Running head: ASSESSMENT OF SEX AND AGGRESSION Assessment of Sex and Aggression t PRBPERTY OF National Criminal Justice Reference'Service (NCJRS) Validation and Revision of the Multidimensional Assessment of Sex and Aggression "

To provide better customer service, NCJRS has made this Federally-funded grant final report available electronically in addition to traditional paper copies. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Abstract The Multidimensional Assessment of Sex and Aggression (the MASA) was initially created to supplement the often poorly represented information in the archival records of sex offenders and to provide sufficient data to classify adult sex offenders. It has now been revised four times, expanding the breadth of its assessment, simplifying its language to make it appropriate for juveniles, and computerizing its administration. This article summarizes some of the recent reliability and validity analyses that have been calculated on a wide variety of samples including college students, community non-criminals, non-sex offending criminals, and adult and juvenile sex offenders. Continued reliability and cross-sample stability of factor structures and the intercorrelations across I its scales suggest that the inventory shows promise as a useful assessment instrument for sex offenders.

, Assessment of Sex and Aggression 4 processed, so that it will be widely used. The generation of a model of sexual aggression that will be useful for guiding decisions requires the analysis of multiple domains, measured on sufficiently large samples. The extensive use of an efficient, reliable, valid, standardized multivariate assessment tool would contribute substantially to the establishment of data bases required to generate such a model.
The option of adapting existing, well standardized inventories to serve this evaluation need has not proven viable. Although the major self-report inventories that are currently available (e.g., MMPI-2, MCMI, MSI) are appropriate for and helpful with the assessments for which they were created, they are suboptimal for the assessment of sex offenders, because they do not assess the most critical sexual aggression domains (see Knight, Rosenberg, & Schneider, 1985;Prentky & Knight, 1991). Each of these major assessment instruments has its own distinct advantages, but is also burdened with significant disadvantages that compromise its utility for this purpose.
The MMPI, which is the most frequently studied psychometric instrument for sex offenders and other criminal populations (see Gearing, 1979;Knight et al., 1985), provides a rich source of empirically validated data for comparative purposes and contains the most extensively researched scales for faking good and bad.
Unfortunately, large numbers of sex offenders produce profiles that match non-offender groups (Marshall & Hall, 1995), and within clearly defined subgroups of sex offenders MMPI profiles have been characterized more by their heterogeneity than by their similarities (Erickson, Luxenberg, Walbek, & Seely, 1987;Hall, Maiuro, Vitaliano, & Proctor, 1986;Marshall & Hall, 1995). Moreover, cluster analytic studies of sex offenders have yielded inconsistent results across studies (Anderson, Kunce, & Rich, This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice. 1979; Kalichman, Szymanowski, McKee, Taylor, & Craig, 1989;Schlank, 1995;Shealy, Kalichman, Henderson, Szymanowski, & McKee, 1991), and when types have emerged, they have often pooled subjects with quite different offense histories (Marshall & Hall, 1995).
The MCMl has fared somewhat better than the MMPl with sex offenders. It is , sensitive to the antisocial and narcissistic features that are so prevalent in correctional settings (McNeil & Meyer, 1990), and it has scales that show the same stability for sex offenders as other deviant populations (Langevin et al., 1988). It has yielded some meaningful cluster groupings of sexual offenders (Bard & Knight, 1986), and the factor structure of the responses crf an inmate sample approximated those found in clinical populations (Langevin et al., 1988). Moreover, some differences among both adult and juvenile sex offender groups on the MCMl parallel the differences found in previous diagnostic literature (Carpenter, Peed, & Eastman, 1995;Chantry & Craig, 1994).
Regretfully, only a handful of studies using the MCMl with sexual offenders have been carried out, and it shares with the MMPl and the MSI the general difficulties that we discuss below.
I,* , The MSI has the distinct advantage of providing information about sexual behavior and pathology not tapped by traditional psychological tests. The scales of the MSI have shown fair internal consistencies in independent assessments of the instrument (Kalichman, Henderson, Sheaiy, & Dwyer, 1992), and its test-retest reliabilities range from .64 to .92 over a 21-day period (Nichols & Molinder, 1984).
Importantly, some of its scales have been related to improvement in sex offender treatment programs (Minor, Marques, Day, & Nelson, 1990;Simkins, Ward, Bowman, & This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice.  Rinck, 1989). Unfortunately, despite its widespread use, relatively (ittle empirical work has been done to assess its validity. Moreover, although there is some evidence of correlations with greater pathology on MMPI scales, these correlations have not been found to be consistent across samples (Kalichman et ai., 1992). The transparency of its items and potential response bias contamination are also potential drawbacks. Distortions and Immaturity, and Justifications). In a cluster analytic studies of sex offenders (Kalichman et ai., 1989;Schlank, 1995) only the Paraphilias Scale has shown cross-study discriminatory power. In addition, none of the MSI sexual dysfunction subscales were related to sex drive or sexual fantasy on the Derogatis Sexual Functioning Inventory (Kalichman et at. 1992), even though these have been found to be important components of sexual aggression (Knight, 1995).
The most telling problems with these three inventories, however, are three shared difficulties that make them all questionable instruments for evaluating sex offenders. First, no data exist on the usefulness of any of their scales for making dispositional decisions at any level for sex offenders, and no data on the most appropriate scale cutoffs or on the hit rates for various dispositional decisions are evident. Second, none of the inventories sufficiently sample all the domains that have been found critical in assessing sexual aggression (see Knight et al., 1985;Prentky & Knight, 1991 admitters, report little psychopathology, and present themselves in a favorable light (Langevin, 1988;Lanyon & Lutz, 1984). An adequate assessment tool for sex offenders must provide better solutions to the duplicity problem.
These problems explain the disenchantment with such instruments that has arisen among practitioners and researchers (e.g., Marshall & Hall, 1995). In our own research program, it also became clear to us that if we were going to integrate adequately the role of sexual behavior, cognitions, and fantasy and offense planning into the taxonomic systems we were developing for sex offenders, we needed to ', develop a self-report inventcry that met these assessment needs. Consequently, we created the Multidimensional Assessment of Sex and Aggression, the MASA, a selfreport inventory that assesses all domains necessary for classification in our taxonomic systems (see Knight, Prentky, & Cerce, 1994). It is the purpose of this article to summarize some of the recent data on the development and validation of this instrument.

Method
Historv and Desiqn of the MASA We have previously described in detail our methodology for constructing the original MASA (Knight et al., 1994 In the first revision of the MASA, which was completed seven years ago, we incorporated the assessment of additional domains that our research program had identified as important for evaluating sex offenders. For instance, items were added that assess those developmental antecedents that we have found (a) to discriminate sexually coercive non-criminals from non-sexually coercive males (Knight, 1993), (b) to predict criminal recidivism (Knight, 1999), (c) to discriminate early-onset from late-onset sex offenders (Knight & Prentky, 1993), and (d) to be correlated with the amount of injury done to victims during sexual assaults (Prentky, Knight, Sims-Knight, Straus, Rokous, & Cerce, 1989). Moreover, in this first revision an attempt to evaluate components of Hare's Psychopathy Checklist (Hare, 1980;Hare, Harpur, Hakstian, Forth, Hart, & Newman, 1990) was introduced, as well as improved lie and fake good This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice. scales.
In the second revision of the MASA we simplified the language and made it suitable for juveniles by incorporating alternative age-appropriate questions both on social competence and on sexual attitudes, behavior, cognitions, and fantasies. In revising the core of the inventory we once again item analyzed scales to assure the highest internal consistency. Our analyses from both the first and second revisions of I I "" the MASA indicated that the scales that we had introduced to assess response bias required more work. Response biases, or responding to a range of questionnaire items on some basis other than the specific item content, plague all of psychometric assessment (Paulhus, 1 9Sl), but especially the assessqent of sex offenders, who present problems not regularly associated with other patient or criminal populations (Marshall & Hall, 1995). Some of these problems stem from the fact that these offenders must talk openly about sexual behavior, a topic that engenders much anxiety and discomfort in our society. More importantly, the offender is asked to admit to behaviors that are not only socially unacceptable, but are, in fact, illegal. To date we have applied the control technique of demand reduction, by guaranteeing subjects anonymity. From the success of our reliability, internal consistency, factor analytic, and cross-group consistency analyses reported below, this has proved to be a successful strategy. If the MASA is, however, to have practical utility, it must incorporate assessments of various response biases, so that their presence can be evaluated and taken into account in situations in which anonymity cannot be guaranteed. This was a major focus of our third and fourth revisions, which we will describe briefly in the discussion section.
This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice. It comprised repetitive offenders against adult women and children. For the purposes of this report, the term rapist (n = 59) refers to an adult male whose sexual offenses were committed exclusively against adult women (Le., 16 years of age or older). A child molester was defined as someone who had at least one sexual offense against a victim During the second wave of testing, parental or legal guardian permission had to be obtained for juveniles before the testing team came on site. When the testing team arrived in the institution to administer the test, interested participants were convened in groups of 7 to 12 subjects. They were informed in more detail about the nature of the study, about the kind of material they would be asked to answer, about the protection of confidentiality they were guaranteed and about the Writ of Confidentiality we had been awarded from NIMH, and about the fee they would be paid for their participation ($1 8.00). A strong plea was made for honesty, and the potential future benefits of adequate assessment for offenders like themselves was stressed. After informed consent statements had been explained by a visiting research team member and signed both by the offender and the research team representative, either the paper-and-pencil or the computer versions of the MASA were distributed and a standard set of instructions was given. If offenders had difficulty reading the inventory, arrangements were made for one of our team to read the inventory to him.

Results
This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice. reliabilities (only two scales--Vandalism in Adulthood and Impulsivity in the Offense yielding reliabilities e .70) indicated that reasonable reliability had been achieved (Knight et al., 1994).

\
For these ratjonal scales we also reported concurrent validity coefficients, derived by correlating each scale with a parallel, independent assessment of the same domain, which was created by rating the information provided in the participants' ' I archival records. These analyses indicated that only the domains of sexualization, sexual aggression, and sexual offense planning failed to show adequate concurrent validity coefficients. A comparison between offenders' answers to the MASA scale$ for these domains and the information garnered from their archival files indicated that far more sexual preoccupation, deviance, compulsiveness, inadequacy, and sadistic fantasies and behaviors were reported on the MASA than were evident in the archival files, suggesting that the MASA provided greater validity and coverage of the relevant information than the criminal and clinical files.
Factor analvses of the MASA: The example of offense planninq. Although the general rational scales had reasonably high internal consistencies, suggesting that then items in each scale assessed the same general construct, we wanted to explore the factor structure within each domain. Consequently, for each of ten separate domains, which are listed in the left hand column of Table 2, we calculated principal components analyses (Hartman, 1967), extracted all factors with eigenvalues greater than 1, and rotated these factors to VARIMAX criteria. We examined these preliminary factors to determine the number of core factors that were cohesive and theoretically meaningful.
We then recalculated the principle components analyses, limiting the extraction to the This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice. , number of pore factors. We will present the analysis of the Offense Planning domain to I illustrate the outcome of this procedure for one domain.
In the original offense planning principal components analysis ten factors were extracted. An exaqination of this solution revealed that the last four factors were either single item or weak, splinter factors. Consequently, a six factor solution was specified, and this solution, which accounted for 65% of the variance is presented in Table 7 . In the clinical and criminal literature offense planning is often discussed as a univocal construct (Rosenberg & Knight, 1988). The factor analysis of the offense planning items challenges this notion and suggests that offense planning is a multidimensional construct, comprising the six relatively independent factors of this analysis.'' A brief consideration of these factors reveals that each represents a theoretically meaningful and separable component of offense planning And pre-offense fantasy. The first factor closely approximates a construct that has often been described in the clinical literature. It involves fantasies that Cohen,Garofalo,Boucher,and Seghorn (I 971) attributed to their compensatory rapist type and Groth, Burgess, and Holmstrom (1977) saw as characteristic of their similarly defined power-reassurance rapist. Hazelwood (1 987) has referred to these as pseudo-unselfish fantasies, and Marshall (1 989) . discussed them in the context of seeking intimacy, which is the descriptor we have chosen. In these fantasies the rapist ignores the agonistic nature of the sexual assault and fantasies that his sexual overtures will elicit a positive response in the victim. The The relatively low intercorrelation among these offense planning characteristics suggests that they are ripe for cluster analyses that might be informative either from a criminal investigative analysis perspective (Knight, Warren, Reboussin, & Soley, 1998) or from a psychotherapeutic vantage (Pithers, 1990). Kdistinct clusters of planning could be identified and replicated, these could also have important implications for crime scene analyses and for structuring relapse prevention interventions. The greater differentiation of offense planning into distinct components of planning also provides potential resolutions to the problems we encountered with the clearly inaccurate global representation of this construct in MTC:R3 (Knight, 1999).
Correlations amonq the factor domains in the oriqinal sample. We generated factor scales for each of the factors in the ten domains by standardizing each item and averaging over all items that loaded >.40 on each factor. To analyze the relations between the various components in the MASA, we correlated the factor scores of each domain with the factor scores of the other domains for the 59 rapists who had taken the original version of the MASA. Table 2  There are several important relationships in this table that should be noted.
Social competence was completely independent of all the other domains. None of the correlations of its two factor scales, Relationships and Independence, reached .01 significance with any other factor scale.
As expected, juvenile and adult antisocial behavior were highly relatgd.
Antisocial behavior was moderately related to expressive aggression, but relatively independent of other factors.
As expected, pervasive anger, expressive aggression, and sadism were all interrelated. Both sadism and pervasive anger were strongly related to expressive aggression, and expressive aggression was strongly related to pervasive anger, but sadism and pervasive anger were only weakly related.
Sexual drive was strongly related to pervasive anger, expressive aggression, sadism, and offense planning.
Pornography use was strongly related to both sexual drive and expressive aggression.
The paraphilias were moderately related to expressive aggression, sexual drive, and offense planning.

First Generalization Studv Analvses
This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice.  = 140), non-sex offending criminals (n = 162), and non-criminals (n = 220).
We repeated the trincipal components analyses described above on each of these groups and were able to replicate most of the factor structures of the original sample. Table 3 presents the average Cronbach alphas for these same factor scales on the   Perfect two group consistency on offense planning is indicated by light diagonal lines.
These same shading designations will be used in Figure 2 to provide a comparative backdrop of adult group agreement for the pattern of correlations of the juveniles.
The diagonal divides the matrix into the same two different assessments of relation that we employed in Table 2. Above the diagonal the criterion is based on the average correlation between factor scales in the two domains, and below the diagonal the criterion refers to the percent of correlations across the two domains that reached significance (gC.01). To meet the average correlation criterion above the diagonal, the average correlation among the factor scales across a factor domain pair must have reached at least .01 significance (e.g., as indicated in Figure  Reliabilitv of the MASA for iuvenile sex offenders. In addressing the question of whether Revision 3 of the MASA was an appropriate assessment tool for juveniles, we examined the internal consistencies and test-retest reliabilities of the factor scales on juvenile sex offenders we tested in our second generalization study. Table 4

I
As can be seen in Table 4, the internal consistencies on these factors were consistently high. For the juvenile sex offenders approximately 90% of the factor scale a's were greater than .79, and 67% were greater than .80. For the adults 92% exceeded .70, and 84% were greater than .80. Although the internal consistencies of the juveniles were slightly lower than those of the adults, they were still high and clearly support the use of these factor scales for juveniles. Both the juveniles and the adults  slightly lower than those of the adults, were nonetheless high and support the use of these same factors for juveniles. It is noteworthy that the factors with low test-retest reliabilities for the juveniles were exclusively in those domains that could be considered less appropriate for a juvenile sample-social competence and adult antisocial. Both juveniles and adults showed high test-retest correlations on both the Voyeurism and Sexual Behavior factors, despite the low internal consistencies found for these factor scales.
Comparison of the relation amonq factor domains for iuveniles and adults.  responding. In all other relations at least two or three adults groups showed a relation between the domains (e.g., the pervasive anger and juvenile antisocial behavior correlations showed 2 adult group agreement for the average correlation and 3 adult group agreement for the percent of correlations greater than 40%). It is interesting'that the only two completely discrepant cells involved the relation between juvenile antisocial behavior and social competence, where there was a relation for the juveniles, but not for any of the adult groups. A finding in our retrospective study sheds some light on this \ I disagreement (Knight & Prentky, 1993). Juvenile antisocial behavior and social competence were found to be the major distinguishing characteristics between offenders charged with sexual offenses as juveniles and those who had no sexually coercive behavior until adulthood. This suggests that these two characteristics might be related to being apprehended for sexual coercion as a juvenile, and thus might produce the correlation in apprehended juvenile samples that we see here. In general, the pattern of correlations strongly supports the comparability of juvenile sex offenders' patterns of responding to other groups' patterns across these factor domains.  all adult groups that paralleled the relations among the adult groups. This consistency with adult groups suggests again that the juveniles' pattern of responding on the MASA was comparable to that of the other groups and indicates that the MASA can serve as an appropriate assessment tool for juveniles.

Discussion
, As is evident in the selective data presented in this article, the MASA shows "" great promise for fulfilling the much needed role of a comprehensive assessment tool for both juvenile and adult offenders, and indeed for both criminal and non-criminal samples. In the original sample the rational scales, which had been created to measure specific theoretical domains found important in the classification of rapists, demonstrated high internal consistency and reasonable cross-temporal stability.
Moreover, these scales either correlated highly with companion scales that had been rated using archival records, or evidenced considerably more frequent admission of sexual deviance, violence, and sadism than was recorded in the archival files, This suggests that the scales were likely to have been tapping true variance and to have captured information not recorded in the archival sources.
The factor analyses calculated on the original MTC sample have not only been informative about potentially important cohesive constructs in the select MTC sample, but they also provided structures that we found to cohere across multiple, radically different samples. The factor analysis of the offense planning items, presented here as a representative example of the results of these analyses, illustrates the potential insights these factor analyses have yielded for assessment, taxonomic structuring,
Although the MASA has already proven a rich source of numerous insights into various components of sexual aggression, it will never be able to serve the role of a viable assessment tool for sexual aggression without solving the critical problem of II'I duplicity. As we indicated in the introduction, the issues of denial and lying are especially problematic for sex offenders. To date we have addressed this response bias problem by applying the control technique of demand reduction by promising and guaranteeing anonymity to all participants in our studies. From the success of our reliability, internal consistency, factor analytic, and cross-group consistency analyses, this has proven to be a successful strategy.
If the MASA is, however, to have practical utility, it must incorporate assessments of various response biases, so that their presence can be evaluated and taken into account in situations in which anonymity cannot be guaranteed. In the early versions of the MASA our use of an abbreviated Marlowe-Crowne scale (Saunders, 1991) and a newly created Sexual Behavior Lie scale, which asked about common sexual behaviors (e.g., masturbation, viewing sexual materials) that non-defensive respondents should admit, but defensive respondents might deny, proved suboptimal.
In our most recent revision to the MASA we have decided to make a concerted effort to address this problem, implementing three independent response bias assessment

27
guaranteeing the anonymity of responses, using instructional sets to fake good ot fake bad versus a set to reply honestly, testing inmates at various stages in their treatment history, and comparhg offenders whose responses closely match key information from their criminal files to those for whom there is a significant discrepancy). process of questionnaire responding rather than simply on respgnse content (e.g., Holden, 1995). The Authorware software we have used to program the MASA has the option of measuring response latencies to questions. In Versions 4 and 5 of the MASA the average differential latencies to selective, reading-length-matched questions that ask about neutral and sensitive areas of various response domains are recorded.
Holden (1 995) has demonstrated that respondents who are dissimulating by denying deviant or negative behaviors or cognitions take longer than honest respondents to "" endorse items that describe negative characteristics about themselves. Not only can the response latency measures be used to detect fakers, but Holden and Hibbs (1995) have provided evidence that the use of item response latencies can account for dissimulation variance not accounted for by standard validity indexes, and thus can add incremental validity to detecting duplicity.
The third strategy uses item response theory (IRT) to generate an appropriateness measurement aimed at identifying dishonest respondents. Zickar and Drasgow (1 996) have demonstrated that when attempting to lower false positive misclassifications this approach more successfully identified faking respondents than the social desirability content approach We have recently begun an analogue study in which we administered the MASA to male college students under three instructional sets-honest responding, faking good (encouraging them to lie to make a positive impression), and faking bad (encouraging them to lie to make a negative impression). Preliminary results indicate that Paulhus' self-deceptive positivity scale has only moderate success in differentiating fake bad subjects from honest subjects and no success in differentiating fake goods.
This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice.
Consistent with the nature of the manipulation, his impression management scale was very successful in differentiating fake bad subjects, but just missed significance in discriminating fake goods from honest respondents. Similarly, our revised sex lie scale had only moderate success in differentiating fake bads and no ability to distinguish fake good subjects. In contrast, Holden's reaction time measure discriminated both types of lying from honest responding. ! 4 We have also completed a study of 95 non-civilly committed sexual offenders at the MTC program, who were quite concerned about the issue of civil commitment and thus were primed to be very defensive. We tested them under conditions of anonymity and feedback to the clinical staff. Although we have not had the oppokuni& to analyze these data, it is already clear from the testing and data entry that the groups responded differently in the two conditions. Our hope is that by simultaneously applying to these data and to the other studies in progress all three of duplicity assessment approaches-content, latency, and IRT appropriateness measurements--we will be able to provide a sufficiently high level of duplicity detection to allow the MASA to be used as a clinical assessment instrument. Preliminary analyses are very promising, but suggest that the confluence of these three strategies will be necessary to achieve adequate detection hit rates. If we are successful in this research, it will be possible to apply the rich information that the MASA has yielded as a research instrument to clinical and dispositional decisions.

30
, status as a rapist, child molester, mixed-age offender has most frequently been made after testing by ratings from the offender's criminal history. The original factor analyses of the ten domains and the replication factor analyses in the sex offender sample both included child molesters. Consequently, the factor scales thus far generated are equally appropriate for both child molesters and rapists. Many of the child molesters tested were dismayed by their perception that critical components of their offense and sexual histories were not adequately assessed by the questions in the test. They made many excellent suggestions for changes and additions. Using both their recommendations and also incorporating domains that we had found important in our child mole,ster typology research (Knight, 1992), we have in Versions 4 and 5 created complete new subroutines assessing fixation on children, sexual preference, identification of the range of victim ages and the sex of victims, amount of contact with children, and offense modus operandi specific to child molesters. We are currently validating these new additions on sex offender samples in Maine, Massachusetts, and Minnesota.
The current version the MASA (Version 5), which exists only in a computerized format, takes full advantage of contingency based questioning and has added modules for a detailed developmental history, greater assessment of adolescent social competence, as well as the more extensive evaluation of domains unique to child molesters, just described. It now has the potential to fill the role of a standardized, comprehensive self-report instrument for sex offenders. If, as now seems possible, the duplicity problem can be adequately addressed, the next step in its development will be the creation of computer algorithms to provide user friendly feedback to aid in clinical and dispositional decision making.
This document is a research report submitted to the U.S. Department of Justice. This report has not been published by the Department. Opinions or points of view expressed are those of the author(s) and do not necessarily reflect the official position or policies of the U.S. Department of Justice.