Académique Documents
Professionnel Documents
Culture Documents
BRIEF REPORT
Frank Schilling
Reinhard Eher
Joachim Nitschke
The Severe Sexual Sadism Scale (SSSS) is a screening device for the file-based assessment of
forensically relevant sexual sadism. The SSSS consists of 11 dichotomous (yes/no) items that code
behavioral indicators of severe sexual sadism within sexual offenses. Based on an Austrian sample of 105
sexual offenders, the present study replicated the 1-dimensional scale structure of the SSSS, as evidenced
by confirmatory factor analysis. More specifically, the scale was commensurate with the 1-parameter
logistic test model (Rasch model). Reliability was estimated to be good. Criterion validity for the clinical
diagnosis of sexual sadism was good. With a cutoff value of 7 points, sensitivity and specificity were
estimated at 56% and 90%, respectively.
Keywords: sexual sadism, scale development, Rasch, psychometrics, sensitivity
765
Method
Participants
Participants were 105 adult male sexual offenders who had been
evaluated between 2002 and 2004 at the Federal Evaluation Centre
for Violent and Sexual Offenders (FECVSO) of the Austrian
Prison Service (Schilling, Ross, Pfafflin, & Eher, 2010). The
FECVSO is a department of the Austrian Federal Ministry of
Justice. Participants were included consecutively if they had a
sexual crime (rape, sexual homicide) as the index offense. In brief,
since the end of 2001 every sexual offender convicted of an
unconditional prison sentence by an Austrian court has to be
reported to the FECVSO. After a file-based risk assessment of
every offender, a substantial proportion of these reported offenders
(about 60%) are routinely seen for risk assessment by experienced
forensic psychiatrists and psychologists at the FECVSO. Selection
of offenders for clinical forensic assessment is done by one of the
following criteria: a total score of more than 5 points in the
Static-99 actuarial risk assessment instrument (Hanson & Thornton, 2000), age under 25, a prison sentence of more than 4 years,
a conviction for a child sexual abuse offense with a nonrelated
victim, and any offender reconvicted for a sexual crime. Consequently, all participants in the present sample were interviewed
clinically (Eher, Matthes, Schilling, Haubner-MacLean, & Rettenberger, 2011).
766
Results
On average, participants in the sample were 33.19 years old
(SD 9.71; range: 15.31 60.28) at the time of the judicial verdict.
The length of the determinate prison sentences of 102 participants
was 50.44 months (SD 46.02; range: 6 258) on average; an
additional two offenders were sentenced to life imprisonment.
Data on length of sentence was not available for one offender.
According to the assessment of the experienced forensic psychiatrists and psychologists at the FECVSO, a subsample of 18
participants (17.1%) were diagnosed as suffering from sexual
sadism according to DSMIVTR criteria.
A subset of six cases was chosen at random and coded independently by the third and fourth authors. Averaging joint occurrences and disparities across the six cases and focusing on the 11
items of the SSSS as derived by Nitschke et al. (2009) yielded an
overall index of interrater agreement of .58 (95% CI [.40,
.77]). According to the criteria given by Landis and Koch (1977),
this would indicate a moderate level of agreement. Concerning the
total score on the SSSS, the consistency variant of the intraclass
correlation coefficient (type [2, 1], single measure) calculated to
.82.
As Table 1 shows, most of the items of the SSSS were positively
correlated. Exceptions were the two least frequent criteria (Item 5
[Offender mutilates sexual parts of the victims body] and Item
8 [Offender mutilates nonsexual parts of the victims body]);
these two items showed negative correlations with some of the
other items. Overall, the mean tetrachoric correlation between the
11 items was .41. In an exploratory factor analysis, a one-factorial
solution (variance explained 4.72) accounted for two thirds
(68%) of the total communality (6.97) as estimated by Thurstones
method (based on maximum correlation coefficients) and nearly
half (43%) of the total variance.
Confirmatory factor analysis of the SSSS items yielded a reasonable fit for a one-factorial solution, with absolute fit indices,
2(44) 54.60, p .13, and root-mean-square error of approximation .05 (90% CI [.00, .09]). The relative fit indices, comparative fit index .89 and TuckerLewis index .87, fell below
the commonly accepted margin of .95, however. As Table 2
indicates, factor loadings ranged from .31 to .97, with the majority
of items (six out of 11) showing factor loadings well above .55. In
the light of the results from the confirmatory factor analysis,
Hypothesis 1 could be retained: The items of the SSSS appeared to
represent a one-dimensional scale.
Hypothesis 2, however, could not be retained: The number of
Guttman errors (i.e., instances in which a less frequent behavior
was present in a single case, whereas a more frequent one was not)
was too high. Across the 11 items of the SSSS, we noted 240
Guttman errors for all 105 cases. This resulted in a coefficient of
reproducibility of .79 a value that fell short of the commonly
accepted threshold of .90. Consequently, the data did not conform
to a deterministic (Guttman) scale. Because some items showed
negative correlations with other items, nonparametric item response theory (Mokken scaling) was not an option either. Consequently, we tested a one-parameter logistic (Rasch) model. The
score distribution log-likelihood of the estimated model (with 21
parameters) was 476.33, compared with a log-likelihood of the
saturated model of 382.24 at 2,047 parameters. The value of the
Akaike information criterion, based on the score distribution loglikelihood of the estimated model, was 994.66. The Akaike information criterion for the saturated model was 4858.48. Item parameters ranged from 2.96 for the most frequent item (Item 6
[Offender engages in gratuitous violence toward the victim]) to
2.66 for the least frequent Item 8 (denoting mutilation of nonsexual
parts of the victims body).
According to the item fit index (standardized weighted mean
square residual, or infit) described by Wright and Masters (1982),
all items showed acceptable fit, with the exception of Item 9
(Victim is abducted or confined), which showed significant
underfit with a weighted mean square residual coefficient of 2.16.
This means that Item 9 was too indiscriminate, with a too high
proportion of Guttman errors. The unweighted mean square residual coefficient (outfit) of 1.23 for Item 9 indicated an acceptable
model fit at p .89.
Table 1
Matrix of Tetrachoric Correlations
Variable
10
11
1
2
3
4
5
6
7
8
9
10
11
.50
.37
.68
.01
.44
.61
.18
.41
.69
.61
.62
.61
.47
.48
.32
.11
.36
.53
.32
.78
.14
.36
.16
.06
.39
.75
.56
.21
.28
.25
.31
.60
.79
.84
.22
.23
.77
.00
.30
.72
.02
.15
.10
.35
.40
.31
.15
.61
.28
.19
.46
.60
.45
.15
.81
p .05.
p .01.
767
Table 2
Factor Loadings (From Confirmatory Factor Analysis) and Item Parameters (From Rasch Modeling) of the
Severe Sexual Sadism Scale
Item
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
Percent present
Factor loading
Infit
61.0
59.0
29.5
24.8
4.8
79.0
9.5
3.8
60.0
35.2
9.5
.73
.56
.70
.92
.31
.44
.46
.40
.46
.97
.85
1.9681
1.7340
0.0250
0.4487
2.2053
2.9562
1.6583
2.6603
1.7947
0.4864
1.9417
0.1415
0.5342
0.3060
1.1040
0.3285
0.9442
0.9157
0.1077
2.1623
1.6244
0.5436
.556
.703
.620
.135
.371
.827
.820
.457
.985
.052
.293
Note. i item difficulty parameter according to Rasch scaling; infit weighted mean square standardized residual (z transform): an item fit statistic
that denotes significant underfit (i.e., lack of discrimination if infit is greater than 1.96, p .975) or overfit (i.e., an almost deterministic response pattern
if infit is less than 1.96, p .025; Wright & Masters, 1982).
Under Andrichs (1988) method (i.e., estimating error variance based on the standard error of measurement of person
parameters), the reliability of the 11-item Rasch scale was
estimated at rtt .86 in the sample. Internal consistency of the
scale was estimated at Cronbachs .75 and Guttmans 2
.78.
The test characteristic curve plots the total score on the SSSS
as a function of the underlying trait level (see Figure 1). The
test characteristic curve had the steepest ascent at a trait level of
0.05, indicating that this trait level differentiated best between cases. A trait level of 0.05 would correspond with the
range between the total scores values of 5 ( 0.32) and 6
( 0.33).
The score distribution of the sexual sadists was swamped by
the score distribution of the nonsadistic sexual offenders due to
sexual sadists representing the minority (17%) in the sample
(see Figure 2). The two graphs in Figure 2 come closest to
intersection at a raw score total of 7.55. The intersection would
Figure 1. Test characteristic curve of the Severe Sexual Sadism Scale (solid line) describing the relationship
between the latent trait () and the total score, with 95% confidence bands (dashed lines).
768
Discussion
The present study replicated the structure of an 11-item set
indicative of severe sexual sadism with a sample of 105 sexual
offenders from Austria (18 of whom had been diagnosed as sexual
sadists). Confirmatory factor analysis supported the view that the
items represent a one-dimensional scale. However, the deterministic (Guttman) properties that were identified in the construction
sample (Nitschke et al., 2009) could not be confirmed. Still, the 11
items were commensurate with the one-parameter logistic (Rasch)
model. Hence, the item set retained the properties of forming a
cumulative scale and, more importantly, of the total score as a
sufficient statistic for the underlying trait. Similarly, OMeara,
Davies, and Hammond (2011) recently applied Rasch modeling to
a brief self-rating scale from a related domain, the Short Sadistic
Impulse Scale, capturing a (nonsexual) sadistic personality trait.
Estimates of reliability as well as sensitivity and specificity were
lower than in the construction sample. Still, at rtt .86, the
reliability of the scale within the present sample met the standard
for clinical decisions (rtt .85) suggested by Rosenthal and
Rosnow (1991). In the current testing sample, the criterion validity
of the SSSS with regard to the clinical DSMIVTR diagnosis of
sexual sadism was good, with a large effect size for distinguishing
participant groups. The cutoff of 4 points suggested by Nitschke et
al. (2009) would lead to 42% false positives and 17% false
negatives, affording a biased selection ratio in the present sample
(with a base rate of 17.1% sexual sadists): Among the individuals
above the cutoff, only one in three would meet the clinical criteria
for sexual sadism. The latent trait level corresponding with the
cutoff value of 4 ( 0.97) was clearly below the margin
yielding optimum information ( 0.05). Consequently, future
research should address the question whether the cutoff of 4 points
is too low. For the higher cutoff of 7 points that afforded a more
acceptable specificity (90%) and a selection ratio above 50%, the
standard error of measurement of the associated trait level (
0.99) was 0.82. The range of 0.99 1 standard error of
measurement covers the interval [0.12, 1.81]. Expressed in terms
of the total score, this would entail the neighboring values of 6
points ( 0.33) and 8 points ( 1.64), respectively.
References
American Psychiatric Association. (2000). Diagnostic and statistical manual of mental disorders (4th ed., text rev.). Washington, DC: Author.
doi:10.1176/appi.books.9780890423349
Andersen, E. B. (1973). A goodness of fit test for the Rasch model.
Psychometrika, 38, 123140. doi:10.1007/BF02291180
Andrich, D. (1988). Rasch models for measurement. London, England:
Sage.
Berner, W., Berger, P., & Hill, A. (2003). Sexual sadism. International
Journal of Offender Therapy and Comparative Criminology, 47, 383
395. doi:10.1177/0306624X03256131
Burgess, A. W., Hartman, C. R., Ressler, R. K., Douglas, J. E., & McCormack, A. (1986). Sexual homicide: A motivational model. Journal of
Interpersonal Violence, 1, 251272. doi:10.1177/088626086001003001
Dietz, P. E., Hazelwood, R. R., & Warren, J. (1990). The sexually sadistic
criminal and his offenses. Bulletin of the American Academy of Psychiatry and Law, 18, 163178.
Doren, D., & Elwood, R. (2009). The diagnostic reliability of sexual
sadism. Sexual Abuse: A Journal of Research and Treatment, 21, 251
261. doi:10.1177/1079063209342072
Eher, R., Matthes, A., Schilling, F., Haubner-MacLean, T., & Rettenberger, M.
(2011). Dynamic risk assessment in sexual offenders using STABLE-2000 and
the STABLE-2007: An investigation of predictive and incremental validity.
Sexual Abuse: A Journal of Research and Treatment. Advance online publication. doi:10.11771/1079063211403164
Ford, L. R., Jr. (1957). Solution of a ranking problem from binary comparisons. American Mathematical Monthly, 64, 28 33. doi:10.2307/
2308513
Hanson, R. K., & Morton-Bourgon, K. E. (2005). The characteristics of
persistent sexual offenders: A meta-analysis of recidivism studies. Journal of Consulting and Clinical Psychology, 73, 1154 1163. doi:10.1037/
0022-006X.73.6.1154
Hanson, R. K., & Thornton, D. (2000). Improving risk assessments for sex
offenders: A comparison of three actuarial scales. Law and Human
Behavior, 24, 119 136. doi:10.1023/A:1005482921333
Hare, R. D. (2003). Hare Psychopathy ChecklistRevised (2nd ed.). Toronto, Canada: Multi-Health Systems.
Kingston, D. A., Seto, M., & Bradford, J. (2009, October). Sexual sadism:
Excessive violence, phallometric arousal, and diagnostic predictors of
recidivism. Paper presented at the 28th Annual Research and Treatment
769
Mokros, A., Osterheider, M., Hucker, S. J., & Nitschke, J. (2011). Psychopathy and sexual sadism. Law and Human Behavior, 35, 188 199.
doi:10.1007/s10979-010-9221-9
Nitschke, J., Osterheider, M., & Mokros, A. (2009). A cumulative scale of
severe sexual sadism. Sexual Abuse: A Journal of Research and Treatment, 21, 262278. doi:10.1177/1079063209342074
OMeara, A., Davies, J., & Hammond, S. (2011). The psychometric properties and utility of the Short Sadistic Impulse Scale (SSIS). Psychological Assessment, 23, 523531. doi:10.1037/a0022400
Richters, J., de Visser, R. O., Rissel, C. E., Grulich, A. E., & Smith,
A. M. A. (2008). Demographic and psychosocial features of participants
in bondage and discipline, sadomasochism or dominance and submission (BDSM): Data from a national survey. Journal of Sexual Medicine,
5, 1660 1668. doi:10.1111/j.1743-6109.2008.00795.x
Rosenthal, R., & Rosnow, R. L. (1991). Essentials of behavioral research:
Methods and data analysis (2nd ed.). New York, NY: McGraw-Hill.
Santtila, P., Sandnabba, N. K., & Nordling, N. (2006). Sadomasochism. In
R. D. McAnulty & M. M. Burnette (Eds.), Sex and sexuality: Vol. 3.
Sexual deviation and sexual offenses (pp. 21 45). Westport, CT: Praeger.
Schilling, F., Ross, T., Pfafflin, F., & Eher, R. (2010). Aktenbasiertes
Screening Instrument Sadismus assoziierter Merkmale (ASISAM): Entwicklung und evaluierung [File-based Screening Instrument of Criteria
Associated With Sadism (ASISAM): Development and evaluation].
Recht und Psychiatrie, 28, 183189.
Taylor, G. W., & Ussher, J. M. (2001). Making sense of S&M: A discourse
analytic account. Sexualities, 4, 293314. doi:10.1177/
136346001004003002
Warm, T. A. (1989). Weighted likelihood estimation of ability in item
response theory. Psychometrika, 54, 427 450. doi:10.1007/
BF02294627
Wright, B. D., & Masters, G. N. (1982). Rating scale analysis: Rasch
measurement. Chicago, IL: Mesa Press.
Yates, P. M., Hucker, S. J., & Kingston, D. A. (2008). Sexual sadism:
Psychopathology and theory. In D. R. Laws & W. T. ODonohue (Eds.),
Sexual deviance: Theory, assessment, and treatment (2nd ed., pp. 213
230). New York, NY: Guilford Press.