This item is only available as the following downloads:
1 ANALYSIS OF CHANGE IN PARENT PERCEPTIONS OF BARRIERS TO REMAINING IN PARENT CHILD INTERACTION THERAPY By ALISON REBECCA ZISSER A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2011
2 2011 Alison Rebecca Zisser
3 To my mother and father, who have offered infinite support and encouragement as I have worked toward my goals.
4 ACKNOWLEDGMENTS I would like to thank Sheila Eyberg, Ph.D., my chair and research advisor, for her mentorship, guidance, and support as I conducted this project. I would also like to thank Stephen Boggs, Ph.D., Allyson Hall, Ph.D., Michael Marsiske, Ph.D. and Ronald Rozen sky, Ph.D., members of my dissertation committee, for the time and energy they have devoted to providing thoughtful feedback as I have progressed through my doctoral qualitative exam and dissertation proposal. I am also very appreciative to Dr Marsiske for his expert guidance as I conducted the statistical analyses for this dissertation. I would like to thank the members of the Child Study Laboratory for their encouragement and feedback as I worked on this dissertation. In particular, I would like to thank Reesa Donnelly, Ph.D., Amanda Seib, M.S. and Katie Rogers, B.A. for their assistance with data collection while I was away from Gainesville. I would especially like to thank my parents, Carolyn and Elliot Zisser, for their love, support, and encouragement throughout my graduate studies and as I worked on this dissertation. Finally, I would like to acknowledge the National Institute of Mental Health (R01 MH 072780) for funding the project from which these data were collected.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 9 LIST OF ABBREVIATIONS ................................ ................................ ........................... 10 ABSTRACT ................................ ................................ ................................ ................... 11 1 INTRODUCTION ................................ ................................ ................................ .... 13 Literature Review ................................ ................................ ................................ .... 13 Treatment Attendance and Attrition ................................ ................................ .. 13 Consideration of Treatment Barriers within Access to Treatment Models ........ 15 Barriers to Treatment Participation and Continuation in Child and Family Therapy ................................ ................................ ................................ ......... 17 Therapeutic Process Factors and the Therapeutic Relationship ...................... 19 Analysis of Change over Time ................................ ................................ ................ 23 Identifying Predictors of Change in Barriers to Treatment Continuation ................. 24 Treatment Credibility and Expectancies ................................ ........................... 24 Child Disruptive Behavior ................................ ................................ ................. 27 Family Socioeconomic Status ................................ ................................ .......... 29 Summary ................................ ................................ ................................ ................ 30 Parent Child Interaction Therapy ................................ ................................ ............ 31 Study Aims and Hypotheses ................................ ................................ ................... 32 2 METHOD ................................ ................................ ................................ ................ 34 Participants ................................ ................................ ................................ ............. 34 Screening Measures ................................ ................................ ............................... 34 Measures of Outcome Predictors ................................ ................................ ........... 36 Outcome Measure ................................ ................................ ................................ .. 39 Procedures ................................ ................................ ................................ ............. 40 Analyses ................................ ................................ ................................ ................. 44 Normality ................................ ................................ ................................ .......... 44 Multicollinearity ................................ ................................ ................................ 44 Reliability ................................ ................................ ................................ .......... 45 Descriptive Statistics ................................ ................................ ........................ 45 Analysis of Change ................................ ................................ .......................... 46 Consideration of Unbalanced Designs ................................ ............................. 47 Implications of Treatm ent Dropout on Subsequent Analyses ........................... 48 Aims 1 and 2 Implementing the Hierarchical Linear Model ............................ 49 Aim 1: Measuring Change in Parent Perceived Barriers Across Time ............. 50
6 Aim 2: Modeling Predictors of Parent P erceived Barriers ................................ 51 Aim 3: Barriers at the Last Completed Session as a Predictor of Dropout ....... 52 3 RESULTS ................................ ................................ ................................ ............... 56 Preliminary Analyses ................................ ................................ .............................. 56 Normality an d Multicollinearity ................................ ................................ .......... 56 Reliability ................................ ................................ ................................ .......... 57 Descriptive Analyses ................................ ................................ ............................... 57 Multilevel Modeling of Change in Perceived Barriers to Treatment Continuation ... 59 Unconditional Means Model ................................ ................................ ............. 62 Change in Barriers over Time ................................ ................................ ........... 63 Line ar time ................................ ................................ ................................ 63 Quadratic time ................................ ................................ ............................ 64 Addressing Attrition in the Model ................................ ................................ ...... 65 Child Disruptive Behavior ................................ ................................ ................. 67 Treatment Expectations ................................ ................................ .................... 69 Parent Socioeconomic Status ................................ ................................ .......... 73 Relationship between SES and Barriers to Treatment Continuation at the Start of Treatment ................................ ................................ ................................ ........ 74 Predicting Drop Out from Treatment ................................ ................................ ....... 74 4 DISCUSSION ................................ ................................ ................................ ....... 100 Predictors of Level and Change of Perceived Barriers ................................ ......... 102 Child Disruptive Behavior ................................ ................................ ............... 102 Mother Expectations and Beliefs in the Credibili ty of Treatment .................... 103 Family Socioeconomic Status ................................ ................................ ........ 106 Attr ition from Treatment ................................ ................................ ........................ 106 Barriers to Treatment Continuation and Models of Health Services Usage .......... 108 Limitations of the Study and Future Directions ................................ ..................... 109 Meeting Assumptions for Hierarchical Linear Modeling ................................ .. 109 Generalizability of Study Results ................................ ................................ .... 112 Measurement of Treatment Barriers ................................ ............................... 114 The Parent Therapist Relationship and Future Research on Treatment Barriers ................................ ................................ ................................ ........ 115 LIST OF R EFERENCES ................................ ................................ ............................. 1 18 B IOGRAPHICAL S KETCH ................................ ................................ .......................... 128
7 LIST OF TABLES Table page 2 1 Cronbach alpha values ac ross occasions of treatment for the Eyberg Child Behavior Inventory (ECBI) and the Barriers to Treatment Participation Scale Revised (BTPS R) ................................ ................................ .............................. 54 3 1 Skewness and kurtosis for BTPS R at each occasion of treatment .................... 77 3 2 Correlations between predictor variables ................................ ........................... 78 3 3 Means for descriptive and demographic variables ................................ .............. 79 3 4 Frequency counts for demographic variables ................................ ..................... 80 3 5 Means for descriptive and demographic variables by group status .................... 81 3 6 Frequency counts for demographic variables by group status ........................... 82 3 7 Means for level 2 variables ................................ ................................ ................. 83 3 8 Resu lts of model tests for parent reported barriers to treatment continuation, Model 1 Model 5 ................................ ................................ .............................. 84 3 9 Results of model tests for parent reported barriers to treatment continuation, Model 6 Model 9 ................................ ................................ .............................. 85 3 10 Results of model tests for parent reported barriers to treatment continuation, Model 10 Model 12 ................................ ................................ .......................... 86 3 11 Results of model tests for parent rep orted barriers to treatment continuation, Model 13 Model 16 ................................ ................................ .......................... 87 3 12 Results of model tests for parent reported barriers t o treatment continuation, Model 17 Model 20 ................................ ................................ .......................... 88 3 13 Results of model tests for parent reported barriers to treatment cont inuation, Model 21 Model 24 ................................ ................................ .......................... 89 3 14 Results of model tests for parent reported barriers to treatment continuation, Model 25 Model 27 ................................ ................................ .......................... 90 3 15 Cha nge in fit across models ................................ ................................ ............... 91 3 16 Change in within subjects and between subjects variance explained across models ................................ ................................ ................................ ................ 92
8 3 17 Summary of stepwise logistic regression predicting baseline predictors of attrition from treatment ................................ ................................ ....................... 93 3 18 Summary of hierarchical logistic regression predicting attrition from treatment .. 94
9 LIST OF FIGURES Figure page 2 1 Procedure for completion of BTPS R ................................ ................................ 55 3 1 Spaghetti plot of Blom normalized BTPS R score for each parent child dyad .... 95 3 2 Curve estimation for pattern of change of Blom normalized BTPS R score ....... 96 3 3 Fitted model of linear trajectory of change in BTPS R ................................ ........ 97 3 4 Fitted model of quadratic trajectory of change in BTPS R scores ...................... 98 3 5 Fitted model of linear change in BTPS R scores ................................ ................ 99
10 LIST OF ABBREVIATION S 2LL 2 log likelihood ADHD Attention Deficit Hyperactivity Disorder AIC BIC BTPS Barriers to Treatment Participation Scale BTPS R Barriers to Treatment Participation Scale Revised CD Conduct Disorder CDI Child Directed Interaction CEQ P Credibility/ Expectancies Questionnaire Parent Version DBD Disruptive Behavior Disorders E CBI Eyberg Child Behavior Inventory HBM Health Belief Model ODD Oppositional Defiant Disorder PCIT Parent Child Interaction Therapy PDI Parent Directed Interaction SES Socioeconomic Status
11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ANALYSIS OF CHANGE IN PARENT PERCEPTIONS OF BARRIERS TO REMAINING IN PARENT CHILD INTERACTION THERAPY By Alison Zisser August 2011 Chair: Sheila Eyberg Major: Psychology Parent perceived barriers to treatment represent obstacles that interfere with investigated how perceived barr iers change over the course of treatment. This study explored changes in parent reports of treatment barriers during Parent Child Interaction Therapy (PCIT) through repeated measurement of parent perceived barriers at each treatment session. A weekly measu re of child disruptive behavior was included as a time varying covariate of parent perceived barriers. Parent expectations for treatment and family socioeconomic status were investigated as predictors of both level of parent perceived barriers and change i n parent perceived barriers. Relations between perceived barriers at the last attended therapy session and treatment attrition were also investigated. Data from 46 mother child dyads participating in a family based, behavioral treatment program for childr en with ADHD were used in this study. Hierarchical linear modeling was implemented to investigate predictors of level and change in parent perceived barriers to treatment participation. Results indicated a linear decrease in
12 parent perceived barriers over the course of PCIT. Level of parent perceived barriers and mid point of treatment. Additionally, on occasions at which paren ts reported an above average level of child disruptive behavior, parents also reported a higher level of perceived barriers. The relationship between change in perceived barriers and average level of child disruptive behavior across treatment varied signif icantly between families. Additionally, parent reported barriers to treatment continuation at the last attended session predicted attrition from treatment at a trend level of significance. Results suggest that perceived barriers are not a fixed construct, but more a reflection of varying influences a family experiences throughout treatment. Addressing parent perceived barriers during the course of treatment and remaining sensitive to predictors of treatment barriers within the therapeutic context may result in decreased attrition from child and family mental health treatment.
13 CHAPTER 1 INTRODUCTION Literature Review Past research in clinical child psychology has focused on the development of effective therapies and more recently on evidence based treatments for childhood disorders. Less attention has been given to the identification of barriers to treatment delivery a nd treatment engagement (Nock & Ferriter, 2005). Research suggests that one in eight children has a mental health problem that significantly impairs familial, academic, or community functioning (Costello, Egger, & Angold, 2005). However, approximately 75% of children with emotional or behavioral disorders do not receive needed mental health services (Ringel & Sturm, 2001). Identification and reduction of treatment barriers are critical in increasing access to child mental health treatment programs. Treatmen t Attendance and Attrition Relatively little research exists on determinants of attendance in psychosocial treatments for children. Child therapy warrants special attention because it differs in significant ways from adult treatments. In child treatment, parents often play a critical transportation, and legal consent (Nock & Ferriter, 2005). Additionally, although the treatment often involves multiple individuals with whom the child is in frequent contact, including parents, adult guardians, siblings, and teachers. This as well as the unique referral status of children (i.e., rarely self referred) also influences motiva tion to attend treatment.
14 Treatment attrition is a significant obstacle to attaining successful outcomes among children who begin outpatient therapy. It is estimated that 40 to 60% of children who enter treatment leave prematurely and against the recommen dations of treatment providers (Wierzbicki & Pekarik, 1993). Besides reducing the number of children who benefit from treatment, attrition leads to other significant consequences (Kazdin, Holland, Crowley, & Breton, 1997). Children who drop out of treatmen t tend to be more impaired than children who remain in treatment (Kazdin, Mazurick, & Siegel, 1994). Additionally, when compared to children who remain in treatment, those who drop out are significantly less likely to show improvement (Boggs et al., 2004; Prinz & Miller, 1994; Santisteban et al.,1996). Finally, dropping out of treatment increases the costs of providing mental health services. Children that eventually drop out of treatment are more likely to have cancelled and missed appointments, which resu lts in increased staff time, non reimbursed appointments, and occupied therapy slots that could be used by other children (Kazdin, Holland, Crowley, & Breton, 1997). Most research on treatment attendance and attrition in the adult and child therapy liter ature has investigated the role of demographic variables (e.g., participant age, socioeconomic status [SES]) in predicting attendance. In the child therapy literature, factors including ethnic minority status, parent psychopathology, low SES, single parent status, parent psychopathology, and severity of child psychopathology are related to low attendance and premature treatment termination (Armbruster & Schwab Stone, 1994; Fernandez & Eyberg, 2009; Gould, Shaffer, & Kaplan, 1985; Kazdin, Mazurick, & Bass, 1 993; Mccabe, 2002; Wierzbicki & Pekarik, 1993). Studies investigating attrition from behavioral treatment programs for children with disruptive
15 behavior disorders have also demonstrated that parents who are more verbally negative and offer less praise to t heir children at pre treatment are more likely to drop out of treatment (Fernandez & Eyberg, 2009; Werba, Eyberg, Boggs, & Algina, 1996). Although these findings provide useful information about characteristics of families that are at higher risk for attri tion, little information is revealed regarding why the families stop attending treatment (Nock & Ferriter, 2005). Nock and Ferriter (2005) propose that barriers to treatment are a more useful way of conceptualizing why families miss appointments or stop at tending treatment than simply identifying demographic variables associated with attrition. Consideration of Treatment Barriers within Access to Treatment Models Treatment barriers have been addressed in various health related disciplines, including publi c health (Janz, Champion, & Strecher, 2002). The Health Belief Model (HBM), a conceptual framework used to explain health behavior, is a prominent model in the public health literature. The HBM posits that barriers represent impediments to following recomm ended behaviors that would contribute to improved health. The HBM reflects a value expectancy theory involving the desire to get well (value) and the belief that a certain available health action will prevent or ameliorate illness (expectation). Expectatio n is further described in terms of the estimate of severity of the condition and the likelihood that the condition may be improved through personal action (Janz et al., 2002). The HBM also suggests that individuals will act to prevent or treat health condi tions if they believe the condition might have serious consequences or if an Additionally, individuals take action if the potential benefits of the action outweigh the p erceived costs (Rosenstock, 1974).
16 The Behavioral Model of Health Services Use (Andersen, 1968) also addresses treatment utilization. This model focuses primarily on determinants of access to care. The behavioral model suggests that the tendency to use he alth services is related to the need for care (Andersen, 1968). Andersen (1968) suggested that predisposing, enabling, and need factors will have varying levels of in fluence on utilization, depending on the type of health service pursued (e.g., medical, dental, mental health). Predisposing factors include demographic characteristics, social standing, and health beliefs (i.e., values, knowledge, and attitudes about hea lth or health services that psychological characteristics may also be important predisposing factors that relate to health service usage (Andersen, 1995; Rosneau, 1994). Enablin families to access care, community enabling factors such as health care facilities, and personal enabling factors such as knowledge of how to access services, must both be pr esent. Income, health insurance, and travel time are all personal enabling factors (Andersen, 1995). Social relationships are also enabling factors, because social relationships both facilitate and sometimes impede health services usage (Bass & Noelker, 19 87; Counte & Glandon, 1991). In the original model of health services use (Andersen, 1968), the concept of need, including both perceived need and evaluated need, was considered a third contributing influence on services use. ervices use has expanded since its original conceptualization to include factors that influence health services utilization and
17 outcome. Additional factors involve consideration of the greater health care system (i.e., health policy and resource organizati on) as determinants of services use. Measures of health outcome, including consumer satisfaction and post treatment health status, as perceived by the consumer and evaluated by the provider, are also included in the revised model (Andersen, 1995). The beh avioral model of health services use presents factors that increase the likelihood of families accessing treatment. The absence of or variations in predisposing factors, enabling factors, and perceived need represent barriers to accessing treatment. For ex ample, a positive impression of mental health care, which is a predisposing factor, would facilitate treatment access. However, a negative impression of mental health care, also a predisposing factor, would likely hamper treatment access and represents a t reatment barrier. Although the behavioral model of health services use largely addresses the initial access to treatment, families continue to function within a system of treatment facilitators and barriers throughout the treatment process. Barriers also v ary considerably from family to family and relate to the extent to which parents and their children view treatment as acceptable (Kazdin, 2000). Barriers to Treatment Participation and Continuation in Child and Family Therapy A barriers to treatment model specific to the child treatment domain was proposed by Kazdin and colleagues (Kazdin, Holland, & Crowley, 1997; Kazdin, Holland, Crowley, & Breto n, 1997). In contrast to Anders e barriers model focuses on ba rriers experienced while engaged in treatment. In this model, barriers to treatment are conceptualized as stemming from the interaction between patient and treatment, shifting focus away from factors specific to the individual child, parent, family, therap ist, or treatment alone.
18 The barriers to treatment model has been investigated in a prospective study of children, ages 3 to 14 years, who were referred to treatment for oppositional, aggressive, and antisocial behaviors (Kazdin, Holland, & Crowley, 1997) Study results indicated that the experience of barriers to participation in treatment was significantly related to premature attrition from therapy and was not explained by established child, parent, or family factors that also contributed to treatment a ttrition (e.g., level of child psychopathology, SES, minority status). Further, parent perceptions of few barriers to treatment acted as a protective factor (i.e., lessened risk of dropout) for families that were at higher risk for attrition based on the e stablished child, parent, and family factors measured at pre treatment (Kazdin, Holland, & Crowley, 1997). Parents who reported higher levels of perceived barriers were also in treatment for significantly fewer weeks and had higher rates of canceling or mi ssing scheduled appointments before premature termination (Kazdin, Holland, Crowley, & Breton, 1997). Parent perceptions of barriers served a partial meditational role in the relationship between attrition and child, parent, or family factors (Kazdin, Holl and, & Crowley, 1997). Relations between higher reported barriers and treatment attrition have been replicated in PCIT treatment within community mental health settings as well (Danko & Budd, 2009). The barriers to treatment model reflects the notion that families experience obstacles to participating in treatment, which increases the likelihood that the family will adhere less well or stop attending treatment (Nock & Ferriter, 2005). The barriers to treatment model posits that barriers to remaining in tre atment can be classified into four domains: (a) obstacles that compete with treatment (e.g., conflict with a significant other or children about attending treatment); (b) treatment demands (e.g., thoughts that
19 treatment is too long, costly, difficult, or d emanding), (c) perceived relevance of perception of support from the the rapist, and liking of the therapist) (Kazdin, Holland, Crowley, & Breton, 1997). Kazdin, Holland, Crowley, and Breton (1997) developed the Barriers to Treatment Participation Scale (BTPS), a measure of perceived barriers to continuing in treatment, design ed to reflect the four proposed domains of barriers (i.e., obstacles that compete with treatment, treatment demands, perceived relevance of treatment, relationship with the therapist). However, factor analysis of the scale using a principal components anal ysis with varimax rotation revealed that performance on the BTPS was best represented by a single factor, suggesting that the scale provides a total barriers score, rather than scores for each of the four domains conceptualized during scale development (Ka zdin, Holland, Crowley, & Breton, 1997). Therapeutic Process Factors and the Therapeutic Relationship Although relations between demographic factors and barriers have been examined in relatively few studies, each with a different participant sample, the l ack of strong relations between treatment barriers scores and demographic factors, such as SES, suggests that therapist patient relationship factors likely contribute to parent perceptions of treatment barriers (Kazdin, Holland, & Crowley, 1997; Stevens, K elleher, Ward Estes, & Hayes, 2006). Therapist patient interaction factors that influence the quality of the therapeutic relationship are therapy process variables (Harwood & Eyberg, 2004). Research has suggested that process variables affect treatment out comes more
20 strongly than baseline patient characteristics (Kolb, Beutler, Davis, Crago, & Shanfield, 1985). Studies investigating relations between the therapeutic relationship in child and family therapy and treatment attrition suggest that the quality of therapeutic relationship is a significant predictor of dropout (Garcia & Weisz, 2002; Harwood & Eyberg, 200 4) and an important process variable in determining outcome (Shirk & Karver, 2003). Garcia and Weisz (2002) found on an instrument measuring parent reasons for ending treatment that the following items loaded highly (i.e., factor loadings of .70 .90) on a therapeutic relationship problems subscale. This subscale in turn significantly predicted attrition from treatment: (a) the therapist did not seem to be doing the right things; (b) the therapist did not seem to understand ; (c) the therapist did not talk about the right problems ; (d) did not like the therapist It is important to consider both the strengths and weaknesses of the barriers to treatment model. One strength is that this model organizes into one model several constructs identified by past research as predictive of treatment attendance and adherence (Nock & Ferriter, 2005). These constructs include parent stre ssors ( Prinz & Miller, 1994 ), therapeutic relationship factors (Garcia & Weisz, 2002; Harwood & Eyberg, 2004), and parent perceptions of treatment relevance (Day & Reznikoff, 1980; Plunkett, 1984). Thus, the barriers to treatment model provides a parsimonious mechanism for conceptua lizing factors that are related to treatment attendance and adherence.
21 The barriers to treatment model also contributes to understanding how child and parent variables such as parent psychopathology, child psychopathology, and child age relate to treatme nt attendance. Although these child and parent variables may not directly contribute to treatment attendance, they likely contribute to parent perceptions of barriers to attending treatment (Nock & Ferriter, 2005). Another important strength of the barrier s model is its ability to explain, at least partially, the relationship between parent and child characteristics and attendance in treatment. Weaknesses inherent in the barriers to treatment model and its measurement with the BTPS also exist. First, this m odel focuses exclusively on perceptions held by treatment or their feelings about the therapeutic relationship (Kazdin, Holland, Crowley, & Breton, 1997). Additionally, the b arriers to treatment model may be more applicable to families of children with externalizing disorders than with internalizing disorders (Kazdin, Holland, Crowley, & Breton, 1997). Most studies investigating the Kazdin et al. model, including the study in which the BTPS was developed, have included children with externalizing problems only (Kazdin, Holland, Crowley, & Breton, 1997). This is problematic because families of conduct disordered children are often described as having higher rates of clinical dys function, fewer resources, and multiple stressors (Kazdin, 1995), resulting in the inclusion of many of these familial characteristics in the barriers to treatment model. A weakness of past research on the barriers model, especially pertinent to the curr ent study, is the reliance on pre treatment or post treatment assessment of parent perceived barriers (Kazdin & Wassell, 1998). Assessment of barriers at pre treatment
22 fails to capture barriers that arise during treatment (e.g., negative opinion of treatme nt provider, frustration with length of treatment). It is also important to consider that families respond to different treatments in different ways, which a pre treatment assessment of treatment barriers fails to capture. Families participating in treatme nt modalities with a predetermined number of sessions or established treatment criteria for termination may show improvement in their treatment goals well before the end of treatment, which could affect their perception of treatment relevance towards the e nd. Further, other families may reach the final session of a time limited treatment without having shown much or any improvement. Differences in rate of response to treatment are not captured by a pre treatment assessment of the barriers to treatment conti nuation. Finally, it is often difficult to collect data from families who have dropped out of treatment. Thus, studies that rely on post treatment measures of parent perceived barriers may disproportionately reflect the impressions of treatment completers. To date, no study has investigated the direction or magnitude of change in parent perceptions of barriers to treatment participation throughout treatment or even whether parent perceptions of barriers change. Because this study refers to barriers relative to continued participation in treatment rather than barriers to treatment access, this study examine whether perceptions of barriers to treatment continuation change for mothers participating in Parent Child Interaction Therapy (PCIT). This study used hierarchical linear modeling to investigate change in parent perceptions of barriers to treatment continuation.
23 Analysis of Change over Time Studies investigating change i n groups or individuals across time use longitudinal designs with repeated, time ordered observations (Wu, Clopper, & Wooldridge, 1999). Pre post design is an example of a longitudinal method in which the outcomes of interest are measured at an initial tim e point and then measured again at a second specified time or at the completion of treatment. Univariate analysis of variance (ANOVA), multivariate ANOVA, and regression are all traditional approaches used to analyze data in pre post designs. Difference sc ores have also been used as a measure of change over time. In their comparison of traditional univariate and multivariate repeated measures designs with multilevel modeling, Wu et al. (1999) discussed the following goals for analysis of longitudinal data: (a) direct investigation of intra individual change; (b) identification of inter individual differences in intra individual change; (c) analysis of the relationship between inter individual changes and intra individual changes; and (d) investigation of the variables that impact intra individual and inter individual change. Although multilevel models are related to traditional ANOVA approaches in that both are from the family of general linear models, the approaches differ in the operationalization of time, the assumptions made, and the presentation of the output data (Wu et al., 1999). Methodological researchers have posited that ANOVA approaches do not sufficiently address the goals for longitudinal research outlined by Wu et al. (1999), especially i n the measurement of intra individual change over time (Rogosa, 1995). Multilevel model analysis, also referred to as mixed effects models and hierarchical linear models (HLM; Bryk & Raudenbush, 1992), allows for the investigation of both
24 inter individual and intra individual change over time (Bryk & Raudenbush, 1987). An additional strength of the multilevel modeling approach is the ability to look at predictors of change as well as time varying covariates, which are predictor variables that are measured o n repeated occasions (Tabachnick & Fidell, 2007). Identifying Predictors of Change in Barriers to Treatment Continuation Identifying predictors of change in barriers is critical to further understand why some families remain in treatment and reach success ful outcomes, while other families drop out early. Research has shown that parent perceptions of barriers to treatment as well as child, parent, and family characteristics such as SES and ethnic minority status are related to treatment attrition (Kazdin, H olland, & Crowley, 1997). Therapists have minimal power to alter demographic risk factors of attrition such as SES or ethnicity. However, if there are predictors of perceived barriers to treatment continuation that therapists can identify, be aware of, and target for change, then risk of treatment dropout may be attenuated. In this study, parent beliefs about treatment credibility and effectiveness were investigated as predictors of change in parent perceived barriers to continuing treatment. The frequency of child disruptive behavior was also investigated as a time varying covariate of parent perceived barriers. SES was explored as a predictor of initial level of barriers as well as a predictor of change in parent perceived barriers. Parent expectations, ch ild disruptive behaviors, and family SES are further discussed here. Treatment Credibility and Expectancies Expectancies about psychotherapy represent anticipatory beliefs that patients bring to treatment and include beliefs about procedures, therapists, o utcomes, and other aspects of the intervention and delivery (Nock & Kazdin, 2001). Treatment
25 credibility given treatment is to the patient (Kazdin, 1979). Related but distinc t constructs, investigators often use the two terms interchangeably and include credibility related items on measures of parent expectancies for treatment (Devilly & Borkovec, 2000). As a result, there has been much more research on relations between expec tancies and treatment outcome than credibility and outcome. Devilly and Borkovec (2000) proposed that treatment credibility is more related to logical thought processes of patients than affective processes, whereas treatment expectancy was thought to relat e more to affective processes than cognitive. Thus, the combined measurement of both treatment credibility and expectancies may provide a more complete evaluation of parent beliefs regarding treatment than if only one was considered. The literature on tre atment expectancies has shown mixed results, with some studies showing positive associations between expectancies and outcome (Bradley, Poser, & Johnson, 1980; Safren, Heimberg, & Juster, 1997; Sotsky et al., 1991) and other studies showing no association (Basoglu et al., 1994; Piper & Wogan, 1970). Additionally, some studies have shown a curvilinear relationship between treatment expectancies and outcome, with patients who have moderate expectations showing greater change than patients with very high or ve ry low treatment expectations (Goldstein, 1962a, 1962b). In reviews of the treatment expectancy literature, authors have suggested that methodological limitations (e.g., reliance on patient self reported outcomes, use of expectancy measures without evidenc e of psychometric quality) likely contribute to the mixed findings in this literature (Arnkoff, Glass, & Shapiro, 2001; Greenberg, Constantino, & Bruce, 2006; Noble, Douglas, & Newman, 2001).
26 Research on credibility and expectancies in the child treatme nt literature has demonstrated that parent beliefs about treatment are related to both attendance and adherence to treatment recommendations. Incongruence between parent expectancies for treatment and the actual structure of therapy (e.g., level of parenta l involvement, number of sessions needed) has been shown to predict attrition from treatment (Day & Reznikoff, 1980; Plunkett, 1984). Treatment credibility and expectancies have also significantly predicted adherence to treatment (Nock, Ferriter, & Holmber g, 2007). One study showed that after controlling for the child, parent, and family factors of age, gender, race, and SES, parent beliefs about treatment at the first therapy session predicted adherence as late as the seventh and eighth week of treatment ( Nock et al., 2007) Nock and Kazdin (2001) investigated relations between parent treatment expectancies and perceptions of treatment barriers within a family therapy modality. Results indicated that parent expectancies for therapy were a linear predicto r of barriers to treatment measured at post treatment, with lower expectations relating to higher perceived barriers. Interestingly, parent expectancies for therapy were a curvilinear predictor of treatment attendance and attrition. Parents whose expectanc ies were very high or very low attended the greatest number of sessions and were least likely to terminate from treatment prematurely. Parent expectancies also contributed significant, unique variance to parent perceived treatment barriers, attendance, and premature termination, over and above the pre treatment family, child, and parent characteristics (i.e., SES, parent stress, parent psychopathology, child symptom severity) that have traditionally been associated with these constructs (Nock and Kazdin, 20 01).
27 Considering the active role expected of parents in family based treatments, it is important to understand the ways in which parent beliefs about treatment affect their perceptions of treatment barriers. Although studies have found a relationship betw een pre treatment expectations and perceived barriers measured post treatment, no research to date has examined whether or how expectations change during treatment trea tment. Altering parent expectations for treatment structure through preparatory strategies has led to positive results including increased knowledge of therapy, higher receptivity to therapy, and more favorable outcome expectancies (Bonner & Everett, 1986; Shuman & Shapiro, 2002). Investigating relations between change in parent expectations and perceived barriers is important in determining whether parent expectations should be targeted at multiple points in treatment. Child Disruptive Behavior Researche rs have identified child behavioral impairment and improvement in disruptive behavior to be related to treatment completion and perceived barriers (Kazdin, Holland, & Crowley, 1997; Kazdin & Wassell, 1998; Perez, 2008; Stevens et al., 2006). Stevens et al. (2006) found that parents of children who met treatment goals of improved behavior reported fewer barriers to treatment than parents of children who did not meet treatment goals. Additionally, children of parents who reported higher levels of perceived ba rriers at post treatment were found to show patterns of slower improvement of disruptive behavior (Perez, 2008). In a study measuring relations between child disruptive behavior and parent reported barriers at the end of PCIT, Budd, Danko, and Legato (2010 ) reported that level of parent perceived barriers significantly predicted level of child disruptive behavior at the last attended session of
28 treatment. Budd et al. (2010) further reported that levels of parent perceived barriers and child disruptive behav ior were significantly and positively related. No study to date has investigated how change in child behavior relates to parent perceived barriers over the course of treatment. Among the various reasons that children are referred for psychological interven tion, disruptive behavior is the most common (Kazdin, Siegel, & Bass, 1990; Kazdin, 2003) and was a major focus in this study. Disruptive behaviors vary from relatively minor infractions such as talking back to significant acts of aggression. Depending on the severity and nature of the presenting problems, children with disruptive behavior disorders (DBDs) may be diagnosed with oppositional defiant disorder (ODD) or conduct disorder (CD), and attention deficit hyperactivity disord er (ADHD) is often diagnose d co morbidly with ODD or CD among pre school aged children referred to treatment (Wilens et al., 2002). Conduct problems are among the most severe childhood conditions because of the resulting impairment across multiple domains of functioning (Lambert, Wah ler, Andrade, & Bickman, 2001). Childhood DBDs are often associated with considerable family dysfunction and impairment (Nock & Kazdin, 2002). Untreated DBDs are costly to society. Children with DBDs account for a larger percentage of health care costs th an children with chronic health conditions such as epilepsy, asthma, and diabetes (Guevara, Mandell, Rostain, Zhao, & Hadley, 2003). Young children demonstrating persistently high levels of disruptive behavior fail to learn prosocial behavior and are at high risk for antisocial behavior and criminal activity in adolescence and adulthood (Loeber & Dishion, 1983; White, Moffit, Earls, Robins, &
29 Silva, 1990). Thus, it is important to maximize attendance and completion in treatments for DBDs in early childhoo d to maximize the likelihood of reversing the developmental trajectory of escalating disruptive behavior in these at risk children. Family Socioeconomic Status Family SES has been examined in the psychological and public health literatures for many years a and education (Bradley & Corwyn, 2002; Brooks Gunn, Klebanov, & Liaw 1995; Howell, 2004; Newacheck, 1992; Valenzuela, 1997). SES has also been associated with ironment, childrearing practices, and familial stability (Hoffman, 1984). Especially pertinent to research in child treatment, SES has been found to be significantly related to attrition from child and family therapy (Armbruster & Schwab Stone, 1994; Ferna ndez & Eyberg, 2009; Gould et al., 1985; Kazdin et al., 1993; Mccabe, 2002; Wierzbicki & Pekarik, 1993). This finding is not surprising, given the cost of therapy in fees, time away from work, travel expenses, and child care. Although treatment costs are represented within the barriers to treatment model (Kazdin, Holland, & Crowley, 1997), no study has found significant relations between SES and treatment barriers (Kazdin, Holland, & Crowley, 1997; Kazdin, 2000; Stevens et al., 2006). This may be due to th e operational definitions of SES in the studies. The Hollingshead Four Factor Index of Social Status (Hollingshead; 1975) is frequently used to measure SES in the child development and treatment literature. One advantage of the Hollingshead Index (HI) is t hat it accounts for both the education and the occupation of caregivers. Research suggests that to understand the complex interactions between SES and parenting behavior (which could include decisions to continue treatment), the entire context in which the child resides must be considered (Callahan & Eyberg, 2010).
30 The socioeconomic indicators of education and income likely impact family behavior differently and each should be considered in research (Smith & Graham, 1995). Although the HI is limited in tha t it includes only the education and occupation of family income earners and not of unemployed parents, it provides a more complete measure SES (i.e., income or education) alone. Additionally, Hollingshead has been used as a measure of family SES in many of the past studies examining barriers to treatment continuation and was implemented in this study to facilitate comparison of results (Garcia & Weisz, 2002; Kazdin, Hollan d, & Crowley, 1997; Nock & Kazdin, 2001; Prinz & Miller, 1994). Summary Past research has identified barriers to treatment as a significant predictor of treatment completion and outcome. However, no study to date has investigated how therapy. Th is study aimed to investigate whether parent report of treatment barriers changes over the course of PCIT, through repeated measurement of barriers at each treatment session and application of multilevel statistical procedures. This study also investigated the effects of parent expectations for treatment as well as previously identified child, parent, and family predictors of change during PCIT. Child predictors include consideration of change in disruptive behavior as a time varying covariate with parent p erceived barriers. Parent predictors include change in expectations for treatment, and family predictors include SES. Before discussion of specific hypot heses, a brief review of Parent Child Interaction Therapy, the treatment model used in this study, is p resented.
31 Parent Child Interaction Therapy Parent Child Interaction Therapy (PCIT) is an evidence based treatment (EBT) for young children with disruptive behavior disorders (DBDs) (Eyberg, Nelson, & Boggs, 2008). PCIT has been used effectively to treat d isruptive behavior across a wide range of problem severity (Zisser & Eyberg, 2010). Therapists work with parents to increase positive attention and warmth with their child in the first phase of treatment, the Child Directed Interaction (CDI). During the se cond phase of treatment, the Parent Directed Interaction (PDI), parents learn specific discipline procedures to manage their child effectively and address inappropriate behavior. Both CDI and PDI begin with a teaching session during which therapists teach parents the basic CDI and PDI skills that will be implemented during therapy. Remaining sessions use bug in the ear technology for therapists to coach parents in vivo while the parent interacts with the child. ough PCIT in several ways. First, therapists observe and code parent child interactions at the start of each session to select the parent skills to target during the session and to determine when parents have met the criteria for moving from one phase of t reatment to the next and for completing treatment. Before each session, parents also fill out the Intensity Scale of the Eyberg disruptive behavior at home. Finally, in additi on to these criteria, treatment does not end own. Thus, PCIT is performance based rather than time limited, and the number of treatment sessions varies widely. The average length of treatment is 15 sessions, although completion in the range of 10 to 20 sessions is not uncommon (Zisser & Eyberg, 2010). The attrition rate in PCIT is about 35% (Fernandez & Eyberg, 2009;
32 Werba et al., 2006), which compares favorably to t he 40 to 60% commonly cited for child psychotherapy (Wierzbicki & Pekarik, 1993). Following PCIT, parents interact more positively with their child and report re port significantly less parenting stress, significantly less depressive symptomatology, and a more internal locus of control than waitlist parents (Eyberg, 2007; Schuhmann, Foote, Eyberg, Boggs, & Algina, 1998). Additionally, the significant changes in dis ruptive behavior and parenting stress are maintained one to three years following the conclusion of treatment (Boggs et al., 2004). Like other child therapy modalities, dropout in PCIT is problematic. Although predictors of treatment attrition in PCIT have been identified, no study has examined change in parent perception of treatment barriers in PCIT. Study Aims and Hypotheses Aim 1: The first aim of this study is to determine whether parent perceptions of barriers to treatment continuation change over the course of treatment. Hypothesis 1.1: Parent perceptions of barriers to treatment continuation will change over the course of treatment and will decrease over the course of treatment. Aim 2: The second aim of this study is to investigate predictors of ch ange in parent perceptions of barriers to continuing in treatment. Hypothesis 2.1: As child disruptive behavior decreases, parent perceptions of barriers to treatment continuation will decrease. Hypothesis 2.2: An increase in treatment expectations from the beginning to the mid point of treatment will be related to a decrease in parent perceived barriers to treatment continuation over the course of treatment.
33 Exploratory Analysis: The role of SES will be explored as a predictor of the initial level of per ceived barriers as well as a predictor of change in parent perceptions of barriers to treatment continuation. SES will be operationally defined by the Hollingshead Index (Hollingshead, 1975). Aim 3: The third aim of this study is to investigate the relat ionship between treatment completion status and parent perceptions of barriers to treatment continuation at the last treatment session attended. Hypothesis 3.1: Treatment dropout status (i.e., dropout, completer) will be predicted based on parent report o f perceived barriers at the last attended session. A higher level of parent perceived barriers will predict treatment dropout, when controlling for baseline demographic factors and the initial level of perceived barriers.
34 CHAPTER 2 METHOD Participants Forty six mother child dyads participated in this study. Participants were part of a larger, on going treatment study of y oung children with ADHD in individual and group PCIT. Participating families met the following inclusion and exclusion criteria: (a) the child was between 4 and 6 years of age at pretreatment; (b) the child met diagnostic criteria for Attention Deficit Hyp eractivity Disorder (ADHD) according to both parent and teacher report; (c) the primary maternal caregiver was willing to participate; (d) the child was not taking medication for ADHD symptomatology; (e) the child was enrolled in a structured daycare, pres chool, or school; (f) the child achieved a standard score of at least 70, and the mother achieved a standard score of at least 75 on cognitive screening measures; and (g) the child had no history of major sensory impairment (e.g., blindness), or pervasive developmental disorder. Families that did not meet study criteria were referred for alternative treatment. Fifty four mother child dyads were recruited, found eligible, and randomized into group and individual PCIT treatment for this study. However, eight families dropped out of the study prior to beginning treatment, resulting in the sample size of 46 mother child dyads. Screening Measures Child Behavior Checklist (CBCL; Achenbach & Rescorla, 2000). The CBCL has different versions for administration to ch ildren in the age ranges of 1.5 to 5 years and 6 to 18 years. Because children in this study were 4 to 6 years of age, both versions were used. A T score > 61 was required on the statistically derived Attention Problem Syndrome scale of both versions of th
35 the Attention Problem Scale are .68 and .86, respectively, and test retest reliabilities are r = .78 and r 5 year old version and .82 for the 6 18 year old version 1 Revised: Long Version (CTRS R: L; Conners, Sitarenios, Parker, & Epstein, 1998). This 59 item teacher rating scale assesses ADHD and common comorbid disorders in children ages 3 to 17 yea rs. A T score > 65 on either the DSM IV Hyperactive/Impulsive subscale or the DSM IV Total Problems subscale of the CTRS R:L was required for study inclusion. Internal reliability coefficients for the DSM IV Hyperactive Impulsive subscale range from .82 to .95 across male and female 3 to 8 year olds (Conners et al., 1998). In this study, Impulsive and DSM IV Total Problems subscales were .88 and .87, respectively. Diagnostic Interview Schedule for Young Children (YC DI SC; Strong, Lucas, & Lucas, 2006). The YC DISC, a computer assisted diagnostic interview for preschool aged children, is a downward extension of the DISC IV P (Shaffer, Fisher, Lucas, Dulcan, & Schwab Stone, 2000). The psychometric properties of the YC DIS C have not 1 It is important to note the particularl Syndrome Scale for the CBCL version compl eted by mothers of 4 and 5 year old participants (i.e., ient reported in the normative sample and the prevalent use of the CBCL as a measure of child inattention and hyperactivity in the child clinical literature (Achenbach & Rescorla, 2001). There are several factors that likely contributed to the low reliabil ity of the scale in this sample. First, all of the children in the sample had significant hyperactive symptomatology as reported by both parent and teacher. Thus, there was limited variability in responses to the Attention Problems Syndrome Scale. Addition ally, the CBCL scale is restricted to three responses (i.e., 0 = Not True (as far as you know); 1 = Somewhat or Sometimes True; 2 = Very True or Often True). Limiting the scale to three points further restricts the range of responses on this scale. Given t hat alpha values are based on correlations among scale items, limited variability will still quite low, t he magnitude of change indicates that these items were particularly problematic for this sample.
36 yet been established, but its high similarity to the DISC IV P suggests that these properties may be similar. One week test retest reliabilities of the DISC IV P with parents of 9 to 17 year old children for the 5 diagnostic categories assessed in this study have been reported at .79 for ADHD, .54 for ODD, .54 for CD, .58 for SAD, and .66 for MDD (Shaffer, Fisher, & Lucas, 1998). The newer YC DISC was used in this study because symptom items are worded more appropriately for young children than its parent measure. Children were required to meet diagnostic criteria for ADHD Hyperactive type or ADHD Combined type for inclusion in this study. Peabody Picture Vocabulary Test Third Edition (PPVT III; Dunn & Dunn, 1997). The PPVT III is a well standardized measure of receptive language in individuals 2.6 years of age and older. Split half reliability coefficients for children have ranged from .86 to .97, with a median of .94, and test retest reliabilities have ranged from .91 to .94. The correlation between the PPVT III and the WISC III Full Scale IQ is r = .90. A PPVT III standard score at or above 70 was required for inclusion of children in this study. Wonderlic Personnel Test (WPT; Dodrill, 1998). The WPT is a 50 item screening scale of adults' intellectual abilities. In a sample of 120 normal adults, the WPT score was highly correlated (r = .93) with the WAIS Full Scale IQ score and was within 10 points of the WAIS IQ score for 90% of the sample. A WPT standard score of 75 or higher was required for inclusion of parents in this study. Measures of Outcome Predictors The Credibility/ Expectancies Questionnaire Parent Version (CEQ P; Nock, Ferriter, & Holmberg, 2007). The CEQ P is a 6 about treatment credibility and effectiveness. The CEQ P was administered at the start
37 of treatment and again at the mid point of treatment in this study. The 6 items are grouped into two components: (a) Treatment Credibility; and (b) Treatment Expectanci es. Four questions ask respondents to indicate the extent to which the treatment makes sense to them or to which they believe their child will show improvement along a 9 point scale ranging from (1) not a lot of sense/ no improvement to (9) a lot of sense/ very much improved. Two questions ask respondents to indicate how much improvement they think and feel will occur in treatment on an 11 point scale ranging from 0% to 100%. Internal consistency coefficients of .82 for the Credibility component, .88 for th e Expectancies component, and .87 for the total scale have been reported (Nock et al., 2007). Initial CEQ P scores and the change in CEQ P scores from the beginning to mid point of treatment were both used as predictors of change in this study. The change in P score at t he start of treatment from their CEQ P score just after the second phase of PCIT was introduced (i.e., PDI Teach session). Higher, positive values for the change score signify greater increase in expectations from beginning to mid point of treatment. Lower negative values for this predictor signify greater decrease in expectations from beginning to mid P at the beginning and mid point of treatment were .89 and .91, respectively. Demographi c and Background Questionnaire. This parent questionnaire provided descriptive information about the child and family including sex, age, race/ ethnicity, occupation, education level, income and medical history. Information on this questionnaire provided t he baseline predictors of treatment completion status and
38 variables for calculation of the Hollingshead Four Factor Index of Social Status (HI; Hollingshead, 1975). Eyberg Child Behavior Inventory (ECBI; Eyberg & Pincus, 1999). The ECBI is a 36 item parent report measure of disruptive behavior in children ages 2 to 16 years. The ECBI includes two scales: The Intensity Scale measures the frequency of specific behaviors on a 7 point scale from (1) never to (7) always and the Problem Scale measures the extent dichotomous yes no scale. Only the Intensity Scale was used in this study. A 12 week test retest reliability of .80 and an internal consistency coefficient of .95 were reported within a com munity sample for the Intensity Scale (Funderburk, Eyberg, Rich, & Behar, 2003). The ECBI was administered at the beginning of each session over the course of treatment. The ECBI was also used as one criterion for treatment completion. Before graduating f Intensity Scale at less than 114, which is one half standard deviation above the normative mean (Eyberg & Pincus, 1999). Intensity Scale raw scores collected at each session were used to investigate the role of disruptive behavior as a time varying covariate with parent perceptions of barriers to treatment completion. A mean intensity score was calculated for each dyad, and a centered variable, which indicates the weekly deviation from th for the Eyberg Child Behavior Inventory Intensity Scale at the first occasion of treatment was .94. Alpha values across occasions of treatment ranged from .89 to .97, with a
39 mean alpha coe fficient of .95. Table 2 1 includes the alpha coefficients for each week of treatment. Hollingshead Four Factor Index of Social Status (HI; Hollingshead, 1975). The HI measures family socioeconomic status, based on the occupation and education level of the employed parents living at home. The HI does not include the education and occupation of students, homemakers, and unemployed individuals in the home. If no household member is currently employed, the HI is calculated for the individual most likely to be the head of the household, including the occupation in which the individual is typically employed. Education is rated on a 7 point scale and occupations are rated on a 9 point scale. The 9 point occupation scale is a categorization of approximately 450 titles from the 1970 United States Census. Occupation and education scores are first weighted. The education score is multiplied by 3, and the occupation score is multiplied by 5. These scores are then summed. For dual income families, scores are averaged to calculate the HI. HI scores can range from 8 to 66. In this study, a person centered SES mean. Outcome Measure Barriers to Treatment Participation Scale Revised ( BT PS R; Colonna Pydyn, Gjesfeld, & Greeno, 2007). The BTPS R is a 20 item measure of barriers to treatment continuation. In this study, parents completed this measure at each treatment session. The 20 items are grouped into two factor analytically derived su bscales: (a) Treatment Expectations; and (b) External Demands. Respondents indicate the extent to which each item relates to their experience in treatment, along a 5 point scale ranging from (1) Never a problem to (5) Very often a problem Internal consist ency coefficients were .90
40 for the 10 item Treatment Expectations composite and .80 for the 10 item External Demands composite (Colonna Pydyn et al., 2007). The BTPS R is based on the Barriers to Treatment Participation Scale (Kazdin, Holland, Crowley, & Breton, 1997). Colonna Pydyn et al. (2007) conducted an exploratory factor analysis of the original BTPS, and found that the items loaded onto the two factors described previously. With the goal of developing a shorter measure of treatment barriers for ad ministration in community settings, the 20 item BTPS R was constructed using the ten highest loading items on each factor. A confirmatory factor analysis of the revised measure supported the 2 factor structure of the new scale. For this study, the BTPS R was chosen for its strong internal consistency and more feasible length for weekly administration. In addition, the BTPS R items are better worded for a weekly administration than its parent measure and better reflect barriers that could change week to we ek. With permission from the authors of the BTPS R, items were modified from past tense to present tense to reflect the weekly perception of barriers more clearly (C. Greeno, personal communication, July 13, 2009 ). Scale anchors were also modified slightly (i.e., Never a problem was changed to Not at all a problem ; and Very often a problem was changed to Very much a problem ). In this R at the first occasion of treatment was .84. Alpha values across occasions of t reatment ranged from .70 to .94, with a mean alpha coefficient of .86. Table 2 1 includes the alpha coefficients for each week of treatment. Procedures Families were referred to the Child Study Laboratory at the University of Florida. Primary referral s ources were general pediatric practices, child psychiatry practices, community mental health practitioners, and school personnel. Before their first
41 evaluation appointment, each family received a welcome packet including directions to the Health Science Ce nter, a demographic questionnaire, the CBCL, and the CTRS. At the first of two pre treatment assessment visits, after completing the Institutional Review Board (IRB) consent procedure, the completed CBCL and CTRS questionnaires were collected, and families were administered the remaining measures used to determine eligibility. Families were also observed in the first of two behavioral observation assessments. If a family failed to meet any of the inclusion criteria, options for treatment outside of the stud y were discussed. At the second pre treatment evaluation, multiple measures of child, parent, and family functioning were collected. Families were also observed in the second of the two pre treatment behavioral observation assessments. At the completion o f treatment, families again attended two post treatment evaluation visits that were similar in content and structure to the pre treatment assessments. Assessments were conducted by graduate students who had been extensively trained in the administration of all assessment procedures and who were blinded to group versus individual treatment assignment. After families completed the pre treatment evaluation, they were randomly assigned to one of two treatment formats, individual PCIT or group PCIT. In both cond itions, parents learned the same set of parenting skills, and in both conditions parent child dyads had individual coaching time with the therapists. The main difference between the group and individual PCIT condition was that three to four families partic ipate together in group PCIT. Additionally, the individual treatment sessions lasted approximately 75 minutes each week, whereas the group treatment sessions lasted
42 approximately 120 minutes each week. Because six to eight families must complete the pre tr eatment assessment before randomization to treatment condition can take place, families waited 2 to 6 weeks following their assessment before treatment began. For this study, mothers in both treatment conditions filled out the brief BTPS R during the last 5 minutes of each session. The procedure for completing and procedure for completion of the BTPS R is also shown schematically in Figure 2 1. To hile filling out the BTPS R, which includes several items therapist stepped out of the room while mothers completed this measure. Mothers then placed the completed measure in a letter sized envelope, sealed the envelope, and were given the option to sign the envelope across the seal. Mothers placed the letter sized envelope within a larger goldenrod envelope designated specifically for that family. The larger envelop previously completed BTPS R forms from past sessions. This procedure allowed the mothers to see that their envelopes from previous sessions had not been opened. Between sessions, the goldenrod envel folder in a locked storage room. After families completed treatment, the envelopes were opened by research assistants who prepared scan sheets for data entry. According to standard treatment protocol, the EC BI was also administered during each week of treatment, during the first 5 minutes of the session. On this instrument, the Families in this study also completed the CEQ P measure of treatment expectancies
43 and credibility on two occasions during treatment, once at the conclusion of the CDI teaching session and once at the conclusion of the PDI teaching session. These two sessions were chosen for administration of the CEQ P b ecause they fall at the beginning and approximate mid point of treatment, at a time when families in both treatment conditions (group or individual PCIT) have been presented the same privacy, the therapist followed the same procedure as with completion of the BTPS R and stepped out of the room while the mother filled out the CEQ P. The mother then placed the completed CEQ P in a letter sized envelope and placed the sealed envelope in the larger goldenrod envelope used to also collect their BTPS R forms. Both individual and group PCIT treatment sessions were conducted by graduate student therapists in clinical psychology, who were experienced PCIT therapists. PCIT lead therapists had taken a graduate level course in PCIT theory, research, and practice. PCIT lead therapists also had previous experience as co therapists on at least two PCIT cases before becoming a lead therapist. Therapists attended weekly group supervision sessions with the treatment developer. Weekly therapy sessions were also recorded for treatment integrity checking by undergraduate research assistants. The treatment manual includes a checklist for each session to code treatment integrity. A randomly selected 50% of s essions from each family were coded for integrity. Treatment integrity was calculated as percentage agreement with the session checklists.
44 Analyses The Statistical Package for the Social Sciences 18.0 (SPSS) was used for data analysis. Explanations of pre liminary statistical analyses and analyses specific to each hypothesis are detailed below. Normality Data were first screened to ensure normality and the absence of multicollinearity. Descriptive statistics and visual representations of the data, includ ing histograms and box plots were examined for normality. To further investigate whether the distribution deviated from normality, values of kurtosis and skewness as well as Kolmogorov Smirnov and Shapiro Wilk tests were used (Field, 2009). Variables with significant skewness and kurtosis were transformed using the Blom transformation (Blom, 1958), to correct for positive skewness and kurtosis. Blom transformation involves the calculation of an equivalent data point in z score metric with rank order maintai ned. The equivalence scores are fitted along the normal curve, thus improving the skewness and kurtosis of the data. The Blom transformation was implemented after other transformation techniques, such as the log linear and square root transformations did n ot significantly improve the normality of the data. Multicollinearity Relationships among predictor variables were investigated with Pearson bivariate correlations to ensure the absence of multicollinearity. Multicollinearity is a term that describes the existence of linear or near linear relationships among variables in a model (Pedhazur, 1997). When multicollinearity is present in a hierarchical linear model, calculation of model estimates may be compromised. When multicollinearity is present in small am ounts, concern about its impact is minimal. However, as the presence of
45 multicollinearity increases, concern regarding the negative impact of multicollinearity on estimation of effects in the model increases as well (Shieh & Fouladi, 2003). Reliability T ests of reliability included measurement of internal consistency through calculation of the alpha coefficient as well as calculation of test retest reliability between the first occasion of treatment and the approximate mid point of treatment (i.e., PDI Te ach) for each of the repeated measures (i.e., BTPS R, CEQ P, ECBI). The test retest statistic was implemented to show how reliable the measurements of parent perceived barriers to treatment continuation, child disruptive behavior, and parent expectancies w ere between two points of time in treatment. High test retest reliability indicates stability of the measure across time, but also could be an artifact of the rank order of responses or the responses remaining constant between the two time points. Low test retest reliability would indicate poor stability of the measure across time. However, change in the construct being measured could result in low test retest reliability as well. Descriptive Statistics Descriptive statistics were calculated to describe key child, mother, and family characteristics and to compare participants in group versus individual conditions. T tests and chi square analyses were used to compare the baseline child, parent, and family characteristics of participants in the group and individual treatment conditions to determine whether treatment groups could be collapsed for subsequent analyses. No significant differences were expected in baseline characteristics of participants in the t wo conditions, given the random assignment to treatment conditions.
46 Analysis of Change Regression plots for change in parent perceived barriers were examined for all participants to investigate the potential patterns of change over time and to determine the most suitable model specification for analysis of the change over time. This refers to whether the pattern of change is linear, quadratic, cubic, or non linear and influences which equation is used to analyze the pattern of change. Investigation of th e regression plots offers preliminary information about whether the pattern of change is linear or whether the model should investigate higher order (i.e., quadratic, cubic, or other non linear) patterns of change. To model session by session change in pe rceived barriers to treatment continuation over the course of treatment, hierarchical linear modeling (HLM; Bryk & Raudenbush, 1992) was used. Specifically, a longitudinal mixed effects model was used to examine the session by session data. Fixed and ran dom effects were investigated with HLM. Fixed effects refer to the constant effect of a predictor across all individuals in the sample. Random effects refer to individual differences in the measured fixed effect. In this study, fixed and random effects wil l be measured at two levels. Level 2 fixed effects refer to between individual differences in criterion variables, across all participants. Level 1 fixed effects measure variations in the criterion variable across all participants and occasions of measurem ent. A level 1 random effect indicates whether a within person relationship differs significantly across individuals. Before discussing the occasion specific (level 1) and person specific (level 2) mixed models that were implemented to investigate change in parent perceived barriers, methodological challenges specific to this study, in particular the unbalanced study design, are addressed.
47 Consideration of Unbalanced Designs A study design is balanced when (a) all participants have an equal number of me asurements with no missing data, and (b) occasions of measurement are all equally spaced apart. A study design is unbalanced when participants vary in the number of measurements or the time between occasions of measurement. Unbalanced designs result from m issing data, because some dyads have fewer data points available than other dyads. The design of this study was unbalanced, because the number of occasions varied between participants. Fortunately, multilevel models are robust enough to manage unbalanced d esigns. Variance in the number of measurements per participant in this study was due to the time unlimited and two phase structure of PCIT. PCIT has two phases (CDI and PDI), and in both phases families vary in the number of sessions needed to master th e phase specific skills and move on to the next phase or complete treatment. Each session for each participant is considered an occasion. The structure of treatment resulted in an unequal number of occasions per participant. Unequal spacing of time betwee n treatment session occasions also leads to an unbalanced design. Although it is intended that families attend PCIT sessions weekly, on a set day of the week, there are cancellations, no shows, and rescheduling of appointments that result in unequal length s of time between occasions for some participants. However, after consideration of statistical power in this study, time was investigated as a fixed factor. 2 Thus, time between sessions was considered as having 2 The language similarities here may introduce some confusion. In this study, we distinguish between fixed/random effects versus fixed/random factors. As expl ained elsewhere, fixed/random effects, in HLM refer to effects that are averaged/true for the entire sample (i.e., fixed) versus those that differ between individuals (i.e., random). A given Level 1 variable can have both a fixed effect (i.e., the average effect of
48 equal meaning for participants and was looked at in terms of the session (i.e., occasion of treatment) at which the measure was completed (i.e., occasion 1, occasion 2, random factor, which would consider occasion of mea surement as the number of days since the beginning of treatment. Implications of Treatment Dropout on Subsequent Analyses The number of occasions of measurement varied in this study not only because of the performance based structure of treatment, but als o because some families dropped out of treatment before completion. One assumption of multilevel models is that when data are missing, there is no systematic cause for the missing data, and the data may be considered Missing at Random (MAR; Rubin, 1976). T he MAR assumption has important implications for this study, given that some of the families dropped out of treatment. If no systematic cause is found for attrition (i.e., there is no significant association between a variable and likelihood of dropping ou t), then the HLM may be conducted without adjustment, and data from treatment completers may be considered adequately representative of the data from treatment non completers (i.e., dropouts). However, if attrition is systematically related to baseline var iables, that would suggest that predictor) and a random effect (i.e., the significant individual differences in the effect of that predictor). Fixed and random factors refer to the nature of the distribution of the predictor. A fixed factor has predefined levels (e.g., di fferent dosages of a drug) that are investigator selected and to which people are assigned. A random factor is one in which the population of levels is defined outside the investigator (e.g., levels of drug that are taken by community dwelling residence), and from which the investigator makes random selections. In this study, the random number of days since randomization. However, to evaluate this reliably, there would have had to b e multiple individuals with the same days since randomization. In this study, it is possible that some days since randomization would be expressed by only a single individual. Thus, the fixed factor approach selected and assigned by the investigator) was more feasible, since a ll (i.e., non attrited) participants would have experienced every level of a fixed factor session.
49 there are inherent differences between treatment completers and non completers, and the data are considered Missing Not at Random (MNAR; Rubin, 1976). When data are considered MNAR, a pattern mixture model approach may be impleme nted for HLM analyses (Little, 1993). Pattern mixture models classify missing participants, which are then averaged across each group. Variables based on the missingness group s are used as covariates in the model, allowing the researcher to investigate the impact of the missing data pattern on the outcome. Pattern mixture models also assess the degree to which missingness interacts with model terms (Hedeker & Gibbons, 1997; Sch afer & Graham, 2002). In this study, participants were classified into two groups based on their missingness status treatment completers and treatment dropouts. Completion status was included in the model as a dichotomous variable to assess change in th e outcome of interest, parent perceptions of barriers to treatment. Because the goal of including completion status in the model was to predict differences in change in barriers over time between participants who dropped out and completed treatment, a part ial pattern mixture approach was utilized in which interactions between completion status and model covariates were not investigated. This approach was further justified by the lack of individual differences in the interaction between completion status and the rate of change (M. Marsiske, personal communication, April 13, 2011). Aims 1 and 2 Implementing the Hierarchical Linear Model As discussed earlier, the HLM includes two levels the level 1 model (also referred to as the within participants or occa sion specific model) and the level 2 model (also referred to as the between participants or person specific model). Change in the
50 level 1 model of parent perceptions of barriers to treatment continuation was first evaluated without predictors. Child disrup tive behavior at each occasion was investigated as a level 1 predictor, and parent expectancies, SES, and average level of child disruptive behavior were examined as level 2 predictors of change. Each model investigated whether the addition of predictors i mproved the fit of the model and the variance explained. Improvements in fit were determined through the reduction of between and within person residual variance, as compared to the baseline model (Bryk & Raudenbush, 1992). Fit was also assessed in this study by looking at the change in the 2 log likelihood ( 2LL, AIC, and BIC communicate goodness of fit of the model t o the data. The 2LL is a variation of the log likelihood (i.e., log likelihood multiplied by negative two) and is a measure of error or unexplained variance in the model. Both the AIC and BIC take into account the number of parameters in the model. Althou gh the AIC and BIC are not inherently interpretable, their values in each model can be compared to determine how adding parameters improves or hinders model fit. Smaller values of 2LL, AIC, and BIC, represent improved model fit. Intraclass correlation coe fficients (ICC) were also reported as measures of variability at the between participants and within participant levels. Aim 1: Measuring Change in Parent Perceived Barriers Across Time A null model (i.e., the unconditional means model ), which contains no predictors, was first tested to define the variance to be explained and was used as a comparison for subsequent models to calculate (a) improvement in fit, and (b) variance explained. Next, linear time was entered into the model as a fixed factor (i.e., O ccasion 1 [CDI
51 Teach], Occasion 2 [CDI Coach 1], Occasion 3 [CDI Coach 2], etc). This model which is referred to as the unconditional growth model, determined the fixed effect of time (i.e., the average effect, averaged across all participants) and the ra ndom effect of time (i.e., measure of size of individual differences in the strength of the effect). Improvement in fit was assessed by comparing the fit statistics (i.e., 2LL, AIC, BIC) to those of the null model. Based on plots that suggested a curvilin ear effect of time, the fixed and random effect of quadratic time was added to determine if a non linear pattern of change existed (i.e., an acceleration or deceleration in change of barriers to treatment continuation). A positive quadratic effect is indic Aim 2: Modeling Predictors of Parent Perceived Barriers Hypothesis 2.1 addressed the role of child disruptive behavior as a time varying covariate in the analysis of change in parent perceived barriers to treatment continuation. A time varying covariate is a predictor variable that is also measured on multiple occasions. In this study, level of child disruptive behavior is a ti me varying covariate. The association of child disruptive behavior and parent perceived barriers was investigated at the between individual and within mean score on the ECBI Intensity Scale across treatment was included in t he model to determine between the ior across treatment. This difference score was also included in the model to test the over the course of time during treatment.
52 Hypothesis 2.2 addressed the role of ch anges in treatment credibility and expectancies in predicting change in parent perceived barriers to treatment. Parent report of treatment credibility and expectations was measured at the end of each teach session in the first and second phases of treatmen t, CDI Teach and PDI Teach. The teach sessions fall at the beginning and approximately the mid point of treatment. A difference score was calculated to indicate change in parent expectations for treatment and was also included in the model as a level 2 pre dictor. The initial expectations score was also included in the model as a level 2 predictor. Interactions between time and initial expectations and time and change in expectations were investigated to determine whether treatment credibility and expectatio ns were significantly related to change in parent perceived barriers. An exploratory analysis was conducted to investigate the role of family SES as a predictor of change in parent perceived barriers. First, the fixed effect of SES was investigated as a level 2 predictor, to determine how SES is related to change in perceived barriers and to test whether the association of SES and change in perceived barriers differs between participants. The interaction between SES and the time function was also investig ated to explore whether SES predicts rate of change in barriers. SES was also investigated as a predictor of initial level of parent perceived barriers A linear regression was used to explore whether SES was a significant predictor of barriers to treatmen t continuation at the first occasion of measurement. Aim 3: Barriers at the Last Completed Session as a Predictor of Dropout The third aim of this study was to investigate the relationship between treatment completion status and parent perceptions of barr iers to treatment continuation at the last treatment session attended. It was hypothesized parent perceived barriers at the last
53 attended session would predict dropout from treatment. A hierarchical logistic regression was implemented with treatment comple tion status (i.e., present at post test or not present at post test) as the dichotomous dependent variable. Prior to conducting the hierarchical logistic regression, a forward step wise logistic regression was run to determine demographic and psychosocial variables that significantly predicted attrition in this sample. The variables found to be significant predictors of attrition in the stepwise logistic regression were then controlled for in the hierarchical logistic regression, by entering these variables in the first block along with a baseline measure of parent perceived barriers to treatment continuation. Although it was proposed that variables that have been cited in the child clinical literature as predictors of attrition, namely ethnicity, SES, and s ingle parent status (Armbruster & Schwab Stone, 1994; Fernandez & Eyberg, 2009; Gould et al., 1985; Kazdin et al., 1993; McCabe, 2002; Wierzbicki & Pekarik, 1993) would be included in the first block, these variables were not included in the first block, b ecause they were not significant predictors of attrition in this sample. The variable parent perceived barriers, measured at the last treatment session attended, was entered in the second block to determine whether perceived barriers at the last session at tended significantly predicted whether the family completed treatment.
54 Table 2 1. Cronbach alpha values across occasions of treatment for the Eyberg Child Behavior Inventory (ECBI) and the Barriers to Treatment Participation Scale Revised (BTPS R) ECBI BTP S R Occasion N Alpha N Alpha 1 45 .94 43 .84 2 42 .96 41 .90 3 39 .96 39 .88 4 37 .95 35 .86 5 37 .95 37 .86 6 35 .96 34 .81 7 37 .96 37 .87 8 36 .97 36 .81 9 36 .96 36 .83 10 35 .96 34 .80 11 34 .97 34 .86 12 31 .94 30 .85 13 31 .93 30 .84 14 29 .95 29 .88 15 23 .96 23 .91 16 15 .96 14 .86 17 13 .96 13 .94 18 9 .96 9 .82 19 8 .90 8 .87 20 7 .93 7 .83 21 3 .89 3 .70 22 1 1 Note. Sample size varied for ECBI and BTPS R at several occasions, due to missing data.
55 Therapist fills in session identification data (i.e., date and session number) and family ID number on BTPS R and attaches letter sized envelope Before session Therapy session Figure 2 1. Procedure for completion of BTPS R Therapist completes session Therapist hands mother BTPS R form with attached letter sized envelope and goldenrod envelope containing envelopes from past weeks Therapist steps out of the room for approximately 5 minutes Mother completes BTPS R form Therapist returns to the therapy room envelope Mother places sealed letter sized envelope in goldenrod envelope Mother places completed BTPS R form in letter sized envelope and seals envelope
56 CHAPTER 3 RESULTS Preliminary Analyses Normality and Multicollinearity Among the predictor and outcome variables examined in this study, weekly measures of child disruptive behavior (i.e., ECBI intensity scor es) were normally distributed for each treatment session. Initial CEQ P score and Hollingshead SES were normally distributed as well. The predictor variable of change in parent expectations and the outcome variable of parent perceived barriers to treatment participation were transformed using the Blom transformation (Blom, 1958), to correct for positive skewness and kurtosis. Prior to transformation, the skewness of change in expectations was 1.51, and kurtosis was 6.52. Following transformation, the skewn ess of change in expectations was 0.01 and the kurtosis was 0.27. Blom transformation was also used for transformation of the BTPS R outcome measure. Prior to transformation, the average skewness of treatment barriers was 2.37, and the average kurtosis w as 6.23. Following transformation, the average skewness was 1.10 and the average kurtosis was 0.64. The skewness and kurtosis for the BTPS R at each occasion of treatment is shown in Table 3 1. leptokurtic, non normal data structure that necessitated transformation. Although possible scores on the BTPS R range from 20 to 100, scores in this sample fell within a more restricted range. Average raw scores on the barriers measure ranged from 21.80 (S D = 3.23) on week 13 of treatment to 26.00 (SD = 8.96) in week 19. Across participants and weeks of treatment, the mean raw score for BTPS R was 23.18 (SD = 5.56).
57 Multicollinearity was assessed by determining Pearson bivariate correlations. Table 3 2 shows the bivariate correlations of the predictor variables and interaction variables across occasions. In terms of absolute values, significant bivariate correlations between predictors ranged from .097 to .489 and fall in the small to large effect size r ange (Cohen, 1988). Correlations in the range of .80 and higher are indicative of substantial multicollinearity (Field, 2009). Thus, although relations exist between predictors in the small to large effect size range, multicollinearity was not overly conce rning in this sample. Reliability Reliability statistics were determined for the BTPS R, CEQ P, and ECBI to show how consistent the measurements of parent perceived barriers to treatment continuation, child disruptive behavior, and parent expectancies are within each respondent and over time. Tests of reliability included a measurement of internal retest reliability. Test retest reliability was estimated through Pearson bivariate correla tions between scores at CDI Teach and PDI Teach for each of the repeated measures (i.e., ECBI, BTPS R, CEQ P). The Blom normalized score for BTPS R at CDI Teach and PDI Teach was used in this analysis. Test retest reliability for all three measures were la rge and significant (ECBI, r = .66, p < .001; Blom normalized BTPS R, r = .66, p < .001; CEQ P, r = .64, p < .001). Descriptive Analyses Descriptive statistics were calculated for child and mother demographic variables as well as predictor and outcome variable measures. Means and standard deviations for
58 demographic and descriptive variables are presented in Table 3 3, and frequency counts f or demographic variables are presented in Table 3 4. Participant children were 65% ( N = 30) male, with a mean age of 4.78 years ( SD = .81). Their parent reported ethnic/ racial background was 61% ( N = 28) Caucasian, 22% ( N = 10) Black or African American 2% ( N = 1) Hispanic, and 15% ( N = 7) bi racial. All children met diagnostic criteria for ADHD Hyperactive Impulsive or Combined Type, which was one of the inclusion criteria for this study. Based on the YC DISC, 78% ( N = 36) met diagnostic criteria for O DD, 24% ( N = 11) for Conduct Disorder (CD), and 15% ( N = 7) for Separation Anxiety Disorder (SAD) in addition to ADHD. No children met diagnostic criteria for Major Depressive Disorder (MDD) on the YC DISC. Seventeen percent ( N = 8) of the children receive d a diagnosis of ADHD without a comorbid diagnosis. Children achieved a mean standard score of 104.11( SD = 13.24) on the Peabody Picture Vocabulary Test (PPVT III). Participant mothers had a mean age of 35.16 ( SD = 7.02), and their self identified ethnic/ racial background was 67% ( N = 31) Caucasian, 22% ( N = 10) Black or African American, 7% ( N = 3) Hispanic, and 4% ( N = 2) Biracial. Sixty seven percent ( N = 31) of mothers were married, 4% ( N = 2) were separated, 13% ( N = 6) were divorced, and 13% ( N = 6) were never married. Marital status was missing for one mother. For highest level of educational attainment, 13% ( N = 6) of mothers had a graduate or professional degree, 28% ( N = 13) had completed college, 41% ( N = 19) had completed some college, 15% ( N = 7) had received a high school diploma, and 2% ( N = 1) did not complete high school. Mothers achieved a mean standard score of 106.78 ( SD = 13.00) on the Wonderlic Personnel Test.
59 At trition from this study was 2 6 %. Thirty four families completed treatment, and 12 families dropped out of treatment. Treatment completers attended an average of 15.65 sessions ( SD = 3.11; CDI sessions, M = 6.03, SD = 1.38; PDI sessions, M = 9.62, SD = 2.46). Families who dropped out o f treatment attended an average of 4.5 sessions ( SD = 4.15), with nine families discontinuing treatment in the CDI Phase, and three families discontinuing treatment in the PDI phase. No significant differences were found between treatment assignment (i.e. group versus individual) on child age, t (44) = .437, p > .05, mother age, t (43) = .066, p > .05, mother IQ, t (43) = 1.66, p > .05, SES, t (44) = .837, p > .05, or completion status (i.e., completer versus drop 2 (1) = 3.39, p > .05). Difference between treatment assignment in initial score on the BTPS R was also non significant, t (44) = 1.58, p > .05. Differences between individual and group treatment could not be measured on variables of maternal education, marital status, and ethnicity/ race because chi square assumptions were not met (i.e., expected frequency for some categories were less than 5). Means and standard deviations for demographic and descriptive variables are presented by treatment assignment in Table 3 5, and fr equency counts for demographic variables by treatment assignment are presented in Table 3 6. Multilevel Modeling of Change in Perceived Barriers to Treatment Continuation Multilevel modeling was implemented to investigate predictors of change in parent per ceived barriers to treatment continuation. Twenty seven models were tested, labeled 1 through 27. Each model was characterized by the successive inclusion of fixed and random effects of the hypothesized predictors of barriers to treatment continuation. Onc e all 27 models were conducted, repeated and random effect structures were assessed to attain the most parsimonious and best fitting model. Attempts to allow
60 correlated errors over time did not converge or improve fit, and the simplified GLM style assumpti ons (i.e., Scaled Identity repeated effects structure, variance components random effects structure) fit well on the repeated and random parts of the model. The 27 models were as follows: Model 1 tested the unconditional means model, which contained no pr edictors. The unconditional means model or null model, was the baseline for comparison of subsequent models and established the variance to be explained. Models 2 through 5 examined the fixed (i.e., general effect, averaged across participants) and random (i.e., estimation of degree of individual differences in the strength of the effect) effects of Linear Time (Occasion, coded 0 22) and Quadratic Time to find the optimal time basis for the model. Model 6 addressed the fixed effect of attrition on level o f perceived barriers over time. Models 7 through 9 addressed the fixed and random effects of attrition on change in perceived barriers over time. Attrition was added to the model to account for differences in level of perceived barriers and rate of change in perceived barriers between treatment completers and treatment drop outs. In models 10 through 27, fixed and random effects of predictor variables of interest (i.e., Average and Centered ECBI, Initial CEQ P, Change in CEQ P, SES, and interactions between predictors and time) were examined. Means and standard deviations for the Level 2 (i.e., between subjects) variables are summarized in Table 3 7. Tables 3 8 through 3 14 summarize the results of the model tests for barriers to treatment continuation. Tabl e 3 15 contains the fit statistics (i.e., 2LL, AIC, BIC) for each model. Table 3 16 summarizes the between and within person variance in each model as well as variance explained by each model. Terms that provide important information about variance in th e models include within 2 ), between person
61 0 2 1 2 ), and variance in 2 2 ). These variance terms indicate whether in each model there is 2 ) or whether 0 2 ), due to individual differences. Additionally, the 1 2 2 2 respectively) convey whether there are individual differences in linear and quadratic rates of change. Table 3 16 also lists the pseudo R 2 terms, which represent estimates of the amount of within and between person variance for which each model accounts. A primary goal of adding predictor s to the HLM is to account for residual variation (i.e., the variance in the model that is unexplained by the predictors in the model). As predictors are added to the model, the reduction in within 2 ) and between 0 2 ) is quantified by the calculation of a pseudo R 2 term for both within person (pseudo R 2 ) and between person (pseudo R 0 2 ) terms (Singer & Willett, 2003). It is important to note that pseudo R 2 terms are flawed in that while R 2 statistics by definition are never negative, pseudo R 2 calculations occasionally produce negative results in situations in which the addition of a predictor reduces residual variance at one level (e.g., Level 1) but then increases residual variance at the other level (e.g., Level 2) (Singer & Willett, 2003; Snijders & Bosker, 1999). Pseudo R 2 terms may be especially flawed when random effects are included in the model (Snijders & Bosker, 1999). Thus, while comparison of pseudo R 2 terms are helpful and informative, change in model f it should be given greater focus when comparing models. Following Blom normalization of the data, values of intercept and rate of change remain in z score metric. As a result, the numerical estimates hold little interpretable
62 meaning. Thus, description of the results focuses on direction and significance of effects as well as change in the variance to be explained as predictors are added to the model in successive steps. The addition of fixed and random effects is summarized as separate steps within the te xt and in separate columns, labeled Models 1 through 27 in Tables 3 8 to 3 14. Before conducting the HLM, the occasion level (i.e., Level 1) data for BTPS R was p lotted as a spaghetti plot ( Figure 3 1) to visually examine session by session change in BTPS R for each family. This type of plot is useful to investigate patterns of change over time to determine the most suitable model specification for analysis of the change in BTPS R over time. Figure 3 1 shows that most families show a linear trend of de crease in BTPS R over time. It is also evident from Figure 3 1 that there are individual differences in this pattern of change. Pattern of change was further examined through a curve estimation model (Figure 3 2), which was suggestive of both linear and qu adratic patterns of change. Although a cubic trend also appears in the plot, the cubic trend appears more trivial than the linear and quadratic trajectories of change. Unconditional Means Model Model 1 represents the Unconditional Means Model, or null mo del. This model assumes no change over time and contains no predictors. As shown in Table 3 8, the within 2 = .250, p < .001, suggesting that there was statistically significant variability in parent report of treatment b arriers over time. The between 0 2 = .427, p < .001, indicated statistically significant differences between mothers in their overall perception of barriers to treatment continuation. Intraclass correlation coefficients (ICC) convey the pe rcentage of variance that is between person (i.e., Level 2) and the percentage of variance that is within
63 person (Bryk and Raudenbush, 1992). In this sample 37% of the variability was within person, and 63% of the variability was between person. Given the significance of the variance components, the addition of Level 1 and Level 2 predictors was warranted to explain differences in barriers to treatment participation between participants and across time. Change in Barriers over Time Linear time In Model 2, shown in Table 3 8, the fixed effect of the Level 1 predictor, Linear Time, was added as a predictor of the dependent variable, BTPS R. No level 2 predictors were added. The main effect was negative and significant, meaning that on average, perceptions of barriers to treatment continuation decreased over time. Calculation of pseudo R 2 showed that addition of the fixed effect explained approximately 15% of within person variance (Table 3 16). No between person variance was explained. Addition of the fix 2 (1) = 85.91, p < .001. The addition of the random effect of linear time, shown in Model 3 of Table 3 8, 2 (1) = 123.55, p < .001. The variance in linear tim 1 2 = .003, p < .001) conveyed that there were individual differences between mothers in the rate of change across time and significant random variation in linear change to be explained. The addition of the random effect of linear time accounted for app roximately 28% of within person variance (Table 3 16). No between person variance was explained. Figure 3 3 illustrates the estimated change in BTPS R when linear time was modeled. This plot illustrates both the linear decrease in barriers over time, as we ll as individual differences in the linear rate of change.
64 Quadratic time Based on residual plots revealing a quadratic trend to perceived barriers, the fixed and random effects of quadratic time were added to the model. Model 4 in Table 3 8 summarizes th e fixed effect of quadratic time on the outcome variable. The addition of 2 (1) = 10.90, p < .01. The main effect was significant and positive, suggesting that there was a significant curvili near trend to change in barriers across treatment. Pseudo R 2 in Table 3 16 shows that including the fixed effect of quadratic time accounted for approximately one percent of within person variance. The pseudo R 0 2 value remained negative, indicating that no between person variance was explained. Inclusion of the random effect of quadratic time, summarized in Model 5 in Table 3 2 (1) = 51.49, p < .01. The main effect of quadratic time on barriers to treatment continuation was not significant. Contrary to the overall quadratic trend suggested by Model 4, it is possible that when the quadratic trajectory of change was permitted to vary between subjects, there was not one true curve to the sample. Change in pseudo R 2 seen in Table 3 16, suggests that approximately 10% of within participant variance was explained with the addition of random quadratic time. Pseudo R 0 2 remains negative and was, thus, not interpreted. Figure 3 4 illustrates the estimated plot when quadratic time was modeled. This plot illustrates that while there is a quadratic trajectory of change in the data, there was not one true curve for change in BTPS R. Some participants show a positive quadratic trajectory of change (i.e., u shape d trajectory in which there is a faster decrease in barriers at the start, followed by a slower rate of change, and a slight increase in barriers towards the end of treatment). Other participants show a negative quadratic
65 trajectory of change (i.e., invert ed u shaped pattern of change), and others show the linear pattern of change. Following the addition of linear and quadratic time to the model, both within person and between 2 = .116, p 0 2 = .569, p < .001. There is also significant variance to be explained in linear and quadratic rate of 1 2 = .003, p 2 2 = 0.00, p < .01. Thus, adding additional Level 1 and Level 2 predictors was warranted. Addressing Attrition in the Model A partial pattern mixture model approach was applied to address the influence of attrition on the outcome variable. Attrition was defined as a dichotomous variable in drop out. Model 6 in Table 3 9, included the fixed effect of attrition. Model fit was not significantly improved when attrition is added to the model, and the main effect of attrition on BTPS R was not significant. This suggests that perhaps level of parent perceived barr iers to treatment continuation does not vary based on parent completion status. There was no change in pseudo R 2 Pseudo R 0 2 remained negative and was thus not interpretable. Model 7, in Table 3 9, included the fixed effect of the interaction between attrition 2 (1) = 4.39, p < .05. The main effect of the interaction between attrition and linear time was positive and significant, which suggests that, on average, the linear slope of change fo r treatment barriers was positively related to whether the dyad completed or dropped out of treatment. Inclusion of this interaction in the model did not explain between or within person variance, nor did it explain variance in linear or quadratic change Figure 3 5
66 contains the estimated plots of linear time for participants who completed treatment and for participants who dropped out of treatment. These plots show the differences in slope of change for treatment completers and treatment drop outs. In the following step, the random effect of the interaction between attrition and linear time was added to the model. However, addition of the random effect of this interaction resulted in a Hessian warning, suggesting that the random effect could not be reli ably estimated. Thus, this step was not included in the tables, and no random effects of the interaction between attrition and linear time could be interpreted. In Model 8, shown in Table 3 9, the fixed effect of the interaction between attrition and qua dratic time was added to the model. Improvement in model fit was 2 (1) = 2.79, p < .1. The main effect was negative and significant at trend, suggesting that that the quadratic trajectory of change was related to whether the dyad c ompleted or dropped out of treatment. This suggests that families who initially showed an accelerated increase in treatment barriers were more likely to drop out. Inclusion of this interaction in the model did not explain between or within person varian ce or variance in the linear or quadratic rate of change. The random effect of the interaction between attrition and quadratic time was added to Model 9, shown in Table 3 9. There was no improvement in model fit, and the main effect of the interaction wa s not significant. Thus, when allowing for individual differences between parent child dyads, there was no longer a relationship between treatment completion status and the quadratic trajectory of change. Improvement in within person variance was negligibl e, and there was no improvement in between
67 person variance. The interaction between attrition and quadratic time was removed from subsequent models. Attrition and its interaction with the linear time slope remained in subsequent models to account for the effect of treatment drop out on level and change in perceived barriers. However, because there was no random variance in the attrition by linear time interaction (i.e., there are no individual differences to explain in these terms), interactions between s tudy predictors and attrition were not investigated. Child Disruptive Behavior In Model 10, shown in Table 3 10, the average value of ECBI across time for each family (hereafter referred to as mean ECBI) was added to the model. Model fit was not signif icantly improved with this addition. The main effect of mean ECBI on barriers to treatment continuation was also non significant. It is possible that mothers who report, on average, a higher level of child disruptive behavior do not also report higher or l ower levels of barriers to treatment continuation. Inclusion of mean ECBI explained no within or between person variance. The amount of within person variance to be 2 = .116, p < .001, and between 0 2 = .571, p < .001, remains significant. In Model 11, shown in Table 3 10, the fixed effect of the time varying covariate, centered ECBI, was entered into the model. The main effect of centered ECBI was positive and significant. This suggests that in those weeks when mothers reported an reported level of child disruptive behavior), they also reported a higher level of perceived barriers. Addition of the fixed effect of centered EC BI resulted in a small
68 increase of explained within person variance of approximately one percent. Addition of 2 (1) = 5.01, p < .05. In Model 12, summarized in Table 3 10, the rando m effect of centered ECBI was not significant, suggesting that relations between child disruptive behavior reported at each session do not vary by family. Thus, significant differences between families in this session by session relationship are not detect able. Inclusion of the random effect did not significantly improve the fit of the model. It is important to note that the random effect of centered ECBI was not retained in subsequent models due to the absence of random variance to be explained for centere d ECBI. Model fit for the addition of the fixed and random effects of centered ECBI is summarized in Table 3 15. The fixed effect of the interaction between mean ECBI and linear time was added in Model 13 and is summarized in Table 3 11. This addition res ulted in no improvement of model fit. The main effect was not significant, suggesting that across mother child dyads, average level of ECBI neither explains nor moderates the rate of change in barriers. Additionally, there was no improvement in variance ex plained for the within or between person effect. The random effect of the interaction between mean ECBI and linear time was added in Model 14 and is summarized in Table 3 11. The random effect was significant ( p < .05) and resulted in significant improvem 2 (1) = 10.13, p < .05. Calculation of pseudo R 1 2 for linear change showed that adding this term to the model accounted for approximately 67% of the random variance in linear time, rendering random variance in the linear slope non sig nificant for subsequent models. This finding suggests that despite the non significant fixed effect suggested by Model 13, the
69 varies significantly across individuals. Chan ge in between person and within person variance was negligible, as compared to the previous model. Due to significant variance within the interaction, the random effect of the interaction was retained in subsequent models. In Model 15, summarized in Tabl e 3 11, the fixed effect of the interaction between mean ECBI and quadratic change was added to the model. The main effect of the interaction was not significant. This suggests that the quadratic rate of change does not vary based on overall level of child disruptive behavior. Model fit did not significantly improve with the addition of this interaction. There was no change in between person and within person variance explained. The random effect of the interaction between mean ECBI and quadratic change wa s added in Model 16, shown in Table 3 11. Inclusion of this term in the model accounted for approximately 20% of random variance in quadratic rate of change, as compared to the previous model. However, due to the lack of variance to be explained within the interaction and limited power in the model, the random effect of the interaction was removed from subsequent models. Following the addition of the interaction between ECBI and quadratic time to the model, there remained significant within person variance, 2 = .114, p < .001, and between 0 2 = .440, p < .001. Treatment Expectations In Model 17, shown in Table 3 12, initial CEQ P (i.e., CEQ P score at first treatment occasion [CDI Teach]) was added to the model. Fit of th e model significantly 2 (1) = 9.83, p < .01. The main effect of initial
70 CEQ P was significant and negative, suggesting that mothers who reported lower expectations and beliefs in the credibility of treatment also reported a higher level of barriers to treatment continuation, on average. Calculation of between person pseudo R 0 2 shown in Table 3 16, revealed an increase in explained between person variance of approximately 20%. There was no change in explained wit hin person variance. In Model 18, shown in Table 3 12, CEQ P change (i.e., the change in CEQ P score from beginning to mid point of treatment) was added to the model. Addition of this 2 (1) = 69.72, p < .001. The main effect of change in CEQ P was significant and negative, suggesting that mothers whose expectations and beliefs in treatment credibility decreased from the beginning to the mid point of treatment also reported a higher lev el of barriers, on average. Within person explained variance improved by approximately 2% and between person explained variance improved by approximately 19%. Significant within and between 2 = .109, p < .001; 0 2 = .260, p < .001, warranting further examination of predictors in the model. In Model 19, presented in Table 3 12, the fixed effect of the interaction between initial CEQ P and linear time was added to the model. Improvement in the model was 2 (1) = 3.74, p < .1. The main effect of the interaction was also significant at trend, suggesting that change in barriers varied based on the initial level of parent expectations. Addition of the fixed effect of the interaction di d not contribute to explanation of between or within person variation. In Model 20, shown in Table 3 12, the random effect of the interaction between initial CEQ P and linear time was added to the model but was found to be non
71 significant. Contrary to th e effect of initial CEQ P on the linear slope of change in barriers suggested by Model 19, it is possible that when the effect of this interaction was permitted to vary between subjects, the main effect no longer held for the sample. In addition, model fit did not improve, and change in between and within person variance was negligible. Due to the lack of variance to be explained in this term, the interaction was not included in subsequent models. Model 21, in Table 3 13, included the fixed effect of the interaction between initial CEQ P and quadratic time. The main effect of the interaction was non significant, suggesting that curvilinear change was not related to initial level of parent expectations. Fit of the model was not significantly improved, and n o between or within person variance was explained. In the following step, the random effect of the interaction between initial CEQ P and quadratic time was investigated. This interaction could not be reliably estimated. Thus, this step was not included in the tables, and no random effects of the interaction between CEQ P and quadratic time could be interpreted. Model 22, in Table 3 13, included the fixed effect of the interaction between CEQ P change and linear time. The main effect of the interaction was non significant, suggesting that perhaps linear change in barriers was not related to change in expectations. Change in model fit and variance between and within subjects was negligible. In the next step, the random effect of the interaction between change in CEQ P and linear time could not be reliably estimated. Thus, this step was not included in the
72 table, and no random effects of the interaction between CEQ P and linear time could be interpreted. The fixed effect of the interaction between change in CEQ P and quadratic time was included in Model 23. The main effect of this interaction was positive and significant at trend, suggesting that the curvilinear trend of change in barriers was related to change in parent expectations and beliefs abo ut the credibility of treatment. 2 (1) = 2.76, p < .1. Although there was no change in between and within person variance explained, calculation of pseudo R 2 indicated that inclusion of this term acc ounted for approximately 17% of the random variance in quadratic time as compared to the previous model. The random effect of the interaction between change in CEQ P and quadratic time was then included in Model 24, shown in Table 3 13. The main effect o f this interaction was positive and significant at trend, suggesting that the relationship between the curvilinear change in perceived barriers and change in expectations differed between treatment participants. Addition of the random effect of this intera ction accounted for an additional 5.72% of the random variance in quadratic time. Change in fit was negligible, as was change in between and within person variance. Due to the lack of variance to be explained in this term, the interaction was not included in subsequent models. Following the investigation of change in parent expectations and its interaction with time, there remained significant within and between person variance to be 2 = .111, p 0 2 = .248, p < .001).
73 Parent Soc ioeconomic Status In Model 25, shown in Table 3 14, the fixed effect of SES was added to the model. Because SES was a level 2 variable (i.e., there is one value per family across treatment), its random effect was not investigated. There was neither a sign ificant improvement in model fit nor a significant main effect when SES was included. Change in within person variance was minimal. However, the pseudo R 2 for between person variance indicated that the fixed effect of SES accounted for approximately 3% of between person variance. Model 26, shown in Table 3 14, included the fixed effect of the interaction between parent SES and linear time. Addition of this interaction did not significantly improve model fit. In addition, the main effect of the interactio n was not significant, suggesting that the linear rate of change in barriers was not related to parent socioeconomic status. Calculation of pseudo R 2 for within and between person variance suggests that addition of this interaction accounted for no withi n person variance and approximately 1% of between person variance. In the next step, the random effect of the interaction between SES and linear time could not be reliably estimated. Thus, this step was not included in the table, and no random effects of the interaction between SES and linear time could be interpreted. Model 27, shown in Table 3 14, included the fixed effect of the interaction between parent SES and quadratic time. Model fit did not significantly improve with the addition of this interac tion. In addition, the main effect of the interaction was not significant suggesting that the curvilinear rate of change was not significantly related to parent socioeconomic status. Between and within person variance accounted for by the model was not significantly improved, as compared to the previous model.
74 The random effect of the interaction between SES and quadratic time, added in the next step, could not be reliably estimated. Thus, this step was not included in table 3 14, and no random effects o f the interaction between SES and linear time could be interpreted. Based on pseudo R 2 calculation, the final model accounted for 43% of the between person variance and 56% of the within person variance. However, there remained significant within and bet ween 2 = .111, p 0 2 = .242, p < .001. Relationship between SES and Barriers to Treatment Continuation at the Start of Treatment Relations between SES and parent perceived barriers to treatment continuation we re further explored by assessing whether SES was predictive of level of parent perceived barriers at the first occasion of treatment. Results of a linear regression indicated that SES was not predictive of BTPS R at the first occasion of treatment, = 17, t (42) = 1.08, p = .29. Predicting Drop Out from Treatment A logistic regression was conducted to test the hypothesis that barriers at the last attended session of treatment predicted drop out from treatment. Prior to this analysis, a logistic regression was conducted to determine whether baseline demographic and descriptive variables (i.e., mother and child race, mother and child age, mother and child IQ, mother marital status, mother education level, SES, mother depressive symptomatology, child comorbid diagnoses) also predicted drop out from treatment. Due to the atheoretical nature of this preliminary analysis (i.e., goal of the analysis was the identification of significant predictors of attrition within the sample, and specific
75 variables were not hypo thesized to be significant prior to the analysis), a forward step wise logistic regression was conducted. Table 3 17 summarizes the results of the forward stepwise logistic regression. In Step 1, Wonderlic score (i.e., estimate of mother IQ) was a signifi cant predictor of drop out from treatment, B = 0.07, p < .05, OR = .93, with higher IQ associated with lower chance of drop out from treatment. This single predictor step was a significant 2 (1) = 5.40, p < .05. Sensitivi ty and specificity were 16.7 and 94.1, respectively meaning that the model correctly identified approximately 17% of families that dropped out of treatment and approximately 94% of families that completed treatment, based on Wonderlic score. A hierarchic al logistic regression was then conducted to test the hypothesis that level of parent perceived barriers to treatment continuation at the last attended session is a significant predictor of drop out. The block entry method was selected, with the covariate from the previous analysis, Wonderlic score, as well as BTPS R score at the first treatment session, entered in the first block and level of barriers at the last attended session entered in the second block. Table 3 18 summarizes the results of this analys is. The first block of the analysis confirmed the result of the previous stepwise logistic regression, indicating that Wonderlic score significantly predicted treatment drop out, B = 0.07, p < .05, OR = .93. Barriers at the first occasion of treatment was not a significant predictor of attrition. Improvement in model fit relative to the null model was 2 (2) = 5.61, p = .06. Sensitivity and specificity for predicting attrition were 16.7 and 97 .1 respectively meaning that the mode l correctly identified approximately 17% of families that dropped out of treatment and approximately 97% of
76 fam ilies that completed treatment, based on Wonderlic score and level of barriers at the first occasion of treatment. The second block investigated whether adding the BTPS R score at the last attended session improved the model. Level of barriers at the last attended session was a significant predictor of drop out from treatment, B = 0.30, p < .05, OR = 1.35. Mother Wonderlic score also predicted attr ition at a trend level of significance, B = 0.07, p = .06, OR = .93. The value of the odds ratio greater than one indicated that as the score on the BTPS R increased, the likelihood of dropping out of treatment increased as well. The second block entered into the logistic regression, which included BTPS R at the 2 (3) = 11.98, p 2 (1) = 6.37, p = .01. When BTPS R at the last attended session was included in the model, sensitivity of predicting drop out increased to 33% and specificity was 94%.
77 Table 3 1. Skewness and kurtosis for BTPS R at each occasion of treatment Occasion N Skewness St. Error of Skewness Kurtosis St. Error of Kurtosis 1 43 1.32 0.36 0.96 0.71 2 42 2.02 0.37 4.31 0.72 3 41 1.97 0.37 4.59 0.72 4 37 1.67 0.39 2.18 0.76 5 37 2.39 0.39 5.84 0.76 6 36 1.63 0.39 1.87 0.77 7 37 2.65 0.39 7.90 0.76 8 37 2.25 0.39 4.74 0.76 9 37 2.06 0.39 3.74 0.76 10 35 1.76 0.40 1.73 0.78 11 36 2.85 0.39 8.50 0.77 12 31 3.01 0.42 10.17 0.82 13 27 2.67 0.45 8.05 0.87 14 25 3.08 0.46 10.78 0.90 15 22 4.14 0.49 18.07 0.95 16 13 2.68 0.62 7.54 1.19 17 13 3.45 0.62 12.15 1.19 18 9 2.64 0.72 7.29 1.40 19 7 1.42 0.79 1.19 1.59 20 6 1.82 0.85 2.98 1.74 21 1 22 1 Note. N = 43 at first occasion due to missing data at first occasion. Sample size varied by occasion due to missing data, attrition, and treatment completion.
78 Table 3 2. Correlations between predictor variables Predictor 1 2 3 4 5 6 7 8 9 10 11 12 13 1 Mean ECBI 2 Centered ECBI .06 3 Mean ECBI x Lin Time .00 .00 4 Mean ECBI x Quad Time .00 .23** .10* 5 Initial CEQ P .17** .00 .02 .01 6 Change in CEQ P .35** .00 .10* .05 .49** 7 Initial CEQ P x Lin Time .00 .01 .18** .01 .00 .02 8 Initial CEQ P x Quad Time .02 .13** .01 .20** .00 .03 .14** 9 Change CEQ P x Lin Time .03 .02 .39** .02 .04 .00 .46** .04 10 Change CEQ P x Quad Time .03 .25** .03 .34** .01 .00 .03 .47** .00 11 SES .16** .00 .10* .05 .18** .11** .01 .03 .02 .01 12 SES x Lin Time .10* .07 .07 .11** .01 .03 .22** .05 .13** .01 .00 13 SES x Quad Time .05 .06 .13** .06 .01 .01 .05 .20** .01 .15** .00 .21** Note. ECBI = Eyberg Child Behavior Inventory Intensity Scale. CEQ P = The Credibility/ Expectancies Questionnaire, Parent Version. SES = Hollingshead Socioeconomic Status. Lin Time = Linear Time. Quad Time = Quadratic Time. p < .05. ** p < .01.
79 Table 3 3. Means for descriptive and demographic variables Variable Mean S.D. Child Age 4.78 0.81 Child PPVT 104.11 13.24 Mother Age 35.16 7.02 Mother Wonderlic 106.78 12.86 Hollingshead SES 43.37 11.91 Note. N = 46. PPVT = Peabody Picture Vocabulary Test Third Edition. Hollingshead SES = Hollingshead Socioeconomic Status.
80 Table 3 4. Frequency counts for demographic variables Frequency % Child Characteristics Sex (male) 30 65 Race/ Ethnicity Caucasian 28 61 African American 10 22 Hispanic 1 2 Bi racial 7 15 ADHD Only 8 17 ADHD and Co morbid Diagnosis Oppositional defiant disorder 36 78 Conduct disorder 11 24 Separation anxiety disorder 7 15 Mother Characteristics Race/ Ethnicity Caucasian 31 67 African American 10 22 Hispanic 3 7 Bi racial 2 4 Other Marital Status Married 31 67 Divorced 6 13 Separated 2 4 Never married 6 13 Highest Education Level Graduate Degree 6 13 Completed College 13 28 Some College 19 41 Completed High School 7 15 Less than High School Diploma 1 2 Note. N = 46. Frequency and percentage of comorbid diagnoses is based on the YC DISC ( Strong, Lucas, & Lucas, 2006) diagnostic interview with parents.
81 Table 3 5. Means for descriptive and demographic variables by group status Individual Group Variable Mean S.D. Mean S.D. Child Age 4.78 0.81 4.83 0.82 Child PPVT 104.11 13.24 102.96 13.53 Mother Age 35.00 7.02 34.79 7.65 Mother Wonderlic 106.78 12.86 103.83 12.48 Hollingshead SES 44.91 11.81 41.96 12.07 Note. Individual N = 22; Group N = 24. PPVT = Peabody Pictur e Vocabulary Test Third Edition ; Hollingshead SES = Hollingshead Socioeconomic Status.
82 Table 3 6. Frequency counts for demographic variables by group status Individual Group Frequency % Frequency % Child Characteristics Sex (male) 15 68 15 63 Race/ Ethnicity Caucasian 16 73 12 50 African American 5 23 5 21 Hispanic 0 0 1 4 Bi racial 1 5 6 25 Mother Characteristics Race/ Ethnicity Caucasian 17 77 14 58 African American 5 23 5 21 Hispanic 0 0 3 13 Bi racial 0 0 2 8 Other 0 0 0 0 Marital Status Married 14 64 17 71 Divorced 3 14 3 13 Separated 1 5 2 8 Never Married 4 18 2 8 Highest Education Level Graduate Degree 3 14 3 13 Completed College 7 32 6 25 Some College 10 46 9 38 Completed High School 2 9 5 21 Less than High School Diploma 0 0 1 4 Note. Individual N = 22; Group N = 24.
83 Table 3 7. Means for level 2 variables Variable Mean S.D. Mean ECBI 117.39 34.24 Initial CEQ P 41.93 7.36 Change in CEQ P 2.19 6.24 Hollingshead SES 43.37 11.91 Note. N = 46, except for Mean and Standard Deviation for Change in CEQ P, where N = 37 due to missing data at PDI Teach. ECBI = Eyberg Child Behavior Inventory ; CEQ P = The Credibility/ Expectancies Questionnaire, Parent Version.
84 Table 3 8. Results of model tests for parent reported barriers to treatment continuation, Model 1 Model 5 Model 1 Model 2 Model 3 Model 4 Model 5 Fixed Effects Intercept 0.11 (.10) 0.36** (.10) 0.36** (.10) 0.34** (.12) 0.40** (.12) Rate of Change Time (linear) 0.04**(.00) 0.04**(.00) 0.03**(.01) 0.05**(.01) Time (quadratic) 0.00**(.00) 0.00 (.00) Random Effects Level1 Within person 0.25** (.02) 0.21**(.01) 0.14**(.01) 0.14**(.01) 0.12**(.01) Level 2 In initial status 0.43** (.10) 0.43**(.10) 0.58**(.13) 0.58**(.13) 0.57**(.13) In linear change 0.00**(.00) 0.00**(.00) 0.00** (.00) In quad. change 0.00**(.00) Note. Standard errors are between parentheses; Model 1 = Unconditional Means Model; Model 2 = Addition of Linear Time (fixed); Model 3 = Addition of Linear Time (random); Model 4 = Addition of Quadratic Time (fixed); Model 5 = Addition of Quadratic Time (random). p < .05. **p < .01.
85 Table 3 9. Results of model tests for parent reported barriers to treatment continuation, Model 6 Model 9 Model 6 Model 7 Model 8 Model 9 Fixed Effects Intercept 0.41**(.13) 0.38** (.13) 0.39** (.14) 0.39** (.13) Attrition 0.06 (.27) 0.16 (.29) 0.17 (.29) 0.11 (.30) Rate of Change Time (linear) 0.05** (.01) .05** (.01) 0.05** (.01) .05** (.01) X Attrition 0.08* (.04) 0.04 (.05) 0.04 (.05) Time (quadratic) 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) X Attrition # 0.01 (.01) 0.01 (.01) Random Effects Level1 Within person 0.12**(.01) 0.12** (.01) 0.12** (.01) 0.11** (.01) Level 2 In initial status 0.57**(.13) 0.58** (.13) 0.59** (.13) 0.57** (.13) In linear change 0.00**(.00) 0.00**(.00) 0.00**(.00) 0.00**(.00) In quadratic change 0.00** (.00) 0.00** (.00) 0.00** (.00) 0.00** (.00) In attrition x quad. change # 0.00 (.00) Note. Standard errors are between parentheses; Model 6 = Addition of Attrition (fixed); Model 7 = Addition of Attrition x Linear Time (fixed); Model 8 = Addition of Attrition x Quadratic Time (fixed); Model 9 = Addition of Attrition x Quadratic Time (random). p < .10. p < .05. ** p < .01. # parameter was not included in subsequent models.
86 Table 3 10. Results of model tests for parent reported barriers to treatment continuation, Model 10 Model 12 Model 10 Model 11 Model 12 Fixed Effects Intercept 0.18 (.42) 0.12 (.42) 0.13 (.42) Attrition 0.13 (.30) 0.19 (.30) 0.18 (.29) Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) Centered ECBI 0.00*(.00) 0.00 (.00) Rate of Change Time (linear) 0.05** (.01) 0.04** (.01) 0.04** (.01) X Attrition 0.08* (.04) 0.08* (.04) 0.08* (.04) Time (quadratic) 0.00 (.00) 0.00 (.00) 0.00 (.00) Random Effects Level1 Within person 0.12** (.01) 0.11** (.01) 0.11** (.01) Level 2 In initial status 0.57** (.13) 0.58** (.13) 0.55** (.13) In linear change 0.00** (.00) 0.00** (.00) 0.00** (.00) In quadratic change 0.00** (.00) 0.00** (.00) 0.00** (.00) In Centered ECBI # 0.00 (.00) Note. Standard errors are between parentheses; Model 10 = Addition of Mean ECBI (fixed); Model 11 = Addition of Centered ECBI (fixed); Model 12 = Addition of Centered ECBI (random). p < .10. p < .05. ** p < .01. # parameter was not included in subsequent models.
87 Table 3 11. Results of model t ests for parent reported barriers to treatment continuation, Model 13 Model 16 Model 13 Model 14 Model 15 Model 16 Fixed Effects Intercept 0.05 (.48) 0.28 (.44) 0.27 (.44) 0.28 (.44) Attrition 0.19 (.30) 0.08 (.27) 0.08 (.27) 0.08 (.27) Mean ECBI 0.00 (.00) 0.01 (.00) 0.01 (.00) 0.01 (.00) Centered ECBI 0.00*(.00) 0.00* (.00) 0.00* (.00) 0.00* (.00) Rate of Change Time (linear) 0.04** (.01) 0.04** (.01) 0.03** (.01) 0.03** (.01) X Attrition 0.08* (.04) 0.07* (.03) 0.07* (.03) 0.07* (.03) X Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) Time (quadratic) 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) X Mean ECBI 0.00 (.00) 0.00 (.00) Random Effects Level1 Within person 0.11** (.01) 0.11** (.01) 0.11** (.01) 0.11** (.01) Level 2 In initial status 0.58**(.13) 0.44** (.10) 0.44** (.11) 0.44** (.11) In linear change 0.00**(.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) In quadratic change 0.00**(.00) 0.00** (.00) 0.00**(.00) 0.00 (.00) In Mean ECBI X linear 0.00*(.00) 0.00*(.00) 0.00*(.00) In Mean ECBI X quadratic # 0.00 (.00) Note. Standard errors are between parentheses; Model 13 = Addition of Mean ECBI x Linear Time (fixed); Model 14 = Addition of Mean ECBI x Linear Time (random). Model 15 = Addition of Mean ECBI x Quadratic Time (fixed); Model 16 = Addition of Mean ECBI x Quadratic Time (random). p < .10. p < .05. ** p < .01 # parameter was not included in subsequent models.
88 Table 3 12. Results of model tests for parent reported barriers to treatment continuation, Model 17 Model 20 Model 17 Model 18 Model 19 Model 20 Fixed Effects Intercept 1.89* (.74) 3.52** (.93) 3.29** (.95) 3.31** (.93) Attrition 0.20 (.25) 0.22 (.34) 0.21 (.35) 0.22 (.34) Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) Centered ECBI 0.00* (.00) 0.00** (.00) 0.00** (.00) 0.00** (.00) Initial CEQ P 0.05** (.01) 0.07** (.02) 0.07** (.02) 0.07** (.02) CEQ P Change 0.35** (.12) 0.35** (.12) 0.35** (.12) Rate of Change Time (linear) 0.04** (.01) 0.03** (.01) 0.03** (.01) 0.03** (.01) X Attrition 0.07* (.03) 0.06 (.03) 0.05 (.03) 0.06 (.03) X Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) X Initial CEQ P 0.00 (.00) 0.00 (.00) Time (quadratic) 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) X Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) Random Effects Level1 Within person 0.11** (.01) 0.11** (.01) 0.11** (.01) 0.11** (.01) Level 2 In initial status 0.34** (.08) 0.26** (.07) 0.27** (.07) 0.26** (.07) In linear change 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) In quadratic change 0.00** (.00) 0.00** (.00) 0.00** (.00) 0.00** (.00) In ECBI X linear 0.00 (.00) 0.00*(.00) 0.00 (.00) 0.00 (.00) In Initial CEQ P X linear # 0.00 (.00) Note. Standard errors are between parentheses; Model 17 = Addition of Initial CEQ P (fixed); Model 18 = Addition of CEQ P Change (fixed); Model 19 = Addition of Initial CEQ P x Linear Time (fixed); Model 20 = Addition of Initial CEQ P x Linear Time (random); Mo del 21 = Addition of Initial CEQ P x Quadratic Time (fixed). p < .10. p < .05. **p < .01.
89 Table 3 13. Results of model tests for parent reported barriers to treatment continuation, Model 21 Model 24 Model 21 Model 22 Model 23 Model 24 Fixed Effects Intercept 3.27** (.95) 3.38**(.96) 3.27** (.95) 3.25** (.95) Attrition 0.21 (.34) 0.20 (.34) 0.20 (.34) 0.20 (.34) Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) Centered ECBI 0.00** (.00) 0.00** (.00) 0.00** (.00) 0.00** (.00) Initial CEQ P 0.07** (.02) 0.07** (.02) 0.07** (.02) 0.07** (.02) CEQ P Change 0.35** (.12) 0.38** (.12) 0.37** (.13) 0.37** (.13) Rate of Change Time (linear) 0.03** (.01) 0.03** (.01) 0.03** (.01) 0.03** (.01) X Attrition 0.06 (.03) 0.06 (.03) 0.06 (.03) 0.06 (.03) X Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) X Initial CEQ P 0.00* (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) X CEQ P Change 0.00 (.00) 0.01 (.01) 0.01 (.01) 0.01 (.01) Time (quadratic) 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) X Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00* (.00) X Initial CEQ P 0.00 (.00) X CEQ P Change 0.00 (.00) Random Effects Level1 Within person 0.11** (.01) 0.11** (.01) 0.11** (.01) 0.11** (.01) Level 2 In initial status 0.26** (.07) 0.26** (.07) 0.26** (.07) 0.26** (.07) In linear change 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) In quadratic change 0.00** (.00) 0.00** (.00) 0.00* (.00) 0.00 (.00) In ECBI X linear 0.00 (.00) 0.00 (.00) 0.00 (.00) 0.00 (.00) In CEQ P change X quadratic # 0.00 (.00) 0.00 (.00) Note. Standard errors are between parentheses; Model 22 = Addition of CEQ P Change x Linear Time (fixed); Model 23 = Addition of CEQ P Change x Quadratic Time (fixed); Model 24 = Addition of CEQ P Change x Quadratic Time (random). p < .10. p < .05. **p < .01. # parameter was not included in subsequent models.
90 Table 3 14. Results of model tests for parent reported barriers to treatment continuation, Model 25 Model 27 Model 25 Model 26 Model 27 Fixed Effects Intercept 2.71*(.93) 2.80*(.91) 2.80*(.91) Attrition 0.20 (.34) 0.20 (.33) 0.20 (.33) Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) Centered ECBI 0.00** (.00) 0.00** (.00) 0.00** (.00) Initial CEQ P 0.07** (.02) 0.07** (.02) 0.06** (.02) Change in CEQ P 0.38** (.13) 0.37** (.12) 0.37** (.12) Rate of Change Time (linear) 0.03** (.01) 0.02* (.01) 0.02* (.01) X Attrition* 0.06 (.03) 0.06 (.03) 0.06 (.03) X Mean ECBI 0.00 (.00) 0.00 (.00) 0.00 (.00) X Initial CEQ P 0.00 (.00) 0.00 (.00) 0.00 (.00) X Change CEQ P 0.01 (.01) 0.00 (.01) 0.00 (.01) X SES 0.00 (.00) 0.00 (.00) Time (quadratic) 0.00 (.00) 0.00 (.00) 0.00 (.00) X Mean ECBI 0.00* (.00) 0.00* (.00) 0.00* (.00) X Initial CEQ P 0.00 (.00) 0.00* (.00) 0.00* (.00) X Change CEQ P 0.00 (.00) 0.00 (.00) 0.00 (.00) X SES 0.00 (.00) Random Effects Level1 Within person 0.11** (.01) 0.11** (.01) 0.11** (.01) Level 2 In initial status 0.25** (.07) 0.24** (.06) 0.24** (.06) In linear change 0.00 (.00) 0.00 (.00) 0.00 (.00) In quadratic change 0.00* (.00) 0.00* (.00) 0.00* (.00) In ECBI X linear change 0.00 (.00) 0.00 (.00) 0.00 (.00) Note. Standard errors are between parentheses; Model 25 = Addition of SES (fixed); Model 26 = Addition of SES x Linear Time (fixed); Model 27 = Addition of SES x Quadratic Time (random). p < .10. p < .05. **p < .01.
91 Table 3 15. Change in fit across models Model 2LL 2LL AIC BIC (1) Null 965.56 971.56 984.62 (2) Add linear time, fixed 879.65 85.91 887.65 905.06 (3) Add linear time, random 756.10 123.55 766.10 787.85 (4) Add quadratic time, fixed 745.20 10.90 757.20 783.30 (5) Add quadratic time, random 693.71 51.49 707.71 738.17 (6) Add attrition, fixed 693.66 0.05 709.66 744.46 (7) Add attrition x linear time, fixed 689.27 4.39 707.27 746.42 (8) Add attrition x quadratic time, fixed 686.48 2.79 706.48 749.99 (9) Add attrition x quadratic time, random* 685.14 1.34 707.14 754.10 (10) Add mean ECBI, fixed 689.01 0.26 709.01 752.52 (11) Add centered ECBI, fixed 684.01 5.00 706.01 753.87 (12) Add centered ECBI, random* 683.33 0.68 707.33 759.54 (13) Add mean ECBI x linear time, fixed 683.91 0.10 707.91 760.12 (14)Add mean ECBI x linear time, random 673.78 10.13 699.78 756.35 (15)Add mean ECBI x quadratic time, fixed 673.51 0.27 701.51 762.42 (16) Add mean ECBI x quadratic time, random* 673.54 0.03 703.54 768.80 (17) Add initial CEQ P, fixed 663.71 9.80 693.71 758.97 (18) Add CEQ P change, fixed 593.99 69.72 625.99 695.01 (19) Add initial CEQ P x linear time, fixed 590.25 3.74 626.25 703.90 (20) Add initial CEQ P x linear time, random 590.25 0.00 626.25 703.90 (21) Add initial CEQ P x quadratic time, fixed 589.59 0.66 625.59 703.23 (22) Add CEQ P Change x linear time, fixed 588.73 0.86 626.73 708.69 (23) Add CEQ P Change x quadratic time, fixed 585.97 2.76 625.97 712.24 (24) Add CEQ P Change x quadratic time, random* 585.85 0.12 627.85 718.43 (25) Add SES, fixed 584.20 1.77 626.20 716.78 (26) Add SES x Linear Time, fixed 581.66 2.54 625.66 720.56 (27) Add SES x Quadratic Time, fixed 581.65 0.01 627.65 726.86 is indicative of a model term that was not included in subsequent models is inidcative of a model term that was included in the next model but then removed from subsequent models
92 Table 3 16. Change in within subjects and between subjects variance explained across models Model 2 0 2 pseudo R 2 pseudo R 0 2 (1) Null 0.25 0.43 (2) Add linear time, fixed 0.21 0.43 0.15 0.00 (3) Add linear time, random 0.14 0.58 0.43 0.35 (4) Add quadratic time, fixed 0.14 0.58 0.44 0.36 (5) Add quadratic time, random 0.12 0.57 0.54 0.33 (6) Add attrition, fixed 0.12 0.57 0.54 0.33 (7) Add attrition x linear time, fixed 0.12 0.58 0.54 0.34 (8) Add attrition x quadratic time, fixed 0.12 0.59 0.54 0.38 (9) Add attrition x quadratic time, random* 0.11 0.57 0.54 0.34 (10) Add mean ECBI, fixed 0.12 0.57 0.54 0.34 (11) Add centered ECBI, fixed 0.11 0.58 0.55 0.13 (12) Add centered ECBI, random* 0.11 0.55 0.55 0.13 (13) Add mean ECBI x linear time, fixed 0.11 0.58 0.54 0.35 (14)Add mean ECBI x linear time, random 0.11 0.44 0.54 0.03 (15)Add mean ECBI x quadratic time, fixed 0.11 0.44 0.54 0.03 (16) Add mean ECBI x quadratic time, random* 0.11 0.44 0.54 0.03 (17) Add initial CEQ P, fixed 0.11 0.34 0.54 0.20 (18) Add CEQ P change, fixed 0.11 0.26 0.56 0.39 (19) Add initial CEQ P x linear time, fixed 0.11 0.27 0.56 0.38 (20) Add initial CEQ P x linear time, random 0.11 0.26 0.56 0.39 (21) Add initial CEQ P x quadratic time, fixed 0.11 0.26 0.56 0.38 (22) Add CEQ P Change x linear time, fixed 0.11 0.26 0.56 0.38 (23) Add CEQ P Change x quadratic time, fixed 0.11 0.26 0.56 0.39 (24) Add CEQ P Change x quadratic time, random* 0.11 0.26 0.56 0.39 (25) Add SES, fixed 0.11 0.25 0.56 0.42 (26) Add SES x Linear Time, fixed 0.11 0.24 0.56 0.43 (27) Add SES x Quadratic Time, fixed 0.11 0.24 0.56 0.43 is indicative of a model term that was not included in subsequent models is indicative of a model term that was included in the next model but then removed from subsequent models
93 Table 3 17. Summary of stepwise logistic regression predicting baseline predictors of attrition from treatment Model Description B SE Wald OR 95% CI P 2LL Improvement Over Null Step 0: Null Model Intercept 1.04 0.34 9.62 0.35 0.002 52.81 Step 1: Mother IQ only Intercept 6.34 3.49 3.30 564.00 0.07 Wonderlic 0.07 0.03 4.34 0.93 [0.87, 1.00] 0.04 47.40 5.40 Note. CI = confidence interval for odds ratio (OR).
9 4 Table 3 18. Summary of hierarchical logistic regression predicting attrition from treatment Model Description B SE Wald OR 95% CI P 2LL Improvement Over Null Improvement Over Previous Model 0: Null Model Constant 1.04 0.34 9.62 0.35 0.002 52.81 Model 1: Covariates Added Constant 7.03 3.84 3.35 1124.81 0.07 Wonderlic 0.07 0.03 4.31 0.93 [0.87, 1.00] 0.04 BTPS R at First Session 0.03 0.06 0.20 0.97 [0.86, 1.10] 0.66 47.19 5.61 Model 2: Covariates + Barriers Constant 3.13 4.35 0.52 22.88 .47 Wonderlic 0.07 0.04 3.53 0.93 [0.87, 1.00] .06 BTPS R at First Session 0.16 0.13 1.58 0.85 [0.66, 1.10] .21 BTPS R at Last Session 0.30 0.15 4.22 1.35 [1.01, 1.81] .04 40.83 11.98 6.37 Note. CI = confidence interval for odds ratio (OR).
95 Figure 3 1. Spaghetti plot of Blom normalized BTPS R score for each parent child dyad. Lines represent scores for each family across occasions
96 Figure 3 2. Curve estimation for pattern of change of Blom normalized BTPS R score
97 Figure 3 3. Fitted model of linear trajectory of change in BTPS R
98 Figure 3 4. Fitted model of quadratic trajectory of change in BTPS R scores
99 (A) (B) Figure 3 5. Fitted model of linear change in BTPS R scores. (A) The fitted model of linear change for treatment completers, (B) The fitted model of linear change for treatment drop outs
100 CHAPTER 4 DISCUSSION When families begin psychotherapy for child behavioral problems, they have already overcome significant barriers to engaging in treatment, including the pervasive stigma that many associate with mental health treatment (Janz et al., 2002). However, once a family is involved in mental health treatment, a new set of barriers emerge barriers to continuing tre atment. Although past research has identified demographic factors such as ethnic minority status, low SES, and single parent status that are associated with treatment attendance and drop out, (Armbruster & Schwab Stone, 1994; Fernandez & Eyberg, 2009; Gou ld et al., 1985; Kazdin et al., 1993; Mccabe, 2002; Wierzbicki & Pekarik, 1993 ), demographic constructs are difficult, if not impossible, to alter within the psychotherapy context. Increasing understanding of barriers to treatment continuation is a more useful way of conceptualizing why families miss appointments or stop attending treatment than identifying demographic constructs associated with attrition, bec ause barriers address treatment related factors relevant to the psychotherapeutic process (Nock & Ferriter, 2005). Many t reatment barriers are also potentially changeable within the therapy context (Nock & Ferriter, 2005). Identification of barriers to treatment continuation and predictors of these barriers provide greater opportunity to address withi n the therapeutic services (Nock & Ferriter, 2005). This study investigated change in parent perceptions of barriers to treatment continuation. Multilevel modeling was implemented to assess whether perceived barriers to treatment continuation changes over the course of PCIT. Predictors of level
101 of perceived barriers and change in perceived barriers were then examined to determine the factors related to this change. F inally, relations between treatment barr iers and treatment attrition were investigated through logistic regression to assess whether the level of treatment barriers at the last attended session of treatment significantly predicted drop out from treatment. Researchers have suggested that repeated measurement of potential barriers throughout treatment could sensitize families to the difficulties of remaining in treatment and potentially jeopardize attendance (Kazdin, Howland, & Crowley, 1997). In this study, however, repeated administration of the BTPS R did not appear to affect decrease in perceived barriers during treatment. Additionally, the treatment drop out rate of 26% in this study compares favorably to average drop out rate of 40 to 60% from child mental health treatment (Wierzbicki & Pekarik, 1993). Results sho wed a significant linear decrease in barriers over time across participants. Some mother child dyads also sh owed a curvilinear pattern of change in which there was a faster decrease in perceived barriers early in treatment, followed by a slowing in the rate of change, and then a slight increase in barriers late in treatment. Past research on treatment barriers h as primarily investigated perceived barriers at the barriers to treatment may change and evolve during treatment. To our knowledge, this is the first study to examin e whether barriers change across time and the first study to examine treatment barriers more than once during PCIT. The results suggest that perceived barriers are not a fixed construct experienced in the same way across
102 treatment, but are more a reflecti on of varying influences a family experiences at different points during treatment. Predictors of Level and Change of Perceived Barriers Child Disruptive Behavior Across families, the average level of mother reported child disruptive behavior across trea tment sessions was unrelated to individual differences between the families perceived barriers to treatment continuation. However, on occasions in which mothers level of disruptive behavior), mothers also reported experiencing higher levels of perceived barriers to continuing in treatment. Conversely, on occasions in which mothers reported below average child disruptive behavior, mothers also reported fewer barriers to treatment. This relationship was roughly the same for all families in the study. Additionally, a lthough on average, the general mean level of child disruptive behavior does not explain rate of change in perceived barriers, for some di sruptive behavior does in fact, explain rate of change in perceived barriers to treatment continuation. hyperactive behavior. Perceived barriers represent factors inte commitment to continuing treatment. It is plausible that on occasions in which children are more disruptive, parents may report more obstacles to remaining committed to treatment particularly s not improving. y contribute to tensions among family members and time spent prac ticing treatment skills. O n occasions that parents ruptiveness may in itself
103 become a barrier that is not directly addressed on the barriers mea sure, but is reflected rises at home make it hard f reatment takes time away from spending time with my children. Past investigations of relations between child externalizing behavior and barriers to treatment continuation have been inconclusive, with research suggesting both significant relations (Budd et al., 2010; Perez, 2008) and non significant relations (Kazd in, Holland, Crowley, & Breton, 1997). Perez (2008) investigated the effect of treatment barriers on change in child disruptive behavior during PCIT and found that at the end of treatment of the barriers experienced during tr eatment barriers associated with less change in child disruptive behavior. In the Perez (2008) study, treatment barriers also accounted for a substantial portion of the individ ual differences between parents in rate of change of perceived barriers to treatment. In our study, the average level of child disruptive behavior during treatment explained a significant portion of individual differences in rate of change in perceived bar riers. Thus, although there does not appear to be a significant relationship between child disruptive behavior and change in barriers across participants, child disruptive behavior accounts for a substantial portion of the individual differences in change in perceived barriers. Mother Expectations and Beliefs in the Credibility of Treatment Both initial beliefs about the effectiveness and credibility of the treatment offered and the extent of change in these beliefs during treatment were generally related to with lower expectations also reported more barriers to continuing treatment. When g the
104 initial phase of treatment, their perception of barriers increased. These findings suggest that higher perceived barriers are related to both the expectations that mothers start treatment with and how their expectations and beliefs in treatment credi bility change during treatment. Taken together, results suggest that mothers with more skepticism regarding the efficacy of treatment for their child also experience treatment as less worthwhile and as a greater burden. Neither initial expectations for treatment nor change in expectations moderated the linear rate of change in parent perceived barriers However, trend level evidence suggested that when expectations for positive treatment outcome increased during CDI, moth s of barriers showed a faster initial decline, followed by a slowing in the rate of change. Th is effect was not significantly different between mother child dyads. It is important to note however, that the moderating effect of expectations on change in barriers was small. R esults of this study are consistent with past research in which lower initial expectations were related to higher perceived barriers at post treatment (Nock & Kazdin, 2001; Ste expectations and beliefs about treatment are signifi cantly related to their motivation in treatment and their adherence to treatment (Nock et al., 2007). Treatment expectations and beliefs in the credibility of treatment a ppear to be important predictors of factors that both interfere with treatment (i.e., treatment barriers) and facilitate treatment (e.g., motivation, adherence to therapist recommendations). Nock and colleagues (2007) posited that by believing strongly in a positive treatment outcome, parents with high
105 expectations may feel greater commitment to becoming an active contributor to positive changes during treatment. The relevance of expectations to treatment barriers is encouragi ng for clinical intervention, because expectations are easily targetable within a therapy or assessment session. Expectations can be measured at the first contact with a family, such as during the intake evaluation before treatment begins. It is n ot uncomm on for families to drop out after the initial contact, before even starting treatment. In fact, 8 of the 54 families recruited for this study dropped out after the initial intake assessment and before barriers or expectations were even assessed. Measuring expectations for treatment at the first contact may facilitate recognition of families at risk of attrition at this early point. Families that report initially low expectations for treatment or a decline in expectations during treatment may benefit from ta rgeted discussion with the clinician about their expectations Brief preparatory interventions targeting treatment expectations have been associated with more accurate expectations and fewer cancelled or missed appointments (Day & Reznikoff, 1980). However past research showing that parents with moderate expectations for a family based treatment were more likely to drop out of treatment than parents with the lowest and highest expectations suggests that parents with even moderate levels of treatment expect ations remain vulnerable to dropout (Nock & Kazdin, 2001). In fact, given the absence of detrimental effects of preparatory strategies targeting treatment expectations, expectations for treatment remain a relevant and useful construct to address with all f amilies engaging in treatment.
106 Family Socioeconomic Status The influence of SES on change in parent perceived barriers to treatment continuation w as an exploratory question in this study. Earlier studies have found that SES is not related to treatment bar riers (Kazdin, Holland, & Crowley, 1997; Kazdin, Holland, Crowley, & Breton, 1997; Stephens et al., 2006). This is the first study to our knowl edge to investigate relations between SES and c hange in barriers. We found SES u nrelated to the level of perceive d barriers or the change in perceived barriers during treatment. Although several items on both t he BTPS and the BTPS R address external demands on the family suggest little overlap between treatment barriers, as measured by the BTPS R and family demographic constructs such as SES. Kazdin, Holland, Crowley, and Breton (1997) p rop osed that the contribution of parent perceived bar riers to the prediction of attrition over and above familial variables, such as SES, supports the incremental validity of the barriers measure. They further proposed that measure s of treatment barriers capture unique information pertinent to critical trea tment related issues such as risk of attrition. Both low perceived relevance of treatment and parent therapist relationship problems, as captured by the BTPS and the BTPS R, may be more significant barriers than external, environmental demands for familie s in treatment (Stephens et al., 2006). Attrition from Treatment Attrition was included in the HLM to account for non random missing data due to drop out from treatment. In this study, the average level of perceived barriers did not differ between mother s who completed treatment and mothers who dropped out
107 However, results indicated that across families, the rate of change of parent perceived barriers differed for families who completed versus dropped out of treatment. Most studies examining parent pe rceived barrier s to treatment continuation measure ba rriers at the time families are leaving or have already left treatment. This methodology leads to missing data when families drop out of treatment without notifying the therapist or do not return for pos t treatment assessment. In this study, with barriers measured at each treatment session, we were able to investigate barriers at the last attended session for all participating families specifically examining whether perceived barriers to treatment completion at the last attended session is related to dropout versus completer status. Our results were consistent with previous studies: After controlling for perceived barriers at the start of treatment and significant demographic predictors of drop out (i.e., mother IQ), the dropout in our sample remained positively associated with perceived barriers at treatment exit. This finding suggests that the level of perceived treatment barriers has more predictive power at the last point of contact than at the s tart of treatment. However, many families that drop out of treatment d o so at an early point in the treatment process. If treatment barriers we re assessed regularly over the course of treatment, an elevation in parent reported barriers might be an early wa rning sign of attrition, allowing the therapist to address concerns to reduce risk of dropout. A survival analysis of predictors of attrition from PCIT would provide greater perspective into relations between per ceived barriers and the point in treatment w hen risk of attrition is highest Although earlier studies documented the relation between perceived barriers and treatment attrition (Kazdin, Holland, & Crowley, 1997; Kazdin, Holland, Crowley, &
108 Breton, 1997; Stevens et al., 2006), these studies examine d barriers only after attrition or completion had already occurred. The difficulty of obtaining data from treatment dropouts leads to over representation of treatment completers in such samples and may bias results toward their experience of the barriers t o treatment. Danko and Budd (2009) attempted to circumvent this selection bias by administering the BTPS R at the session in which parents indicated their intent to discontinue treatment. However, families often drop out of treatment without notice. This s tudy assessed barriers at each treatment session to maximize responses from dropouts as well as completers. Thus, data on treatment barriers were obtained from all families at the point of termination, successfully circumventing selection bias in the analy ses of relations between perceived barriers and attrition. Barriers to Treatment Continuation and Models of Health Services Usage Anderse care was related to predisposing factors (i.e., demographic characteristics, social standing, and health beliefs), enabling factors (i.e., knowledge of how to access services, travel distance), and perceived need. Although Anderse sing services, results of this study support the relevance of these factors to continued participation in a treatment program. In particular, BTPS R items specifically relate to predisposing health beliefs and enabling factors. However, it is important to note that the BTPS R items assess more the presence of negative beliefs regarding treatment (e.g., Treatment is not what I expected Treatment does not seem to be working ) and absence of enabling factors ( e.g., Crises at home make it hard for me to ge t to a session My job gets in the way of coming to a session ) Relations between level of barriers at the last session attended
109 and attrition from treatment suggest that these factors relate not only to treatment access but al so to treatment terminatio n when predisposing and enabling factors are absent. Additionally, the decrease in barriers over time suggests that predisposing and enabling factors change over the course of treatment. ment and barriers to treatment cont inuation further support the importance of beliefs as a predisposing factor for continued engagement in services Results indicating that the level of perceived barriers is related both to the expectations that mothers st art treatment with and to how their expectations ch remain important throughout the treatment process, from initiation of services through termination. Limitations of the Study and Future Dir ections This study was designed to examine whether and how parent perceptions of barriers to treatment continuation change over the course of treatment. Several aspects of the study design necessitated a hierarchical linear modeling approach, including th e longitudinal study question and the unbalanced nature of the data. A statistical tool that would accommodate unbalanced data due to the unequal number of occasions per mother child dyad and missing data was imperative. Hierarchical linear modeling was se lected because of its ability to accommodate unbalanced study design (Fitzmaurice, Laird, & Ware, 2004). S eve ral assumptions inherent to HLM such as large sample size were challenging to meet and constitute limitations for this study. Meeting Assumptions for Hierarchical Linear Modeling This study included 46 mother child dyads and an average of approximately 16 occasions per dyad. Use of HLM necessitates large samples to provide adequate power
110 for the numerous parameters that are estimated. In multilev el studies, sample size at the group level is generally more problematic than sample size at the occasion level, because group level sample size is always smaller than occasion level sample size (Maas & Hox, 2005). Additionally, group level sample size dir ectly influences the ability to detect random effects between group members. Simulation studies generally indicate that power is greater when there is a greater number of groups (i.e., Level 2 units) and fewer cases per group (i.e., Level 1 units) (Tabachn ik & Fidell, 2007). Simulation studies suggested that a minimum of 30 groups with 30 occasions per group were necessary for adequate power (Bassiri, 1988; Kim, 1990; as cited in Hoffmann, 1997). Kreft and De Leeuw (1998; as cited in Field, 2009) generally recommend more than 20 groups Despite the lac k of consensus regarding minimally sufficient sample size for accurate estimation, our review of the HLM literature suggests that the th determining adequate sample size for HLM is correct However, a larger number of within group occasions appears to attenuate the effect of a limited number of Level 2 units (Hertzog, Lindenberger, Ghisletta, & Von Oertzen, 200 6; Maas & Hox, 2005). The ratio of parameters (i.e., predictors) estimated (23 total) to number of groups (46 participants) in this study substantially limited the pow er available to measure between group differences reliably. It is possible that some re sults were nonsignificant or only at trend level due to insufficient power for the number of parameters estimated. Despite the power limitations of this study, the results still offer valuable information regarding change in perceived barriers as well as f actors related to the experience of barriers in PCIT. The significant findings in this study may be considered valid. Future
111 investigation of the study que stions with a larger sample will permit further analysis of differences in the level and rate of change of barriers between participants as well as offer more reliable measurement of non significance versus significance. HLM also requires that variables are normally distributed. Because of the leptokurtic distribution of perceived ba rriers and chan ge in expectations/ credibility variable, transformation of the data with the Blom approach (Blom, 1958) was necessary. The Blom approach scales the data in z rank order. How ever, differences in scaling o f the variables in the model limited interpretation of effect size. Thus, this study focused on change in the variance of barriers to treatment continuation explained by the predictors and significance of the effect. In addition to being non normally distr ibuted, the barriers measure had a restricted range of scores in this study. The average score on the BTPS R was slightly lower than the average score reported by Danko and Budd (2009) in their assessment of parent perceived barriers to PCIT in a community mental health clinic. It is not clear why mothers reported such low levels of barriers to treatment, although there are several possible explanations for the low scores. First, m others may have perceived themselves as truly experiencing few barriers. Seco nd, mothers could have been hesitant to disclose concerns about treatment process issues despite the steps that were taken to ensure th ey believed their responses would remain confidential. Third, because of the frequent administrati on of the BTPS R questi onnaire, the mothers may have put less thought into their responses each week. Finally, administration of the measure in the
112 questionnaire items, if they were in a hurry to leave. I t will be important in future research to consider whether change in barriers is more optimally measured on a biweekly basis, at a different time point in the session, and in a manner that promotes greater anonymity. Most of th e missing data in this st udy were due to drop out from treatment and were accounted for by controlling for attrition in the HLM. However, two other sources of missing data were not addressed in the multi level model therapist error (i.e., the therapist did not administer the for m to the parent) and missing questionnaires. Of the 586 occasions of measurement in this study, 10 occasions of measurement of barriers to treatment continuation were missing because of missing forms, and two occasions of measurement were missing because o f therapist error in failing to administer the measure Missing data due to missing forms and therapist error is relatively minor when considering the number of occasions of measurement in this study. However, it will be important for future studies to ree valuate protocol for administration of study measures to minimize missing forms and therapist error. Generalizability of Study Results Generalizability of the findings to the general clinical population is limited for several reasons. First, for weekly administration of a barriers measure to be useful for predicting and preventing dropout in community based mental health centers, therapists would likely be the data collector or at least would need to be aware of participant responses on the measu re of perceived barriers. Parents responses would be known to their therapist could compromise their comfort in disclosing their responses o n scale items related to patient therapist relationship factors
113 measure in a manner that maintains the utility of the information a nd patient comfort in disclosure c alls for new approaches to data collec tion, such as online surveys or partneri ng with other clinic staff or clinic administrators. Another threat to generalization of study results is that only maternal ratings for the study outcomes and predictors were i nvestigated. F ather ratings were not included because there were too few fathers participating in treatment for independent analysis. Additionally, data from two parents of the same child could not be included because the observations from the mother and f ather would not be independent of each other. s into the struggles and barri ers experienced by the family are valuable, ratings by an additional informant familiar with the family, such as the therapist, would provide an additional unique perspective into both level and change in barriers over the course of treatment (Kazdin, Holland, & Crowley, 1997). Generalizability is also limited because of the limited racial ethnic diversity of the sample, largely due to the geographic locat ion of the study (i.e., rural and suburban communities in North Central Florida). The restriction on number of predictors per sample size also precluded the investigation of race/ ethnicity as a predictor. It is possible that families from different geogra phic locations, such as urban municipalities, would show different levels and patterns of treatment barriers due to differences in environmental demands and access to treatment (Cambell, Kearns, & Patchin, 2006). Level and change in barriers to treatment c ontinuation might also vary based on the setting in which service s are delivered. Mother child dyads in this study were participants in a research study, in which state of the art equipment was available and monitoring of integrity of treatment administrat ion was continuous. As research
114 participants, mothers in this study also received treatment for their child behavior at no cost, as well as free child care for siblings, a $5 .00 gas card, and a parking voucher at ea ch attended session. These cost saving provisions are rarely offered in community treatment settings and could further limit generalizability of the findings. Treatment continuation was also incentivized in that mothers were aware at the start of treatment that they would rece ive payment for completion of post treatment assessments. It is possible that this incentive to complete treatment may be related to the lower rate of attrition in this study. Past research suggests that even when no cost treatment and both monetary and l ogistic supports are provided SES is still the greatest predictor of attrition from PCIT (Fernandez & Eyberg, 2008). A lthough it is unlikely that these cost saving measures substantially affected attrition in this study, families may have perceived fewer barriers to treatment continuation as a result of the logistical supports provided, such as childcare for siblings and free treatment. Measurement of Treatment Barriers There were also limitations related to the measurement of perceived barriers To incre ase the feasibility of weekly assessment of barriers, this study used the revised BTPS R (BTPS R; Colonna Pydyn et al., 2007), which is half the length of the original BTPS (Kazdin, Holland, Crowley, & Breton, 1997). The decreased length of the BTPS R limi ted the potential range of barriers reported by parents. The weekly administration of the barriers measure in this study as compared to administration only at the end of treatment in past studies also may have influenced the manner in which barriers were
115 reflecting on the entire treatment experience would differ from how barriers are perceived on a week to week basis. Both the BTPS R and CEQ P contain factors measuring d ifferent aspects of the targeted construct s The BTPS R contains two factors, expectations for treatment and scale items on this factor primarily relate to parent beliefs a bout treatment and the therapeutic relationship (e.g. reatment expectations and beliefs in the credibility of treatment. Because of the limited power and high number of parameters in this study, the individual factors on the BTPS R and CEQ P were not evaluated separately. Future evaluation of differences betw een factors on the BTPS R will offer insight into differences in how external demands versus parent beliefs about treatment may differ in their change over time and prediction of attrition. Similar evaluation of the CEQ P factors will provide perspective i nto how expectations and beliefs regarding treatment credibility may differ in predicting level and change in barriers. The Parent Therapist Relationship and Future Research on Treatment Barriers The importance of relationship factors in family based beha vioral treatment programs has been highlighted by research investigating relations between parent therapist process variables and treatment attrition. In one study, therapist verbal behavior early in treatment was predictive of completion status for 73% of participating families (Harwoood & Eyberg, 2004). Prinz and Miller (1994) found that families who received an enhanced version of a family based treatment for child disruptive behavior,
116 which included time to discuss non child related stressors, showed si gnificantly less attrition from treatment as compared to participants in the standard family treatment. It is important to note, however, that families that dropped out from either treatment condition most often attributed their dropout to environmental or logistical obstacles, factors likely related to SES (Prinz & Miller, 1994). Thus, attending to process in therapy and the relationship between therapist and parent may attenuate drop out in treatment, while logistical and environmental demands remain a ba rrier to treatment continuation for some families. Future investigation of change in the two subscales of the BTPS R overall perceptions of barriers, the therapeutic relationship, and external demands. After investigation of each variable in the model, there remained significant variance not accounted for by the hypothesized predictors, both between parent child dyads and across treatment occasions. It is plausible w ith a larger sample size, which would facilitate measurement of random effects, that differences between parent child dyads in perceived barriers to treatment continuation would be better explained. However, questions remain regarding the source of varianc e between occasions in barriers to treatment continuation. Given past research in PCIT documenting that pre treatment maternal verbalizations to their child an indication of parenting style, were predictive of attrition (Fernandez & Eyberg, 2009), the mat ch between parent beliefs about parenting and the approach to parenting espoused in treatment may be an impor tant determinant of the change in parent receptivity to treatment and perceived barriers to treatment over time. Additionally, inclusion of thera pist verbal behaviors as a pre dictor of change in barriers may differentiate between families for whom the
117 therapist parent relationship is more important and between time points in treatment when communication is especially critical. There are several n otable strengths of this study including the measurement of perceived barriers for all treatment participants regardless of completion status, the identification of change in perceived barriers over the course of PCIT, and the recognition of moderators of this change. Results add to existing understanding of barriers to treatment continuation by showing that perceived barriers change over the course of treatment and that fluctuations in child disruptive behavior relate to this change. Still, the significant variance remaining to be explained in this study suggests that there are additional predictors of barriers t hat have not yet been identified Further investigation into barriers to treatment continuation is necessary to increase understanding of relations between treatment barriers, its predictors, and its relationship with treatment outcome.
118 LIST OF REFERENCES Achenbach, T.M., & Rescorla, L.A. (2000). Manual for ASEBA preschool forms & profiles Child Behavior Checklist for ages 6 18 Burlington, VT: University of Vermont, Research Center for Children, Youth, & Families. Achenbach, T. M., & Rescorla, L. A. (2001). Manual of the ASEBA school age forms & profile Child Behavior Checklist for ages 1.5 5 Burlington, VT: University of Vermont, Research Center for Children, Youth, & Families. Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on Automatic Control, 19 716 723. doi:10.1109/TAC.1974.1100705 Andersen, R.M. (1968). Behavioral model of Series, 25 Chicago, IL: University of Chicago, Center for Health Administration Studies. Andersen, R.M. (1995). Revisiting the behavioral model and access to medical care: Does it matter? Journal of Health a nd Social Behavior 36, 1 10. doi:10.2307/2137284 Armbruster, P. & Schwab Stone, M.E. (1994). Sociodemographic characteristics of dropouts from a child guidance clinic. Hospital and Community Psychiatry, 45 804 808. Retrieved from http://ps.psychiat ryonline.org/ Arnkoff, D. B., Glass, C. R., & Shapiro, S. J. (2002). Expectations and preferences. In J. C. Norcross (Ed.), Psychotherapy relationships that work: Therapists contributions and responsiveness to patient (pp. 325 346). New York, NY: Oxford University Press. (1994). Pre treatment predictors of treatment outcome in panic disorder and agoraphobia treated with alprazolam a nd exposure. Journal of Affective Disorders, 30 123 132. doi:10.1016/0165 0327(94)90040 X in home services: An expanded conceptual framework. Journal of Health and Social Behavior, 28, 184 196. doi:10.2307/2137131 Blom, G. (1958). Statistical estimates and transformed beta variables New York, NY: John Wiley. Boggs, S.R., Eyberg, S.M., Edwards, D., Rayfield, A., Jacobs, J., Bagner, D., & Hood, K. (2004). Outcom es of parent child interaction therapy: A comparison of dropouts and treatment completers one to three years after treatment. Child & Family Behavior Therapy 26(4), 1 22. doi: 10.1300/J019v26n04_01
119 Bonner, B.L., & Everett, F.L. (1986). Influence of clien t preparation and problem severity on attitudes and expectations in child psychotherapy. Professional Psychology: Research and Practice 17 223 229. doi:10.1037/0735 7028.17.3.223 Bradley, R.H., & Corwyn, R.F. (2002). Socioeconomic status and child de velopment. Annual Review of Psychology, 53, 371 399. doi:10.1146/annurev.psych. 53.100901.135233 Bradley, I., Poser, E. G., & Johnson, J. A. (1980). Outcome expectation ratings as predictors of success in weight reduction. Journal of Clinical Psychology 36 500 502. Retrieved from http://onlinelibrary.wiley.com/journal/10.1002/ %28ISSN%291097 4679 Brooks Gunn J, Klebanov PK, & Liaw F. (1995). The learning, physical, and emotional environment of the home in the context of poverty: The Infant Health and Development Program. Children and Youth Services Review, 17, 251 76. doi:10.1016/0190 7409(95)00011 Z Bryk, A.S., & Raudenbush, S.W. (1987). Application of hierarchical linear models to assess ing change. Psychological Bulletin, 101 147 158. doi:10.1037/0033 2909.101.1.147 Bryk, A.S. & Raudenbush, S.W. (1992). Hierarchical linear models for social and behavioral research: Applications and data analysis methods Newbury Park, CA: Sage. Budd, K.S., Danko, C., & Legato, L. (2010, August). Barriers to treatment completion in community based PCIT. Inside the black box: Treatment process for youth externalizing behaviors. Symposium conducted at the meeting of the American Psychological Association San Diego, CA. Callahan, C.L., & Eyberg, S.M. (2010). Relations between parenting behavior and SES in a clinical sample: Validity of SES Measures. Child and Family Behavior Therapy, 32, 125 138. doi: 10.1080/07317101003776456 Cambell, C.D., Kearns, L.A., & Patchin, S. (2006). Psychological needs and resources as perceived by rural and urban psychologists. Professional Psychology: Research and Practice, 37 45 50. doi: 10.1037/0735 7028.37.1.45 Cohen, J. (1988). Statistical power analysis for th e behavioral sciences (2nd ed.). Hillsdale. NJ: Erlbaum. Colonna Pydyn, C., Gjesfjeld, C.D., & Greeno, C. (2007). The factor structure of the Barriers to Treatment Participation Scale (BTPS): Implications for future barriers scale development. Administration and Policy in Mental Health, 34, 563 569. doi: 10.1007/s10488 007 0139 6
120 Conners, C.K., Sitarenios, G., Parker, J.D., & Epstein, J.N. (1998). Revision and restandardization of the Conners Teacher Rating Scale (CTRS R): Factor structure, r eliability, and criterion validity. Journal of Abnormal Child Psychology, 26 279 291. doi:10.1023/A:1022606501530 Costello, E.J., Egger, H., & Angold, A. (2005). 10 year research update review: The epidemiology of child and adolescent psychiatric disord ers: I. methods and public health burden. Journal of the American Academy of Child & Adolescent Psychiatry 44 972 986. doi:10.1097/01.chi.0000172552.41596.6f Counte, M.A., & Glandon, G.L. (1991). A panel study of life stress, social support, and the health services utilization of older persons. Medical Care 29 348 361. doi:10.1097/00005650 199104000 00004 Danko, C., & Budd, K.S. (2009, October). Characteristics of PCIT completers versus dropouts in a community mental health setting. Poster sessi on presented at the National Parent Child Interaction Therapy Conference, Traverse City, MI. Day, L., & Reznikoff, M. (1980). Preparation of children and parents for treatment at a Journal of Co nsulting and Clinical Psychology 48, 303 304. doi:10.1037/0022 006X.48.2.303 Devilly, G.J. & Borkovec, T.D. (2000). Psychometric properties of the credibility/ expectancy questionnaire. Journal of Behavior Therapy and Experimental Psychiatry 31 73 8 6. doi:10.1016/S0005 7916(00)00012 4 Dodrill, C.B. (1981). An economical method for the evaluation of general intelligence in adults. Journal of Consulting & Clinical Psychology 49, 668 673. d oi:10.1037/0022 006X.49.5.668 Dunn, L.M. & Dunn, L.M. (1997) Examiner's manual for the PPVT 3: Peabody Picture Vocabulary Test, Third Edition. Circle Pines, MN: American Guidance Services. Ensminger, M. E., & Fothergill, K. E. (2003). A decade of measuring SES: What it tells us and where to go from here. In M. H. Bornstein, & R. H. Bradley (Eds.), Socioeconomic status, parenting, and child development (pp. 13 27). Mahwah,NJ: Erlbaum. Eyberg, S.M. (2007, November). Final report on the PCIT maintenance treatment project. Paper p resented at the ABCT Preconference on Social Learning and the Family, Philadelphia. Eyberg, S.M., & Pincus, D. (1999). Eyberg Child Behavior Inventory and Sutter Eyberg Student Behavior Inventory: Professional manual Odessa, FL: Psychological Assessment Resources.
121 Fernandez, M.A., & Eyberg, S.M. (2009). Predicting treatment and follow up attrition in parent child interaction therapy. Journal of Abnormal Child Psychology, 37 431 441. doi:10.1007/s10802 008 9281 1 Field, A. (2009). Discovering statistics using SPSS, third edition SAGE Publications Inc., Thousand Oaks, CA. Fitzmaurice, G.M., Laird, N.M., & Ware, J.H. (2004). Applied longitudinal analysis Hoboken, NJ: Wiley Interscience. Funderburk, B.W., Eyberg, S.M., Rich, B., & Behar, L. (2003). Further psychometric evaluation of the Eyberg and Behar rating scales for parents and teachers of preschoolers. Early Education & Development 14 67 81. doi:10.1207/s15566935eed1401_5 Garcia, J. A., & Weisz, J. R. (2002). When youth mental healt h care stops: Therapeutic relationship problems and other reasons for ending youth outpatient treatment. Journal of Consulting and Clinical Psychology 70, 439 443. doi:10.1037/0022 006X.70.2.439 Goldstein, A. P. (1962a). Therapist patient expectancies in psychotherapy New York, NY: Pergamon Press. Goldstein, A. P. (1962b). Participant expectancies in psychotherapy. Psychiatry 25, 72 79. Gould, M.S., Shaffer, D., & Kaplan, D. (1985). The characteristics of dropouts from a child psychiatry clinic. Journal of the American Academy of Child Psychiatry, 24 316 328. doi:10.1016/S0002 7138(09)61093 1 Greenberg, R.P., Constantino, M.J., & Bruce, N. (2006). Are patient expectations still relevant for psychotherapy process and outcome? Clinical Psycholo gy Review 26 657 678. doi:10.1016/j.cpr.2005.03.002 Guevara, J. P., Mandell, D. S., Rostain, A. L., Zhao, H., & Hadley, T. R. (2003). National estimates of health services expenditures for children with behavioral disorders: An analysis of the medica l expenditure panel survey. Pediatrics, 112 e440 e446. doi:10.1542/peds.112.6.e440 Harwood, M., & Eyberg, S.M. (2004). Therapist verbal behavior early in treatment: Relation to successful completion of parent child interaction therapy. Journal of Clinic al Child and Adolescent Psychology, 33, 601 612. doi:10.1207/s15374424jccp3303_17
122 Hedeker, D., & Gibbons, R.D. (1997). Application of random effects pattern mixture models for missing data in longitudinal studies. Psychological Methods, 2, 64 78. doi :10.1037/1082 989X.2.1.64 Hertzog, C., Lindengerger, U., Ghisletta, P., & Von Oertzen, T. (2006). On the power of multivariate latent growth curve models to detect correlated change Psychological Methods, 11, 244 252 doi:10.1037/1082 989X.11.3.244 Hoffman, L.W. (1984). Work, family, and the socialization of the child. In R. D. Parke (Ed.), The family: Review of child development research (Vol. 7, pp. 223 282). Chicago, IL: University of Chicago Press. Hofmann, D. A. (1997). An overview of the log ic and rationale of hierarchical linear models. Journal of Management 23 723 744. doi:10.1177/014920639702300602 Hollingshead, A. B. (1975). Four factor index of social status New Haven, CT: Yale University Press. Howell, E. (2004). SCHIP Washington, D.C.: The Urban Institute. Janz, N.K., Champion, V.L., & Strecher, V.J. (2002). The health belief model. In K. Glanz, B.K. Rimer, & F.M. Lewis (Eds.), Health behavior and health education: Theory, research, and practice (pp. 45 66). San Fransisco: Jossey Bass. Kazdin, A.E. (1979). Therapy outcome questions requiring control of credibility and treatment generated expectancies. Behavior Therapy 10 81 93. doi:10.1016/ S0005 7894(79)80011 8 Kazdin, A.E. (1995). Conduct disorders in childhood and adolescence (2 nd ed.). Thousand Oaks, CA: Sage. Kazdin, A.E. (2000). Perceived barriers to treatment participation and treatment Acceptability among antisocial children and their families. Journal of Child and Family Studies 9, 157 174. doi: 1 0.1023/A:1009414904228 Kazdin, A. E. (2003). Problem solving skills training and parent management training for conduct disorder. In A. E. Kazdin & J. R. Weisz (Eds.), Evidence base d psychotherapies for children and adolescents (pp. 241 262). New York, NY: Guilford. Kazdin, A.E., Holland, L., & Crowley, M. (1997). Family experience of barriers to treatment and premature termination from child therapy. Journal of Consulting and Clin ical Psychology 65 453 463. doi: 10.1037/0022 006X.65.3.453
123 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 8, 1051 1062. doi: 10.1111/j.1469 7610.1997.tb01621.x 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 1 6. doi:10.1207/s15374424jccp2201_1 Kazdin, A.E., Mazurick, J.L., & Siegel, T.C. (1994). Treatment outcome among children with externalizing disorders who terminate prematurely versus those who complete psychotherapy. Journal of the American Academy of Ch ild and Adolescent Psychiatry 33 549 557. doi:10.1097/00004583 199405000 00013 Kazdin, A.E., Siegel, T., & Bass, D. (1990). Drawing upon clinical practice to inform research on child and adolescent psychotherapy. Professional Psychology : Research an d Practice, 21, 189 198. doi:10.1037/0735 7028.21.3.189 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. doi:10.1037/0735 7028.29.4.332 Kolb, D. L., Beutler, L. E., Davis, C. S., Crago, M., & Shanfield, S. (1985). Patient and therapy process variables relating to dropout and change in psychotherapy. P ychotherapy, 22, 702 710. doi:10.1037/h0085556 Lambert, E. W., Wahler, R. G., Andrade, A. R., & Bickman, L. (2001). Looking for the disorder in conduct disorder. Journal of Abnormal Psychology 110, 110 123. doi:10.1037/0021 843X.110.1.110 Little, R.J.A. (1993). Pattern mixture models for multivariate incomp lete data. Journal of the American Statistical Association 88 125 134. doi:10.2307/2290705 Loeber, R., & Dishion, T.J. (1983). Early predictors of male delinquency: A review. Psychological Bulletin 94 68 99. doi:10.1037/0033 2909.94.1.68 McCabe, K.M. (2002). Factors That Predict Premature Termination Among Mexican American Children in Outpatient Psychotherapy. Journal of Child and Family Studies 11 347 359. doi:10.1023/A:1016876224388 Newacheck, P. W. (1992). Characteristics of child ren with high and low usage of physician services. Medical Care 30 30 42. doi:10.1097/00005650 199201000 00003
124 Noble, L.M., Douglas, B.C., & Newman, S.P. (2001). What do patients expect of psychological services? A systematic and critical review of empirical studies. Social Science and Medicine 52 985 998. doi:10.1016/S0277 9536(00)00210 0 Nock, M.K., & Ferriter, C. (2005). Parent management of attendance and adherence in child and adolescent therapy: A conceptual and empirical review. Clinica l Child and Family Psychology Review 8 149 166. doi: 10.1007/s10567 005 4753 0 Nock, M.K., Ferriter, C., & Holmberg, E. (2007). Parent beliefs about treatment credibility and effectiveness: Assessment and relation to subsequent treatment participatio n. Journal of Child and Family Studies 16 27 38. doi:10.1007/s10826 006 9064 7 Nock, M.K. & Kazdin, A.E. (2001). Parent expectancies for child therapy: Assessment and relation to participation in treatment. Journal of Child and Family Studies 10, 15 5 180. doi:10.1023/A:1016699424731 Nock, M. K., & Kazdin, A. E. (2002). Parent directed physical aggression by clinic referred youths. Journal of Clinical Child and Adolescent Psychology 31, 193 205. doi: 10.1207/S15374424JCCP3102_05 Pedhazur, E.J. ( 1997). Multiple regression in behavioral research: Explanation and prediction 3 rd edition. Orlando: Harcourt Brace College Publishers. Perez, J.C. (2008). Predictors of patterns of change in child disruptive behavior and parenting stress during parent child interaction therapy and its relation to treatment outcome (Unpublished doctoral dissertation). University of Florida, Gainesville, FL. Piper, W. E., & Wogan, M. (1970). Placebo effect in psychotherapy: An extension of earlier findings. Journal of Consulting and Clinical Psychology 34, 447. doi:10.1037/h0029345 psychiatric service. Journal of Clinical Psychology 40, 372 377. doi:10.1002/1097 4679(198401)40:1< 372::AID JCLP2270400169>3.0.CO;2 Q Prinz, R.J., & Miller, G.E. (1994). Family based treatment for childhood antisocial behavior: Experimental influences on dropout and engagement. Journal of Consulting and Clinical Psychology 62 645 650. doi:10.1037/0022 006X.62.3.645 Ringel, J., & Sturm, R. (2001). National estimates of mental health utilization and expenditure for children in 1998. Journal of Behavioral Health Services & Research 28 319 332. doi:10.1007/BF02287247
125 Rogosa, D. (1995 supplemental questions. In J.M. Gottman (Ed.), The analysis of change (pp. 3 66). New Jersey: Erlbaum. Rosenstock, I.M. (1974). Historical origins of the health belief model. Health Education Monographs, 2 328 335. Rosnau, P.V. (1994). Reflections on the cost consequences of the new gene technology for health policy. International Journal of Technology Assessment in Health Care 10 546 561. doi:10.1017/S0266462300008151 Rubin, D.B (1976). Inference and missing data. Biometrika 63 581 592. doi:10.1093/biomet/63.3.581 relationship to pretreatment symptomatology and outcome of cognitive beh avioral group treatment for social phobia. Journal of Consulting and Clinical Psychology 65, 694 698. doi:10.1037/0022 006X.65.4.694 Santisteban, D.A., Szapocznik, J., Perez Vidal, A., Kurtines, W.H., Murray, E.J., & LaPerriere, A. (1996). Efficacy of i ntervention for engaging youth and families into treatment and some variables that may contribute to differential effectiveness. Journal of Family Psychology 10 35 44. doi:10.1037/0893 322.214.171.124 Schafer, J.L., & Graham, J.W. (2002). Missing data: Our view of the state of the art. Psychological Methods 7 147 177. doi:10.1037/1082 989X.7.2.147 Schuhmann, E.M., Foote, R., Eyberg, S.M., Boggs, S., & Algina, J. (1998). Parent child interaction therapy: Interim report of a randomized trial with sh ort term maintenance. Journal of Clinical Child Psychology, 27 34 45. doi:10.1207/s15374424jccp2701_4 Shaffer, D., Fisher, P., & Lucas, C.P. (1998). NIMH DISC IV. Diagnostic interview schedule for children, parent informant New York, NY: Columbia Unive rsity. Shaffer, D., Fisher, P., Lucas, C.P., Dulcan, M.K., & Schwab Stone, M.E. (2000). NIMH Diagnostic Interview Schedule for Children Version IV (NIMH DISC IV): Description, differences from previous versions, and reliability of some common diagnoses. Journal of the American Academy of Child & Adolescent Psychiatry 39 28 38. doi:10.1097/00004583 200001000 00014 Schwarz, G. (1978). Estimating the dimension of a model. The Annals of Statistics 6, 461 464. doi:10.1214/aos/1176344136
126 Shieh, Y., & Fouladi,R.T. (2003). The effect of multicollinearity on multilevel modeling parameter estimates and standard errors. Educational and Psychological Measurement, 63, 951 985. doi: 10.1177/0013164403258402 Shuman, A.L., & Shapiro, J.P. (2002). The effec ts of preparing parents for child psychotherapy on accuracy of expectations and treatment attendance. Community Mental Health Journal 38 3 16. doi:10.1023/A:1013908629870 Singer, J. D., & Willett, J. B. (2003). Applied longitudinal data analysis: Mod eling change and event occurrence New York, NY: Oxford University Press. Smith, T.E., & Graham, P.B. (1995). Socioeconomic stratification in family research. Journal of Marriage and the Family, 57 930 940. doi:10.2307/353413 Snijders, T., & Bosker, R (1999). Multilevel analysis: An introduction to basic and advanced multilevel modeling Thousand Oaks, CA: Sage. Sotsky, S. M., Glass, D. E., Shea, M. T., Pilkonis, P. A., Collins, J. R., Elkin, I., . Moyer, J. (1991). Patient predictors of respon se to psychotherapy and pharmacotherapy: Findings in the NIMH Treatment of Depression Collaborative Research Program. American Journal of Psychiatry 148 997 1008. Retrieved from http://ajp.psychiatryonline.org/ Stevens, J., Kelleher, K.J., Ward Estes, J ., & Hayes, J. (2006). Perceived barriers to treatment and psychotherapy attendance in child community mental health centers. Community Mental Health Journal 42 449 458. doi:10.1007/s10597 006 9048 5 Strong, R.A., Lucas, C.P., & Lucas, L.J. ( 2006). The Young Child Diagnostic Interview Schedule for Children [Computer Software]. New York, NY: New York University School of Medicine. Tabachnick, B. G., & Fidell, L. S. (2007). Using multivariate statistics, 5th ed. Boston: Allyn and Bacon. Valenzuela M. (1997). Maternal sensitivity in a developing society: The context of urban poverty and infant chronic undernutrition. Developmental Psychology, 33 845 855. doi:10.1037/0012 16126.96.36.1995 Weinberger, J., & Eig, A. (1999). Expectancies: The ignored common factor in psychotherapy. In I. Kirsch (Ed.), How expectancies shape experience (pp. 357 382). Washington, DC: American Psychological Association. Werba, B.E., Eyberg, S.M., Boggs, S., & Algina, J. (2006). Predicting outcome in parent child intera ction therapy: Success and attrition. Behavior Modification, 30 618 646. doi:10.1177/0145445504272977
127 White, J.L., Moffitt, T.E., Earls, E., Robins, L., & Silva, P.A. (1990). How early can we tell? Predictors of childhood conduct disorder and adolesce nt delinquency. Criminology 28 507 533. doi:10.1111/j.1745 9125.1990.tb01337.x Wierzbicki, M., & Pekarik, G. (1993). A meta analysis of psychotherapy dropout. Professional Psychology: Research and Practice 24, 190 195. doi:10.1111/j.1745 9125.1990 .tb01337.x Wilens, T.E., Biederman, J., Brown, S., Tanguay, S., Monuteaux, M., Blake, C., & Spencer, T.J. (2002). Psychiatric comorbidity and functioning in clinically referred preschool children and school age youths with ADHD. Journal of the American A cademy of Child & Adolescent Psychiatry, 41, 262 268. doi:10.1097/00004583 200203000 00005 Wu,Y. W., Clopper, R. R., & Wooldridge, P. J. (1999). A comparison of traditional approaches to hierarchical linear modeling when analyzing longitudinal data. R esearch in Nursing and Health, 22 421 432. doi:10.1002/(SICI)1098 240X(199910)22:5<421::AID NUR8>3.0.CO;2 Q Zisser, A., & Eyberg, S.M. (2010). Treating oppositional behavior in children using parent child interaction therapy. In J.R. Weisz & A.E. Kazdin (Eds.) Evidence based psychotherapies for children and adolescents ( 2nd ed ., pp. 179 193). New York, NY: Guilford.
128 BIOGRAPHICAL SKETCH Alison Rebecca Zisser was born and raised in Jacksonville, Florida. She earned her Bachelor of Arts degree in psychology and history and graduated magna cum laude in 2005 from Washington University in St. Louis. While a student at Washington University in St. Louis, Alison conducted an honors thesis under the mentorship of Alan Lambert, Ph.D. After receiving her u ndergraduate degree, Alison spent five months in and conducting research on AIDS prevention. After returning to the United States, Alison worked for six months as an ear ly intervention therapist with developmentally delayed children. In August, 2006, Alison enrolled in a dual Master of Science and Health Psychology and was a research as sistant in the Child Study Laboratory under the mentorshi p of Sheila Eyberg, Ph.D., ABPP, during her graduate studies. research focused on parental factors affecting child mental health treatment, such as parent psychopathology, expectations for t reatment, and perceived barriers. Alison completed She received her Ph.D. from the University of Florida in the summer of 2011.