connection between noncompliance and a negative consequence. After placing the child on the chair, the parent says, Stay on the chair until I tell you that you can get off. This statement serves to estab lish the parents contro l during the time out procedure. The parent is direct ed to ignore all negative child beha viors that occur while the child is in time out. This can prove to be difficult for the parent as the child often engages in various forms of attention-getting behavior, such as emotional manipulation (, I dont love you anymore, My stomach hurts, and Im sorr y, I promise Ill listen) or negative physical behavior (, wetting pants). The ch ild is required to sit on the ch air for 3 minutes, plus 5 seconds of quiet at the end. These 5 sec onds of quiet ensure that the ch ild does not come away from the time out with the impression that his or her be havior on the chair was the cause for ending time out. Once the 5 seconds of quiet has elapsed, the parent is instructed to walk over to the child and ask, Are you read to [origina l command]? If the child says no, begins to argue, or ignores the parent, the parent says, All right, then stay on the chair until I tell you that you can get off. The parent then immediately leaves the area of the chair and begins the 3-minute time period again. If the child indicates that he or she is ready to obey, eith er by saying yes or by getting off the chair in a compliant manner, the parent wa lks the child back to the task. The parent then
47 indicates to the child to obey the command (, pointing to the block that th e child was originally instructed to put in its bo x). When the child obeys, th e parent only gives a brief acknowledgement (Fine.). Because the child had to be sent to timeout before complying, a labeled praise is not given. Inst ead, the parent follows up with a nother direct command. At this point, the child is likely to obe y the command, allowing the parent the opportunity to give an enthusiastic labeled praise fo r minding immediately, to explai n the reason that the childs compliance is good, and to return to a CDI. In the event that the child does not rema in in the chair during timeout, a back-up procedure is used to shape the childs behavior to remain in the chair. This back-up procedure is the time-out room, an empty room that is easi ly accessible from the playroom. In the home, parents are instructed to find a room at least 4 feet by 4 feet in size that can be lighted, to use as a time-out room. Common choices are utility rooms, bathrooms, or walk-in closets that can be cleared out for a few weeks. If the child gets off the chair before the 3 minut es elapse, the parent is instructed to give a time-out room warning one time ever, saying, You got off the chair before I told you that you could. If you get off the chair again, you will have to go to the time-out room. After the room warning, if the child leaves the chair again before the parent gi ves permission, the child goes to the time-out room. The parent says, You got off the chair before I told you could, so you have to go to the time-out room. Once the child is in the room, the parent cl oses the door and begins the one-minute, plus 5 seconds of quiet, time period. After the time has elapsed, the child is escorted back to the chair and told, Stay in the chair until I tell you that you can get off. The 3minute time out on the chair then begins again. The purpose of the time-out room is to serve as a back-up time procedure used only to teach the ch ild to remain on the time-out chair until given
48 permission to get off. If the child were allowed to get off the time-out chair without permission, the chair would be ineffective for use in the tim e-out procedure. The child must remain on the chair until the parent gives permission, by asking the child if he or she is ready to obey the original command. After the parents have practiced and gained e xperience with the PDI procedure in session, they are directed to begin practic ing the PDI at home, first in the context of their daily CDI home sessions. Specifically, parents be gin practicing PDI at the end of the CDI special time, during which they direct the child in putting away the toys that were used during the play. As parents begin to feel confident following through with the PDI procedures at home, they are instructed to use direct commands at specific times throughout th e day with the intent to follow through with a labeled praise if the child complies or the timeout procedure if the child does not comply. In treatment sessions, the parents PDI skills are coded during the fi rst 5 minutes to assess their progress toward mastery. If 75% of the parents commands are given corr ectly and the parent follows through with praise or timeout correctly at least 75% of the time, then they have met mastery criteria for PDI. Several studies have provided empirical support for the effectiveness of PCIT for treating children with disruptive behavior. Studies have compared PCIT to wait-list controls (McNeil, Capage, Bahl, & Blanc, 1999; Schuhmann, Foote, Eyberg, Boggs, & Algina, 1998), classroom controls (McNeil, Eyberg, Eisenstadt, Newc omb, & Funderburk, 1991), modified PCIT (Nixon, Sweeney, Erickson, & Touyz, 2003), and group pare nt training (Eyberg & Matarazzo, 1980). Outcome studies in PCIT have evidenced changes in parents behavior toward their children, including increased reflectiv e listening, physical proximity, prosocial verbalization, and decreased criticism and sarcas m (Eisenstadt, Eyberg, McNe il, Newcomb, & Funderburk, 1993).
49 Significant improvements in pare nt psychopathology, personal distre ss, and parenting locus of control have been reported af ter PCIT as well (Schuhmann et al., 1998). The effects of PCIT have been found to generalize to untreated si blings (Brestan, Eyberg, Boggs, & Algina, 1997; Eyberg & Robinson, 1982), to other settings, su ch as school (McNeil et al., 1991), and across time (Boggs, Eyberg, Edwards, Rayfield, Jac obs, Hood, & Bagner, 2004; Funderburk et al., 1998; Eyberg et al., 2001; Hood & Eyberg, 2003). The Schuhmann et al. (1998) study has been the largest PCIT outcome study to date. In the Schuhmann et al. study, following an initial assessment, 64 clinic-referred families of children with ODD were randomly assigned to an immedi ate treatment (IT) or wait-list control (WL) group. After treatment, parents in the IT group in teracted more positively with their child and were more successful in gaining their childs compliance than parents in the WL group. Parents in the IT group also reported less parenting stress and a more internal locus of control than parents in the WL group. Both c linically and statistic ally significant improvements were reported in child behavior at the end of treatment. Families that completed treatment reported high levels of satisfaction with both the content and pro cess of PCIT. Four-month follow-up data showed that gains were maintained on all parent-report measures. Nixon, Sweeney, Erickson, and Touyz (2003) al so conducted a large outcome study with 54 disruptive preschool ch ildren randomly assigned to one of three conditions: standard PCIT; modified PCIT or no-treatment waitlist cont rol group. The modified PCIT used didactic videotapes, telephone consultations and face-to-face sessions to abbreviate treatment. Children were assessed at pretreatment, posttreatment, and at six-month follow-up. After treatment, mothers in the standard and abbreviated treatm ent groups reported less oppositional behavior, as measured by the ECBI, compared with the wait list control group. Moth ers in the standard
50 treatment group also reported less severe behavi or problems in the home compared to waitlist controls, though mothers in the abbreviated group did not. Mothers in the standard and abbreviated groups praised their children more a nd gave fewer commands compared to waitlist controls during parent-child inte ractions. Additionally, mothers in the standard group also gave fewer criticisms and their children were more compliant compared to waitlist controls. Gains were maintained at 6-month follow-up and clinically significant reductions in number of maternal commands were also maintained in bot h the abbreviated and st andard treatment groups. Studies have also examined the generalizati on of treatment effects to the school setting. McNeil et al. (1991) examined the generalization of PCIT treatment effects from the clinic to the school setting in 10 children with disruptive behavior occurring both at home and in the classroom. Families received 14 weeks of PCIT, and no advice or direct classroom intervention regarding school misbehavior was provided. Both teacher ratings and school observational measures indicated significantly greater improve ments in disruptive behavior for the treated group than a group of 10 normal controls and a group of 10 behavior problem controls drawn from the classrooms of the treated children. Re sults were less supportive for generalization on measures of hyperactivity and distractabilit y. Nixon (2001), however, has demonstrated generalization of PCIT to meas ures of hyperactivity and temper ament. He randomly assigned 34 families to either PCIT or a wait-list control group (WL) and compared them to a group of 21 nondisturbed preschoolers at post-treatment and at 6-month follow-up. At post-treatment, the PCIT preschoolers were reported by their mothers to have reduced hyperactivity and more flexible temperament and were less likely to meet criteria for ADHD than the WL group. At 6 months post-treatment, children in the PCIT group were comparable to children in the normal comparison group on measures of oppositional and hyperactive behavior.
51 Funderburk et al. (1998) conduc ted 12and 18-month follow-up school assessments for 12 children, including those who had completed PCIT in the McNeil et al. (1991) study. At the 12month follow-up, children in the treated group maintained post-treatment improvements in teacher ratings and observational measures of disruptive behavior and showed further improvements in social competency. They were indistinguishable from 72 randomly selected control children from their classrooms on measure of disruptive behavior and social competence. At the 18-month follow-up, children maintained their improvements in compliance but demonstrated declines on other measures into the pre-treatment range. Generalization of PCIT to family functioning was found in a study conducted by Eyberg and Robinson (1982). In a sample of 7 families that completed PCIT, significant improvements were seen in several aspects of family functioning, as well as child behavior. Mothers showed less anxiety and pessimism, increased involvement and interest in others, a nd a greater degree of internal control after treatment. Maternal ad justment ratings also improved, as did observed behavior of the childrens untreated siblings Brestan, Eyberg, Boggs, and Algina (1997) examined parent perceptions of untreated siblings after PCIT compared to untreated families. After completing treatment, fathers rated the be havior problems of unt reated siblings as occurring less frequently and moth ers rated these behaviors as less problematic than ratings from untreated control families. The effects of PCIT have been shown to la st after treatment has ended. In a six-week follow-up of 14 families, all families maintained treatment gains on observation measures of child compliance, parent-rating scale measures of disruptive behavior, internalizing problems, activity level, maternal stress, a nd child self-report of self-esteem (Eisenstadt et al., 1993). At a 2-year follow-up with these families, parent ratings of child behavior problems, child activity
52 level, and parenting stress remained similar to post treatment levels, and most children remained free of disruptive behavior diagnoses (Eyberg et al., 2001). The magnitude of the effects in parent ratings of child disruptive behavior and parenting stress at 1and 2-year follow up were large, ranging from 2.89 to .70, suggesting that th e maintenance of gains from PCIT two years out is a legitimate finding. These parents also continued to report high satisfaction with the process and outcome of PCIT. Boggs et al. (2004) compared outcomes for families who completed PCIT and those who dropped out of the Schuhmann et al. (1998) study. Families were located and telephone and mail assessments were conducted. Length of follow-up ranged from 10 to 30 months after initial treatment intake, with the average length of fo llow-up just under 20 months. Results indicated significantly poorer long-term out comes for those who dropped out of treatment. Children and families that completed treatment maintained treatment gains over this period, whereas the dropouts showed disruptive behavior and pare nting stress at pr e-treatment levels. Hood and Eyberg (2003) attempted to locate 50 families that had completed PCIT 4 to 6 years earlier. Of the 29 families that could be located, 23 participated in telephone and mail follow-up assessments of child disruptive behavi or and parenting locus of control. Results indicated that the significant changes made during treatment were maintained for the children, now ages 6 to 12, and their mothers. Child beha vior at post-treatment assessment and length of time since treatment were strong predictors of child behavior at long-term follow-up. The investigators found that childrens disruptive behavior decreased with ti me since treatment. One study has specifically examined predictors of dropout in PCIT. Werb a et al. (in press) explored both observational and self-report data as predictors of treatment outcome and dropout. They found that mothers who were highly demanding during parent child interactions were more
53 likely to drop out of treatment prematurely. They also found that maternal distress at pretreatment (a composite score consisting of parent stress and depression) differentiated between treatment completers and dropouts, w ith dropouts having higher levels of maternal distress at pretreatment. Taken together, findings from the above studies suggest that, despit e success in changing both child and parent functioning, treatment outcome in PCIT is vulnerable to similar risk factors reported in other treatment outcome studies in children with disruptive behavior. Studies in PCIT have demonstrated that families who prem aturely drop out of PCIT yield poorer outcomes, including child disruptive behavior and parenting stress returning to pretreatment levels (Boggs et al, 2004). One study specifically examined pr edictors of treatment outcome and dropout in PCIT and found that parent factors, particularly maternal di stress and demandingness during play observations, were predictive of premature dropout from PCIT. The larger investigation from which this study was borne attempted to prevent attrition by implementing supportive components into treatment that specifically addr essed potential barriers to participation in treatment. Therefore, examining the influence of child, parent, and family factors on therapeutic change and dropout in PCIT is particularly im portant to see if the supportive component was effective in retaining families in treatment. Hypotheses The m ain aim of this study is to examine th e effects of child, parent, and family risk variables on the rate of change in child disruptive behavi or and parenting stre ss during the course of an intervention specifically aimed at affecting change in processes and interactions that have inadvertently reinforced disruptive child behavi or. Change in child disruptive behavior and parenting stress will be examined weekly to expl ore whether rate of change in each phase of treatment and across treatment as a whole in child disruptive behavior and parenting stress varies
54 by phase of treatment among families referred for disruptive behavior. Differences in rate of change in child disruptive behavi or and parenting stress are expe cted between phases, as number of sessions in each phase of treatment differe d from family to family. Certain families completed the CDI phase within five sessions while other families remained in the CDI phase for 10 or more sessions. Families also differed in number of sessions in the PDI phase, with some families graduating from PDI within five sessions and other remaining in PDI for more than 10 sessions. Families who completed each phase of PCIT in fewer number of session are expected to demonstrate steeper rates of change in comparison to families who required more sessions in each phase of treatment. Child, parent, and fam ily risk factors are pr edicted to explain the differences in rates of change between th e CDI phase and PDI phase of PCIT. A secondary aim of this study is to investig ate whether the patterns of change in child disruptive behavior and parenting stress predict change in treatm ent outcome and dropout in this study. Findings from previous studies ( Kazd in & Mazurick, 1994; Pekarik, 1992) examining predictors of dropout from treatment found that early versus late dropout was distinguished by different sets of child, parent, and family predictors. Moreover, Pekarik (1992) found no differences in parent ratings of child disruptive behavior between completers and late dropouts in his sample, suggesting that late dropouts achieve d similar gains in treatment as completers. Thus, early dropouts in this study are expected to show patterns of little to no change. Late dropouts are expected to show patte rns of change indicative of impr ovement but at a slower rate when compared to completers. Two sets of analyses are proposed for th e current study. The first set will focus on examination of patterns of change in child disruptive behavior and parenting stress in each phase of PCIT and across treatment as whole, through application of the multilevel model approach.
55 The following child, parent, and family risk factors will be examined to determine their effect on patterns of change in each phase of treatment a nd across treatment as whole, in child disruptive behavior and parenting stress: child risk factors include: (a) presence of ADHD and (b) child gender; parent risk factor of interest is maternal depression; family risk factors that will be examined are: (a) socioeconomic status as measured by the Hollingshead Index, and (b) perceived barriers to participation in treatme nt reported by treatment completers at the posttreatment assessment. Because premature dropout is a significant problem in child therapy, families who prematurely dropped out of treatment will be included in the multilevel model analysis. Multilevel modeling uses all avai lable data for each participant, even when there are missing data at certain time points for a particular participant. Therefore, trajectories for dropouts will be possible to analyze in context of individual risk factors. It is recognized that only treatment completers have contributed on pe rceived barriers to participation in treatment; therefore that particular multilevel model analysis will onl y include data from families who completed treatment. The second set of analyses will focus on invest igating whether patterns of change in child disruptive behavior and parenting stress determined from the first set of analyses can predict change in treatment outcome, including early vers us late dropout from treatment. A traditional linear regression approach will be applied to examine whether individual rates of change in child disruptive behavior and parenting stress predict change in four treatment outcome variables: (1) change in attachment security from preto post-treatment; (2) cha nge in child disruptive behavior from preto post-treatment, (3) change in parenting daily hassl es from preto posttreatment; and (4) change in maternal ratings of perceived social support from preto post-
56 treatment. A logistic regressi on approach will be applied to determine whether patterns of change in child disruptive behavior and parenting stress for dropouts can predict early versus late dropout from the study. Below are the hypotheses proposed for this study, organized by set of analysis. Patterns of Change in Child Disrupt ive Behavior and Parenting Stress Term ination criteria require that a child be ra ted within the non-clin ical range on a measure of child disruptive behavior (a raw score of 114 which is a standard deviation below the normative mean). Therefore, it is expected that all completers will show improvements in child disruptive behavior over the course of PCIT. I ndividual differences in the rates of change in child disruptive behavior and parenting stress in each phase of treatment and across treatment as a whole are predicted: Hypothesis 1: Children with a co-morbid rese arch diagnosis of ADHD are hypothesized to demonstrate greater changes in levels of child disruptive behavior and in parenting stress in the PDI phase than the CDI phase. Hypothesis 2: Children who are more securely attached at pre-treatment are hypothesized to make greater change in th e CDI phase than the PDI phase. Hypothesis 3: A relationship between changes in maternal depression and changes in child disruptive and parenting stress are hypothesized High levels of ma ternal depression at pre-, mid-, and post-treatment are expected to be associated with less change (flatter slope) in child disruptive behavior and parenting st ress, whereas levels of maternal depression that decrease from preto mi dto post-treatment are expect ed to be associated with a decrease in child disruptive behavi or and parenting stress durint PCIT. Hypothesis 4: Patterns of change for boys and girls will be examined separately to determine whether gender differences exist in patterns of change in child disruptive behavior and parenti ng stress during PCIT. Hypothesis 5: Families who retrospectively report a greater number of perceived barriers to participation in treatment are expected to show less change (fla tter slope) in PCIT compared to families who report fewere barriers. Hypothesis 6: Children from socioeconomically disadvantaged homes are hypothesized to show flatter slope of change in child disrupt ive behavior during PCIT compared to children from less disadvantaged homes.
57 Hypothesis 7: Mothers who are socioeconomi cally disadvantaged and retrospectively report greater number of barriers to participation in treateme nt are hypothesized to show a flatter slope of change in pa renting stress during PCIT, compar ed to mothers whoe report fewer barriers and are not socioeconomically disadvantaged. Predicting Treatment Outcome from Patterns of Change in Child Disruptive Behavior and Parenting Stress Hypothesis 1: Children who show steeper patterns of change in child disruptive behavior in CDI than PDI will show greater change in attachment security from preto post-treatment. Hypothesis 2: Children who show steeper patterns fo change in child disruptive behavior during PCIT will show greater change in maternal ratings of parenting stress from preto post-treatment, compared to children who show flatter slopes during PCIT. Hypothesis 3: Mothers who show steeper patterns of change in parenting stress related to perceptions of child as difficult to manage during CDI than PDI will show greater change in attachment security from preto post-treatment. Hypothesis 4: Mothers who show steeper patterns of change in parenting stress related to dysfunctional parent child interactions and perceptions of child as difficult to manage in PCT will show greater change in child disruptieve behavior from preto post-treatment. Hypothesis 5: Mothers who show steeper patterns of change in parenting stress related to dysfunctional parent child interactions and di stress related to parent ing role during PCIT will show greater change in maternal ratings of perceived social support from preto posttreatment. Predicting Early versus Late Dropout in PCIT from Patterns of Change Hypothesis 6: It is hypothesized that fa milies w ho demonstrate flatter patterns of change in child disruptive behavior, parenting stre ss related to dysfunctional parent child interactions, and prenting stress related to perceptions of having a difficult child will be more likely to drop out of treatment in the CDI phase.
58 Table 1-1 DSM-IV-TR Criteria for ADHD Diagnostic Criteria for AttentionDeficit/Hyperactivity Disorder Either (1) or (2): six or more of the followi ng symptoms persisting at least 6 months to a degree which is maladaptive and inc onsistent with developmental level Inattention Hyperactivity-Impulsivity Often fails to give close a ttention to detail Often fidgets with hands or feet or squirms in seat Often has difficulty sustaining attention in tasks or play activities Often leaves seat in classroom or in other situations in which remaining in seat is expected Often does not seem to listen when spoken to directly Often runs about or climbs excessively in which it is inappropriate Often does not follow through on instructions and fails to finish schoolwork, chores, or duties in the workplace Often has difficulty playing or engaging in leisure activities quietly Often has difficulty organizing tasks and activities Is often on the go or often play acts as if driven by a motor Often avoids, dislikes, or is reluctant to engage in tasks that require sust ained mental effort Often talks excessively Often loses things necessary for tasks or activitiesOften blurts out answers before questions have been completed Is often easily distracted by extraneous stimuli Often has difficulty awaiting turn Is often forgetful in daily activities Often interrupts or intrudes on others
59 CHAPTER 2 METHOD Participants Participants were 100 boys and girls between th e ages of 3 and 6 years who were enrolled in a larger treatm ent outcome study of PCIT. Inclusion criteria in the larger study included meeting diagnostic criteria for a DSM-IV diagno sis of Oppositional Defiant Disorder and living with at least one parent able to participate in treatment. A diagnosis of ODD was determined by a combination of the Diagnostic Interview Schedule for Children (DISC) criteria for ODD and a Child Behavior Checklist (CBCL) a ggression scale cutoff score of T > 61. Both child and parent were also required to obtain a standard scor e (SS) of 75 or higher on a cognitive screening measure. Children with a history of severe sensory or mental impairment (e.g., deaf, blind, autistic) were excluded from the study. Howe ver, children who had been stable on a psychoactive medication for a 30-day period prior to screening and who did not plan to change their medication regimen during treatment were included. Children diagnosed with co-morbid childhood disorders, such as ADHD, CD and Sepa ration Anxiety Disorder, were not excluded from the study. Thirty-one girls and 69 boys were enrolled in the study. Sixty-four families completed treatment, and 36 families dropped out of treatmen t. Mean ages for all participants are as follows: children (n = 100), 4.8 years (range 3 to 6 years); mothers (n = 100), 35 years; and fathers (n = 54) 38 years. Ei ghty-one percent of ch ildren are identified as white, 7% are identified as black, and 10% are identified as bi-racial and 2% as Hisp anic. Eighty-two percent of mothers and 51% of fathers id entified themselves as white. Of the mothers and fathers who participated in treatment, 57% and 91%, respecti vely, were married. Seventeen percent of the mothers were single, another 17% were divorced, and 6% were separated.
60 Measures Demographic Questionnaires Parent questionnaires w ere used to collect demographic information about the child and family including age, gender, ethnicity, parent occupation and education, and marital status. These questionnaires were filled out by mothers and fathers at the pre-treatment evaluation. Child Cognitive Status The Peabody Pictur e Vocabulary Test Third Edition (PPVT-III; Dunn & Dunn, 1997) is a well-standardized test that measures receptiv e language in individuals, ages 2.6 years through adulthood. Raw scores are converted to standard scores (SS) with a mean of 100 and a standard deviation of + 15. Split half reliability coefficients for children range from .86 to .97, with a median of .94. Test-retest reliabilities range from .91 to .94. The correlation between the PPVTIII and the WISC-III Full Scale IQ is .90. The st andard score of the PPVT-III was used to determine a childs eligibility to participate in the larger study. Parent Cognitive Status The Wonderlic Personnel Test (W PT; Dodrill, 1981) is a 51-item test designed as a screening scale of adult s intellectual abilities. The test score is the number of items answered correctly in 12 minutes. In a sample of 120 norma l adults, the Wonderlic es timate of intelligence correlated .93 with the WAIS Fu ll Scale IQ score, and the Wonde rlic score was within 10 points of the WAIS IQ score for 90% of the particip ants. Sex, education, le vel of intelligence and emotional adjustment were not found to affect the observed correlations These findings were replicated by Dodrill and Warner (1988) and ha ve been extended to psychiatric (Hawkins, Faraone, Pepple, & Seidman, 1990) and academic se ttings (McKelvie, 1989). The Wonderlic standard score was used to determine parent eligibility to part icipate in the study.
61 Outcome Variables Patterns of Change Child Disruptive Behavior: The Ey berg Child Behavior Inventory (ECBI; Eyberg & Pincus, 1999) a 36-item parent ra ting scale of disruptive behavior s in children between the ages of 2 and 16, will be used to measure changes in child disruptive behavior over the course of treatment. It contains two scales: The Intens ity Scale measures the frequency of childrens behavior on a 7-point scale from (1) never to (7) always, and the Problem Scale measures the degree to which the childs behaviors are problematic for the parent on a yes-no scale. The Intensity and Problem Scales of the ECBI have shown internal consistency coefficients of .95 and .93; inter-rater (mother-f ather) reliability coefficients of .69 and .61; and test-retest reliability coefficients of .80 and .85 across 12 weeks. Cronb achs alpha for the sample in this study was .82 for the Intensity Scale. Treatment outcome results from several studies have shown the ECBI to be a sensitive measure of treatment change in clinic-referred children (Webster-Stratton & Hammond, 1997). A recent study (Funderburk, Eyberg, Rich, & Behar, 2003) examined test-retest reliability of the ECBI across a longer interval (10 months) in an untreated school sample and yielded reliability estimates of .75 for both Intensity and Problem scales. In addition, paired t -tests showed that scores were as likely to decrease as to increase over time for th e intensity score. Demonstration of long-term stability of the ECBI provides evidence of measurement stability over longer periods of time and provides reassurance that changes in ECBI scores over time reflect meaningful change rather than measurement error. The ECBI was administered at the pre-treatme nt evaluation and at the beginning of each session during the course of treatment. The ECBI was also used as a criterion measure to determine a familys readiness to complete treatment. The completion criterion for graduation
62 was an Intensity Scale raw score of < 114, which is a standard deviation ( SD ) above the normative mean (Colvin, Eyberg, & Adams, 1999). Intensity Scale raw scores collected at each session will be used to examine patterns of change in childrens disruptive behavior over the course of treatment. Parenting Stress. The short form of the Parenting Stress Index (PSI-SF; Abidin, 1995) is a 36-item parent self-report scale containing thre e factor-analytically derived subscales (Parental Distress, Parent-Child Dysfunctional Interaction, and Difficult Child). The short form subscales have shown Cronbachs alphas of .80 to .91 and 6-m onth test-retest reliabili ties of .68 to .85. On the long form of the PSI, higher scores have been associated with increased severity of conductdisordered behavior (Eyberg, Boggs, & Rodri guez, 1992; Ross, Blanc, McNeil, Eyberg, & Hembree-Kigin. 1998). The PSI-SF was collected at the pre-treatment evaluation and at each treatment session. Change in Parent and Child Outcomes Change in Child Behavior Problems. The Child Behavior Checklists (CBCL/2-3; Achenbach, 1992 and CBCL 4-18; Achenbach, 1991a) we re adm inistered at the pre-treatment evaluations and the CBCL aggression scale on bot h forms was used to determine a research diagnosis of ODD. The CBCL/4-18 is a comprehensive instrume nt designed to assess the frequency of a variety of child behaviors during the past 6-month period. It c onsists of 118 behavior-problems items rated by the parent on a 3-point scale from (0) not true, to (2) very true or often true. The CBCL/2-3 is similar in format to the CBCL/4-18 and consists of 99 items. Both questionnaires include factor analyzed narrow band scal es (e.g., Anxious/Depressed, Delinquent, and Aggressive) and two broadband scales of extern alizing and internalizi ng behavior problems.
63 The broadband problem scales of the CBCL/4-18 ha ve mean test-retest reliabilities of .89 and .75 over a one-week and one-year period, resp ectively. Test-retest reliability for the CBCL/2-3 has been reported to range from .79 to .92 for the problem scales over a one-week period and .56 to .76 over a one-y ear period (Crawford & Lee, 1991). The pre-treatment CBCL Externalizing T score will be subtracted from the post-treatment CBCL Ex ternalizing T score to obtain a change score for each child. The CBCL Exte rnalizing change score will be examined as an outcome variable in the regression analyses. Change in Maternal Stress. The Parent Daily Hassles (PDH; Crnic & Greenberg, 1990) is a 20-item self-report questionnai re measuring stressful events in parenti ng and parent-child interactions. Parents rate the frequency, degree, and intensity of each hassle. The Frequency and Intensity scales have internal consistency coefficients of .81 a nd .90. The authors found that the cumulative effects of relatively minor stresses were important predictors of conduct-disordered behavior in a non-referred sample. Greater mother-reported hassles have also been related to greater difficulty managing toddler behavior (Belsky, Woodworth, & Crnic, 1996). The pretreatment PDH intensity scale will be used to meas ure level of maternal st ress and it relationship with rate of change in child disruptive behavior. Change in Maternal Perceived Social Support. The Multidimensional Scale of Perceived Social Support (MSPSS; Zimet, Dahlem, Zimet, & Fa rley, 1998) is a 12-item selfreport questionnaire designed to assess the percei ved adequacy of social support from family, friends, and significant others. Items are scored on a 7-point scale that ranges from very strongly disagree to very strongly agree. Cronbachs alpha for the three subscales ranges from .85 to .98 in various samples (Dahlem, Zimet, & Walker, 1991). Intern al reliability for the total score has ranged from .80 to .92. Test-retes t reliability over three months ranged from .72
64 to .85 for the subscales and was .85 for the total scal e. Factor validity has been demonstrated for the three-subscale factor structur e and concurrent validity for two of the three subscales has been shown. The pre-treatment maternal ratings of to tal perceived social suppo rt will be used to examine the effect of perceived social support on th e rate of change in ch ild disruptive behavior and parenting stress over the course of treatment. Change in Attachment Security. The Attachment Q Sort ta sk was used to assess the quality of attachment between mother and child, as rated by the mother, at pre-, midand posttreatment. The Q sort consists of a 90 cards that describe a speci fic behavior characteristic of children between the ages of 12 mont hs to 4 years of age. Mothers were asked to consider their childs typical behavior and rate their child on ea ch specific behavior by sorting the card into one of several piles ranging from most like my child to least like my child. The mothers were asked to complete three different sorts. In the first sorting task, mothers were able to put as many cards in a particular pile, based on their obser vations of their childs behavior. They were then asked to resort the cards again, placing the card in the pile they felt most appropriately described their child. In the fi nal sort, mothers were limited to the number of cards they could put in each pile (10 cards per pile with 9 piles total). Mothers so rting were then compared to an ideal prototypical secure child de rived from expert sorting and rate d a score from -1 to +1. The Q sort score is a correlation i ndicating the strength of the relationship between the profile derived from the mothers sort and the ideal profile and it reflects the similarity of the rated childs profile to the ideal secure child profile. In a meta-analytic study examining the reliabil ity and validity of the attachment Q sort measure across 139 studies and 13,835 children, van Ijzendoorn a nd colleagues (van Ijzendoorn, Vereijken, Bakermans-Kranenburg, and Riksen-W alraven, 2004) found adequate convergent
65 validity (r = .23) with the strange situation paradigm and discriminant t validity (r = .16) with a measure of temperament. They also found that average ratings of attachment were .10 points higher in clinical samples (.31) compared to norm ative samples of children (.21). It is important to note that these findings were based on observer ratings rather than self-report ratings, which demonstrated weak convergent reliability to th e strange situation paradigm (r = .16) and poor discriminant validity from a measure of temperament (r = .36). Change in Q sort was examined as an outcome, based on patterns of change in child disruptive behavior and parenting stress in CDI and PDI. Pre-trea tment scores were substracted from post-treatment scores to obtain change scores. Treatment Completion, Early versus Late Dropout. Patterns of change in child disruptive behavior and parenting stress will be examined as predictors of treatment completion and premature termination. Treatment status an d dropout status were coded dichotomously (0 = dropout, 1 = completer and 0 = CDI dropout and 1 = PDI dropout). Predictors of Patterns of Change Child Factors Child co-morbidity: Th e Diagnostic Interview Schedule for Children (NIMH DISC-IV-P; Shaffer, Fisher, Dulcan, Davies, Piacentini, Schwab-Stone, Lahey, Bourdon, Jenson, Bird, Canino, & Reiger, 1996) is a structured diagnostic interview for administration to parents. It includes all common disorders of children in accordance with the DSM-IV that are not dependent on specialized observations or te st procedures. Individual modules can be administered separately. One-week test-retest re liability on administration to parents of 9 to 17 year old children has been reported at .79 fo r ADHD, .54 for ODD, and .54 for CD (Fischer, Rolf, Hasazi, and Cummings, 1998). A comorbid diagnosis of ADHD will be examined as child
66 factor to test whether a co-morbid diagnosis of ADHD influences the rate of change in child disruptive behavior over the course of treatment. Child gender: Child gender will be examined as a dichotomous predictor in the multilevel models of change in child disrup tive behavior and parent ing stress. Boys were assigned a value of 0 and girls assigned a value of 1. Parent Factors Maternal D epression : The Beck Depression Inventory-II (BDI-II; Beck, Steer & Brown, 1996) is a 21-item self-report scale of adu lt depressive symptomatology. The BDI-II was modified from its previous versi on to make it more consistent w ith the DSM-IV criteria in item content and time frame. Severity of depression can be scored according to four levels: minimal (0-13), mild (14-19), moderate ( 20-28) and severe (29-63) (Beck, Steer, Ball, & Ranieri, 1996). Internal consistency (.91) and one-week test-retest reliability (.93) have been reported for the BDI-II. The BDI was collected a pre-, mid-, and post-treatment. These ratings will be entered as a co-varying predictor to determine whether changes in maternal depression across PCIT were associated with patterns of change in child disruptive be havior and parenting stress. Family Factors Socioeconomic Status: Socioecono mic status (SES) will be calculated according to the Hollingshead Index of Social Status (Hollingshead, 1975). The Hollingshead index is computed according to a four-factor index th at includes parent educational attainment and occupation. Pretreatment SES was entered into the multilevel model as a predictor of patterns of change in child disruptive behavior a nd parenting stress. Perceived Barriers to Partic ipation in Treatment. The Barriers to Participation in Treatment Scale (BTPS; Kazdin, Holland, Crowley, & Bret on, 1997) assesses perceived barriers to attending and completing treatment. The 52-item scale can be administered in person or by
67 telephone. Items can be divided into four subs cales, including (a) stressors and obstacles that compete with treatment, (b) treatment demands a nd issues, (c) perceived relevance of treatment, and (d) relationship with the ther apist; however, results from a pr eliminary factor analysis found that most items loaded onto a sing le factor of perceived barriers Therefore, only a total score will be calculated for this study. Internal consis tency for the total barrie rs score for the parentcompleted BTPS was .86. In other studies, th e BTPS has predicted higher rates of attrition, fewer weeks in treatment, and higher rates of cancellations and no shows (e.g., Kazdin & Wassell, 1998). Ideally, differences between completers and dropouts should be examined; however, few dropouts have returned the BTPS. Therefore, relationships between the BTPS total score with rate of change over the course of treatment will be examined in completers. Procedures Fa milies were referred to the Child Study Laboratory from the University of Florida Health Science Center Psychology Clinic. Primary referral sources included pe diatric neurology, child psychiatry, and general pediatric practices. Fa milies were also referred from community mental health practitioners and presc hools. A few families were self -referred. Before their first evaluation appointment, each family was sent a welcome packet that included directions to the Health Science Center, a parking pass, and a de mographic questionnaire. At the first of two pretreatment assessment visits, informed consent was obtained from all families and the limits of confidentiality were reviewed. Pa rents and children were then screened to determine if they met inclusion criteria. Screening measures include d the aggression scale of the CBCL, which was scored to determine the first of the combinati on criteria necessary for a research diagnosis of ODD. If the CBCL aggression score was at or above a T score of 61, the family proceeded with the screening. Clinical and structured diagnos tic interviews were completed, and the assessor determined if the child met DSM criteria for a diagnosis of ODD. The parent and the child were
68 then administered a cognitive screening measure, which required a SS of 75 or higher. If a family failed to meet any of the inclusi on criteria (CBCL aggression scale T score > 61, DISC ODD diagnosis, and cognitive screening standard scores > 75), they were referred back to the Psychology Clinic. During the pre-treatment evaluation, multiple measures of child, parent and family functioning were collected. During the structur ed diagnostic interview, used in screening, information was also obtained on four additi onal Axis 1 disorders: Attention Deficit Hyperactivity Disorder (ADHD) Separation Anxiety Disorder (SAD), Conduct Disorder (CD) and Major Depressive Disorder (MDD). The first assessment visit also included observations of the parent-child interaction in three structured play situations. In between the first and second pre-treatment evaluations, the family was contac ted by telephone over five consecutive days to obtain daily diary information on behavior problems and disciplin e and to administer several questionnaires. For children enrolled in school, school evaluation was con ducted also during the pre-treatment evaluation and cons isted of direct behavior obser vations in the classroom and administration of teacher questionnaires. Duri ng the second evaluation appointment, parents were administered a structured interview of prio r service utilization, including medication management, other forms of child or family ther apy, and school services such as individualized education plans and special classroom placement. The mother was also administered a card sorting task to assess the childs attachment and another set of parent child interaction observations was conducted. A mid-assessment evaluation was conducted betw een the CDI and PDI phase of treatment. The mid-assessment evaluation included administ ration of the card-sorting task to assess the childs attachment after CDI a nd parent questionnaires, includ ing the BDI and other measures
69 not examined in this study. At the completi on of treatment, families underwent the posttreatment evaluation, which was similar in structur e and content to the pre-treatment evaluations, including clinical interview, di agnostic interview, play observati ons, school observations if the child was in school at pre-treatment, a nd parent questionnaire s and rating scales. Once families completed the pre-treatment evaluation, they began treatment. Families remained in treatment until (a) parents met both CD I and PDI criteria, (b) the ECBI Intensity raw score was < 114, (c) children were rated by parents as meeting fewer than 4 ODD symptoms on the DSM-IV ODD Rating Scale, and (d) parents felt ready and confident to end treatment. If these completion criteria were met, the family graduated from treatment and moved into the post-treatment evaluation and follow-up phase of the study. The length of time (as measured by number of total treatment sessions ) to complete treatment in this study varied from as few as 9 sessions to as many as 31 sessions. On aver age, families completed PCIT in 17 sessions, progressing through CDI in 6 sessions (range 3 to 14 CDI sessions) and PDI in 11 sessions (range 5 to 24 PDI sessions). Analyses Descriptive Statistics Characte ristics of the sample will be analyzed and descriptive statistics will be reported for child, mother, family, and treatment variables to be analyzed in this study. Mean and standard deviations will be reported for participant characteristics (child and mother age, mother education, family socioeconomic status, child and mother cognitive abilities as measured by screening tools used in the study). Relationships among the rredictor and treatment outcome variables will be explored and reported, in addi tion to their means,standard deviations, and distributions (skewness/kurtosis). Frequencie s will be reported for the following child and mother characteristics: racial/et hnic make-up of the sample, child gender, mother marital status,
70 child research diagnoses, treatment status, and children taking pr escribed psychotropic medication. Reliability Sam pling reliability statistics will also be re ported for the ECBI In tensity Scale and the four PSI-SF scales. Sampling reliability in multil evel modeling is a measure of how reliable the measurements or scores of child disruptive behavi or and parenting stress are within each person over time. Generally speaking, sampling reliability improves as variation between participants increases, variation within participants decrease s, and the number of occas ions of measurements increases. Reliability estimates provided by the multilevel modeling software range between 0 and 1, with estimates closer to 1 indicating adeq uate sampling reliability of measurements taken for each participant. Analysis of Change To m odel change in child disruptive behavior and parenting stress over the course of treatment, a multilevel model approach will be used to analyze the session-by-session data collected over the course of treatment. More specifically, a random regression coefficient (RRC) model will be used to examine the session-by-session data. These session-by-session data include the Intensity scale raw scor e on the ECBI and the four subscale T scores on the PSI-SF (PSI-SF Total, Parenting Distress subscale, Pare nt Child Dysfunctional In teraction subscale, and Difficutl Child subscale). Prior to specifying th e occasion-specific (level-1) and person-specific (level-2) RRC models that will be used to ex amine change in PCIT, methodological challenges particular to this study will be discussed, namely, the unbalanced design of the study. Dealing with the Unbalanced Design A balanced design refers to a study in whic h all occasions of measurement are equally spaced apart and all participants have an equal number of measurements with no missing data
71 points. For example, a longitudi nal study assesses part icipants at baseline, 3 months, 6 months and 12 months, and all participants are assessed at each data point. An unbalanced design is the opposite, in which the occasions of measurem ent are unequal between participants and participants vary in number of measurements, th erefore leading to missing data. The design of this study is unbalanced, as spacing between occasions of measurement are unequal among participants, number of occasions varies between participants, and participants are missing data for certain data points. As is the case in most treatment studies, data for a particular participan t at a particular time point may not be available for various reasons. A therapist may have forgo tten to administer an ECBI at a given session or a pa rent may have forgot ten to fill out the backside of the questionnaire. Fortunately, multilevel modeling uses all available data for each participant; therefore, participants with missing da ta will be included in the analysis. Data for a participant may also be missing due to premature dropout from the study. A type of multilevel modeling, called pattern-mixture random effects modeling (Hedeker & Gibbons, 1997), handles such missing data by dividing par ticipants on the basis of their missing-data pattern. The grouping variable is then used to examine the effect of th e missing-data pattern on the outcome under investigation. To examine the relationship between dropout and completer patterns of change over the course of treatment, a grouping variable will be created for dropouts. Thirty-six families prematurely dropped out of PCIT Of the 36 dropout families, 19 dropped out in the CDI phase, completing an average of 4 sessions (range 1 to 11), and 17 families dropped out in the PDI phase, completing an average of 7 CDI sessions (ra nge 4 to 12) and 7 PDI sessions (range 1 to 20). In the analysis, dropout families will be classified as either an early or late dropout The
72 early dropout group will include families who dropped out before meeting criteria for CDI completion. The late dropout group will include families who at least one PDI session. The second reason for the unbalanced design is related to the time-unlimited and phasic nature of the treatment. PCIT has two phases (CDI and PDI), and families require different numbers of sessions in each phase of treatment before they can move to the next phase or complete treatment. This results in an unequal number of participants per occasion (per session, such that a different set of participants will be providing data for estimation of rate of change at each time point) and an unequal number of occasi ons (sessions) per participant. In addition, length of time between occasions is unequal. Ea ch treatement session is an occasion. Though it was planned that each participant would be seen on a weekly ba sis during treatment, cancellations, no-shows, and rescheduling of appo intments resulted in varying lengths of time between sessions for each participant. Fort unately, multilevel modeling does not require an equal number of participants at each occasion (session), and it does not assume an equal number of occasions for each particip ant. In addition, multilevel modeling can accommodate unequal spacing between occasions for each participant. That is, the spacing between occasions and the number of occasions per participant can be spec ified as random, rather than fixed, effects, meaning that these variables are allowed to vary from person to person (Cudeck & Klebe, 2002). By taking into account that thes e components vary from person to person, their effects on the outcome variable are essentially controlled. In this study, number of days between sessions will be used to account for th e unequal spacing of sessions and will be allowed to vary between participants. Choosing a Model Regression plots for all partic ipants should be exam ined to assist in determining the potential patterns of change over time and thereby assist in determining the most appropriate
73 model specification for analysis of change over time. Model speci fication refers to the equation chosen to analyze the pattern of change. Equations can specify wh ether the pattern of change is linear, quadratic, cubic or non-li near. The most appropriate mode l specification is one that is both parsimonious and well fitted to the data (Wille tt, 1988). Examination of individual regression plots provides preliminary information about the shape of the pattern of change and whether the model should include higher-order (i.e. quadratic, cubic, or nonlinear) equations to capture curvilinear or non-linear patterns of ch ange. Studies with large sample sizes (N > 200), in which examination of each individual trajec tory would be burdensome, include high-order equations at the outset and then conduct model comparison tests to determine the best-fit pattern of change model. For this study, the moderate sample size do es not hinder examination of individual regression plots to determine the inclus ion of higher-order specif ications in the model. Individual regression plots were examined by phase of treatment and across treatment as a whole to determine the necessity of investigating higher order patterns of change. A curvilinear pattern was evaluated initi ally and was not found to be signifi cant. Hence, higher-order patterns were abandoned in favor of the more parsimoni ous linear model, which was significant (see Result section). The Random Regression Coefficient Model The term random regression coe fficient (RRC) model refers to a general type of model typically used with within-participants data, such as the kind collected in this study. As with all other types of multilevel models, the RRC model in cludes two levels th e level 1 model (also referred to as the occasion-specific or within-p articipants model) and the level 2 model (also referred to as the person-specific or between-partic ipants model). Patterns of change in child disruptive behavior and parenting stress will first be examined without child, parent, or family predictors in the level-1 model. The effects of child, parent, and family predictors will then be
74 examined in the level-2 or between-participants mode l. Patterns of change will first be examined by phase at the individual level (level 1), so th ere will be a trajectory for ECBI scores and PSISF scores for CDI and PDI. The level 1 model Change in c hild disruptive behavior and parenting stress will first be modeled at the within-participants level. The general RRC equa tion for the level-1 model is as follows: Yti = 0 i + 1 iTti + eti (2-1) Let Yti denote the ECBI Intensity scale score or PSI-SF T score for the ith participant on the tth occasion, with t = 1,, n. The variable Tti denotes the value in time (elapsed time between sessions) for the tth measurement of participant i Recall that occasions and length of time between occasions varies between participan ts and are a random effects. The term 0 i is the intercept, which is a measure of the participan ts average level of di sruptive behavior or parenting stress at time 0 ( Tti = CDI Teach). The slope, which re presents the rate of change in child disruptive behavior and parenti ng stress over time, is expressed as 1 i. Finally, the term eti denotes the residual variance within participant i across t observations. A decrease in ECBI Intensity scale scores and PSI-SF subscale scores over time has been hypothesized in this study. Using variable labels instead of letters, the leve l-1 model to analyze chan ge in child disruptive behavior and pare nting stress is: ECBIti = 0 i + 1 iTimeti + eit PSI Totalti = 0 i + 1 iTimeti + eit PSI PDti = 0 i + 1 iTimeti + eit PSI PCDIti = 0 i + 1 iTimeti + eit PSI DCti = 0 i + 1 iTimeti + eit
75 Modeling predictors The between-participants (level 2) m odel ex amines individual differences in change in child disruptive behavior and parenting stress over the course of PCIT. The general RCC equation for the level-2 model is as follows: 0 i= 00 + 01Zi + u0 i (2-2) 1 i= 10 + 11Zi + u1 i (2-3) The first model specifies that the intercept, 0 i, is a function of the grand mean, 00, which represents the average ECBI or PSI-SF score at Tti = 0 = CDI Teach, the average difference, 01, between individuals in their average ECBI or PSI-SF, and a residual term, u0 i, which represents how far each individual deviates from the inte rcept. The second model specifies the slope, 1 i, according to the average slope, 10, for the sample, the average difference, 11, around the slope, and the residual term, u1 i, which represents how far each indi vidual deviates from the average slope. The term, Zi, in both equations refers either to the grouping variable or the predictor variable. In this study, the groupi ng variable will refer to dropout status (completer versus early dropout versus late dropout). Pr edictor variables include child factors (presence of comorbid ADHD, gender, and pre-treatment CBCL externalizing T score as a measure initial severity of behavior problems), parent factors (maternal BDI II score, PDH Frequency score, and maternal perceived social support score), and family factors (Hollingshead score and perceived barriers to participation in treatment total score.) Using variable labels instead of letters, below is an example of the level-2 model in cluding a predictor of interest: 0 i= 00 + 01ADHDi + u0 i 1 i= 10 + 11ADHDi + u1 i Essentially, these equations result in an inte rcept and slope accordi ng to ADHD diagnosis. The terms, 01 and 11, examine whether individual differenc es in overall level of disruptive behavior and parenting stress and individual differences in rate of change in disruptive behavior
76 and parenting stress over time can be explained by a co-morbid diagnosis of ADHD. Each predictor variable will be entered into the level-2 m odel as demonstrated above. Modeling maternal depression as a time varying covariate A tim e-varying covariate is a predictor variable that is also measured at repeated occasions, as maternal depression was measured in this st udy, at pre-, mid-, and pos t-treatement. It is hypothesized that changes in maternal depression will be related to changes in child disruptive behavior and parenting stress over the course of treatment. Specifically, an increase in maternal depression as measured by the BDI is expected to result in an increase in child disruptive behavior and parenting stress while a decrease in the BDI is expect ed to result in a decrease in the ECBI and the PSI-SF. This relationship between the BDI and the ECBI and PSI-SF is modeled as: Yti = 0 i + 1iTti + 2i (Xti meanXti) + eit (2-4) Similar to the level-1 model presented above, Yti represents the ECBI intensity scale score or PSI-SF score for participant i at occasion t and T is the value in time. The intercept and slope terms, 0 i and 1i still reflect the average ECBI or PSISF score and the average rate of change in scores over time, respectively. The term, Xti, is the time varying covariate, and 2i, represents the rate of change in the outcome variable in relationship to cha nges in the time varying covariate. In this case, Xti is the total maternal BDI score collected at pre-, mid-, and posttreatment. Each mothers pre-, mid-, and posttreatment BDI score will be subtracted from her overall mean BDI score. This results in a measure of changes in the mothers depression level relative to the mothers overall le vel of depression across treatment. Maternal depression will also be examined as a level-2 predictor and the model with maternal depression as a timevarying covariate is below: 2i = 20 + 21BDIi + u2 i (2-5)
77 The slope, 2i, is defined by the average rate of change in BDI scores for the sample, 20, the average difference in rate of change in BDI scores between participants, 21, and a residual error term, u2 i. Predicting Change in Treatment Outcome and Dropout from Patterns of Change in Child Disruptive Behavior and Parenting Stress A secondary aim of the study is to investigate whethe r the patterns of change derived from the multilevel model predict change in treatment outcome and dropout from this study. Relations between patterns of change and change in predictors will be examined using a traditional linear regression approach. Slope estimates from the treatment phase patterns of change will be entered as predictors in a li near regression model. Change scores will be calculated for each four outcome variables (the attachment Qsort, the CBCL, the Parent Daily Hassles Frequency scale, and the MSPSS) measuri ng change from preto post-treatment. A sample equation is presented below: Yti = 00 + 10(CDI Slope)ti + 20PDI Slope) + (2-6) To examine patterns of change that predict early versus late dropout, slope estimates from the patterns of change for dropouts will be entere d into a logistic regression model to determine whether patterns of change in CDI versus PDI can predict early ve rsus late dropout from treatment. The equation for the logistic regression model is presented below: Z(prob=1) = 1/1 + e -( + (slope value) + ) (2-7) e 2.718
78 CHAPTER 3 RESULTS Descriptive Analyses Child and mother dem ographic variables were analyzed with descriptive statistical methods. Descriptive statistics we re also calculated for predicto r and outcome variable measures. These data are presented in Tables 3-1 and 3-2 located at the end of the chapter. Thirty-one girls and 69 boys were enrolled in the study. Mean ages for all participants are as follows: children (n = 100), 4.38 years (range 3 to 6 years); mothers (n = 100), 33 years; and fathers (n = 54) 38 years. Se venty-six percent of children ar e identified as white, 8% are identified as black, 12% are identified as bi-racia l, and 3% as Hispanic. Eighty-three percent of mothers and 51% of fathers identified themselves as white. Of the mothers and fathers who participated in treatment, 58% and 91%, respecti vely, were married. Seventeen percent of the mothers were single, another 17% were divorced, and 6% were separated. Sixty-four families completed treatment, a nd 36 families dropped out of treatment. Treatment completers finished PCIT in appr oximately 17 sessions (C DI sessions, M = 5.76, SD = 1.85, PDI sessions, M = 11.17, SD 4.56). Of those who dropped out of treatment, nineteen families stopped coming to treatment in the CDI phase of PCIT (mean number of sessions = 4.16, SD = 2.56), and 17 families stopped coming to treatment in the PDI phase of PCIT (mean number of sessions = 14.05, SD 6.74). No di fferences were found in pre-treatment ECBI intensity scale scores or PSI total scores betw een treatment completers and dropouts, or between early versus late dropouts. Comple ters had significantly better scor es on the final ECBI intensity scale scores and PSI total scores compared to dropouts. No differenc es were detected in the final ECBI and PSI scores between early and late dr opouts; however, maternal report of disruptive behavior in children was closer to the clinical cut-off in the late dropout group (M = 139, SD =
79 41) compared to the early dropout group (M = 158.63, SD = 31.91). Maternal ratings of total parenting stress remained high in dropout group s (early drop-out M = 101.52, SD 24 and late drop-out, M = 93.11, SD = 28). Mothers in the sample were found to be of average intelligence (Wonderlic, M = 106.28, SD = 12.36). Twenty-six percent of mothers were college graduates, 6% had graduate education, and 40% received some college or vocational training. Eighteen pe rcent had high school diplomas, 6% had completed some high school and 2% had completed some middle school education. On average, families in the study earned approximately $2,615/month; however, the range of monthly income between partic ipants was large ($99 to $10,217). Mean SES (Hollingshead Index) for the total sample was 37.2 (SD = 13.1). Comparison of means between completers and dropouts on these demographical va riables revealed a sign ificant difference in SES between the two groups (F = 3.83, p = .05). Maternal ratings of childrens di sruptive and externalizing beha vior prior to treatment were in the clinical range (ECBI Intensity scale score M = 170.4, SD = 26.2, CBCL Externalizing T score M = 72.9, SD = 7.7). Mothers also reported significant levels of parenting stress (PSI Total score M = 103.8, SD = 17.7, PDH Frequency M = 50.7, SD = 10.1) at pre-treatment. Average ratings of maternal at tachment were low (Q-sort M = .03, SD = .21, range -.470 to 1.46 ), indicating that most children in the sample were not securely attached to their mothers at pretreatment, and mothers in general reported low levels of perceived social support prior to treatment (MSPSS M = 33.7, SD = 15.2). Mean ratings of maternal depression at pre (n=100, M =13.5, SD = 15.2, range 0 to 42), mid (n=57, M = 10.4, SD = 7.8, range 0-34), and posttreatment (n = 59, M = 10.4, SD = 7.8, range 0 to 42) were in the minimal range. No differences were detected between treatment comple ters and dropouts among these variables.
80 Only treatment completers reported on barrier s to participation in treatment. As can be seen in Table 3-2, mean post-treatment BTPS scores for mothers in the treatment completer group overall was 57.10 (SD = 13.4). The mean for the 1997 sample of treatment completers and dropouts in the Kazdin et al. study was 65.97 (SD=13.83), and total sc ore on the BTPS in that sample predicted poor performance in trea tment, including higher rate of attrition, fewer weeks in treatment, and higher rates of cancelled appointments. Children in the study were found to be of average intelligence (PPVT, M = 102.21, SD = 12.91). Inclusion criteria for the study dictated th at a child must meet diagnostic criteria for Oppositional Defiant Disorder (ODD), which wa s determined by the DISC interview and a CBCL externalizing T score > 61. Ninety-eight percent of the sample carried a primary research diagnosis of ODD and 45% of the sample carried a co-morbid research diagnosis of CD. Seventy percent of the sample carried a co-m orbid research diagnosis of ADHD, 24% of the sample carried a co-morbid diagnosis of SAD, a nd 4% of the sample met diagnostic criteria for MDD. Children in the study on average carried 2.45 research diagnoses. Only 20% of the sample carried one diagnosis at pre-treatment while 43% met criteri a for three diagnoses and 27% met criteria for two diagnoses. A small number of children met criteria for more than 3 diagnoses (10%). Table 3-1 below presents di agnostic data for the total sample. Twenty-six percent of the children in the study were on some kind of psychotropic medication prior to entering the study. As noted in the method s ection, children were required to be on a stable dose of their medication for 30 days prior to being evaluated for eligibility to participate in the study. Patterns of Change Analyses Multilevel modeling was used to analyze the pa tterns of change in ch ildrens disruptive behavior and mothers ratings of parenting stress across treatm ent and within the CDI and PDI
81 phases of PCIT. Unconditional models, or models without predictors such as time or treatment phase, were first run to analyze the variati on in the outcome variab les of interest. Any unexplained variance would pr ovide a basis for inclusion of predictors in the analysis of change models for child disruptive beha vior and parenting st ress. Results for the unconditional models for the ECBI and PSI scales are listed in Table 3-3 and 3-4. The unconditional means model examines change in ECBI Intensity Scale and PSI scale scores within individuals across occasions (day s between sessions starting with Day 0 = CDI Teach). Time is not included in this model as a w ithin-person or level-1 pred ictor. Initial status in the ECBI Intensity scale and PSI s cale scores at Day 0 are significant ( t = 46.395, df = 97.368, p < .001), suggesting that there are si gnificant differences between indi viduals in initial levels of disruptive behavior and overall pa renting stress at the start of treatment. Interclass correlations were calculated to obtain the proportion of total variation in the outcome variables of child disruptive behavior and parenting stress that lies between part icipants. Calculations yielded value of .60 for the ECBI and a value of .72 for the PSI total scor e. These values indicate that 60% of the total variation in ECBI Intensity Scal e and 72% of the PSI tota l score are attributable to differences between children at the start of treatment. At the bottom of the table are deviance statistics, which provide a measure of the goodness-of-fit of the model to the data. The general rule is the smaller the value the better the fit of the model. As can be seen, the deviance statistics are relatively large in value, indicating the model may not be the most parsimonious in explaining the variance in the outcome measures. Unconditional means models were also run for the total and subscales of the PSI the Parent Distress (PD) subscale, the Parent-Child Dysfunctional Interacti on (PCDI) subscale, and the Difficult Child (DC) subscale. As can be s een in Table 3-4, seventy-eight percent, 74%, and
82 63% of the total variation in each the respective subscales are attr ibutable to differences between mothers. The intercept values for each of th e subscales also indicate that mothers were experiencing high levels of stress in their parenting role, were e xperiencing negative interactions with their children, and perceived their children to be difficult to manage. The next models run in the analyses were unconditional growth models, in which the Time variable was added as a predictor to the model. Time in this st udy was defined as the number of days in treatment with Day 0 = CDI Teach. Days in between each session were calculated per family to determine the number of the day the next treatment session fell. All families began treatment on Day 0 (CDI Teach). The next treatment session, CDI Coach 1, however, fell on a different day for per family. For example, a number of families completed the CDI Coach 1 session on Day 7, whereas other families completed the same session on Day 10 or Day 5. Hence, the coding of time captured the unequal spacing between occasions (sessions), allowing for time to vary between families in the analys is. Results for these models for each of the outcome measures are listed in Table 3-5. Comparison of intercept estimat es between the unconditional and growth models reveals that some of the intercept estimates for outcome measures of interest in creased in value. Hox (2002) notes that the unconditional means-model tends to overestimate the occasion-level variance and underestimate the chil d-level variance, thereby infl ating the intercept parameter estimate. Inclusion of the Time variable models the occasion-level variance in the disruptive behavior and parenting stress measures and results in more realistic estimates for both the occasion-level and child-level variances. The inte rcept, which is the expected level of child disruptive behavior and parenting stress at Day 0 in treatment, remains statistically significant at the p < .001 level, suggesting that indi vidual differences remain between children in initial levels
83 of disruptive behavior and between mothers in initial levels of parenting stress at the start of treatment. Examination of predictors of interest at the next level may explain these differences. Rate of change, or slope, estimates for all outcome measures were also sta tistically significant at the p < .001 level and indicate that le vels of childrens disruptive behavior and mothers parenting stress are decreasing over time. The variance components in the unconditional grow th models also indicate there is still unexplained residual variance both within-per sons and between-subjects. The term, R2 reflects the proportional reduction in the within-person variance component, 2 i, between the unconditional means and growth models with the a ddition of time in the model. Each percentage listed in the table reflects the amount of within-person variation in the ECBI and PSI scales associated with linear time, here conceptualized as treatment. Therefore, fifty-five percent of the within-person variation in the PSI total score, or overall levels of parenting stress, is associated with time in treatment. The level-2 variance components re flect individual differences between subjects in initial status ( 2 0) and rates of change (2 1) over time. Variance estimates for initial status should be interpreted with caution because the intercept estimates changed with the addition of time in the model. Variance estimates for rates of change are statistically significant ( 2 < .05) for all outcome measures, suggesting individual differences in rates of change over time in childrens disruptive behavior and maternal parenting stress. Examination of predictors of interest may further explain these differences. Predictor Models Several hypotheses were proposed about the eff ects of certain parent and child predictors on patterns of change in disruptive behavior and parenting st ress. A few of these hypotheses also proposed predictions about rates of change betw een treatment phases, CDI versus PDI.
84 Therefore, a model examining treatment phase wa s first analyzed to determine if differences exist in the rates of change according to treatment phase. Child and parent predictors of interest were then entered as predictors in the cha nge models to determine whether they explain remaining variation within-p ersons and between-subjects. Treatment Phase Model Treatm ent phase was entered into the change model as a dichotomous variable (CDI = 1 and PDI = 2). To refresh, all families received tr eatment in the same order, with the CDI phase first and graduation to the PDI phase once parents demonstrated mastery of the PRIDE skills. Some families remained in one phase of treatment longer than others; therefore, the number of occasions per treatment phase was conceptualized and analyzed as a time-varying predictor because length of time in each phase of treatment varied between families. Interpretation of the parameter estimates di ffers with the inclusion of a time-varying predictor in the level 1 model. The level 1 model equation for this analysis is written below. Yit = 00 + 10 (Dayit) + 20 (Txphaseit) + 30 (Day X Txphaseit) (3-1) The intercept value remains largely similar in interpretation: 00 represents the average ECBI or PSI score on the first day of treatment. The slope parameter, 10, is now a conditional rate of change and represents the average rate of change in ECBI or PSI scores controlling for the effect of phase of treatment. The time varying predictor, treatment phase ( 20), represents the average difference, over time, in ECBI or PSI scores between the CDI and PDI phase and can be conceptualized as a main effect Last is the interaction term between time (measured as number of the day in treatment) and treatment phase, 30. This term looks at whether rate of change in ECBI or PSI scores differs over tim e between each phase of treatment. Results of the treatment phase model for each outco me variable are listed in Table 3-6. It is important to note that two m odels were analyzed: the main effect model which included the
85 treatment phase variable and the interaction effect model which included the interaction term between time and treatment phase. This follows typical protocol in analyzing multilevel models. In Table 3-6, the results for the interaction e ffect model are listed for all outcome measures, except the Difficult Child subscale of the PSI. Th e main effect model for the DC subscale was significant, whereas the inclusion of the intera ction term rendered both the main effect and interaction term insignificant. Hence, only the si gnificant finding is shown below. Both the main effect and interaction models yi elded non-significant fi ndings for the Total, Parent Distress, and Parent-Child Dysfunctiona l Interaction subscales, hence, only the interaction model results are shown. With respect to the ECBI, the effect of tr eatment phase was found to significantly vary over time, indicating that the rate of change in ECBI scores over time differed between the CDI and PDI phase of treatment. As can be seen in Graph 3-1, the slope for the CDI phase of treatment is steeper than the slope for the PDI phase of treatment. A si gnificant result was also found for the Difficult Child subscale of the PSI, in which treatment phase was found to significantly affect the average rate of change over time in moth ers perceptions of their children as difficult to manage between the CDI and PDI pha se of treatment. Graph 3-1 illustrates this average .97 point difference in Difficult Child su bscale ratings between the CDI and PDI phase of treatment, such that scores changed at a sli ghtly steeper rate in CDI (of an average of .97 points) compared to PDI. Hypothesis 1: Children with ADHD are hypoth esiz ed to demonstrate greater changes in levels of disruptive behavior and parenting stress in PDI than CDI. To test these hypotheses, the predictor, ADHD, was entered as a level2 predictor into the existing treatment phase model for the ECBI and DC subscale of the PSI. Results for the ECBI and DC subscale are presented in Table 3-7. A separate model was also run for the other PSI
86 scales to explore whether a dia gnosis of ADHD affected the rate of change in parenting stress over the course of PCIT. These results are presented in Table 3-8. Results were not supportive of the hypothesis that children with ADHD exhibited greater change in disruptive behavior or that mothers ex hibited greater change in maternal stress related to difficulty managing in children in the PDI pha se of treatment. Treatment phase remained a significant predictor in both models, suggesting that another substantive level-2 predictor may explain differences in rates of change in each outcome over time be tween CDI and PDI. Despite lack of support for the hypothesis a diagnosis of ADHD was a significant predictor with respect to initial levels of child disruptive behavi or at the start of treatment. Further examination of this finding revealed th at children with ADHD had significantly higher initial levels of disruptiv e behavior, as rated by their mothers, at start of treatment compared to children without an ADHD diagnosis. By substituti ng the actual values into the level-2 model, the average ECBI score at the start of treatment for children with ADHD differed from children without a diagnosis of ADHD by 17.54 (see Graph 3-2). It is important to note that children changed at similar rates over the course of PCIT regardless of a diagnos is of ADHD. This is reflected in the non-significant finding for the term 11 (ECBI = .09, p = .74; DC = -.007, p = .702). For the remaining outcome measures, a diagnos is of ADHD proved to be only significant with respect to initial levels of maternal stress related to the parenting ro le (PSI PD subscale). Again, mothers of children with a diagnosis of ADHD reported a higher level of parenting distress compared to mothers of children with no diagnosis of ADHD. Ac tual values of the parameter estimates were entered into the level-2 model and the average point-difference in the Parent Distress subscale is noted in Graph 3-2. Despite this average difference at the start of
87 treatment, mothers experienced similar rates of reduction in levels of parent distress over the course of PCIT. To measure the strength of the effect of the predicto r in the model, pseudo-R2 statistics were calculated for the significant findings. Desp ite their statistical signi ficance, a diagnosis of ADHD explains approximately 8% of the variation in initial levels of child disruptive behavior and 4% of the variation in initia l levels of parenting distress. These percentages, along with the significant variance components, su ggest the exploration of other predictors to account for the differences in initial levels and rate s of change in all outcome measures. Hypothesis 2: Children who are more securely attached at pre-treatment are hy pothesized to make greater change in CDI than PDI. Pre-treatment attachment Q sort scores were entered into the treatment phase model as a level-2 predictor to determine whether a childs le vel of attachment predicted differences in rate of change in child disruptive behavior between phases of treatment. Table 3-9 presents results. Results were not supportive of the hypothesis that children who were more securely attached at pre-treatment would demonstrate greate r change in CDI versus PDI. In addition, pretreatment attachment did not affect the average ra te of change in child disruptive behavior over the course of PCIT (11 = -.18, p = .72) nor did it explain any differences in initial levels of child disruptive behavior ( 01 = -27.45, p = 11). The interaction between days in treatment and phase of treatment ( 20) remained significant, suggesting that other predictors may explain the differences in rate of change between the CDI and PDI phase of treatment.
88 Hypothesis 4: The patterns of change for boys and girls w ill be examined separately to determine whether gender differences exist in the pattern of change in child disruptive behavior and parenting stress during PCIT. Gender was entered into the unconditional grow th model to determ ine whether gender differences exist in the patterns of change in ch ild disruptive behavior and parenting stress during PCIT. Results for gender are presented in Table 3-10. Gender was not a significant predic tor with respect to initial le vels and rates of change in child disruptive behavior or pare nting stress related to parent-ch ild dysfunctional interactions and difficulty managing childrens behavior. Gender did, however, significantly predict differences in initial levels of overall paren ting stress and in stress related to the parenting role, with mothers of girls reporting higher levels of stress on both outcome measures than mothers of boys. Graph 3-3 for these significant findings are presented at the end of the chapter. Despite the statistical significance of thes e findings, the effect sizes for gender are relatively small. Gender explained 5% of the vari ation in initial levels of overall parenting stress between boys and girls and 6% of th e variation in initial levels of stress related to the parenting role between boys and girls. Th ese percentages suggest that othe r substantive predictors can be included in the model to further explain variation in initial levels of parenting stress. In addition, the variance components for rate of change rema in significant, also indicating the need to explore other predictors to account for the variation in rates of change over the course of PCIT. Hypothesis 5: Families who report a greater nu mber of b arriers to treatment participation are expected to show less change (a flatt er slope of change) in PCIT compared to families who report fewer barriers. Families who successfully completed treatment f illed out the Barriers to Participation in Treatment (BPTS) questionnaire at post-treatment These retrospective sc ores were entered as predictors into the change model to determine whether families who reported greater number of barriers demonstrate less change in PCIT compared to families who reported fewer barriers. The
89 Time variable was re-coded to reflect the retrospec tive nature of this par ticular analysis. In previous and subsequent analyses Day 0 represents the first day in treatment and the number of the day of the following session indi cates forward movement in time. Time was re-coded so that Day 0 represents the last day in treatment. This allows for the post-treatment ratings to predict differences in patterns of change toward fina l status in PCIT and accurately reflects the retrospective nature of this analysis. Post-treatment ratings of barri ers to participation in treatm ent were found to significantly affect rates of change in child rens disruptive behavior over the course of PCIT. To determine the direction of the prediction, the estimate values for rate of cha nge were entered in the level-2 model equation below: 2 i = 10 + 11 (BPTSi) + (3-2) As the BPTS score is a continuous variable with no determined cut-off in the literature to delineate better or worse scores, the sample mean plus or minus the sample standard deviation (57.10 + 13.4), was used to delineate a high (70.5), average (57.1), and low score (43.7) on the BPTS. These values were plugged in to the level2 equation above to calcul ate prototypical slope values at each level: 2 i = 1.02 .01 (70.5) = .31 2 i = 1.02 .01 (57.1) = .45 2 i = 1.02 .01 (43.7) = .58 A high BPTS score, which reflects a mothers perception of greater number of barriers to participation in treatment, predicted less change ov er the course of treatment, compared to a low BPTS score, as seen in the estimated slope values of .31 and .58. This finding supports the proposed hypothesis; however, it is also evident by the estimated slope values that families at all levels of perceived barriers demonstrated promis ing reductions in childrens disruptive behavior
90 over the course of PCIT. In addition to the significant finding, retrospective report of perceived barriers to treatment accounted for 26% of the variation in rate of change in child disruptive behavior. Graph 3-4 provides a visual depiction of the rates of change at each level of perceived barriers to participation in treatment. Estimates for the change model are presented in Table 3-11 Hypothesis 6: Children from socioeconomicall y disadvantaged homes are hypothesiz ed to show a flatter slope of change in PCIT compared to children who are from less disadvantaged homes. Socioeconomic status was found to significantly predict individual differences in initial status in childrens disruptive be havior but not in rates of change of disruptive behavior over the course of treatment. Results are presented in Table 3-12. To calculate meaningful scores to interpret the direction of the prediction with resp ect to initial status, the Hollingshead Index score was also divided in the same manner as the BPTS score. 1 i = 00 + 01 (SESi) + (3-3) High and low values were calculated by plugging in the sample mean, plus or minus the sample standard deviation, into the above equation: 1 i = 180.13 .57 (24.15) = 166.36 1 i = 180.13 .57 (37.94) = 158.5 1 i = 180.13 .57 (51.73) = 150.64 As could be anticipated, an inverse relationship is seen between SES and childrens disruptive behavior, with a low Hollingshead Index (24.15) associated with a worse (more disruptive) initial ECBI score and initial ECBI score improving as SES index improves. Despite these differences in initial status, children cha nged at similar rates over the course of treatment. Low, average, and high values were also ca lculated for the slopes to demonstrate this phenomenon (see Graph 3-5). 2 i = -.51 + .003 (24.15) = -.44 2 i = -.51 + .003 (37.94) = -.40 2 i = -.51 + .003 (51.73) = -.35
91 Hypothesis 7: Mothers who are socioeconomically disadvantaged and report a greater number of barriers to treatment p articipation are hypothesized to show a flatter slope of change in PCIT on parenting stress rati ngs compared to mothers who report fewer barriers to treatment participation and who are not socioeconomically disadvantaged. Separate change models were analyzed to examine SES and BPTS as predictors of patterns of change in parenting stress, due to the retros pective nature of the BP TS ratings. The reverse coding of the Time variable (Day 0 = final day in treatment) was used to analyze BPTS as predictors of patterns of change in pare nting stress. The prospective coding of Time (Day 0 = first day in treatment) was used to analyze SES as predictor of patterns of change in parenting stress. Socioeconomic status was not found to be a si gnificant predictor in th e model with respect to rate of change or initial st atus for any of the PSI scales. N on-significant findings are presented in Table A-6 in Appendix A. The Barriers to Part icipation in Treatment (BPTS) did significantly predict differences in rate of change for all the parenting stress scales, as well as approach significance with respect to initia l levels of parent distress. Re sults are presented in Table 3-13. To determine the direction of the prediction, the sample mean, plus and minus the sample standard deviation were entered into the level 2 equation below to calcula te meaningful high and low values for the BTPS. 2 i = 10 + 11 (BPTSi) (3-2) Values for each of the total and subscales are listed below. For the Total PSI Scale: 2 i = .42 .005 (43.7) = .20 2 i = .42 .005 (51.7) = .17 2 i =.42 .005 (70.5) = .07 For the Parent Distress Subscale: 2 i = .11 .001 (43.7) = .07 2 i = .11 .001 (51.7) = .06 2 i = .11 .001 (70.5) = .04
92 For the Parent-Child Dysfunc tional Interaction Scale: 2 i = .11 .001 (43.7) = .07 2 i = .11 .001 (51.7) = .06 2 i = .11 .001 (70.5) = .04 For the Difficult Child Subscale: 2 i = .21 .002 (43.7) = .12 2 i = .21 .002 (51.7) = .11 2 i = .21 .002 (70.5) = .07 As hypothesized, mothers who reported a greater number of barriers demonstrated slower change (flatter slopes as indicated by the values of .07 for the total and DC scales and .04 for the PD and PCDI scales) in overall and specific levels of parenti ng stress; however, all mothers demonstrated change in the desired direction on all the scales of the PSI. Graph 3-6 provides a visual illustration of the patterns of change for each parenting stress scale. With respect to the significant findings in this model, pseudo-R2 statistics demonstrate the proportion of variance these predictors account for in the outcome measures. Socioeconomic status accounts for 18% of the vari ation in initial levels of overall parenting stress, and barriers to participation in treatment accounts fo r 7% of the variation in rate of change in overall parenting stress. For the Parent Distress subscale, barriers to participation in treat ment accounted for 4% of the variation in initial levels and 25% of the va riation in rate of change. Finally, the BPTS accounted for 32% and 13% of the variation in rate of change for the PCDI and DC subscales of the PSI, respectively.
93 Hypothesis 3: A relationship between changes in maternal depression and changes in child disruptive behavior and parenting stress are h ypothesized. High BDI scores at pre-, mid-, and post-treatment are expected to be a ssociated with less change (flatter slope) in child disruptive behavior and parenting stress, whereas BDI scores that decrease over the course of treatment are expected to be associated with a decrease in child disruptive behavior and parenting stress over the course of PCIT. Maternal ratings of depressi on were collected on three occas ions during the course of treatment: at the pre-treatment assessment, at the mid-treatment assessment, and at the posttreatment assessment. These individual ratings were subtracted from each mothers pretreatment depression score to obtain a with in-person centered value at each time point, representing each mothers deviation from her ini tial rating of depressive symptoms. Hence, the predictor, maternal depression, was allowed to va ry over time. Depression was entered into the change model as a time-varying pr edictor to determine whether ch anges in ratings of maternal depression were related to changes in children s disruptive behavior a nd in parenting stress during PCIT. Only subjects with complete data were included in this analysis (n = 63), as missing data was not well tolerated in the level1 model. Results are presented in Tables 3-14 and 3-15. Interpretation of the parameter estimates di ffers with the inclusion of a time-varying predictor in the level 1 model. The level 1 mode l equation for this analysis is written below. Yit = 00 + 10 (Dayit) + 20 (Centered BDIit) + 30 (Day X BDIit) (3-4) The intercept value remains largely similar in interpretation: 00 represents the average ECBI or PSI score on the first day of treatment. The slope parameter, 10, is now a conditional rate of change and represents the average rate of change in ECBI or PSI scores controlling for the effect of maternal depression. The pre-trea tment BDI score is ente red into the model ( 20) and controls for the effect of in itial depression ratings on changes in ECBI scores over time. The time varying predictor, 20, represents the average difference, over time, in ECBI or PSI scores
94 as mothers BDI scores increase or decrease. Last is the interaction term between time (measured in days) and the time-varying pred ictor of matern al depression, 30. This term looks at whether rate of change in ECBI or PSI scores differs over time in association with changes in mothers depression (as hypothesized). Regarding childrens disruptive behavior, chan ges in maternal depression were found to significantly affect the average rate of change in ECBI scores over time, but its effect on ECBI scores was not found to vary over time, such that ra te of change in ECBI scores did not vary in relation to changes in maternal depression. On av erage, childrens ECBI ratings were found to decrease at a rate of -.34 over th e course of treatment, controlling for the effects of maternal depression. Controlling for the effect of time (treatment) and mothers pre-treatment BDI score, a one unit change in mothers depression was asso ciated with an average 1.30 point-difference in childrens ECBI scores over time. The direction of this point-difference is dependent on the direction of the unit change in maternal depression. A decrease in mothers ratings would result in a negative number, which would render the valu e for the point-difference negative (and in the direction desired) for the average ECBI score. For the most part, mothers depression ratings decreased with time; however, if there was no change in mothers BDI score (equaling a value of 0 for 20), the average ECBI score changed at a rate of -.34. An in crease in mothers depression ratings, which rarely occurred, resulted in a positive value for the point-difference and was associated with less change in the average ECBI score over time. Graph 3-7 demonstrate the effect of maternal depression on change in childrens disruptive behavior, according to mothers initial levels of depression and change in mothers depression over the course of treatment. As can be seen in the graphs, childrens initi al levels of disruptive behavior increased with severity of maternal ratings of depressive symptoms. Decreases in
95 mothers ratings of depression we re associated with reduction of disruptive behaviors, with greater decreases in maternal depression resultin g in greater reductions in childrens disruptive behavior. Less change in mothers ratings of depr ession over the course of treatment resulted in flatter slopes for the ECBI ove r the course of treatment. The parenting stress models yielded slightly different results than the ECBI model. The main effect model, which only looked at th e effect of maternal depression on the average rate of change in PSI scores over time, was significant for the total and individual subscales of the PSI. The interaction effect between the centered BD I scores and time (days in treatment) was significant only with respect to the Difficult Ch ild Subscale of the PSI and rendered the main effect of maternal depression on this subscale non-significant. Th e effect of maternal depression on the Difficult Child subscale did vary over time and the rate of change on this subscale did change in relationship to changes in mothers depression over time, providing support for the hypothesis. Overall, decreases in mothers depression ratings were associated with decreases in mothers ratings of stress relate d to difficulty managing their chil ds behavior (see Graphs 3-8). Though the effect of maternal depression did no t vary over time with respect to individual changes in ECBI scores, a significant associa tion was present between changes in maternal depression and average change in childrens disruptive behavior over time. This association was in the predicted direction, with less change in mothers depression re sulting in less overall change for ECBI scores. For the PSI scales in which the interaction term was not significant, similar results were noted for the main effect of maternal depression. Su pport for the interaction effect was found with respect to the Difficult Child subscale of the PSI, which measures parenting stress related to mother s perception of how difficult her child is to manage. Mothers, who experienced a decrease in depressive symp toms over the course of treatment, reported an
96 improved perception of the manageability of their child. This and the main effect findings with the ECBI and other PSI scales provide support for the supposition that childrens behavior in this study changed not only in relationship to treatment but also in association with changes in maternal depression, a variable in the literature that has been proposed to affect not only the development of disruptive behavior in children, but also succe ssful outcome in treatment of disruptive behavior (Kazdin, 1999; Webster-Stra tton and Hammond, 1990). Predicting Treatment Outcome from Patterns of Change The second set of hypotheses in this study focu sed on whether patterns of change that em erged from the first set of analyses could pred ict specific outcomes in treatment. A few of the hypotheses proposed greater change in certain outc ome measures based on differences in rate of change between the CDI and PDI phases of treatm ent. The predicted slope estimates calculated for each participant were entered into linear re gression models, as proposed, to determine if patterns of change predicted change in treatme nt outcome; however, certain analyses failed to run or ran into issues with tolerance limits being met or predictors being removed from the regression model due to being constants. Henc e, the results from these analyses are not presented below. To evaluate changes in treatment outcomes ba sed on patterns of change in child disruptive behavior and parenting stress, two sets of al ternative analyses were conducted a multilevel analysis in which the intercept was redefined to re present final status at the end of treatment and traditional linear regression analyses. Both statis tical approaches address the questions presented in the hypotheses in different respects. The multilevel model conceptualizes an association between patterns of change in child disruptive behavior and parenting stress in CDI and PDI and change in treatment outcome from pr eto post-treatment. It is e xpected that different patterns of change between CDI and PDI will predict outcomes on specific parent and child variables.
97 Traditional regression similarly examines the pred ictive relation between changes in predictors and change in treatment outcome. Traditional regression is limited to including only treatment completers and change is modeled on a limited nu mber time points (preto post CDI and preto post PDI for ECBI and PSI scores and preto post-treatment for specific outcomes). For the multilevel model approach, the Time variable was re-coded to reflect the aim of predicting specific child and parent outcomes; hence, Time was re-coded to direct focus on predictors of patterns of change toward final status in treatment. With the Time coding reversed (Day 0 = final day in trea tment), the intercept, ( 00), now represents the average level of child disruptive behavior and parenting stress at the final session of PC IT. Rates of change in child disruptive behavior and parenting stress are conditio nal rates of change and represent the average rate of change in ECBI or PSI scores controllin g for the effect of phase of treatment. Treatment phase ( 20) remains a time varying predic tors and represents the average difference, over time, in ECBI or PSI scores between the CDI and PDI phase and can be conceptualized as a main effect. Last is the interaction term between time (meas ured as number of the day in treatment) and treatment phase, 30. This term looks at whether rate of change in ECBI or PSI scores differs over time between each phase of treatment. To examine whether patterns of change in child disruptive behavior and parenting stress predict therapeutic change in treatment outcome, change scores for the specific parent and child outcomes were calculated and entered into the multilevel model as predictors. A significant relationship between the intercept (final status) and/or pattern of change in child disruptive behavior and parenting stress a nd the change score would suggest that patterns of change during PCIT relate to therapeutic change in treatment outcomes. The outcomes of interest include change in parent-child attachment from preto post-treatment, change in child disruptive
98 behavior from preto post-treatment, change in parent daily hassles from preto post-treatment, and change in perceived social s upport from preto post-treatment. Five multilevel models examining the effects of treatment phase and specific outcomes on patterns of change in child di sruptive behavior and parenting stress were analyzed. The unconditional growth model including only the re-coded Time variable was first analyzed and found to be significant, as noted in Table 3-16, suggesting differen ces in patterns of change and final status in child disruptive behavior and parenting stress during PCIT. Treatment phase and the specific outcome predictors were then entered into the un conditional growth model. Two models returned promising findings in the predicte d direction. The three remaining models were non-significant (results presen ted in Appendix A). Promising findings are discussed below. Hypothesis 1: Children who show a steeper slop e on the ECBI Intensity Scale in CDI than PDI will show greater change in attachment security from preto post-treatment. Change in attachment security from preto post-treatment and treatment phase were entered as predictors into the pa ttern of change model for child disruptive behavior. Results are presented in Table 3-17. A trend was found between patterns of change in child disruptive behavior and change in attachment security from preto post-tre atment. Treatment phase did not predict differences in patterns of change as noted below. An interaction effect between treatment phase and change in attachment was also exam ined and returned non-significant findings ( 21 = 3.53, SE = 9.79, t = -.361, p = .719). Only the main effects model with the trend finding is presented in the table below. It was hypothesized that children who demonstr ate faster change in CDI than PDI would demonstrate greater change in attachment security from preto post-treatment. Given that patterns of change in child disr uptive behavior did not differ by treatment phase, differences in patterns of change for child disruptive behavior were examined across treatment as a whole.
99 Prototypical slopes were calculated in order to determine the direction of the prediction in the multilevel model. A low and high value for the attachment security (Qsort) change score was calculated by adding and subtrac ting the sample standard deviation for the Qsort change score from the sample Qsort change score mean (M = .232 + .305). These values were entered into the level-2 model equations for slope (rate of change) estimates. 2 i = 10 + 11 (Qsorti) + (3-6) 2 i = .24 + .29 (-.023) = .23 2 i = .24 + .29 (.587) = .40 The intercept estimate ( 00 = 132.35) reflects the average level of child disruptive behavior at the completion of PCIT, and the slope estimate ( 10 = .24), reflects the average rate of change in disruptive behavior during PCIT. Families who experienced greater change in attachment from preto post-treatment were found experience faster behavior change ( 2 i = .40) during PCIT, compared to families who experienced le ss change in attachment from pre to posttreatment ( 2 i = .23). Though change in attachment did not predict significant differences in final status at the end of PC IT, plugging the same low and high values into the level 2 intercept model demonstrates that families who experienced grea ter change in attachment from preto posttreatment completed treatment with levels of disruptive behavior well within the normative range ( 1 i = 119) compared to families who expe rienced less change in attachment ( 2 i = 133), whose children were one point above the clinical cut-off. Graph 3-9 pr ovides a visual presentation of the relation between patterns of ch ange in child disruptive behavi or and change in attachment security from preto post-treatment. Hypothesis 3: Mothers who show steeper s lope on the Parent-Child Dysfunctional Interaction subscale of the PSI-SF during CDI than PDI will show greater change on attachment security during PCIT. Change in attachment security from preto post-treatment and treatment phase were entered into the pattern of change model for the PCDI subscale of the parenting stress measure.
100 A trend was found for the interaction effect betw een treatment phase and change in attachment security. Results for the mode l are presented in Table 3-18. High and low values for the Qsort change scor e were entered into the level 2 equation for the interaction term, treatment phase x change in Qsort, to determine the direction of the trend. As can be seen by the prototypical estimates calculated below, families who experienced greater change in attachment from pre to post-treatmen t demonstrated faster change in parent-child dysfunctional interaction. Graph 310 below depicts this trend. 2 i = 20 (Tx Phase) + 21 (Qsorti) + (3-7) Treatment Phase CDI (1) 2 i = .062 (1) 2.05 (-.023) (1) = .05 2 i = .062 (1) 2.05 (.58 7) (1) = -1.21 Treatment Phase PDI (2) 2 i = .062 (2) 2.05 (-.023) (2) = .09 2 i = .062 (2) 2.05 (.58 7) (2) = -2.41 Though the prototypical slope values suggest that mothers dem onstrated faster change in parent-child dysfunctional interaction in PDI, this effect was not statistically significant, as noted by the non-significant main effect for treatment pha se in the table above. The interaction trend appears to be driven by change in attachment from preto post-treatment, with mothers who demonstrate greater change in attachment e xperiencing faster change in parent-child dysfunctional interaction dur ing PCIT as a whole; however, it is also important to note the main effect for change in attachment was also not signi ficant, raising concern that the interaction trend may be a spurious finding. Traditional Linear Regression Analyses Traditional linear regression m odels were al so analyzed to examine whether change in child disruptive behavior and pa renting stress in CDI and PDI predicted change in specific
101 outcomes. Change scores for CDI and PDI were calculated for ECBI and the three subscales of the PSI-SF (PD, PCDI, and DC) by subtracting the score from the CDI and PDI Teach session from the score from the final CDI and PDI sessi on. The CDI change score and the PDI change score were entered as predictors in the linear regression models. Cha nge scores were also calculated for the four outcomes of interest: chan ge in attachment security (Qsort), change in child disruptive behavior (CBCL Externalizing scale), change in pare nting stress (PDH), and change in perceived social support (MSPSS). The first regression model examined whether ch ange in child disruptive behavior in CDI and PDI predicted change in attachment security from preto post-treatment. Both predictors, the CDI ECBI change score and the PDI ECBI ch ange score, were ente red simultaneously into the linear regression model. Change in E CBI during CDI was found to significantly predict change in attachment security from preto post-treatment ( = .34, t = 2.286, p = .03, R2= .12). Change in ECBI in PDI was not found to predict cha nge in attachment security from preto posttreatment ( = .16, t = 1.05, p = .30). The value of the regression coefficient ( =.34) for CDI ECBI change suggests a positive linear relationshi p between change in child disruptive behavior and change in attachment security, such that as change in child disrupt ive behavior increases, change in attachment security also increas es. Results are presented in Table 3-19. Change in Parent-Child Dysfunctional Inte raction (PCDI) in CDI and PDI was also examined as predictors of change in attachment security from preto post-treatment. The CDI and PDI change score in PCDI were not significant predic tors of change in attachment security (CDI change, = -.14, t = -.92, p = .36; PDI Change, = .15, t = 1.01, p = .32). Change in Parent Child Dysfunctional Interac tion in CDI and PDI was also examined as predictors of change in child disruptive behavior from pre to post-treatment. Both the CDI ( =
102 .36, t = 2.67, p = .01) and PDI change scores ( = -.33, t = -2.42, p = .02) were found to significantly predict change in child disruptive behavior as m easured by the CBCL Externalizing scale and accounted for 25% of th e variance in change in child di sruptive behavior. The inverse relationship between CDI change and change in child disruptive behavior suggests that given a .36 unit increase in change in PC DI, change in CBCL decreases by 1 unit. Examination of the mean CDI change in PCDI reveals little change in PCDI during PCIT (M = .50, + 4.3, range -4 to 5); hence, it predicted minima l change in child disruptive beha vior. The negative regression coefficient for PDI change suggests a negative linear relationship between change in PCDI during PDI and change in child disruptive behavior, with a decrease in mothers ratings of parent child dysfunctional interaction associated with a d ecrease in child disrupti ve behavior. Results are presented in Table 3-20. Change in the Difficult Child subscale in CDI and PDI was also examined as predictors of change in child disruptive be havior; however, CDI and PDI change in DC was not found to significantly predict change in ch ild disruptive behavior. Inspecti on of the mean change in CDI reveals little change in the Di fficult Child subscale (M = 2.5, + 4.4, range -2 to 7) during CDI. Mean change in PDI was only slightly larger (M = 6.2, + 7.3, range -1 to 13). Change in Parent Distress in CDI and PDI was examined as predictors of change in maternal ratings of perceived social support (MS PSS) from preto post-treatment. Change in CDI and PDI in parent distress was not found to si gnificantly predict change in maternal ratings of perceived social support (CDI, = .12, t = .77, p = .45; PDI, = .15, t = .95, p = .35). Results presented in Table 3-21. It is important to note that the sample sizes for each of the five linear regression models were reduced due to missing data. Sample si zes for each model discussed above were as
103 follows: 44, 44, 45, 45, and 43. These sample sizes are significantly reduced from the larger sample (N = 100). One reason for th e deletion of more than half of the sample from the analysis was attrition. Another reason for deletion of cases was missing data for unidentified reasons (failure to gather data during session, failure of mother to correctly fi ll out questionnaire). Hence, the results from the traditional linear regression analyses focus on only a sub-sample. Predicting Dropout from Patterns of Change A noted advantage of multilevel m odeling is its ability to use all avai lable data, including data collected on participants who prematurely dropped out of the study. Hence, the data from treatment dropouts were included in the growth a nd phase of treatment models run earlier, and their individual rates of change were estimated along with treatment completers. (A few cases were dropped from these analyses due to inco mplete data. Three were dropouts and one a completer.) The individual slope estimates for dropouts (N=35) were entere d as a predictor into the logistic regression analysis below. The dropouts in the sample were further classified as either early (0) or late dropouts (1), correspond ing with the phase of treatment in which the family dropped out. It was hypothesized that fa milies who show less change on the ECBI Intensity Scale, the Difficult Child and Parent-C hild Dysfunctional Interaction subscales in CDI would be more likely to dr op out of treatment in CDI. The logistic regression model did not yield significant resu lts, suggesting that the CDI slopes for dropouts were not predictive of attrition in the CDI phase of treatment. In addition, the model only correctly classified 19% (o r 3 out of 16) of the early dropouts A traditional logistic regressi on analysis was also run, examining whether change in ECBI Intensity Scale, Difficult Child, and Parent-Child Dysfunctional In teraction subscales by treatment phase predicted early versus late dropout in PCIT. Ch ange scores were calculated for each of the independent variables according to phas e of treatment; hence, the predictors included
104 the CDI change score and PDI cha nge score for the ECBI, DC, and PCDI scales of the PSI-SF. Three separate models were run for each of the pr edictors. The CDI and PDI change scores were entered simultaneously into th e logistic regression model. None of the models returned significant findi ngs, suggesting that patt erns of change in child disruptive behavior and parenting stress do not predict premature dropout from PCIT. Results are presented in Table 322. The ECBI change model (N = 26) correctly classified 77% of families (10 out of 13) who dropped from trea tment during CDI and 54% of families (7 out of 13) who dropped from treatment during PDI. The PCDI change model (N = 25) also correctly classified 77% of families (10 out of 13) w ho dropped from treatment during CDI and 58% of families (7 out of 12) who dropped from PDI. The DC change model (N = 24) correctly classified 69% of families (9 out of 13) who dropped from treatment during CDI and 45% of families (5 out of 11) who dropped from treatment during PDI.
105 Table 3-1 Sample Characteristics Demographic Means (S.D.) Child Age 4.38 (1.10) Mother Age 33.43 (9.44) Hollingshead Index 37.20 (13.07) Total Monthly Income 2,615.43 (2,139.05) Mother Wonderlic Personnel Test 106.28 (12.36) Child Peabody Picture Vocabulary Test 102.21 (12.91) Frequency Child Characteristics Mother Characteristics Gender Mother Race Male 69 White83 Female 31 Black7 Child Race Hispanic4 White 76 Bi-racial5 Black 8 Mother Marital Status Hispanic 3 Married58 Asian 1 Separated6 Bi-racial 12 Divorced17 Widowed1 Children on Medication Single17 Yes 26 Mother Education No 57 Junior High School 2 Some High School6 High School Diploma18 Some College/Technical40 College Graduate26 Graduate School6 Table 3-1. Continued Treatment Characteristics Treatment Status Number of CDI Sessions Completed Number of PDI Sessions Completed Completers 64 5.76 (1.85)11.17 (4.56) Dropouts 36 Dropout Status Early 19 4.16 (2.56) Late 17 14.05 (6.74) Child Co-morbid Diagnosis ADHD CD Yes 70 Yes45 No 22 No51 MDD SAD Yes 3 Yes24 No 86 No67
106 Table 3-2. Correlations for Predictor and Outcome Variables Variables 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 1 Pre ECBI -2 Pre PSI Tot 36 -3 PSI PD .00 52 -4 PSI PCDI .22 59 .23 -5 PSI DC .47 .60 .19 .38 -6 SES -.21 -.13 .17 -.07 -.03 -7 Pre Qsort -.31 -.29 -.02 -.35 -.29 .14 -8 Post Qsort -.05 -.31 -.35 -.15 -.15 .06 .07 -9 Pre PDH .21 .25 .16 .17 .34 -.22 -.20 -.08 -10 Post PDH .22 .19 .20 .01 .22 08 .13 -.17 .16 -11 Pre CBCL .55 .42 .26 .17 .36 -.23 -.25 .01 .11 .06 -12 Post CBCL .20 .39 .26 .11 .01 -.09 -.12 -.16 -.16 .47 .40 -13 Pre MSPSS .02 .43 .47 .24 .26 -.25 -.05 -.10 .30 .17 .25 .33 -14 Post MSPSS 16 .23 .26 .04 .13 -.36 -.00 -.23 .21 .27 .35 .19 .66 -15 Pre BDI .17 .58 .44 .27 .33 -.15 -.13 -.09 .24 .03 .45 .33 .54 .44 -16 Mid BDI .31 .26 .57 .09 .33 -.09 -.18 .00 .20 .13 .51 .20 .64 .55 .78 -17 Post BDI .26 .09 -.00 .05 .29 -.11 -.19 .07 .26 .39 .37 .23 .39 .51 .41 .72 -18 Post BPTS -.05 -.00 .15 .17 .05 .19 -.00 -.11 .22 .34 .17 .03 .09 .17 -. 01 .10 .23 -Table 3-2 continued 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Mean 170.4 103.8 33.9 26.6 43.1 37.2 .03 .31 50.7 40.4 72.9 56.1 33.7 32.4 13.5 10.4 7.02 57.1 SD 26.2 17.7 14.2 6.9 6.4 13.1 .21 .23 10.1 7.4 7.7 9.5 15.2 13.9 9.2 7.8 8.3 13.4 Skewness -.04 -.22 2.05 .22 .38 .17 1.03 .50 1.28 -.11 -.27 .63 .65 .59 .79 .69 1.97 1.38 SD .26 .27 .25 .25 .26 24 .25 .31 .25 .32 .24 .31 .24 .31 .24 .32 .31 .31 Kurtosis -.23 .38 6.27 -.46 .07 -.74 4.12 1.9 4.54 -.41 .87 -.29 -.04 .64 .43 .08 4.71 1.76 SD .52 .53 .50 .50 .51 48 .49 .60 .49 .62 .48 .60 .48 .62 .48 .62 .61 .60 N 83 81 90 90 89 100 96 61 96 57 99 61 100 58 100 57 59 61
107 Table 3-3. Unconditional Means Models for the ECBI Parameter Unconditional Means (SE) Fixed Effects Initial Status, 0 i Intercept 00 138.23** (3.03) Variance Components Level 1 Within-person 2 i 560.41 (23.67) Level 2 In initial status 2 0 819.50(28.62) Pseudo R2 statistics and Goodness-of-fit R2 .59 Deviance 12,432.49 The t -statistic is statistically significant at the p < .05 or ** p < .001. The chi square statistic is significant at the p < .001 level. Table 3-4 Unconditional Means Models for the PSI Total and Subscales Parameter Unconditional Means (SE) PSI Total PD PCDI DC Fixed Effects Initial Status, 0 i Intercept 00 95.35** (1.87) 29.38** (.88) 26.02** (.68) 39.98** (.73) Variance Components Level 1 Within-person 2 i 122.72 (11.08) 19.44 (4.41) 14.49 (.3.81) 27.99 (5.29) Level 2 In initial status 2 0 322.93 (17.97) 71.98 (8.48) 42.56 (6.52) 47.96 (6.92) Pseudo R2 statistics and Goodness-of-fit R2 .72 .79 .75 .63 Deviance 10,425.98 8.023.38 7,614.66 8,409.10 The t -statistic is statistically significant at the p < .05 or ** p < .001; The chi square statistic significant at the p < .001 level
108 Table 3-5. Unconditional Growth Model for the ECBI and PSI Scales Parameter Unconditional Growth Means (SE.) ECBI PSI Total PD PCDI DC Fixed Effects Initial Status, 0 i Intercept 00 158.47** (3.21) 102.91** (1.87) 31.67** (.94) 27.52** (.68) 43.78** (.76) Rate of Change, 2 i Intercept 10 -.38** (.04) -.15** (.02) -.05** (.01) -.03** (.006) -.07** (.01) Variance Components Level 1 Withinperson 2 i 210.29 (14.50) 54.47 (7.38) 10.94 (3.31) 10.00 (3.16) 11.92 (3.45) Level 2 In initial 2 0status 939.22 (30.65) 321.86 (17.94) 81.30 (9.02) 41.65 (6.45) 51.89 (7.20) In rate of change 2 1 .12 (.35) .03 (.17) .005 (.07) .003 (.05) .008 (.08) Pseudo R2 statistics and Goodness-of-fit R2 .62 .56 .44 .31 .57 Deviance 11,469.64 9,672.72 7,536.90 7,339.65 7,616.20 The t -statistic is statistically significant at the p < .05 or ** p < .001; The chi square statistic significant at the p < .001 level
109 Table 3-6. Phase of Treatment Predictor Model Parameter Change Estimates (S.E.) ECBI PSI Total PD subscale PCDI subscale DC subscale Fixed Effects Initial Status, 0 i Intercept 00 164.74** (4.49) 101.47** (2.91) 31.55** (1.41) 28.17** (1.03) 42.80** (.90) Rate of Change, 2 i Time 10 -.61** (.12) -.15** (.06) -.06** (.02) -.04 (.02) -.08* (.01) Tx Phase 20 -3.94 (3.28) 1.35 (1.94) .87 (.26) -.52 (.58) .97* (.48) Tx Phase X Time 30 .13* (.07) -.004 (.03) .008 (.01) .006 (.01) Variance Components (S.D.) Level 1 Withinperson 2 i 208.55 (14.44) 54.30 (7.37) 10.90 (3.30) 10.03 (3.16) 11.81 (3.43) Level 2 In initial 2 0status 932.27 (30.53) 324.56 (18.02) 82.15 (9.06) 42.10 (6.49) 52.02 (7.21) In rate of change 2 1 .12 (.34) .03 (.18) .004 (.07) .002 (.05) .008 (.09) Pseudo R2 statistics and Goodness-of-fit R2 .01 .01 Deviance 11,465.55 11,457.40 7,547.58 7,353.75 7,621.65 The t -statistic is statistically significant at the p < .05 or ** p < .001; The chi-square statistic significant at the p < .001 level.
110 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 1700 2 0 40 6 0 80 100 1 20 140 160 180 200 22 0 240 26 0 280 30 0 320 3 40 360 3 80 40 0Days in TreatmentECBI Intensity Scale CDI Slope PDI Slope 0 5 10 15 20 25 30 35 40 45 500 20 40 60 80 1 00 120 140 16 0 180 200 22 0 2 40 260 280 3 00 320 340 360 3 80 400Days in TreatmentPSI DC Scale CDI Slope PDI Slope Graph 3-1. Patterns of Change in CDI and PDI. A) Patterns of change in child disruptive behavior in CDI and PDI. B) Patterns of change in parent ing stress related to having a difficult child to manage.
111 Table 3-7. Effect of Diagnosis of ADHD and Treatment Phase on ECBI and PSI DC Parameter Change Estimates (S.E.) ECBI DC subscale Fixed Effects Initial Status, 0 i Intercept 00 146.60** (6.90) 40.03** (1.99) ADHD 01 17.54* (7.65) 3.68 (2.22) Rate of Change, 2 i Slope 10 -.61** (.14) -.07** (.02) ADHD 11 .09 (.16) -.007 (.02) Tx Phase 20 1.50* (.54) Tx Phase Time 30 .14* (.06) Tx Phase Time ADHD 21 -.07 (.08) Variance Components Level 1 Within-person 2 i 207.81 (14.41) 11.77 (2.43) Level 2 In initial status 2 0 859.68 (29.32) 49.86 (7.06) In rate of change 2 1 0.12 (.35) .008 (.08) Pseudo R2 statistics and Goodness-of-fit R2 .62 .57 R2 0 .08 .04 R2 1 .00 .00 Deviance 11,451.27 7,604.59 The t -statistic is statistically significant at the p < .05 or ** p < .001; = The chi square statistic significant at the p < .001 level
112 Table 3-8. Effect of Diagnosis of AD HD on Total, PD and PCDI PSI Scales Parameter Change Estimates (S.E.) PSI Total PD PCDI Fixed Effects Initial Status, 0 i Intercept 00 97.02** (3.60) 28.48** (1.49) 27.03** (1.32) ADHD 01 7.74a (4.19) 4.21* (1.86) .64 (1.54) Rate of Change, 2 i Slope 10 -.14** (.04) -.04** (.01) -.03* (.01) ADHD 11 -.01 (.04) -.005 (.02) .006 (.01) Variance Components Level 1 Within-person 2 i 54.46 (7.38) 10.94 (3.31) 10.00 (3.16) Level 2 In initial status 2 0 311.08 (17.64) 78.08 (8.83) 41.59 (6.45) In rate of change 2 1 .03 (.18) .005 (.07) .003 (.05) Pseudo R2 statistics and Goodness-of-fit R2 .56 .44 .31 R2 0 .03 .04 .00 R2 1 .00 .00 .00 Deviance 9,669.54 7,532.99 7,339.24 The t -statistic is statistically significant at the p < .05 or ** p < .001; = The chi square statistic significant at the p < .001 level; a p value approached significant at .07.
113 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160 170 1800 2 0 40 60 8 0 100 1 20 140 16 0 1 80 200 22 0 2 40 260 28 0 300 32 0 340 360 38 0 400Days in TreatmentECBI Intensity Scale No ADHD Treatment Slope ADHD Treatment Slope 0 5 10 15 20 25 30 350 20 4 0 6 0 8 0 1 0 0 1 2 0 140 160 1 8 0 2 0 0 2 2 0 240 260 2 8 0 3 0 0 320 340 3 6 0 3 8 0 4 0 0Days in TreatmentParent Distress Scale No ADHD Treatment Slope ADHD Treatment Slope Graph 3-2. Patterns of Change by ADHD Diagnosis. A) Patterns of change in child disruptive behavior with and without ADHD diagnosis. B) Pattern of change in parent distress with and without ADHD diagnosis.
114 Table 3-9. Effect of Attachment Q sort and Ph ase of Treatment on Child Disruptive Behavior Parameter Change Estimates (S.E.) ECBI Fixed Effects Initial Status, 0 i Intercept 00 160.94** (3.01) Q-Sort 01 -27.45 (16.81) Rate of Change, 2 i Slope 10 -.55** (.10) Q-Sort 11 -.18 (.50) Tx phase Time 20 .09* (.04) Tx phase Time Qsort 21 .04 (.24) Variance Components Level 1 Within-person 2 i 208.12 (14.43) Level 2 In initial status 2 0 877.51 (29.62) In rate of change 2 1 .12(.35) Pseudo R2 statistics and Goodness-of-fit R2 .62 R2 0 .07 R2 1 .00 Deviance 11,452.29 The t -statistic is statistically significant at the p < .05 or ** p < .001; The chi square statistic significant at the p < .001 level
115 Table 3-10. Effects of Gender on Change in PCIT Parameter Change Estimates ECBI PSI Total PD subscale PCDI subscale DC subscale Fixed Effects Initial Status, 0 i Intercept 00 151.28** (9.43) 92.11** (5.41) 25.26** (2.71) 24.75** (2.00) 40.92** (2.25) Gender 01 5.54 (6.13) 8.33* (3.65) 4.92* (1.88) 2.13 (1.36) 2.21 (1.54) Rate of Change, 2 i Days in Treatment 10 -.46** (.12) -.23** (.06) -.05* (.02) -.04* (.02) -.11** (.03) Gender 11 .06 (.07) .06 (.04) .004 (.01) .01 (.01) .02 (.03) Variance Components (S.D) Level 1 Withinperson 2 i 210.32 (14.50) 54.47 (7.38) 10.94 (3.31) 10.00 (3.16) 11.92 (3.45) Level 2 In initial status 2 0 933.26 (30.55) 307.31 (17.53) 76.20 (8.73) 40.69 (6.38) 50.84 (7.13) In rate of change 2 1 .12 (.34) .03 (.18) .004(.07) .003 (.05) .007 (.09) Pseudo R2 statistics and Goodness-of-fit R2 .62 .55 .44 .31 .57 R2 0 .05 .06 R2 1 .00 .00 Deviance 11,467.94 9,665.22 7,529.52 7,335.81 7,611.16 The t -statistic is statistically significant at the p < .05 or ** p < .001; The chi-square statistic significant at the p < .001 level.
116 0 10 20 30 40 50 60 70 80 90 100 110 1200 20 40 60 80 10 0 120 140 160 180 200 220 240 260 280 300 3 20 340 3 60 380 400Days in TreatmentPSI Total Scale Male Treatment Slope Female Treatment Slope 0 5 10 15 20 25 30 35 400 20 4 0 6 0 80 10 0 120 140 16 0 18 0 200 22 0 24 0 260 280 30 0 32 0 340 360 38 0 400Days in TreatmentParent Distress Scale Male Treatment Slope Female Treatment Slope Graph 3-3. Patterns of Change by Gender. A) Patte rns of change in overall maternal parenting stress by gender. B) Patterns of ch ange in parent distress by gender.
117 Table 3-11. Barriers to Participa tion in Treatment Change Model Parameter Change Estimates ECBI Fixed Effects Initial Status, 0 i Intercept 00 60.21* (24.35) BTPS 01 .85* (.42) Rate of Change, 2 i Intercept 10 1.02** (.20) BTPS 11 -.01** (.002) Variance Components Level 1 Withinperson 2 i 218.73 (14.79) Level 2 In initial status 2 0 793.38 (28.17) In rate of change 2 1 .09(.30) Pseudo R2 statistics and Goodness-of-fit R2 .62 R2 0 .37 R2 1 .26 Deviance 5,565.31 The t -statistic is statistically significant at the p < .05 or ** p < .001; The chi-square statistic significant at the p < .001 level
118 Graph 3-4. Predicting Change in Ch ild Disruptive Behavior from Mo thers Ratings of Barriers to Participation in Treatment.
119 Table 3-12. Effect of SES on Change in Childrens Disruptive Behavior Parameter Change Estimates ECBI Fixed Effects Initial Status, 0 i Intercept 00 180.13** (9.96) SES 01 -.57** (.13) Rate of Change, 2 i Intercept 10 -.51** (.13) SES 11 .003 (.002) Variance Components Level 1 Withinperson 2 i 210.24 (14.50) Level 2 In initial status 2 0 877.32 (29.62) In rate of change 2 1 .12(.34) Pseudo R2 statistics and Goodness-of-fit R2 Y R2 .62 R2 0 .07 R2 1 .00 Deviance 11,463.26 The t -statistic is statistically significant at the p < .05 or ** p < .001; The chi-square statistic significant at the p < .001 level
120 -80 -60 -40 -20 0 20 40 60 80 100 120 140 160 1800 20 4 0 60 8 0 1 00 120 1 40 16 0 1 80 20 0 220 24 0 260 28 0 300 320 340 360 380 400Days in TreatmentEstimated ECBI Values High SES Estimate Mid SES Estimate Low SES Estimate Graph 3-5. Initial Status and Ra te of Change in Child Disrup tive Behavior by Socioeconomic Status. 1 i = 180.13 -.57 ( 24.15 ) = 166.36 1 i = 180.13 -.57 ( 37.94 ) = 158.5 1 i = 180.13 -.57 ( 51.73 ) = 150.64
121 Table3-13. Effect of Perceived Barriers to Part icipation in Treatment on Change in Parenting Stress Parameter PSI Total PD subscale PCDI subscale DC subscale Fixed Effects Initial Status, 0 i Intercept 00 51.62** (12.06) 15.25* (4.73) 14.71* (4.37) 21.53** (4.35) BTPS 01 .56* (.20) .18* (.07) .15* (.07) .22* (.07) Rate of Change, 2 i Slope 10 .42** (.11) .11* (.03) .11* (.03) .21** (.05) BTPS 11 -.005** (.002) -.001* (.0004) -.001* (.0004) -.002** (.001) Variance Components Level 1 Within-person 2 i 48.41 (6.96) 12.52 (3.54) 5.99 (2.45) 11.56 (3.40) Level 2 In initial 2 0status 332.06 (18.22) 75.69 (8.70) 40.22 (6.34) 55.48 (7.45) In rate of change 2 1 .02 (.15) .004 (.06) .003 (.05) .005 (.07) Pseudo R2 statistics and Goodness-of-fit R2 .55 .44 .31 .57 R2 0 .26 .11 .13 .33 R2 1 .10 .14 .03 .24 Deviance 4,613.21 3,736.35 3,276.66 3,670.86 The t -statistic is statistically significant at the p < .05 or ** p < .001; The chi-square statistic significant at the p < .001 level
122 Graph 3-6. Patterns of Change in Parenting Stre ss by Levels of Barriers to Participation in Treatment. A) Patterns of change in overall maternal parenting stress by levels of barriers to participation. B) Pa tterns of change in maternal parent distress by levels of barriers to participation. C) Patterns of change in matern al parent child dysfunctional interaction by levels of barri ers to participation. D) Patterns of change in maternal ratings of difficult child by levels of barriers to participation.
123 Graph 3-6. Cotinued.
124 Graph 3-6. Continued.
125 Graph 3-6. Continued.
126 Table 3-14. Effect of Maternal Depre ssion in Childrens Disruptive Behavior Parameter Change Estimates (S.E.) Main Effect Model Interaction Effect Model Fixed Effects Initial Status, 0 i Intercept 00 141.50** (6.31) 141.51** (6.37) Rate of Change, 2 i Slope 10 -.34** (.04) -.34** (.04) Pre-BDI Score 20 .87** (.39) .87** (.40) Level-1 Centered BDI 30 1.30** (.54) 1.32 (.69) Time X Centered BDI 40 -.0002 (.004) Variance Components (S.D.) Within-person 2 i 459.52 (21.43) 460.15 (21.45) Level 2 In initial status 2 0 562.72 (23.72) 561.80 (23.70) In rate of change 2 1 .15(.02) .15 (.02) Pseudo R2 statistics and Goodness-of-fit R2 .62 R2 0 .40 R2 1 .20 Deviance 1,400.20 1,400.20
127 0 20 40 60 80 100 120 140 1600 2 0 40 60 8 0 100 1 20 1 40 160 1 8 0 2 00 2 2 0 2 4 0 2 60 280 3 0 0 3 20 340 3 60 3 80 400Days in TreatmentECBI Intensity Scale Minimal Sxs with No Change Minimal Sxs with Some Change Minimal Sxs with Increase Over Time -20 0 20 40 60 80 100 120 140 160 1800 20 40 60 80 100 120 140 16 0 180 20 0 220 240 260 280 300 320 340 360 380 400Days in TreatmentEstimated ECBI Values Mild Sxs with Minimal Change Mild Sxs with Variable Change Mild Sxs with Decrease Over Time Graph 3-7. Patterns of Change in Child Disrupti ve Behavior by Change in Maternal Depression during PCIT. A) Effect of minimal matern al depressive symptoms on patterns of change. B) Effect of mild maternal depr essive symptoms on patterns of change. C) Effect of moderate maternal depressive symp toms on patterns of change. D) Effect of severe maternal depressive symptoms on patterns of change. 141.5 -.34 + .87 ( 14.79 ) + 1.30 ( 0 10 -7 ) 141.5 .34 + .87 ( 14.79 ) + 1.30 ( 0 -2 -5 ) 141.5 .34 + .87 ( 14.79 ) + 1.30 ( 0 -3 -16 ) 141.5 -.34 + .87 ( 6.09 ) + 1.30 ( 0 10 18 ) 141.5 .34 + .87 ( 6.09 ) + 1.30 ( 0 -1 0 ) 141.5 .34 + .87 ( 6.09 ) + 1.30 ( 0 -4 -6 )
128 -20 0 20 40 60 80 100 120 140 160 1800 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 34 0 360 380 40 0Days in TreatmentEstimated ECBI Values Moderate Sxs with Minimal Change Moderate Sxs with Change Over Time -40 -20 0 20 40 60 80 100 120 140 160 180 2000 20 40 60 80 100 120 140 160 180 200 220 240 260 28 0 300 320 340 360 380 400Days In TreatmentEstimated ECBI Values Severe Sxs with Variable Change Severe Sxs with Moderate Change Severe Sxs with Complete Reduction of Sxs Graph 3-7 Continued. 141.5 -.34 + .87 ( 20.01 ) + 1.30 ( 0 -1 -1 ) 141.5 .34 + .87 ( 20.01 ) + 1.30 ( 0 -11 -21 ) 141.5 -.34 + .87 ( 33.06 ) + 1.30 ( 0 -8 0 ) 141.5 .34 + .87 ( 33.06 ) + 1.30 ( 0 -14 -16 ) 141.5 .34 + .87 ( 33.06 ) + 1.30 ( 0 -17 -40 )
129 Table 3-15 Effect of Maternal Depression on Parenting Stress Parameter PSI Total PSI PD Main Effect Interaction Effect Main Effect Interaction Effect Fixed Effects Initial Status, 0 i Intercept 00 85.92** (3.65)) 86.60** (3.63) 22.19** (1.54) 22.32** (1.54) Rate of Change, 2 i Slope 10 -.11** (.02) -.11** (.02) -.02** (.01) -.02** (.005) Pre-BDI 20 1.16** (.18) 1.09** (.17) .65** (.08) .63** (.08) Level-1 Centered BDI 30 1.10** (.25) 0.62 (.41) .39** (.09) .19 (.16) Time X Centered BDI 40 .004 (.003) .001 (.001) Variance Components Level 1 Withinperson 2 i 105.22 (10.26) 94.24 (9.71)) 16.01 (4.00) 15.62 (3.95) Level 2 In initial status 2 0 146.22 (12.09) 153.91 (12.41) 28.86 (5.37) 29.70 (5.45) In rate of change 2 1 .07 (.004) .07 (.01) .0003 (.02) .0002 (.01) Pseudo R2 statistics and Goodness-of-fit R2 .56 .44 R2 0 .54 .65 R2 1 .57 .94 Deviance 1,210.46 1,207.40 931.69 927.80
130 Table 3-15. Continued Parameter PSI PCDI PSI DC Main Effect Interaction Effect Main Effect Interaction Effect Fixed Effects Initial Status, 0 i Intercept 00 24.03** (1.50) 24.02** (1.50) 39.72** (1.61) 40.15** (1.64) Rate of Change, 2 i Slope 10 -.02** (.004) -.02** (.004) -.07** (.01) -.07** (.01) Pre-BDI 20 .17** (.08) .17** (.08) .35** (.11) .31** (.11) Level-1 Centered BDI 30 .21** (.07) .19 (.13) .43** (.13) .11 (.16) Time X Centered BDI 40 .0002 (.0006) .003** (.001) Variance Components Level 1 Withinperson 2 i 12.94 (3.60) 12.96 (3.60) 18.94 (4.35) 13.67 (3.70) Level 2 In initial status 2 0 42.48 (6.52) 42.47 (6.52) 28.14 (5.30) 30.80 (5.55) In rate of change 2 1 .0002 (.01) .0002 (.01) .002 (.05) .003 (.06) Pseudo R2 statistics and Goodness-of-fit R2 .31 .57 R2 0 .02 .46 R2 1 .93 .75 Deviance 914.36 914.27 984.49 977.49
131 0 5 10 15 20 25 30 35 40 450 20 40 60 80 1 00 1 20 140 160 180 2 0 0 2 2 0 2 40 2 60 2 80 3 00 3 20 3 40 3 60 3 80 4 00Days in TreatmentDifficult Child Scale Minimal Sxs with No Change Minimal Sxs with Some Change Minimal Sxs with Increase Over Time -10 0 10 20 30 40 500 20 40 60 80 100 120 1 40 160 180 200 220 240 260 280 300 320 34 0 360 380 400Days in TreatmentDifficult Child Scale Mild Sxs with Minimal Change Mild Sxs with Variable Change Mild Sxs with Decrease Over Time Graph 3-8. Patterns of Change in Parenting Stress Related to Difficult Child by Changes in Maternal Depression during PCIT. A) Eff ect of changes in minimal maternal depressive symptoms. B) Effect of change s in mild maternal depressive symptoms. C) Effect of moderate matern al depressive symptoms. D) Effect of severe maternal depressive symptoms. 40.15 .07 + .31 ( 7 ) + .11 + .003 ( 0 -1 0 Da y) 40.15 -.07 + .31 ( 7 ) + .11 + .003 ( 0 -4 -6* Da y) 40.15 -.07 + .31 ( 7 ) + .11 + .003 ( 0 10 18 Da y) 40.15 .07 + .31 (17) +.11 + .003 (0, -2, -5 Day) 40.15 .07 + .31 (17) +.11 + .003 (0, 10, -7 Day) 40.15 .07 + .31 (17) +.11 + .003 (0, -3, -16 Day)
132 -10 0 10 20 30 40 50 600 20 4 0 60 80 100 120 1 40 1 6 0 180 2 00 220 2 40 2 6 0 280 300 3 20 3 4 0 360 3 80 4 0 0Days in TreatmentDifficult Child Scale Moderate Sxs with Minimal Change Moderate Sxs with Change Over Time -30 -20 -10 0 10 20 30 40 50 600 20 4 0 60 80 100 120 14 0 16 0 18 0 20 0 2 2 0 240 260 28 0 30 0 32 0 34 0 3 6 0 380 400Days in TreatmentDifficult Child Scale Severe Sxs with Variable Change Severe Sxs with Moderate Change Severe Sxs with Complete Reduction of Sxs Graph 3-8 Continued. 40.15 -.07 + .31 ( 23 ) +.11 + .003 ( 0 -1 -1 Da y) 40.15 .07 + .31 ( 23 ) +.11 + .003 ( 0 -11 -21* Da y) 40.15 -.07 + .31 ( 23 ) +.11 + .003 ( 0 -8 0 Da y) 40.15 .07 + .31 ( 23 ) +.11 + .003 ( 0 -14 -16* Da y) 40.15 .07 + .31 ( 23 ) +.11 + .003 ( 0 -17 -40* Da y)
133 Table 3-16. Unconditional Growth M odels with Reverse Time Code Parameter Unconditional Growth Means (SE.) ECBI PSI Total PD subscale PCDI subscale DC subscale Fixed Effects Final Status, 0 i Intercept 00 121.10** (5.07) 88.71** (3.00) 27.13** (1.31) 24.66** (.97) 36.69** (1.29) Rate of Change, 2 i Intercept 10 .34** (.05) .13** (.02) .04** (.01) .03** (.008) .07** (.01) Variance Components Level 1 Withinperson 2 i 212.29 (14.57) 44.52 (6.67) 12.01 (3.47) 6.15 (2.48) 9.78 (3.13) Level 2 In final status 2 0 1251.35 (35.37) 448.24 (21.17) 84.86 (9.21) 46.40 (6.81) 82.85 (9.10) In rate of change 2 1 .12 (.34) .03 (.16) .004 (.06) .003 (.05) .006 (.08) Pseudo R2 statistics and Goodness-of-fit R2 .62 .64 .38 .58 .65 Deviance 6,143.81 5,039.59 4,083.99 3,629.39 3,952.95
134 Table 3-17. Effect of Change in Qsort on Patter ns of Change in Parent Child Dysfunctional Interaction Parameter Change Estimates (S.E.) PCDI Fixed Effects Final Status, 0 i Intercept 00 25.27** (1.66) Qsort Change 01 1.51 (3.83) Rate of Change, 2 i Slope 10 .02** (.007) Qsort Change 11 .02 (.02) Tx Phase 20 .06 (.46) Tx Phase by Qsort Change 21 -2.05a (1.09) Variance Components Level 1 Within-person 2 i 6.12 (2.47) Level 2 In initial status 2 0 45.24 (6.72) In rate of change 2 1 .002 (.05) Pseudo R2 statistics and Goodness-of-fit R2 Y R2 .58 R2 0 .03 R2 1 .13 Deviance 3621.15 a Slope estimate p value of .059 (t = -1.88). **Estimates significant at the p .01 level.
135 Graph 3-9. Effect of Change in Attachment Security from Preto Post-treatment on Patterns of Change in Child Disrupti ve Behavior during PCIT
136 Table 3-18. Relation between Patterns of Change in Child Disruptive Behavior and Change in Attachment Security from Preto Post-treatment Parameter Change Estimates (S.E.) ECBI Fixed Effects Final Status, 0 i Intercept 00 132.35** (9.25) Qsort Change 01 -22.58 (13.31) Rate of Change, 2 i Slope 10 .24** (.07) Qsort Change 11 .29a (.16) Tx Phase 20 -3.00 (2.93) Variance Components Level 1 Within-person 2 i 199.46 (14.12) Level 2 In initial status 2 0 1731.24 (41.61) In rate of change 2 1 .11 (.34) In Rate of change by tx phase 2 2 200.61 (14.16) Pseudo R2 statistics and Goodness-of-fit R2 .62 R2 0 .02 R2 1 .05 R2 2 .02 Deviance 6,125.01 a Slope estimate p value of ..074 (t = 1.82). **Estimates significant at p < .01.
137 Graph 3-10. Effect of Treatment Phase and Change in Attachment Security from Preto Posttreatment on Patterns of Change in Pa rent Child Dysfunc tional Interaction
138 Table 3-19. Predicting Change in Child Disrupt ive Behavior from Change in Attachment Change in Qsort Unstandardized Coefficients Standardized Coefficients Predictors B S.E. t Significance ( p) ECBI CDI Change .004 .002 .34 2.29 .03 ECBI PDI Change .001 .001 .16 1.05 .30 PCDI CDI Change -.01 .01 -.14 -.92 .36 PCDI PDI Change .01 .01 .15 1.01 .32 Table 3-20. Predicting Change in Parent Child Dysfunctional Interaction from Change in Child Disruptive Behavior Change in CBCL Unstandardized Coefficients Standardized Coefficients Predictors B S.E. t Significance ( p) PCDI CDI Change .84 .32 .36 2.67 .01 PCDI PDI Change -.79 .32 -.33 -2.43 .02 DC CDI Change .40 .35 .18 1.16 .25 DC PDI Change -.01 .21 -.01 -.03 .98 Table 3-21. Predicting Change in Maternal Perc eived Social Support from Change in Parent Distress Change in MSPSS Unstandardized Coefficients Standardized Coefficients Predictors B S.E. t Significance ( p) PD CDI Change .12 .15 .12 .77 .45 PD PDI Change .34 .35 .15 .95 .35 Table 3-22. Logistic Regression Mo del Predicting Early vs. Late Dropout from Change in Child Disruptive Behavior and Pare nting Stress in CDI and PDI Predictor S.E. Wald df p Exp ( ) ECBI CDI Change -.02 .02 .89 1 .35 .99 ECBI PDI Change -.001 .01 .01 1 .91 1.0 DC CDI Change -.01 .07 .09 1 .86 .98 DC PDI Change -.06 .04 2.0 1 .15 .94 PCDI CDI Change .06 ..06 .96 1 .33 1.06 PCDI PDI Change -.07 .08 .74 1 .39 .93
139 CHAPTER 4 DISCUSSION The aim s of this study were to examine predicto rs of different patterns of change in child disruptive behavior and parenti ng during PCIT and to examine di fferent patterns of change as predictors of outcomes on specific parent and chil d variables. Multilevel modeling was used to examine the patterns of change in child disruptive beha vior and parenting st ress across PCIT and between each phase of PCIT (Ch ild Directed Interaction and Pa rent Directed Interaction). Predictors of the established patt erns of change in child disruptiv e behavior and parenting stress across the two phases of treatment, CD I and PDI, were then examined. Patterns of change examining the effect of time (conceptualized as days in treatment) demonstrated significant patterns of changes in individual child di sruptive behavior and parenting stress related to involvement in PCIT. Unconditional multilevel models which examined patterns of change across PCIT as a whole demonstrated that children experienced improvements in child disruptive behavior and parents experienced dec lines in levels of parenting stress. Variances around slopes, how ever, indicated significant inter-individual differences in the patterns of change in child disruptive behaviors and parenting stress across PCIT. Therefore, specific child, parent, family, and treatment predictors were examined and found to predict different patterns of change in child disruptive behavior and parenting stress during PCIT. Predictors of Patterns of Change Treatm ent phase was found to predict differences in the patterns of change in childrens disruptive behaviors and in paren ting stress, as was anticipated. Children demonstrated faster behavior change during CDI than PDI and mothers ratings of having difficulty managing their childs behavior demonstrated fa ster change during CDI than PD I. Length of time (number of
140 days spent) in CDI may explain differences in patterns of cha nge between phases. Recall that families progressed at different speeds through e ach phase of treatment. Therefore, families differed in the number of CDI sessions completed and number of PDI sessions completed. Families who were in CDI at a particular day in treatment were found to be changing at a faster rate than families who had already proceeded to PDI on that particular day. Families who proceed quickly from CDI to PDI appear to s low down possibly because that is where the parenting problems lies for that family pred ictability, consistency, and follow-through in disciplining problem behaviors. Meanwhile, families who take more time in CDI are moving faster in terms of behavior change. These findings lend support to the aim of CD I, which is to decr ease child disruptive behaviors by improving the warmth and security in the parent-child relationship, and they highlight the power a repaired parent-child rela tionship has on the change in child disruptive behaviors. Poor attachment and poor parenting practices have been consistently linked to the severity of disruptive behaviors (Patterson, 1982; Loeber and Schmaling, 1985); therefore CDI is the first phase of treatment in PCIT. During the CDI Teach sessi on, which orients parents to the structure of format and goals of CDI, parents are informed that use of the PRIDE skills taught in CDI throughout their day, in interactions with their children, will help increase their childrens good behavior and improve the quality of the parent -child relationship. It appears that as child disruptive behavior improves in response to the PRIDE skills, mo thers perceptions of their children also shift in a positive direction, w ith mothers perceiving their children as less problematic as they progress th rough CDI and becoming more competent in their use of the PRIDE skills.
141 Interestingly, pre-treatment attachment security ratings did not predict different patterns of change between CDI and PDI, as was hypothesized. The range of Q sort ratings for this sample (.24 to -.18), which represent th e strength of the association of the individual dyads attachment security to the ideal secure at tachment profile, suggests little similarity between mothers and children in our sample to the ideal, securely attached mother-child dyad. Hence, all families had room for improvement in the quality of the parent child relationshi p. It is important to note that PRIDE skills remain an integral component in PDI. The PRIDE skills are ut ilized to establish a warm, positive interaction between mothers and children prior to implementing a demand on the child to comply with a direct command and are conceptualized to re-establish warmth in the parent-child interaction followi ng a time-out session, which places strain on the parent-child relationship. Hence, the quality of the parent-child relationshi p continues to be shaped after families have proceeded to PDI, and in this sample, all families were in need of improving the quality of the attachment to be more secure and warm. Maternal Depression Changes in m aternal depression from preto midto post-treatment were found to predict different patterns of change in child disruptive behavior a nd parenting stress. Children of mothers whose ratings of depression remained c onsistently high during PCIT demonstrated less behavior change in children during PCIT. Likewise, children of mothers whose ratings of depression decreased during PCIT demonstrated faster change in child disruptive behavior during PCIT. Studies examining the effect of matern al depression on treatment outcome have demonstrated that high levels of maternal depressi on at start of treatment predicts less change in child disruptive behaviors (Kazdin, 1995; Kazdin & Wassel, 1999; Webster-Stratton and Hammond, 1990). The one study found in the literatur e that examined maternal depression as
142 predictor of patterns of change in child disruptive did not find that pre-treatment levels of maternal depression predicted differe nt patterns of change in thei r sample (Hartmanet al., 2003). The authors postulated that no effect was found in th eir study due to the small percentage of their sample reporting moderate to severe levels of depression. Findings from this study contradict the finding from the Hartman et al (2003) study, imp licating changes in maternal depression as a significant predictor of patterns of change in child disruptive behavior and parenting stress, and lending support for the inclusion of enhancements in treatment protocols to attenuate the adverse impact parent factors may have on progress in treatment. Findings from this study illustrate the tandem effect between maternal depression and child disruptive behavior discussed in the developm ental literature (Munson et al., 2001), but in the context of intervention. Changes in maternal depression predicted average differences in patterns of change in child disruptive behavior. Child ren whose mothers maintained high levels of maternal depression or experienced an increase in depression during PCIT exhibited slower rates of change in child disruptive behavior, wher eas children of mothers who reported minimal symptoms or experienced a decrease in depression during PCIT exhibited fa ster rates of change in child disruptive behavior. The relationship between cha nges in maternal depression and patterns of change in mothers ratings of having difficu lty managing their child varied over time, such that an increase or a decrease in mothers ratings of depression from one time point to another was associated with an increase or decrease in parenting st ress as it relates to difficulties managing child behaviors. For a mother who started treatm ent reporting minimal depressive symptoms and experienced an 18 point increase in depressive symptoms from mi d-treatment to post-treatment, also experienced a significant increase in paren ting stress. For mothers who started treatment
143 reporting moderate to severe depressive sympto ms and experienced a significant decrease in symptoms to minimal levels from preto midto post-treatment experienced a tandem decrease in parenting stress during PCIT. If change in maternal depression was variable from preto midto post-treatment, patterns of change in mothers ratings of having a difficult child were also found to vary in the same direction. This link between changes in maternal depre ssion and patterns of change in parenting stress is important given that le vels of parenting stress have been found to influence disciplinary practices that promote and escalate aggressive and oppositional child behavior (Patterson, Reid, and Dishion, 1992). If mothers levels of depre ssion maintain higher levels of parenting stress, or vice versa, childrens behavior appears less likely to change dur ing treatment, as two of the main predictors in the development and maintena nce of disruptive behaviors remain essentially untreated or unaffected during tr eatment. Mothers who are stresse d and depressed are likely to use more punitive, harsh, and critical forms of discipline (Downey and Coyne, 1990) and may increase their risk for abusing their child Recent advances in the child therapy lit erature have focused on developing and implementing enhancement components to child interventions in an attempt to remedy or attenuate the adverse impact pa rent risk factors have on ther apeutic change and treatment completion (e.g., Kazdin and Whitley, 2003; McKay and Bannon, 2004; Nock and Ferriter, 2005; Sanders and McFarland, 2000). Findings from this study provide further support for the inclusion of enhanced components in child therapy to address parent factors that adversely affect success in treatment. Other studies have demons trated positive outcomes in treatments that included enhancements to addr ess maternal depression (Sanders and McFarland, 2000) and parenting stress (Kazdin and Whitley, 2003). In both of the cited studies, the enhancements were
144 structured intervention components that directly treated maternal depression and parenting stress. The PCIT protocol in this study included a supportive component in each session in which therapists were encouraged to provide support and problem-solve with parents any significant stressors that were perceived as affecting participation in treat ment. Though this study did not directly examine therapist supportive behaviors as predictors of patte rns of change, some mothers showed exhibited signifi cant declines in depressive sy mptoms and hence, significant declines in parenting stress. Inclusion of this supportive co mponent may have assisted in attenuating the adverse impact of maternal de pression, leading to positive outcomes for both mother and child and lending weight to the import to include components in any intervention that reduces the negative impact of parent risk factor s on treatment progress. Family Predictors Post-treatm ent ratings of percei ved barriers to participation in treatment were examined as a predictor of patterns of change in child disruptive behavior and parenting stress for families who completed treatment. Families who reporte d greater number of perceived barriers to participation in treatment were found to have pa tterns of slower change in child disruptive behavior and parenting stress acr oss PCIT as a whole. Families who reported fewer perceived barriers to participation in treatment demonstrated patterns of faster change in child behavior and parenting stress. These findings emphasize the im portant of assisting families in treatment to reduce these barriers that impede th eir success in treatment. The tr eatment protocol used in this study included components to address and assi st in removing or al leviating barriers to participation in treatment. This included providing families with a voucher to cover the expense of parking to arranging payment of bus or cab fare for families who relied on public transportation to attend treatment sessions. Th erapists occasionally ev en provided therapy in families homes.
145 It is important to keep in mi nd that these findings are limited to the treatment completers in the sample. Graduation from treatment required meeting termination criteria, which included final ratings of child disruptive behavior standard deviations below the normative mean on the ECBI, and children who graduated from treatment reached non-clinical levels of disruptive behaviors. Treatment completers who reported a greater number of perceived barriers finished treatment close to termination criteria (avera ge final ECBI status of 119), with treatment completers who reported fewer barriers well below termination criteria (final ECBI status of 108 for average group and 97 for lowest group). Hen ce, families who reported greater number of barriers to participation in trea tment still experienced improvements in child disruptive behavior during PCIT, only at slower rates than familie s who reported fewer barriers. This finding supports other studies in the chil d therapy literature that suggest that some treatment is better than no treatment at all (Nock and Ferriter, 2005). A great concern for families who perceive greater number of barriers to participation is their ability to attend treatment c onsistently in order to receive the most benefit. In an earlier study, Kazdin and Wassell (1998) found that familie s who reported greater number of barriers also had poor treatment attendance (Kazdin and Wassell, 1998), and families who dropped out of treatment in the same study had poor treatment attendance as measured by higher cancellation and no-show rates. In a recent review of the literature, Nock and Ferriter (2005) note that premature termination from treatment does not necessarily mean that families do not improve. In fact, they cite Kazdin a nd Wassell (1998) who found that 34 % of the dropouts in their study improved significantly prior to leaving treatment. Nock and Ferriter (2005) note that families with poor treatment attendance may still demonstr ate change if they remain adherent to the treatment in between attended treatment sessions. PCIT is a skills-based approach and families
146 are expected to complete a homework assignm ent on a daily basis during the week between sessions. Families are expected to practice th e CDI and PDI skills by incorporating a special play time at home, during which they actively pr actice the PRIDE skills and time out procedures. Hence, families who may not be able to attend treat ment consistently, as a result of experiencing a greater number of barriers impeding participation, may still reap be nefits if they are adhering to treatment outside of the treatment session. Furt her investigation is necessary to confirm this relation between treatment adherence, barriers to participation, and patterns of change in PCIT; however, findings in this study suggest that tr eatment completers experience improvements in child disruptive behavior and pa renting stress, even if they ex perience high numbers of barriers to participation in treatment. Gender Gender was the f inal child predictor entered into the model, and it was examined to determine whether gender predicte d different patterns of change in child disruptive behavior and parenting stress during PCIT. Boy and girls in this sample were found to have similar patterns of change in child disruptive behavior during PCIT, as well as similar levels of disruptive behavior at start of treatment. Gender was also not pred ictive of differences in patterns of change for parenting stress. Mothers of boys and girls in this sample had similar patterns of change in their stress levels related to parent ing, despite significant differences in overall parenting stress and stress related to the paren ting role at start of treatment. (Mot hers of girls in this sample reported higher levels of overall parenting stress and stress related to the parenting role, at the start of treatment.) The role of gender has not been well studied in the treatment outcome literature or the developmental literature. One reas on often cited is the low prevalence of disruptive behaviors in preschool girls; hence the there are fewer gi rls in study samples compared to boys, making it
147 difficult to look at gender differences. WebsterStratton (1996) examined the effect of gender on treatment outcome in a study of 64 girls and 1 58 boys, ages 4 to 7 years, and did not find an effect between gender and time. Boys and girls were found to improve si milarly over time, even with mothers perceiving boys as having more di sruptive behaviors at pre-treatment. Findings from this study were similar, in that gender did not predict di fferent patterns of change in disruptive for boys and girls. Boys and girls were actually found to ha ve similar levels of disruptive behavior at start of treatment in this sample. The lack of significant findings for gender differe nces is not disappointin g. It suggests that PCIT works as well for boy as it does for girls. Any differences that do exist in patterns of change appear related to other ch ild, parent, and family predictors. Predicting Treatment Outcomes from Patterns of Change in Child Disruptive Behavior and Parenting Stress Patterns of change in child di sruptive behavior and parenti ng stress were hypothesized to predict specific parent and child outcom es. A regression an alysis using individual patterns of change as predictors of change in attachment security, child disruptive behavior, daily parenting stress, and perceived social support from pr eto post-treatment was originally proposed; however, some cases failed to meet necessary assumptions for the regression analysis to run, calling into question the appropriateness of using predicted values fr om the patterns of change as data for another analysis. The researcher conducted alternative analyses in an attempt to address the questions proposed in the hypotheses, which included a rede fined multilevel model analysis and traditional linear and logistic regression models. The redefined multilevel model examined change in parent and child outcome variables as predictors and placed emphasis on the level of disruptive
148 behavior and parenting stress at the end of treatment by revers ing the coding for time (days in treatment). Two of the five models examined returned significant and promising results. Families who exhibited faster change in CDI in child disruptive behavior were predicted to exhibit greater change in attachment security from preto post-treatment. Results from the multilevel mode did not support differences in patterns of change between CDI and PDI. Change in attachment security, however, did pr edict different patterns of change in child disruptive behavior across PCIT as a whole. Families who showed greater improvement in attachment security from preto post-treatment demonstrated fa ster behavior change, where as families who demonstrated less change in attachment security exhibited slower behavior change. Results from the traditional regression model anal ysis, which examined change in ECBI in CDI and PDI as predictors of change in attachment s ecurity from preto po st-treatment, supported the hypothesis that greater change in child disruptive in CD I predicted greater change in attachment security. Families who exhibited faster change in parent child dysfunctional interaction in CDI were predicted to exhibit greater change in attachment security from pr eto post-treatment. Results from the multilevel model returned a trend for the interaction between change in attachment security and treatment phase, however, not in th e predicted direction. The results from the multilevel model suggest an association between ch anges in attachment se curity and patterns of change in PDI, with greater change in attachme nt security predicting faster change in parent child dysfunctional interaction duri ng PDI. The results from th e traditional linear regression analysis, examining change in parent child dysfunctional interacti on in CDI and PDI as predictors of change in attachment security, returned non-significant findings, suggesting that
149 change within each phase of treatment was not predictive of treatment outcome for attachment security. Results from the traditional linear regression model predicting change in child disruptive behavior from change in parent child dysfunctional interaction in CDI and PDI were significant. Both change in CDI and change in PDI in parent child dysfunctional interaction predicted change in attachment security. The overall lack of change in parent child dysfunctional interaction during CDI resulted in less change in child disruptive behavior fr om pre to post-treatment. In contrast, improvement in parent child dysf unctional interaction during PDI resulted in improvement in child disruptive behavior from pre to post-treatment. The findings that patterns of cha nge in PDI, as predicted by change in attachment security, predicted greater change in parent child dysfunc tional interaction and th at change in PDI in parent child dysfunctional interaction predicted improvements in child disruptive behavior were initially surprising. Upon further consideration, however, these findings make sense in light of the structure of PCIT. The CDI phase addres ses only one component of the parent child relationship. The PDI phase completes transfor mation in parent child interaction by shaping parenting style to include warmth as well as limits and consistent follow through with consequences and discipline. Indeed families did not demonstrate much change in parent child dysfunctional interaction during CDI, suggesting th at PDI was a necessary component to assist families in developing fully functional parent-child interactions. It is recognized that both alternative analyses failed to address the questions posed in the hypotheses adequately. It was hoped that patterns of change could pr edict to outcome in specific parent and child variables; how ever, statistical limitations re ndered analysis of this kind unfeasible. Findings from the alternative analyses provide preliminary support for relation
150 between therapeutic change and change in spec ific child and parent treatment outcomes. Findings from the multilevel models suggest an association between changes in outcomes and patterns of change, while results from the traditio nal regression models i ndicate that change in a particular phase of treatment is predictive of cha nge in specific outcomes. Further investigation is warranted to establish the re lation between patterns of change in PCIT and change in specific outcomes. Examination of patterns of change in betw een dropouts and treatment completers did not yield significant findings, indicatin g that treatment status did not predict differences in patterns of change in child disruptive across PCIT. Dropout status also did not pred ict different patterns of change in child disruptive behavior among dropouts; hence, families who dropped from treatment in CDI did not exhibit a unique patt ern of change from families who dropped from treatment in PDI. Patterns of change in child di sruptive behavior in CDI and PDI were able to be examined as predictors of dropout status and were not predictive of early versus late dropout in PCIT. These findings are disappointing given ev idence in the literature indicating that early dropouts are differentiated by unique set of predicto rs than families who drop late in treatment. It was anticipated that families who dropped in CDI would have exhibited patterns of change indicating minimal change in child disruptive beha vior; however, in light of perceived barriers to participation in treatment findings, families w ho dropped early from treatment may be better differentiated based on number of barriers to partic ipation rather than on tr eatment status alone. Limitations and Challenges of the Current Study There were m any challenges present in this st udy that required foret hought in determining the appropriate statistical approach and creativeness in application of both traditional and stateof-the-art statistical approaches The unbalanced nature of the data set required the use of a statistical approach that could accommodate the l ack of balance resulting from three sources: (1)
151 unequal spacing of occasions; (2) unequal number of occasions per participant; (3) missing data. Multilevel modeling was chosen sp ecifically for its ability to acco mmodate unbalanced data sets and utilize all data point s available to return estimates c onsidered more valid and reliable compared to estimates derived from a selected da ta set (one that exclude s participants due to missing data; Zaidman-Zait and Zumbo, 2005). Significant findings reported in this study ar e limited to the current sample for a few reasons. First, the geographic location and racial/ethnic make-up of the sample limit generalizability of significant findings to largely White samples from suburban and rural locations. The small sample of other races and ethnicities included in the study precluded examination of patterns of change by race/ethnicity. It is also plausible that families from other geographical locations (urban versus suburban ve rsus rural) may have different patterns of change. Again, such examination was precluded gi ven the nature of this sample, which included families from the suburban area of Gainesville and the outlying rural towns. Patterns of change may also differ according to location of service de livery of treatment. Families were participants in a research study, in which monitoring for the integrity of treatment delivery was on-going, with state-of-the art equipment and set-up. Fam ilies receiving PCIT in community settings may demonstrate different patterns of change predicted from a different set of predictors. Methodological issues also limit findings from this study. Most concerning are the number of significance tests run during th e course of the study, which raises concern over inflation of Type I error in the current findings. The resear cher ran sixty-six models during the course of analyzing the data, equaling 217 significance tests ( 397 parameters total including variance components). Under the assumption of independe nt measures and observations, the experimentwise error rate for the current study equals 1, calling into questi on the validity of the significant
152 findings reported in the study. Under the as sumption of dependence among observations, the experiment-wise error rate of 1 is less than the product of the number of comparisons and per comparison probability (.05), which lends some r eassurance that findings in the study are not solely a result of artifact due to sheer number of tests run; however, the researcher remains cautious in interpretation of current findings and en courages readers to consider the significance of findings, the size of the effect for significan t findings, and the large number of tests run. For example, treatment phase was a significant pr edictor of different patterns of change in child disruptive behavior and pare nting stress related to having a di fficult child to manage at the p < 05 level; however, treatment phase only explains 1% of the within person variance after linear time, which explains 62% in child disruptive beha vior and 57% in stre ss related to having a difficult child to manage. This small percen tage in accounted within -person variance does not lend much weight to the effect of the significa nt finding and certainly ra ises caution regarding the validity of the finding given the number of signifi cance tests run in the study. Closer inspection of the signifi cant finding for barriers to part icipation in treatment as a significant predictor of different patterns of chan ge in child disruptive behavior and parenting stress reveals small to moderate effect sizes. Re trospective ratings of ba rriers to participation explained 16% of the between-participants variance in rate of change in ch ild disruptive behavior during PCIT, a small effect size, whereas retr ospective BPTS ratings explained 32% of the variance in rate of change in parent child dysf unctional interaction during PCIT, a medium effect size. Alongside the .001 level of significance of these findings, th ese findings can be cautiously interpreted as promising and deserving of further examina tion alongside other important predictors not investig ated in this study.
153 A cumulative model approach, which examines the effects of multiple predictors on rates of change in a specific outcome, would assist in reduc ing the number of m odels and significance tests run, thereby reducing inflati on of Type I error. Future in vestigations of predictors of change in therapy should approach model building with a priori pr edictions driven by theory. Hence, only predictors linked meaningfully to the specific outcome are examined. Such an approach lends itself to a model comparison approach, which is considered to provide meaningful results beyond significance tes ting (Judd, McClelland, and Culane, 1995). The current study did not test a cumulative model nor did it compare competing models of predictors of patterns of change. Future investigations s hould attempt to examine predictors cumulatively and with an eye toward testing models for expl anatory power in modeli ng change in therapy. A limited number of predictors were chosen for examination in this study. The researcher attempted to include meaningful predictors based on findings in th e current literature; however, results from the current study clearl y indicate the need to examine additional predictors. At least for this sample, gender, diagnosis of ADHD, a nd socioeconomic status were not predictive of different patterns of change in child disruptive behavior and pa renting stress and the predictors that were found to significantly predict differences can only boast sm all to medium effect sizes at best, suggesting that important predictors we re excluded from the models tested here. Differences in patterns of change in this curr ent study may have been more strongly predicted by therapist qualities and characteristics or other tr eatment factors, such as treatment attendance and adherence. The inclusion of thes e factors in a cumulative model of change in disruptive behavior and parenting stress may result in a more robus t model of change. Othe r factors to consider include family structure and other family charact eristics, such as family cohesion or conflict,
154 other measures of environmental disadva ntage, such poverty level, and neighborhood characteristics such as crime and violence. It is important to note that this study modeled cha nge solely using maternal ratings for measures of both outcomes and predictors. Thou gh parents are considered the best resource for information regarding their child compared to ot her informants, such as teachers (Loeber, Green, and Lahey, 1991), independent measures, such as behavioral observations, provide additional information that may converge or diverge from info rmation gleaned from parent report alone. It is possible that patterns of cha nge would differ based on informant (mother, father, and therapist) and sources of data used (parent and therapist report or observa tional data). The findings here reflect patterns of change based on maternal ra tings of child disruptive behavior and parenting stress and all predictors examined in the study. Lack of a control sample makes it difficult to speculate on the effect of PCIT on patterns of change in child disruptive behavior and parenting stress. Regression to the mean is often cited by researchers, as a confounding variable to find ings demonstrating change in response to intervention (Cunningham, 2006). Regression to the mean is thought to represent a phenomenon in which a person who presents with significant levels on a particular measure at time 1 will move towards the population mean at subsequent measurements. Certain factors increase the risk for regression artifacts in a study. Fi rst is a nonrandom sample, selected based on performance on a pre-test that differs from the population. Another factor is imperfect correlation between measurements, with measures with little to no correlation more likely to regress toward the mean on consecutive measurem ents (Rogosa, 1988). Measures in this study tended to correlate with one anot her; however, inclusi on of a control group in future studies
155 would assist in determining artifact from true ch ange in child disruptive behavior and parenting stress during PCIT. Future Directions Prelim inary findings from this study support different patterns of change in child disruptive behavior. Patterns of change in ch ild disruptive behavior di ffered between treatment phase, with families appearing to change at a faster rate in CDI than PDI. This significant finding is confounded by order, as CDI typically precedes PDI in the PCIT protocol. Future studies may want to examine whether the difference in patterns of change is present regardless of order of treatment phase, and wh ether faster change is presen t in CDI regardless of order. Changes in maternal depression during PCIT also predicted different patterns of change in child disruptive behavior and parenting stress. While this finding lends credence to the transactional relationship between maternal depression and child disrupt ive behavior, further investigation is needed to determine whether ch anges in maternal depression led to changes in child disruptive behavior and pare nting stress or vice versa. Retrospective maternal ratings of perceived barriers to partic ipation in treatment predicted different patterns of change in child disruptive behavior and parenting stress, with families who experienced greater number of ba rriers experiencing slower ra tes of change during PCIT. Despite slower rates of change, families who ex perienced greater number of barriers successfully completed treatment, with childrens levels of disruptive behavior returning to normative levels. This finding suggests that measures taken to mi nimize impact of barriers to participation were successful; however, further invest igation is merited, particularly to examine the role of treatment attendance and adherence in patterns of change in child disr uptive behavior during PCIT.
156 Examination of other child, parent, family, e nvironmental, and treatment-related variables as predictors is merited, in order to reach a bett er understanding of predicto rs of change in PCIT, and child therapy as whole. As noted in the limitations, the predictors examined in this study were limited and most of the predictors examined were not found to affect patterns of change in child disruptive behavior or pare nting stress. Of those that were significant, e ffect sizes ranged from small to medium, indicating that a signifi cant portion of the variance remains unaccounted for, providing further support for the need for further investigation with other predictors of interest. In a similar vein, cumulative models that examine the effects of multiple predictors simultaneously on patterns of change are neede d. The results in this study are limited by the large number of models run, which inflated Type I error and raises concern that significant findings in this study were a result of chance. Future studies should include meaningful cumulative models supported by theory and concep tual frameworks that posit how specific child, parent, and family factors affect change in therapy. The results of this study are limited to change s in maternal ratings of child behavior and parenting stress. Future studies should include measurements of direct behavior observation to provide convergence between parent reports of ch ange and observed chan ge. Would patterns of change in child disruptive behavior differ according to informant and measure? A future study would include data provided by fathers to a llow comparison of patterns of change based on ratings provided by mothers and fathers. Ex amination of observational data would provide patterns of change in observed behavior and provide convergent validity to patterns of change based on parent ratings of behavior.
157 Finally, inclusion of a control group in a future study would help elucidate whether patterns of change in child di sruptive behavior result from in tervention or time alone. Child, parent, and family predictors can be examined to determine which combination of variables predict a pattern of change in wh ich child disruptive behavior pers ists and the pattern of change in which disruptive behaviors desi st and return to normative levels Findings from such a study would address the role of regression to the mean and its true eff ect on patterns of change across time and across treatment.
158 APPENDIX A TABLES OF NON-SI GNIFICANT FINDINGS Table A-1. Patterns of Change in Child Disrupti ve Behavior and Parent ing Stress by Treatment Phase Parameter Change Estimates ECBI PSI Total PD subscale PCDI subscale DC subscale Fixed Effects Final Status, 0 i Intercept 00 125.97** (7.23) 89.41** (3.30) 26.08** (1.57) 26.12** (1.30) 36.42** (1.51) Tx Phase 01 Rate of Change 2 i Days in Treatment 10 .31** (.06) .12** (.06) .04* (.01) .02* (.006) .06** (.01) Tx Phase 11 -2.86 (.2.94) -.32 (1.18) .55 (.52) -.77 (.48) .12 (.53) Variance Components (S.D) Level 1 Withinperson 2 i 199.49 (14.12) 43.08 (6.56) 11.79 (3.43) 5.81 (2.41) 9.51 (3.08) Level 2 In initial status 2 0 1738.72 (41.70) 368.23 (19.19) 75.34 (8.68) 61.81 (7.86) 75.61 (8.69) In rate of change 2 1 .12 (.34) .02 (.13) .005(.07) .001 (.04) .005 (.06) Pseudo R2 statistics and Goodness-of-fit R2 .62 .64 .38 .58 .65 R2 0 -----R2 1 -----Deviance 6128.59 5029.72 4079.61 3,606.25 3945.17 ** Estimates significant at the p < 001
159 Table A-2. Relation between Patterns of Change in Child Disruptive Behavior and Change in Parenting Stress (PDH) from Preto Post-Treatment Parameter Change Estimates ECBI Fixed Effects Final Status, 0 i Intercept 00 123.63** (7.13) PDH 01 .23 (.43) Rate of Change, 2 i Days in Treatment 10 .28** (.07) PDH 11 -.005 (.005) Variance Components (S.D) Level 1 Within-person 2 i 212.56 (14.58) Level 2 In initial status 2 0 1242.83 (35.25) In rate of change 2 1 .11 (.33) Pseudo R2 statistics and Goodness-of-fit R2 .62 R2 0 -R2 1 -Deviance 6141.99 ** Estimates significant at p < .001; Variance component significant at the p < .001
160 Table A-3. Relation between Patterns of Change in Parent-Child Dysfunctional Interaction and Change in Attachment Security from Preto Post-treatment Parameter Change Estimates PCDI Fixed Effects Final Status, 0 i Intercept 00 25.45** (1.45) Qsort 01 -2.88a (3.49) Rate of Change, 2 i Days in Treatment 10 .02** (.007) Qsort 11 .04 b (.02) Variance Components (S.D) Level 1 Within-person 2 i 6.15 (2.48) Level 2 In initial status 2 0 45.76 (6.76) In rate of change 2 1 .002 (.05) Pseudo R2 statistics and Goodness-of-fit R2 .58 R2 0 -R2 1 -Deviance 3627.17 ** Estimates significant at p < .001; Variance component significant at the p < .001; a Estimate p value = .413; b Estimate p value = .123
161 Table A-4. Relation between Patterns of Change in Parenting Stress and Change in Child Disruptive Behavior from Preto Post-treatment Parameter Change Estimates PCDI subscale DC subscale Fixed Effects Final Status, 0 i Intercept 00 33.35** (2.19) 34.66** (2.62) CBCL 01 -.14a (.13) -.12 c (.14) Rate of Change, 2 i Days in Treatment 10 .03* (.02) .06** (.02) CBCL 11 -.0002 b (.0008) -.0004 d (.001) Variance Components (S.D) Level 1 Within-person 2 i 6.15 (2.48) 9.78 (3.13) Level 2 In initial status 2 0 44.84 (6.69) 81.57 (9.03) In rate of change 2 1 .002 (6.69) .006 (.08) Pseudo R2 statistics and Goodness-of-fit R2 .58 .65 R2 0 --R2 1 --Deviance 3626.48 3950.55 ** Estimates significant at the p < .001; Estimate p value = .062; Variance component significant at p < .001; a Estimate p value = .300; b Estimate p value = .804; c Estimate p value = .346; d Estimate p value = .769
162 Table A-5. Relation between Pattern of Change in Difficult Child and Change in Perceived Social Support from Preto Posts-treatment Parameter Change Estimates DC Fixed Effects Final Status, 0 i Intercept 00 36.99** (1.21) MSPSS 01 .14a (.09) Rate of Change, 2 i Days in Treatment 10 .06** (.01) MSPSS 11 -.002 b (.001) Variance Components (S.D) Level 1 Within-person 2 i 9.78 (3.13) Level 2 In initial status 2 0 80.76 (8.98) In rate of change 2 1 .006 (.08) Pseudo R2 statistics and Goodness-of-fit R2 .65 R2 0 -R2 1 -Deviance 3951.32 ** Estimates significant at the p < .001; Variance component significant at p < .001; a Estimate p value = .141; b Estimate p value = .143
163 Table A-6. Patterns of Change in Child Disrupti ve Behavior and Parent ing Stress by Treatment Phase Parameter Change Estimates PSI Total PD subscale PCDI subscale DC subscale Fixed Effects Final Status, 0 i Intercept 00 108.36** (5.45) 34.01** (2.56) 29.45** (1.96) 43.80** (2.44) SES 01 -.14 (.13) -.06 (.06) -.05 (.05) -.0005 (.06) Rate of Change 2 i Days in Treatment 10 -.23 ** (.06) -.05* (.02) -.04* (.02) -.11** (.03) SES 11 .002 (.001) .002 (.001) .0004 (.0004) .001 (.001) Variance Components (S.D) Level 1 Withinperson 2 i 54.45 (7.38) 10.94 (3.31) 10.01 (3.16) 11.92 (3.45) Level 2 In initial status 2 0 317.53 (17.82) 80.54 (8.97) 41.16 (6.42) 51.82 (7.20) In rate of change 2 1 .03 (.18) .005(.07) .003 (.05) .007 (.09) Pseudo R2 statistics and Goodness-of-fit R2 .56 .44 .31 .57 R2 0 ----R2 1 ----Deviance 9670.06 7536.12 7338.18 7614.14 ** Estimates significant at p < 001; *Estimates significant at p < .01.
164 LIST OF REFERENCES Abikoff, H. & Klein, R.G. ( 1992). Attention-deficit hyperactivity and conduct disorder: Com orbidity and implications for treatment. Journal of Consulting and Clinical Psychology, 60, 881-892. Achenbach, T. (1991a). Child Behavior Checklist (CBCL). Burlington: University of Vermont. Achenbach, T. (1992). Manual for the Child Behavior Checklist (2-3) and 1992 Profile Burlington, VT: University of Vermont Press. Aguilar, B., Sroufe, L.A., Egeland, B., & Carl son, E. (2000). Distinguishing the life-coursepersistent and adolescent-limited antisocial behavior types: From birth to 16 years. Development and Psychopathology, 12, 109-132. American Psychiatric Association. (2000). Diagnostic and Statistical Manual of Mental Disorders, 4th Edition, Text Revision. Washington, DC: APA. Armbruster, P. & Kazdin, A.E. (1994). Attrition in child psyc hotherapy. In T.H. Ollendick & R.J. Prinz (Eds.), Advances in clinical child psychology (Vol. 16, pp. 81-108). New York: Plenum. Azar, S., & Wolfe, D. (1989). Child abuse and neglect. In E. Mash & R. Barkley (Eds.), Treatment of childhood disorders (pp. 451-489). New York: Guilford. Baker, D. B. (1994). Parenting stress and ADHD: A comparison of mothers and fathers. Journal of Emotional and Behavioral Disorders, 2 46-50. Barkley, R. A. (1990). ADHD: A Handbook for Diagnosis and Treatment New York, Guilford. Barkley, R.A, Fischer, M., Edelbrock, C.S., & Smallish, L. (1990). The adolescent outcome of hyperactive children diagnosed by research criteria I: An 8-year prospective follow-up study. Journal of the American Academy of Child and Adolescent Psychiatry, 29, 546557. Bates, J., Bayles, K., Bennett, D. S., Ridge, B., & Brown, M. M. (1991). Origins of externalizing behavior problems at eight years of age. In E. J. Pepler & K. H. Rubin (Eds.), The development and treatment of childhood aggression. (pp. 197-216). New York: Academic Press Baumrind, D. (1967). Child care practices ant eceding three patterns of preschool behavior. Genetic Psychology Monographs, 75, 43-88. Baumrind, D. (1991). The influence of parentin g style on adolescent competence and substance use. Journal of Early Adolescence, 11, 56-95. Beck A.T., Steer, R. A., & Brown, G. K. (1996). Beck Depression Inventory-II. San Antonio, TX: Psychological Corporation, Harcourt, Brace &. Company.
165 Biederman, J., Munir, K., & Knee, D. (1987). Conduct and oppositional defiant disorder in clinically referred ch ildren with attention deficit disorder: A controlled family study. Journal of the American Academy of Child & Adolescent Psychiatry, 26, 724-727. Brestan, E., & Eyberg, S. M. (1998). Effective psychosocial treatments of conduct disordered children and adolescents: 19 y ears, 82 studies, and 5,272 kids. Journal of Clinical Child Psychology, 27, 180-189. Brestan, E., Eyberg, S. M., Boggs, S., & Algina, J. (1997). Parent-Child Interaction Therapy: Parent perceptions of untreated siblings. Child and Family Behavior Therapy, 19, 13-28. Brinkmeyer, M.Y. & Eyberg, S.M. (2003). Pare nt child interaction therapy for oppositional children. In A.E. Kazdin & J.R. Weisz (Eds.) Evidence-based psychotherapies for children and adolescents. New York: Guilford. Broidy, L.M., Nagin, D.S., Tremblay, R.E., Bates, J.E., Brame, B., Dodge, K.A., Fergusson, D., Horwood, J.L., Loeber, R., Laird, R., Lynam, D.R., Moffit, T.E., & Petit, G.S. (2003). Developmental trajectories of childh ood disruptive behaviors and adolescent delinquency: A six-site cross-national study. Developmental Psychology, 39 222-245. Calzada, E. J., & Eyberg, S. M. (in press). No rmative parenting in a sample of Dominican and Puerto Rican mothers of young children. Campbell, S.B. (1991). Behavior problems in preschool children: Clinic al and developmental issues. New York: Guilford Press. Campbell, S.B. (1995). Behavior problems in presc hool children: A review of recent literature. Journal of Child Psychology and Psychiatry, 36, 113-149. Campbell, S. B., Pierce, E. W., Moore, G., Marakovitz, S., & Newby, K. (1996). Boys' externalizing problems at elementary school age: Pathways from early behavior problems, maternal control, and family stress. Development and Psychopathology, 8, 701-719. Campbell, S.B., Shaw, D.S., & Gilliom, M. ( 2000). Early externalizi ng behavior problems: Toddlers and preschoolers at ri sk for later maladjustment. Development and Psychopathology, 12, 467-488. Cecil, H. Stanley, M.A., Carrion, P.G., & Swann, A. (1995). Psychometric properties of the MSPSS and NOS in psychiatric outpatients. Journal of Clinical Psychology, 51, 593-602. Crnic, K. A., & Greenberg, M. T. (1990). Minor parenting stresse s with young children. Child Development, 61, 1,268-1,637. Coie, J.D. & Dodge, K.A. (1998). Aggression and an tisocial behavior. In W. Damon (Series Ed.) & N. Eisenberg (Vol. Ed.), Handbook of child psychology: Vo l. 3. Social, emotional, and personality development (5th ed., pp. 779-862). New York: Wiley.
166 Colvin, A., Eyberg, S., & Adams, C. (1999a). Re standardization of the Eyberg Child Behavior Inventory. Manuscript in progress. Connor, D.F. (2002). Preschool attention deficit h yperactivity disorder: A review of prevalence, diagnosis, neurobiology, and stimulant treatment. Developmental and Behavioral Pediatrics, 23, S1-S9 Ct, S., Tremblay, R., Nagin, D., Zoccolillo, M., & Vitaro, F. (2002). Childhood behavioral profiles leading to adolescent conduct disorder : Risk trajectories for boys and girls. Journal of the American Academy of Child and Adolescent Psychiatry, 41, 1086-1094 Cummings, E. M., & Davies, P. (1994). Children and marital conflict: The impact of family dispute and resolution. New York: Guilford. Dahlem, N.W., Zimet, G.D., & Walker, R.R. (1991). The multidimensional scale of perceived social support: A confirmation study. Journal of Clinical Psychology, 47, 756-761. DeVito, C. & Hopkins, J. (2001). Attachment, parent ing and marital dissatis faction as predictors of disruptive behavior in preschoolers. Development and Psychopatholoy, 13, 215-231. Dodge, K. A. (1990). Developmental psychopath ology in children of depressed mothers. Developmental Psychology, 16 3-6. Dodrill, C. B. (1981). An economical method for the evaluation of general intelligence in adults. Journal of Consulting & Clinical Psychology, 49, 668-673 Downey, G., & Coyne, J. C. (1990). Children of depressed parents: An integrative review. Developmental Psychology, 108, 50-76. Dumas, J.E. & Wahler, R.G. (1983). Predictors of treatment outcome in parent training: Mother insularity and socioeconomic disadvantage. Behavioral Assessment, 5, 301-313. Dunn, L., & Dunn, L. (1997). Peabody Picture Vocabulary Test, 3rd Edition Circle Pines, MN: American Guidance Service. Eisenstadt, T. H., Eyberg, S., McNeil, C., Ne wcomb, K., & Funderburk, B. (1993). Parent child interaction therapy with behavior problem child ren: Relative effectiv eness of two stages and overall treatment outcome. Journal of Clinical Child Psychology, 22, 42-51. Eyberg, S. (1992). Parent and teacher behavi or inventories for th e assessment of conduct problem behaviors in children. In L. Va ndeCreek, S. Knapp, & T.L. Jackson (Eds.), Innovations in clinical practice: A sourcebook (pp. 261-270). Sarasota, FL: Professional Resource Press/Professional Resource Exchange Inc. Eyberg, S. M., Boggs, S. R., & Rodriguez, C. M. (1992). Relationships between maternal parenting stress and child with disruptive behavior. Child and Family Behavior Therapy, 14, 1-9.
167 Eyberg, S., Funderburk, B., Hembree-Kigin, T ., McNeil, C., Querido, J., & Hood, K. (2001). Parent-Child Interaction Ther apy with behavior problem children: One and two year maintenance of treatment effects in the family. Child and Family Behavior Therapy 23, 1-20. Eyberg, S. M., & Matarazzo, R. G. (1980). Traini ng parents as therapists: A comparison between individual parent-child interaction tr aining and parent gr oup didactic training. Journal of Clinical Psychology, 36, 492-499. Eyberg, S., & Pincus, D. (1999). ECBI & SESBI-R: Eyberg Child Behavior Inventory and SutterEyberg Student Behavior Invento ry-Revised professional manual. Odessa, FL: Psychological Assessment Resources Eyberg, S. M., & Robinson, E. A. (1982). Parent child interaction training: Effects on family functioning. Journal of Clinical Child Psychology, 11, 130-137. Fagot, B. I., & Pears, K. C. (1996). Changes in attachment during the third year: Consequences and predictions. Development and Psychopathology, 1, 15-30. Franz, C. E., McClelland, D. C., & Weinberg er, J. (1991). Child antecedents of social accomplishment in midlife adults : A 36-year prospective study. Journal of Personality & Social Psychology, 60 586-595 Frick, P. J., Lahey, B. B., Loeber, R., Stouthame r-Loeber, M., Christ, M. A. C., & Hanson, K. (1992). Familial risk factors to oppositiona l defiant disorder and conduct disorder: Parental psychopathology and maternal parenting. Journal of Consulting and Clinical Psychology, 60, 49-55. Foote, R., Eyberg, E., & Schuhmann, E. (1998). Pa rent-child interaction approaches to the treatment of child behavior problems. In T.H. Ollendick & R.J. Prinz (Eds.), Advances in Clinical Child Psychology. (Vol. 20, pp. 125-151). New York: Plenum Press. Funderburk, B. W., Eyberg, S. M., Newcomb, K ., McNeil, C. C., Hembre e-Kigin, T., & Capage, L. (1998). Parent-child interaction therapy w ith behavior problem children: Maintenance of treatment effects in the school setting. Child Family Behavior Therapy, 21, 17-38. Funderburk, B. W., Eyberg, S. M., Rich, B. A., & Behar, L. (2003). Further psychometric evaluation of the Eyberg and Behar rating scal es for parents and teach ers of preschoolers. Early Education and Development, 14, 67-81. Gelfand, D., & Teti, D. (1990). The eff ects of maternal depression on children. Clinical Psychology Review, 10 262-272 Greenberg, M.T., Speltz, M.L., DeKlyen, M., & E ndriga, M.C. (1992). Attachment security in preschoolers with and without exte rnalizing problems: A replication. Development and Psychopathology, 5, 91-213.
168 Gottman, J.M. & Rushe, R.M. (1993). The analysis of change: Issues, fallacies, and new ideas. Journal of Consulting and Clinical Psychology, 61 907910. Gould, M.S., Shaffer, D., & Kaplan, D. (1985). The characterist ics of dropouts from a child psychiatry clinic. Journal of the American Acade my of Child Psychiatry, 24, 316-328. Hartman, R.R., Stage, S.A., & We bster-Stratton, C. (2003). A growth curve analysis of parent training outcomes: Examining the influence of child risk factors (inattention, impulsivity, and hyperactivity problems), parental and family risk factors. Journal of Child Psychology and Psychiatry, 44, 388-398 Harwood, M. & Eyberg, S. (2004). Child-directed in teraction: Prediction of change in impaired mother-child functioning. Jounral of Abnormal Psychology, 44, 335-347. Hedeker, D. & Gibbons, R.D. (1997). Application of random-effects pattern-mixture models for missing data in longitudinal studies. Psychological Methods, 2, 64-78. Hinshaw, S.P., Lahey, B.B., & Hart, E.L. (1993) Issues of taxonomy and comorbidity in the development of conduct disorder. Development and Psychopathology, 5, 31-49. Hood, K. K., & Eyberg, S. M. (2003). Outcomes of parent-child interact ion therapy: Mothers' reports of maintenance three to six years after treatment. Journal of Clinical Child and Adolescent Psychology, 32 419-429. Hox, J. (2002). Multilevel an alysis: Techniques and appli cations. Mahwah, NJ: Lawrence Erlbaum Associates. Jensen, P.S., Martin, D., & Cantwell, D.P. (1997). Comorbidity in ADHD: Implications for research, practice, and DSM-V. Journal of the American Academy of Child and Adolescent Psychiatry, 36, 1056. Kazdin, A.E. (1995). Child, parent, and family dysfunction as predictors of outcome in cognitive-behavior treatmen t of antisocial children. Behaviour Research and Therapy, 33, 271-281. Kazdin, A. E. (1996). Combined and multim odel treatments in child and adolescent psychotherapy: Issues, challeng es, and research directions. Clinical Psychology-Science & Practice, 3 69-100 Kazdin, A.E., Holland, L., & Crowley, M. (1997). Fa mily experience of barriers to treatment and premature termination from child therapy. Journal of Consulting and Clinical Psychology, 65, 453-463. Kazdin, A.E., Holland, L., Crowley, M., & Breton, S. (1997). Barriers to treatment participation scale: Evaluation and validation in the context of child outpatient treatment. Journal of Child Psychology and Psychiatry, 38, 1051-1062.
169 Kazdin, A.E., Mazurick, J.L., & Bass, D. (1993). Risk for attrition in treatment of antisocial children and families. Journal of Clinical Child Psychology, 22, 2-16. Kazdin, A.E. & Wassell, G. (1998). Treatment completion and therapeutic change among children referred for outpatient therapy. Professional Psychology: Research and Practice, 29, 332-340. Kazdin, A.E. & Wassell, G. (1999). Barriers to tr eatment participation an d therapeutic change among children referred for conduct disorder. Journal of Clinical Child Psychology, 28, 160-172. Kazdin, A.E. & Wassell, G. (2000). Therapeutic change in children, parents, and families resulting from treatment of children with conduct problems. Journal of the American Academy of Child and Adolescent Psychiatry, 39, 414-420. Kazdin, A.E. & Whitley, M.K. (2003). Treatment of parental stress to enhance therapeutic change among children referred for a ggressive and antisocial behavior. Journal of Consulting and Clinical Psychology, 71, 504-515. Lahey, B. B., Piacentini, J., McBurnett, K., Stone, P., Hartdagen, M.A., & Hynd, G. (1988). Psychopathology in the parents of children with conduct disorder and hyperactivity. Journal of the American Academy of Child and Adolescent Psychiatry, 27, 163-170. Lavigne, J.V., Gibbons, R.D., Christoffel, K.K., Arend, R., Rosenbaum, D., Binns, H., Dawson, N., Sobel, H., & Isaacs, C. (1996). Preval ence rates and correla tes of psychiatric disorders among preschool children. Journal of the American Academy of Child and Adolescent Psychiatry, 35, 204-214. Loeber, R. (1988). The natural histories of juvenile conduct problems: Evidence for developmental progressions. In B.B. Lahey & A.E. Kazdin (Eds.) Advances in clinical child psychology (Vol. 11, pp. 97-115). New York: Plenum Press. Loeber, R. (1991). Antisocial behavior : More enduring than changeable? Journal of the American Academy of Child & Adolescent Psychiatry, 31, 393-397. Loeber, R., & Schmaling, K. B. (1985). Empirical evidence and covert patterns of antisocial conduct problems. Journal of Abnormal Child Psychology, 12 337-352. Mash, E. J., & Johnston, C. J. (1990). Determin ants of parenting stress: Illustrations from families of hyperactive children and fam ilies of physically abused children. Journal of Clinical Child Psychology, 19, 313-328. McKabe, K.M., Rodgers, C., Yeh, M., & Hough, R. (2004). Gender differences in childhood onset conduct disorder. Development and Psychopathology, 16, 179-192. McKay, M. & Bannon, W. (2004). Engaing families in child mental health services. Child and Adolescent Psychiatric Clin ics of North America, 13, 905-921.
170 McMahon, R.J. (1994). Diagnosis, assessment and treatment of externalizing problems in children: The role of longitudinal data. Journal of Consulting and Clinical Psychology 62, 901-917. McNeil, C.B., Capage, L., Bahl, A., & Blanc, H. (1999). Importance of early intervention for disruptive behavior problems: Comparison of treatment and waitlist control groups. Early Education and Development, 10, 445-454. McNeil, C., Eyberg, S., Eisendstadt, T., Newcomb, K., & Funderburk, B. (1991). Parent child interaction therapy with behavior problem child ren: Generalization of treatment effects in the school setting. Journal of Clinical Child Psychology, 20, 140-151. Moffit, T. E. (1990). Juvenile delinquency a nd attention deficit di sorder: Developmental trajectories from age 3 to 15. Child Development, 61, 893-910. Moffitt, T.E. (1993a). Adolescent-limited and lif e-course-persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100, 674-701. Moffitt, T.E. (1993b). The neuropsychology of conduct disorder. Development and Psychopathology, 5, 135-151. Moffitt, T.E. & Caspi, A. (2001). Childhood predicto rs differentiate life-course-persistent and adolescence-limited antisocial pathways among males and females. Development and Psychopathology, 13, 355-375. Moffitt T.E., Caspi, A., Dickson, N., Silva, P., & Stanton, W. (1996). Childhood-onset versus adolescent-onset antisocial conduct problems in males: Natural history from ages 3 to 18 years. Development and Psychopathology, 8, 399-424. Morgan, J., Robsinson, D., & Aldridge, J. ( 2002). Parenting stress and externalizing child behaviour. Child and Family Social Work, 7, 219-225. Munson, J.A., McMahon, R.J., & Spieker, S.J. (2001). Structure and variability in the developmental trajectory of childrens ex ternalizing problems: Impact of infant attachment, maternal depressive symptomatology, and child sex. Development and Psychopathology, 13, 277-296. Nagin, D. & Tremblay, R. (1999). Trajecto ries of physical aggression, opposition, and hyperactivity on the path to physically vi olent and non-violent juvenile delinquency. Child Development, 70, 1181-1196. Nixon, R. (2001). Changes in hyperactivity and temperament in behaviourally disturbed preschoolers after Parent Child Interaction Therapy (PCIT). Behaviour Change, 18, 168176. Nixon, R.V., Sweeney, L., Erickson, D.B., & To uyz, S.W. (2003). Parent-child interaction therapy: A comparison of standard and abbr eviated treatments for oppositional defiant preschoolers. Journal of Consulting and Clinical Psychology, 71, 251-260.
171 Nock, M. & Ferriter, C. (2005). Parent manageme nt of attendance and adherence in child and adolescent therapy: A concep tual and empirical review. Clinical Child and Family Psychology Review, 8, 149-166. Olson, S. L., Bates, J. E., & Bayles, K. (1990). Early antecedents of childhood impulsivity: The role of parent-child interaction, c ognitive competence, and temperament. Journal of Abnormal Child Psychology, 18 317-334 Owens, E.B. & Shaw, D.S. (2003). Predicting growth curves of externalizing behavior across the preschool years. Journal of Abnormal Child Psychology, 31 575-590. Patterson, G.R. (1982). A social learning approach: Vol. 3. Coercive family process. Eugene, OR: Castalia. Patterson, G. (1988). Stress: A change agent for family process. In N. Garmezy & M. Rutter (Eds.), Stress, Coping and Development in Children Baltimore MD: Johns Hopkins University Press. Patterson, G. R., Capaldi, D. M., & Bank, L. (1991). An early starter model for predicting delinquency. In D. J. Pepler, & K. H. Rubin (Eds.), The development and treatment of childhood aggression (pp. 139-168). Hillsdale NJ: Lawrence Erlbaum. Patterson, G., & Forgatch, M. (1990). Initiation and maintenance of process disrupting singlemother families. In G. Patterson (Ed.), Depression and aggression in family interaction (pp. 209-246). Hillsdale, NJ: Erlbaum. Patterson, G.R., Reid, J.B., & Dishion, T.J. (1992). Antisocial boys. Eugene, OR: Castalia. Pekarik, G. (1992). Posttreatment adjustment of c lients who drop out early vs. late in treatment. Journal of Consulting and Clinical Psychology, 48, 379-385. Perez, J.C., Bell, S., Adams, R., Ga rzarella, L. & Eyberg, S.M. (2002). Tracking treatment progress of families with oppositional preschoolers. Gainesville: University of Florida, Department of Clinical & Health Psychology, Child Study Lab. Web address http://www.hp.ufl.edu/%7Es eyberg/labresearch.htm Raudenbush, S.W (1995). Hierarchical linear models to study the effects of social context on development. In J.M. Gottman (Ed.), The analysis of change (pp. 165-201). New Jersey: L. Erlbaum Associates. Raudenbush, S.W. (2001). Comparing personal trajectories and drawing causal inferences from longitudinal data. Annual Review of Psychology 52, 501-525. Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models (2nd Ed.). Thousand Oaks, CA: Sage.
172 Rogosa, D. (1995). Myths and methods: Myths a bout longitudinal research plus supplemental questions. In J.M. Gottman (Ed.), The analysis of change (pp. 3-66). New Jersey: L. Erlbaum Associates. Rubin, K., Burgress, K., Dwyer, K., & Has tings, P. (2003). Predicting preschoolers externalizing behaviors from toddler temp erament, conflict and maternal negativity. Developmental Psychology, 39, 164-176. Sanson, A., Oberklaid, F., Pedlow, R., & Prior, M. (1991). Risk indi cators: Assessment of infancy predictors of pre-school behavioural maladjustment. Journal of Child Psychology and Psychiatry, 32, 609-626. Sanders, M. & McFarland, M. (2000). Treatment of depressed mothers with disruptive children: A controlled evaluation of cognitiv e behavioral family intervention. Behavior Therapy, 31, 89-112. Schachar, R. & Wachsmuth, R. (1990). Hype ractivity and parental psychopathology. Journal of Child Psychology and Psychiatry, 31 381-392. Schuhmann, E.M., Foote, R., Eyberg, S.M., Boggs, S., & Algina, J. (1998). Parent child interaction therapy: Interim report of a randomized trial wi th short-term maintenance. Journal of Clinical Child Psychology, 27, 34-35. Shaffer, D., Fisher, P., Dulcan, M., Davies, M., Piacentini, J., Schwab-Stone, M., Lahey, B., Bourdon, K., Jensen, P., Bird, H., Camro, G ., & Regier, D. (1996). The MMH Diagnostic Interview Schedule for Children, Version 2.3 (DISC 2.3): Description, acceptability, prevalence rates, and performance in the MECA study. Journal of the American Academy of Child & Adolescent Psychiatry, 35 865-877. Shaw, D.S., Gilliom, M., Ingoldsby, E.M. & Nagin, D.S. (2003). Trajectories leading to schoolage conduct problems. Developmental Psychology, 39, 189-200. Shaw, D.S., Keenan, K., & Vondra, J.I. (1994) Developmental precurs ors of externalizing behavior: Ages 1 to 3. Developmental Psychology, 30, 355-364. Shaw, D.S., Owens, E.B., Vondra J.L., Keenan, K., & Winslow, E.B. (1996). Early risk factors and pathways in the development of early disruptive behavior problems. Development and Psychopathology, 8, 679-699. Silverthorn, P. & Frick, P.J. (1999) Developmental pathways to an tisocial behavior: the delayedonset pathway in girls. Developmental and Psychopathology, 11, 101-126. Silverthorn, P., Frick, P.J., & Reynolds, R. (2001) Timing of onset and correlates of severe conduct problems in adjudicated girls and boys. Journal of Psychopathology and Behavioral Assessment, 23, 171 181.
173 Spieker, S.J., Larson, N.C., Lewis, S.M., Keller, T.E., & Gilchrist, L. (1999). Development trajectories of disruptive behavi or problems in preschool children of adolescent mothers. Child Development, 70, 443-458. Wahler, R. G., & Dumas, J. (1989). Family f actors in childhood psychopathology: A coercionneglect model. In T. Jacob (Ed.), Family interaction and psychopathology (pp. 381-428). New York: Plenum Press. Werba, B.E., Eyberg, S.M., Boggs, S.R., & Algina, J. (in press). Predicting outcome in parentchild interaction therapy: Responsiveness and attrition. Behavior Modification. Webster-Stratton, C. (1996). Early-onset conduc t problems: Does gender make a difference? Journal of Consulting and Clinical Psychology, 64, 540-551. Webster-Stratton, C., & Hammond, M. (1988). Matern al depression and its relationship to life stress, perceptions of child behavior problems, parenting behaviors, and child conduct problems. Journal of Abnormal Child Psychology, 11 123-129 Webster-Stratton, C., & Hammond, M. (1990). Predictors of treatment outcome in parent training for families with conduct problem children. Behavior Therapy, 21 319-337. Webster-Stratton, C., & Hammond, M. (1997). Tr eating children with early-onset conduct problems: A comparison of child and parent training intervention. Journal of Consulting find Clinical Psychology, 65 93-109. Wierzbicki, M. & Pekarik, G. (1993). A meta-analysis of psychotherapy dropout. Professional Psychology: Research and Practice, 24, 190-195. Willet, J.B. (1988). Questions and answers in the measurement of change. In American Educational Research Association (Ed.), Review of research in education. (pp. 345-422). Itasca, IL: F.E. Peacock Publishers. Willet, J.B. (1994). Measurement of change. In T. T. Husn & N. Postlethwaite (Eds.) The international encylopedia of education (2nd Edition). (pp. 671-678). Oxford, England: Pergamon. Williford, A., Calkins, S., & Keane, S. (2007). Pr edicting changes in parenting stress across early childhood: Child and maternal factors. Journal of Abnormal Child Psychology, 35, 251263. Zaitman-Zait, A. & Zumbo, B. (2005). Multilevel (HLM) models for modeling change with incomplete data: Demonstrating the eff ects of missing data and level-1 model misspecification. Paper presented at the Hier archical Linear M odeling (SIG) of the American Educational Research Association conference: Montreal, Quebec, Canada. Zimet, G.D., Dahlem, N.W., Zimet, S.G., & Fa rley, G.K. (1988). The multidimensional scale of perceived social support. Journal of Personality Assessment, 52, 30-41.
174 BIOGRAPHICAL SKETCH Jaim ee is a native Floridian. She received her Bachelor of Arts degree in psychology from Florida Atlantic University in 1998. She began her graduate training at the University of Florida and received a Master of Science degree in clinical and health Psychology in 2001. Jaimee entered doctoral candidacy in summer 2004 and co mpleted her internship in psychology at the Columbia University Medical Center/New York Presbyterian Hospital from 2005 to 2006. She took a hiatus from graduate school to give bi rth to her daughter, Sophia, and returned to complete her degree summer 2008. Jaimee resides in Lantana, Florida and plans to remain with the Treasure Coast Early Steps Program, where sh e has been working with infants and toddlers with developmental dela ys and disabilities.