Psychological Assessment 2012 American Psychological Association

2013, Vol. 25, No. 1, 194 203 1040-3590/13/$12.00 DOI: 10.1037/a0029966

A Taxometric Exploration of the Latent Structure of Hoarding

Kiara R. Timpano Joshua J. Broman-Fulks

University of Miami Appalachian State University

Heide Glaesmer and Cornelia Exner Winfried Rief

University of Leipzig University of Marburg

Bunmi O. Olatunji Meghan E. Keough

Vanderbilt University University of Washington
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

Christina J. Riccardi Elmar Brhler

This document is copyrighted by the American Psychological Association or one of its allied publishers.

Florida State University University of Leipzig

Sabine Wilhelm Norman B. Schmidt

Massachusetts General Hospital, Boston, Massachusetts, and Florida State University
Harvard Medical School

Despite controversy regarding the classification and diagnostic status of hoarding disorder, there remains
a paucity of research on the nosology of hoarding that is likely to inform the classification debate. The
present investigation examined the latent structure of hoarding in three, large independent samples. Data
for three well-validated measures of hoarding were subjected to taxometric procedures, including
MAXimum EIGenvalue, Mean Above Minus Below A Cut, and Latent-Mode factor. Two symptom
measures, one of which closely mirrors the proposed diagnostic criteria for hoarding disorder, and a
measure of hoarding beliefs were analyzed. Sample 1 (n 2,501) was representative of the general
German population, while Samples 2 (n 1,149) and 3 (n 500) consisted of unselected undergraduate
students. Findings across all three samples and taxometric procedures provided converging evidence that
hoarding is best conceptualized as a dimensional construct, present in varying degrees in all individuals.
Results have implications across research and treatment domains, particularly with respect to assessment
approaches, treatment response determination, and policy decisions. These findings underscore the need
for further investigations on the nosology of hoarding, to help validate this construct as we move forward
with respect to our research and treatment efforts, as well as the potential inclusion of hoarding disorder
in the Diagnostic and Statistical Manual of Mental Disorders (5th ed.; DSM5; American Psychiatric
Association, 2012).

Keywords: hoarding, taxometrics, dimensional latent structure

This article was published Online First September 17, 2012.
Kiara R. Timpano, Department of Psychology, University of Miami; Joshua J.
Broman-Fulks, Department of Psychology, Appalachian State University; Heide
Glaesmer, Abteilung fr Medizinische Psychologie und Medizinische Soziologie,
University of Leipzig, Leipzig, Germany; Cornelia Exner, Department of Clinical
Psychology, University of Leipzig; Winfried Rief, Department of Psychology,
University of Marburg, Marburg, Germany; Bunmi O. Olatunji, Department of
Psychology, Vanderbilt University; Meghan E. Keough, Department of Psychia-
try, University of Washington; Christina J. Riccardi, Department of Psychology,
Florida State University; Elmar Brhler, Abteilung fr Medizinische Psychologie
und Medizinische Soziologie, University of Leipzig; Sabine Wilhelm, Department
of Psychiatry, Massachusetts General Hospital, Boston, Massachusetts, and Har-
vard Medical School; Norman B. Schmidt, Department of Psychology, Florida
State University.
The authors have no conflicts of interest or financial disclosures, with
the exception of Sabine Wilhelm, who has received research support
from the National Institute of Mental Health (NIMH), the Obsessive-
Compulsive Foundation, and the Tourette Syndrome Association. She
has also received medications and placebos from Forest Laboratories
for an NIMH-funded study; she is a presenter for the Massachusetts
General Hospital Psychiatry Academy in educational programs, sup-
ported through independent medical education grants from pharmaceu-
tical companies; she has received royalties from Elsevier Publications,
Guilford Publications, New Harbinger Publications, and Oxford Uni-
versity Press; and she has received speakers honorarium from various
academic institutions.
We thank Carissa Orlando for her assistance in providing ratings of the
taxometric plots.
Correspondence concerning this article should be addressed to Kiara R.
Timpano, Department of Psychology, University of Miami, 5665 Ponce de
Leon Boulevard, Coral Gables, FL 33146. E-mail: kiaratimpano@
gmail.com

194
 LATENT STRUCTURE OF HOARDING
Hoarding disorder is defined as a persistent and extreme diffi-
culty with discarding possessions, which, combined with severely
cluttered living spaces and excessive acquiring, can lead to clinical
levels of distress and/or impairment (Frost & Hartl, 1996). Current
prevalence estimates range from 2 6%, indicating that it is a far
more common phenomenon than initially hypothesized (Iervolino
et al., 2009; Timpano, Exner, et al., 2011). Hoarding disorder has
additionally been linked with a chronic course (Tolin, Meunier,
Frost, & Steketee, 2010), high rates of comorbidity, and poor
physical health (Tolin, Frost, Steketee, Gray, & Fitch, 2008). The
serious ramifications associated with hoarding are not just limited
to the individual, as symptoms may affect family members and
communities (Tolin, Frost, Steketee, & Fitch, 2008). Despite the
development of a cognitive behavioral intervention (Steketee,
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Frost, Tolin, Rasmussen, & Brown, 2010), this syndrome is gen-
erally acknowledged as a challenge to treat. Considered collec-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
tively, these data underscore our understanding of hoarding as a
profound public health burden.
Hoarding symptoms have been traditionally associated with
obsessive-compulsive disorder (OCD) or included as a symptom
of obsessive compulsive personality disorder; however, there is a
growing consensus that this classification might not be empirically
warranted (Mataix-Cols et al., 2010). In light of clinical, genetic,
and neuropsychological research conducted over the last 10 years,
the Diagnostic and Statistical Manual of Mental Disorders (5th
ed.; DSM5; American Psychiatric Association, 2012) Obsessive-
Compulsive Spectrum Sub-Workgroup has put forth the recom-
mendation to identify hoarding as a discrete disorder within
DSM5 (American Psychiatric Association, 2012; Mataix-Cols, et
al., 2010; Pertusa et al., 2010). Nevertheless, many questions
regarding the nosology of hoarding remain. Finding answers to
these issues is of paramount importance, for as Cronbach and
Meehl (1955) have noted, classification provides the field with an
operationalization of a given phenomenon and, as such, represents
the foundation for theory. Therefore, effectively capturing hoard-
ing disorder within a scientifically informed nomenclature would
significantly contribute to our understanding of its phenomenol-
ogy, etiology, and treatment.
One relevant nosological consideration pertains to the latent
structure of hoarding symptoms. That is, do individuals with
hoarding disorder reflect a distinct group or category, or do clinical
hoarding symptoms reflect an extreme end of a larger spectrum of
saving and acquiring behaviors? The proposed DSM5 definition
of hoarding reflects a categorical approach (American Psychiatric
Association, 2012); however, the true nature and underlying struc-
ture of hoarding remains an empirical question to be tested. The
idea of a dimensional versus categorical structure of hoarding has
implications across research and treatment arenas. For example,
from a measurement perspective, a dimensional view calls for the
assessment of hoarding symptoms across a continuum, whereas a
categorical view necessitates instruments that accurately classify
individuals as category members or nonmembers. In the same
vein, should hoarding be determined to have a continuous struc-
ture, then the diagnostic thresholds currently being discussed in
relation to DSM5 will in reality be relatively arbitrary. The latent
structure of hoarding can also inform etiological investigations and
treatment approaches. As Haslam (1997) and others (A. M. Rus-
cio, Borkovec, & Ruscio, 2001) have discussed, dimensional phe-
nomenon may be associated with different etiological models than
195
taxonic phenomenon, the former reflecting a combination of mul-
tiple, possibly graded factors, whereas the latter may be attribut-
able to a single gene or discrete environmental effect.
Examining the extant research on hoarding disorder, it remains
unclear whether the syndrome would be best classified as a distinct
category or should be conceptualized as a dimensional phenome-
non. Considering the degree of impairment associated with hoard-
ing disorder, along with the almost fantastical levels of clutter
noted in some cases, it is plausible to conclude that clinical
hoarding represents a unique or distinct phenomenon. That is, one
hypothesis may be that clinical hoarding is qualitatively different
from the collecting, saving, and acquiring behaviors seen in the
nonclinical population (Nordsletten & Mataix-Cols, 2012). Pro-
viding support for this hypothesis are neuropsychological findings
that outline specific deficits in executive functioning among indi-
viduals with clinical hoarding (Tolin, Villavicencio, Umbach, &
Kurtz, 2011). Individuals with hoarding also often endorse ex-
tremely unusual, maladaptive beliefs about, and strong emotional
attachment toward their possessions (Gilliam & Tolin, 2010).
Finally, from a clinical perspective, one study found that treatment
providers easily distinguished clinical hoarding from normal sav-
ing behaviors in children (Plimpton, Frost, Abbey, & Dorer, 2009).
In contrast, there are also a number of findings that support a
dimensional perspective. These data present the alternative, yet
viable, hypothesis that individuals with hoarding may differ from
nonclinical controls as a matter of degree, not type. Relevant
to this is the fact that all individuals have at some point in their life
saved and acquired objects. Some have suggested that hoarding
may even represent an adaptive mechanism that emerged through
evolutionary processes (Leckman & Mayes, 1998). Second, factor
analytic studies of hoarding symptoms, as well as associations
between hoarding symptoms and various mood and anxiety con-
structs, indicate comparable solutions across clinical and nonclini-
cal samples (Coles, Frost, Heimberg, & Steketee, 2003; Frost,
Steketee, & Grisham, 2004). Proposed risk factors, such as stress-
ful or traumatic life events, are also associated with hoarding
symptoms in a similar manner, regardless of the type of sample
investigated (Timpano, Keough, Traeger, & Schmidt, 2011; Tolin,
Meunier, et al., 2010). Similarly, a series of symptom provocation
imaging studies demonstrated that the same brain regionspre-
dominantly frontal limbic regionsare associated with hoarding
symptoms in both healthy controls and individuals with clinical
hoarding (Mataix-Cols et al., 2003, 2004).
Taxometrics is a series of computational procedures and epis-
temological principles designed to discern the latent structure of
phenomena (Meehl, 1995; Schmidt, Kotov, & Joiner, 2004). Spe-
cifically, taxometrics evaluates whether the relations among spec-
ified indicators of a proposed category or taxon fit a categorical
or a continuous solution. To date only one taxometric investigation
is relevant to hoarding. Olatunji, Williams, Haslam, Abramowitz,
and Tolin (2008) examined the latent structure of obsessive com-
pulsive (OC) symptoms in a large sample of students, taking into
account six different symptom types (hoarding, washing, checking,
obsessing, neutralizing, and ordering) and three beliefs associated
with OCD. The findings supported a dimensional model of the
latent structure for all symptom types, with the exception of
hoarding, which showed some evidence of taxonicity. A signifi-
cant limitation of this investigation with respect to hoarding is that
it relied on the three hoarding items of the Obsessive-Compulsive
 196 TIMPANO ET AL.
InventoryRevised (OCIR; Foa et al., 2002) to assess hoarding
behaviors. While the OCIR represents a valid assessment instru-
ment for OCD, it does not adequately capture hoarding. As dis-
cussed above, there is a growing understanding that hoarding
symptoms should not necessarily be considered OCD, and mea-
sures designed to assess OCD may not effectively reflect the core
components of hoarding: clutter, difficulty discarding due to strong
urges to save and/or distress, and acquisition. Although the OCIR
includes items that generally address clutter, acquisition, and sav-
ing behaviors, this measure has not been validated for use with
hoarding samples (i.e., hoarding populations considered separately
from clinical OCD samples). Also, from a methodological per-
spective, the OCIR only includes three hoarding items, which give
rise to potential interpretative problems for taxometric analyses in
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
which indicators have limited ranges.
The aims of the current investigation were to examine the latent
This document is copyrighted by the American Psychological Association or one of its allied publishers.
structure of hoarding symptoms using a series of taxometric pro-
cedures. We sought to extend previous research by using three
separate and well established measures of hoarding to capture this
multifaceted construct. In addition to considering two separate
symptom measures of hoarding, one of which closely mirrors the
proposed diagnostic criteria for hoarding disorder, we also ana-
lyzed a measure of hoarding beliefs. Findings from clinical re-
search studies indicate that the average age of hoarding patients is
around 50 years; however, age of onset research suggests that
hoarding symptoms emerge prior to the age of 20 (Tolin, Meunier,
et al., 2010). Analyses were therefore conducted using three large,
independent samples, including a community population with a
large age range and two early adulthood student samples. This
sampling procedure was deemed suitable given the estimated
population prevalence of hoarding (Timpano, Exner, et al., 2011).
Considering the suggestive evidence from factor analytic studies
and the similarity of findings when clinical and nonclinical sam-
ples are compared (Coles et al., 2003; Frost et al., 2004), we
hypothesized that the taxometric procedures would yield a contin-
uous solution of hoarding across samples.
Method
Participants
The present investigation relied on three large, independent
samples, which are characterized in Table 1. Sample 1 was repre-
sentative of the general German population, while Samples 2 and
3 consisted of unselected undergraduate students.
Table 1
Descriptive Statistics for the Three Samples
Type of M age Age range % Range of total
Sample N sample (years) (years) Female
Sample 1 2,501 Community 48.8 1494 51.4
Sample 2 1,149 Student 19.5 1833 68.3
Sample 3 500 Student 19.1 1830 66.3
Note. HRS Hoarding Rating Scale (Tolin, Frost, & Steketee, 2010); SIR Saving InventoryRevised (Frost et al., 2004); SCI Saving Cognitions
Inventory (Steketee, Frost, & Kyrios, 2003).
Measures
Hoarding Rating Scale (HRS; Tolin, Frost, & Steketee,
2010). The HRS is a five-item measure of the key hoarding
dimensions. Participants are required to answer items using a
9-point Likert-type scale, ranging from 0 (none) to 8 (extreme).
Items ask about (a) clutter, (b) difficulty with discarding posses-
sions, (c) difficulty with excessive acquisition, (d) emotional dis-
tress, and (e) functional impairment. HRS scores have been shown
to have good measurement characteristics, including evidence for
testretest reliability, internal consistency ( .97.87), conver-
gent and divergent validity, as well as construct validity in both
hoarding and nonclinical samples (Tolin, Frost, & Steketee, 2010).
The present investigation relied on the German-HRS (Timpano,
Exner, et al., 2011).
Saving InventoryRevised (SIR; Frost et al., 2004). The
SIR is a 23-item self-report questionnaire assessing a range of
hoarding behaviors in both clinical and nonclinical samples. Par-
ticipants are required to answer items using a 5-point Likert-type
scale from 0 (none) to 4 (almost all/extreme/very often). The
measure includes three factor analytically derived subscales, in-
cluding acquisition (seven items), clutter (nine items), and diffi-
culty discarding (seven items). SIR scores have been shown to
have good psychometric indices, including strong internal consis-
tency coefficients ( .97.87), as well as convergent and diver-
gent validity in both clinical (Frost et al., 2004) and student
samples (Coles et al., 2003).
Saving Cognitions Inventory (SCI; Steketee, Frost, &
Kyrios, 2003). The SCI reflects a 24-item measure examining
attitudes and beliefs associated with hoarding symptoms. The scale
includes four subscales: emotional attachment (10 items), memory
(five items), control (three items), and responsibility (six items).
Items are answered on a 7-point Likert-type scale from 1 (not at
all) to 7 (very much). SCI scores have been found to have good
internal consistency, as well as good convergent and discriminant
validity (Steketee et al., 2003). It has been used with both clinical
(Muroff et al., 2009) and nonclinical (Luchian, McNally, &
Hooley, 2007) samples.
Procedure
Sample 1 was collected with the assistance of an independent
demographic consulting company (USUMA, Berlin, Germany), as
part of a large cross-sectional survey of a representative subgroup
of the general German population. A detailed description of pro-
cedures and sampling strategies can be found in a previously
Indicator Indicator correlations
validity within:
Scale score M SD Taxon Complement
HRS 034 2.70 .67 .09 .37
SIR 081 2.41 .18 .14 .29
SCI 24148 2.04 .48 .23 .29
 LATENT STRUCTURE OF HOARDING 197

published report (Timpano, Exner, et al., 2011). Briefly, all sub-
jects were personally visited by a trained study assistant, informed
about the study procedures, and asked to sign an informed consent
document prior to completing a battery of questionnaires (includ-
ing the HRS).
With respect to the student samples, participants completed
either the SIR (Sample 2) or the SCI (Sample 3) in exchange for
partial course credit. Informed consent (approved by the Institu-
tional Review Board) was obtained prior to completing a battery of
self-report questionnaires, and all studies were run in accordance
with American Psychological Association standard ethical guide-
lines.
Indicator Selection
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Data Analytic Strategy
The latent structure of hoarding was investigated by submit-
ting the hoarding indicators identified in each sample to three
taxometric proceduresMAXimum EIGenvalue (MAXEIG),
Mean Above Minus Below A Cut (MAMBAC), and Latent-
Mode factor analysis (L-Mode). Simulated taxonic and dimen-
sional data plots were generated using Monte Carlo data that
matched the unique distributional characteristics (e.g., skew,
kurtosis, N) of the research data to assist in the interpretation of
results. To derive parameter estimates and generate categorical
comparison data, cases were assigned to groups using an iter-
ative method wherein analyses were initially run to obtain the
mean base rate. Analyses were then repeated a second time
using the mean base rate to classify cases into conjectured taxon

Composite indicators were derived based on the theoretical and complement groups. The relative fit of the research data
This document is copyrighted by the American Psychological Association or one of its allied publishers.

model of hoarding and prior factor analytic studies (Frost & Hartl,
1996; Frost et al., 2004). The descriptive characteristics and inter-
nal consistency levels for each measure and sample are presented
in Table 2, along with interindicator correlations. The five HRS
items map onto several of the proposed DSM5 hoarding criteria,
including Criterion A (difficulty discarding personal possessions),
Criterion C (clutter), Criterion D (distress and/or impairment), and
the acquiring specifier (Mataix-Cols et al., 2010; Tolin, Frost, &
Steketee, 2010). Each item represented its own indicator. The SIR
has been found to assess three behavioral facets of hoarding,
including clutter, difficulty discarding, and acquiring. Items load-
ing on each factor were summed to create three indicators. The SCI
reflects four different cognitive aspects related to hoarding, includ-
ing emotional attachment to possessions, concerns about memory,
control over possessions, and responsibility toward possessions.
Similar to our approach with the SIR, items loading on each SCI
factor were summed to create four indicators reflecting hoarding-
relevant cognitions.
plots to the simulated taxonic and dimensional curves was
assessed visually by two independent raters (96% agreement)
and using the Comparison Curve Fit Index (CCFI; J. Ruscio,
Ruscio, & Meron, 2007). The algorithms for the taxometric
procedures were obtained from J. Ruscio (2011).
MAXEIG. MAXEIG (Waller & Meehl, 1998) is a multivar-
iate procedure that calculates and plots eigenvalues from all
indicators across successive intervals of an input indicator. The
MAXEIG procedure offers a supplemental Inchworm Consistency
Test (ICT), which repeats MAXEIG analyses on the research data
with an increasing number of overlapping windows. The ICT
improves the interpretability of MAXEIG plots and is particularly
beneficial for elucidating the presence of a low base-rate taxon
(Waller & Meehl, 1998). The ICT was performed using 50, 100,
150, and 200 windows with .90 overlap.
MAMBAC. The MAMBAC (Meehl & Yonce, 1994) proce-
dure is based on the assumption that if two discrete groups exist
(i.e., taxon and nontaxon), mean differences between groups will

Table 2
Descriptive Statistics, Reliability, and Pearson Correlation Coefficient for the Indicators Derived From Each of the Three
Hoarding Measures

Indicator correlation
Indicator M SD Indicator range Alpha 1 2 3 4

Study 1: Hoarding Rating Scale

1. Clutter 1.41 1.91 08
2. Discarding 1.39 1.71 08 .53
3. Acquisition 0.88 1.38 08 .47 .56
4. Distress 0.87 1.36 08 .49 .57 .69
5. Impairment 0.88 1.38 08 .52 .58 .69 .79

Study 2: Saving InventoryRevised

1. Clutter 6.17 5.72 031 .91
2. Discarding 7.99 5.39 027 .88 .68
3. Acquisition 7.60 4.79 027 .79 .55 .55

Study 3: Saving Cognitions Inventory

1. Emotion 23.57 11.91 1070 .92
2. Control 11.64 4.41 321 .73 .52
3. Responsibility 15.11 6.82 638 .82 .73 .62
4. Memory 13.80 6.63 534 .83 .64 .49 .68
Note. All correlation coefficients significant at p .001.
 198 TIMPANO ET AL.
occur on valid indicators of group membership. MAMBAC anal-
yses in the present study were conducted using 100 evenly spaced
cuts beginning 25 cases from either extreme.
L-Mode. L-mode (Waller & Meehl, 1998) is a multivariate
taxometric procedure that conducts a factor analysis on the pro-
posed indicators, and plots the distribution of scores on the first
principal factor. Taxonic variables tend to produce a bimodal
distribution of factor scores. In contrast, dimensional variables
generate factor score plots that are unimodally distributed.
CCFI. The CCFI is an objective numerical gauge of whether
data plots more closely match simulated taxonic or simulated
dimensional plots. The CCFI procedure compares the root-mean-
square residual (RMSR) estimates of the research data with the
simulated taxonic and dimensional curves. The result is a CCFI
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
score for each of the procedures that ranges from 0 to 1, which can
be averaged across taxometric procedures to produce a mean CCFI
This document is copyrighted by the American Psychological Association or one of its allied publishers.
score for each sample. Scores 0.45 are interpreted as consistent
with a latent dimension and scores 0.55 are indicative of taxonic
latent structure. Caution is generally recommended in interpreting
CCFI scores between 0.45 and 0.55, as they do not provide strong
support for either model (J. Ruscio, Walters, Marcus, & Kaczetow,
2010). Several recent studies have provided strong support for the
utility of the CCFI in accurately interpreting taxometric output (J.
Ruscio, 2007; J. Ruscio & Kaczetow, 2009; J. Ruscio & Marcus,
2007; J. Ruscio, et al., 2007; J. Ruscio & Walters, 2009).
Results
HRS-Suitability Analyses
The five HRS indicator variables (M 2.70, SD .67) ex-
ceeded minimum validity criteria (1.25 SD; Meehl, 1995) and
demonstrated nuisance correlations (i.e., within conjectured group
correlations) within tolerable limits (taxon M .09, complement
M .37, full sample M .59). To ensure that the hoarding data
were capable of producing interpretable output, plots generated
from simulated taxonic and dimensional data with similar distri-
butional characteristics to the research data were created. As can
be seen in Figure 1, the MAXEIG, MAMBAC, and L-Mode
procedures passed the initial suitability test. Thus, the indicators
and taxometric procedures were deemed appropriate for analysis.
HRS-Results
A visual analysis of the five MAXEIG plots indicated that all
five were relatively flat or rising toward the right side of the plot
and failed to demonstrate the peaks that would be typical of
taxonic data. Overall, the ICT aided in interpretability of the plots,
with support for a dimensional solution increasing as the number
of overlapping windows increased. As can be seen in the top
section of Figure 1, the averaged MAXEIG curve more closely
resembled the simulated dimensional plots.
Similarly, the application of the MAMBAC procedure to the
five hoarding indicators generated 20 plots, 16 of which were
concave or rising toward the right side of the plot and lacking
typical peaks indicative of taxonicity. The remaining four plots
demonstrated peaks near the center of the plots, suggesting a
base-rate near .50. A comparison of the data plots with simulated
taxonic and dimensional plots revealed that vast majority of the
data plots were more consistent with the simulated dimensional
plots than the simulated taxonic plot (see the middle section of
Figure 1 for the averaged MAMBAC plot superimposed on sim-
ulated taxonic and dimensional plots). Thus, the MAMBAC pro-
cedure provided convergent evidence for a latent hoarding dimen-
sion.
L-Mode analysis of the HRS indicators generated a single factor
score density plot. As can be seen in lower section of Figure 1, the
L-mode distribution did not contain two clearly delineated peaks
and was relatively consistent with the simulated dimensional plot.
As noted above, the CCFI provides an index of whether the data
plots more closely resemble simulated taxonic or dimensional
plots. The mean CCFI score across MAXEIG, MAMBAC, and
L-Mode analyses was .36, providing additional objective evidence
that the HRS data plots were more comparable to the simulated
dimensional plots, and indicating that hoarding, as measured by
the HRS in a large representative population sample, is dimen-
sional at the latent level.
SIR-Suitability Analyses
The three SIR indicators (M 2.41, SD 0.18) met minimum
validity criteria and demonstrated low nuisance correlations (taxon
M .14, complement M .29, full sample M .62). As can be
seen in Figure 2, the MAXEIG, MAMBAC, and L-Mode proce-
dures passed the initial suitability test, with the simulated taxonic
and dimensional plots being of the expected shape and distinguish-
able from each other.
SIR-Results
A visual analysis of the three MAXEIG plots indicated that all
three were relatively flat and failed to demonstrate the peaks that
would be typical of taxonic data. The ICT aided in interpretability
of the plots, with support for a dimensional solution increasing as
the number of overlapping windows increased. Although one plot
demonstrated a potential small peak on the right side at 200
windows, this peak appeared to be spurious as it was not evident
at other levels of overlapping windows. As can be seen in top
section of Figure 2, the averaged MAXEIG curve more closely
resembled the simulated dimensional plots. Similarly, the
MAMBAC procedure generated six plots, all of which were con-
cave in shape and failed to evidence typical peaks indicative of
taxonicity. A comparison of the data plots with simulated taxonic
and dimensional plots revealed that the data plots were highly
consistent with the simulated dimensional plots (see the middle
section of Figure 2). Further, the factor score density plot created
by L-Mode did not evidence a clearly delineated second peak and
was consistent with the simulated dimensional plot. The mean
CCFI score for MAXEIG, MAMBAC, and L-Mode analyses of
the SIR data provided strong support for a dimensional structure
(CCFI .33). Thus, the results of SIR taxometric analyses pro-
vided additional evidence that hoarding is dimensional at the latent
level.
SCI-Suitability Analyses
A similar approach was used to examine the suitability of the
SCI data for taxometric analysis. The four SCI indicators met
 LATENT STRUCTURE OF HOARDING 199
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 1. Averaged MAXimum EIGenvalue (MAXEIG; Waller & Meehl, 1998; A), Mean Above Minus
Below A Cut (MAMBAC; Meehl & Yonce, 1994; B), and Latent-Mode factor analysis (L-Mode; Waller &
Meehl, 1998; C) curves for the Hoarding Rating Scale (HRS; Tolin, Frost, & Steketee, 2010) data (Sample 1)
superimposed on simulated taxonic (left) and dimensional curves (right).

minimum validity criteria (M 2.04, SD 0.48) and demon-
strated nuisance correlations within tolerable limits (taxon M
.23, complement M .29, full sample M .62). The simulated
taxonic and dimensional plots were clearly distinguishable and
passed the initial suitability test.
SCI-Results
MAXEIG analyses produced four plots, only one of which
appeared to demonstrate a peak that would be suggestive of a
taxon. The remaining three plots were relatively flat and con-
sistent with a latent dimension. Refer to the top section of
Figure 3 for the averaged MAXEIG data curve. Visual com-
parison of the data plots with simulated comparison plots re-
vealed that the data plots were highly consistent with the
simulated dimensional plots. MAMBAC analyses produced 12
plots, 10 of which lacked a distinct peak, were highly consistent
with the simulated dimensional plots, and were interpreted as
supporting a dimensional structure. The two remaining plots
 200 TIMPANO ET AL.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 2. Averaged MAXimum EIGenvalue (MAXEIG; Waller & Meehl, 1998; A), Mean Above Minus
Below A Cut (MAMBAC; Meehl & Yonce, 1994; B), and Latent-Mode factor analysis (L-Mode; Waller &
Meehl, 1998; C) curves for the Saving InventoryRevised (SIR; Frost et al., 2004) data (Sample 2) superimposed
on simulated taxonic (left) and dimensional curves (right).

were ambiguous, demonstrating slight rises toward the right Discussion

side of the graph. The data plots were generally consistent with
the simulated dimensional plots. The L-mode data plot demon-
strated a unimodal distribution and was consistent with the
simulated dimensional plot. The mean CCFI score for
MAXEIG, MAMBAC, and L-Mode analyses of the SCI data
provided strong support for a dimensional structure (CCFI
.36).
Answers regarding the nosology of hoarding will be of key impor-
tance in the coming years, as decisions are made about the classifi-
cation of hoarding in our diagnostic system (Mataix-Cols et al., 2010).
As it stands, there is a paucity of extant research on the latent structure
of hoarding, and as a result, different studies have used either a
dimensional or categorical definition without a clear empirical basis
 LATENT STRUCTURE OF HOARDING 201
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
This document is copyrighted by the American Psychological Association or one of its allied publishers.

Figure 3. Averaged MAXimum EIGenvalue (MAXEIG; Waller & Meehl, 1998; A), Mean Above Minus
Below A Cut (MAMBAC; Meehl & Yonce, 1994; B), and Latent-Mode factor analysis (L-Mode; Waller &
Meehl, 1998; C) curves for the Saving Cognitions Inventory (SCI; Steketee, Frost, & Kyrios, 2003) data (Sample
3) superimposed on simulated taxonic (left) and dimensional curves (right).

for doing so. In a similar fashion, the proposed DSM5 definition of
hoarding assumes a categorical latent structure. The present investi-
gation sought to address this issue by conducting a detailed taxometric
analysis of hoarding symptoms and associated beliefs.
Taxometrics do not rely upon traditional significance testing.
Rather, a notable strength of these procedures is that one can
evaluate the consistency of taxometric findings across a set of
different procedures (Schmidt et al., 2004). In our report, we
specifically integrated a series of three proceduresincluding
MAXEIG, MAMBAC, and L-Modeand considered the collec-
tive findings across three well validated measures of hoarding,
including the HRS, SIR, and SCI. Overwhelmingly, the results
were consistent with a dimensional solution. That is, differences in
hoarding behaviors and beliefs seem to be quantitative versus
qualitative. These results are in contrast to the report by Olatunji et
al. (2008), which examined the taxonicity of various subtypes of
OCD. However, it is likely that indicator selection (i.e., the OCIR
hoarding subscale) contributed to these differences in findings.
 202 TIMPANO ET AL.
There are a number of implications for the study and treatment
of hoarding that emerge from our findings supporting a dimen-
sional structure. Etiological models should be able to account for
a continuum of hoarding symptoms, ranging from nonclinical
levels to extreme representations with high levels of distress and
impairment. Etiological studies should consider additive or inter-
active effects between various risk factors. That is, a dimensional
latent structure implies that the phenomenon may be multiply
determined and would be best characterized by an additive or
interactive, graded etiology, rather than discrete, single environ-
mental or genetic effects. This finding is in line with a recent twin
study of hoarding, which found support for both genetic and
nonshared environmental factors in the etiology of hoarding
among females (Iervolino et al., 2009).
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
Assessment approaches to hoarding, from both a research and
clinical perspective, will need to be carefully considered. Measure-
This document is copyrighted by the American Psychological Association or one of its allied publishers.
ment lies at the crux of successful empirical and clinical endeav-
ors, and therefore the field as a whole needs to ensure that a given
phenomenon is being accurately captured. Our data recommend
against the use of arbitrary cut-points to identify individuals with
hoarding. A dichotomization of study samples would have direct
implications for construct validity and would also result in the loss
of statistical power for empirical investigations. Existing cutoff
scores certainly offer clinical utility; however, our data recom-
mend that they be used as a gauge for identifying significant and
clinically meaningful levels of hoarding, rather than as a means to
place individuals in discrete categories. In line with current shifts
within the literature (Widiger & Trull, 2007), dimensional classi-
fication approaches to hoarding that disperse scores broadly may
be considered. Our findings also have implications for treatment
response and relapse, such that treatment response should not be
understood as a discrete shift from hoarder to nonhoarder status.
Instead, our data support the notion that patients may move along
a continuum from clinically significant hoarding symptoms to
nonsignificant levels and that this movement may be gradual or
partial. Finally, these assessment considerations have implications
for public policies regarding hoarding, particularly with regard to
disability determination criteria. Classifying patients into discrete
disordered and nondisordered categories may be less helpful (or
accurate) in such instances, and future research will be required to
help determine how to best reconcile our understanding of the
dimensional nature of hoarding with current policies and criteria.
The present findings should be considered in light of several
limitations. First and foremost, it is possible that given the low
base rate of hoarding we may not have had sufficient representa-
tion to detect the putative taxon. Although each of the three
samples included individuals with hoarding symptoms that fell
within the clinical range, future research should consider using
diagnostic interviews in conjunction with taxometric analyses.
This would help confirm that the study sample does in fact contain
a sufficient number of clinically diagnosed hoarding patients,
which would in turn ensure that hoarding symptoms range from
mild to extreme. While purely clinical samples are often less
well suited to taxometric procedures, given the restricted range in
both symptom type and severity, the ideal sample for taxometric
analyses includes about an equal number of potential taxon and
complement members (Schmidt et al., 2004). The fact that we used
excellent indicators of hoarding in all three samples, along with the
fact that Sample 1 examined a large representative community
sample, ameliorates this limitation somewhat. Nevertheless, these
sampling issues remain an important consideration for future re-
search. A second limitation is our selection of indicators. It is
possible that alternative sets of indicators could identify disconti-
nuity or taxonicity. To this end, we selected two measures of
hoarding to capture different features of the current cognitive
behavioral conceptualization of hoarding (i.e., behaviors and be-
liefs); yet it is plausible that other facets of this phenomenon might
represent equally valuable indicators, which may, in turn, support
more taxonic findings. For example, once the neuropsychological
profile is better established for clinical hoarding, these measures
might represent interesting and informative indices for taxometric
analysis.
In conclusion, the current investigation found support for a
dimensional model of hoarding. This is the first taxometric study
on hoarding to rely on well-validated measures of different facets
of the phenomenon. In general, our findings underscore the need
for further investigations on the nosology of hoarding, to help
validate this construct as we move forward with respect to our
research and treatment efforts, as well as the potential inclusion of
hoarding disorder in DSM5.
References
American Psychiatric Association. (2012). Proposed draft revisions to
DSM disorders and criteria (DSM5 development). Retrieved from
http://www.dsm5.org
Coles, M. E., Frost, R. O., Heimberg, R. G., & Steketee, G. (2003).
Hoarding behaviors in a large college sample. Behaviour Research and
Therapy, 41, 179 194. doi:10.1016/S0005-7967(01)00136-X
Cronbach, L. J., & Meehl, P. E. (1955). Construct validity in psychological
tests. Psychological Bulletin, 52, 281302. doi:10.1037/h0040957
Foa, E. B., Huppert, J. D., Leiberg, S., Langner, R., Kichic, R., Hajcak, G.,
& Salkovskis, P. M. (2002). The Obsessive-Compulsive Inventory:
Development and validation of a short version. Psychological Assess-
ment, 14, 485 496. doi:10.1037/1040-3590.14.4.485
Frost, R. O., & Hartl, T. L. (1996). A cognitive-behavioral model of
compulsive hoarding. Behaviour Research and Therapy, 34, 341350.
doi:10.1016/0005-7967(95)00071-2
Frost, R. O., Steketee, G., & Grisham, J. R. (2004). Measurement of
compulsive hoarding: Saving InventoryRevised. Behaviour Research
and Therapy, 42, 11631182. doi:10.1016/j.brat.2003.07.006
Gilliam, C. M., & Tolin, D. F. (2010). Compulsive hoarding. Bulletin of the
Menninger Clinic, 74, 93121. doi:10.1521/bumc.2010.74.2.93
Haslam, N. (1997). Evidence that male sexual orientation is a matter of
degree. Journal of Personality and Social Psychology, 73, 862 870.
doi:10.1037/0022-3514.73.4.862
Iervolino, A. C., Perroud, N., Fullana, M. A., Guipponi, M., Cherkas, L.,
Collier, D. A., & Mataix-Cols, D. (2009). Prevalence and heritability of
compulsive hoarding: A twin study. American Journal of Psychiatry,
166, 1156 1161. doi:appi.ajp.2009.08121789
Leckman, J. F., & Mayes, L. C. (1998). Understanding developmental
psychopathology: How useful are evolutionary accounts? Journal of the
American Academy of Child & Adolescent Psychiatry, 37, 10111021.
doi:10.1097/00004583-199810000-00010
Luchian, S. A., McNally, R. J., & Hooley, J. M. (2007). Cognitive aspects
of nonclinical obsessive-compulsive hoarding. Behaviour Research and
Therapy, 45, 16571662. doi:10.1016/j.brat.2006.08.014
Mataix-Cols, D., Cullen, S., Lange, K., Zelaya, F., Andrew, C., Amaro, E.,
. . . Phillips, M. L. (2003). Neural correlates of anxiety associated with
obsessive-compulsive symptom dimensions in normal volunteers. Bio-
logical Psychiatry, 53, 482 493. doi:S0006322302015044
 LATENT STRUCTURE OF HOARDING
Mataix-Cols, D., Frost, R. O., Pertusa, A., Clark, L. A., Saxena, S.,
Leckman, J. F., . . . Wilhelm, S. (2010). Hoarding disorder: A new
diagnosis for DSMV? Depression and Anxiety, 27, 556 572. doi:
10.1002/da.20693
Mataix-Cols, D., Wooderson, S., Lawrence, N., Brammer, M. J., Speckens,
A., & Phillips, M. L. (2004). Distinct neural correlates of washing,
checking, and hoarding symptom dimensions in obsessive-compulsive
disorder. Archives of General Psychiatry, 61, 564 576. doi:10.1001/
archpsyc.61.6.564
Meehl, P. E. (1995). Bootstraps taxometrics: Solving the classification
problem in psychopathology. American Psychologist, 50, 266 275. doi:
10.1037/0003-066X.50.4.266
Meehl, P. E., & Yonce, L. J. (1994). Taxometric analysis I: Detecting
taxonicity with two quantitative indicators using means above and below
a sliding cut (MAMBAC procedure). Psychological Reports, 74, 1059
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.
1274.
Muroff, J., Steketee, G., Rasmussen, J., Gibson, A., Bratiotis, C., &
This document is copyrighted by the American Psychological Association or one of its allied publishers.
Sorrentino, C. (2009). Group cognitive and behavioral treatment for
compulsive hoarding: A preliminary trial. Depression and Anxiety, 26,
634 640. doi:10.1002/da.20591
Nordsletten, A. E., & Mataix-Cols, D. (2012). Hoarding versus collecting:
Where does pathology diverge from play? Clinical Psychology Review,
32, 165176. doi:10.1016/j.cpr.2011.12.003
Olatunji, B. O., Williams, B. J., Haslam, N., Abramowitz, J. S., & Tolin,
D. F. (2008). The latent structure of obsessive-compulsive symptoms: A
taxometric study. Depression and Anxiety, 25, 956 968. doi:10.1002/
da.20387
Pertusa, A., Frost, R. O., Fullana, M. A., Samuels, J. F., Steketee, G., Tolin,
D. F., . . . Mataix-Cols, D. (2010). Refining the diagnostic boundaries of
compulsive hoarding: A critical review. Clinical Psychology Review, 30,
371386. doi:S0272-7358(10)00020-6
Plimpton, E. H., Frost, R. O., Abbey, B. C., & Dorer, W. (2009). Com-
pulsive hoarding in children: 6 case studies. International Journal of
Cognitive Therapy, 2, 88 104. doi:10.1521/ijct.2009.2.1.88
Ruscio, A. M., Borkovec, T. D., & Ruscio, J. (2001). A taxometric
investigation of the latent structure of worry. Journal of Abnormal
Psychology, 110, 413 422. doi:10.1037/0021-843X.110.3.413
Ruscio, J. (2007). Taxometric analysis: An empirically grounded approach
to implementing the method. Criminal Justice and Behavior, 34, 1588
1622. doi:10.1177/0093854807307027
Ruscio, J. (2011). Documentation of program code to perform taxometric
analyses and simulate comparison data in the R language. Retrieved
from http://www.tcnj.edu/~ruscio/taxometrics.html
Ruscio, J., & Kaczetow, W. (2009). Differentiating categories and dimen-
sions: Evaluating the robustness of taxometric analyses. Multivariate
Behavioral Research, 44, 259 280. doi:10.1080/00273170902794248
Ruscio, J., & Marcus, D. K. (2007). Detecting small taxa using simulated
comparison data: A reanalysis of Beach, Amir, and Baus (2005) data.
Psychological Assessment, 19, 241246. doi:10.1037/1040-3590.19.2
.241
Ruscio, J., Ruscio, A. M., & Meron, M. (2007). Applying the bootstrap to
taxometric analysis: Generating empirical sampling distributions to help
203
interpret results. Multivariate Behavioral Research, 42, 349 386. doi:
10.1080/00273170701360795
Ruscio, J., & Walters, G. D. (2009). Using comparison data to differentiate
categorical and dimensional data by examining factor score distribu-
tions: Resolving the mode problem. Psychological Assessment, 21, 578
594. doi:10.1037/a0016558
Ruscio, J., Walters, G. D., Marcus, D. K., & Kaczetow, W. (2010).
Comparing the relative fit of categorical and dimensional latent variable
models using consistency tests. Psychological Assessment, 22, 521.
doi:10.1037/a0018259
Schmidt, N. B., Kotov, R., & Joiner, T. E. (2004). Taxometrics: Toward a
new diagnostic scheme for psychopathology. Washington, DC: Ameri-
can Psychological Association. doi:10.1037/10810-000
Steketee, G., Frost, R. O., & Kyrios, M. (2003). Cognitive aspects of
compulsive hoarding. Cognitive Therapy and Research, 27, 463 479.
doi:10.1023/A:1025428631552
Steketee, G., Frost, R. O., Tolin, D. F., Rasmussen, J., & Brown, T. A.
(2010). Waitlist-controlled trial of cognitive behavior therapy for hoard-
ing disorder. Depression and Anxiety, 27, 476 484. doi:10.1002/da
.20673
Timpano, K. R., Exner, C., Glaesmer, H., Rief, W., Keshaviah, A., Brahler,
E., & Wilhelm, S. (2011). The epidemiology of the proposed DSM5
hoarding disorder: Exploration of the acquisition specifier, associated
features, and distress. Journal of Clinical Psychiatry, 72, 780 786,
878 879. doi:10.4088/JCP.10m06380
Timpano, K. R., Keough, M. E., Traeger, L., & Schmidt, N. B. (2011).
General life stress and hoarding: Examining the role of emotional
tolerance. International Journal of Cognitive Therapy, 4, 263279.
doi:10.1521/ijct.2011.4.3.263
Tolin, D. F., Frost, R. O., & Steketee, G. (2010). A brief interview for
assessing compulsive hoarding: The Hoarding Rating Scale-Interview.
Psychiatry Research, 178, 147152. doi:10.1016/j.psychres.2009.05.001
Tolin, D. F., Frost, R. O., Steketee, G., & Fitch, K. E. (2008). Family
burden of compulsive hoarding: Results of an Internet survey. Behaviour
Research and Therapy, 46, 334 344. doi:S0005-7967(08)00002-8
Tolin, D. F., Frost, R. O., Steketee, G., Gray, K. D., & Fitch, K. E. (2008).
The economic and social burden of compulsive hoarding. Psychiatry
Research, 160, 200 211. doi:S0165-1781(07)00292-2
Tolin, D. F., Meunier, S. A., Frost, R. O., & Steketee, G. (2010). Course
of compulsive hoarding and its relationship to life events. Depression
and Anxiety, 27, 829 838. doi:10.1002/da.20684
Tolin, D. F., Villavicencio, A., Umbach, A., & Kurtz, M. M. (2011).
Neuropsychological functioning in hoarding disorder. Psychiatry Re-
search, 189, 413 418. doi:S0165-1781(11)00502-6
Waller, N. G., & Meehl, P. E. (1998). Multivariate taxometric procedures:
Distinguishing types from continua. Newberry Park, CA: Sage.
Widiger, T. A., & Trull, T. J. (2007). Plate tectonics in the classification of
personality disorder: Shifting to a dimensional model. American Psy-
chologist, 62, 71 83. doi:2007-01685-001
Received September 2, 2011
Revision received July 16, 2012
Accepted July 20, 2012