The Prediction of Dropout in Secure Residential Youth Care: Why We Should Distinguish Between Different Types of Dropout
Masterthesis Forensische Orthopedagogiek Graduate School of Child and Education, University of Amsterdam GGzE Onderzoeksgroep Forensische Geestelijke Gezondheidszorg Lindy C. P. van Gerwen [10673431] First supervisor: Dr. Marc J. Noom Second supervisor: Drs. Trudy van der Stouwe Tilburg, August 2016
Acknowledgements
I would like to give thanks to dr. Marc Noom, my supervisor from the University of Amsterdam and GGzE Centre for Child and Adolescence Psychiatry, for his guidance and useful critiques which helped me to improve my thesis. I would also like to give thanks to the staff of the Research Group Forensic Mental Health Care for sharing their knowledge, and especially dr. Ilja Bongers for guiding me through the process of this research. Finally, I would like to give thanks to my family for their support and my sister for her guidance and useful critiques.
Abstract
Aim: The current study identifies predictors of different types of dropout (self-initiated dropout, step-out, and push-out) in secure residential youth care. Method: The sample contained 247 juveniles that were admitted to De Catamaran, hospital for youth forensic psychiatry and orthopsychiatry. Logistic regression analyses with either the Child Behavior Checklist (CBCL), Youth Self-Report (YSR) or Structured Assessment of Violence Risk in Youth (SAVRY) identified predictors of dropout in general. Subsequently, multinomial logistic regression analyses using the same instruments identified predictors of different types of dropout. These predictors were further specified with post-hoc analyses. Results: The presence of historical and individual risk factors make it more likely that juveniles drop out of treatment than to complete it. However, other relationships were found when looking at different types of dropout. First, historical risk factors increase the chance of self-initiated dropout than treatment completion, while emotional problems and self-reported somatic complaints decrease this chance. Second, the presence of individual risk factors makes it more likely that juveniles end the treatment as a step-out than as a completer, while the protective factors of the SAVRY decrease this chance. Third, both historical and individual risk factors contribute to a higher chance of push-out than treatment completion. And again, protective factors of the SAVRY decrease this chance. Conclusion: The current study provides empirical support for the distinction between different types of dropout. This knowledge can not only help recognize the risks of different types of dropout, but can also help prevent different types of dropout.
The Prediction of Dropout in Secure Residential Youth Care
Residential treatment can contribute to a positive development of juveniles with severe behavioral and emotional problems (Knorth, Harder, Zandberg, & Kendrick, 2008).
Unfortunately, a significant amount of juveniles end their treatment prematurely, a
phenomenon also known as dropout1 (De Haan, Boon, De Jong, Hoeve, & Vermeiren, 2013; Johnson, Mellor, & Brann, 2008). Treatment dropout can have negative consequences, such as a delay in psychosocial development, an increase of behavioral and emotional problems, and engagement in delinquent activities (De Haan et al., 2013; Van der Molen et al., 2015). In order to prevent these negative consequences, it is important to identify factors that can
predict treatment dropout at an early stage. Therefore, the aim of the present study was to gain insight into predictors of treatment dropout in secure residential youth care.
The prediction of treatment dropout has been the focus of ample studies (e.g. De Haan et al., 2013, Eisengart, Martinovich, & Lyons, 2008; Knorth et al., 20082). Multiple
researchers found that the presence of behavioral problems can increase the risk of treatment dropout (De Haan et al., 2013; Eisengart et al., 2008; Friars & Mellor, 2007; Johnson et al., 2008; Piotrkowski & Baker, 2004; Sunseri, 2003). According to Sunseri (2003), this relationship might be explained with the deviancy training hypothesis. This hypothesis suggests that antisocial or high risk juveniles are particularly vulnerable to aggravation of peers. When these juveniles are placed together, they might stimulate each other’s antisocial and delinquent behavior, such as running away from the facility (Bayer, Hjalmarsson, & Pozen, 2009; Sunseri, 2003). If these behaviors occur, treatment dropout increases (Sunseri, 2003). More specific behavioral problems that have been identified as risk factors for
1 In the current study a distinction is made between three types of treatment dropout: self-initiated dropout (premature ending of the treatment at initiative of the juvenile), step-out (mutual agreement between juvenile and treatment staff to end the treatment prematurely), and push-out (premature ending of the treatment at initiative of the treatment staff).
treatment dropout are substance abuse and aggressive behavior (Eisengart et al., 2008; Harder et al., 2013; Knorth, Klomp, Van den Bergh, & Noom, 2007; McIntosh, Lyons, & Weiner, 2010). For juveniles with substance related problems it is assumed that they seek easier access to substances by running away from the facility. Therefore, these juveniles are more likely to end the treatment themselves (Eisengart et al., 2008). For aggressive juveniles the treatment is more often ended by the treatment staff, possibly because the aggressive behavior is
threatening to others and can evoke feelings of anger, fear or impotence in residential workers (Knorth et al., 2007). Although many research has shown that the existence of behavioral problems positively influences the chance of treatment dropout, it should also be mentioned that some researchers found that the presence of behavioral problems did not affect the chance of treatment dropout (Van der Ploeg & Scholte, 2003; Vos, 2008). Perhaps because these studies focused on dropout in residential youth care while others (e.g. Friars & Mellor, 2007) examined dropout in outpatient youth care.
Unfortunately, the existing literature contains three problems that hinder a clear view on the possible predictors of treatment dropout. The first problem is the inconsistency regarding emotional problems and its contribution to treatment dropout. Some researchers found that the presence of emotional problems can increase the risk of treatment dropout (Friar & Mellor, 2007; Johnson et al., 2008; Piotrkowski & Baker, 2004). On the other hand, Johnson and colleagues (2008) also found that anxiety problems can decrease the risk of treatment dropout. Their explanation for this result was that anxious patients might find it too daunting to leave the treatment or might be extra motivated to stay in treatment. Furthermore, other researchers found that the chance of dropout is not affected by the presence of emotional problems (Van der Ploeg & Scholte, 2003; Vos, 2008). According to De Haan and colleagues (2013) these differences might be caused by different target groups and designs.
A second problem that arises in the existing literature is that the concept of dropout is ambiguous. Johnson and colleagues (2008) concluded that most researchers use the number of sessions that is attended by the juvenile to distinguish between completers and dropouts. However, they found two disadvantages with this use. First, the cut-off score to distinguish between completers and dropouts is often arbitrary, and therefore differs in various studies. Second, the duration of the treatment is not necessary related to dropout. Another way to define treatment dropout is with the use of three different types, namely self-initiated dropout, step-out, and push-out. This definition is preferable, since it can be assumed that each type of dropout has its own risks and needs a different approach (Eisengart et al., 2008; Sunseri, 2003). For example, Johnson and colleagues (2008) found that one-third of the push-out juveniles had complex problems (i.e. having more than two diagnoses and problems within family, school, social competences, and/or life events) and one-third of the step-out juveniles had either complex problems or a depressive disorder. Researchers that focused on self-initiated juveniles found that these juveniles more often had severe behavioral problems or substance related problems (Eisengart et al., 2008; Sunseri, 2003). Although research has shown that each type of dropout has its own risks, research using these types is scarce.
A third problem that hinds a clear view on treatment dropout is that previous studies primarily focus on risk factors, instead of the combination of risk and protective factors. However, both risk and protective factors could help to better understand problem behavior (Lodewijks, De Ruiter, & Doreleijers, 2010; Lösel & Farrington, 2012; Pomp, 2009). For instance, with the inclusion of protective factors the amount of variance that is explained by risk factors increases (Lodewijks et al., 2010). In addition, protective factors can buffer risk factors and make it possible for juveniles to leave an antisocial or emotional disturbed
pathway (Lodewijks et al., 2010; Lösel & Farrington, 2012; Pomp, 2009). This desistance can be predicted by protective factors like a strong social support and strong attachment to
prosocial adults (Lodewijks et al., 2010). Furthermore, motivation can act as a protective factor since motivation is one of the most important client factors associated with treatment outcome. In other words, a better motivation has a positive effect on treatment outcome
(Karver, Handelsman, Fields, & Brickman, 2005, 2006; Knorth, Harder, & Kalverboer, 2012). Due to inconsistencies and ambiguities in the existing literature, the current study aims at identifying both risk and protective factors which influence the chance of treatment dropout within secure residential youth care. Based on prior research, four hypotheses are specified: (1) juveniles with behavioral problems, such as rule breaking behavior, are expected to have an increased risk of self-initiated dropout, (2) juveniles with higher levels of emotional problems, such as depression, are expected to have an increased risk of step-out, (3) juveniles with higher levels of behavioral problems, such as aggressive behavior, are expected to have an increased risk of push-out, and (4) juveniles with protective factors, such as a strong social support, prosocial involvement, and a positive attitude towards the intervention, are expected to have an increased chance of completing the treatment.
Method Setting
Data were gathered on juveniles admitted to De Catamaran, hospital for youth forensic psychiatry and orthopsychiatry located at GGz Eindhoven in the Netherlands. De Catamaran offers psychological and psychiatric assessment and treatment to juveniles between the age of 14 and 23 years. Patients of de Catamaran have been involved in the Dutch criminal justice system and/or are a risk to themselves or others through their behavior. Therefore, juveniles within de Catamaran can be sentenced under the Dutch juvenile civil law, Dutch juvenile criminal law, or admitted voluntarily.
Participants
Analyses were performed within an existing database of GGzE, center for Child & Adolescence Psychiatry De Catamaran, hospital for youth forensic psychiatry and orthopsychiatry. The amount of female juveniles that were admitted to the hospital was relatively small. To prevent a possible gender effect, only male juveniles were included in the sample, resulting in 370 eligible juveniles. Furthermore, based on the absence of
measurements on the instruments that were used, juveniles without any measurements were excluded from further analyses (n = 123).
After this exclusion, the sample contained 247 male juveniles who were admitted between October 2004 and March 2015, and discharged between February 2006 and December 2015. At admission the juveniles had an average age of approximately 17 years, and at discharge their age was around 18 years. The average treatment duration was about 19 months. Furthermore, 106 juveniles were sentenced under the Dutch juvenile civil law, 115 were sentenced under the Dutch juvenile criminal law, and 26 were voluntarily admitted. Regarding ethnicity, a small majority of the juveniles (n = 138) were of Dutch ethnicity. Of the remaining 109 juveniles, 71 juveniles were immigrants (i.e. at least one parent is born abroad or at least one parent and the juvenile himself is born aboard) and of 38 juveniles the ethnicity was unknown. Detailed information about the characteristics of the juveniles is displayed in Table 1 and Table 2.
Table 1. Juvenile’s Descriptives
Selected sample (n = 247
Excluded sample (n = 123)
Variable M (sd) M (sd) t
Age at the start of the treatment 16.84 (1.81) 17.67 (1.46) -4.34* Age at the end of the treatment 18.43 (2.13) 19.12 (2.13) -2.89* Duration of the treatment in months 18.70 (13.29) 17.03 (15.74) 1.01 Note: *p < .005
Table 2. Juvenile’s Characteristics Selected sample (n = 247) Excluded sample (n = 123) Variable n (%) n (%) Test
Judicial status Civil law 106 (43%) 11 (9%) 2 = 146.07*
Criminal
law 115 (46%) 23 (19%)
Voluntary 26 (11%) 89 (72%)
Ethnicity Dutch 138 (56%) 17 (14%) Fisher’s exact =
.99
Immigrant 71 (29%) 9 (7%)
Unknown 38 (15%) 97 (79%)
Note: *p < .001
Measures
Background information. Background information was collected with structured file analyses. This information included age at the start of the treatment, age at the end of the treatment, duration of the treatment, judicial status, and ethnicity. The structured file analyses were performed by trained researchers or by interns under supervision of a trained researcher.
Emotional and behavioral problems. Emotional and behavioral problems were measured with two instruments, a proxy report and a self-report. First, the Dutch version of the Child Behavior Checklist 4/18 was used to assess emotional and behavioral problems (CBCL; Verhulst, Van der Ende, & Koot, 1996; Achenbach & Rescorla, 2001). The CBCL is completed by parents or caregivers whom are asked to describe the juvenile’s functioning during the previous six months. In this study, the CBCL was completed by the group worker. Items that measure specific emotional and behavioral problems are judged on 3-point scales (0 = not true, 1 = somewhat or sometimes true, or 2 = very true or often true). The CBCL consists of two broadband scales: Internalizing Problems and Externalizing Problems. The broadband scale Internalizing Problems measures emotional problems and contains three
narrowband syndrome scales: Anxious/Depressed, Withdrawn/Depressed, and Somatic Complaints. An example of the Internalizing Problems is: ‘Can stand up for him-/herself’. The broadband scale Externalizing Problems measures behavioral problems and contains two narrowband syndrome scales: Rule Breaking Behavior and Aggressive Behavior. An example of the Externalizing Problems is: ‘Is disobedient at school’. Psychometric analyses indicate a good reliability of the instrument. For the broadband scale, Cronbach’s alpha’s of .90
(Internalizing Problems) and .94 (Externalizing Problems) indicate an excellent reliability. Alpha’s of the syndrome scales ranged between .78 and .94, indicating a good to excellent reliability (Achenbach & Rescorla, 2001; George & Mallery, 2003).
Second, the Dutch version of the Youth Self-Report was used to measure emotional and behavioral problems (YSR; Verhulst, Van der Ende, & Koot, 1997; Achenbach & Rescorla, 2001). The YSR is the self-report version of the CBCL. Juveniles are asked to assess their own emotional and behavioral problems in the previous six months. The YSR is scored on 3-points scales (0 = not true, 1 = somewhat or sometimes true, or 2 = very true or often true). It contains two broadband scales: Internalizing Problems and Externalizing Problems. The former is used to measure emotional problems, while the latter is used the measure behavioral problems. The YSR has the same narrowband syndrome scales as the CBCL. Psychometric analyses showed that the YSR is a reliable instrument: the broadband scales had alpha’s of .90 and the alpha’s of the syndrome scales ranged between .71 and .85 (Achenbach & Rescorla, 2001).
Risk and protective factors. Risk and protective factors were measured with the Dutch version of the Structured Assessment of Violence Risk in Youth (SAVRY; Lodewijks, Doreleijers, De Ruijter, & De Wit-Grouls, 2003). With the SAVRY the risk of violent
behavior and recidivism is assessed based on behavior shown in the previous six months. Risk factors are measured with three broadband scales: Historical Risk Scale (e.g. ‘Previous violent
behavior’), Social/Contextual Risk Scale (e.g. ‘Rejection by peers’), and Individual Risk Scale (e.g. ‘Negative attitudes’). The risk factors are scored on three-points scales (0 = low risk, 1 = moderate risk, 2 = high risk). Furthermore, protective factors are measured with the broadband scale Protective Factors (e.g. ‘Prosocial involvement’), which are measured on a dichotomous scale (0 = present and 2 = absent). Psychometric analyses indicate moderate to good reliability, with alpha’s ranging between .62 and .86 (Lodewijks, Doreleijers, & De Ruijter, 2008).
Dropout. The absence or present of dropout, as well as the type of dropout was measured with the discharge status of juveniles. Four forms of discharge statuses were
defined: completer, self-initiated dropout, step-out, and push-out. First, a juvenile is classified as completer when the treatment is finished and all treatment goals are accomplished. Second, self-initiated dropout occurs when the juvenile end the treatment at own initiative (e.g. not returning to the facility after furlough). Third, a juvenile is classified as step-out when there is a mutual agreement between the juvenile and treatment staff to end the treatment prematurely (e.g. after the safety risk is decreased the juvenile is being transferred to a specialized clinic that better fits the treatment goals). Fourth, push-out occurs when the treatment staff decide to end the treatment prematurely (e.g. when there is no possibility of a therapeutic relationship).
Procedure
Data were obtained from the Routine Outcome Monitoring (ROM) database containing both the CBCL and YSR questionnaires. The ROM-procedure started in January 2005 and was used to monitor juveniles and the overall progress of the treatment during hospitalization. The questionnaires were being re-taken at intervals of six months, in June and December. In the current study only CBCL questionnaires from 2009 and later were included, since the CBCL was not included in the ROM-procedure until 2009. The CBCL is filled out digitally.
For the YSR a slightly different procedure was followed. Prior to December 2012 the YSR questionnaire was only used for scientific research and not part of the ROM-procedure. In this time juveniles were given instruction by letter and verbally and had to sign an
informed consent form with which they gave permission to use the data for scientific research under the condition that anonymity was ensured by the way of coded data. After this
permission the YSR was filled out on a paper and pencil version by the juvenile every six months under supervision of a research assistant. Juveniles were at all times allowed to stop or withdraw without any explanation from the study. When the YSR was added to the ROM-procedure in December 2012 the measurement became an integral part of the treatment. From then on informed consent was acquired passively: consent to the use of data for research ends was included with the signing the treatment agreement. The YSR today is filled out digitally.
The SAVRY was not included in the ROM-procedure and a different procedure was followed. Three months after admission the historical items of the SAVRY were scored based on medical and criminal files. Six months after admission the other risk factors and the
protective factors were scored based on information from the electronic patient records, including treatment plans and reports by staff members. With this information the historical items were also complemented. When juveniles stayed less than six months, the SAVRY was scored at the time of release. Scoring of the SAVRY was done by trained professionals or interns. Individual scoring took place after an interrater reliability of at least 80 percent was reached. Interns stayed under supervision of a trained professional (even after reaching an interrater reliability of 80 percent).
The discharge status was scored by trained researchers and/or interns. Information necessary to score the discharge status was gathered from the electronic patient records. Scoring was based on a mutual agreement between two informants (i.e. two researchers or a researcher and an intern).
For the purpose of the current study, only T1 questionnaires were included.
Furthermore, even though the CBCL was only included from 2009 onwards, the sample of the CBCL is larger than of the YSR. Possibly because of a lower response rate or the exclusion of the Young Adult Self-Report (YASR) questionnaires. In the past the YASR was filled out by juveniles that were 18 years or older instead of the YSR. The YASR is the young adult version of the YSR and directed at juveniles and adults in the age of 18 to 30 years. Since de Catamaran is a youth care institution, the YASR is no longer part of the measurements and all juveniles fill out the YSR. Because of this, the YASR was not included in the current study.
Data analyses
All analyses were performed with the Statistical Package for the Social Sciences (SPSS) 19.0. Attrition analyses. First, an attrition analysis was performed with t-test for
independent samples (for continuous variables) and Chi-square-tests (for categorical
variables) to identify the differences in background variables between the selected sample (i.e. juveniles with a measurement on at least one of the instruments) and the excluded sample (i.e. juveniles without any measurements). The Fisher’s Exact Test was used when the assumption was violated that the minimum expected cell frequency is less than five.
Second, attrition analyses were performed with the missing data within the SAVRY. The missing data was approached on two levels: a group level and an individual level. For this, the method of Spice and colleagues (2010) was followed. On group level items were excluded on which more than 10% of the juveniles had a missing value. This led to the exclusion of five items: the historical risk items 3 (Early expression of previous violent behavior), 6 (Has been a witness of violence in the family) and 7 (History of child abuse), the social/contextual item 11 (Interference with delinquent peers), and the individual item 21 (Lack of empathy/remorse). On the individual level, for juveniles who exceeded the 10%
limit (i.e. had more than one item missing) on a scale, the juvenile’s total score for that specific scale was not computed. This occurred for 12 juveniles: seven within the Historical Risk Scale, 3 within the Social/Contextual Risk Scale, and 2 within the Individual Risk Scale.
Logistic regression analyses. To test the assumption that it is important to distinguish between different dropout types, the predictors of dropout in general were first identified with direct logistic regression analyses. The assumption regarding no multi-collinearity was
checked using the Tolerance-statistic and Variance Inflation Factor (VIF) and the assumption of no outliers was checked during the analyses with the Casewise List Table. Next, three models were tested. The first model contained the Internalizing Problems and Externalizing Problems of the CBCL as independent variables, and treatment dropout as dependent variable. The second model consisted of the Internalizing Problems and Externalizing Problems of the YSR as independent variables, and treatment dropout as dependent variable. The third model contained the Historical Risk Scale, Social/Contextual Risk Scale, Individual Risk Scale, and Protective Factors of the SAVRY as independent variables, and treatment dropout as
dependent variable. The reference category in all three models was the completer group and all models were corrected for the number of months the juveniles was admitted before measurement. The fit of the models was tested with the Chi-square model test and the effects of the independent variables with the Wald-statistic. A significance level of p < .05 was used.
Group’s descriptives. To map the differences between juveniles in the four discharge status groups one-way between-groups analysis of variance (ANOVA’s) were used for
continuous background variables. Since the four groups were unequal, the Welch-statistic was used to interpret the results of the ANOVA’s. Post-hoc analyses using the Tukey-HSD were computed to indicate the place of possible differences and with the Eta Squared statistic the magnitude of the differences was calculated. For categorical background variables the
Chi-square statistic was used. If the Chi-Chi-square assumption that the minimum expected cell frequency is less than five was violated, the Fisher’s Exact Test was used.
Multinomial logistic regression analyses. Multinomial logistic regression analyses were conducted to estimate the probability that a juvenile would become a certain dropout type (i.e. self-initiated dropout, step-out, push-out) based on multiple independent variables (Starkweather & Moske, 2011). In these analyses the three types of dropout were used as the dependent variable and completer was used as reference category. The same three models as with the direct logistic regression analyses (CBCL, YSR, SAVRY) were tested. The fit of the model was tested with the Chi-square model test, using the Pearson statistic. After testing the model fit, the effects of the independent variables were determined with the Wald-statistic. A significance level of p < .05 was used. All models were corrected for the number of months the juvenile was admitted before measurement.
Post-hoc analyses. Multinomial logistic regression analyses were used as post-hoc tests to determine the effect of the syndrome scales on different types of treatment dropout. Two models were tested: CBCL and YSR. These models had the five syndrome scales (Anxious/Depressed, Withdrawn/Depressed, Somatic Complaints, Rule Breaking Behavior, and Aggressive Behavior) as independent variables and the dropout types as dependent variable. The completer group was used as reference category. The same statistics as with the multinomial logistic regression analyses as mentioned above were used to test the fit of the model and the effect of the independent variables.
Results Attrition analyses
To test if the selected sample (n = 247) differs from the excluded sample (n = 123) ANOVA’s for continuous variables and Chi-square-tests for categorical variables were computed. The groups were significant different on all background variables except on treatment duration
and ethnicity. As shown in Table 1 and 2, the selected sample was younger at the start and at the end of the treatment, and more often sentenced under the Dutch juvenile civil law or Dutch juvenile criminal law. The excluded sample was more often admitted voluntarily.
Direct logistic regression analyses
Assumptions. There was no indication of multi-collinearity with Tolerance statistics ranging between .84 and .96, and VIF values ranging between 1.04 and 1.91. Outliers were checked with the regression analyses and no outliers were found.
Logistic regression analyses. Direct logistic regression analyses were used to assess the impact of a number of factors on the likelihood that a juvenile would drop out of
treatment. There was no distinction made between different types of dropout. Three models were tested with treatment dropout as dependent variable. Both the first model (containing the CBCL Internalizing Problems and Externalizing Problems as independent variables) and the second model (containing the YSR Internalizing Problems and Externalizing Problems as independent variables) were not statistically significant: 2 (3) = 5.76, p = .443, 2 (3) = 6.18, p = .103 respectively.
The third model (containing the SAVRY Historical Risk Scale, Social/Contextual Risk Scale, Individual Risk Scale, and Protective Factors) was reached a statistical significance: 2 (5) = 21.21, p = .001. The model explained 13% of the variances in treatment dropout, and correctly classified 65% of the cases. Furthermore, two independent variables made unique statistically significant contributions. The first and strongest predictor was the Historical Risk Scale, recording an odds ratio of 1.22 (95% CI 1.08 – 1.38). This indicates that when the score on the Historical Risk Scale increases with one point, the juvenile is 1.22 times more likely to drop out of treatment than to complete the treatment. The second predictor was the Individual Risk Scale. This independent variable recorded an odds ratio of 1.13 (95% CI 1.02
– 1.27). Therefore, when the score on the Individual Risk Scale increases with one point, juveniles were 1.13 times more likely to drop out of treatment than to complete treatment. The analyses details are displayed in Table 3.
Table 3. Logistic Regression Predicting Likelihood of Dropping Out of Treatment Versus Completing Treatment
Model 2 Nagelkerke R
squared
Independent variable Odds Ratio [95% C.I. for OR] CBCL (n = 195)a 5.76 .04 Internalizing Problems .98 [.95 – 1.01] Externalizing Problems 1.02* [1.00 – 1.05] YSR (n = 190)b 6.18 .04 Internalizing Problems .97* [.93 – 1.00] Externalizing Problems 1.03 [.99 – 1.06] SAVRY (n = 217)c 21.12** .13 Historical Risks 1.22** [1.08 – 1.38] Social/Contextual Risks .88 [.72 – 1.06] Individual Risks 1.13* [1.02 – 1.27] Protective Factors .89 [.78 – 1.02] Note: a dropout n = 100, completer n = 95; b dropout n = 98, completer n = 92; c dropout n = 118, completer n = 99; * p < .05, ** p < .005; Corrected for number of months admitted before measurement; The reference category is: Completer.
Descriptives and characteristics of juveniles within each discharge status group
ANOVA’s were performed to map the differences between juveniles in different discharge statuses on continuous background variables. The Welch-statistic was used to the interpret the results. There was only a statistically significant difference on treatment duration: F (3, 18.19) = 8.11, p =.000. The effect size, calculated using the Eta Squared, was .09, indicating a
moderate effect (DeVellis, 2003). Furthermore, post-hoc comparisons using the Tukey-HSD test indicated that treatment duration was the shortest for self-initiated dropout juveniles and the longest for step-out juveniles. Detailed descriptive information is displayed in Table 4.
For categorical background variables Chi-square tests were performed. Because the assumption that the minimum expected cell frequency should be less than five was violated,
results were interpreted with the Fisher’s Exact test. There was a significant difference on ethnicity (p = .022), indicating that completers were more often of Dutch ethnicity. More detailed characteristic information is displayed in Table 5.
Table 4. Descriptive of Juveniles Within each Discharge Status Group Self-initiated
dropout1
Step-out 2 Push-out3 Completer4
Variable M (sd) M (sd) M (sd) M (sd) F test Group difference Age at start of treatment 16.76 (1.62) 16.83 (1.40) 17.07 (2.21) 16.83 (1.85) .23 Age at end of treatment 17.87 (1.91) 18.86 (2.03) 18.38 (2.43) 18.60 (2.20) 2.27 Treatment duration (in months) 13.52 (11.07) 24.08 (16.08) 14.78 (11.62) 21.25 (12.69) 8.11* 1<3<4<2 Note: * p < .001
Table 5. Characteristics of Juveniles Within each Discharge Status Group
Self-initiated dropout
Step-out Push-out Completer
Variable n (%) n (%) n (%) n (%) Fisher’s
Exact test Judicial status Voluntary 4 (7%) 5 (14%) 3 (7%) 14 (13%) 7.29
Civil law 29 (53%) 12 (33%) 16 (35%) 49 (44%) Criminal law 22 (40%) 19 (53%) 27 (58%) 47 (43%) Ethnicity Dutch 27 (49%) 19 (52%) 21 (46%) 71 (64%) 14.25* Immigrant 22 (40%) 11 (31%) 14 (30%) 24 (22%) Unknown 6 (11%) 6 (17%) 11 (24%) 15 (14%) Note: * p < .05
Multinomial logistic regression analyses
The first model (CBCL) tested was statistically significant: 2 (9) = 46.63, p = .000. The full model explained 23% of the variances in dropout types. However, only the Internalizing Problems made a unique contribution to the model. An odds ratio of .94 (95% CI .90 – .99) was found, indicating that with a one-point increase on the Internalizing Problems, the juvenile was .94 times less likely to end the treatment as a self-initiated dropout than as a completer. In other words, with a one-point increase on the Internalizing Problems, juveniles are 1.063 times more likely to end the treatment as a completer than as a self-initiated dropout.
The second model (YSR) also reached statistically significance: 2 (9) = 49.32, p = .000, and explained 24% of the variances in dropout types. Again, only the Internalizing Problems made a unique contribution to the model with an odds ratio of .92 (95% CI .87 – .97) for self-initiated dropout. So, with a one-point increase on self-reported emotional
problems, juveniles were 1.09 times more likely to complete the treatment rather than end it at own initiative.
The third model (SAVRY) did also prove to be statistically significant: 2 (15) = 71.83, p = .000, with 32% of the variances explained by the model. Three of the independent variables (Historical Risk Scale, Individual Risk Scale, and Protective Factors) made unique contributions to the model. The Historical Risk Scale recorded an odds ratio of 1.25 (95% CI 1.07 – 1.47) for self-initiated dropout, and 1.19 (95% CI 1.02 – 1.04) for push-out. For each one-point increase on this scale, juveniles are 1.25 times more likely to end the treatment as a self-initiated dropout and 1.19 times more likely to end the treatment as a push-out than as a completer. The Individual Risk Scale had an odds ratio of 1.18 (95% CI 1.00 – 1.04) for step-out, and 1.23 (95% CI 1.06 – 1.43) for push-out. This also indicates that for each one-point increase on the Individual Risk Scale, juveniles were 1.18 times or 1.23 times more likely to
end the treatment as a step-out or push-out than as a completer. The Protective Factors recorded an odds ratio of .78 (95% CI .64 – .99) for step-out and .81 (95% CI .68 – .97) for push-out. Therefore, for each one-point increase, the chance that juveniles end their treatment as a completer than as step-out or push-out increased with 1.28 or 1.23 times respectively. For detailed analyses outcomes, see Table 6 and Table 7.
Table 6. Multinomial Logistic Regression Predicting the Likelihood of Different Types of Dropout using the CBCL and YSR Self-initiated dropout (n= 35a/38b) Step-out (n = 27a/24b) Push-out (n = 38a/36b)
Model 2 Nagelkerke R squared Variable Odds ratio
[95% C.I. for OR]
Odds ratio
[95% C.I. for OR]
Odds ratio
[95% C.I. for OR] CBCL (n = 195) 46.63** .23 Internalizing Problems .94* [.90 –.99] .97 [.92 – 1.02] .96 [.92 – 1.00] Externalizing Problems 1.01 [.99 – 1.04] .97 [.94 – 1.01] 1.01 [.98 – 1.04] YSR (n = 190) 49.32*** .24 Internalizing Problems .92** [.87 – .97] .98 [.93 – 1.03] .96 [.92 – 1.00] Externalizing Problems 1.02 [.98 – 1.09] .95 [.90 – 1.00] 1.00 [.96 – 1.04] Note: a sample size of CBCL; b sample size of YSR; * p < .05, ** p < .01, *** p <.005; the reference category is: Completer
Table 7. Multinomial Logistic Regression Predicting the Likelihood of Different Types of Dropout using the SAVRY Self-initiated dropout (n = 40) Step-out (n = 29) Push-out (n = 39) Model 2 Nagelkerke R squared
Variable Odds ratio
[95% C.I. for OR]
Odds ratio
[95% C.I. for OR]
Odds ratio
[95% C.I. for OR] SAVRY
(n = 217)
71.83*** .32 Historical Risk Scale 1.25** [1.07 – 1.47] 1.16 [.98 – 1.38] 1.19* [1.02 – 1.40] Social/Contextual Risk Scale .83 [.65 – 1.07] 1.01 [.75 – 1.36] .79 [.61 – 1.03] Individual Risk Scale 1.01 [.87 – 1.17] 1.18* [1.00 – 1.40] 1.23** [1.06 – 1.43] Protective Factors .87 [.73 – 1.03] .78* [.64 – .99] .81* [.68 – .97] Note: * p < .05, ** p < .01, *** p <.005; the reference category is: Completer
Post-hoc analyses using multinomial logistic regression
Post-hoc analyses were performed to get more insight into which specific emotional problems contributes to the protective value for self-initiated dropout. Two models were tested. The first model contained the five syndrome scales of the CBCL, three of them belonging to the Internalizing Problems (Anxious/Depressed, Withdrawn/Depressed, and Somatic
Complaints), and two to the Externalizing Problems (Rule Breaking Behavior and Aggressive Behavior). The first model reached statically significance (2 (18) = 50.73, p = .000) but none of the independent variables made unique contributions to the model.
The second model that was tested contained the five syndrome scales of the YSR, of which three belonged to the Internalizing Problems (Anxious/Depressed,
Withdrawn/Depressed, and Somatic Complaints), and three to the Externalizing Problems (Rule Breaking Behavior and Aggressive Behavior). The full model was statistically significant: 2 (18) = 61.49, p = .000, and explained 30% of the variances. Only Somatic Complaints made a unique contribution to the model. For self-initiated dropout an odds ratio of .79 (95% CI .64 – .96) was found. For step-out there was also an odds ratio of .79 (95% CI .64 – .98). Thus, with an increase of one point on this syndrome scale, juveniles are 1.27 times more likely to end the treatment as a completer than as self-initiated dropout or step-out. Analyses details are displayed in Table 8.
Table 8. Post-hoc Analyses Conducted with the CBCL and YSR Self-initiated dropout (n= 35a/38b) Step-out (n = 27a/24b) Push-out (n = 38a/36b)
Model 2 Nagelkerke R squared Variable Odds ratio
[95% C.I. for OR]
Odds ratio
[95% C.I. for OR]
Odds ratio
[95% C.I. for OR] CBCL (n = 195) 50.73*** .25 Anxious/Depressed .94 [.84 – 1.04] 1.00 [.89 – 1.12] .96 [.87 – 1.06] Withdrawn/Depressed .91 [.80 – 1.04] .90 [.79 – 1.03] .90 [.78 – 1.01] Somatic Complaints 1.00 [.85 –1.18] 1.02 [.85 – 1.22] 1.08 [.93 – 1.25] Rule Breaking Behavior 1.03 [.94 – 1.07] .99 [.89 – 1.09] 1.04 [.96 – 1.14] Aggressive Behavior 1.00 [.94 – 1.07] .97 [.90 – 1.05] 1.00 [.93 – 1.06] YSR (n = 190) 49.32*** .24 Anxious/Depressed .88 [.76 – 1.03] 1.06 [.93 – 1.22] .92 [.80 – 1.06] Withdrawn/Depressed 1.09 [.92 – 1.29] 1.01 [.84 – 1.22] .97 [.82 – 1.14] Somatic Complaints .79* [.64 - .96] .79* [.64 – .98] 1.02 [.86 – 1.20] Rule Breaking Behavior .99 [.89 – 1.11] .95 [.84 – 1.08] 1.03 [.93 – 1.15] Aggressive Behavior 1.03 [.92 – 1.14] .95 [.84 - 108 .96 [.87 – 1.07] Note: a sample size of CBCL; b sample size of YSR; * p < .05, ** p < .01, *** p <.005; the reference category is: Completer
Discussion
In line with previous research (De Haan et al., 2013; Johnson et al., 2008), the current study has indicated that a large percentage of the juveniles (i.e. 55%) drop out of treatment in secure residential youth care. Because of this, the current study aimed to identify predictors of
treatment dropout. The results have shown that we were able to predict treatment dropout based on multiple factors. However, we assumed that it is important to distinguish different types of dropout because each type might have its own risk and protective factors (Eisengart et al., 2008; Sunseri, 2003). This assumption was confirmed, given that the predictors of dropout in general were not similar for each type of dropout. Furthermore, relationships between protective factors and dropout were only identified when we looked at different types of dropout. These findings have supported the importance of distinguishing between different types of dropout.
Predicting treatment dropout in general
For dropout in general both the presence of historical risk factors and individual risk factors increased the chance that a juvenile would drop out of treatment rather than complete the treatment. These findings might be explained with two theories: the adolescence-limited/life-course-persistent theory of Moffitt (1993) and the attachment theory of Bowlby (2005). First, the presence of historical risk factors might indicate the capability and willingness of
juveniles to change their behavior. According to Moffitt (1993) antisocial behavior can occur in two ways: adolescence-limited and life-course-persistent. The first is characterized by adolescence-onset antisocial behavior and desistance of this behavior after transition into adulthood. Adolescence-limited antisocial behavior is, to some degree, considered as normal in the life of juveniles. On the other hand, life-course-persistent antisocial behavior is
adulthood. Life-course-persistent juveniles more often had childhoods of inadequate
parenting, a difficult temperament, and anger management problems. Because the persistence and severity of their antisocial behavior, these juveniles are less sensitive to corrections (Moffitt, 1993; Moffitt & Capsi, 2001). To further illustrate this theory, Mulder, Brand, Bullens and Van Marle (2011) found that factors such as past delinquent behavior and experiences of poor parenting skills increased the chance of recidivism. Thus, historical risk factors, such as past violent behavior, past non-violence offending, and past intervention failures might be a warning signal that juveniles are on the life-course-persistent pathway. Therefore, their behavior is harder to change and they might not feel the need for treatment, resulting in an increased chance of treatment dropout.
The second theory might also explain the predictive value of the historical risk factors but in combination with the individual risk factors. More specifically, it is assumed that early caregiver disruption, which is measured with the historical risk factors, is partly responsible for the increased risk of treatment dropout. Bowly (2005) states that the early attachment is determined for the quality of relationships in further life. This attachment can be threatened by factors such as early caregiver disruption and increase the chance of an insecure
attachment. The difficulty of establishing a therapeutic relationship with insecurely attached juveniles (Zegers, Schuengel, Van Ijzendoorn, & Janssens, 2006), might explain the
contribution of individual risk factors. Namely, insecurely attached juveniles more often show high-risk behavior (e.g. substance abuse and acting out) as an attempt to receive attention (Warr, 2007) and show poor compliance. These factors are all negatively associated with the formation of a therapeutic relationship (Byers & Lutz, 2015; Eltz, Shirk, & Sarlin, 1995; Harder et al., 2013) and measured with the individual risk factors. Since the therapeutic relationship is one of the most robust indicators of change in treatment (Zegers et al., 2006),
juveniles with whom a therapeutic relationship is harder to establish might be at risk of treatment dropout.
The prediction of self-initiated dropout
The first hypothesis that juveniles with behavioral problems, such as rule breaking behavior, have an increased risk of self-initiated dropout was not confirmed. Contrary to the
expectations, the presence of historical risk factors made it more likely that juveniles end their treatment as a self-initiated dropout than as a completer. Self-initiated dropout juveniles are characterized by antisocial behavior that leads to ending the treatment at own initiative. The historical risks factors might be important because antisocial attitudes and beliefs find their origin in early childhood (Moffitt, 1993). Moreover, the Historical Risk Scale is correlated with violent recidivism and the strongest predictor within the SAVRY of both general and violent recidivism (Dolan & Rennie, 2008). Combining these findings with the theory of Moffitt (1993), the presence of historical risk factors might indicate a life-course-persistent pathway. Juveniles are less willing to chance their behavior and more likely to continue antisocial behavior (e.g. running away from the facility or engagement in delinquent acts), resulting an increased chance of self-initiated dropout.
Furthermore, results have shown that the presence of emotional problems and self-reported somatic complaints decreased the chance that a juvenile ended the treatment as a self-initiated dropout than as a completer. Possibly because these problems create a certain level of distress. Brookman-Freeze and colleagues (2008) showed that juveniles’ self-reported severity of symptoms was positively correlated to the frequency of treatment visits.
According to them, treatment motivation depends on the awareness of the symptoms by the juveniles themselves. When a certain level of distress is perceived, juveniles are more
inwards (Achenbach & Rescorla, 2001) it might be easier for the juvenile to recognize them, create a feeling of treatment need, and result in a decreased chance of self-initiated dropout.
The prediction of step-out
The second hypothesis that juveniles with higher levels of emotional problems, such as depression, have an increased risk of step-out was not confirmed. This finding is in contrary to the finding of Johnson and colleagues (2008). However, they examined dropout in
outpatient treatment and with younger juveniles. So, the contrary results might be caused by differences in target groups and settings. Moreover, the current findings were in line with more comparable studies (Van der Ploeg & Scholte, 2003; Vos, 2008): within a residential setting, emotional problems do not seem to be a risk factor for treatment dropout.
Moreover, the results have indicated that individual risk factors increased the chance that juveniles end their treatment as a step-out than as completer. Possibly because the length of the treatment is negatively associated with motivation for the treatment. The urge to practice new learned skills becomes stronger as the length of the treatment increases.
Juveniles are more motivated to move to settings (e.g. home training) where they can practice their skills (Dixon, 2007). In the current study the length of the treatment was on average the longest for step-out, around two years. So, it assumable that juveniles lose their motivation for residential treatment and therefore are more likely to become a step-out. On the other hand, juveniles might be protected against step-out when they are motivated enough to accomplish all their treatment goals, even after a long stay in the facility. This can explain why protective factors as measured with the SAVRY were found to increase the chance that a juvenile completes the treatment, rather than to step out of the treatment. After all, within this scale factors related to a better motivation towards the treatment are measured.
The prediction of push-out
The fourth hypothesis that juveniles with higher levels of behavioral problems, such as aggression, have an increased chance of push-out was not confirmed. In the current study the presence of both historical and individual risk factors increased the chance that the juvenile would end the treatment as a push-out than as a completer. Since push-out juveniles are characterized by the incapability of the formation of a therapeutic relationship, the attachment theory can offer an explanation. As mentioned above, the historical risk factors might be indicative for the early attachment, whereas the individual risk factors might represent the consequences of an insecure attachment (e.g. disruptive behavior and poor compliance).
Results have also indicated that the chance of push-out decreased when protective factors as measured with the SAVRY were present. This scale contains factors such as a positive attitude towards the treatment, prosocial involvement, and strong social support, which all positively affect the formation of a therapeutic relationship (Harder et al., 2013). As a result, juveniles with these factors might be protected against becoming a push-out.
Predicting treatment completion
The fourth hypothesis that juveniles with protective factors, such as a strong social support, prosocial involvement, and a positive attitude towards the intervention, have an increased chance of completing the treatment was partly confirmed. The protective factors as measured with the SAVRY decreased the chance of both step-out and push-out. As mentioned above, it is assumed that this protective value occurs because this scale contains aspects that are associated with a higher motivation and a stronger therapeutic relationship. However, for self-initiated dropout the protective factors as measured with the SAVRY did not had any
predictive value. Here, the presence of emotional problems and self-reported somatic complaints were identified as protective factors. For these juveniles the awareness of the
problems might prevent them of ending the treatment at own initiative, but is not directly related to the motivation and willingness to chance.
The impact of behavioral problems
Contrary to the findings of previous studies (De Haan et al., 2013; Eisengart et al., 2008; Friars & Mellor, 2007; Johnson et al., 2008; Sunseri, 2003), behavioral problems were not found to be a risk factor in the current study. One possible explanation for this is that previous studies have different target groups than the current study. For example, in previous studies (Eisengart et al., 2008; Friar & Mellor, 2007; Johnson et al., 2008) dropout was examined in outpatient treatment, whereas the current study was directed at juveniles in secure residential treatment. Research has shown that juveniles in residential youth care are characterized by their severe behavioral problems (Harder et al., 2013). Therefore, in residential youth care the presence of behavioral problems might not be predictive anymore for a specific type of dropout. This result is in line with previous research on dropout in residential youth care (Van der Ploeg & Scholte, 2003; Vos, 2008).
Another explanation for the differences in results is that previous studies contained a sample of younger juveniles (Eisengart et al., 2008; Johnson et al., 2008). Younger juveniles are in a different stage regarding biological, cognitive, and social-emotional development (Santrock, 2005). Therefore, a younger sample might not be comparable with the current sample. In summary, different target groups and settings might explain why findings of previous studies were not replicated in the current study.
Clinical implications
With the current study more insight is gain into predictors of different types of dropout in secure residential treatment. Paying attention and effort to these warning signals could pay
profits in the treatment of juveniles. For example, juveniles with a history of antisocial behavior might be at risk for self-initiated dropout. When the treatment staff can create a feeling of need for the treatment, this type of dropout might be prevented. Furthermore, efforts regarding the therapeutic relationship might be rewarding for push-out juveniles, while for step-out juveniles the treatment staff should give them enough challenges to keep them motivated. The current study can be seen as a start in the prediction of different types of dropout. Further research is necessary to explore more specific predictors to enhance the usability of the results in clinical practice.
Limitations
The current study has several limitations that should be taken into account. A first limitation of the study is that due to a small number of female juveniles, only males were included in the sample. That is why the results cannot be generalized to females. In addition, research has shown that males and females in residential youth care differ on multiple factors. For example, female juveniles have higher levels of mental health problems (Fazel, Doll, & Langstrom, 2008; Russel & Marston, 2010), their antisocial behavior is much more difficult to predict (Landsheer & Van Dijkum, 2005), and they show fewer clear links between childhood aggression and offending during adolescence (Broidy et al., 2003). Regarding dropout, female juveniles with conduct disorder are less likely to complete treatment (Wells and Faragher, 1993) and aggressive female juveniles are more likely to drop out of high school (Stack, Serbin, Schwartzman, & Ledingham, 2005). Unfortunately, research on female dropout is scarce. This scarcity might be caused by the mostly male population in both
forensic and non-forensic youth care (Cauffman, 2008; Jurrius, Bauer, Rutjes, & Stams, 2011) and the finding that male and female juveniles tend to share the same risk factors for
antisocial behavior (Goldweber, Broidy, & Cauffman, 2009). Nevertheless, it is interesting to further examine dropout in a female sample.
A second limitation of the current study is that the sample mostly consisted of juveniles that were sentenced under the Dutch juvenile civil law or Dutch juvenile criminal law. Therefore, the results might not be representative for voluntarily admitted juveniles. Previous research has shown that forced placed juveniles show a lack of motivation (Englebrecht, Peterson, Scherer, & Naccarato, 2008), are more often unaware of their
problems, and show resistance to change (Karver, Handelsman, Fields, & Bickman, 2006). It was beyond the scope of this study to test these findings. Further research in a comparable sample is necessary, since previous studies (Englebrecht et al., 2008; Karver et al., 2006) had other target groups and research designs.
A third limitation of the current study is that there was a great amount of missing values within the SAVRY. Exclusion of some items were necessary to retain reliability of the scales. However, this can also negatively affect the external validity of the instrument.
Psychometric analyses are after all performed with all items. Besides this, the items that were excluded might have made a contribution to the prediction. For example, a history of child maltreatment and exposure to violence in the home are related to both an insecure attachment (Bowlby, 2005) and to life-course-persistent antisocial behavior (Moffitt, 1993). To advance our understanding of dropout, it might be interesting to include these factors in the prediction of dropout.
Recommendations for future research
To advance our understanding of different types of dropout, further research is necessary. This research should focus on the limitations of the current study, such as the inclusion of females, voluntarily admitted juveniles, and analyses with all SAVRY items. Besides
focusing on the limitations of the current study, it is recommended that further research uses more specific instruments to predict dropout. The instruments in the current study give an overall view of important risk factors, such as historical risk factors, but lack specificity. In other words, the findings in the current study indicate that historical risk factors are important but do not show which factors within this scale are responsible for the predictive value. Specification can create more concrete clinical implications and can lead to a better prevention of dropout.
Conclusion
The current study shows that treatment dropout can be predicted at an early stage. Rather than to look at dropout in general, a distinction between different types of dropout should be made. Also, the inclusion of protective factors showed to be useful since it gives information on the factors that need to be reinforced during treatment. In conclusion, dropout can occur in different forms. To prevent them, it is important that both researchers and clinicians have in mind that these different types of dropout might need a different approach.
References
Achenbach, T.M., & Rescorla, L.A. (2001). Manual for the ASEBA School-Age Forms & Profiles. Burlington, VT: University of Vermont, Research Center for Children, Youth, & Families.
Albrecht, G., Veerman, J. W., Damen, H., & Kroes, G. (2001). The Child Behavior Checklist for group care workers: A study regarding the factor structure. Journal of Abnormal Child Psychology, 29(1), 83-89. doi:10.1023/A:1005255614039
Baruch, G., Vrouva, I., & Fearon, P. (2009). A follow-up study of characteristics of young people that dropout and continue psychotherapy: Service implications for a clinic in the community. Child and Adolescents Mental Health, 14(2), 69-75.
doi:10.1111/j.1475-3588.2008.00492.x
Bowlby, J. (2005). A secure base: Parent-child attachment and healthy human development. London, United States of America: Basic Books.
Broidy, L. M., Nagin, D. S., Tremblay, R. E., Bates, J. E., Brame, B., Dodge, K. A., . . . Vitaro, F. (2003). Developmental trajectories of childhood disruptive behaviors and adolescent delinquency: A six-site, cross-national study. Developmental Psychology, 39(2), 222-245. doi:10.1037/0012-1649.39.2.222
Brookman-Frazee, L., Haine, R. A., Gabayan, E. N., & Garland, A. F. (2008). Predicting frequency of treatment visits in community-based youth psychotherapy. Psychological Services, 5(2), 126-138. doi:10.1037/1541-1559.5.2.126
Byers, A. N. & Lutz, D. J. (2015). Therapeutic alliances with youth in residential care:
Challenges and recommendations. Residential Treatment for Children & Youth, 32(1), 1-18. doi:10.1080/0886571X.2015.1004285
De Haan, A. M., Boon, A. E., De Jong, J. T. V. M., Hoeve, M., & Vermeiren, R. R. J. M. (2013). A meta-analytic review on treatment dropout in child and adolescent
outpatient mental health care. Clinical Psychology Review, 33(5), 698-711. doi:10.1016/j.cpr.2014.04.005
DeVellis, R. F. (2003). Scale development: Theory and applications (2nd edn). Thousand Oaks, California: Sage.
Dixon, J. (2007). Young people leaving residential care: Experiences and outcomes. In A. Kendrik (Red.), Residential child care: Prospects and challenges (pp. 76-90 ). London and Philadelphia: Jessica Kingsley Publishers.
Dolan, M. C. & Rennie, C. E. (2008). The Structured Assessment of Violence Risk in Youth as predictor of recidivism in a United Kingdom cohort of adolescent offenders with conduct disorder. Psychological Assessment, 20(1), 35-46. doi:10.1037/1040-3590.20.1.35
Eisengart, J., Martinovich, Z., & Lyons, J. S. (2008). Discharge due to running away from residential treatment: Youth and setting effects. Residential Treatment for Children & Youth, 24(4), 327-343. doi:10.1080/08865710802174418
Eltz, M. J., Shirk, S. R., & Sarlin, N. (1995). Alliance formation and treatment outcome among maltreated adolescents. Child Abuse & Neglect, 19(4), 419-431.
doi:10.1016/0145-2134(95)00008-V
Englebrecht, C., Peterson, D., Scherer, A., & Naccarato, T. (2008). “It’s not my fault”: Acceptance of responsibility as a component of engagement in juvenile residential treatment. Children and Youth Services Review, 30(4), 466-484.
doi:10.1016/j.childyouth.2007.11.005
Fazel, S., Doll, H., & Långström, N. (2008). Mental disorders among adolescents in juvenile detention and correctional facilities: a systematic review and metaregression analysis of 25 surveys. Journal of the American Academy of Child & Adolescent
George, D., & Mallery, P. (2003). SPSS for Windows step by step: A simple guide and reference. 11.0 update (4th ed.). Boston: Allyn & Bacon.
Goldweber, A., Broidy, L. M., &Cauffman, E. (2009). Interdisciplinary perspectives on persistent female offending: A review of theory and research. In J. Savage (Ed.), The development of persistent criminality (pp. 205-230). New York: Oxford University Press.
Harder A. T., Knorth, E. J., & Kalverboer, M. E. (2012). Securing the downside up: Client and care factors associated with outcomes of secure residential youth care. Child and Youth Care Forum, 41(3), 259-276. doi:10.1007/s10566-011-9159-1
Harder, A. T., Knorth, E. J., & Kalverboer, M. E. (2013). A secure base? The adolescent–staff relationship in secure residential youth care. Child and Family Social Work, 41(3), 305-317. doi:10.1111/j.1365-2206.2012.00846.x
Ivan, Z. & Bereczkei, T. (2006). Parental bonding, risk-taking behavior and life history theory. Journal of Cultural and Evolutionary Psychology, 4(3), 267-275. doi:10.1556/JCEP.4.2006.3-4.5
Johnson, E., Mellor, D., & Brann, P. (2008). Differences in dropout between diagnoses in child and adolescent mental health services. Clinical Child Psychology and Psychiatry, 13(4), 515-530. doi:10.1177/1359104508096767
Jurrius, K., Bauer, J., Rutjes, L., & Stams, G. J. (2011). Databank biedt overzicht van
cliënttevredenheidsonderzoek. Jeugd en Co Kennis, 5(4), 30-39. doi:10.1007/s12450-011-0038-z
Karver, M. S., Handelsman, J. B., Fields, S., & Bickman, L. (2005). A theoretical model of common process factors in youth and family therapy. Mental Health Services Research, 7(1), 35–51. doi:10.1007/s11020-005-1964-4.
Karver, M. S., Handelsman, J. B., Fields, S., & Bickman, L. (2006). Meta-analysis of therapeutic relationship variables in youth and family therapy: The evidence for different relationship variables in the child and adolescent treatment outcome literature. Clinical Psychology Review, 26(1), 50–65. doi:10.1016/j.cpr.2005.09.001 Knorth, E. J., Harder, A. T., Zandberg, T., & Kendrick, A. J. (2008). Under one roof: A
review of selective meta-analysis on the outcomes of residential child and youth care. Children and Youth Services Review, 30(2), 123-140.
doi:10.1016/j.childyouth.2007.09.001
Knorth, E. J., Klomp, M., Van den Bergh, P. M., & Noom, M. J. (2007). Aggressive adolescents in residential care: A selective review of treatment requirements and models. Adolescence, 42(167), 461-485. Retrieved from
https://www.researchgate.net/profile/Erik_Knorth/publication/5801167_Aggressive_a dolescents_in_residential_care_a_selective_review_of_treatment_requirements_and_ models/links/02e7e51d66907881e2000000.pdf
Landsheer, J. A. & Van Dijkum, C. (2005). Male and female delinquency trajectories from pre through middle adolescence and their continuation in late adolescence.
Adolescence, 40, 729-748. Retrieved from
http://web.b.ebscohost.com/ehost/pdfviewer/pdfviewer?sid=fc32a4cc-eea6-4333-83e7-7760475b70d2%40sessionmgr107&vid=1&hid=123
Lodewijks, H. P. B., De Ruiter, C., & Doreleijers, T. A. H. (2008). SAVRY risk assessment in violent Dutch adolescents. Relation to sentecing and recidivism. Criminal Justice and Behavior, 35(6), 696-709. doi:10.1177/0093854808316146
Lodewijks H, de Ruiter C, Doreleijers T. (2010) The impact of protective factors in desistance from violent reoffending: A study in three samples of adolescent offenders. Journal of
Lodewijks, H., Doreleijers, T., De Ruiter, C., & De Wit-Grouls, H. (2003). Handleiding voor de gestructureerde risicotaxatie van gewelddadig gedrag bij jongeren. Zutphen: Rentray.
Lösel, F. & Farrington, D. P. (2012). Direct protective and buffering protective factors in the development of youth violence. American Journal of Preventive Medicine, 43(2), 8-23. doi:10.1016/j.amepre.2012.04.029
McIntosh, A., Lyons, J. S., Weiner, D. A., & Jordan, N. (2010). Development of a model for predicting running away from residential treatment among children and adolescents. Residential Treatment for Children & Youth, 27(4), 264-276.
doi:10.1080/0886571X.2010.520634
Merlo, J., Chaix, B., Ohlsson, H., Beckman, A., Johnell, K., Hjerpe, P., ... & Larsen, K. (2006). A brief conceptual tutorial of multilevel analysis in social epidemiology: using measures of clustering in multilevel logistic regression to investigate contextual phenomena. Journal of epidemiology and community health, 60(4), 290-297. doi:10.1136/jech.2004.029454
Moffitt, T. E. (1993). Adolescence-limited and life-course-persistent antisocial behavior: A developmental taxonomy. Psychological Review, 100(4), 674-701. doi:10.1037/0033-295X.100.4.674
Moffitt, T. E. & Caspi, A. (2001). Childhood predictors differentiate life-course persistent and adolescence-limited antisocial pathways among males and females. Developmental and Psychopathology, 13(2), 355-375. doi:10.1017/S0954579401002097
Mulder, E., Brand, E., Bullens, R., & Van Marle, H. (2011). Risk factors for overall recidivism and severity of recidivism in serious offenders. International Journal of Offender Therapy and Comparative Criminology, 55(1), 118-135.
Piotrkowski, C. S. & Baker, A. J. L. (2004). Predicting discharge disposition among adolescents in residential treatment. Residential Treatment for Children & Youth, 21(3), 69-87. doi:10.1300/J007v21n03_04
Van der Molen, E., Vermeiren, R., Krabbendam, A., Beekman, A., Doreleijers, T., & Jansen, L. (2015). Een bijzonder kwetsbare groep: Psychopathologie en sociaal functioneren van jonge vrouwen die geplaatst zijn geweest in een gesloten jeugdinstelling. Kind & Adolescent Praktijk, 14(1), 34-41. doi:10.1007/s12454-015-0009-4
Van der Ploeg, J. D. & Scholte, E. M. (2003). Effecten van behandelingsprogramma’s voor jeugdigen met ernstige gedragsproblemen in residentiële settings: eindrapport. Amsterdam: Nippo.
Verhulst, F. C., Van der Ende, J., & Koot, H. M. (1997). Handleiding voor de Youth Self- Report (YSR). Rotterdam: Afdeling Kinder- en Jeugdpsychiatrie, Sophia
Kinderziekenhuis/Academisch Ziekenhuis Rotterdam/Erasmus Universiteit Rotterdam.
Verhulst, F. C., Van der Ende, J., & Koot, H. M. (1996). Handleiding voor de CBCL/4-18. Rotterdam: Afdeling Kinder- en Jeugdpsychiatrie, Sophia
Kinderziekenhuis/Academisch Ziekenhuis Rotterdam/Erasmus Universiteit Rotterdam.
Vos, E. M. (2008). Uitval binnen de residentiële jeugdzorg. Op welke kenmerken verschillen de uitvallers van de reguliere beëindigers? (masterthesis).
Rennie, C. E., & Dolan, M. C. (2010). The significance of protective factors in the assessment of risk. Criminal Behaviour and Mental Health, 20(1), 8-22. doi:10.1002/cbm.750 Russel, M. A. & Marston, E. G. (2010). Profiles of mental disorder among incarcerated
adolescent females. Court Review: The Journal of the American Judges Association, 46, 16-23. Retrieved from
http://digitalcommons.unl.edu/cgi/viewcontent.cgi?article=1334&context=ajacourtrevi ew
Santrock, J. W. (2005). Adolescence. New York: McGraw-Hill.
Spice, A., Viljoen, J. L., Gretton, H. M., & Roesch, R. (2010). Psychological assessment for adult sentencing of juvenile offenders: An evaluation of the RSTI and the
SAVRY. International journal of forensic mental health, 9(2), 124-137. doi:10.1080/14999013.2010.501846
Stack, D. M., Serbin, L. A., Schwartzman, A. E., & Ledingham, J. (2005). Girls’ aggression across the life course: Long-term outcomes and intergenerational risk. In D. J. Pepler, K. C. Madsen, C. Webster, & K. S. Levene (Eds.), Development and treatment of girlhood aggression (pp. 253-283). New York, London: Psychology Press. Starkweather, J. & Moske, A. K. (2011). Multinomial logistic regression. Information
retrieved on 12-06-2016, from:
http://www.unt.edu/rss/class/Jon/Benchmarks/MLR_JDS_Aug2011.pdf Sunseri, P. A. (2003). Predicting treatment termination due to running away among
adolescents in residential care. Residential Treatment for Children & Youth, 21(2), 43-59. doi:10.1300/J007v21n02_03
Warr, M. (2007). The tangled web: Delinquency, deception, and parental attachment. Journal of Youth and Adolescence, 36(5), 607-622. doi:10.1007/s10964-006-9148-0
Wells, P. & Faragher, B. (1993). In-patient treatment of 165 adolescents with emotional and conduct disorders. A study of outcome. The British Journal of Psychiatry, 162(3), 345-352. doi:10.1192/bjp.162.3.345
Zegers, M. A. M., Schuengel, C., Van Ijzendoorn, M. H., & Janssens, J. M. A. M. (2006). Attachment representations of institutionalized adolescents and their professional
caregivers: Predicting the development of therapeutic relationships. American Journal of Orthopsychiatry, 76(3), 325-334. doi:10.1037/0002-9432.76.3.325
