The Pennsylvania State University

The Graduate School

College of the Liberal Arts

 

PREDICTING RECIDIVISM AMONG JUVENILE OFFENDERS PARTICIPATING IN AN ADVENTURE-BASED THERAPY INTERVENTION

 

A Thesis in

Psychology

by

Sandra L. Newes

 

ă 2001 Sandra L. Newes

 

Submitted in Partial Fullfillment

of the Requirements

for the degree of

 

Doctor of Philosophy

 

 

 

August 2001

 

 

 

We approve the thesis of Sandra L. Newes.

 

                                                                                                            Date of Signature

 

__________________________________________                        _______________

Thomas D. Borkovec

Distinguished Professor of Psychology

Thesis Adviser

Chair of Committee

 

 

__________________________________________                        ________________

Gordon C. Hall

Professor of Psychology and Crime, Law, and Justice

 

 

_________________________________________                          _________________

James T. Herbert

Professor of Education

 

 

_________________________________________                          __________________

Dene S. Berman

Clinical Professor, School of Professional Psychology

Associate Clinical Professor, School of Medicine

Wright State University

Special Member

 

 

_________________________________________                          __________________

Daniel F. Perkins

Associate Professor of Agricultural and Extension Education

 

 

________________________________________                            __________________

Keith A. Crnic

Professor of Psychology

Head of the Department of Psychology

 

 

ABSTRACT

 

 

Juvenile crime is a problem of national concern.  Given the high level of recidivism among this population, it is important to identify variables that may be predictive of recidivistic behavior.  The present study examined whether the predictive value for variables found previously in the non-intervention literature is the same or different in the context of an intervention designed to positively impact recidivism.  In addition, this study examined whether any additional predictive value was attributed to novel predictors. 

Specifically, this study investigated whether the demographic/ historic variables of age of first convicted offense and number of previous offenses, and the psychological characteristics of pre-test levels of MMPI-A scales 4 and 9, pre-test and adjusted post-test levels of the Tennessee Self-Concept Scale Total, and pre-test and adjusted post-test levels of the sensation seeking Disinhibition subscale were predictive of recidivism among juvenile offenders court mandated to an adventure-based therapy program.  Results indicated that age of first offense and the adjusted post-test Disinhibition subscale were significant predictors.  Examinations such as this have applied implications, as they may ultimately allow us the ability to more effectively provide individuals with the most appropriate treatment or correctional setting.   

 

 

 

 

 

 

TABLE OF CONTENTS

 

List of Tables…………………………………………………………………….       vi

Acknowledgements………………………………………………………………      vii

Chapter 1:  OVERVIEW…………………………………………………………      1

Statement of Purpose……………………………………………………..      1

 

Chapter 2:  THE PREDICTION OF RECIDIVISM……………………………...      3

            Prevalence of Recidivism…………………………………………………      3

            Historic and Demographic Variables……………………………………..       3

            Psychological Characteristics……………………………………………..      4

                        Personality…………………………………………………………    4

                        MMPI……………………………………………………………...   7

                        Self-concept……………………………………………………….    8

                        Sensation seeking………………………………………………….    12

 

Chapter 3:  ADVENTURE-BASED THERAPY AND ITS THEORETICAL

                   CONNECTIONS TO THE PREDICTOR VARIABLES…………….    17

            Adventure-Based Therapy and Recidivism………………………………..      18       

            Definition of Adventure-Based Therapy…………………………………..       19

            Characteristics of Adventure-Based Therapy……………………………...      21

                        Behavioral regulation………………………………………………     21

                        Risk appraisal………………………………………………………   22

                        Self-concept………………………………………………………..   22

            Baucom(1996):  Investigation of AT and recidivism……………………...        24

            Hypotheses…………………………………………………………………  29

 

Chapter 4:  METHODS……………………………………………………………   30

            Participants………………………………………………………………...    30

            Project Challenge…………………………………………………………..   30

            Procedure…………………………………………………………………..  32

            Recidivism Data……………………………………………………………    32

            Historic /Demographic Data……………………………………………….     33

            Measures…………………………………………………………………...  33

                        Minnesota Multiphasic Personality Inventory-Adolescent………..         33

                        Sensation Seeking Scale…………………………………………...    34

                        Tennessee Self-Concept Scale……………………………………..    35

            Analysis Plan………………………………………………………………    35

                        Preliminary Analyses………………………………………………     35       

                                    Gender, Ethnicity, and Recidivism………………………...      35

                                    Independent sample t-tests…………………………………    37

                                    Test of collinearity…………………………………………     37

                                    Analysis of covariance (ANCOVA)……………………….     38

                        Logistic Regression………………………………………………...    38

                        Interpretation of the Logistic Regression Analyses………………..        41

                                    Overall model……………………………………………..     41

                                    Individual predictors………………………………………     42

                                    Predictive accuracy for individual cases………………….        42

                        Linear Regression…………………………………………………     43

 

Chapter 5:  RESULTS……………………………………………………………..    44

            Preliminary Analyses………………………………………………………     44

                        Descriptive statistics……………………………………………….     44

                        Chi-square………………………………………………………….  44

                        Independent sample t-tests…………………………………………    46

                        Test of collinearity…………………………………………………     46

                        Analysis of covariance……………………………………………..    49

            Logistic Regression………………………………………………………..     49

                        Summary…………………………………………………………...   58

            Linear Regression………………………………………………………….    59

 

Chapter 6:  DISCUSSION…………………………………………………………   60

            Significant Results…………………………………………………………     60

                        Age of First Offense……………………………………………….    60

                        Adjusted Post-test DIS Scale………………………………………   67

            Non-Significant Results……………………………………………………     71

                        Tennessee Self-Concept Scale (TSCS)…………………………….    71

                        Number of previous offenses………………………………………     77

                        MMPI-A scales 4 and 9……………………………………………   83

                                    Sample differences…………………………………………    86

            Conclusions………………………………………………………………...   89

                        Future predictor models……………………………………………    90

            Study Limitations and Future Research……………………………………      92

            Importance of Conducting Rigorous Therapy Outcome Studies of AT……        98

 

REFERENCES……………………………………………………………………..  102

                       

                       

                                   

           

 

           

                       

                       

 

 

 

 

 

 

 

LIST OF TABLES

 

 

TABLE                                                                                                                     PAGE

  

    1                  Gender……………………………………………………………..   36       

 

    2                  Ethnicity……………………………………………………………   36

 

    3                  Comparison of Predictor Variables by Group……………………..      45

 

    4                  Comparison of Predictor Variables by Gender ……………………     47

 

    5                  Comparison of Predictor Variables by Ethnic Group……………..        48

 

    6                  Final Model Summary Statistics…………………………………...     51

 

    7                  Classification Table - Block 1, cut value=.5……………………….      52                                                                   

 

    8                  Classification Table – Block 2, cut-value=.5      ……………………..  53

 

    9                  Histogram of Estimated Probabilities:  Block 1……………………       54

 

    10                Histogram of Estimated Probabilities:  Block 2……………………       55

 

    11                Classification Table – Block 1, cut-value=.6………………………      57

 

    12                Classification Table – Block 2, cut-value=.6………………………      57

 

                                                                                                                                                                                                                                                                                               

 

 

 

 

 

 

 

 

 

 

 

 

 

 

ACKNOWLEDGEMENTS

 

 

            This has been a long and difficult road.  Getting to this point truly would not have been possible without the help and support of numerous people.  First, I’d like to acknowledge and send my most heartfelt thanks to my chair, Dr. Thomas Borkovec.  I am moved beyond words by what you have done for me.  You have taught me a tremendous amount, on many levels.  I would also like to thank Dr. Keith Crnic for your pivotal role in helping me to reach this point.  Without your help, this would not be happening.

            I would also like to individually thank each of my committee members.  Dr. Gordon Hall, thank you for your support, your expertise, your helpfulness, and your calming influence.  Dr. Jim Herbert, thank you for willingness to see this process through in its entirety.  Dr. Dene Berman, thank you for immediately lending your support, your knowledge, and your rationality, as well as for formally joining this process when you did.  Dr. Daniel Perkins, thank you for your novel perspective, your ideas, and for providing in several ways the means for me to continue moving forward. Genuinely, each one of you has played a vital and unique role in this process.  

            Friends have also helped to keep me going every step of the way.  Without Eric Helfen’s solid presence, again, this would not have occurred.  There are so many others it would be impossible to thank you all individually.  I am very fortunate. 

            Finally, thank you to my family.  Your complete love and support has been amazing, and your confidence has been unwavering.  Thank you for all of this.

 

Chapter 1

 

 

OVERVIEW

Chapter one of this review consists of an overview of the paper, including a statement of purpose and potential contributions of the present study.  Chapter two is a discussion of the prediction of recidivism, focusing on a review of the relevant literature specific to the variables utilized in this study including historic/demographic factors, personality, self-concept, and sensation seeking. Chapter three is primarily a discussion of the characteristics of adventure-based therapy most relevant to the goals of this study, including the targeting of behavioral regulation, improved risk appraisal, and self-concept change, and this chapter closes with the specific hypotheses of the study.  Chapter four delineates the methodology of the present study, including participant variables, program information, recidivism information, measure descriptions, and the analytic strategy.  The results are reported in chapter five.  The discussion of these results, including implications, future research suggestions, and study limitations, are presented in chapter six.  

 

Statement of Purpose

The present study was among the first to examine the combined effects of historic/demographic variables and psychological characteristics (including personality variables, self-concept, and aspects of sensation seeking) in predicting recidivism among juvenile offenders who have participated in an adventure-based therapy program. 

While numerous studies have shown historic/demographic variables to predict recidivism (age of first offense and number of previous offenses being the most robust predictors), and several studies have looked at the impact of psychological characteristics on the prediction of recidivism, no other study to date has examined the combined predictive value of these specific variables for juveniles who participate in a treatment intervention designed to decrease recidivism.  The present study developed a predictor model for recidivism in such an unexplored context by utilizing a sample of juvenile offenders mandated to an adventure-based therapy program.

To accomplish this goal, a predictor model for juvenile offenders involved in the Project Challenge Program (of Project Adventure), an adventure-based therapy program for juvenile offenders in the state of Georgia, was examined.  Specifically, this study investigated whether there are juveniles who are more or less likely to recidivate based on such potentially robust predictors as demographic/historic variables (age of first convicted offense and number of previous offenses) and pre-test levels of MMPI-A scales 4 and 9.  In addition, pre-test levels of the Sensation Seeking Scale Disinhibition subscale (SSS- DIS; Zuckerman, 1979), pre-test levels of the Tennessee Self-Concept Scale (TSCS; Roid & Fitts, 1994), and adjusted post-test scores for these variables were added to the model to determine if these variables added to the successful prediction of recidivism.  Examinations such as this have applied implications, as they may ultimately allow us the ability to more effectively provide individuals at high risk for recidivism with the most appropriate treatment or correctional setting.

 

 

 

 

 

Chapter 2

 

THE PREDICTION OF RECIDIVISM

               

Prevalence of Recidivism

 Juvenile crime is a problem of national concern.  To illustrate the magnitude of this problem, FBI crime reports indicate that in 1995, 2.5 million arrests were made of juveniles, and recidivism rates among such juvenile offenders have been shown to be approximately 15%.  Moreover, those persons involved in the juvenile justice system who had four or more referrals (high recidivists) also committed 50-60% of the most serious offenses (cited in Risler, 1998).  Given these figures, clearly juvenile recidivism carries a heavy societal impact.

Given the levels of recidivism among juvenile offenders, the prevention of repeated criminal behavior within this population is important in the overall reduction of criminal activity.  One approach suggested in the literature as a basis for creating prevention programs involves the development and testing of multidimensional predictor models of recidivism (Van Voorhis, 1994), models which incorporate distinct aspects of individual functioning.  The following sections focus on the prior empirical literature which has examined several such predictors of criminal behavior.

 

Historic and Demographic Variables

Several well-established historical and demographic variables are predictive of future recidivism.  Age of first arrest or conviction and number of offenses committed are reported as being among the most robust predictors of recidivism overall, with younger first time offenders being more likely to recidivate and the likelihood of recidivism increasing with each offense (Dembo, Wiliams, Getreu, Genung, Schmeidler, Berry, Wish, & LaVoie, 1991; Rutter, Giller, & Hagell, 1998; Van Voorhis, 1994; Zamble & Quincey, 1997).  Although these historic variables have been shown to be robust predictors of recidivism in a number of studies, it has also been noted that offenders are a heterogeneous group (Van Hooris, 1994).  Therefore, any prediction or classification model based on such historic factors alone may be overlooking important additional factors that may be impacting recidivistic activity.  One such domain that has been investigated involves individual differences in specific and theoretically relevant psychological characteristics.

 

Psychological Characteristics

            Personality

One psychological characteristic that has been linked to delinquency and recidivism is personality.  Findings indicate that those aspects of personality commonly associated with a psychopathic personality style are the most predictive of recidivism (Heaven, 1994, 1996; Webster & Jackson, 1997).  Such characteristics include hyperactivity, deviance, and impulsivity (Rutter, Giller, & Hagell, 1998), as well as aggression, egocentricity, antisocial behavior, lack of empathy, monotony avoidance, and poor socialization (Af Kernberg, Humble, & Schalling, 1992; Eysenck & Eysenck, 1985).

Other characteristics associated with psychopathic personality organization have been found in the literature to be linked with recidivistic activities as well.  In a longitudinal study utilizing a birth cohort, researchers found the above noted personality characteristics among adolescent recidivists, as well as a lack of harm avoidance (i.e., risk taking), a lack of traditionalism (i.e., social deviance), and negative emotionality, defined as “a tendency to experience aversive affective states such as anger, anxiety, and irritability” (Krueger, Schmutte, Caspi, & Moffitt, 1994, p. 187).  

Caspi, Moffitt, Silva, Stouthamer-Loeber, Kruger, and Schmutte (1994) provide an explanation of how the combination of the personality characteristics seen in the above review influence recidivism, suggesting that it is likely that “such individuals with chronically high levels of negative emotions perceive interpersonal events differently than other people” (p. 187).  They note that delinquent individuals are predisposed to construe events in a cognitively biased way, excessively perceiving threats in the acts of others.  According to this theoretical view, these types of biased perceptions contribute to further negative emotionality, creating a spiraling relationship of mutual influence between cognitively based misperceptions and negative emotion.  When combined with impulsivity (weak restraint), such negative emotions may be readily translated into criminally related activity for such persons (i.e., they are “more quick on the draw”; Caspi et al., 1994, p. 187). 

Wallbank (1985) adds to this formulation, postulating that the psychopathic personality style may be related to behavioral disinhibition.  He notes that “psychopaths are not driven to antisocial behavior by strong urges for money, sex, or violence but….given some strong inducement to respond, they have little capacity for behavioral inhibition” (p. 712).

While plausible, these formulations lack a clear explanation as to why such negative emotionality would lead to recidivistic activity, activity that typically carries a high level of associated risk.  Zuckerman (1994) discusses recidivism as being directly related to risk appraisal, arguing that appraisals of risk for activities typically decrease with repeated exposures.  Therefore, he argues that increased experience with risky criminal behavior (related to the above noted levels of impulsivity and negative emotionality among recidivists) leads to habituation.  Such habituation to the risk associated with criminal activities leads to a further lack of restraint, or behavioral disinhibition.  Zuckerman postulates that this process, based on the interaction of these factors, increases the probability of future criminal acts.    

As such, Zuckerman (1994) provides us with a theory that expands on the writings of Caspi et al (1994), particularly with regards to recidivistic activities.  In addition, Zuckerman’s theory of differential risk appraisal and behavioral disinhibition also provides us with an explanation for those demographic variables most predictive of recidivism noted above (i.e., younger age of first offense and number of previous offenses), as the presence of these demographic and historical factors are likely to be associated with more frequent exposures and thus with increased habituation to criminal acts.  These constructs of differential risk appraisal and behavioral disinhibition are re-visited in the later discussion of sensation seeking and delinquent behavior.

 

The earlier noted characteristics associated with recidivistic tendencies, specifically psychopathic personality organization and behavioral disinhibition, can be assessed by the Minnesota Multiphasic Personality Inventory (MMPI; Butcher, Williams, Graham, Tellegen, Ben-Porath, & Kraemmer, 1992) and the Sensation Seeking Scale (SSS; Zuckerman, 1979).  A discussion of the empirical work in this area using the MMPI follows.  As sensation seeking is also theoretically linked to the intervention, it is reviewed following the discussion of the last construct of interest in this investigation, self-concept and its relationship to delinquency.

 

MMPI

The MMPI is a widely used measure of personality.  The measure contains ten scales measuring clinical symptomatology, as well as validity, content, and supplementary scales.  Scale 4 (Psychopathic Deviate) and Scale 9 (Mania) of the MMPI clinical scales are commonly associated with psychopathic personality organization (Caid, 1986; Lindgren, Harper, Richman, & Stehbans, 1986), and elevations of scales 4 and 9 are often associated with criminal activity.  Not surprisingly, scales 4 and 9 have also been found in the literature to be the MMPI-A scales most commonly associated with delinquent behavior (Archer, 1992; Cashel, Sewell, & Hillmon, 1998; Pena, Megargee, & Brody, 1996; Zuckerman, 1994).  In addition, these scale elevations have been found to be predictors of recidivism (Weaver & Wooten, 1992). 

Interestingly, however, Lindgren, et al. (1986) examined the utility of scales 4 and 9 of the MMPI in predicting future problems, including but not limited to recidivism, for adolescents referred to a 30-day residential evaluation program.  These scales were unable to correctly classify these individuals into the high or low future problems group.  One reason for this contradiction may be based on the sample.  In the previously cited studies, the delinquent sample consisted of juvenile offenders, defined as those who had been involved in criminal activity and brought to the attention of the justice system.  In contrast, Lindgren et al.’s (1986) sample consisted of adolescents designated as “delinquent” or “a child in need of assistance.”  Thus, while this sample may have included some individuals who were actively involved in criminal activity, it may well be that this sample exhibited substantially less criminality than samples in previously cited studies. 

In addition, behaviors leading to a participant being classified as a member of the “continued problems” group included school expulsion and behavior problems, along with more criminally related activities.  This classification system may also have impacted the results, as it is possible that the group may have been more heterogeneous in its characteristics and its outcome measures than in previous studies.  However, their study raises questions about the ubiquitousness of scales 4 and 9 as predictors of recidivism, suggesting that it is important to look at whether these scale elevations have utility in predicting recidivism in different contexts and with different samples.  The present study was an effort to explore this possibility, where the context involved participation in a therapeutic intervention designed to reduce recidivism.

 

Self-concept

Self-concept is another psychological characteristic that has been found to be associated with delinquency.  The adolescent offender literature has shown repeatedly that juvenile offenders have lower self-concepts than do non-delinquent adolescents  (Byrd, O’Connor, Thackrey, & Sacks, 1993; Levy, 1997; Levy, Sullenberger & Vyas, 1991; Lund & Salary, 1980).  However, only a few studies have investigated the relationship between self-concept and recidivism.  These studies are briefly reviewed, and theoretical formulations regarding the connection between low self-concept and recidivism are also described. 

In a college sample, recidivists were found to have the lowest self-concepts compared to non-recidivists and first offenders (Fitts, 1965; Levy, 1997).  Relatedly, individuals with the lowest self-concept overall have been found to commit the most crimes as well as crimes of greater severity (Watson, 1979).  Some investigators have described what they call a “delinquent self-concept” (Byrd, O’Connor, Thackrey, & Sacks; 1993), defined as the discrepancy between an individual’s reported self-concept and their report of how they believe a delinquent person might respond to the same measure (i.e., what they believe a delinquent’s self-concept may look like).  Surprisingly, in a study utilizing such a formulation, it was found that frequent offenders (i.e., recidivists) had less delinquent self-concepts than did non-frequent offenders (Byrd et al., 1993).

In considering this potential discrepancy, it is of note that Byrd et al. (1993) operationalized self-concept using somewhat idiosyncratic measures.  Specifically, they utilized the Private Self-Consciousness subscale of the Self-Consciousness Scale (SCS; Fenigstein, Sheier, & Buss, 1975), a measure where participants endorse relevant self-referential statements, as “a measure of self-awareness” and theoretically associated scores on this measure with self-concept  (p. 196).  In addition, they used the Role Construct Repertory Grid (RCGS; Kelly, 1955) as a measure of self-concept that assessed “the important constructs that particular individuals use to make distinctions in the interpersonal world” (Byrd et al., p. 196).  As these measures operationalized self-concept in a unique way, it is potentially problematic to compare this study to those that have used recognized and well-established measures of self-concept.  In addition, the methodology of Byrd et al. involved comparing an individual’s self-report of their own self-concept with their self-report of what they believed to be a delinquent self-concept.  The degree of accordance between the two was conceptualized as an index of the delinquency of the individual’s own self-concept.  This is a different approach than those studies that have compared actual reported self-concepts between groups.   Therefore, such discrepant findings may be explained by differing methodologies.

            The theoretical connection between self-concept and delinquent behavior has been discussed in a number of ways.  Labeling theory proposes that once a person behaves in ways that society characterizes as deviant, he or she is then “labeled.”  If the individual chooses to accept and internalize the label, he or she is likely to continue to act in ways that are consistent with this label (Schur, 1973).  Recidivism may be related to the continuation of this process, whereby an individual’s self-concept becomes increasingly deviant as society continues to “label” him or her.  As this process continues, an individual is also more likely to associate with others who have similarly deviant self-concepts, increasing further the likelihood of criminal acts.

            Relatedly, consistency theory proposes that individuals are motivated to act in ways that maintain consistency in their beliefs about themselves and the world (Heider, 1958).  Thus, a person whose self-concept is deviant (based on labeling theory) is motivated to act in a deviant manner as a means of maintaining consistency in their self-concept.  In addition, acting in such a manner confirms his or her beliefs about his or her self and the world, another tenet of consistency theory.  Simplistically, “the juvenile offender sees himself as bad and worthless and he acts accordingly” (Fitts & Hammer, 1969, p. 132).

What is lacking in these theoretical formulations, however, is a clear explanation of why a poor self-concept may manifest itself in criminal behavior.  Oetting, Deffenbacher, and Donnermeyer (1998) discuss this issue, noting that it is neither the personality of an individual or his nor her self-concept that leads directly to delinquent behavior, but rather the ways in which these characteristics interfere with or influence primary socialization processes (e.g., peers, family, school) at all ages increase the likelihood of delinquent behavior.   Thus, an individual with a poor or deviant self-concept may be unable to engage in normal socialization processes.  Such a lack of socialization can lead to social marginalization and influence the process of “social norming.”  Therefore, he or she may be likely to further internalize negative or deviant attitudes. 

This again sets up a process of mutual influence between psychological characteristics and the environment.  Oetting et al. (1998) argue that given this process, such an individual may be more likely to engage in criminal acts, not only as a result of a poor self-concept but also as a result of the way in which this self-concept interferes with socialization.  Interestingly, it may also be true that personality interferes with the socialization process in a similar fashion.  Thus, this formulation may provide us with not only an explanation for the connection between self-concept and criminal activity but also with an additional theoretical postulation for the connection between criminal activity and personality.

Importantly, self-concept change is one characteristic of successful treatments for juvenile offenders (Greenwood, 1986).  As lower self-concept has been associated with higher recidivism levels, and positive change in self-concept has been found to be an element of a successful treatment, it is possible to speculate that change in self-concept during the course of an intervention may be an important predictor of recidivism.  The present study examined this question in the context of an adventure-based therapy intervention, an activity-focused treatment intervention designed to positively impact recidivism.   Importantly, one of the theoretical tenets of this type of intervention is that it impacts positively on self-concept, and thus it can be hypothesized that such change will occur.  This contention is revisited in the later discussion of adventure-based therapy.

 

Given that the previous discussion of psychological characteristics associated with recidivism noted a theoretical connection between behavioral disinhibition, differential risk appraisal, and recidivistic activity, the addition of these constructs to a multidimensional predictor model of recidivism was thought to potentially provide valuable additional insights into other factors that may be contributing to recidivism.  In addition, incorporating these constructs allowed for the development of a predictive model that encompasses several theoretical conceptions.  Therefore, a discussion of the association of sensation seeking, a personality construct that includes differential risk appraisal and behavioral disinhibition, with delinquency and potentially recidivism is warranted.

 

Sensation seeking

Sensation seeking, defined as “the need for varied, novel, and complex sensations and experiences and the willingness to take physical and social risks for the sake of such experience” (Zuckerman, 1979, p.10), has been shown in a number of studies to be correlated with delinquent behavior (Simo & Perez , 1991; Wallbank, 1985; Zuckerman, 1994), as well as with psychopathic behaviors and attitudes in non-delinquents (Levenson, Kiehl, & Fitzpatrick, 1995).  No research, however, has looked directly at the relationship of sensation seeking with recidivism.  This is somewhat surprising given that sensation seeking, as measured by the Sensation Seeking Scale (SSS; Zuckerman, 1979), has been modestly correlated with elevations on MMPI scales 4 and 9 (Montag & Birenbaum, 1986; Zuckerman, 1994), elevations previously noted as being associated with recidivism.  Zuckerman (1994) suggests that these associations between the MMPI scales and the SSS are due to the measurement of nonconformity, the need for excitement, and the need for high-energy activities directly assessed by both measures.  It is notable, however, that given the low correlations (.08-.25 for scale 4; .40-.47 for scale 9), it is likely that these measures are assessing related but separate constructs.  

Importantly, not only has sensation seeking been found to be associated with delinquency and those personality characteristics associated with delinquency, sensation seeking has also been found to predict delinquency (Farley, 1981; McGee, 1991; White, LaBouvie, & Bates, 1985).  These associations suggest the possibility that sensation seeking may also have utility in predicting recidivism. 

One potential explanation for the above findings may be related to behavioral disinhibition.  As noted above, tendencies toward behavioral disinhibition may impact criminal behavior (Zuckerman, 1994).  In addition, behavioral disinhibition has also been theoretically associated with psychopathic personality functioning (Wallbank, 1985) and differential risk appraisal (Zuckerman, 1994), both related previously to recidivism.  Notably, one of the subscales of the SSS, the Behavioral Disinhibition Scale (DIS) measures this construct explicitly.  In addition, the DIS scale also measures reported behaviors (e.g., drinking, gambling, criminal behavior, fast driving) and thus can also be conceptualized as a measure of risk-taking behaviors commonly associated with sensation seeking. 

Zuckerman (1994) theoretically associates sensation seeking and risk appraisal.  He postulates that sensation seeking leads to lower cognitive appraisals of risk, which in turn leads to greater risky behavior.  He notes that if no negative consequences are forthcoming, this process may result in an even lower risk appraisal in the future.  Therefore, Zuckerman believes that it may not be the risk itself that motivates high sensation seekers, but rather the differential appraisal of risk which may be contributing to a higher incidence of risky behavior among such persons.  Studies have supported this contention, finding that high sensation seekers tend to appraise situations as less risky than do non-sensation seekers, and sensation seeking was correlated with criminal risk-taking (r=.53, p<.001).  In addition, sensation seeking was found to predict 27% of the variance in criminal risk taking behavior (Zuckerman, 1994).   Thus, as with personality, the data suggest that risk appraisal is an important mechanism involved in the relationship of sensation seeking and recidivism.

In order to incorporate the constructs of risk taking and behavioral disinhibition into the model, the DIS subscale of the SSS was used in the present study.  Given the potential relationship of these constructs to recidivism, using the DIS subscale was a theoretically based way to examine the predictive utility of those specific aspects of sensation seeking most closely associated with recidivism.  Previous studies looking at the relationship between the DIS scale and delinquency lend support to this contention. 

In a number of studies, the DIS scale has been found to be the SSS subscale most strongly associated with delinquent behaviors and psychopathic personality characteristics (Blackburn, 1978; Daderman & af Klinteberg, 1997; Levenson, 1990; Levenson, Kiehl, & Fitzpatrick, 1995; Newcomb & McGee, 1991; Perez & Torrubia, 1985; White et al., 1985).  The DIS scale has also been found to predict future criminal behavior.  In a longitudinal study, Newcomb and McGee (1991) found that in a sample of high school students, scores on the DIS scale predicted both delinquent acts and lower “law abidance” four and five years later.  A second longitudinal study of adolescents reported similar results for the predictive value of the DIS scale on self-reported delinquency (White et al., 1985).  Strikingly, they found that DIS scores decreased among those participants who did not participate in delinquent acts.  In contrast, they found those who did participate in delinquent acts exhibited increases over time in the DIS scale scores.   

It is interesting to speculate as to the impact of such change in levels of behavioral disinhibition on recidivistic activity.   If high sensation seeking is a trait that is associated with delinquency, and the specific aspect of sensation seeking most commonly associated with criminal behavior is behavioral disinhibition, perhaps change in behavioral disinhibition among juvenile offenders may lead to lower levels of delinquent behavior.  The above findings of White et al. (1985) regarding the change over time of the DIS scale within their sample lend support to this statement. 

If it is possible that change in behavioral disinhibition can decrease criminal activity, it may also be possible that change in this measure over the course of a treatment intervention is predictive of future recidivism.  Interestingly, White et al. (1985) suggest that “involvement in delinquent activities represents an overt expression of a high level of underlying disinhibition needs” and  “a consideration of possible prevention programs should address the question of how to modify levels of disinhibition more directly” (p. 118). 

The present study investigated the hypotheses that pre-test levels of DIS add to the prediction of recidivism and that change in levels of behavioral disinhibition would be predictive of recidivism, utilizing pre-test scores on the DIS scale and adjusted post-test change scores over the course of the intervention on the DIS scale as predictors within the context of an adventure-based therapy program.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Chapter 3

 

ADVENTURE-BASED THERAPY AND ITS THEORETICAL CONNECTIONS TO THE PREDICTOR VARIABLES

 

Given the noted elevations in levels of sensation seeking and the increased levels of behavioral disinhibition among delinquent individuals, a treatment program that provides high levels of sensation and stimulation, while therapeutically targeting levels of behavioral disinhibition, may be an effective way to decrease criminal behavior in such individuals.  Therefore, it was useful to investigate the predictive value of the previously discussed demographic/historical variables and psychological characteristics on recidivism in such a context.  

Zuckerman (1994) notes that sensation seeking needs are “displaceable” and can be satisfied through a variety of activities.  Relatedly, Newcomb and McGee (1991) state that “If sensation seeking contributes to problem behaviors, then therapeutic environments and treatment plans could help to meet these needs in less injurious and perhaps more socially approved ways” (p. 615).  They also believe that among high sensation seekers, “one way to change behavior is through the substitution of stimulating activities” (Newcomb & McGee, 1991, p. 626).  Therefore, a treatment program that also directly teaches more socially appropriate ways of dealing with high sensation seeking needs (e.g., adventure activities) may provide additional benefits overall for such a high sensation seeking population. 

In addition, a treatment program that targets those constructs seen earlier in this discussion as being theoretically associated with recidivism (i.e., differential risk appraisal and self-concept) may also be an effective way to impact recidivism among juvenile offenders.  One type of treatment which attempts to incorporate each of these elements is adventure-based therapy (AT).  A brief introduction to adventure-based therapy, as well as a discussion of the characteristics of adventure-based therapy relevant to the predictors delineated within this study is now offered.

 

Adventure-Based Therapy and Recidivism

Adventure-based therapy (AT) is a type of intervention commonly used with juvenile offenders.  As it has been used as an alternative to juvenile incarceration, a number of studies have been conducted looking at the effectiveness of AT in preventing recidivism among juvenile offenders (Castellano & Soderstrom, 1992; Kelly & Baer, 1971; Minor & Elrod, 1990, 1992, 1994; Willman & Chun, 1973).  While the prevailing view in the AT literature appears to be that AT can effectively reduce recidivism rates (Bandoroff, 1989; Gillis, 1992; Gillis & Thompson, 1996; Hattie, Marsh, Neill, & Richards, 1997), there are serious methodological flaws inherent in all of the studies conducted in this area.  Consequently, we can make no empirically-based statements regarding the effectiveness of this type of intervention.  Well-controlled outcome research is necessary in order to begin answering such a question.   For a review of this literature, see Newes (2000).

Despite its lack of empirical validation, AT is a type of intervention whose usage is clearly on the rise.  For example, in 1993, a directory of experiential therapy and adventure-based counseling programs was published that listed 257 programs.  Given the increasingly widespread use of AT, particularly as an alternative to juvenile incarceration, looking at predictors of recidivism in such a context is an important avenue to explore.  Such an examination can begin allowing us to determine whether predictors of recidivism found previously in the non-intervention literature for juvenile offenders are or are not also predictors in this intervention context.  As no studies have looked at predictors of recidivism in such a framework with appropriate statistical analyses, the present study provides a valuable contribution to both the juvenile recidivism literature and the adventure-based therapy literature.  A definition of AT and an overview of those characteristics of AT theoretically associated with recidivism follows.

 

Definition of Adventure‑Based Therapy

Also referred to as "wilderness therapy," "therapeutic adventure,” “adventure therapy," and "adventure‑based counseling," AT is a therapeutic modality combining the therapeutic benefits of adventure experiences with those of more traditional modes of therapy.  While commonly associated with wilderness excursions, the wilderness component is not essential.  AT can take place in non-wilderness settings, and also can utilize adventure-based activities that are unrelated to a wilderness experience (e.g., ropes course activities, group initiatives).  Regardless of setting, all activities take place within a group context.  This necessitates that participants communicate, cooperate, and problem-solve together in order to successfully achieve group goals. 

Ringer (1994) more specifically defines AT as a class of change-oriented, group‑based experiential learning processes that occur in the context of a contractual, empowering, and empathic professional relationship.  Adding to this definition, as its basis AT uses planned activities chosen to facilitate individual or group level change in specific targeted areas.  A substantial emphasis is put towards reflecting on and processing the experiences, with therapist intentionality focused on therapeutic growth.  Individual and group psychotherapy sessions are also commonly incorporated into an AT intervention.  Elements of this definition are not unique to AT and can be assumed generally in many therapeutic traditions.  However, the emphasis on "group‑based experiential learning processes" in an active setting is a combination differentiating AT from other forms of therapy.  For a more comprehensive discussion of AT characteristics, the reader is referred to Newes (2000).

One key theoretical tenet of AT is that the combination of perceived risk in the activities and an unfamiliar environment leads to stress.  Gass (1993) refers to this as positive stress, or “eustress”.  It is believed that this eustress results in an intensive activation of interpersonal and intrapersonal processes.  This process is thought to unbalance the client, moving them away from their familiar behavioral templates and necessitating new behavioral choices.  Herbert (1996) postulates that this process is a mechanism of change in AT, noting that “Stressful experiences that are likely to occur throughout an adventure based program serve as impetus for individual change (p. 5).” 

A second key theoretical tenet of AT is that it provides an opportunity for self re-evaluation, and a means to disprove negative self-evaluations.  To accomplish this, a series of activities are chosen by the therapist.  The choice of particular activities is based on specified therapeutic goals and expected outcomes, and activities are appropriately sequenced by the therapist in order to ensure the maximum probability of success.  Successful achievement of an unfamiliar and difficult task, combined with therapeutic processing, is thought to provide a concrete means by which to negate a negative self-concept and increase self-esteem for participants.

As noted above, AT also contains elements that are directly related to constructs theoretically associated with delinquent behavior and recidivism, including behavioral regulation, risk appraisal, and self-concept.  As will be seen, these two primary theoretical tenets are at the basis of these elements as well.  These relationships will now be discussed in more detail.

 

Characteristics of Adventure-Based Therapy

Behavioral regulation

One characteristic of AT is the emphasis on learning behavioral regulation and thus decreasing behavioral disinhibition.  It is postulated that adventure activities, with the feedback and the consequences available through such experiences, provide learning that enables participants to begin regulating their own behavior (Bandoroff, 1989). Amesberger (1998) expands on this goal, noting that ultimately AT is aimed at facilitating:

“the reflection on internalized norms and values with the aim to support a person to find new and more suitable structures for his or her life.  Destructive and dysfunctional behaviors or emotions should be recognized in their effects, as well as helpful and effective ones (p. 29).”

 

AT researchers also believe that the group context of AT provides an opportunity for the receipt of peer feedback that can have increased meaning for an individual (Davis-Berman & Berman, 1994; Gass, 1993).  Thus, it is thought that this environment provides a powerful vehicle for the recognition of the effects of behavioral disinhibition as well as a potent medium in which to practice new ways of behaving (Davis-Berman & Berman, 1994; Gass, 1993).

 

Risk appraisal

Improved risk appraisal skills are also a goal of AT.  Challenges are often structured to appear impossible, dangerous, or risky to the group (Gass, 1993).  In actuality, the challenges are low in actual risk but high in perceived risk (Davis-Berman & Berman, 1994; Gass, 1993, Herbert, 1996, 1998), with the term "risk" referring to not only physical but also intra­- and interpersonal risk (Gass, 1993).  It is believed that involvement with activities that have a high level of perceived risk may provide an opportunity for delinquent adolescents to learn to more effectively appraise risky situations (Bandoroff, 1989).  Relatedly, Priest (1993) has suggested that participants will be able to influence their probability of success in an adventure experience if they have realistic perceptions of risk involved in the choices they make.  Given that inadequate risk appraisal skills are thought to be associated with criminal activity (Zuckerman, 1994), an intervention with this type of characteristic may therefore help to decrease recidivistic tendencies.

           

Self-concept

Change in self-concept is another therapeutic goal of AT (Bandoroff, 1989; Davis-Berman & Berman, 1994; Ewert, 1989).  Accomplished primarily through the provision of initial successes, or "mastery tasks,” AT is thought to foster feelings of capability and counteract negative self‑evaluations, learned helplessness, and dependency (Kimball & Bacon, 1993).  The achievement of an unfamiliar task, a task a client had previously thought him or herself incapable of, is believed to further facilitate this process (Davis-Berman & Berman, 1994; Gass, 1993).  Bandoroff (1989) argues that such achievements can provide tangible evidence to disprove negative self-conceptions.  In addition, such successes can provide concrete examples from which both therapist and client can draw from, thus allowing for the ongoing reinforcement of a more positive self-concept (Davis-Berman & Berman, 1994; Gass, 1993).

Generally, it is suggested in the AT literature that clients who make new behavioral choices in order to complete a novel challenge they had interpreted as carrying a high level of risk are thought to see themselves differently, with the ultimate goal being the recognition of their own self-imposed limitations and a change in their self image (Davis-Berman & Berman, 1996; Herbert, 1996, 1998).  Herbert notes that “It is the realization that persons challenge themselves and, in doing so, (re)learn something about themselves” that facilitates change (1996, p.5).   In addition, this change in competencies through the development of adventure-based skills may have the additional and related benefit of providing more socially acceptable outlets for sensation seeking needs.  This has been noted previously as a potentially important way to prevent criminal behavior among high sensation seekers (Newcomb & McGee, 1991; Zuckerman, 1994).  This process may also positively impact self-concept. 

 

Given these associations between AT and constructs associated with recidivism, it is possible that such an intervention may be useful in lowering recidivism rates.  Therefore, in an attempt to develop effective predictor models, it was important to look at whether previous predictors of recidivism found in the literature also have predictive utility in such a context.  It was also important to examine whether other theoretically based predictors could contribute additional predictive value to any such model.  Importantly, the exploration of questions such as this may ultimately allow for us to begin more appropriately providing offenders with the most effective treatment or correctional setting,

 

One study to date has attempted to utilize the above reviewed constructs in a predictive model for recidivism among juvenile offenders who have participated in an adventure-based therapy program.  As will be seen, however, the methodology is lacking in scientific rigor, and thus we can make no solid conclusions about the predictive value for recidivism of any of the variables examined in this study.

 

Baucom (1996): Investigation of AT and Recidivism

Baucom (1996), in an unpublished master’s thesis which utilized data employed in the present study, examined whether a variety of demographic variables and the difference scores from pre- to post-test on any of the MMPI-scales, the SSS total and subscales, the Tennessee Self-Concept Scale (TSCS; Roid & Fitts, 1994) total and subscales, and the BDI (BDI; Beck Depression Inventory; Beck, Rush, Shaw, & Emery, 1979) predicted recidivism.  The sample for Baucom’s study was comprised of juveniles court-mandated to an AT program.  Consistent with the findings examined in the earlier review of demographic variables, it was found that those participants who were re-convicted within three years (recidivists) were younger than non-recidivists at the start of the program.  In addition, Baucom (1996) found that recidivism was predicted by change in scale 9 of the MMPI (recidivists had a larger change score), change in the family-self subscale of the TSCS (recidivists had a larger change score), and a smaller change score on the SSS total.  Baucom reported that this combination of variables accounted for 15% of the variance in predicting recidivism.

Baucom’s (1996) use of change scores as predictors, however, presents a significant methodological difficulty.  While intuitively the use of difference scores appears to be a valid change measure, Cohen and Cohen (1983) argue against the use of such scores, stating that the difference between pre and post-test levels is a statistically uninterpretable and unreliable measure of change.  With that caveat in mind, an examination of the mean scores between groups provides a more complete picture of Baucom’s findings. 

While recidivists may have shown a greater difference between pre-test and post-test on scale 9, it is important to note that the recidivist group both started the program and ended the program with a higher mean score on this scale than did the non-recidivist group.  A similar pattern was found for the SSS total.  This suggests a need to control for differences in pre-test levels of severity, as the use of difference scores does not allow conceptually for a direct and meaningful comparison between groups.  On the TSCS family-self subscale, Baucom (1996) reported that recidivists had decreased scores at post-test than at pre-test, while non-recidivists showed increased scores on this scale over the course of the program.  Given this opposite direction of the change between groups, this comparison using difference scores also lacks meaning. 

These problems with Baucom’s (1996) use of difference scores suggest that another method is warranted.  An alternative approach, one that also allows for the conceptual examination of change over the course of the intervention (Tabachnick & Fidell, 2000), is to use post-test scores that are adjusted for pre-test levels of severity.  This is a more rigorous way to approach this data, and the present study employed such scores as predictors. 

Baucom’s (1996) study suffers from other methodological errors as well.  First, Baucom erroneously utilized multiple regression techniques in predicting a dichotomous outcome (i.e., recidivism vs. non-recidivism) and logistic regression is necessary to accurately examine this data (Hosmer & Lemeshow, 1989; Tabachnik & Fidell, 2000).

Second, Baucom did not include pre-test levels of any of these measures.  By including first in a regression equation pre-test scores for variables hypothesized (based on previous findings) to predict recidivism, the variance in prediction that can be accounted for by pre-test levels is controlled for prior to the examination of change.  Finally, by including such a high number of variables in the regression equation (28 total), Baucom introduced error variance into the equation and also reduced her statistical power (Tabachnik & Fidell, 2000).  Thus, it is possible to question whether her findings may be based on chance, and it is also possible that significant predictors may have been overlooked.  Limiting the number of variables is necessary to increase the power of the analyses.

The present study attempted to overcome Baucom’s (1996) methodological errors, thus building on her findings.  A more rigorous and conservative approach to the data was to utilize logistic regression with a limited number of pre-test and adjusted post-test variables for which empirical and theoretical support could be found within the literature.  Therefore, this study included first in the logistic regression equation pre-test levels of those measures one can expect to predict recidivism based on previous empirical findings.  After controlling for the variance attributed to these expected and robust predictors of juvenile recidivism, a limited number of theoretically-based post-test scores which were adjusted for pre-test levels of severity were then included.

 

In summary, the present study developed a predictor model using variables found previously in the literature as being predictive of recidivism in combination with variables theoretically linked to both recidivism and AT for individuals who participated in an AT intervention.  Through the evaluation of such a model, it was determined whether those variables found previously to be predictive of recidivism were also predictive in the context of an intervention designed to positively impact recidivistic behavior.  In addition, the examination of this model allowed us to determine if there was any additional predictive value for those predictors included in the model which remained as yet unexplored.  Ultimately, the overriding purpose of this study was to create a multidimensional predictor model for juvenile offenders particular to this context.   

Specifically, due to the predictive value for recidivism found previously in the literature for these variables, the demographic/historic variables of age of first convicted offense and number of previous offenses made up one component of the model and were hypothesized to predict recidivism.  For the same reason, pre-test levels of the MMPI-A scale 4 (Psychopathic Deviate) and scale 9 (Mania) were also hypothesized to predict recidivism and were added to the model.  These predictors were included first in the model in order to increase our confidence that the amount of variance in recidivism that was predicted by these potentially robust predictors was correctly identified.

Pre-test and adjusted post-test scores of the sensation seeking DIS scale were also used in the model.  As sensation seeking has been found to be predictive of delinquent behavior, and the DIS subscale been shown to be the specific aspect of sensation seeking most closely associated with such behavior, pre-test levels of the sensation seeking DIS subscale were used in the predictor model and were hypothesized to predict recidivism. 

In addition, given that the DIS scale measures behavioral disinhibition and risky behavior and that these constructs have been theoretically linked to recidivistic behaviors, it was possible to speculate that change in these constructs as measured by the DIS scale would be predictive of recidivism.  Moreover, as AT is theoretically associated with change in behavioral disinhibition and risk appraisal, it was reasonable to speculate that meaningful change in these areas may actually occur during the course of such an intervention.

Given these factors, adjusted post-test scores for the sensation seeking DIS scale were hypothesized to predict recidivism and were used in the present study as an additional dimension of the predictor model.  Through the utilization of pre-test levels of the DIS scale in the model as well, the variance in the adjusted post-test DIS scores that could be attributed to pre-test levels of this variable was accounted for.  Thus, any predictive value of change over the time of the intervention could be more clearly and rigorously examined.

 The final variables added to the predictor model were pre-test and adjusted post-test scores for the TSCS total.  Given the connection between delinquency and low self-concept, as well as the fact that self-concept is predictive of high vs. low levels of criminal activity, it was hypothesized that pre-test levels of the TSCS total would predict recidivism.  In addition, as self-concept may be associated with recidivism, and the AT intervention is theoretically postulated to impact self-concept, change in self-concept over the course of the intervention may be predictive of recidivism.  Thus, it was hypothesized that adjusted post-test scores for the TSCS total would be predictive of recidivism.  As with the DIS scale, the inclusion of pre-test levels of the TSCS in the model served the added purpose of allowing us to more clearly see the impact of the change in the TSCS from pre-test to post-test in predicting recidivism in this context.

The specific hypotheses for the present study are now presented.

 

Hypotheses

 

1) The demographic/historic variables of age of first convicted offense and number of self-reported previous offenses will be predictive of recidivism.

 

2) Pre-test levels of the MMPI-A scales 4 and 9 will be predictive of recidivism.

3) Pre-test levels of the DIS sub-scale of the SSS will be predictive of recidivism.

4)  Adjusted post-test levels of the DIS scale will be predictive of recidivism.

5)  Pre-test levels of the TSCS total will be predictive of recidivism.

6)   Adjusted post-test levels of the TSCS total will be predictive of recidivism.

                                                        

 

 

 

 

 

 

Chapter 4

 

      METHOD

 

Participants

Participants were 100 individuals (31 females, 69 males) who were court mandated to participate in the Project Challenge program at Project Adventure in Covington, GA, between 1992 and 1996.  They were 47% Caucasian and 44% African-American (ethnicity data were missing for 9 participants), with an average age of 14.82 (sd=1.05).  Forty-seven (16 female, 31 male) were reconvicted within three years following program completion, and were thus designated as "recidivists" (the remaining 15  females and 38 males were designated "non-recidivists"). Recidivists were 53% Caucasian and 36% African-American, and non-recidivists were 42% Caucasian and 51% African-American.  Power for the final regression analysis (alpha - .05; R2 = .2) was .87.

 

Project Challenge

The Project Challenge program of Project Adventure is a six-week (38 days) residential adventure-based therapy program for adjudicated juveniles.  The program was specifically designed as an adventure-based therapeutic alternative to juvenile incarceration and operates under contract with the Georgia Department of Youth Services (Gillis, 2000).  Participants who commit violent crimes and sex offenses are excluded from participation, as well as those with “extensive criminal records” (Terry, 2001).  

The primary goal of the program is to reduce recidivism.  Other therapeutic goals include the development of more positive attitudes towards community, family, peers, adults, and school (Schoel, Prouty, & Radcliffe, 1988).  In addition, the program attempts to increase self-control, self-esteem, self-discipline, decision-making skills, and problem-solving skills (Terry, 2001).  Five to six programs are run per year, and participants enter the program as a group (5 to 10 participants).  As such, participant groups in the present study both began and completed the program at the same time, and these groups remained intact for the duration of the program.   

Adventure-based activities believed to have therapeutic value comprised the primary form of intervention in Project Challenge.  These were employed continuously for the duration of the program.  Activities included team initiatives, ropes course experiences, camping, backpacking, hiking, and therapeutic games.  Upon completion, all activities were discussed within the group. As is consistent with a residential adventure-based therapy model, group and individual psychotherapy sessions were also conducted (Davis-Berman & Berman, 1994; Gass, 1993). 

The Project Challenge facilitators followed a specific, yet flexible protocol indicating the range of activities that were appropriate to be initiated at particular times/phases during the program.  Decisions as to the specific activity choice were based on the facilitators’ determination of where the group was in its developmental process.

During weeks one and three, participants camped the entire time in a remote site on the property of Project Adventure.  Adventure-based activities participants engaged in during this time included the learning of basic survival and camping skills, hiking, journal writing, and group initiatives thought to have therapeutic value.  During week five, participants went on a backpacking trip in a remote wilderness setting.  This trip was also one-week in duration.  

In weeks two, four, and six, participants lived together in a residential group home.  During this component, adventure-based activities were still utilized as the primary form of intervention, but the activities consisted more of ropes course work and group initiatives.  As is customary in any form of therapeutic residential program for youth (Hollin & Howells, 1996), academics were also incorporated into Project Challenge.  This occurred only during the time the time participants spent in the residential setting.  These academic sessions occurred 2-3 times per week for 2-3 hours per session.  Community-based tutors were employed to supervise the academic activities.

 

Procedure

            Participants were administered the pre-test battery of questionnaires within the first three days of entry into the Project Adventure program.  These questionnaires were administered in a cafeteria setting free from distractions.  Post-test questionnaires were administered in the same setting within the three days prior to program completion.  Questionnaires were scored by the Director of Project Adventure, a licensed clinical psychologist, or trained graduate assistants under his supervision.

 

 Recidivism Data

            Recidivism was operationally defined in the present study as conviction of a crime committed after the completion of the Project Challenge program.  Recidivism data for participants in Project Challenge were available from the state of GA for three years following program completion for each individual.

 

Historic/Demographic Data

Historic/demographic data examined for participants included number of self-reported previous offenses (including those for which the individual was arrested or charged and not convicted), and age of first convicted offense. 

 

Measures

Minnesota Multi-Phasic Personality Inventory- Adolescent

     The MMPI-A (Butcher, Williams, Graham, Archer, Tellegen, Ben-Porath, & Kraemmer, 1992) is a self-report measure developed for the measurement of personality and pathology in adolescents.  Normative data are available for individuals between the ages of 14-18 years old.    This 478-item measure is made up of four basic scales: validity, clinical, content, and supplementary scales.  These scales are then comprised of individual sub-scales.  This study used pre-test levels of two clinical scales, scale 4 (Pd; Psychopathic Deviate) and scale 9 (Ma; Mania). 

The MMPI-A overall has demonstrated acceptable test-retest reliability and internal consistency, comparable to the MMPI-2 (Butcher et al., 1992).  Test-retest correlations for scales 4 and 9 over a one-week interval were reported as .80 and .70 respectively (Butcher et al., 1992), and .68 and .54 over a one-year interval (Stein, McClinton, & Graham, 1998).  Coefficient Alphas for scale 4 were reported as .63 for boys and .68 for girls in the normative sample (Butcher et al., 1992).  For the clinical sample, Coefficient Alphas were reported as .53 for boys and .64 for girls.  For scale 9, Coefficient Alphas were reported as .61 for both boys and girls in the normative sample, and .59 and .67 for boys and girls respectively in the clinical sample.  The validity of the MMPI-A has been established in comparisons with the MMPI-2 and other well-established and psychometrically sound measures of personality and psychopathology (Butcher et al., 1992; Cashell, Rogers, Sewell, & Nolliman, 1998).

 

Sensation Seeking Scale

            The SSS (Zuckerman, 1979) measures an individual's tendencies toward sensation seeking and is based on theories of individual differences in optimal levels of stimulation and arousal (Zuckerman, 1994).  It is a 40- item self-report measure with forced choice responses grouped into four subscales.

The Disinhibition (DIS) subscale of the SSS was utilized in the present study.  The 10-item DIS subscale measures self-reported incidents of disinhibited behavior.  It has been related to more “traditional” forms of sensation seeking (e.g., drinking, partying, sex, gambling), as well as nonconformity to social codes and “sociopathy” (Zuckerman, 1979).  Internal reliability scores for the DIS subscale are reported to be .74-.78, and test-retest reliability is reported to be .91 (Zuckerman, 1994).  Validity of the DIS has been established through comparisons with a variety of personality scales, particularly with individual scales measuring extraversion and asocial tendencies; as well as nonconformist behavior scales and sociopathy scales.  Correlations with these types of scales are reported to be in the moderate to the high range (Zuckerman, 1994).

 

Tennessee Self-Concept Scale

     The Tennessee Self-Concept Scale (TSCS; Roid & Fitts, 1994) is a 100-item self-report measure of self-concept.  It has been shown to be applicable to clinical samples and yields a wide range of multidimensional information regarding an individual’s self-concept.  Test items are designed to tap self-perceptions from an internal frame of reference as well as an external frame of reference.  Validity data are also provided via 10 items which measure self-criticism, a factor which functions as a lie scale.  The internal frame of reference component measures three manifestations of an individual’s personal perception of self: identity, behavior, and self-satisfaction.  The external frame-of-reference aspect of the test identifies the extent to which an individual uses outside sources in forming his or her self-perceptions.  Roid and Fitts (1988) reported internal consistencies ranging from .70 to .87 for the TSCS, with an alpha coefficient of .91 for adolescents.  Studies have also demonstrated test-retest reliability and validity data, and published normative data are available (Roid & Fitts, 1988).  The TSCS total was used in the present study.

 

 

Analysis Plan

 

Preliminary Analyses

 

            The sample was divided into recidivist and non-recidivist groups and means and standard deviations were computed for each group on all variables. 

Gender, Ethnicity, and Recidivism

To determine if there were significant differences between the recidivist and non-recidivist groups based on gender or ethnicity, chi-square analyses using Fisher’s Exact Test were computed.  Significant differences between recidivists and non-recidivists on either of these dimensions were to have resulted in the significant variables(s) being included in the regression equation as a possible predictor of recidivism.  This did not occur.  The gender and ethnic composition of the recidivist and non-recidivist groups can be found in Table 1 and Table 2. 

 

 

Table 1

GENDER

 

Recidivist

Non-Recidivist

Total

Female

16

15

31

Male

31

38

69

Total

47

53

100

 

 

 

Table 2

ETHNICITY

 

Recidivist

Non-Recidivist

Total

African-American

17

27

44

Caucasian

25

22

47

Total

42

49

 

NOTE:  Information on ethnic background was missing for 9 participants

Independent sample t-tests

            Independent sample t-tests were conducted to explore differences based on gender or ethnicity for each of the predictor variables.  Additional independent sample t-tests were employed to examine differences between recidivists and non-recidivists on each of the predictor variables.

 

Test of collinearity

            The next step in the analytical strategy was to determine the magnitude of the relationships between the variables. To this end, a correlation matrix was developed using Pearson product moment correlations (Pearson r).  While the creation of this correlation matrix did allow for an exploration of variable relationships, this step was primarily intended to make certain the data did not violate the regression assumption of the absence of collinearity between variables.  In the case that two variables exhibited correlations of .90 or above, one of the related variables was excluded from the regression equation (Tabachnik & Fidell, 2000). 

In the present study, high correlations were discovered between the pre-test and adjusted post-test DIS scores (r = .96) and pre-test and adjusted post-test scores of the TSCS (r = .98), respectively.  This is reflective of the use of the pre-test variables in the creation of the adjusted post-test scores and necessitated the removal of one of each correlated variable pair.  Following the recommended strategy of eliminating the correlated variable that is of less relevance theoretically (Tabachnick & Fidell, 2000), the pre-test scores of both the DIS and the TSCS were dropped from further analyses.  The rationale behind this decision is explained in more detail in the results.  Given the removal of these variables from the analyses, there is no further mention of the use of pre-test DIS and pre-test TSCS scores as predictors in the discussion of the methods employed in the present study.  

 

Analysis of covariance (ANCOVA)

Adjusted post-test scores were created prior to conducting the regression analysis.  Using analysis of covariance (ANCOVA), post-test scores for the DIS and TSCS were adjusted using DIS and TSCS pre-treatment scores, respectively, as the covariates.  The covariates controlled for differences in pre-treatment levels of severity on these measures.  This procedure provided the individual adjusted DIS and TSCS post-test scores which were used as predictors in the regression equation.

 

Logistic Regression

As the next step in the analyses, a combined sequential and stepwise logistic regression was conducted.  These regression analyses allowed for the determination of whether those variables found previously in the literature to be predictive of recidivism were also predictive in the context of an intervention.  These analyses also allowed for the evaluation of any additional predictive value that was gained by adding previously unexamined predictors to the model.

Logistic regression was specifically required, as the hypotheses of the study called for the prediction of a dichotomous outcome with both continuous and nominal predictors (Hosmer & Lemeshow, 1989; Tabachnik & Fidell, 2000).  The combination of sequential and stepwise approaches in the logistic regression equation entailed the sequential entry of variables in two blocks, with a stepwise progression employed within each block.         

When using a sequential progression, the researcher specifies the order of entry of variables into the model. This can be done with individual predictor variables or with blocks of variables.  Typically, this ordering is based on theoretical postulations or existing empirical evidence, and the predictor or block of predictors expected to be the strongest is entered first.  Through this process, all the predictive variance associated with the first predictor variable or block is attributed to that variable.  Variables entered later in the sequence are attributed only the amount of predictive variance that they contribute to the model above and beyond the predictor or block that was entered prior.  Thus, any amount of shared variance is attributed exclusively to the earliest predictor or block entered into the equation. 

Alternatively, when employing stepwise procedures, each variable is entered into the model simultaneously, with the result being that no variable is given precedence over another variable within the block.  Inclusion and removal of predictors into the model is based solely on statistical criteria and a forward or backward stepwise progression can be specified.

When the more common forward progression is utilized, the variable with the most significance (i.e., the strongest predictor) within the block is entered first into the model.  This iteration is referred to as step 1 within the particular block.  Should it be found that other predictors within the block also meet statistical criteria, the most significant of the remaining variables is entered next into the model at step 2, and so on.  This iterative process continues until all predictors meeting the specified statistical criteria have been entered.  In contrast, when using a backward progression the models starts with all variables in the equation and variables are deleted one at a time if they do not meet the specified criteria.  Importantly, if each progression yields the same model, confidence in the results is increased.  This study utilized both a forward and a backward progression, and the model produced was equivalent. 

Specific to the methodology of the present study, the choice of a combined sequential and stepwise progression was appropriate for the following reasons:

1) As the historic and demographic variables of age of first offense and number of previous offenses have been identified previously in the literature as being robust predictors of recidivism, it was necessary to appropriately control for the amount of predictive variance associated with these variables prior to the evaluation of any of the other hypothesized predictors.  Thus, a sequential progression was appropriate, and these two variables were entered as the first block in the regression equation.  The remaining variables (pre-test MMPI-A scale 4, pre-test MMPI-A scale 9, adjusted post-test DIS scores, and adjusted post-test TSCS scores) were entered as the second block in the model, as there was no clear empirical or theoretical basis by which to support further sequencing of the variables.  

2) Although there was a basis in the literature for the aforementioned variable blocks, there was no hypothesized ordering of the predictors within each block.  Therefore, stepwise progressions were employed within the blocks in order to allow each predictor included within a specific block equal opportunity statistically for entrance into the model.

Overall, the sequential aspect of the logistic regression allowed for evaluation of the overall change in predictive utility of the model as it underwent a developmental progression between blocks, and also provided information as to the impact of individual predictors on the goodness of fit of the model.  The addition of stepwise procedures within blocks ensured that no variable within a block was given precedence over another in entering the model.  Instead, the variables in each block were entered into the model simultaneously (i.e., the predictors within each block started as statistically equivalent).

 

Interpretation of the Logistic Regression Analyses           

At each stage of variable entry, the predictive utility of the model was evaluated against the model that was presented just prior.  At the first iteration, the new model was compared to the null model.  At each additional stage of the model (i.e., upon the entry of a new block of predictor variables), the process described below for interpretation of the overall model and interpretations for each individual variable was repeated. 

 

Overall model

The first step in interpreting the model at each block was the evaluation of the overall goodness of fit.  To accomplish this, model chi-squares were examined to determine the significance level of the overall predictive model as compared to the null model.  Negelkerke R2  was also utilized as an approximate estimation of R2 , thus providing some indication of the amount of variance in prediction accounted for by the model.  In addition, as the model progressed (block 2), the significance level of the change in the likelihood ratios provided information regarding the level of improvement in the overall model as each block was added. 

 

Individual predictors

 The next step was to more closely examine the individual predictors included in a specific block.  This allowed for the determination of which predictors were the most influential (DeMaris, 1995; Tabachnick & Fidell, 2000).  Using maximum likelihood estimations, log odds were statistically transformed into odds ratios.  The examination of these odds ratios provided information as to the change in the odds of successfully predicting outcome with a one- unit change in the predictor.  In the instance that multiple significant predictors were found within each block, those statistically significant predictors that most impacted the odds of successfully predicting outcome at each stage were interpreted as the most important (SPSS Inc., 1999; Tabachnick & Fidell, 2000).   The final overall model included all predictors that yielded significant odds ratios. 

 

Predictive accuracy for individual cases

The final step in evaluating the model at each developmental stage was to determine the predictive accuracy of the model for individual cases.  This was determined through an inspection of a 2 x 2 case classification table.  The examination of this table provided information as to the percentage of actual cases correctly classified by the model, utilizing a standard default cut-off probability criterion of .5 for group membership classification (Tabachnick & Fidell, 2000). 

Upon completion of model development, further information as to the predictive accuracy of the overall model was gained through an examination of a histogram of estimated probabilities.  This histogram gave information regarding the actual cases, including their group classification (correct or incorrect) and the predicted probability score they received.  This form of individual case analysis also provided information about the clustering of individual cases at specific probabilities. 

The histogram revealed that there appeared to be a clustering of misclassified cases with an estimated probability of membership into their actual group of slightly over .5.  Given that .5 is simply an arbitrary default designation, in such an instance Tabachnick and Fidell (2000) recommend changing the cut-off criteria for group membership in the direction that increases accuracy of prediction and re-running the analyses.  In line with this recommendation, such a procedure was employed, and case classification results using both the .5 and the .6 probability cut-point for group membership are reported in the results.

 

Linear Regression

A final exploratory step in the analyses was conducted to examine whether any of the predictor variables were predictive of length of time until recidivism or differences in offense severity from initial offense to recidivism offense for the recidivist group.  Two linear regressions were employed:  (1) All of the predictors as independent variables and length of time until recidivism as the dependent variable, (2) all predictor variables as the independent variable and difference in offense severity as the dependent variable. 

 

 

Chapter 5

 

 

RESULTS

 

 

            A combined sequential and stepwise logistic regression analysis was performed to assess the predictive value for recidivism of a model which first included two demographic/historic predictors (age of first offense, number of previous offenses) and then added six psychological characteristic variables (pre-test MMPI-A scale 4, pre-test MMPI-A scale 9, pre-test and adjusted post-test DIS scores, pre-test and adjusted post-test scores TSCS scores).  Linear multiple regression analyses were also conducted to explore the usefulness of these variables for recidivists in predicting: 1) differences in offense severity from initial offense to recidivism offense, and 2) length of time until recidivism.  Analyses were conducted using SPSS, version 10.0.

 

Preliminary Analyses

            Descriptive Statistics

            Means and standard deviations for the recidivist and non-recidivist groups were calculated for each variable.  These are presented in Table 3.          

 

Chi-square

The chi-square analyses revealed no significant differences between recidivism groups on gender or ethnicity (gender: X2(1)= .59, p =.51; ethnicity: X2(1)=1.94, p =.21).  Thus, the regression analyses were collapsed across gender and ethnicity. 

 

Table 3

 

Comparison of Predictor Variables

by Group

Recidivists (N=47) vs. Non-Recidivists (N=53)

 

Mean

Std. Dev.

Std. Error

t-value

Sig. Level

Age of First Offense

R

14.68

.97

.14

-3.23

.002**

NR

15.38

1.16

.16

Number of Previous Offenses

R

4.72

2.97

.43

1.05

.293

NR

4.15

2.44

.34

Pre-test MMPI-A Scale 4

R

61.3

7.88

1.15

.98

.332

NR

59.58

9.49

1.3

Pre-test MMPI-A Scale 9

R

56.19

11.68

1.7

.693

.490

NR

54.64

10.68

1.47

Post-test DIS

R

6.47

1.07

.16

4.90

.000***

NR

5.53

.85

.12

Post-test TSCS

R

359.43

15.69

2.29

-1.23

.222

NR

362.85

12.11

1.67

              NOTE:  R = recidivist; NR= non-recidivist

                     *p < .05

                    **p < .01

                   ***p < .001

 

 

 

 

Independent sample t-tests

No significant differences between gender and ethnicity groups were found on any of the predictors that were ultimately included in the final model.  There were significant differences (t=3.17; p < .01) between African-Americans (mean=59.11, sd=12.13) and Caucasians (mean=51.79; sd=9.85) on pre-test MMPI-A scale 9 scores.  While statistically significant, these differences are commonly seen in the literature and recent meta-analytic findings indicate that these differences may not be substantive (Hall, Bansal, & Lopez, 1999).   Means and standard deviations for each variable by gender and ethnic group can be found in Table 4 and Table 5.

Additional independent sample t-tests were conducted to explore differences between recidivists and non-recidivists for each of the predictor variables (see Table 3).  Age of first offense (t=-3.23, p < .01) and post-test DIS scores (t=4.9; p<.001) were found to differ significantly between recidivists and non-recidivists, with recidivists having a lower mean age than non-recidivists and scoring higher on the post-test DIS subscale than non-recidivists.  This parallels the results of the logistic regression. 

 

Test of collinearity

As the next preliminary step in the analyses, a correlation matrix for all of the predictor variables was constructed using Pearson product moment correlations. This step allowed for the magnitude of the correlations between variables to be assessed in order to determine the existence of severe mulitcollinearity.  Examination of the correlations revealed that pre-test scores and adjusted post-test scores on the DIS subscale were highly correlated (r =.96, p < .001), as were the pre-test and adjusted post-test scores for the TSCS (r =.98, p < .001).  This is reflective of the use of the pre-test variables in the creation of the adjusted post-test scores.  Given that such powerful correlations violated the absence of collinearity regression.

Table 4

 

Comparison of Predictor Variables

by Gender

Males (N=69)  vs. Females (N=31)

 

Mean

Std. Dev.

Std. Error

t-value

Sig. Level

Age of First Offense

Male

15.00

1.08

.13

.38

.707

Female

15.10

1.21

.22

Number of Previous Offenses

Male

4.04

2.35

.28

1.89

.062

Female

5.13

13.22

.60

Pre-test MMPI-A Scale 4

Male

59.65

9.26

1.11

1.22

.225

Female

62.00

7.57

1.38

Pre-test MMPI-A Scale 9

Male

55.48

11.54

1.39

-.13

.899

Female

55.17

10.52

1.92

Post-test DIS

Male

6.00

.98

.12

-.38

.705

Female

5.91

1.25

.23

Post-test TSCS

Male

362.44

12.95

1.56

-1.18

.242

Female

358.85

16.06

2.93

                    NOTE:

                     *p < .05

                               **p < .01

                              ***p < .001

 

 

 

 

Table 5

Comparison of Predictor Variables

by Ethnic Group

African-American (N=44)  vs. Caucasian (N=47)

 

Mean

Std. Dev.

Std. Error

t-value

Sig. Level

Age of First Offense

African-American

15.23

1.08

.16

1.59

.115

Caucasian

14.84

1.22

.18

Number of Previous Offenses

African-American

4.64

2.58

.39

.72

.472

Caucasian

4.23

2.72

.40

Pre-test MMPI-A Scale 4

African-American

60.32

8.48

1.28

.23

.820

Caucasian

59.89

9.17

1.34

Pre-test MMPI-A Scale 9

African-American

59.11

12.13

1.83

3.17

.002**

Caucasian

51.79

9.85

1.44

Post-test DIS

African-American

5.80

.93

.14

-1.39

.167

Caucasian

6.12

1.23

.18

Post-test TSCS

African-American

361.99

16.25

2.45

.59

.555

Caucasian

360.19

12.61

1.84

                    NOTE:

                     *p < .05

                               **p < .01

                              ***p < .001

 

 

 

           

assumption, these findings necessitated the exclusion of one of each pair of correlated variables from the regression equations (Tabachnick & Fidell, 2000). 

It was determined that pre-test DIS scores and pre-test TSCS scores would be dropped from further analyses.  The decision to drop the pre-test scores instead of the post-test scores was warranted based on the following rationale:  Given that change in both the DIS and the TSCS was hypothetically linked to both recidivism and adventure-based therapy, the DIS and TSCS adjusted post-test scores were of more relevance theoretically to the goals of the study.  Thus, as both the adjusted post-test DIS subscale and the adjusted post-test TSCS scale were included in the study in order to test a theoretical postulation, findings related to these scores were potentially of more importance than those related to pre-test scores on these measures.  This strategy is in line with Tabachnick and Fidell’s (2000) suggestion of excluding variables with lesser theoretical significance in instances of mulitcollinearity. 

 

Analysis of covariance

            The adjusted post-test DIS and TSCS scores created through the ANCOVA procedure were saved and used in the regression equations as predictors.  

 

Logistic Regression

            The hypothesized predictors were entered in two blocks, and a forward stepwise progression using maximum likelihood ratio statistics (forward LR) was employed within each block.  A similar backward stepwise progression was also conducted and yielded equivalent findings. 

As it has been noted in the literature, standard significance levels of p <.05  are considered to be overly stringent for use in stepwise progressions.  Therefore, the statistical criterion for inclusion into the model was set at the suggested level of .15 (Bendell & Affifi, 1977; Hosmer & Lemeshow, 1989; Tabachnick & Fidell, 2000).  This level is recommended in order to ensure the entry of variables into the model with coefficients that are significantly different from zero.  In line with this suggestion, those variables with p < .15 were interpreted as contributing significantly to the predictor model in the logistic regression analyses.     

            At the first stage of model development, the demographic/historic variables of age of first offense and number of previous offenses were entered as block 1.  There was significant model fit overall (X2=10.06, df=1, p<.01), indicating that the model at this stage was able to significantly predict recidivism when compared to the null model.  Nagelkerke’s R2 (.128) indicated that the amount of variance in predicting recidivism that could be accounted for at this stage was approximately 13 %. 

Of the two hypothesized predictors entered in this block, age of first offense was a significant predictor (p < .01) while number of previous offenses was not significant (p <.29).  Removal of age of first offense significantly decreased the predictive utility of the model (p < .01) with a change in –2 Log likelihood of 10.06, df=1.  Odds ratios for this stage of the model showed that as age of first offense increases by one year (assuming that all other variables in the model stay the same), the odds of becoming a recidivist decrease by 54%.  Final model summary statistics can be seen in Table 6.

                                   

 

Table 6

 

Final Model Summary Statistics

 

 

 

 

VARIABLE NAME

BETA

STD. ERROR

SIG. LEVEL

ODDS RATIO Exp(B)

BLOCK 1

Age of First Offense

.568

.225

.012*

.57

Number of Previous Offense

 

.288

 

BLOCK 2

Pre-test MMPI-A Scale 4

.941

Pre-test MMPI-A Scale 9

.975

Post-test DIS scale

-1.05

.28

.000***

2.86

Post-test TSCS scale

 

.446

 

NOTE:

Final model chi-square (X2) = 28.97, df=2

p < .001

Nagelkerke’s R2 = .336

 

 

 

 

Utilizing the default probability of group membership cut-off criterion of .5 (SPSS Inc., 1999, Tabachnick & Fidell, 2000), a 2 x 2 case classification table was constructed, and the predicted group membership was compared to the actual reported group membership for each individual case.  Based on probability estimations, at this stage the model correctly classified 59.6% of the recidivists and 71.7 % of the non-recidivists, with a 66 % success rate overall (see Table 7).

 

 

Table 7

 

 

Classification Table- Block 1

Probability of group membership cut-value=.5

 

PREDICTED GROUP

Percentage Correctly Classified

Recidivist

Non-recidivist

 

ACTUAL

GROUP

Recidivist

28

19

59.6%

Non-Recidivist

15

38

71.7%

                                                                                                        Total          66.0%

 

 

At the second stage of model development (block 2), pre-test scores on MMPI-A scale 4, pre-test scores on MMPI-A scale 9, post-test DIS scores, and post-test TSCS scores were entered into the model.  As hypothesized, post-test DIS scores were found to significantly improve the model (X2= 6.81, p <.01), and Nagelkerke’s R2 (.336) showed that the model at this step accounted for approximately 34 % of the variance in predicting outcome.  The predictive power of the model is significantly decreased with the removal of this variable (p <.001), showing a change in –2 Log likelihood of 18.91 (df=1).  Odds ratios indicate that the odds of becoming a recidivist increase by a factor of 2.8 per one unit increase in post-test DIS scores.  The odds ratios for age of first offense shift slightly with the addition of the post-test DIS variable to the model, with the odds of not becoming a recidivist changing from 54% to 57% as first offense occurs one year later.  Contrary to the stated hypotheses, pre-test scores on MMPI-A scale four (p < .95), pre-test scores on MMPI-A scale 9 (p< .96), and post-test TSCS scores (p< .45) were not significant predictors (see Table 6).  

With the addition of the post-test DIS scores, the full model was able to correctly classify 63.8 % of recidivists and 77.4 % of the non-recidivists with an overall predictive accuracy rate of 71% (see Table 8).  An examination of the histogram of estimated probabilities for both Block 1 and Block 2 revealed a slight clustering of non-recidivists with estimated probabilities of successfully predicting group membership just above .5 (See Table 9 and Table 10).  

 

 

 

Table 8

 

 

Classification Table- Block 2

Probability of group membership cut-value=.5

 

         PREDICTED GROUP

Percentage Correctly Classified

Recidivist

Non-recidivist

 

ACTUAL

GROUP

Recidivist

30

17

63.8%

Non-Recidivist

12

41

77.4%

                                                                                                                Total           71.0%                                                                                                               

 

 

 

 

 

Table 9

 

Histogram of Estimated Probabilities:  Block 1             

 

       Observed Groups and Predicted Probabilities

 

       8 ô                            n                               ô

         ó                            n                               ó

         ó                            n           n                   ó

F        ó                            n           n                   ó

R      6 ô                            nn          n                   ô

E        ó                            nn          n                   ó

Q        ó                       n    rn n        n                   ó

U        ó                       n    rn n        n                   ó

E      4 ô                       r n  rr nnnn     n    n              ô

N        ó                       r n  rr nnnn     n    n              ó

C        ó              n    r   r n  rrrnnnnnnn  r nn nn             ó

Y        ó              n    r   r n  rrrnnnnnnn  r nn nn             ó

       2 ô             rn  r r   rnrrnrrrrrnrrnn nrnnnnnn             ô

         ó             rn  r r   rnrrnrrrrrnrrnn nrnnnnnn             ó

         ó         n rnrrn r r   rrrrrrrrrrrrrrnnnrnrnnrr   n         ó

         ó         n rnrrn r r   rrrrrrrrrrrrrrnnnrnrnnrr   n         ó

Predicted ňňňňňňňňňňňňňňôňňňňňňňňňňňňňňôňňňňňňňňňňňňňňôňňňňňňňňňňňňňňň

  Prob:   0            .25            .5             .75             1

  Group:  rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn

 

          NOTE:

          Predicted Probability is of membership for non-recidivist

                    The Cut Value is .50

                    Symbols: r - recidivist

                                    n - non-recidivist

                    Each Symbol Represents .5 Cases.

 

 

 

 

 

 

 

 

 

 

 

 

Table 10

 

 

Histogram of Estimated Probabilities:  Block 2

 

 

             Observed Groups and Predicted Probabilities

 

       8 ô                                                            ô

         ó                                                            ó

         ó                                                            ó

F        ó                                                            ó

R      6 ô                                                            ô

E        ó                                                            ó

Q        ó                                          n                 ó

U        ó                                          n                 ó

E      4 ô                                          n                 ô

N        ó                                          n                 ó

C        ó     r       r  nn n        n      n nn n n nnn             ó

Y        ó     r       r  nn n        n      n nn n n nnn             ó

       2 ô     r    rn r  nr rn n n nrr nnn rn nnnrnrnnnnn  nnn n     ô

         ó     r    rn r  nr rn n n nrr nnn rn nnnrnrnnnnn  nnn n     ó

         ór r rrrrrrrrnr rrr rr rnr rrr rrrnrrrnrrrnrrrnrn nnnnnnrnnn ó

         ór r rrrrrrrrnr rrr rr rnr rrr rrrnrrrnrrrnrrrnrn nnnnnnrnnn ó

Predicted ňňňňňňňňňňňňňňôňňňňňňňňňňňňňňôňňňňňňňňňňňňňňôňňňňňňňňňňňňňňň

  Prob:   0            .25            .5             .75             1

  Group:  rrrrrrrrrrrrrrrrrrrrrrrrrrrrrrnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn

 

     NOTE: 

          Predicted Probability is of membership for non-recidivist

          The Cut Value is .50

          Symbols: r - recidivist

                          n - non-recidivist

          Each Symbol Represents .5 Cases.

 

        

Following the recommendations discussed in the methods chapter for such an instance (Tabachnick & Fidell, 2000), the cut-off criterion was changed to .6, and the analyses were re-run.  Applied implications of changing this criteria are typically based on the costs of making type I versus type II errors (Tabachnick & Fidell, 2000).  Stated another way, were there greater consequences for making one type of error versus another (e.g., misclassifying a non-recidivist as a recidivist versus overlooking a potential recidivist), determinations as to the most appropriate probability cut-point for individual cases would be made based on these types of considerations (e.g., with higher consequences, a cut-point resulting in more conservative probability estimations being used to determine group membership in either direction would be warranted).  This type of instance is most commonly seen in applied medical research, where the consequences of misdiagnosis may be quite high, and is considered more an exploratory step in experimental research (Tabachnick & Fidell, 2000).   

Returning to the results of the analyses, at Block 1 a re-examination of the case classification results when using a cut-off criteria of .6 indicated that the overall successful prediction rate was not substantially altered, exhibiting a 1% increase in accurate prediction overall (67%).  However, the number of cases correctly classified was distinctly different across groups.  The accurate prediction of recidivism was shown to increase, rising from 59.6% to 85.1%.  Alternatively, the accurate prediction of membership into the non-recidivist group decreased, falling from 71.7% to 50.9% (see Table 11).

A similar pattern was seen at Block 2, with the accurate prediction of recidivism occurring 63.7% of the time when using a .5 probability cut-off, and increasing to 76.6% when using a .6 cut-off.  The correct prediction of non-recidivism at Block 2 decreased when using a .6 cut-off, declining from 77.4% to 66%.  Interestingly, at Block 2 the prediction rate remained the same when the probability cut-off was changed (71%).  Changing the cut-off criteria did not significantly alter the overall predictive accuracy, but rather shifted the accuracy between groups.  The classification table for the full model using the .6 probability cut-off can be found in Table 12.

 

Table 11

 

 

 

Classification Table- Block 1

Probability of group membership cut-value=.6

 

PREDICTED GROUP

Percentage Correctly Classified

Recidivist

Non-recidivist

 

ACTUAL

GROUP

Recidivist

40

7

85.1%

Non-Recidivist

26

27

50.9%

                                                                                                                Total           67.0%                                                                                                                  

 

 

 

 

 

 

Table 12

 

 

 

Classification Table- Block 2

Estimated probability of group membership cut-value=.6

 

         PREDICTED GROUP

Percentage Correctly Classified

Recidivist

Non-recidivist

 

ACTUAL

GROUP

Recidivist

36

11

76.6%

Non-Recidivist

18

35

76.6%

                                                                                                                Total          66.0%                                                                                                               

Studentized residuals and Cook’s distance for individual cases were examined to explore the potential influence of outliers in the data.  One case exhibiting a studentized residual above two standard deviations (Std. R = 2.34) was identified.  Goodness of fit statistics for the final model were recomputed with this case excluded.  Model fit did increase somewhat (X2=22.28, p <.001), and Nagelkerke’s R2 (.379) indicated that the absence of this outlier increased the amount of variance in predicting outcome accounted for by the final model from approximately 34% to approximately 38%.  Close examination of the data for this case provided no reason to suspect data inaccuracies (Tabachnick & Fidell, 2000).  Thus, the more conservative approach of retaining the case in the final analyses was warranted. 

 

Summary

The final model with all significant predictors included age of first offense and post-test DIS scores, and was shown to account for approximately 34% of the variance in predicting outcome (Nagelkerke’s R2=.336).  Based on these findings, the final model would predict that those more likely to recidivate would commit their first offense at a younger age and show higher DIS scores upon completion of the intervention than non- recidivists. 

With this two variable model, the data indicate that successful prediction of group membership occurs 71% of the time overall.  This overall percentage does not change based on estimated probability cut-off point.  Changing the cut-off probability criteria for group membership from .5 to .6 enhanced correct prediction into recidivist group while decreasing accurate prediction of the non-recidivist group at both stages of model development. 

 

Linear Regression

            Prior to conducting the regression analyses, Levene’s test for Equality of Variances indicated the predictors were free from heteroscedasticity.  Individual regression analyses with length of time until recidivism and differences in offense severity as dependent variables yielded no significant predictors. 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Chapter 6

 

 

 

DISCUSSION

 

 

            The results of this study revealed that age of first offense and adjusted post-test Disinhibition (DIS) subscale (Zuckerman, 1979) scores were significant predictors of recidivism.  Minnesota Multiphasic Personality Inventory- Adolescent (MMPI-A; Butcher, et al., 1992 ) scales 4 and 9 and number of previous offenses were not significant predictors.  As was noted in the results, pre-test DIS scores and pre-test Tennessee Self-Concet Scale (TSCS; Roid & Fitts, 1994) scores were excluded from the final analysis for reasons of multicollinearity.

 

Significant Results

Age of first offense

The results showed that age of first convicted offense was a significant predictor, indicating that those persons who committed their first offense at a younger age are more likely to recidivate.  Based on probability estimations of .5, the predictor model which included this variable alone was able to correctly predict the group membership of 66% of all participants, with a higher success rate for non-recidivists (71.7%) than recidivists (59.6%).  Overall, the model at this stage of development accounted for 13% of the variance in predicting outcome.  This replicates previous findings in the literature, and age of first offense is commonly thought to be one of the most robust predictors of recidivism (Andrews & Bonta; 1994; Dembo et al., 1991; Draine, Solomon, & Myerson, 1994; Gendreau, Little, & Goggin, 1996; Katsiyannis & Archwamety, 1997; Myner, et al., 1998; Nagin & Farrington, 1992b; Nagin & Paternoster, 1991; Tolan & Thomas, 1995). 

Importantly, this is the first time that age of first offense has been examined as a predictor of recidivism for juveniles who participated in an intervention.  The fact that the predictive utility of this variable persisted in spite of intervention efforts has applied implications.  In particular, this finding highlights the importance of youth prevention programs that delay or deter age of onset.  If such programs can have immediate benefit in delaying criminal involvement, long-term societal benefit might occur through the potential reduction of overall frequency and intensity of criminal behavior among such offenders (Nagin & Farrington, 1992a).  Future research into the factors that underlie the relationship between age of first offense and recidivism is necessary in order to increase the effectiveness of such programs.  This contention will be explored further below.

Given that both the present study and previous research indicate that age of first offense has predictive utility for recidivism, future efforts at developing predictor models for juvenile recidivism would be remiss in not including this variable.  Clearly, younger age of first offense is a significant risk factor and warrants active attention in applied settings.  Significant resources should be employed in an attempt to avert future offenses with individuals who are convicted of their first criminal act at a young age. It has been suggested that involving family members, schools, and community organizations in intervention efforts may prove may improve success rates for such individuals (Smith & Aloisi, 1999; Tollett & Benda, 1999). 

An examination of several theories explaining the relationships between age of first offense and recidivism will follow.  These theories will be also referenced in the later discussion of other results of the study.  Through this theoretical analysis, specific areas for future research will be highlighted.  A discussion of the applied significance of such future research in designing effective prevention and intervention programs will also be offered.   

Theoretical explanations of the relationship between age of first offense and repeat criminal activity tend to focus primarily on three postulations:  1) delinquent behavior is related to stable characteristics, suggestive of a delinquent population predisposed to criminal activity, 2) dynamic and situational (proximal) characteristics lead to earlier criminal behavior, or 3) early criminal behavior is causally linked to future criminal behavior (i.e., past involvement increases the possibility of future involvement) (Nagin & Paternoster, 1991; Nagin & Farington, 1991; Tolan & Thomas, 1995).  Although it is noted in the literature that these three explanations are “distinctly different” (Nagin & Paternoster, 1991; p. 165), definitive support for one above the others is lacking (Tolan & Thomas, 1995). They are perhaps not mutually exclusive (Nagin & Farrington, 1992a; Nagin & Farrington, 1992b; Nagin & Paternsoter, 1991; Smith et al., 1991), and it may be that each of these influences interactively impact criminal behavior (Nagin & Farrington, 1992b; Smith et al., 1991).  Future research using predictor models that include variables which follow these theoretical lines is one way to approach this question.  This will be further discussed. 

Those theorists ascribing to the first school of thought believe that juveniles who commit offenses at a younger age may simply be more predisposed to criminal activity (Gotttfredson & Hirschi, 1990).  They contend that this predisposition remains stable over the life course, associating age of first offense and recidivism in this manner (Gotttfredson & Hirschi, 1990; Wilson & Hernstein, 1985).  Wilson and Hernstein note that “the offender offends not just because of immediate needs but also because of enduring personal characteristics” (p. 209).  Other evidence has also been found for this postulated predisposition towards criminal behavior, relating it primarily to lack of control (i.e., impulsivity) and risk taking (Gotttfredson & Hirschi, 1990; Nagin & Farrington, 1992b; Wallbank 1985; Zuckerman, 1994). 

If indeed the predictive power of age of first offense is simply indicative of a criminal predisposition which leads to earlier participation in criminal acts as well as repeated offending, it has been suggested that the utility of this predictor is limited to the identification of high risk individuals (Gottfredson & Hirschi, 1990; Nagin & Farrington, 1992b; Tolan & Thomas, 1995).  It has also been noted that intervention with such individuals will only be effective to the extent that it keeps them “out of harms way” for some period of time (Gottfredson & Hirschi, 1990; Nagin & Farrington, 1992b; Tolan & Thomas, 1995). The fact that the present study found age of first convicted offense to be predictive despite participation in an intervention can perhaps be seen as support for such a stable criminal disposition resistant to treatment efforts.

However, such a statement overlooks the fact that any existing criminal

predisposition that underlies the relationship between age of first offense and recidivism would also involve a set of personal characteristics.  Hence, instead of applying this theoretical contention in this seemingly deterministic fashion (i.e., those who commit their first offense at an earlier age are predisposed to offend again) and concluding the investigation there, this postulation of a criminal disposition can instead be viewed as a theoretical basis for future research into the component characteristics.

Without this contention having been explored, it is possible that the intervention  may not have impacted individuals with recidivistic tendencies in the most effective manner.  This is an alternative explanation for the findings of the present study with regards to age of first offense, one which highlights again the importance of such research.

The second theory relates age of first offense and recidivism via situational and environmental characteristics. While this seemingly does not preclude the above noted predisposition towards criminal behavior, this theory would associate age of first onset with recidivism through more dynamic and proximal factors such as peer influence, family factors, SES, and substance abuse.  Support has been found in the literature for this contention as well, although it is unclear whether these factors are linked causally to early criminal acts (Hollin & Howells, 1996; Nagin & Paternoster, 1991; Nagin & Farrington, 1992a, Nagin & Farrington, 1992b; Rutter, Giller, & Hagell, 1998; Tolan & Thomas, 1995). 

While the findings from the present study may be seen as supporting theory one, it is also possible that environmental factors impacted recidivism as well.  Moreover, there may have been an interaction between any existing criminal predisposition and environmental factors.  That is, such factors could be more likely to lead to criminal behavior for individuals with a criminal disposition.  While the empirical basis for such speculation is beyond the scope of this study, it is an important area for future research.

The third theoretical frame is that committing criminal acts at a younger age has a direct causal influence on future criminal behavior.  More specifically, the act of committing the first crime may have “a behavioral influence in the sense that the experience of crime increases the likelihood of further offending by changing something about the offender’s personal characteristics or life chances” (Nagin & Paternoster, 1991, p. 166).  Theories associating age of first offense with habituation to criminal acts or a disinhibitory behavioral response would fall under this umbrella.  However, causal links between specific psychological processes and criminal behavior remain as yet unproven (Nagin & Farrington, 1992a).  Future research is required in order to explore the directionality of any such relationship (i.e., does criminal behavior lead to psychological processes that generate further criminal behavior, or do some specific psychological processes lead to criminal behavior). The connection between criminal behavior, behavioral disinhibition, and risk appraisal will be explored further in the discussion of the findings relating to the DIS scale.

Empirical investigation designed to identify the specific mechanisms by which age of first offense is related to recidivism can provide evidence in support of one or more of the theories discussed earlier.  As each of these theories suggest different factors underlying this relationship, future research is necessary to determine specifically what these factors may be.  It may be that a combined theoretical approach will prove most efficacious.

Continued exploration of predictor models including variables that are based the above theories is one way to approach future research.  Such models can, first, include individual characteristics, dynamic external characteristics, or psychological processes exclusively in order to test these differing theoretical postulations.  Multidimensional models that incorporate all of these characteristics should also be examined to test for effects of a combined model.  Furthermore, including age of first offense in these models at different points in the regression would determine the variables’ predictive value in the presence of other related predictors. 

In addition, process research may prove helpful in delineating the specific components of the AT intervention, along with how they might impact on any of these identified characteristics.  Such process research can then investigate the mediational relationship of targeted psychological processes to outcome, as well as testing the efficacy of specific treatment components.

Empirical results from this type of investigation have important applied implications.  Findings can be used to identify high-risk youth prior to their having committed the first offense, with the goal being to monitor such individuals and channel them into appropriate prevention programs.  Such findings can also be applied to the ongoing development and improvement of prevention programs that may deter or delay first criminal acts by actively impacting on those characteristics.  Additionally, the identification of protective or inhibitory factors for criminal behavior is also important, as prevention programs can attempt to bolster such factors.  Research using samples of juveniles exhibiting high-risk characteristics who do not engage in delinquent activity is one way to approach this issue.   

Importantly, this study indicates that age of convicted first offense is predictive of recidivism even in an intervention setting.  Thus, future investigation of individual characteristics or environmental influences that may moderate the effects of this variable on recidivism for juveniles who have already engaged in criminal activity is also necessary.   Should such moderators be discovered, intervention programs can incorporate specific components designed to impact on these areas (e.g., should family relationships be found to be a moderator of recidivism for young offenders, intervention programs can incorporate or strengthen a component focused on this area).  Ideally, such a process will result in intervention programs that are more effective at reducing recidivism for individuals who are convicted of their first offense at a young age.  

 

Adjusted post-test DIS scale

It was also found in the present study that adjusted post-test DIS scores were predictive of recidivism.  According to the model, the odds of becoming a recidivist increased with higher adjusted post-test DIS scores, subsequent to the passage of time and the occurrence of an intervention.  The participant group as a whole decreased significantly in DIS over the course of the intervention (F(1,98)=9.87, p<.05), but recidivists ended the program with significantly higher DIS scores than did non-recidivists (t=-3.23, p =.05) .  The addition of DIS scores significantly increased the amount of outcome variance predicted by the model over that predicted by age of first offense from 13% to 34%.  Classification of participants at this point in model development, with a probability of group membership cut-point of .5, was 71% correct overall.  For recidivists, the model successfully predicted group membership for 63.8% of participants.  For non-recidivists, 77.4% were correctly classified.      

As previous research has found relationships between the DIS scale and delinquency (Blackburn, 1978; Daderman & afKlintenberg, 1997; Levenson, 1990; Levenson, Kiehl, & Fitzpatrick, 1995; Newcomb & McGee, 1991; Perez & Torrubia, 1985; White et al., 1985), this finding was expected.  The results relating to this measure may support theoretical postulations put forth in the introduction associating recidivism with constructs measured by the DIS scale, namely behavioral disinhibition and differential risk appraisal (Zuckerman, 1994). 

Briefly, behavioral disinhibition (i.e., impulsivity) is theoretically linked to criminal behavior through a diminished capacity to refrain from such activities (Wallbank, 1985). Notably, this same “lack of self-control” was also brought forth in the discussion of age of first offense (Gotttfredson & Hirschi, 1990; Nagin & Farrington, 1992).  Zuckerman (1994) explains the circular relationship of behavioral disinhibition, differential risk appraisal and recidivism.  He argues that risk appraisals of criminal activity typically decrease through repeated exposure, which is brought on by behavioral disinhibition.  This increased experience with criminal behavior leads to habituation.  Increased habituation to criminal acts then results in further behavioral disinhibition and decreased risk appraisal, leading to a greater likelihood of repeat offending (Zuckerman, 1994).  

As habituation, disinhibition, and risk taking are postulated as being associated with the relationship between age of first offense and future criminal acts, it is interesting to consider the fact that higher DIS scores and age of first offense together were the best predictors in the present study. Given the theoretical similarities, it may be that these constructs and age of first offense are associated.

To expand upon this, it is possible that preexisting higher levels of disinhibition or differential risk appraisal lead an individual to earlier criminal behavior.  Following Zuckerman’s theory (1994), such an individual is likely to develop increased habituation to criminal acts,  ultimately leading in the aforementioned circular fashion to recidivistic behavior.  This process would support the theory noted previously that associates younger age of first offense with a predisposition to criminal acts (Gotttfredson & Hirschi, 1990; Wilson & Hernstein, 1985).   That is, if “early starters” are more crime prone, this may be related to those characteristics assessed by the DIS scale. 

 Alternatively, it may be that the actual process of committing crimes results in increased disinhibition and decreased risk appraisal.  At this point the identical circular process would occur.  This would support theories of a causal link between involvement in crime and recidivism via habituation.  However, this argument provides no explanation of what leads one to crime at an early age in the first place.  Linking back to the theories put forth regarding age of first offense, it can be speculated that environmental factors,  underlying personal characteristics, and their interaction may impact on criminal recidivism.

As there are no data testing this process, it remains hypothetical and the directionality of the relationship remains unclear.   Nevertheless, either argument provides a possible explanation for why both age of first offense and higher adjusted post-test DIS scores were significant predictors of recidivism in the present study.  What remains to be discussed, however, is the fact that DIS was predictive in the context of an intervention.  This will be addressed through the discussion of the treatment implications associated with these findings. 

The results of this study with regards to the DIS scale have potentially important applied significance.  As recidivists were found to have higher post-test DIS scores upon the completion of an intervention, a treatment that actively targets change in risk appraisal and behavioral disinhibition (those constructs assessed by the DIS scale) may prove to be effective.  This sentiment has also been stated previously in the literature  (Newcomb & McGee, 1991; White et al., 1985). 

As was noted in the introduction, AT may provide such a treatment opportunity.  In fact, change in these areas (i.e., behavioral disinhibition, risk appraisal) is incorporated into the theoretical foundations of AT (Amesberger, 1998, Bandoroff, 1989; Davis-Berman & Berman, 1994; Gass, 1993; Priest, 1993).  Importantly, the fact that higher DIS scores were found to be predictive of recidivism in the context of an AT intervention suggests the possibility that the intervention may have effected change in these areas, particularly for non-recidivists.  However, this study is lacking in the rigorous methodological control necessary to reach any such conclusions.  Internally valid outcome designs which rule out history, maturation, repeated testing, etc. must be conducted to unambiguously test this hypothesis. 

The findings of this study also have potential theoretical implications for AT.  Given the theoretical link between AT and the DIS scale constructs, the findings suggest that further empirical testing of the theory underlying AT as it relates to behavioral disinhibition and risk appraisal may provide support for that relationship.  Future well-controlled research should examine the impact of AT in these areas, using specific measures that capture these theoretical elements.  Should change in these areas be empirically verified, dismantling or component control designs can then be implemented to test specific components of the treatment for their impact on these characteristics.  

Predictor models that include separate measures of behavioral disinhibition and risk appraisal can also be tested in order to determine which of these constructs is more strongly related to recidivism.  These results can similarly be applied to both the theory and the intervention, as well as to informing future outcome research designed to establish cause-and-effect relationships.   

Results of these types of investigation can be applied to expand or refine the theory.  This approach can be applied systematically to test each theoretical tenet.  Through such a circular process which focuses on rigorous empirical testing and ruling out of rival hypotheses, ultimately the process and mechanism of change in AT can be more thoroughly delineated.  Such testing is necessary given the present state of the AT literature.

 

Non-Significant Results

            The hypotheses that the TSCS, MMPI-A scales 4 and 9, and number of previous offenses would predict recidivism were not supported.

           

Tennessee Self-Concept Scale (TSCS)

Previous research has indicated a relationship between self-concept and recidivism (Evans, Levy, Sullenberger, & Vyas, 1991; Fitts, 1965; Levy, 1997; Watson, 1979).  Self-concept is also at the theoretical foundation of AT, namely that AT causes changes in self-esteem which lead to behavioral and psychological changes that impact favorably on outcome (Bandoroff, 1989; Davis-Berman & Berman, 1994; Gass, 1993; Herbert, 1996, 1998).  Thus, the lack of findings in this area is unexpected and has potential theoretical and applied implications for AT. 

Existing AT theory delineates a process by which change in self-concept occurs.  Through the provision of success or “mastery tasks”, AT is thought to counteract negative self-evaluations and foster feelings of capability (Kimball & Bacon, 1993).  The successful achievement of unfamiliar tasks that the individual had previously thought him or herself incapable of is also thought to facilitate this process (Davis-Berman & Berman, 1994; Gass, 1993).  Thus, the environment is thought to provide tangible evidence to support the development and reinforcement of increasingly positive self-concepts. 

Accepting that there are no flaws in this theory, this provides grounds for speculation that perhaps the Project Challenge program itself was lacking.  Although Project Adventure is a recognized pioneer in the AT field, it is possible that the Project Challenge program did not provide the necessary style or quality of intervention to effect self-concept change.  This has been noted as a problem in the delinquency treatment literature, and Rutter et al. (1998) note “Some failures to modify delinquent behavior are due to a lack of effective implementation of the intervention rather than to a lack of efficacy of the intervention method itself” (p. 318).  Following this line of thought, it is possible that the results may have been different if the same study were run in another program that more effectively influences self-concept.

Of more theoretical relevance, however, is the fact that the TSCS adjusted post-test score was not predictive of recidivism.  This casts doubt on self-concept change as a theoretical tenet of AT.  Secondary analyses, employed to examine change from pre-test to post-test on the TSCS through a repeated measures ANOVA, indicated a trend towards significance  F(1,98), p=.07.  This leaves open the possibilities that AT was not sufficiently effective in generating positive changes or that the theory is perhaps incorrect. Self-concept change that has been seen in previous AT studies (Marsh, Richards, & Barnes, 1986; Hazelworth & Wilson, 1990; Davis-Berman & Berman, 1989; Myers, 1982; Kelley, 1993; Kelley, Coursey, & Selby, 1997; Herbert, 1998) may not be causatively attributable to the intervention.  It is possible that these findings could be based on alternative explanations such as post-intervention euphoria or environmental factors, as well as other basic considerations such as history and maturation.  The above points once again to the necessity of empirically testing the theoretical elements of AT.  Without a scientific foundation, the theory as it stands now is in essence just speculation based on anecdotal evidence.

Other possibilities do exist, however, that may explain the lack of findings with regards to the TSCS.  One may be that self-concept is not predictive of recidivism for individuals who participate in an intervention.  As this is the first study to directly examine this question, it may be that hypothesizing predictive utility for self-concept in this setting was incorrect.  The intervention itself also may have functioned to eliminate any existing predictive value of self-concept for recidivism.  Another possibility may be that self-concept is not predictive of recidivism for this particular population in this particular setting.

One explanation that could underlie these suppositions can be found in labeling theory and consistency theory.  Essentially, these theories suggest that the environment of the AT intervention may operate in a way opposite from what AT theorists propose, reinforcing deviant self-concepts as opposed to facilitating self-concept change.  However, this proposition does not necessarily negate the AT theory.  Instead, it is suggestive that perhaps this type of intervention may be less effective with juvenile offenders than with other types of individuals who do not hold such deviant self-concepts. 

As was discussed in the introduction, labeling theory proposes that once an individual behaves in ways that society characterizes as deviant, he or she is then “labeled.”  If the individual chooses to accept and internalize this label, he or she is likely to act in ways that are consistent with this label (Schur, 1973).  Relatedly, consistency theory proposes that individuals are motivated to act in ways that maintain consistency in their beliefs about themselves and the world (Heider, 1958).

According to these theories, it is likely that individuals who entered the program with delinquent self-concepts would have behaved in ways that were consistent with this self-concept as they began the program.  They may then have been “labeled” by the group, further reinforcing an already existing deviant self-concept.  Given that AT is an intervention based primarily on activities that take place within a group context, this process of self-concept reinforcement derived from interactions with the environment (i.e., the group) may have continued.  Associating primarily with other offenders may also have facilitated this process. 

In essence, it is therefore possible that this context may have reinforced a pre-existing deviant self-concept as opposed to altering it.  Proponents of the labeling perspective would support this contention, and it has been argued that “correctional intervention can label or stigmatize persons, creating and sustaining an identity of being a criminal among adolescents, many of whom are experimenting with delinquent behavior for a relatively brief time” (p. 51, Tollett & Benda, 1999).

This is an empirically testable proposition for future research, and some evidence exists in support of this argument.  A negative labeling scale was used in a study of chronic juvenile offenders to measure the degree to which parents and peer groups regarded an individual as someone who “gets into trouble or breaks rules” (Smith, Visher, & Jarjoura, 1991).  Results of this investigation showed that scores on this scale were predictive of participation in future delinquent acts.  Whether by reinforcing the delinquent label or by some other mechanism has, however, not been determined. 

Future research could use similar measures to test the hypothesis that AT is less effective for individuals with strongly deviant self-concepts.  A well-controlled study could be designed to test whether strength of negative labels was impacted by an AT intervention or whether the intervention differentially impacted individuals based on the strength of this label.  In addition, the inclusion of a self-concept measure could test the relationship between these two constructs.  The study could be run with one sample of juvenile offenders and one with non-offenders and the results compared.  Mixed sample studies could also be conducted in order to determine if the inclusion of non-offenders impacted the power of the intervention to effect negative (i.e., deviant) labels.  

In addition, predictor models could be employed using strength of negative labels as a predictor for a designated outcome.  Process research could also be undertaken, exploring what specific aspects of the intervention may be impacting on negative labels and/or deviant self-concepts.  Such research has theoretical implications, with the results being applicable to both AT and labeling theory.   

Such research could have important applied implications as well.  Positive change in self-concept has been found to be an aspect of treatment interventions that is successful in reducing recidivism among juvenile offenders (Greenwood, 1986).  If it is found that AT does strengthen negative labeling for juvenile offenders, and that this results in a lack of positive self-concept change, this type of intervention in its original form may in fact be contraindicated for such a population.  This possibility has not yet been explored.    

Methodological issues may also explain the lack of support for this hypothesis.  An examination of additional previous research with juvenile delinquents suggests that specific aspects of self-concept that are assessed by the TSCS subscales may be more closely associated with recidivism than overall self-concept (as measured by the TSCS).  Specifically, social and self-alienation deficits and disruptive familial relationships (Krueger, Schmutte, Caspi, & Moffit, 1994; Pena, Megargee, & Brody, 1996) have been linked to recidivism.  In addition, poor moral development has been theoretically linked to the development of delinquency (Evans et al., 1991).  Finally, positive personal and family self-concepts have been shown to contribute to be protective factors against delinquency (Levy, 1997). 

These particular constructs can be assessed by the social-self, the personal-self, the moral-ethical-self, and the family-self scale of the TSCS.  Thus, examining the predictive utility of these subscales, instead of the overall TSCS scores, may have yielded alternative results.  In fact, Baucom (1996) found that the change scores for the family-self subscale of the TSCS was a significant predictor, with non-recidivists showing change in a positive direction and recidivists showing change in a negative direction during the time of the intervention.  While Baucom’s methodology is suspect (as discussed in the introduction), this is indicative that exploration of the predictive utility for the subscales of the TSCS may warrant future research.

To explore support for this within the data, two-way repeated measures ANOVA (recidivism status by pre/post) was conducted on each of the TSCS subscales.  Correlations (Pearson r) between adjusted post-test scores for these measures and recidivism were also examined.  Interestingly, significant change in a positive direction was found for the moral-ethical self (F(1,98)=5.83, p<.05) and the personal-self (F(1,98)=4.27, p<.05) subscales. There was no significant change for the social-self subscale, but there was a significant interaction (F(1,98)=.59, p<.05).  Recidivists decreased on this measure from pre-test (mean=63.71) to post-test (mean=61.91), and non-recidivists increased from pre- (mean=61.61) to post- (mean=65.00).   A significant negative correlation was also found between recidivism and adjusted post-test social-self scores (r= -.24). 

While no conclusions can be drawn from these findings due to issues of methodological control, results of these secondary analyses imply that future research examining the predictive utility of these subscales may shed light on the findings of the present study with regards to self-concept.  Any such findings could also be applied to further development of this theoretical tenet of AT. 

  

Number of previous offenses

            The most puzzling finding of this study is that number of previous offenses was not predictive of recidivism.  This variable has predicted recidivism in numerous studies and, along with age of first offense, is thought to be the most robust predictor of recidivistic behavior (Gottfredson & Hirschi, 1990; Myner, et al., 1998; Nagin & Paternoster, 1991; Nagin & Farrington, 1992a; Nagin & Farrington, 1992b; Tolan & Thomas, 1995; Tollett & Benda, 1999).  In fact, it has been observed that “competent research regularly shows that the best predictor of crime is prior criminal behavior” (Gottfredson & Hirschi, 1990, p. 107).  This leads to speculation as to what may have caused these differences.  One possibility is that differing results may be linked to the intervention context.

Surprisingly, the predictive utility of number previous offenses with juvenile offenders who participate in a correctional intervention has received minimal attention.  Only one other such study was located (Tollett & Benda, 1999), and the findings parallel those of the previous literature (i.e., number of previous offenses was found to be predictive of recidivism).  Discrepant findings can perhaps be explained by differing criminal history within the samples (e.g., frequency and severity).  As this is also a potential explanation for the lack of findings relating to MMPI-A scales 4 and 9, this issue will be more fully explored later in the discussion of the MMPI-A results.

            The exploration of how the intervention may have altered the predictive power of number of previous offenses can be best approached from a theoretically-based perspective.  Theories of the relationship between number of previous offenses and recidivism parallel those of age of first offense.  To briefly review, these theories postulate: 1) a stable criminal predisposition exists that leads one to early offending and repeated criminal acts, 2) dynamic, situational, and proximal (i.e., environmental) influences lead one to criminal behavior, or 3) there is a causal link between committing a crime and committing other crimes.  As was noted, the possibility remains that all three types of influences interactively lead to criminal behavior.  

            Following the theories mentioned above, it may be that the intervention impacted on personal characteristics in some way that altered a criminal disposition.  It may also be that the intervention impacted in some way that led to better coping with environmental and situational influences (i.e., delinquent peers, family distress).  Relatedly, participation in the intervention may have facilitated behavioral change, which then led to different environmental circumstances (i.e., decreased substance abuse or abstinence, more positive relationships with family).  Finally, the intervention may have effected change in psychological processes causally linked to recidivism, such as habituation and disinhibition.  The fact that higher scores on the DIS scale were predictive of recidivism allows for speculation that the intervention may have acted on these processes in some way. 

Should the intervention have impacted in any of these ways, it may have led to a decreased propensity towards future criminal acts regardless of past history of criminal offense.  Thus, the predictive power of number of previous offenses may have been altered through the intervention.  However, given that theory underlying the relationship between recidivism and both age of first offense and number of previous offenses is the same, these explanations do not answer the question of why the predictive power of number of previous offenses would be altered while age of first offense remains a significant predictor.

This presents a conundrum.  One possibility may be that those offenders who were convicted of their first criminal act at a younger age may have “more” of those characteristics potentially linked theoretically to early and repeat offending.  That is, they may have an increased predisposition to criminal behavior, a greater number of environmental and situation circumstances that lead to crime may exist for such individuals, or they may be more disinhibited to begin with than those who do not begin their criminal careers quite so early.  It may also be that for those who commit their first offense at a younger age, these characteristics may be more “entrenched.” Thus, such individuals may be more resistant to change.

However, this explanation would only hold true if those who were younger at first convicted offense and those who committed a higher number of self-reported previous offenses were not the same people.  Given that previous criminal acts are assumed to be predictive of future criminal acts (Gottfredson & Hirschi, 1990; Myner, et al., 1998; Nagin & Paternoster, 1991; Nagin & Farrington, 1992a; Nagin & Farrington, 1992b; Tolan & Thomas, 1995; Tollett & Benda, 1999), one would expect a correlation between these two variables.  This secondary analysis was run (Pearson r), and no significant relationship was found.  This indicates that those with a high number of previous offenses are not necessarily those who committed their first convicted offense at a younger age.  Thus, it remains a possibility that those who were convicted of their first criminal act at an earlier age were either less impacted or impacted differently by the intervention in the present study than were those who self-reported a higher number of previous offenses.   

Differential effects of the intervention for those who commit crimes at a younger age, those who self-report a large number of previous offenses, and those who have engaged in both is an area for future research.  Such research has applied implications, as currently both age of first offense and number of previous offenses are considered to be indicators of high risk for recidivism (Tollett & Benda, 1999). 

One methodologically-based explanation for the lack of results for this variable is that the use of self-reported number of previous offenses is a different measure than the more commonly employed number of arrests or convictions (Andrews & Bonta, 1994; Bonta, Law, & Hanson, 1998; Caspi, et al., 1994; Dembo, et al., 1991; Dembo et al., 1998; Hollander & Turner, 1985; Loeber, Farrington, Stouthamer-Loeber, Moffitt, & Caspi, 1998; Tollett & Benda, 1999).  It is possible that these differing operationalizations of the construct may have led to the discrepant findings.

Official statistics are thought to be more heterogeneous than are self-report measures of the same construct and result in decreased variability.  Arrest and conviction records have been found to be more indicative of serious crimes, chronic offending, and specific types of offenses (Cerkovich, Giordano, & Pugh, 1985; Elliot, Dunford, & Huizinga, 1987) than are self-report.   This results in bias within official arrest and conviction statistics, and “there is sufficient evidence in the available research to question the assumption that arrest offenses are representative of crimes known to police or that arrested offenders are representative of suspected offenders” (Elliot et al., 1987, p. 93). 

It is believed that self-report provides a more valid measure, as it is free from distortion and captures a wider range of less serious delinquent behaviors than official arrest records (Cerkovich, et al., 1985; Elliot, et al, 1987; Tolan & Thomas, 1991).  In fact, it was found that the majority of self-reported offenses were associated with property theft of less than $5.00 (Nagin & Paternoster, 1991).  It was also found that there are a large number of juveniles with high rates of self-reported minor offense involvement that refrain from committing the more serious offenses that lead to arrest and conviction (Cernkovich et al., 1985).  Thus, a higher number of self-reported previous offenses may be indicative of less serious offenses than a higher number of arrests or convictions. 

Importantly, such minor offenses may not be associated with recidivism in the same manner.  Instead, self-report measures that include minor criminal acts may be more reflective of social circumstance, opportunity, or developmental phase than more serious offenses typically found in official records (Nagin & Farrington, 1992a). Thus, the use of a self-report measure in this study  may have contributed to the lack of findings with regards to number of previous offenses. 

Notably, this is not to imply that no serious crimes were committed by those individuals in the sample, nor to imply that those crimes were not related to recidivism.  Rather, it is possible that the predictive power of this variable may be diluted when the more representative self-report is utilized in a statistically-based predictor model.  This may be more salient with a sample such as that in the present study, one with a presumably less extensive criminal history than those in previous studies.  Additional issues related to criminal history will be addressed in the following section.

In sum, the lack of results with regards to number of previous offenses indicates that, for juvenile offenders who participate in an intervention, number of previous offenses may not be as clearly indicative of high risk for recidivism as previously thought.  While these results may be related to the above noted measurement issues, it is also possible that the intervention acted in some way that altered the mechanism or mechanisms by which number of previous offenses is related to recidivism. 

Future research first needs to examine what characteristics may underlie the relationships between number of previous offenses and recidivism.  Paralleling the suggestions put forth for future research on age of first offense, this can be done through the use of theoretically-based predictor models. Findings of such an investigation have implications for the theories explaining the connection between number of previous offenses, age of first offense, and recidivism.  Results can be used to determine the relative strength of the three different theoretical postulations in explaining this relationship.  In addition, further investigation can be used to help determine whether these three theories interactively influence recidivism.   

Such future research has applied significance as well, as it can highlight specific areas upon which to focus intervention efforts.  In addition, results can be used to inform studies focusing on why intervention may have a differential impact for those who are first convicted of a crime at a younger age versus those who self-report a high number of crimes.

 

MMPI-A scales 4 and 9

The hypothesis that pre-test levels of MMPI-A scales 4 and 9 would predict recidivism was unexpectedly not supported.  Previously, these scales have been found to be associated with recidivism (Archer, 1992; Cashell, Sewell, & Hillmon, 1998; Pena, Megargee, & Brody, 1996; Weaver & Wooten, 1992; Zuckerman, 1994).

Prior to considering the implications of this finding, it was important to rule out any possibility that the MMPI-A scores used in the present study were invalid.  Given the fact that participants were mandated to the Project Challenge program (of Project Adventure) through the juvenile justice system, individuals within the sample may have been invested in actively trying to create a favorable impression with the juvenile authorities (i.e., “faking good”).  This tendency towards impression management must be considered when conducting assessments of juvenile offenders in correctional settings (Hoge & Andrews, 1996).  Thus, validity profiles (scales F, L, and K) of each participant were closely examined.  All were found to be valid.

According to the F scales, there were no invalid profiles within the sample.  Scores on scales L and K indicated that 9 out of 100 participants may have answered in a somewhat defensive fashion, primarily in the direction of trying to make themselves look more positive.  This is common among adolescent populations, however, and thus these indicators alone were not cause to suspect profile invalidity (Archer, 1992; Butcher, et al., 1992).  Rather, such indicators necessitated further investigation. 

According to the MMPI-A manual (Butcher et al., 1992), profiles with elevations on scales L or K are only to be interpreted in the context of other scales (i.e., elevations on either of these scales in combination with overall scale elevations or depressions suggest an increased likelihood of invalid profiles).  For those nine participants with elevations on scales L or K, there were no additional indicators of invalid profiles.  As per the manual recommendations (Butcher et al., 1992), while these individuals may have answered in a somewhat defensive fashion, their overall profiles provided no reason to suspect response invalidity.  Thus, we can be confident that defensive responding did not skew the results.

Assuming that the MMPI-A findings are valid, they first suggest the possibility that these scales are not predictive of recidivism in this setting.  While it is true that predictive relationships between these scales and recidivism have been found for juveniles through regression analyses (Weaver & Wooten, 1992), it was not in an intervention setting.  The same relationship may not exist in such.   All other prior findings of a relationship between MMPI-A scales 4 and 9, delinquent behavior, and recidivism have been established via correlational procedures and the comparison of recidivists to non-recidivists (Archer, 1992; Cashell, Sewell, & Hillmon, 1998; Pena, Megargee, & Brody, 1996; Zuckerman, 1994).  Future research examining the predictive utility of MMPI-A scales 4 and 9 via regression procedures is necessary to compare results across settings. 

While there may be a well-established relationship between these scales and delinquent behavior, this relationship may be less robust when techniques using these scales as predictors are employed.  Similar to the classic research mantra “correlation is not causation,” it may be that there are underlying factors associated with the scales and recidivistic behavior that fuel this association as opposed to a more direct predictive relationship for the scale scores themselves.

The second possibility is that the intervention altered any existing predictive relationship in some fashion.  Due to their stability, MPPI scores are thought to be resistant to change through a brief intervention such the one in the present study (Butcher, 1995).  Thus, we have limited reason to posit that significant change on these measures occurred from pre-test to post-test.  Secondary repeated measures ANOVA supported this, showing no significant pre-post change for either of the scales. 

Although there may not have been significant change, it is still possible that the intervention acted somehow on processes measured by this scale that are related to recidivism.  In fact, both scale 4 and scale 9 scores encompass traits that have been speculated previously as being causally linked to recidivism, namely disinhibition and risk taking. The fact that both scales have been found to be related to sensation seeking (of which DIS is a subscale), and DIS was found to predict recidivism can be seen as some support for this argument. Thus, perhaps the intervention acted on individuals in a way that was not enough to effect change in the scale scores overall, but did impact on some of these particular processes enough to alter the predictive relationship.  

Interestingly, Baucom (1996) found the change score for MMPI-A scale 9 to be predictive of recidivism.  However, this is in the opposite direction of what one might expect, as mean scores for both of her groups slightly increased.  Nevertheless, Baucom’s findings also allow for speculation that some process encompassed by scale 9 was impacted by the intervention, and this may have altered any existing predictive relationship.  This finding is interesting, as scale 9 is more closely associated with impulsivity (i.e., behavioral disinhibition) than is scale 4 (Archer, 1992).  Given that impulsivity has been associated with recidivism, this may be an important characteristic upon which to direct future research attention.  This discussion also highlights once again the importance of empirically testing the impact of AT on behavioral disinhibition. 

Sample differences

Differing criminal histories in the sample of this study versus those in previous studies is another possible explanation for the lack of MMPI-A findings, as well null findings on number of previous offenses.  Notably, the vast majority of studies examining predictors of recidivism employ samples of incarcerated or institutionalized individuals (Cernkovic et al., 1985; Rutter et al, 1998). Common practice within the juvenile justice system is to consider previous criminal history and current offense severity when considering the restriction level of correctional placement (Rutter et al., 1998).  Given the highly restrictive nature of incarceration and institutionalization, it can thus be assumed that individuals within these settings have extensive and severe criminal histories.

In contrast, Project Challenge is a much less restrictive setting.  Therefore, in line with common practice, judges would be more likely to deem those individuals with a less serious criminal record as suitable for participation in the Project Challenge program.  In fact, participants with a history of serious crime, violent crime and sex offenses in particular, were excluded from participation.  Thus, for Project Challenge participants, it is possible that the MMPI-A scales may be less strongly associated with recidivism than for samples in previous studies with more severe or extensive criminal histories.

Only one MMPI-A study reviewed did not use an incarcerated or institutionalized sample (Lingren et al., 1986).  Interestingly, this study also found no predictive value of MMPI-A scales 4 and 9 for recidivism.  This study provides support for the contention that differences in criminal history in the present study versus other previous studies may have led to the null findings for the MMPI-A scales.

Similarly, number of previous offenses may have been more predictive for recidivism among samples of incarcerated or institutionalized individuals than for Project Challenge participants with lesser criminal histories.  In fact, Cernkovich et al. (1985) believe that these differences are so pervasive that “any comparisons of institutionalized and non-institutionalized offenders are inappropriate because they involve comparing apples to oranges” (p. 706).

Relatedly, there may be differences between those juveniles who get caught and convicted (i.e., incarcerated) as compared to those that do not.  Differences around offense type, severity, and chronicity have been noted previously in the discussion of self-report versus official statistics. Previous research has also found that when police contact occurs due to criminal behavior, the probability of arrest and conviction is related to factors such as age, race and SES of the offender, local departmental policies, presence or absence of an accomplice or a complainant, and demeanor and attitude of the suspect (Elliot et al., 1987).  This is indicative of a potential arrest bias, and Elliot et al. (1987) note that “There is sufficient evidence to question whether arrested offenders are representative of offenders in general” (p. 93). 

It has also been posited that there are underlying individual characteristics that differentiate persons with more extensive and severe criminal histories from those with lesser histories (Nagin & Farrington, 1992b).  Drawing from those theories discussed which attempt to explain the relationship between criminal acts and future criminal acts, these characteristics may include a criminal predisposition, environmental factors, or underlying causal links.   

While obviously Project Challenge participants were “caught,” the fact that they were likely to have had less severe and extensive criminal histories may indicate differences between these individuals and more “chronic offender types” who committed more severe offenses.  Any such differences may have impacted how they responded to the intervention, and how this response related to recidivism. Future research is necessary to examine differential responses to the AT intervention based any of the above noted factors (e.g., age, race, SES, individual characteristics), as well as criminal history.  Predictor models can be employed which incorporate these types of characteristics as well.  Such research has important applied implications, as it may indicate for whom this type of intervention is most successful.

 

 

 

 

 

Conclusions

The findings with regards to the DIS scale and the TSCS scale both have theoretical and applied relevance for AT.  Thus, these are thought to be the most important results of the study.

The results indicating that DIS scores predict recidivism in this context is a novel contribution to both the recidivism and the AT literature. Behavioral disinhibition and risk appraisal are linked to the theoretical foundations put forth within the AT literature.  Thus, the fact that results were found indicating predictive utility of the DIS scale implies that empirical support may be revealed for this theoretical postulation.  Future researchers in the AT area are encouraged to focus efforts in exploring this area, testing both the theory and the treatment. 

            It may be that behavioral disinhibition is a potential point of impact for AT. Within both AT and other intervention settings, this finding should be explored further. The fact that behavioral disinhibition is posited as a mechanism in the relationship between crime and future crime, and that this predictive relationship may have been altered by the intervention for number of previous offenses, lends further support for this contention. In addition, the fact that it was the adjusted post-test scores of this measure that predicted recidivism implies that there may have been change on this measure which possibly impacted recidivistic tendencies.

This finding also suggests that treatment programs should consider attempting to impact adjudicated juveniles in the areas of behavioral disinhibition and risk appraisal.  AT practitioners should consider delineating what components of their program may have impacted on these constructs and test the efficacy of such components.  It is also important to explore any factors that may lead to differential impacts for participants, and how these may relate to DIS scale constructs.  This can be done using predictor models and through well-controlled outcome studies.

In contrast, the fact that adjusted post-test TSCS scores were not found to be predictive of recidivism casts doubt on the centrality of this construct within AT theory.  This suggests that testing of the theoretical tenets of AT is important.  It is possible that specific aspects of self-concept, as measured by the TSCS subscales, may be more strongly associated with recidivism in this context than is self-concept overall.  This is an important area for future consideration, as it may suggest specific areas of self-concept upon which to focus research and intervention efforts. In addition, it is also important to explore whether self-concept is differentially impacted by the AT intervention for those with delinquent self-concepts. 

 

Future predictor models

Overall, such models can be tested both in and out of intervention settings and the results compared.  Differences would have applied implications as they may highlight possible points of impact for interventions.  These can be followed by research more specifically exploring any such areas.

A number of other suggestions were provided for predictor models that can expand the results of this study.  The DIS scale should be included in future models and tested with other variables to see if its predictive power is increased or decreased when paired with other measures.  Younger age of first offense appears to be a significant risk factor for recidivism and should also be included in future models.  Differences in the predictive utility of age of first offense and number of previous offenses can be explored as well, and compared across settings and populations. 

Characteristics that are theorized as underlying the relationship between this variable and recidivism should be tested for their predictive value, both models following one theoretical frame and multidimensional models.  Such models should also include age of first offense, in order to examine whether the predictive power of this variable remains when paired with potentially associated characteristics. The relationship of number of previous offenses and recidivism can be similarly explored. 

Identical models can be tested with different populations in the same setting to determine if predictive relationships are different between groups.  Conducted in an intervention setting, this may suggest ways that the intervention may differentially impact different populations.  The predictive utility of MMPI-A scale 4 and scale 9 should also be compared across settings and between with samples of differing criminal histories.  Such investigations may provide further information to help understand discrepant findings in the literature.

Specific to AT, the utility of the TSCS subscales in predicting recidivism within an AT context should be investigated.  This may yield results that can inform AT theory, and highlight potential areas on which to focus intervention efforts.  The predictive power of a negative label in AT can also be explored, both for its relationship to a specific outcome and its relationship to self-concept.  These results can also be used to inform AT theory as it relates to self-concept. 

 

 

Study Limitations and Future Research

            It is important to repeat at the onset of this discussion that in no way was this study designed as a treatment outcome study.  Thus, no scientifically-based inferences about the relationship between the intervention and recidivism, nor whether the intervention was an agent of change, can be drawn from the results.  In line with the hypotheses, the only conclusions that can be drawn are specifically limited to whether the particular predictors investigated have utility in this particular type of intervention setting.

 Speaking globally, in order to expand both the conclusions and generalizability, future research in this area must provide substantially more rigorous methodological control.  Issues of methodological control did create potential limitations in the present study.  These include possible confounds due to differences in participant experience during the intervention, as well as confounds that may exist due to limited available information about therapist and participant characteristics. 

            As was noted earlier, each individual group of participants entered and completed the Project Challenge program at the same time.  Thus, the groups remained intact throughout the time of the intervention.  While each group followed a particular protocol specifying that aspects of the program be initiated at particular times (e.g., on-site camping during weeks one and three, backpacking during week five), these protocols were of necessity flexible.  This flexibility allowed for the choice of activities within the particular phase of the program to be determined by the facilitators.  Decisions as to what activities to employ were based on the facilitators’ judgment of what was needed by the group at that particular point in its development. 

While the range of activity choices in each phase of the program was determined by the protocol, it is likely that each group’s development followed a somewhat different course.  Given that the choice of activity is determined by where the group was in its process, it is likely that the experiential activities chosen by the facilitators during each specific phase of the program may have differed somewhat.  Thus, different groups of participants may have had slightly differing experiences.

            While intuitively this may appear to be a confound, it is important to recognize that flexible protocols are at the heart of all well-controlled studies of psychotherapy.  According to Borkovec (1994), the most well-designed protocols include specific outlines and techniques to be used at each phase of treatment, but such manuals also must provide “the needed flexibility to respond to the unique nature of each client while still remaining within the spirit of the therapy’s protocol (p. 264).”  Given that, some difference in client experience is inherent in the use of any intervention protocol.

            However, Borkovec (1994) also adds a cautionary note, stating that “there is no guarantee that the therapists actually do adhere to the protocol with every client or in every portion of every session (p. 265).”  As such, he regards integrity checks by independent staff as essential in determining protocol adherence.  Researchers in the delinquency field echo this statement, observing that “A pervasive finding across evaluation studies is that intervention ‘integrity’ is an important ingredient for success; in other words, that the intervention has actually been implemented as intended” (Rutter, Giller, & Hagell, 1998; p. 318). 

Notably, while Project Challenge facilitators were indeed following a treatment protocol, this study contained no such checks.  Thus, we have no absolute assurance that protocols were adhered to, and it remains a possibility that confounds exist due to variations in participant experience.  Future studies within an intervention context should employ such integrity checks to determine whether, in fact, protocols are being followed.   

            Relatedly, differences in therapist training and experience may have also contributed to differences in participant experience.  There is no information provided regarding the staff that worked with each group of participants, thus there is no basis upon which to evaluate the level to which differences in staff training and experience may have led to differences in the quality of service provision.  Any such differences may have also created differing participant experiences (Borkovec, 1994). 

This problem is not unique to this particular study, however.  Such problems are commonly seen in psychotherapy research as well.  In fact, Borkovec and Miranda (1999) note that “The therapy research field has made little progress in addressing this question” (p. 149).  Future studies within any intervention context should make attempts to control for such therapist-related elements. 

            One additional potential confound may exist in the area of participant characteristics.  While grouping participants into recidivist and non-recidivist groups was appropriate to the goals of the study, these groupings do not allow for the exploration of qualitative differences that may exist between participants in these groups.  Such qualitative differences may lead to different participant responses to the intervention.

The present study made some effort in this area by exploring the predictive utility of offense classification ratings (with null results).  Other participant characteristics may also have impacted the results, however.  While such an examination was beyond the scope of this study, differences based on such variables as participant background, previous treatment history, additional aspects of personality, psychopathology, or numerous other variables may have existed which influenced not only an individual’s tendency to recidivate, but also how an individual responds to an intervention designed to impact recidivism.

Previous research indicates several individual differences that may prove interesting to explore in relation to AT and recidivism.  Personality characteristics indicative of a psychopathic personality organization are those most predictive of recidivism (Webster & Jackson, 1997; Rutter, Giller & Hagell, 1998).  Such characteristics include hyperactivity, deviance, impulsivity, aggression, and antisocial behavior.  While this study attempted to operationalize such traits through the use of MMPI-A scale 4, a more specific examination of these traits may lead to different results.

Recidivists have also been found to have fewer years of education, lower scores on intelligence tests, lower school attendance, reduced school achievement, and are more likely to come from broken homes and abuse substances (Bonta, Law, & Hanson, 1998; Dembo, et al., 1991; Dembo et al., 1998; Hodgins, 1993; Myner, et al, 1998).  In addition, psychopathology has also been linked to recidivism.  Specifically, it has been found that adolescents with a mood disorder, particularly depression, are at greater risk for recidivism (Myner, et al., 1998; Puig-Antich, 1982).  Adding variables such as these to future predictor models may help to expand the findings of the present study. 

Most central to this discussion, however, while these characteristics have been found to be associated with recidivism, none of them have been looked at for their predictive utility in an intervention context.  Given that treatment has been found to be associated with a positive effect on the recidivism rates of juvenile offenders (Andrews, Zinger, Hoge, Bonta, Gendreau, & Cullen, 1990; Garrett, 1985; Hollin, 1990; Lipsey & Wilson, 1993; Roberts & Camasson, 1991), and client characteristics have been shown to interact with psychotherapy outcome (Arnkoff, Victor, & Glass, 1993; Beutler, Engle, Mohr, Daldrup, Bergan, Meredith, & Merry, 1991; Beutler, Machado, & Allstetter Neufeldt, 1994; Shoham-Salomon, 1991; Shoham, Bootzin, Rohrbaugh, & Urry, 1995), it is important to examine the impact of such participant characteristics in the context of an intervention and perhaps an AT intervention in particular.  This is another relevant area for further exploration.   

Selection bias may also have existed in the present study, thus impacting the representativeness of the sample and threatening both internal and external validity.  While those individuals with a history of violent crime were excluded from participating in Project Challenge, it is unknown what other criteria may have been used by individual judges to identify appropriate participants.  Thus, particular judges may have been more or less likely to selectively send individuals to Project Challenge.  Alternatively, particular judges may have been more or less likely to select individuals who committed specific offenses or individuals with certain “types” of offense histories.  Given that all adjudicated juveniles who participated in Project Challenge may not have been subjected to the same criteria, it is a possibility that the sample may have been biased.

There also may be limited applied significance for these findings.  This can first be seen when considering the classification criteria for membership into the recidivist or the non-recidivist group.  As a reminder, the cut-off probability criteria of .5 is simply an arbitrary designation.  Therefore, exploring differences in classification that result from a different cut-off criteria is appropriate (Tabachnick & Fidell, 2000).  In the present study, such a shift was made from .5 to .6.  While the overall predictive accuracy rates did not change substantially, this did alter the predictive accuracy across the two groups.  Thus, were such data to be used in any applied sense, decisions would of necessity be made as to whether it was more important to predict recidivism or non-recidivism and the cut-off point would be based on such considerations. 

The limitation is found in the close examination of the histogram of estimated probabilities for group membership.  Ideally, with a strong predictor model, individual cases should cluster at both ends of the continuum (i.e., one cluster should have a high probability of group membership and one cluster should have a low probability).  In the present study, while percentages of individuals correctly classified by probability estimations indicate a reasonable prediction rate, there is minimal bimodal clustering of this sort.  Instead the distribution is more flat, showing a broad range of probability estimations for individual cases.  This is indicative that the model may not be as strong as the percentages may lead one to believe. 

The second issue of applied significance relates to the small differences in mean scores between recidivists and non-recidivists for the significant predictors (age of first offense and DIS scores).  While statistically significant, the differences were less than one unit for both. Thus, the two groups would perhaps in reality be indistinguishable from each other in an applied setting on either of these measures. However, this does not diminish the importance of further research pursuits of these variables, nor of taking these risk factors into account in applied situations.

One final possible limitation of this study exists based on the use of archival data.  While archival data is commonly used and was appropriate for the design of this study, without having had primary control of the data, we are limited in our certainty that the data are free from unknown influences, biases, and errors.  Although every care was taken to be as certain as possible that these issues did not exist (e.g., discussions with the clinical director in charge of data collection and scoring, discussions with Project Adventure staff, the use of quantitative measures, objective outcome measures), the fact remains that such issues are inherent when using archival data sources (Kazdin, 1992). 

 

Importance of Conducting Rigorous Therapy Outcome Studies of AT

The final part of this discussion will be devoted to expanding upon the problems associated with the lack of well-controlled outcome research on AT.  Suggestions for future research in this area will also briefly be put forth.  As has been noted repeatedly, the literature evaluating outcomes of AT programs indisputably lacks the methodological control necessary in order to rule out rival explanations for any reported outcome results.  While the continued investigation of predictor models in an AT context is important, such studies are lacking in the ability to establish cause-and-effect relationships.  Thus, without such outcome research, there can be no isolation of the mechanisms of change in AT.

Referring back to earlier comments about the same issue, without such testing of AT outcomes and theory, we also have no solid foundation upon which to base future studies in this area.  Davis-Berman and Berman (1994) also discuss the problems permeating the AT literature, asserting that:

“These potential problems are raised not to discourage the pursuit of methodologically sound research studies, but rather, to build he argument that effective research on wilderness programs must be innovative and creative, yet methodologically sound.  We must develop ways to meet scientific criteria, yet not compromise the uniqueness of our programs.  This, we believe, is one of the major challenges for wilderness programs in the 1990’s.” (p. 178). 

 

Ultimately, it is necessary that research in any AT context must start at the beginning, at the most basic level of research design.  Given the methodological inadequacies in this literature, it is essential that future research in this area be based upon solid principles of methodological control.  The psychotherapy research field has made substantial progress in this area and researchers in the AT field are strongly encouraged to borrow from this literature.  In particular, Borkovec (1994) and Kazdin (1992) discuss the specific methodological and design considerations that must be taken into account in conducting scientifically-based psychotherapy research.  Additionally, Borkovec and Miranda (1999) provide an excellent summary of the importance of such research, as well as offering suggestions applicable to the implementation of such controlled research in applied settings. 

To apply the ideas of these respected researchers to the AT field, in summary, future AT research must use random assignment, utilize control groups, employ an appropriate level of standardization, create homogenous groups with large enough sample sizes, provide therapist characteristics, utilize relevant pre- and post-test measures, and analyze the data in an appropriate fashion.  After appropriate control has been provided through the use of such techniques, any significant change must then be replicated in a different setting. 

Borkovec (1994) also puts forth suggestions for the use of dismantling and component control designs.  When using these designs, treatment conditions are created in which a particular element is present in one condition but not others.  As all groups are being presented with the same treatment, plus or minus this particular element, potential confounding factors are held constant across groups and the effects of one particular factor are isolated.  Through the use of such designs, the independent variable is held constant to a greater degree than is true with other types of design strategies.  Such designs could be easily adapted to AT settings.

Importantly, it is only through increased methodological rigor and more sophisticated research designs that specific cause-and-effect relationships regarding the impact of an AT intervention can be established.  This is an ongoing process, a systematic approach whereby new studies build on previous findings.  Should well-controlled research empirically establish that AT does facilitate change, we can then begin closer examination of the particular elements of the intervention that are impacting the change process.  Through such systematic exploration, the active mechanism of such change can eventually be isolated.

Though significant effort may be required to conduct such studies, should AT be empirically established as an effective intervention, scientific evaluation of AT can lead to improvement in treatment overall.  Regarding this issue, Borkovec and Castonguay (1998) note “The identification of increasingly specific cause-and-effect relationships leads to better theoretical understanding of the nature of the psychological problems being treated and the nature of the mechanism of change underlying any causative roles for a therapy, its elements, its parameters, or elements added to it.  From this knowledge, hypotheses about modifications or additions to a therapy emerge and can be tested” (p. 139).

Concluding this argument with a more pragmatically-based point, it is important to recognize the need for such research in the present-day climate of empirically-supported treatments and third party payment.  Simply put, therapists are facing increased demands to use treatments that have been empirically proven to work.  In this context, the underlying question becomes “If X treatment had been proven to work and Y treatment has not, how can the continued use of Y treatment be justified?” (Newes, 2001, p. 6).   While this climate may not as yet have pervaded the AT community, it seems imminent that AT practitioners will be soon subjected to the same pressures.  Thus, it is vital that this need be recognized and attended to by future AT researchers.    

In conclusion, the results of this study highlight the need for further well-controlled research, research that focuses on documenting outcomes and testing the specific theoretical elements of AT.  Through such a process, studies such as this can be grounded in empirically-tested theory, allowing for increasingly sophisticated conclusions to be drawn.  In this fashion, the course of theory development and outcome documentation becomes circular, with each new study building upon those that have come previously, as well as informing the next.  Borrowing from the ideas of Borkovec (1994, 1997), in the end, such a process can contribute to the overarching goal of increasing our basic knowledge of human behavior.

 

 

 

 

 

REFERENCES

 

            Andrews, D.A., Zinger, I., Hoge, R.D., Bonta, R.D., Gendreau, P., & Cullen, F.T. (1990a).  Does correctional treatment work?  A psychologically informed meta-analysis.  Criminology, 28, 419-429.

            Andrews, D.A. & Bonta J. (1994).  The psychology of criminal conduct.  Cincinnati:  Anderson Publishing Co.

            Archer, R.P. (1992) MMPI-A:  Assessing adolescent psychotherapy.  New Jersey:  Lawrence Erlbaum Assoc.

            Arnkoff, D.B., Victor, B.J., & Glass, C.R. (1993).  Empirical research on factors in psychotherapeutic change.  In G. Striker and J.R. Gold (Eds.).  Comprehensive handbook of psychotherapy integration (pp. 27-42).  New York:  Plenum Press.

Arthbutnot, J., Gordon, D.A., & Jurkovic, G.J. (1987).  Personality.  In H.C. Quay (Ed) (1987), pgs 139-183.  The Handbook of Juvenile Delinquency.   New York:  John Wiley and Sons.

af Klinteberg, B., Humble, K., & Schalling, D. (1993). Personality and psychopathy of males with a history of early criminal behavior. European Journal of Personality, 6, 245-266.

Ball, S.A., Carroll, K., & Rounsaville, B.J. (1994). Sensation seeking, substance abuse, and psychopathology in treatment-seeking and community cocaine abusers. Journal of Consulting and Clinical Psychology, 62, 1053-1057.

Bandoroff, S. (1989).  Wilderness adventure-based therapy for delinquent and pre-delinquent youth: A review of the literature.  (ERIC Document Reproduction Service No.  ED 377 428).

Basta, J.M. & Davidson, W. S. (1988).   Treatment of juvenile offenders:  Study outcomes since 1980.    Behavioral Sciences and the Law, 6(3), 355-384.

Baucom, L.  (1996).   Towards a success profile for adventure-based  program    participants by evaluating Project Challenge recidivists.   Unpublished master’s thesis.  Georgia College,  Midegeville, GA.

Beck, A.T. & Steer, R.A. (1987).  Beck Depression Inventory Manual.   San Antonio, TX:  The Psychological Corporation.

Beck, A.T., Steer, R.A., & Garbin,  M.G. ((1988).  Psychometric Properties on the Beck Depression Inventory:  Twenty-five years of evaluation.  Clinical Psychology Review, 8, 77-1000.

Berman, D. (1995). Adventure therapy: Current status and directions. Journal of Experiential Education, 18 (2), 61-62.

Beutler, L.E., Engle, D., Mohr, D., Daldrup, R.J., Bergan, J., Meredith, K., & Merry, W. (1991).  Predictors of differential response to cognitive, experiential, and self-directed psychotherapeutic procedures.  Journal of Consulting and Clinical Psychology, 59, 333-340.

Beutler, L.E., Machado, P.P.P., & Allester Neufeldt, S. (1994).  Therapist variables.  In A.E. Bergin & S.L. Garfield (Eds.), Handbook of psychotherapy and behavior change (4th Ed., pp. 229-269).  New York:  Wiley.

Blackburn, R. (1998). Criminality and the interpersonal circle in mentally disordered offenders. Criminal Justice and Behavior, 25, 155-176.

Blackburn (1996).  Mentally Disordered Offenders.  In C.R Hollin (ED.).  Working with offenders:  Psychological practices in offender rehabilitation. New York:    John Wiley and Sons. 

Blackburn, R. & Coid, J.W.  (1998). Psychopathy and the dimensions of personality disorder in violent offenders. Personality and Individual Differences, 25, 129-145.

Bonta, J., Law, M., Hanson, K. (1998). The prediction of criminal and violent recidivism among mentally disordered offenders: A meta -analysis. Psychological Bulletin, 123, 123-142.

Bonta, J. & Motiuk, L.L. (1985).  Utlization of an intervew-based classification instrument:  A study of correctional halfway houses.  Criminal Justice and Behavior, 12, 333-352.

Borkovec, T.D. (1994).  Between-group therapy outcome research:  Design and methodology.  In L.S. Onken & J.D. Blaine (Eds.), NIDA Research Monograph #137, pp. 248-289.  Rockville, MD:  National Institute of Drug Abuse.

Borkovec, T.D. (1997).  On the need for a basic science approach to psychotherapy research.  Psychological Science, 8(3), 145-147.

Borkovec, T.D. & Castonguay, L. G. (1998).  What is the scientific meaning of empirically-supported therapy?  Journal of Consulting and Clincal Psychology, 66(1), 136-142.

Borkovec, T.D. & Miranda, J. (1999).  Between-group psychotherapy outcome research and basic science.  Journal of Clinical Psychology, 55(2), 147-158. 

Brent, D., Kolko, D.J., Birmaher, B., Baugher, M., Bridge, J., Roth, C., & Holder, D. (1998).  Predictors of treatment efficacy in a clinical trial of three psychosocial treatments for adolescent depression. Journal of the American Academy of Child and Adolescent Psychiatry, 37, 906-914.

Butcher, J.N., Williams, C.L., Graham, J.R., Archer, R.P., Tellegen, A., Ben-Porath, Y.S., & Kraemmer, B. (1992).  MMPI_A (Minnesota Multi-phasic Personality Inventory- Adolescent:  Manual for administration, scoring, and interpretation.  Minneapolis:  University of Minnesota Press.  

Butcher, J.N. (1995).  Clinical Personality Assessment.  New York:  Oxford University Press

Byerly, F.C., Carlson, W.A. (1982).  Comparisons among inpatient, outpatients, and normals on three self-report depression inventories.  Journal of Clinical Psychology, 38, 797-804.

Byrd, K.R., O’Connor, K., Thackery, M., & Sacks, J.M. (1992). The utility of self-concept as a predictor of recidivism among juvenile offenders. The Journal of Psychology, 127, 195-201.

Caid, C.D. (1986).  Factors associated with disposition recommendations and recidivism for male juvenile offenders.  Unpublished master’s thesis, University of South Carolina, Columbia.

Carton, S., Jouvent, R., Bungenger, C., & Wildocher, D. (1992). Sensation  seeking and depressive mood. Personality and Individual Differences, 13, 843-849.

Carton, S., Morand, P., Bungenera, C., & Jouvent, R. (1995). Sensation-seeking and emotional disturbances in depression: relationships and evolution. Journal of Affective Disorders, 34, 219-225.

Caspi, A., Moffitt, T.E., Silva, P.A.,  Stouthamer_Loeber, M., Krueger, R.F., & Schmutte, P.S. (1994). Are some people crime-prone? Replications of the personality-crime relationship across countries, genders, races, and methods. Criminology, 32, 163-195.

Cashel, M.L., Rogers, R., Sewell, K.W., & Nolliman (1998).  Preliminary validation of the MMPI-A for a male delinquent sample:  An investigation of clinical of clinical correlates and discriminant validity.  Journal of Personality Assessment, 71(1), 49-69.

Castellano, T.C. & Soderstrom, I.R. (1992).  Therapeutic wilderness programs and juvenile recidivism:  A program evaluation.  Journal of Offender Rehabilitation, 17(3/4), 19-46.

Cernkovich, S.A., Giordana, P.C., Pugh, M. (1985).  Chronic offenders:  The missing cases in self-report delinquency research.  The Journal of Criminal Law and Criminology, 76(3), 705-732.

Chisholm, S.M., Crowther, J.H., & Ben Porath, Y. (1997).  Selected MMPI-2 scales’ ability to predict premature termination and outcome from psychotherapy.  Journal of Personality Assessment, 69(1), 127-144.

Cohen, J. & Cohen, P. (1983).  Applied multiple regression/correlation analysis for the behavioral sciences.  New York:  Lawrence Erlbaum Associates.

Daderman, A. & af Klintenberg, B. (1997).  Personality dimensions characterizing severely conduct disordered male juvenile delinquents.  Stockholm University:  Reports from the Department of Psychology, No. 831.

Davis, G.L., Hoffman, R.G., Quigley, R. (1988). Self-concept change and positive peer culture in adjudicated delinquents. Child and Youth Care Quarterly, 17, 137-145.

Davis-Berman, J. & Berman, D. (1994).  Wilderness therapy.  Dubuque, Iowa:  Kendall Hunt.

DeMaris, A. (1995).  A Tutorial in logistic regression.  Journal of Marriage and the Family, 57, 956-968.

Dembo, R., Schmeidler, J., Williams, L., Berry, E., Getreu, A., Wish, E.D., Genung, L., & LaVoie, L. (1991). Recidivism among high-risk youths: Study of  a cohort of juvenile detainees. The International Journal of the Addictions, 26, 121-177.

Dembo, R., Schmeidler, J., Nini-Gough, B., Chin Sue, C., Borden, P., & Manning, D. (1998). Predictors of recidivism to a juvenile assessment center:  A three eyar study.  Journal of Child and Adolescent Substance Abuse, 7(3), 57-77.

Draine, J., Solomon, P., & Meyerson, A. (1994).  Predictors of reincarceration among patients who received psychiatric services in jail.  Hospital and Community Psychiatry, 45, 163-167.

Duncan, R.D., Kennedy, W.A., & Patrick, C.J. (1995). Four-factor model of recidivism in male juvenile offenders. Journal of Clinical Child Psychology, 24, 250-257.

Elrod, H.P. & Minor, K.I. (1992).  Second wave evaluation of a multi-faceted intervention for juvenile court probationers.  International Journal of Offender Therapy and Comparative Criminology, 36(3), 247-263.

Evans, R.C., Levy, L., Sullenberger, T., & Vyas, A. (1991).  Self-concept and delinquency:  The ongoing debate.  Journal of Offender Rehabilitation, 16(3-4), 59-74.

Ewert, A. (1989).  Outdoor adventure pursuits:  Foundations, models, and theories.  Columbus, OH:  Publishing Horizons.

Eysenck, H.J. (1975).  The Measurement of personality.  Lancaster:  Medical and Technical Publishers. 

Eysenck, H.J. & Eysenck, M.W. (1985).  Personality and individual differences.  New York:  Plenum Press.

Farley, F.H. & Farley, S.V. (1972).  Stimulus seeking motivation and delinquent behavior among institutionalized delinquent girls.  Journal of Consulting and Clinical Psychology, 39, 94-97.

Farley, F.H. (1981).  Basic process individual differences:  A biologically-based theory if individualization for cognitive, affective, and creative outcomes.  In F.H. Farley & N.J. Gordon (Eds).  Psychology and Education:  The state of the union.  Berkeley, CA:  McCutchon Publishing. 

Fenigstein, A., Scheier, M.F., & Buss, A.H. (1975).  Public and private self-consciousness:  Assessment and theory.  Journal of Consulting and Clinical Psychology, 43, 522-527. 

Fitts, W.H. (1964).  Tennessee Self-Concept Scale.  Los Angeles, CA:  Western Psychological Services.

Fitts, W.H. (1965).  Tennessee self-concept scale manual.  Nashville:  TN:  Counselor Recordings and Tests. 

Fitts, W.H. (1988).  The Tennessee Self-Concept Scale. Los Angeles, CA: Western Psychological Services.

Garrett, C.J. ((1985).  Effects of residential treatment of adjudicated adolescents:  A meta-analysis.  Journal of Research on Crime and Delinquency, 22, 287-308.

Gillis, H.L. (1992).  Therapeutic uses of adventure-challenge-outdoor-wilderness:  Theory and research, 35-47.  Keynote Presentation given at the meeting of the Association for Experiential Education.

            Gass, M.A. (1993).  Adventure therapy:  Therapeutic applications of adventure programming.  Dubuque; Iowa:  Kendall Hunt.

            Gendreau, P., Little, T., & Goggin, C. (1996).  A meta-analysis of the predictors of adult offender recidivism:  What works!  Criminology, 34, 575-607.

            Gillis, H.L. (1992).  Therapeutic uses of adventure-challenge-outdoor-wilderness:  Theory and research.  Keynote Presentation given at the annual International Conference of the Association for Experiential Education.

Gillis, H.L. & Simpson, C. (1992).  Project Choices:  Adventure-based residential drug treatment for court referred youth.  Journal of Addictions and Offender Counseling, 12, 12-27.

            Gillis, H.L. & Thomsen, D. (1996).  A research update (1992-1995) of adventure-based therapy:  Challenge activities and ropes courses, wilderness expeditions and residential camping programs.  [on-line].  Available:  http://fdsa.gcsu.edu:6060/lgillis..

            Gottfredson, M. & Hirschi, T. (1990).  A general theory of crime.  Stanford, CA:  Stnadford Unveristy Press.

Greenwood, P.W. (1986).  Intervention strategies for chronic juvenile offenders.  New York:  Greenwood Press.

Haapanen, R.A. & Jessness, C.F. (1982).  Early identification of the chronic offender:  Final Report.  Sacramento:  Department of the Youth Authority.

Haapasalo, J. (1990). Sensation seeking and Eysenck’s personality dimensions in an offender sample. Personality and Individual Differences, 11, 81-84.

Hall, G.C., Bansall, A., & Lopez, I.R. (1999).  Ethnicity and psychopathology:  A meta-analytic review of thirty-one years of comparative MMPI/MMPI-2 research.  Psychological Assessment, 11(2), 186-197.

Han. T. (1997).  A Meta-analytic review of the effects of adventure programming on Locus of Control.  Unpublished master’s thesis.  Georgia State College.  Midegeville, Georgia. 

Harpur, T.J., Hare, R.D., & Hakstain, R. (1989). Two-factor conceptualization of psychopathy: Construct validity and assessment implications. Psychological Assessment: A Journal of Consulting and Clinical Psychology, 1, 6-17.

Harris, P.W. & Jones, P.R. (1999). Differentiating delinquent youths for program planning and evaluation. Criminal Justice and Behavior, 26, 403-434.

Harrington, R., Ruttter, M. & & Frombonne, E. (1996).  Developmental pathways in depression:  Multiple, meanings, antecedents and endpoints.  Development and Psychopathology, 8, 601-616.

Hattie, J., Marsh, H.W., Neill, J.T., & Richards, G.E. (1997).  Adventure education and Outward Bound:  Out of class experiences that make a lasting difference.  Review of Educational Research, 67(1), 48-87.

Heaven, P. (1994).  Family of origin, personality, and self-reported delinquency.  Journal of Adolescence, 17, 445-459.

Heaven, P.C.L. (1996). Personality and self-reported delinquency: A longitudinal analysis. Journal of Child Psychology and Psychiatry, 37, 747-751.

            Herbert, J.T. (1996).  Use of adventure based counseling programs for persons with disabilities.  Journal of Rehabilitation, 62(4), 3-9

Herbert, J.T. (1998).  Therapeutic effects of participating in an adventure-based therapy program.  Rehabilitation Counseling Bulletin, 41(3), 201-216.

Hodgins, S.(1993)  Mental disorder and crime.  Newbury Park:  Sage.

Hodgins, S. & Cote, G. (1993). The criminality of mentally disordered offenders. Criminal Justice and Behavior, 20, 115-129.

Hoge, R.D. & Andrews, D.A. (1996).  Assessing the youthful offender.  Plenum Press:  New York.

Hollander, H.E. & Turner, F.D. (1985). Characteristics of incarcerated delinquents: Relationship between development disorders, environmental and family factors, and patterns of offense and recidivism. Journal of the American Academy of Child Psychiatry, 24, 221-226.

Hosmer, D.W. & Lemeshow, S. (1989).  Applied logistic regression.  New York:  Wiley.

Hollin, C.R. (1990).  Cognitive-behavioral interventions with young offenders.  Elmsford, NY:  Pergamon Press.

Hollin, C.R. & Howells, K. (1996).  Clinical approaches to working with young offenders.  John Wiley & Sons:  Chichester.

Hume, M.P., Kennedy, W.A., Patrick, C.J., & Partyka, D.J. (1996).  Examination of the MMPI-A for the assessment of psychopathy in incarcerated adolescent male offenders. International Journal of Offender Therapy and Comparative Criminology, 40 (3), 224-233.

Hundley, P.L. (1989). Relationships of DUI recidivism to moral reasoning, sensation seeking, and MacAndrew alcoholism scores. Psychological Reports, 65, 1171-1174.

Jefferson, T.W. & Johnson, J.H. (1991). The relationship of hyperactivity and sensation seeking to delinquency subtypes. Criminal Justice and Behavior, 18, 195-201.

Katsiyannis, A. & Achwamety, T. (1997).  Factors related to recidivism among delinquent youth in a state correctional facility.  Journal of Child and Family Studies, 6(1), 43-55.

Kelley, F.J. & Baer, D.J. (1971).  Physical challenge as a treatment for delinquency.  Crime and Delinquency, 17, 437-445.

Kelley, M.P. (1993).  The therapeutic potential of outdoor adventure:  A review, with a focus on adults with mental illness.  Therapeutic Recreation Journal, 27(2),  110-125.

Kelley, M.P., Coursey, R.D., & Selby, P.M. (1997).  Therapeutic adventures outdoors:  A Demonstration of benefits for people with mental illness.  Psychiatric Rehabilitation Journal, 20(4), 61-73. 

Kelly, G.A. (1955).  The Psychology of personal constructs.  New York:  Norton.

Kimball, R. (1983). The wilderness as therapy. Journal of Experiential Education. 5, (3), 6-9.

Kimball, R. & Bacon, S. (1993). The wilderness challenge model. In M. Gass (Eds.), Adventure Therapy: Therapeutic applications of adventure-based therapy programming. Dubuque, IA : Kendall/Hunt.

Kraft, R., & Sakofs, M. (1985). The theory of experiential education. Boulder, CO: Association of Experiential Education.

Krueger, R.F., Schmutte, P.S., Caspi, A., & Moffit, T.E. (1994). Personality traits are linked to crime among men and women: Evidence from a birth cohort. Journal of Abnormal Psychology, 103, 328-338.

Lefkowitz, D.M. (1995). Student offenders: A proactive counseling strategy. College Student Journal, 29, 427-429.

Levenson, M.R., Kiehl, K.A., & Fitzpatrick, C.M. (1995).  Assessing psychopathic attributes in a noninstitutionalized population. Journal of Personality and Social Psychology, 68, 151-158.

Levy, K. S. C. (1997). The contribution of self-concept in the etiology of adolescent delinquency. Adolescence, 32, 671-686.

Loeber, R., Farrington, D.P., Stouthamer-Loeber, Moffitt, T.E., Caspi, A. (1998).  The development of male offending:  Key findings from the first decade of the Pittsburgh youth study.  Studies on Crime and Crime Prevention, 7(2), 141-169.

Lindgren, S.D., Harper, D.C., Richman, L.C., & Stehbens, J.A. (1986).  “Mental imbalance” and the prediction of recurrent delinquent behavior.  Journal of Clinical Psychology, 42, 821-825.

Little, G.L. & Robinson, K.D. (1989).  Relationship of DUI reasoning, sensation seeking, and the MacAndrew Alcoholism Scales.  Psychological Reports, 65, 1171.

Lipsey, M.W. & Wilson, D.B. (1993).  The efficacy of psychological, educational, and behavioral treatment:  Confirmation from meta-analysis.  American Psychologist, 48, 1181-1209.

Lueger, R.J. & Hoover, L. (1984). Use of the MMPI to identify subtypes of delinquent adolescents. Journal of Clinical Psychology, 40, 1493-1495.

Mash E.J. & Barkely, R.A.(1998).  Treatment of childhood disorders.  New York:  The Guilford Press.

McCord, D.M. (1995). Toward a typology of wilderness-based residential treatment program participants. Residential Treatment for Children and Youth, 12 (4), 51-61.

Minor, K.I. & Elrod, H.P. (1990).  The Effects of a multifaceted intervention on the offense activities of juvenile probationers.  Journal of Offender Counseling and Rehabilitation, 15(2), 87-108.

Marsh, H. W., Richards, G. E., & Barnes, J. (1986). Multidimensional self-concepts: A long-term follow-up to the effect of participation in an Outward Bound program. Personality and Social Psychology Bulletin, 12, 475-492.

Minor, K.I. & Elrod, H.P. (1994).  The Effects of a probation intervention on juvenile offenders’ self-concepts, loci of control, and perceptions of juvenile justice.  Youth and Society, 25(4), 490-511.

Montag, I. & Birenbaum, M. (1986). Psychopathological factors and sensation seeking. Journal of Research in Personality, 20, 338-348.

Myner, J., Santman, J., Cappelletty, G.G., & Perlmutter, B.F. (1998). Variables related to recidivism among juvenile offenders. International Journal of Offender Therapy and Comparative Criminology, 42, 65-80.

Nagin, D. & Farrington, D.P. (1992a).  The onset and persistence of offending.  Criminology, 30, 163-189.

Nagin, D. & Farrington, D.P. (1992b).  The stability of criminal potential  from childhood to adulthood.  Criminology, 30, 163-189.

Nagin, D. & Paternoster, R. (1991).  On the relationship of past to future participation in delinquency.  Criminology, 9, 163-189.

Neill, J.T. & Richards, G.E. (1998).  Does adventure education really work?  A Summary of recent meta-anlyses.  Unpublished manuscript.

Newcomb, M.D. & McGee, L. (1991).  Influence of sensation seeking on general deviance and specific problem behaviors from adolescence to young adulthood.  Journal of Personality and Social Psychology, 61(4), 614-628.

Newes, S.L. (2000).  Adventure-based therapy:  Theory, characteristics, ethics, and research.  Manuscript in preparation.

Newes, S.L. (2001a).  The application of empirically-supported treatment criteria to adventure-based therapy research:  Where do we stand and why should we care?  Manuscript submitted for publication.

Newes, S.L. (2001b).  Future directions in adventure-based therapy research:  methodological considerations and design suggestions.  Manuscript submitted for publication.

Oetting, E.R., Deffenbacher, J.L., & Donnermeyer, J.F. (1998). Primary Socialization Theory. The role played by personal traits in the etiology of drug use and deviance II.  Substance Use and Misuse, 33, 1337-1366.

Pena, L., Megargee, E.I., & Brody, E. (1996). MMPI-A patterns of male juvenile delinquents. Psychological Assessment, 8, 388-397.

Priest, S. (1993).  A New model for risk-taking in adventure programming.  Journal of Experiential Education, 16(1), 50-53.

Puig-Antich, J. (1982)Major depression in conduct disorder in puberty. Academy of Child Psychiatry, 21, 118-128.

Quay (1987).  Handbook of juvenile delinquency.  New York:  John Wily and Sons.

Quist, R.M. & Matshazi, D.G.M. (2000).  The child and adolescent functional assessment scale (CAFAS).  A dynamic predictor of juvenile recidivism.  Adolescence, 35, 181-192.

Reid, W.H. & Mathews, W.M. (1980).  A Wilderness experience treatment program for antisocial offenders.  International Journal of Offender Therapy and Rehabilitative Criminology, 24, 171-178.

Ringer, M. (1994).  Adventure therapy: A map of the field: Towards a definition of adventure-based therapy: Workshop Report.  Unpublished manuscript.   

Risler, E.A. (1988).  Evaluating the Georgia legislative waivers effectiveness in deterring juvenile crime.  Research on Social Work Practices, 8, 657-668.

Roberts, A.R. & Camasso, M.J. (1991).  Juvenile offender treatment programs and cost-benefit analysis.   Juvenile and Family Court Journal, 42, 37-47.

Roid, & Fiits, W. (1994).  Tennessee Self-Concept manual.  Western Psychological Association.         

Rutter, M., Giller, H., Hagell, A. (1998).  Antisocial behavior by young people.  New York:  Cambridge University Press.

Sheppard, D., Smith, G.T., & Rosenbaum, G. (1988). Use of MMPI subtypes in predicting completion of a residential alcoholism treatment program. Journal of Consulting and Clinical Psychology, 56, 590-596.

Shore, M.F., Massimo, J.L., & Ricks, D.F. (1965).  A factor analytic study of psychotherapeutic change in delinquent boys.  Journal of Clinical Psychology, 21, 208-212.

Shoham-Salomon, V. (1991).  Introduction to special section on client-therapy interaction research.  Journal of Consulting and Clinical Psychology, 59, 203-204.

Shoham, V., Bootzin, R.R., Rohrbaugh, M.J., & Urry, H. (1995).  Paradoxical vs. relaxation treatment for insomnia:  The moderating role of reactance.  Sleep Research, 24, 365.

Simo, S. & Perez, J. (1992). Sensation seeking and antisocial behavior in a junior student sample. Personality & Individual Differences, 12, 965-966.

Smith, D.A., Visher, C.A., & Jarjoura, G.R. (1991).  Dimensions of delinquency:  Exploring the correlates of participation, frequency, and persistence of delinquent behavior.  Journal of Research in Crime and Delinquency, 28, 6-32.

Smith, W.R. & Aloisi, M.F. (1999).  Prediction of recidivism among “second timers” in the juvenile justice system:  Efficiency in screening chronic offenders.  American Journal of Criminal Justice, 23, 201-222.

Tabachnik, B.G. & Fidell, L.S. (2000).  Using multivariate statistics.  Allyn and Bacon:  Boston.

Tehrani, J.A., Brennan, P.A., Hodgins, S., & Mednick, S.A. (1998). Mental illness and criminal violence. Social Psychiatry and Psychiatric Epidemiology, 33, 581-585.

Terry, N. (2001).  Project Adventure Programs for Resilient Adolescents.  Unpublished manuscript.

Tinklenberg, J.A., Steiner, H., Huckaby, W.J., & Tinklenberg, J.R. (1996). Criminal recidivism predicted from narratives of violent juvenile delinquents. Child Psychiatry and Human Development, 27, 69-79.

Tolan, P.H. & Thomas, P. (1995).  The implications of age of onset for delinquency II:  Longitudinal data.  Journal of Abnormal Child Psychology, 23(2), 157-181.

Tollett, C.L. & Benda, B.B. (1999).  Predicting “survival” in the community among persistent and serious juvenile offenders:  A 12-month follow-up study.  Journal of Offender Rehabilitation, 28, 49-76.

Van Voorhis, P.V.  (1994)  Psychological classification of the adult male inmate.  Albany:  State University of New York Press.

Wallbank, J. (1985). Antisocial and prosocial behavior among contemporary Robin Hoods.  Personality and Individual Differences, 6, 11-19.

Wallace, B. (1993).  Day persons, night persons, and variability in hypnotic susceptibility.  Journal of Personality and Social Psychology, 64, 827-841.

Wasmund, W.C. & Brannon, J.M. (1987). Integrating affective change: A re-evaluation of self-concept and peer group treatment. Residential Treatment for Children and Youth, 4, 93-101.

Watson, K.W. (1979).  Social work stress and personal belief, Child Welfare, 58, 3-12.

Weisz, J.R. (1998). Empirically supported treatment for children and adolescents:  Efficacy, problems, and prospects.  In K.S. Dobson and K.D. Criag (1998).  Empirically supported therapies.  Thousand Oaks, CA:  Sage Publishing.

Weaver, G.M. & Wootton, R. R. (1992).  The use of the MMPI special scales in the assessment of delinquent personality. Adolescence, 27, 545-554.

Webster, C.D. & Jackson, M.A. (1997).  Impulsivity.  New York:  The Guilford Press. 

Welch, G., Hall, A., & Wolkly, F. (1990).  The Replicable dimensions of the Beck Depression Inventory.  Journal of Clinical Psychology, 46, 817-827.

White, H.R., Labouvie, E.W., & Bates, M.E. (1985). The relationship between sensation seeking and delinquency: A longitudinal analysis. Journal of Research in Crime and Delinquency, 22, 197-211.

Willman, H.C. & Chun, R.Y.F. (1973).  Homeward Bound:  An Alternative to the institutionalization of adjudicated juvenile offenders.  Federal Probation, 37, 52-57.

Wilson, J. & Herrnstein, R. (1985).  Crime and human nature.  New York:  Simon and Schuster.

Zamble & Quincey (1997).  The Criminal Recidivism Process.  New York:  Cambridge University Press. 

Zuckerman, M. (1979).  Sensation seeking:  Beyond the optimal level of arousal.  New Jersey:  Lawrence Erlbaum Associates.

Zuckerman, M. (1983).  A Biological theory of sensation-seeking.  In M. Zuckerman, (Ed).  Biological basis of sensation-seeking, impulsively and anxiety (p. 37-76).  Hillsdale, NJ:  Erlbaum.

Zuckerman (1994).  Behavioral and biosocial bases of sensation seeking.  Cambridge:  Cambridge University Press.