See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/11946903

The Rorschach Suicide Constellation: Assessing Various

Degrees of Lethality

Article in Journal of Personality Assessment · April 2001

DOI: 10.1207/S15327752JPA7602_13 · Source: PubMed

CITATIONS READS

49 3,088

5 authors, including:

James Christopher Fowler Dan Holdwick

Baylor College of Medicine State of Michigan
140 PUBLICATIONS 3,685 CITATIONS 13 PUBLICATIONS 846 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by James Christopher Fowler on 31 May 2014.

The user has requested enhancement of the downloaded file.

JOURNAL OF PERSONALITY ASSESSMENT, 76(2), 333–351
Copyright © 2001, Lawrence Erlbaum Associates, Inc.

The Rorschach Suicide Constellation:

Assessing Various Degrees of Lethality

J. Christopher Fowler and Craig Piers

The Erik H. Erikson Institute for Training and Research
at the Austen Riggs Center,
Stockbridge, Massachusetts

Mark J. Hilsenroth and Daniel J. Holdwick, Jr.

Department of Psychology
University of Arkansas

Justin R. Padawer
Department of Psychology
University of Tennessee

In this article we examine the relation between the Rorschach Comprehensive Sys-
tem’s Suicide Constellation (S–CON; Exner, 1993; Exner & Wiley, 1977) and
lethality of suicide attempts during the course of patients’ hospitalization at the
Austen Riggs Center (Stockbridge, MA). Patient records were rated as nonsuicidal (n
= 37), parasuicidal (n = 37), or near-lethal (n = 30) based on the presence and lethality
of self-destructive acts. Diagnostic efficiency statistics utilizing a cutoff score of 7 or
more positive indicators successfully predicted which patients would engage in near-
lethal suicidal activity relative to parasuicidal patients (overall correct classification
rate [OCC] = .79), nonsuicidal inpatients (OCC = .79), and college students (OCC =
.89). Although these predictions were influenced by relatively high base rates in the
hospital population (14.5%), base rate estimates were calculated for other hypotheti-
cal populations revealing different prediction estimates that should be considered
when judging the relative efficacy of the S–CON. Logistic regression analysis re-
vealed that an S–CON score of 7 or more was the sole predictor of near-lethal suicide
attempts among 9 psychiatric and demographic variables.
334 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

Suicide ranked as the ninth leading cause of death in America during 1994.
Macintosh (1996) estimated that 31,000 individuals would die from the direct re-
sults of suicide in 1998. The fact that suicidal behavior is the most frequently en-
countered of all mental health emergencies (Schein, 1976) motivates researchers to
develop assessment measures and prediction models to assess individuals at risk
for suicide. Over the past 40 years scientists have identified psychological (Beck,
Steer, Kovacs, & Garrison, 1985; Exner & Wiley, 1977; Maltsberger, 1986) demo-
graphic (Farberow & Shneidman, 1961; Goldstein, Black, Narsallah, & Winokur,
1991; Suokas & Lonnqvist, 1991), biochemical (Arana & Hyman, 1989), and
sociocultural (Maris, 1981) correlates and predictors of suicide.
The majority of this research has focused on predicting completed suicides,
whereas far fewer studies have predicted failed suicide attempts (medically seri-
ous self-destructive activity). Thus, the greater emphasis on completed suicides
has, in effect, eclipsed efforts to identify and predict serious suicidal behavior in
clinical populations. Medically serious, near-lethal attempts often have lasting
physical, emotional, and social consequences for the individual and their families.
The fact that 8 to 10 times as many failed suicides as completed suicides occur in
the U.S. population (Buzan & Weissberg, 1992), further highlights the need to im-
prove methods of predicting which patients will make a near-lethal suicide attempt
from those who think about, threaten, or make suicidal gestures.
Recently, Elliot, Pages, Russo, Wilson, and Roy-Byrne (1996) discriminated
between emergency room (ER) patients who made medically serious suicide at-
tempts requiring major medical interventions, versus those patients who were
quickly stabilized then released following their suicide attempt. Using explor-
atory regression analysis to identify medically serious suicide attempts from
over 40 psychiatric and demographic variables, they produced a six-variable
model that classified approximately 77% of patients into the correct groups.
Their model demonstrated that patients who made a medically serious suicide
attempt were more likely to present at the ER with a substance induced mood
disorder and were more likely to have few prior suicide attempts. They also
found that these patients had a low incidence of physical or sexual abuse history,
little or no polysubstance use, and few major psychiatric disorders. Although
retrospective rather than predictive, their study nonetheless marked an important
refinement in the assessment of suicide risk by reframing the target population
to include individuals who failed in their effort to die.
In this article we explore the efficacy of the Rorschach Comprehensive Sys-
tem’s Suicide Constellation (S–CON; Exner, 1993; Exner & Wiley, 1977) in ac-
curately predicting which seriously disturbed psychiatric patients will engage in
medically serious or near-lethal suicide attempts relative to other patients who
will engage in parasuicidal activity or no suicide attempts. The ability to predict
near-lethal suicidal activity from patients’ Rorschach responses may have partic-
ular relevance for mental health professionals who use the Rorschach in short
 ASSESSING SUICIDE LETHALITY 335

term hospital settings where relatively high numbers of patients are at risk of
suicide.
Although the Rorschach is the most widely used method for assessing suicide
risk in clinical settings (Bongar, 1991), research has yielded equivocal results. One
reason for its limited efficacy is the fact that most studies focused primarily on sin-
gle signs such as the color-shading blend (Applebaum & Colson, 1968;
Applebaum & Holtzman, 1962), symbolic content (Sapolsky, 1963), or a combi-
nation of the two (Fleisher, 1957; Hertz, 1948; Piotrowski, 1950; Sakheim, 1955).
An overarching problem with single sign approaches is the instability and lack of
statistical power of single variables in predicting or classifying any phenomena,
much less a complex behavior such as suicide. Essentially, scales with a greater
number of items from a conceptually related category will be more stable and more
reliable than a single item from the same pool (Nunnally & Burnstein, 1994).
Most content approaches have failed to demonstrate any significant relation
with suicidal behavior, whereas the more complex content or sign approaches have
not been replicated. Notable exceptions are the color-shading blend (Applebaum
& Colson, 1968; Applebaum & Holtzman, 1962; Hansell, Lerner, Milden, &
Ludolph, 1988), and transparency and cross-section responses (Blatt & Ritzler,
1974; Rierdan, Lang, & Eddy, 1978). However, these studies have shortcomings
in their small sample sizes, heterogeneous groups, and variability in time between
Rorschach administration and the index suicide attempt.
Exner and Wiley (1977) rectified a number of methodological problems when
they applied more rigorous standards in the development of the S–CON. This sta-
tistically generated constellation yielded a single score from a composite of 11
variables and ratios. In the original study of 59 completed adult suicides, the au-
thors found a total S–CON score of 8 positive indexes correctly identified 75% of
the patients who died as a result of the suicide attempt. In a follow-up study, Exner
(1993) added a 12th variable, finding a cutoff score of 8 positive indexes correctly
identified 74% of those who completed suicide. To date, four studies (Affra, 1982;
Exner, 1993; Exner & Wiley, 1977; Silberg & Armstrong, 1992) assessed predic-
tion of suicide using variables from S–CON.
Despite demonstrable results in predicting completed suicides, the utility of the
Rorschach as a predictive tool in suicide assessment continues to be criticized
(Frank, 1994). The most recent criticism (Wood, Nezworski, & Stejskal, 1996)
cited psychometric and methodological problems associated with the comprehen-
sive system, including a pointed criticism of the S–CON. This study was designed
to accurately assess the predictive validity of the S–CON by capitalizing on meth-
odological recommendations made by Farberow (1974), Exner and Wiley (1977),
and Wood, Nezworski, and Stejskal (1996). To assess the predictive validity of the
S–CON it was necessary to: (a) compute interrater reliability of the Rorschach
variables comprising the S–CON under the conditions outlined by Weiner (1991);
(b) explore the capacity of the S–CON to discriminate among psychiatric inpa-
336 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

tients who engage in near-lethal suicide attempts versus those that engage in
“parasuicidal” or no suicidal activity; (c) assess the clinical efficacy of the S–CON
in correctly predicting, in advance, which individuals will later engage in near-le-
thal suicidal behavior when compared with other clinical and nonpatient groups;
and (d) explore the relations between total S–CON scores and discrete behaviors
such as drug overdoses, self-inflicted lacerations, and emergency medical trans-
fers due to suicide attempts.

METHOD

Sampling and Group Classification

The initial sample consisted of all patients (N = 227) admitted to The Austen Riggs
Center (Stockbridge, MA) from January 1993 to May 1998. Patient records were
first masked to disguise patient identity then downloaded from the center’s elec-
tronic database. These files contained patient identification numbers, diagnostic
codes, reports of all suicidal and self-destructive behavior, detailed descriptions of
each specific behavioral manifestations of self-destructive activity, physical
trauma incurred as a result of self-destructive acts, medical procedures performed
and information regarding emergency medical transfers. Primary diagnoses were
established in a consensus case conference at the culmination of the initial evalua-
tion and treatment phase. Diagnoses were made using available sources of informa-
tion including an integration of interview data from the admission consultation,
initial contact with the therapist and psychopharmacologist, consultations with out-
patient therapists, prior hospital records, and interviews with relatives to clarify
family history of psychiatric disorders, life history, and premorbid level of func-
tioning. All patients were assessed by a board certified psychiatrist and licensed
psychologist. Diagnoses were assigned according to the diagnostic criteria of the
Diagnostic and Statistical Manual of Mental Disorders, (4th ed. American Psychi-
atric Association [APA], 1994). This method of diagnostic practice approximates
the longitudinal expert evaluation using all data (LEAD) standard of diagnosis
(Pilkonis, Heape, Ruddy, & Serrao, 1991; Skodol, Rosnik, Kellman, Oldham, &
Hyler, 1988; Spitzer, 1983). It is important to note that in all cases the initial diagno-
ses were determined prior to and independent of the Rorschach data.
Behavioral records were classified into three groups (nonsuicidal, parasuicidal,
and near-lethal suicidal) prior to collecting the archival Rorschach records. This
blind analysis ensures that ratings for lethality were made free of bias from Ror-
schach data. Criteria for inclusion into the nonsuicidal group included no suicidal
activity during index hospitalization, length of stay of 6 months or more, a com-
plete Rorschach administered during the first 30 days of the index hospitalization,
and minimum age of 18.
 ASSESSING SUICIDE LETHALITY 337

Criteria for discriminating between parasuicidal and near-lethal suicidal pa-

tients were determined by assessing the database of incident reports of self-de-
structive behavior. The methodology used for categorization is similar to the
Lethality of Suicide Attempt Rating Scale (LSARS; Smith, Conroy, & Ehler,
1991), but differs in that our discriminations were dichotomous whereas the
LSARS makes finer discriminations. Assessments were based on a thorough eval-
uation of: (a) actual suicidal behavior (ingesting toxins, hanging, cutting, etc.); (b)
the likelihood of an attempt being interrupted (from timing the suicide attempt so
death can be interrupted, to insuring that no interruption of the suicide attempt is
probable); and (c) relative lethality of drug overdose based on an extensive list of
drugs, dosage, and their relative toxicity. Behaviors such as minor drug overdoses
(10mg of Ativan, mixed with alcohol) or self-inflicted lacerations requiring super-
ficial stitches were coded as parasuicidal behavior. Behaviors such as ingesting
250 Tylenol (500mg) capsules and deep medial wrist cutting requiring emergency
transfer warranted inclusion into the near-lethal suicidal group. Independently
coded ratings by the first and second authors for all records demonstrated high re-
liability for classification of patients into nonsuicidal, parasuicidal and near-lethal
groups (κ = .96). After reliability estimates were established all disagreements in
ratings were discussed until consensus was reached.
Starting from the initial pool of all 227 records, selection criteria were then
used to limit and define the sample to be studied. Because all patients admitted
to the center are expected to undergo psychological testing as part of their initial
assessment, all patients were initially eligible for inclusion in the study. In an ef-
fort to ascertain some degree of confidence that the nonsuicidal group was repre-
sentative of patients who do not make suicidal or parasuicidal attempts, our
inclusion criteria were conservative. Patients with a length of stay less than 6
months (n = 68) did not allow for adequate sampling of behavior to be confident
in classifying them as nonsuicidal. Patients who refused projective testing or
who had incomplete or illegible Rorschach protocols (n = 29) were eliminated
because we could not extract the structural variables for the analysis. Because
we were interested in predicting near-lethal and parasuicidal activity in a clini-
cally relevant time frame, suicidal activity that occurred after 60 days
postadministration were eliminated (n = 18), including the only completed sui-
cide that occurred 1 year after the patient was administered the Rorschach.
Among these 18 patients, over two-thirds had made a suicide attempt or engaged
in parasuicidal behavior over a year after being administered projective testing,
therefore seriously limiting the temporal relevance of the Rorschach. Finally,
there were a small number (n = 8) of medical records that lacked adequate detail
to ensure accurate classification.
The final sample of 104 adult inpatients, consisted of 97 women and 7 men with a
mean age of 29.3 (SD = 9.7) at admission. Four men were classified in the
nonsuicidal group, 2 in the parasuicidal and 1 in the near-lethal group. The average
338 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

number of years of education completed by the patients was 14.9 (SD = 1.95). A total
of 90 patients were single (never married), 10 were married, and 4 had been divorced
or widowed. Four patients were Asian and the remaining 100 were White.
The college students that served as a nonclinical comparison group consisted of
participants enrolled in undergraduate psychology classes at a large Southern uni-
versity. Participants in this group were screened for a history of psychiatric dis-
tress, psychotherapy, or psychiatric hospitalization. A sample of 50 participants
were administered the Rorschach by advanced graduate students in an APA ap-
proved clinical psychology doctoral program. This group included 25 men and 25
women whose years of education ranged from 13 to 18 (M = 14.8) and whose mean
age was 22.6.

Procedure

The administration and scoring of Rorschach protocols followed the procedures

and guidelines articulated by Exner (1993). All patients were administered the Ror-
schach within 30 days of admission during the initial evaluation and treatment
phase of hospitalization. Rorschach protocols utilized in this study were drawn
from an archival search of hospital psychological evaluation files. To make tempo-
rally relevant predictions from Rorschach protocols patient records were included
only if the index suicidal activity took place within 60 days following the adminis-
tration of the Rorschach. The rationale for the 60-day interval was based on a previ-
ous finding (Exner & Wiley, 1977) that suicidal activity is highly influenced by
current stressors and that the mental state associated with suicide often shifts within
a relatively short period of time.
For the purposes of a larger study of suicidal and self-destructive behaviors,
all Rorschach protocols were rescored (on the comprehensive system and psy-
choanalytic content scales) by the first author who was blind to all identifying
information as well as group assignment. Twenty protocols were then randomly
selected and scored independently by the second author. The 20 sets of scored
protocols were compared on a response by response basis, to ascertain interrater
reliability (Weiner, 1991). Kappa coefficients and percentage of agreement were
calculated between the two judges for all major scoring categories that comprise
the ratios making up the S–CON. The total percentage of agreement and Kappa
coefficients for all scoring categories are presented in Table 1. After reliability
estimates were established all disagreements in ratings were discussed until con-
sensus was reached.

Analyses

The data analyses proceeded in five steps. First, demographic and psychiatric
variables as well as Rorschach productivity (R) were contrasted across inpatient
 ASSESSING SUICIDE LETHALITY 339

TABLE 1
Rorschach Interrater Reliability

Rorschach Variable % Agreement κ

Location 100 1.00

Determinants 99 0.97
Form level 100 1.00
Content 99 0.98
Pairs 100 1.00
Z score 100 1.00
Special scores 97 0.96

groups using analysis of variance (ANOVA) to identify potential sources of

covariance. Second, ANOVA compared total S–CON scores across the inpatient
and nonclinical groups. When analyses revealed significance differences (p <
.05), post-hoc analysis using Tukey’s honestly significant difference (HSD) was
used for pairwise comparisons. The third phase of analysis assessed the incre-
mental validity of the S–CON by entering other Rorschach indexes (schizophre-
nia index, depression index, and coping deficit index) with the S–CON in a
stepwise logistic regression analyses. Logistic regression was used because it is
a more sensitive and conservative statistic when dichotomous and categorical
variables are under investigation (Davis & Offord, 1997). Fourth, diagnostic ef-
ficiency statistics (Kessel & Zimmerman, 1993) were calculated for the S–CON
at four different points: greater than or equal to six, seven, eight, and nine. These
comparisons determine the S–CON’s ability to differentiate near-lethal suicidal
patients from parasuicidal patients, near-lethal patients from nonsuicidal patients
and near-lethal patients from college students. The five different statistics calcu-
lated were: (a) sensitivity (SN; the ability of a “positive” S–CON score to cor-
rectly identify individuals exhibiting near-lethal suicide attempts), (b) specificity
(SP; the ability of a “negative” S–CON score to correctly identify those individ-
uals with no near-lethal suicide attempts), (c) positive predictive power (PPP;
the probability that an individual will engage in a near-lethal attempt when the
S–CON score is “positive”), (d) negative predictive power (NPP; the probability
that an individual will not engage in a near-lethal suicide attempt when the
S–CON score is “negative”), and (e) overall correct classification (OCC: the
overall “hit rate” correctly classified by the S–CON). Phase 5 included correla-
tion and logistic regression analyses of the Rorschach, demographic, and psychi-
atric variables to explore their relation to specific incidence of self-inflicted
lacerations, overdose attempts, and emergency transfers to a medical facility.
The rationale for this final step is based on previous recommendation to include
dimensional analyses of specific real-world behaviors that eliminate the issue of
rater judgment regarding lethality (Wood et al., 1996).
340 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

RESULTS

ANOVAs contrasting the three clinical groups (see Table 2) revealed no significant
differences in patient age, level of education, FSIQ, Rorschach productivity, or
global assessment of functioning (GAF) scores. The groups were also well matched
on sex (predominantly women) and marital status (predominantly single, never mar-
ried). Based on previous research linking serious psychological impairment to
comorbid Axis I and Axis II disorders (Oldham et al., 1995), the total number of Axis
I and Axis II disorders were also contrasted, F(2, 100) = 2.4, p < .09. Diagnostic char-
acteristics of the inpatient sample highlight considerable similarities among the
groups. Approximately 97% of the patients were diagnosed with major depressive
disorder and a Cluster B Axis II disorder. Of these patients, 100% were diagnosed
with at least two comorbid Axis I disorders. The fact that there were no significant di-
agnostic differences suggests that level and type of psychopathology were not sa-
lient predictors or covariants for suicidal behavior. Furthermore, the similarities
between groups suggest that results reported next are based on characteristics as-
sessed by the Rorschach’s capacity to predict near-lethal suicide attempts rather than
simply measuring the level of psychopathology in these clinical groups.
ANOVAs of total S–CON score (see Table 3) revealed significant differences
among the three clinical groups, F(2, 100) = 14.3, p < .00001. Post-hoc analysis
(Tukey’s HSD) confirmed that the near-lethal group mean scores were signifi-
cantly higher than the parasuicidal and nonsuicidal patient groups. To determine
the magnitude and relative importance of this F value, we calculated the effect size
using a formula for η (Cohen & Cohen, 1983). η is a parameter estimate similar to
a correlation coefficient derived from the ANOVA model. Cohen (1977) noted
that the relative magnitude of correlations are equivalent to small, medium, and
large effect sizes, r = .20, .50, and .80, respectively in the psychological sciences.

TABLE 2
Contrasts for Demographic Variables

Nonsuicidala Parasuicidala Near-Lethalb

M SD M SD M SD F p

Age 28.9 9.90 29.8 9.90 29.2 9.30 0.09 .91

Education 15.0 1.90 14.4 2.00 15.3 1.80 1.50 .22
FSIQ 110.6 12.90 107.7 13.60 114.4 12.10 1.90 .15
Rorschach R 25.1 12.30 23.4 10.20 23.5 13.90 0.24 .79
Total no. of Axis I & II 3.3 1.00 3.7 1.20 3.7 1.20 2.40 .09
GAF 39.1 6.88 39.1 5.66 40.0 5.75 0.24 .78

Note. Rorschach R = total number of Rorschach responses; FSIQ = Wechsler Full Scale IQ; GAF =
global assessment of functioning.
an = 37. bn = 30.
 ASSESSING SUICIDE LETHALITY 341

TABLE 3
Inpatient Group Comparisons of Rorschach Constellation Scores

Nonsuicidala Parasuicidala Near-Lethalb

M SD M SD M SD F p

S–CON 5.2 1.5 5.0 1.6 7.0 1.7 14.30 .00001

DEPI 4.0 1.3 3.9 1.4 4.2 1.0 0.55 .57
CDI 2.8 1.6 2.2 1.6 2.1 1.1 3.00 .06
SCZI 2.7 1.7 2.9 1.7 3.3 1.8 1.20 .30

Note. S–CON = Rorschach Suicide Constellation; DEPI = Depression Index; CDI = Coping Deficit
Index; SCZI = Schizophrenia Index.
an = 37. bn = 30.

Estimating the effect size for the ANOVA revealed a moderate effect (η = .47). A
separate ANOVA was conducted by randomly selecting 50 inpatients then con-
trasting this group’s total S–CON scores with the college student’s total scores.
Expectable differences were found between the inpatient group and the college
students’ scores on this constellation, F(3, 150) = 108.05, p < .00001, with a large
effect size (η = .84).
Because previous research has linked suicidal activity to the presence of de-
pression, deficits in coping, and thought disorder, it seemed possible that Ror-
schach measures associated with these phenomena may also predict the level of
lethality among our patient population. Thought disorder as measured by the
Schizophrenia Index (SCZI), F(2, 100) = 1.2, p = .30, depression as measured by
the Depression Index (DEPI), F(2, 100) = .55, p = .57, and deficits in resources to
cope with current external stressors, as measured by the Coping Deficit Index
(CDI), F(2, 100) = 3.0, p =.06, failed to differentiate level of lethality among the
clinical groups. Although the CDI approached statistical significance, the effect
size was small (η = .21) and the mean scores were well below the cut-off point for
this scale. Next, the four indexes (S–CON, SCZI, DEPI, and CDI) were entered
into a stepwise logistic regression analysis using near-lethal suicide attempts as the
predictor variable. Total S–CON score was the only variable that made entry into
the prediction equation (β = .7532, Wald = 16.78, R = .35, p < .00001).
Next, diagnostic efficiency statistics were calculated using total S–CON score
as the criterion variable for the groups. Results (found in Table 4) indicate that a to-
tal S–CON score of seven or greater predicts those patients who will later make a
near-lethal suicide attempt versus those patients who will engage in parasuicidal
activity or no suicidal activity. Scores below seven on the S–CON resulted in
much lower prediction rates. For example, an S–CON score greater than or equal
to six resulted in lower levels of PPP in predicting near-lethal suicide attempts
from parasuicidal patients (PPP = .63) and from nonsuicidal patients (PPP = .57).
342 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

By contrast, the overall correct classification rate for an S–CON score greater than
or equal to seven was 79% for near-lethal versus parasuicidal and nonsuicidal in-
patients indicate strong predictive validity. An S–CON score of seven or more led
to 81% true positive rate (PPP) for predicting near-lethal suicide attempts versus
parasuicidal activity and for nonsuicidal patients. The high NPP (the probability
that an individual will not engage in a near-lethal suicide attempt when the S–CON
score is below 7) indicates a high rate of true negative predictions as well. The re-
sults of diagnostic efficiency for comparison between the inpatient near-lethal
group and the college sample were also robust. Although the OCC for this analysis
was 89% (with PPP = 1.0; NPP = .85), it should be noted that none of the college
students were classified as near-lethal.
Because the Austen Riggs Center admits a relatively high number of suicidal
patients, the base rate for near-lethal suicide attempts may be significantly higher
than general psychiatric hospitals (44.8% base rate in this study; 14.5% base rate in
the center population over the 5-year period of the study). Therefore, an effort was

TABLE 4
Diagnostic Efficiency Statistics for Various Levels of Lethality

SN SP PPP NPP OCC

Near-lethala vs. parasuicidalb

S–CON ≥ 6 .83 .60 .63 .82 .70
S–CON ≥ 7 .70 .87 .81 .78 .79
S–CON ≥ 8 .27 .97 .89 .62 .66
S–CON ≥ 9 .20 .97 .86 .60 .63
Near-lethala vs. nonsuicidalb
S–CON ≥ 6 .83 .49 .57 .78 .64
S–CON ≥ 7 .70 .87 .81 .78 .79
S–CON ≥ 8 .27 .95 .80 .61 .64
S–CON ≥ 9 .20 1.0 1.0 .61 .64
Near-lethala vs. college studentsc
S–CON ≥ 6 .83 1.0 1.0 .91 .94
S–CON ≥ 7 .70 1.0 1.0 .85 .89
S–CON ≥ 8 .27 1.0 1.0 .69 .72
S–CON ≥ 9 .20 1.0 1.0 .68 .70

Note. SN = sensitivity, the ability of the S–CON to correctly identify patients exhibiting near-lethal
suicide; SP = specificity, the ability of the S–CON to correctly identify individuals with no near-lethal
suicide attempts as not having near-lethal suicide attempts; PPP = positive predictive power, the
probability that an individual has engaged in near-lethal attempt when the S–CON identifies him or her
as having had a near-lethal suicide attempt; NPP = negative predictive power, the probability that an
individual has not engaged in a near-lethal suicide attempt when the S–CON identifies him or her as not
having made a near-lethal attempt; OCC = overall correct classification, the overall proportion of near-
lethal suicide patients and individuals with no near-lethal suicide attempts correctly classified by the
S–CON; S–CON = Rorschach Suicide Constellation.
an = 30. bn = 37. cn = 50.
 ASSESSING SUICIDE LETHALITY 343

TABLE 5
Diagnostic Efficiency Statistics for Hypothetical Populations With Low Base Rates

SN SP PPP NPP OCC

Base rate of 20%

Near-lethala vs. parasuicidalb
S–CON ≥ 7 .70 .87 .57 .92 .84
Base rate of 14.5%
Near-lethala vs. parasuicidalb
S–CON ≥ 7 .70 .87 .48 .94 .85
Base rate of 6.6%
Near-lethala vs. parasuidalb
S–CON ≥ 7 .70 .87 .28 .98 .86
Base rate of 2%
Near-lethala vs. parasuidalb
S–CON ≥ 7 .70 .87 .10 .99 .87

Note. SN = sensitivity, the ability of the S–CON to correctly identify patients exhibiting near-lethal
suicide; SP = specificity, the ability of the S–CON to correctly identify individuals with no near-lethal
suicide attempts as not having near-lethal suicide attempts; PPP = positive predictive power, the
probability that an individual has engaged in near-lethal attempt when the S–CON identifies him or her
as having had a near-lethal suicide attempt; NPP = negative predictive power, the probability that an
individual has not engaged in a near-lethal suicide attempt when the S–CON identifies him or her as not
having made a near-lethal attempt; OCC = overall correct classification, the overall proportion of near-
lethal suicide patients and individuals with no near-lethal suicide attempts correctly classified by the
S–CON; S–CON = Rorschach Suicide Constellation.
an = 30. bn = 37.

made to address the potential inflation of true positive prediction associated with
relatively high base rates (Meehl & Rosen, 1955). We computed diagnostic effi-
ciency statistics for a cutoff score of seven or greater for four hypothetical samples
with base rates of 2%, 6.6%, 14.5%, and 20%. Working from the assumption that
sensitivity and specificity remain constant, we calculated positive and negative
predictive power as well as overall correct classification rates. The results in Table
5 indicate that PPP declines as a function of lower base rates, resulting in a higher
proportion of false positive predictions in populations with low base rates. The op-
posite trend occurs with improvement in NPP in all populations (higher true nega-
tive predictions) for near-lethal suicide attempts.
In each hypothetical sample, the use of the S–CON to predict near-lethal sui-
cide attempts is a marked improvement over guessing that all patients in that envi-
ronment will make a near-lethal suicide attempt from knowledge of the base rate.
In the actual study, the base rate for near-lethal suicide attempts at the Austen
Riggs Center from 1993 to 1998 was 14.5%. Thus, a clinician without the S–CON
scores would be correct in predicting near-lethal attempts only 14.5 out of 100
times, whereas the same clinician with access to S–CON scores could correctly
344 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

TABLE 6
Dimensional Correlates of Total S–CON Scores With Suicidal Behavior

Self-Inflicted
Overdose Incidencea Emergency Transferb Lacerationsc

r p r p r p

Total S–CON .32 .001 .28 .004 .08 .40

Total SCZI .24 .01 .17 .08 .08 .37
Total DEPI .04 .69 .13 .18 –.04 .65
Total CDI –.25 .01 –.22 .02 –.36 .0001

Note. N = 104. S–CON = Rorschach Suicide Constellation; SCZI = Schizophrenia Index; DEPI =
Depression Index; CDI = Coping Deficit Index.
aM = .87, SD = 1.57. bM = 1.48, SD = 2.05. cM = 5.78, SD = 7.53.

predict near-lethal suicide attempts 48 times out of 100 when a score of seven or
greater was present.
Although the primary assessment of the study focused on the accurate prediction
of near-lethal suicide attempts, it is of considerable interest to examine the relations
among the comprehensive system’s major indexes and specific self-destructive be-
haviors. Total S–CON, SCZI, DEPI, and CDI scores were correlated with overdose
attempts, cutting and emergency medical transfers (see Table 6). Results indicate
that total S–CON scores were significantly correlated with overdose attempts, r =
.32, p = .001, and emergency medical transfers, r = .28, p = .004. Given the high num-
ber of superficial wrist cutting episodes in our sample, this nonsignificant correla-
tion, r = .08, p = .40, provides good discriminant validity for the S–CON. In addition,
the level of thought disorder manifested on the Rorschach is correlated with drug
overdoses, SCZI: r = .24, p = .01. The degree to which patients appear to be over-
whelmed by psychological and emotional stressors is negatively correlated with
drug overdoses, CDI: r = –.25, p = .01, emergency medical transfers, CDI: r = –.22, p
= .02, and self-inflicted lacerations, CDI: r = –.36, p = .0001.
The first logistic regression analysis (see Table 7) examined emergency medi-
cal transfers using a stepwise conditional entry process. First, demographic and
psychiatric variables (age; GAF scores; total number of Axis I and II diagnoses;
current substance dependence or abuse disorder; Wechsler FSIQ scores; the Ror-
schach CDI, DEPI, and SCZI) were entered to test potential prediction of emer-
gency medical transfers. None of the eight variables predicted emergency transfer
to a medical center. When total S–CON score was entered into the equation under
the same stepwise conditional entry process it contributed to the prediction of
emergency medical transfers. By computing the logits for predicting emergency
medical transfers (logit[ER] = B[S–CON score] – constant), we can translate them
into probabilities using the formula: elogit(ER) / (1 + elogit(ER) ). Although the general
probability that the average inpatient would be transferred to an emergency medi-
 ASSESSING SUICIDE LETHALITY 345

cal center is 37.5%, knowledge of a patient’s score on the S–CON greatly im-
proves the prediction of which patients are more likely to be transferred. For
example, the probability that a patient with an S–CON score of three will be trans-
ferred is only 19%, whereas a patient with a score of seven has a 50% probability
of being transferred. Finally, a patient with a score of nine has approximately a
67% probability of being transferred to a medical facility for engaging in self-de-
structive behavior.
The second logistic regression analysis (see Tables 8 & 9) examined the rela-
tion between the dependent variables (age; GAF scores; total number of Axis I and
II diagnoses; current substance dependence or abuse disorder; Wechsler FSIQ
score; the Rorschach CDI, DEPI, and SCZI) and the incidence of drug overdose.
Of the eight variables entered into the stepwise conditional analysis only FSIQ
predicted the presence or absence of drug overdose (B = .0461). Total S–CON
score was then entered into the equation under the same stepwise conditional entry
process to determine if it contributes anything more to the prediction of drug over-
dose (B = .4285). Computing the logits for predicting overdose attempts requires a
more complex equation (logit[OD] = B[FSIQ] + B[S–CON] – constant). Trans-
lating the logits into probabilities for three FSIQ scores (FSIQ = 80, 100, 120) and

TABLE 7
Logistic Regression Analyses of Emergency Transfer Status As a Function
of Total S–CON Score

–2 Log
Step Dependent Variable Likelihood Model χ2 p B Wald p Level Expected B

1 Demographic and ns
psychiatric variablesa
2 Total S–CON 89.082 6.183 .01 .3617 5.50 .01 1.4357

Note. S–CON = Rorschach Suicide Constellation.

aAge, full scale IQ, GAF scores, Total Axis I and II disorders, substance abuse diagnosis, Rorschach Coping

Deficit Index, Depression Index, and Schizophrenia Index.

TABLE 8
Logistic Regression Analyses Predicting Patient Overdose Attempts

–2 Log
Step Dependent Variable Likelihood Model χ2 p B Wald p Level Expected B

1 FSIQa 71.70 4.578 .03 .0461 4.2971 .0382 1.0472

2 Total S–CON 66.098 5.602 .02 .4285 4.9950 .0254 1.5350

Note. S–CON = Rorschach Suicide Constellation.

aStep 1 also included age, GAF scores, Total Axis I and II disorders, substance abuse diagnosis, Rorschach

Coping Deficit Index, Depression Index, and Schizophrenia Index.

346 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

TABLE 9
Probability Matrix for Overdose Outcome As a Function of Full Scale IQ
and S–CON Scores

Full Scale IQ

S–CON 80 100 120

3 .03 .05 .10

5 .06 .12 .20
7 .14 .24 .37
9 .27 .42 .58

Note. S–CON = Rorschach Suicide Constellation.

four S–CON scores (3, 5, 7, and 9) results in 12 probabilities (see Tables 8 & 9).
Although the general probability that inpatients would make an overdose attempt
is 25%, knowledge of a patient’s FSIQ and S–CON scores improve the prediction
of which patients are more likely to overdose. For example, patients who score in
the borderline range of intellectual functioning (FSIQ = 80) and who attain an
S–CON score of three have only a 3% chance of taking an overdose. On the other
hand, those patients who score in the superior range of intellectual functioning
(FSIQ = 120 or higher) and who attain a S–CON score of nine or greater have a
58% chance of taking an overdose.

DISCUSSION

Prediction and early intervention of suicide and near-lethal suicidal activity is the
ideal goal of empirical analysis and clinical practice. Although current knowledge
of the determinants of suicide prohibits absolute accuracy in prediction, the Ror-
schach S–CON provides important data to clinicians regarding the relative risk for
near-lethal suicidal activity. The results of this study, when considered in the con-
text of past research, points to the unique characteristics that render the S–CON
valuable in the assessment of suicide risk.
First, the S–CON is a strong predictor of future near-lethal suicidal behavior in
the current sample of highly disturbed and at-risk patients. This finding is impor-
tant because it strengthens Exner and Wiley’s (1977) observation that the S–CON
captures aspects of psychological experience dominant in people who may eventu-
ally act on lethal self-destructive impulses. The fact that the diagnostic efficiency
statistics generated a cutoff score of seven that differentiated near-lethal from
parasuicidal patients is also important because it points to a broader utility of this
constellation in assessing various degrees of lethality. A S–CON score of 7 or
greater predicted near-lethal suicidal behavior that surpassed prediction based on
 ASSESSING SUICIDE LETHALITY 347

knowledge of the population base rates (Meehl & Rosen, 1955). The fact that the
S–CON did not predict parasuicidal behavior demonstrates a measure of
discriminant validity of this constellation. Furthermore, other comprehensive sys-
tem indexes and behavioral indicators of pathology (such as number of diagnoses
and GAF scores) failed to predict near-lethal suicide attempts. This result suggests
that the total S–CON score is associated with ecologically valid, real-world behav-
iors of serious suicide attempts, rather than assessing general impulsivity or self-
destructive trends.
In the broader scope of suicide research these Rorschach findings compare fa-
vorably to Elliot et al.’s (1996) study of medically serious suicide attempts in a
general hospital population. Although both studies found comparable degrees of
overall prediction for type of suicide attempt, true positive prediction of medically
serious attempts was only 28% compared to 81% prediction of near-lethal at-
tempts in our sample. Furthermore, their exploratory regression analyses of 40
variables may capitalize on chance factors in the data, thus making the final regres-
sion model potentially unstable (Tabachnick & Fidell, 1996). For this reason their
study requires replication to validate their final model, whereas the S–CON now
has at least five distinct studies justifying Exner’s (1993) S–CON equation. The
Elliot study was also retrospective, whereas this study was prospective by predict-
ing near-lethal suicide attempts within 60 days following the administration of the
Rorschach. This study may, therefore, generalize better to the realities facing clini-
cians working in short term acute care hospitals where a high percentage of pa-
tients are at risk for attempting suicide.
A reasonable argument could be made that simple paper–pencil question-
naires may be equally sensitive at predicting suicide and suicide attempts and
can be far less expensive. To explore this issue, we will briefly discuss two of
the better known self-report measures, the Linehan Reasons for Living Inven-
tory (LRLF; Linehan, Goodstein, Nielsen, & Chiles, 1983) and the Beck Hope-
lessness Scale (BHS; Beck, 1986). Although neither paper–pencil measure has
differentiated near-lethal from parasuicidal patients, it may be useful to draw
some comparisons between these measures and the S–CON. The LRLF has been
used with a variety of populations to differentiate parasuicidal adults from
nonsuicidal individuals. In a recent study, the LRLF (when combined with three
other self-report measures) was quite successful in discriminating among pa-
tients who had previously made a suicide attempt from a matched clinical con-
trol group (Osman et al., 1999). However, when the entire sample of inpatients
were followed postdischarge, the LRLF did not predict future incidence of sui-
cide attempts. A second limitation of the LRLF is the fact that past studies have
relied on individual self-reports and ratings of parasuicidal behavior and suicidal
ideation as target variables, rather than verifiable lethal or near-lethal suicide at-
tempts. Thus, the scale appears to have significant utility in identifying some of
the cognitive components that are associated with suicidal ideation and affects
348 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

following a suicide attempt, yet has not demonstrated predictive validity of ei-
ther completed suicide, or near-lethal attempts.
The BHS is considered a powerful predictor of later suicide (Bongar, 1991).
Beck and colleagues (Beck et al., 1985) found in study of 207 hospitalized patients
that hopelessness was highly predictive of suicide: Ten out of 11 patients (90.9%)
with BHS scores of 10 or greater died as a direct result of suicide at a 10-year fol-
low-up. In our effort to contrast the S–CON prediction of near-lethal attempts
against the BHS prediction of suicides, we computed diagnostic efficiency statis-
tics (Kessel & Zimmerman, 1993) on Beck’s data. The BHS data yields better SN
(.91) and NPP (.99) than the S–CON. However, the PPP (.12) and OCC (.51) dem-
onstrates that a score of 10 or greater leads to a high number of false positive pre-
dictions. When we make base rate adjustments for the S–CON to match the Beck
study base rate (see Table 5, base rate of 6.6%), we find that the PPP (.28) of the
S–CON is still twice that of the BHS. In addition, the OCC rate (.86) for the
S–CON is substantially higher. It is arguable, based on this limited comparison of
the S–CON and the paper–pencil measures, that the use of the comprehensive sys-
tem’s S–CON is justifiable in clinical settings in which a relatively high base rate
of attempted suicides occur during and shortly after discharge. The data from the
S–CON may prove valuable to clinicians who are faced with complex discharge
decisions and who need to weigh the risks and benefits of extending hospital stays
for patients at risk for suicide.

CONCLUSIONS

Although this study demonstrates strong predictive validity of the S–CON, a num-
ber of cautions are warranted. First, the preponderance of women in the psychiatric
sample raises questions about the of these findings to largely male samples such as
prison populations. Further study will be needed to ascertain if these findings can
be replicated in male samples. Second, the base rate of this study may be higher than
those found in other hospital populations. Use of the S–CON in other hospital set-
tings will therefore require careful consideration of base rates relative to the predic-
tive power of the S–CON in each setting. Third, we have drawn clear distinctions
between near-lethal and parasuicidal individuals for the purpose of this study, but
we know from experience that all patients who think about suicide and make sui-
cidal gestures are at risk for killing themselves. Because unanticipated life events,
losses, and disappointments can be devastating (especially for those contemplating
suicide) a score on any scale will fall short of accurately predicting all suicidal be-
havior. In addition, we know that borderline patients with histories of self-mutilat-
ing behavior (clearly definable as parasuicidal) are at higher risk of killing
themselves than non-self-mutilating borderline patients (Stone, Hurt, & Stone,
1987). It should therefore be obvious that confidence in using the S–CON to predict
future suicidal behavior must be tempered with responsible caution.
 ASSESSING SUICIDE LETHALITY 349

ACKNOWLEDGMENTS

We thank Gary Harrington for his contribution to data collection and data base
management; Greg Meyer and Mark Blais for advice in statistical analysis; and the
Erik H. Erikson Institute at the Austen Riggs Center for financial, technical, and
conceptual support.

REFERENCES

Affra, S. (1982). Predicting adolescent suicidal behavior and the order of Rorschach measurement.
Journal of Personality Assessment, 46, 563–568.
American Psychiatric Association. (1994). Diagnostic and statistical manual of mental disorders (4th
ed.). Washington, DC: Author.
Applebaum, S. A., & Colson, D. B. (1968). A re-examination of the color-shading Rorschach test re-
sponse. Journal of Projective Techniques and Personality Assessment, 32, 160–164.
Applebaum, S. A., & Holtzman, P. S. (1962). The color-shading response and suicide. Journal of Pro-
jective Techniques and Personality Assessment, 26, 155–161.
Arana, G. W., & Hyman, S. (1989). Biological contributions to suicide. In D. G. Jacobs & H. N. Brown
(Eds.), Suicide: Understanding and responding: Harvard medical school perspectives on suicide
(pp. 72–86). Madison, CT: International Universities Press.
Beck, A. T. (1986). Hopelessness as a predictor of eventual suicide. In J. J. Mann & M. Stanley (Eds.),
Psychobiology of suicidal behavior (pp. 90–96). New York: Academy of Sciences.
Beck, A. T., Steer, R. A., Kovacs, M., & Garrison, B. (1985). Hopelessness and suicide: A 10-year pro-
spective study of hospitalized patients with suicidal ideation. American Journal of Psychiatry, 142,
559–563.
Blatt, S. J., & Ritzler, B. A. (1974). Suicide and the representation of transparency and cross-sections on
the Rorschach. Journal of Consulting and Clinical Psychology, 42, 280–287.
Bongar, B. (1991). The suicidal patient: Clinical and legal standards of care. Washington, DC: Ameri-
can Psychological Association.
Buzan, R. D., & Weissberg, M. P. (1992). Suicide: Risk factors and prevention in medical practice. An-
nual Review of Medicine, 43, 37–46.
Cohen, J. (1977). Statistical power analysis for the behavioral sciences. New York: Academic.
Cohen, J., & Cohen, P. (1983). Applied multiple regression/correlation analysis for the behavioral sci-
ences. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
Davis, L. J., & Offord, K. P. (1997). Logistic regression. Journal of Personality Assessment, 68,
497–507.
Elliot, A. J., Pages, K. P., Russo, J., Wilson, L. G., & Roy-Byrne, P. P. (1996). A profile of medically se-
rious suicide attempts. Journal of Clinical Psychiatry, 57, 567–571.
Exner, J. E. (1993). The Rorschach: A comprehensive system: Vol. 1 (3rd ed.). New York: Wiley.
Exner, J. E., & Wiley, J. (1977). Some Rorschach data concerning suicide. Journal of Personality As-
sessment, 41, 339–348.
Farberow, N. L. (1974). Use of the Rorschach in predicting and understanding suicide. Journal of Per-
sonality Assessment, 38, 411–419.
Farberow, N. L., & Schneidman, E. S. (1961). The cry for help. New York: Mcgraw-Hill.
Fleisher, M. S. (1957). Differential Rorschach configurations of suicidal patients: A psychological study of
threatened, attempted, and successful suicides. Unpublished doctoral dissertation, Yeshiva University.
Frank, G. (1994). On the prediction of suicide from the Rorschach. Psychological Assessment, 74,
787–794.
350 FOWLER, PIERS, HILSENROTH, HOLDWICK, PADAWER

Goldstein, R. B., Black, D. W., Narsallah, A., & Winokur, G. (1991). The prediction of suicide. Archives
of General Psychiatry, 48, 418–422.
Hansell, A. G., Lerner, H. D., Milden, R. S., & Ludolph, P. S. (1988). Single-sign Rorschach indicators:
A validity study using a depressed inpatient population. Journal of Personality Assessment, 52,
658–669.
Hertz, M. R. (1948). Suicide configurations in Rorschach records. Journal of Projective Techniques and
Personality Assessment, 12, 3–58.
Kessel, J. B., & Zimmerman, M. (1993). Reporting errors in studies of diagnostic performance of self-
administered questionnaires: Extent of the problem, recommendations for standard presentation of
results, and implications for the peer review process. Psychological Assessment, 5, 395–399.
Linehan, M. M., Goodstein, J. L., Nielsen, S. L., & Chiles, J. K. (1983). Reasons for staying alive when
you are thinking of killing yourself: The reasons for living inventory. Journal of Consulting and
Clinical Psychology, 51, 276–286.
Maltsberger, J. T. (1986). Suicide risk: The formulation of clinical judgment. New York: New York Uni-
versity Press.
Maris, R. W. (1981). Pathways to suicide: A survey of self-destructive behaviors. Baltimore: John
Hopkins University Press.
McIntosh, J. L. (1996). 1994 official USA statistics overhead/slide set. Washington, DC: American As-
sociation of Suicidology.
Meehl, P. E., & Rosen, A. (1955). Antecedent probability and the efficiency of psychometric signs, pat-
terns, or cutting scores. Psychological Bulletin, 52, 194–216.
Nunnally, J. C., & Burnstein, I. H. (1994). Psychometric theory. New York: McGraw-Hill.
Oldham, J. M., Skodol, A. E., Kellman, H. D., Hyler, S. E., Doidge, N., Rosnick, L., & Gallaher, P. E.
(1995). Co-morbidity of Axis I and Axis II disorders. American Journal of Psychiatry, 152,
571–578.
Osman, A., Kopper, B. A., Linehan, M. L., Barrios, F. X., Guiterrez, P. M., & Bagge, C. L. (1999). Vali-
dation of the adult suicidal ideation questionnaire and the reasons for living inventory in an adult
psychiatric inpatient sample. Psychological Assessment, 11, 115–123.
Pilkonis, P., Heape, C., Ruddy, J., & Serrao, P. (1991). Validity in the diagnosis of personality disorders:
The use of the LEAD standard. Psychological Assessment, 3, 46–54.
Piotrowski, Z. (1950). A Rorschach compendium, enlarged and revised. Psychiatric Quarterly, 24,
543–595.
Rierdan, J., Lang, E., & Eddy, S. (1978). Suicide and transparency responses on the Rorschach: A repli-
cation. Journal of Consulting and Clinical Psychology, 46, 1162–1163.
Sakheim, G. A. (1955). Suicidal responses on the Rorschach: A validation study. Journal of Nervous
and Mental Disorders, 122, 322–344.
Sapolsky, A. (1963). An indicator of suicidal ideation on the Rorschach test. Journal of Projective Tech-
niques and Personality Assessment, 27, 332–335.
Schein, M. H. (1976). Obstacles in the education of psychiatric residents. Omega, 7, 78–82.
Silberg, J. L., & Armstrong, J. G. (1992). The Rorschach test for predicting suicide among depressed ad-
olescent inpatients. Journal of Personality Assessment, 59, 290–303.
Skodol, A., Rosnik, L., Kellman, D., Oldham, J., & Hyler, S. (1988). Validating structured DSM–III–R
personality disorder assessment with longitudinal data. American Journal of Psychiatry, 145,
1297–1299.
Smith, K., Conroy, R. W., & Ehler, B. D. (1991). Lethality of suicide attempt rating scale. In B. Bongar’s
(Ed.), The suicidal patient: Clinical and legal standards of care (pp. 257–284). Washington, DC:
American Psychological Association.
Spitzer, R. (1983). Psychiatric diagnosis: Are clinicians still necessary? Comprehensive Psychiatry, 24,
399–411.
 ASSESSING SUICIDE LETHALITY 351

Stone, M. H., Hurt, S. W., & Stone, D. K. (1987). The PI 500: Long-term follow-up of borderline inpa-
tients meeting DSM–III criteria. Vol. 1. Global outcome. Journal of Personality Disorders, 1,
291–298.
Suokas, J., & Lonnqvist, J. (1991). Selection of patients who attempted suicide for psychiatric consulta-
tion. Acta Psychiatr Scand, 83, 179–182.
Tabachnick, B. G., & Fidell, L. S. (1996). Using mutilvariate statistics (3rd ed.). New York: Harper
Collins.
Weiner, I. B. (1991). Editor’s note: Interscorer agreement in Rorschach research. Journal of Personality
Assessment, 56, 1.
Wood, J. M., Nezworski, M. T., & Stejskal, W. J. (1996). The comprehensive system for the Rorschach:
A critical examination. Psychological Science, 7, 3–10.

J. Christopher Fowler
The Austen Riggs Center
25 Main Street
Stockbridge, MA 01262

Received April 1, 2000

Revised June 6, 2000

View publication stats