Taxometrics and developmental psychopathology 2003.pdf · 2003. 8. 3. · Taxometrics and developmental psychopathology 503 cally more interested in the processes through Western

Development and Psychopathology, 15 (2003), 501–527Copyright 2003 Cambridge University PressPrinted in the United States of AmericaDOI: 10.1017.S0954579403000270

Taxometrics anddevelopmental psychopathology

THEODORE P. BEAUCHAINEUniversity of Washington

AbstractDevelopmental psychopathologists have criticized categorical classification systems for their inability to account forwithin-group heterogeneity in biological, etiological, developmental, and cultural influences on behavior. Dichotomizingcontinuous scores of symptom severity is also inadvisable statistically. Perhaps because of a resulting wariness ofcategorizing, few explorations into the ontological status of traits or disorders as dimensional versus discrete havebeen conducted. It is argued here that the limitations of categorizing have little to do with the ontological status oftraits and that developmental psychopathologists should be concerned with identifying discrete behavioral syndromes.Common taxometric methods for resolving discrete traits are described, and questions of concern to developmentalpsychopathologists are outlined that can be addressed through taxometrics studies. These include (a) identifyingchildren who are at risk for future psychopathology, (b) identifying discrete subtypes within current diagnosticclasses, (c) locating sensitive periods in the development of discrete pathological traits, (d) discovering moderatorsof treatment outcome, and (e) elucidating mechanisms of equifinality and multifinality. Although most behavioraltraits probably are distributed continuously, identifying those that are discrete will advance the science of developmentalpsychopathology. Disorders for which taxometric analyses might be applied include anxiety, attention deficithyperactivity disorder, autism spectrum disorders, conduct problems, depression, and schizophrenia.

There are many reasons to prefer continuous lected functions (e.g., Wakefield, 1992, 1993,1997, 1999), or are they defined by somewhatmodels of psychopathology over categorical

models. Some of these are philosophical, where- arbitrary distinctions derived from social val-ues (e.g., Lilienfeld & Marino, 1995, 1999)?as others are methodological and pragmatic.

Philosophical arguments address one of the Should a set of symptoms be considered a dis-order when induced by a high risk environment,most fundamental yet elusive questions facing

behavioral scientists, namely, what constitutes or is evidence of independent internal mecha-nisms necessary (Wakefield, Pottick, & Kirk,a disorder? Do mental disorders reflect failures

of biological systems to perform naturally se- 2002)? Can environmental risk and internalmechanisms even be considered as separatecausal agents (e.g., Bremner & Vermetten,2001)? Could all of the 365 categories in theThis article was written in honor of Paul Everett Meehl,

who died in February 2003. May his passing reinvigorate Diagnostic and Statistical Manual of Mentalthe ideals of empirical rigor and scientific realism that he Disorders (DSM-IV; American Psychiatric As-so dutifully and tirelessly advanced. Rarely is a field so

sociation, 1994) possibly be distinct (e.g., Houts,indebted to the work of one individual. Preparation of this2002; Kendell, 1989)? Does the DSM-IV frame-article was supported in part by Grant MH63699 from the

National Institute of Mental Health. The author expresses work pathologize normal behavior (e.g., Rich-thanks to Geraldine Dawson, Liliana J. Lengua, and M. ters & Cicchetti, 1993)?Jamila Reid for their helpful contributions to this work. Each of these questions instantiates the dif-

Address correspondence and reprint requests to: The-ficulties faced in deciding which behaviors orodore P. Beauchaine, Department of Psychology, Univer-behavioral syndromes to define as disordered.sity of Washington, Box 351525, Seattle, WA 98195-

1525; E-mail: [email protected]. In fact, whether one endorses or rejects the

501

T. P. Beauchaine502

enterprise of diagnosis altogether may follow tion about children’s adjustment (see Cum-mings, Davies, & Campbell, 2000; Hinshawfrom his or her answers to these and similar

questions. Such answers are likely to depend & Park, 1999). Although space limitations pre-clude a full review of the limitations of cate-upon what philosophers of science have termed

essentialist versus nominalist views of human gorical diagnosis, criticisms include assertionsthat the DSM framework (a) places the locusbehavior (Flanagan & Blashfield, 2002). Es-

sentialists argue that mental disorders reflect of disorder within the individual, ignoring con-textual information about familial, neighbor-objective underlying causal realities that are

independent of human values. Wakefield’s (1992, hood, and other influences; (b) assumes bio-logical bases for psychiatric disorders despite1993, 1997, 1999) harmful dysfunction analy-

sis reflects an essentialist philosophy, as dis- claims of being descriptive and atheoretical;(c) is devoid of developmental guidelines forturbances in behavior are assumed to be caused

by biological dysfunctions that impact evolu- assessment and diagnosis; (d) fails to accountfor cultural differences in the expression oftionary fitness. Essentialist views of behavior,

particularly those that invoke evolutionary ac- maladaptation and distress; (e) is limited in itsclinical application because of symptom het-counts of adaptation and maladaptation, have

been criticized as being teleological because erogeneity within diagnostic categories and ho-mogeneity across diagnostic categories; and (f)the link between current adjustment and past

evolutionary fitness is often tenuous (e.g., Beau- assumes that behavioral syndromes are categor-ical entities with discrete etiologies (Cantwell,chaine, 1999; Richters & Hinshaw, 1999).

In contrast, nominalists argue that psychi- 1996; Clark, Watson, & Reynolds, 1995; Cum-mings et al., 2000; Hinshaw, Lahey, & Hart,atric disorders reflect deviations from socially

constructed prescriptions for behavior and that 1993; Jensen & Hoagwood, 1997; Sonuga–Barke, 1998; Sroufe, 1997). Finally, dichoto-there are no objective means of demarcating

normality from abnormality. This postmodern mizing continuous scores of symptom severity,a necessary consequence of categorizing, resultsphilosophy is exemplified in the writings of

Szasz (2000) and Lilienfeld and Marino (1995, in reduced reliability and statistical power andmay produce misleading outcomes in research1999), who suggest that most psychiatric dis-

orders are characterized by unclear boundaries assessing the precursors, correlates, and se-quelae of psychopathology (see MacCallum,and a lack of defining features. As such, nom-

inalists are likely to interpret behaviors as fall- Zhang, Preacher, & Rucker, 2002).For these reasons, many developmental psy-ing along a continuum of social acceptability,

to consider diagnostic cutoffs as arbitrary, and chopathologists prefer to use empirically derivedinstruments rather than categorical diagnoses into be wary of diagnosis altogether. Nominalist

philosophies are sometimes preferred over es- assessment. Such instruments, of which the ChildBehavior Checklist (Achenbach, 1991) is ansentialist philosophies because they carry no

assumptions about unobservable phylogenic example, evaluate behavioral traits across mul-tiple continuous dimensions. Symptoms are as-mechanisms of behavior. This preference may

be prudent when clear links between symp- sessed using factor-analytically derived sub-scales, providing for comparisons of children’stoms and biological dysfunction are lacking.

Although some developmental psychopa- scores to age- and gender-matched norms. Em-pirical assessment instruments carry no assump-thologists have weighed in on this philosophi-

cal debate, most discussions within the field tions about biological substrates, environmen-tal influences, or etiological origins of a givenregarding the merits of dimensional versus dis-

crete models of psychopathology have occurred behavioral profile. Moreover, symptom over-lap among subscales is considered to be clini-at an entirely different level. These discussions

have focused on the methodological and prac- cally relevant information rather than nuisancecontamination across diagnostic categories (seetical constraints imposed by a categorical di-

agnostic system and the utility of dimensional Cummings et al., 2000).These characteristics are appealing becauseassessment approaches toward capturing im-

portant developmental and contextual informa- developmental psychopathologists are typi-

Taxometrics and developmental psychopathology 503

cally more interested in the processes through Western intellectual tradition is replete withinstances in which the establishment of rigidwhich maladaptation emerges than in the de-

scriptive aspects of behavior (Cicchetti, 1993; and artificial boundaries has resulted in ste-reotypes, prejudiced policies, and impedimentsRutter & Sroufe, 2000; Sroufe & Rutter, 1984).1

Equifinality, for example, suggests that a to scientific progress (Waters & Beauchaine,2003). Yet we should not confuse preferencesgiven disorder can be the end state of numer-

ous developmental pathways, and multifinal- for continua that are based on philosophical,methodological, or pragmatic concerns withity suggests that children in similar high risk

situations can diverge toward quite disparate the ontological status of behavioral traits.Whether a particular trait or disorder representsend states, only some of which will be disor-

dered (Cicchetti & Rogosch, 1996). Although a discrete entity is an empirical question thatcannot be settled through methodological con-both categorical and empirically derived in-

struments are descriptive, empirical assessments vention or philosophical debate (see Meehl,1992, 1995; Sonuga–Barke, 1998). Based on aare much more flexible for examining diverse

developmental trajectories. This is because lon- growing number of formal taxometrics inves-tigations, evidence from the adult psychopath-gitudinal evaluations of behavioral functioning

that are indexed to age-matched norms provide ology literature suggests that at least sometraits and disorders are distributed as discreteprofiles of emerging and diminishing symp-

toms over time. Such profiles are invaluable classes, including endogenous depression(Ambrosini, Bennett, Cleland, & Haslam,toward conceptualizing psychopathology (Cic-

chetti & Rogosch, 1996; Kagan, 1997) and in- 2002; Beach & Amir, 2003; Haslam & Beck,1994), schizotypy (Blanchard, Gangestad,clude developmental information that cannot

be extracted from categorical classifications. Brown, & Horan, 2000; Golden & Meehl,1979; Korfine & Lenzenweger, 1995; Lenzen-Indeed, continuous scales provide for the as-

sessment and tracking of symptoms that may weger, 1999; Lenzenweger & Korfine, 1992;Tyrka, Cannon, Haslam, Mednick, Schulsinger,be subthreshold for a given diagnostic cate-

gory (Hinshaw et al., 1993). Schulsinger, Parnas, 1995; Tyrka, Haslam, &Cannon, 1995), dissociative experiences (Wal-ler, Putnam, & Carlson, 1996; Waller & Ross,

Evaluating the Ontological Status1997), psychopathy (Harris, Rice, & Quinsey,

of Traits1994), and Type A behavior patterns (Strube,1989). In contrast, very few taxometrics inves-Clear articulation by developmental psycho-

pathologists of the limitations of categorical tigations have appeared in the child psycho-pathology or developmental psychopathologydiagnostic systems represents a major contri-

bution to the study of human behavior. The literatures (for exceptions see Erlenmeyer–Kimling, Golden, & Cornblatt, 1989; Fra-ley & Spieker, 2003; Skilling, Quinsey, &

1. As an interesting aside, psychology is not the only dis- Craig, 2001; Woodward, Lenzenweger, Kagan,cipline to wrestle with the importance of description

Snidman, & Arcus, 2000). It is therefore un-versus process as topics of inquiry. Several months be-clear when in development these discrete be-fore his untimely death in May 2002, the renowned

paleontologist Stephen Jay Gould delivered a series of havioral traits emerge. Perhaps this lack ofpublic lectures on evolutionary theory. At one of these taxometrics research within developmentallectures Gould fielded the question, “Which is your fa- psychopathology reflects the aforementionedvorite dinosaur?” Gould’s response was incisive: “I don’t

preference for continuous models of behaviorallike any of them,” he stated. In elaborating, Gould notedfunctioning and an associated wariness of cat-that a preoccupation with the descriptive features of

dinosaurs has distracted many paleontologists from ask- egorical diagnostic systems. Yet there are aing questions that are of considerably more scientific number of reasons why inquiries into the onto-value. Gould then explained that as much importance logical status of behavioral traits and disordershas been traditionally placed on description as on the

should be of central interest to developmentalmechanisms of evolutionary change that produce phe-psychopathologists. Moreover, the limitationsnotypic variability, even though questions concerning

the latter are of far greater significance. of categorizing have very little to do with the

T. P. Beauchaine504

ontological status of a trait or disorder be- may be unsatisfying for those with predomi-nantly applied interests, who might still ask,cause individual differences in symptoms are

nearly always observed and merit scientific “so what?” Are there any reasons to suspectthat prevention or intervention strategies mightscrutiny, regardless of whether the trait is dis-

crete or continuous. Thus, our choice of cate- differ based on knowing that a high risk traitis distributed discretely? Meehl (1992, 1995)gorical versus empirical assessment does not

depend on the outcome of taxometrics investi- argued that this is precisely the case and thattherapeutic strategies might differ substan-gations. In the sections to follow, these points

will be elaborated. Moreover, it will be sug- tially based on knowing if a person belongsto a high risk taxon group. Consider the diath-gested that theory-driven taxometrics research

holds the potential to address long-standing yet esis for schizophrenia. As noted above, sev-eral authors have confirmed that schizotypy,unanswered questions of unique concern to

developmental psychopathologists and can or a constellation of observable symptomsthat appears to mark a genetic liability forenrich our understanding of emerging disor-

ders in a way that cannot be achieved by con- schizophrenia spectrum disorders, is distrib-uted as a discrete class, with a base rate some-ducting similar studies with adults.where around 5% (Blanchard et al., 2000;Golden & Meehl, 1979; Korfine & Lenzen-

Why Look for Typologies?weger, 1995; Lenzenweger, 1999; Lenzenweg-er & Korfine, 1992; Tyrka, Cannon, et al.,Before continuing, it may be useful to address

a question that often emerges in discussions 1995). Note that this base rate is considerablyhigher than the 1.1% estimate of the prevalenceabout taxometrics research, namely, “who

cares?” What does it matter whether a trait or of schizophrenia in the general population(Regier, Narrow, Rae, Manderscheid, Locke,a disorder reflects a distinction in kind or a

difference in degree? The simplest answer to & Goodwin, 1993). Of course, this is expectedfor a genetic trait that is not fully penetrant.this question is that identifying taxa enables

us to establish nonarbitrary cutoffs that distin- In other words, the genetic liability may be anecessary but insufficient condition for devel-guish between those with and without a trait

or disorder (Beauchaine & Waters, 2003; oping schizophrenia. Indeed, although the out-come and course of childhood-onset schizo-Meehl, 1995). In Plato’s words, knowing that

a trait is distributed discretely allows us to phrenia has been understudied (see Dulmus &Smyth, 2000), the expression of the disorder“carve nature at its joints.” Moreover, evi-

dence of taxonicity offers strong support for is clearly sensitive to contextual influences.Expressed emotion is observed at high levelsthe construct validity of a trait or diagnostic

entity, particularly when variables from multi- in families of children with schizophrenia spec-trum disorders (Hamilton, Asarnow, & Tomp-ple levels of analysis (e.g., physiological, be-

havior–observational, self-report) are used as son, 1999) and exerts a strong influence onboth course and prognosis (Falloon, Boyd, Mc-markers of the putative taxon (Beauchaine &

Beauchaine, 2002; Meehl, 1995). Note that Gill, Williamson, Razani, Moss, Gilderman, &Simpson, 1985; Hogarty, Anderson, Teiss,this also addresses the philosophical question

of whether a trait or diagnostic category marks Kornblith, Greenwald, Ulrich, & Carter, 1991).The implication is that premorbid identifica-a discrete entity. Such questions are much

more appropriate at the level of disorder than tion of children with a genetic diathesis forschizophrenia might facilitate targeted familyat the level of a diagnostic system (see Flana-

gan & Blashfield, 2002). Arguments against interventions that improve outcome and courseand perhaps delay or prevent the onset of thethe validity of the DSM are often applied to

the entire classification system, which cannot disorder (see Cornblatt, Obuchowski, Roberts,Pollack, & Erlenmeyer–Kimling, 1999). Tax-possibly be valid or invalid in a binary sense.

Rather, some diagnostic categories are of higher ometric analyses of cognitive and neuromotorperformance variables suggest that schizotypyconstruct validity than others.

Yet “merely” carving nature at its joints can be identified in children and adolescents,


with a base rate similar to that found in adults(Erlenmeyer–Kimling et al., 1989). Thus, bothapplied and basic interests are served by iden-tifying taxonic traits and by distinguishing be-tween taxon and nontaxon group members.More specific advantages of taxometrics in-vestigations will be presented in later sections.

Traditional Approaches to Searchingfor Boundaries

Given that identifying taxa is of both basic Figure 1. An admixture of two discrete normaland applied import, how do we go about do- distributions of n = 1000, each indicated by

hatched bars and dashed lines. The effect size sepa-ing it? Before addressing this question, tworating the distributions is 2 standard deviations.issues must be considered. First, what pre-Note that despite this large effect, there is no evi-cisely is meant by the term taxon, and second,dence of bimodality in the combined distribution,

what are common misconceptions about the indicated by filled solid bars. Note also that therequirements of testing taxonic hypotheses? combined distribution is near normal, as indicated

by the solid curve.The term taxon has multiple uses acrossscientific disciplines, and it is therefore diffi-cult to arrive at a universal definition (Meehl, admixed distribution for evidence of bimodal-

ity (Kendell, 1989). However, two discrete dis-1999). In the strongest sense, taxa represententities that are clearly different in kind by tributions often appear to be unimodal when

mixed, even at quite large effect sizes (Gray-some functional criterion, such as separatespecies as defined by reproductive isolation son, 1987; Murphy, 1964; Waller & Meehl,

1998). This is illustrated in Figure 1, where(Mayr, 1942). As is often the case in the be-havioral sciences, however, the term has been two distributions with a mean separation of 2

standard deviations are mixed. Inspection ofused more loosely to refer to almost anymethod of classifying, be it categorical or em- the admixed distribution reveals no evidence

of bimodality. Indeed, the combined distribu-pirical (Achenbach, 1993). In taxometrics re-search, a taxon is typically defined by a bound- tion is near normal with a kurtosis value (−.52)

that is acceptable for almost all statistical testsary that separates taxon group members fromnontaxon group members, that is, from evi- that assume normality. It is worth emphasiz-

ing the magnitude of the effect represented indence that the two groups emerge from sepa-rate or discrete distributions. However, be- Figure 1. Recall that Cohen (1988) defined a

large effect as .8 standard deviation units ofcause distinct distributions are likely to overlapconsiderably, given both within-groups vari- separation between means (Cohen’s d). Thus,

the effect size illustrated in Figure 1 is 2.5ability and measurement error, traditional sta-tistical methods are of little use in detecting times larger than Cohen’s definition of a large

effect. Clearly, bimodality is a very weak cri-taxa. Rather, formal taxometric methods arerequired. Unfortunately, a number of ineffec- terion for inferring taxonicity.

Figure 1 also illustrates the fallibility of thetive strategies for identifying taxa continue toappear in the psychology and psychiatry liter- commonly held belief that discrete traits and

disorders are marked by clear and distinct bound-atures. Some consideration of these strategiesis therefore warranted. aries (see also Lilienfeld & Marino, 1995; Meehl,

1995; Sonuga–Barke, 1998). There are severalreasons why this is unlikely to be the case.

BimodalityFirst, the mechanisms responsible for placingindividuals into taxon and nontaxon groupsEarly efforts to differentiate types from con-

tinua involved plotting the univariate distribu- are almost always latent (unobservable) con-structs that are assessed by imperfect manifesttions of traits or symptoms and inspecting the

T. P. Beauchaine506

indicators. Even genetic disorders of high pene- several alternative methods aimed at testingtaxonic conjectures have been developed. Un-trance are not manifested equivalently across

individuals. Huntington’s chorea, for exam- doubtedly the most popular of these in thepsychological literature is the set of over 300ple, is a progressive and degenerative disorder

of the nervous system that results in mood la- algorithms collectively referred to as clusteranalysis (Blashfield & Aldenderfer, 1988).bility, emotional instability, and motor control

abnormalities. Although the disorder is trans- These methods divide data sets into multiplepartitions, either by maximizing between groupsmitted by an allelic variant of a single dom-

inant gene on chromosome 4, its course is variance or by minimizing within-groups vari-ance. Although details about specific clustervariable across individuals, with age of onset

dispersed across a wide range (Brooks, Mur- analytic methods cannot be presented here be-cause of space constraints, there is ample evi-phy, Janota, & Lishman, 1987). Thus, the la-

tent cause of the disorder is necessarily taxo- dence to suggest that they do not provide strongtests of taxonic structure because they alwaysnic (either one has the genetic variant or one

does not), but the manifest indicators are vari- yield subgroups that differ significantly on theindicator variables, regardless of whether trueable (see also Meehl, 1995, 2001).

Second, indicators of taxon group member- taxa exist (see Blashfield & Aldenderfer, 1988;Klein & Riso, 1993). Moreover, even whenship may be distributed normally among non-

taxon group members. Using procedures to be known taxa are embedded in simulated datasets, clustering algorithms perform poorly whendescribed below, Woodward et al. (2000) re-

ported evidence that extreme behavioral reac- there is overlap among groups and often failto identify the correct number of clusters (Atlastivity among infants is distributed discretely.

About 10% of 4-month-olds were classified & Overall, 1994; Krieger & Green, 1999). Thisstate of affairs exists despite significant ef-into a highly reactive group based on exces-

sive arching, crying, hyperextension, and leg forts over the last two decades to develop ac-curate “stopping rules.” Thus, cutoffs for groupmovements in response to a series of stimuli.

Moreover, group membership was predictive membership that are derived from clusteranalyses are just as likely to be arbitrarily lo-of extreme behavioral inhibition 4 years later.

Yet it would be naive to expect a clear bound- cated along a continuous dimension as theyare to distinguish between true taxa. Becauseary between taxon and nontaxon group mem-

bers for specific indicators of behavioral reac- clustering algorithms are structure imposingrather than structure seeking, these methodstivity or fear. As an adaptive emotion, fear is

observed in the behavioral repertoires of all carry the real risk of identifying false joints innature and provide very weak tests of typologi-infants, with individual variability that charac-

terizes most evolved traits. Thus, infants in the cal models (Beauchaine & Beauchaine, 2002;Meehl, 1979).Woodward et al. sample who were nontaxon

group members exhibited a wide range of be-havioral reactivity.

Other approachesThird, most behavioral indicators used in

psychological and psychiatric research are mea- Several additional techniques have appearedin the psychology and psychiatry literatures insured with imperfect precision and reliability,

contributing to distributional overlap even when efforts to identify homogeneous subgroups ofindividuals within larger samples. These in-true taxa exist. This final point will be recon-

sidered in later sections of this article that out- clude mixture analysis (e.g., Fleiss, 1972), la-tent class analysis (e.g., Lazarsfeld & Henry,line strategies for future taxometrics work.1968), and more recently developed methodsfor identifying groups through analyses of la-

Cluster analysistent growth trajectories (e.g., Nagin, 1999).The operating characteristics of both mixtureFollowing recognition that bimodality and

clear boundaries are unlikely to be observed analysis and latent class analysis have beenexplored by a number of researchers, as sum-in the presence of discretely distributed traits,


marized by Klein and Riso (1993). In brief, of afflicted people. He also believed that acore set of observable phenotypic indicators,these techniques suffer from some of the same

limitations as cluster analysis. Both divide expressed as schizotypy, marked the geneticvariant, regardless of whether a person devel-most data sets into partitions that are consis-

tent with a discrete latent class interpretation, oped schizophrenia. Through careful observa-tion that included extensive clinical work withyet there is no way to determine whether the

identified classes are truly discrete. This is schizophrenia probands and their family mem-bers, Meehl (1962) identified four schizotypicalso the case when identifying groups using

latent growth trajectories (Nagin, 1999; Nagin characteristics as putative markers of the schiz-otaxic genotype: anhedonia, or a limited ca-& Tremblay, 2001). Thus, although these tech-

niques may have heuristic utility, none can de- pacity to experience pleasure2; interpersonalaversiveness, or social fear, distrust, and an-termine the ontological status of a trait or a

disorder as discrete versus continuous. ticipation of rejection; ambivalence, or seem-ingly concurrent motivation toward interper-sonal approach and withdrawal; and cognitive

Schizotaxia, Schizotypy,slippage, or somewhat loose control of associ-

Schizophrenia Revisitedations. Although unknown to Meehl at thetime, we might add to this list smooth pursuitIn his 1994 address for a Distinguished Pro-

fessional Contribution award from the Ameri- and saccade eye tracking abnormalities, whichare also observed in the relatives of schizo-can Psychological Association (APA), Meehl

(1995) summarized the results of a now 30- phrenia probands (e.g., Curtis, Calkins, Grove,Feil, & Iacono, 2001), and have been pro-year effort to develop techniques that could

distinguish between types and continua. The posed to result from a single dominant gene(e.g., Avila, McMahon, Elliott, & Thaker,resulting algorithms, referred to collectively

as coherent cut kinetics (CCKs), were devised 2002; Levy, Holzman, Matthysse, & Mendell,1993).with the intent of identifying latent taxa when

they exist and rejecting the taxon hypothesis It bears repeating that Meehl expected anunknown fraction of schizotypes to decom-when they do not, thereby addressing an in-

herent limitation of alternative methods of pensate into diagnosable schizophrenia. More-over, he recognized that schizotypy would beclassifying. Meehl’s (1995) paper outlined the

two most commonly used CCK procedures, difficult to identify with specificity becausethe distribution of schizotypic characteristicsMean Above Minus Below A Cut (MAMBAC;

Meehl & Yonce, 1994), and Maximum Co- among genetic positives was likely to overlapconsiderably with the distribution of similar traitsvariance (MAXCOV; Meehl & Yonce, 1996;

Waller & Meehl, 1998)–HITMAX. These tech- in the general population. Thus, new methodswould be required to disentangle the admixedniques will be described in some detail below;

but before doing so, a description of the con- distributions of schizotypes and genetic nor-mals. This is a classic case of a well-articu-text in which they were developed will be pre-

sented. This is important because the sus- lated theory serving to motivate and informthe development of a new methodological ap-tained effort of Meehl and his colleagues to

address one of the most fundamental prob- proach, and more importantly, a new way oflems in research on psychopathology exempli-fies the ingenuity, tenacity, and rigor required

2. In more recent writings, Meehl (1975, 1989, 1990) re-when tackling difficult scientific questions (Wat- canted the assertion that anhedonia marks a genetic lia-ers & Beauchaine, 2003). bility for schizophrenia, suggesting instead that hedonic

capacity is a dimensionally distributed trait. However,Meehl (1962) was one of the first propo-Blanchard et al. (2000) reported that anhedonia wasnents of a diathesis–stress model of psycho-distributed discretely among a large sample of collegepathology. He believed that a genetic diathesisstudents, with a base rate similar to that found in taxo-

for schizophrenia, which he labeled schizo- metrics investigations using other schizotypy markers.taxia, was transmitted through a single gene Meehl’s original position may have therefore been cor-

rect.but resulted in schizophrenia for only a subset

T. P. Beauchaine508

thinking about psychopathology. Although asingle gene locus has never been identified forschizophrenia, the schizotypic traits outlinedby Meehl have been shown in repeated taxo-metrics investigations to mark a group that isdistributed as a discrete latent class (Blanch-ard et al., 2000; Erlenmeyer–Kimling et al.,1989; Golden & Meehl, 1979; Korfine & Len-zenweger, 1995; Lenzenweger, 1999; Lenzen-weger & Korfine, 1992; Tyrka, Cannon, et al.,1995; Tyrka et al., 1995). It is to describing

Figure 2. An illustration of the MAXSLOPE pro-the most common CCK methods that this arti-cedure using a hypothetical bivariate distribution

cle now turns. of anhedonia and social anxiety. Note that theslope of the smoothed regression line is relativelyflat within groups, but it steepens where schizo-CCK Algorithmstypes (n = 250) and controls (n = 250) are mixed.The maximum slope is obtained at the value of so-

MAXSLOPE procedure cial anxiety that best discriminates between groups.

Meehl and colleagues’ CCK algorithms includeseveral interrelated procedures that identify requirement is that anhedonia and social anxi-taxonic structure by progressing along succes- ety are correlated less within groups than be-sive cuts of an indicator variable and examin- tween groups. If only one group is present,ing the statistical behavior of related variables then no discontinuity in slope is observed.in contiguous regions of the cut (Meehl, 1999;Meehl & Yonce, 1994, 1996).3 A description

MAXCOV procedureof MAXSLOPE (Grove & Meehl, 1993) pro-vides an intuitively appealing illustration of Although easier to explain, MAXSLOPE isthe CCK approach. Consider two hypothetical less commonly used than MAXCOV, a relatedgroups, those with and without a genetic di- procedure that operates on triads of indicatorsathesis for schizophrenia. Assume that reliable rather than variable pairs (see Meehl & Yonce,measures of anhedonia and social anxiety are 1996; Waller & Meehl, 1998). MAXCOV ex-obtained and that, for illustrative purposes, the amines the covariance, or unstandardized cor-groups are roughly equal in size. In MAX- relation of two indicators across successiveSLOPE, the indicators are plotted against one intervals of a third indicator, and a smoothedanother and a smoothed regression line is fit- plot is fitted through the resulting function.ted to the scatterplot (see Figure 2). The slope To illustrate, assume that in addition to hav-of the regression line (dy/dx) is then calcu- ing reliable indicators of social anxiety (x)lated at successive points of the x variable (so- and anhedonia (y), we collect eye-tracking ab-cial anxiety). If two discrete groups are pres- normality data (z) from groups with and with-ent and the effect size is adequate, then a out a genetic diathesis for schizophrenia. Adiscontinuity is observed in the regression hypothetical scatterplot of social anxiety andslope, which is maximized at the level of so- anhedonia is presented in the top left portioncial anxiety (x) that best discriminates be- of Figure 3, with their covariance across inter-tween groups. In other words, the correlation vals of eye tracking abnormality appearingbetween social anxiety (x) and anhedonia (y), below. Similar to the MAXSLOPE example,represented by the regression slope, is highest if two discrete groups are present and the ef-at the point of greatest admixture. The only fect size is adequate, then a discontinuity is

observed in the covariance of variables x andy (anhedonia and social anxiety), which is max-3. In total, there are 13 CCK procedures (Meehl, 1999).imized at the level of variable z (eye tracking)This article emphasizes the most commonly used and

extensively tested of these algorithms. that best differentiates between groups. This


Figure 3. An illustration of the MAXCOV procedure using hypothetical distributions ofanhedonia and social anxiety across intervals of eye tracking abnormality. (a) In the taxoniccase the greatest covariance of anhedonia and social anxiety is observed within the intervalof eye tracking abnormality that best discriminates between groups of schizotypes (n = 250)and controls (n = 250). (b) In the continuous case there is no peak in the covariance function(n = 500). Note that the overall correlation between anhedonia and social anxiety is .50 inboth samples.

is referred to as the HITMAX value, because used to assign individual cases or observationsto the taxon and nontaxon groups using Bayes’group membership assignments are most ac-

curate when it is used as a demarcating bound- theorem (see Beauchaine & Beauchaine, 2002;Meehl & Yonce, 1996; Waller & Meehl, 1998).ary. As illustrated in the bottom left portion

of Figure 3, the covariance function exhibits a This versatility has led to MAXCOV beingthe most commonly used CCK procedure.marked peak at the HITMAX value. At lower

taxon base rates, the location of this peak mi-grates toward the right (see Meehl, 1995; Meehl

MAMBAC procedure& Yonce, 1996). In contrast, given continuousdata the covariance function of x and y across When testing taxonic hypotheses, MAXCOV

is often used in conjunction with yet anotherintervals of z is relatively flat, as shown in thebottom right portion of Figure 3. CCK algorithm, MAMBAC (Meehl & Yonce,

1994). In this procedure, indicators are ana-For each trivariate combination of indica-tors, MAXCOV yields estimates of the taxon lyzed in variable pairs. One variable is first

sorted in ascending order, which also sorts thebase rate, the sample sizes of the taxon andnontaxon groups, the false positive and false other variable with some degree of efficiency

if both are valid markers of the analyzed trait.negative rates of group membership, and theHITMAX value. Moreover, MAXCOV can be Continuing with the previous example, sort-

T. P. Beauchaine510

Figure 4. An illustration of the MAMBAC procedure using hypothetical distributions ofanhedonia and social anxiety. In the taxonic case a peak in the MAMBAC plot is observedat the level of social anxiety that best discriminates between groups of schizotypes (n =500) and controls (n = 500). In the nontaxonic (continuous) case the MAMBAC plot is Ushaped (n = 1000).

ing on social anxiety (x) will also sort on an- As such, many of these investigations havebeen atheoretical, using imprecise rating scalehedonia (y) because the two variables are cor-

related. Next a sliding cut is moved across all measures that fail to tap putative mechanismsof taxonicity directly. Under these conditions,values of social anxiety, and the mean of an-

hedonia is calculated both above and below there is a greater chance of detecting taxa thatare spurious and a lesser chance of detectingthe cut. At each point the mean above the cut

is subtracted from the mean below the cut, taxa that truly exist. Furthermore, null find-ings have frequently been reported as support-and the resulting function is plotted. If a taxon

group is present, a marked peak is produced ing continuous models of psychopathology, alogical inconsistency that would probably notin the MAMBAC function at the value of so-

cial anxiety (x) that best discriminates between emerge through the editorial process, givenmore commonly used statistical methods. Inthose in the taxon group and those in the non-

taxon group (Meehl, 1995; Meehl & Yonce, the sections to follow, each of these issues willbe elaborated and general recommendations1994). In contrast, continuously distributed data

produce a U-shaped function (see Figure 4). for future taxometrics research will be pro-vided. Next, implications for addressing ques-tions of specific interest to developmentalEvaluating Taxonic Hypothesespsychopathologists will be outlined.

In the 30 years since Meehl (1973) introducedthe first of his CCK algorithms, much has been

Selecting valid indicatorslearned about conducting taxometrics investi-gations. Based on an increasing number of stud- Indicator validity refers to the degree to which

a variable marks the latent construct it is pur-ies appearing in the adult literature, severaltheoretical papers by Meehl and others, and ported to measure. In the case of schizotypy,

the results of numerous studies have confirmedMonte Carlo simulations examining the effi-ciencies of CCK procedures, it is now possi- that anhedonia, perceptual aberration, and so-

cial withdrawal mark a discretely distributedble to offer a number of recommendations forfuture taxometrics work. Above all, taxome- group of individuals who are at elevated risk

for schizophrenia, thereby establishing the va-trics investigations require that indicators beselected with great attention to validity and mea- lidity of each variable as an indicator of the

schizotypy construct (Blanchard et al., 2000;surement precision, the latter of which is rarelyconsidered in psychological research. More- Erlenmeyer–Kimling et al., 1989; Golden &

Meehl, 1979; Korfine & Lenzenweger, 1995;over, because CCKs require larger samplesthan are available in most studies, taxometric Lenzenweger, 1999; Lenzenweger & Korfine,

1992; Tyrka, Cannon, et al., 1995). However,questions have often been pursued in an adhoc fashion with large data sets of convenience. if the validity of indicators is contingent


upon their performance in previous taxome- Frick, DeShazo, McCoy, Ellis, & Loney, 2000;Beauchaine, 2001; Beauchaine, Katkin, Strass-trics investigations, how does one choose

markers of a hypothesized latent taxon that berg, & Snarr, 2001; Frick, Bodin, & Barry,2000).has yet to be identified? As outlined by Meehl

(1995), the search for taxonic entities is abootstrapping endeavor because there are al-

Increasing measurement precisionmost never gold standards or litmus tests forconfirming the validity of symptoms as mark- Although validity is often conflated with mea-

surement precision, the two are not equiva-ers of a discrete diagnostic entity. If therewere such gold standards, then there would be lent. Precision refers specifically to our ability

to measure a construct without error. Manyno need to conduct taxometrics investigationsin the first place. psychological constructs are low in measure-

ment precision, yet have adequate validity.Given that the validity of an indicator can-not be known a priori, it is important that can- This is particularly true for constructs as-

sessed via self- and other-report. For example,didates be selected based on strong theory(Meehl, 1999; Waters & Beauchaine, 2003). whereas depression may be assessed with rea-

sonable validity using scores from the Chil-This way, if no taxon is identified, it is un-likely to be a function of indicator invalidity. dren’s Depression Inventory (CDI; Kovacs,

1992), there is considerable measurement er-For example, suppose I suspect that a subsetof boys with conduct disorder (CD) are at ele- ror in any given score and it is inappropriate

to make equivalency statements across scores.vated genetic risk for developing psychopathy.This is similar to Meehl’s (1962) schizotaxia Thus, although significant impairment might

be inferred in a child with a CDI score of 30example, because both genetic and environmen-tal factors contribute to the psychopathy phe- (validity), it cannot be assumed that two dif-

ferent children with scores of 30 suffer fromnotype (see Lykken, 1995), yet the specific lociof the genotype have not been identified. In or- equivalent levels of depression (precision). In-

deed, Likert scales reflect an ordinal level ofder for CCKs to provide a reasonable test ofthe taxonic hypothesis, it is imperative that I measurement and do not carry the precision

of interval or ratio scales (Stevens, 1951).select indicators that are reasonably specificto the psychopathy construct. Thus, I should This discussion is important because high

measurement precision is essential for taxo-avoid using broadband externalizing symp-toms such as disobedience, oppositionality, or metrics investigations. Extensive Monte Carlo

simulations suggest that MAXCOV, the mostimpulsivity as indicators, because such symp-toms are observed across a wide range of chil- commonly used CCK, is effective in detecting

latent taxa only at effect sizes (d) of 1.2 ordren and adolescents and are only sometimesindicative of psychopathology (see, e.g., Hin- larger (Beauchaine & Beauchaine, 2002; Meehl,

1995). To place this in context, a typical ef-shaw, 2003). These symptoms are thereforeunlikely to mark genetic risk for psychopathy fect size in psychological research is around

.6, or half as large (Cohen, 1988). Thus, givenwith specificity. Given that psychopathy maybe distributed as a discrete class in adulthood an average effect size, existing taxa will go

undetected, even if valid indicators are used.(Harris et al., 1994), it would be best to selectvariables that distinguish between externaliz- CCKs simply do not have the discriminating

power to detect small and medium effects.ing adults with and without psychopathy andto consider variables that mark developmental Several strategies can be used to increase

measurement precision. Factor analysis of in-precursors of psychopathic traits. Candidatesmight include manipulativeness, callousness, dividual items into common components is

one such strategy. The effect is to reduce errorlack of empathy, and certain physiologicalmarkers of underarousal. Each of these has associated with single items by isolating shared

variance across items (see Nunnally & Bern-been tied more specifically to either psychop-athy or severe conduct problems than to broad- stein, 1994). Using factor scores as indicators

should therefore enhance the sensitivity of CCKsband externalizing symptoms (e.g., Barry,

T. P. Beauchaine512

to underlying taxa. In fact, there are taxome- Robins & Guze, 1970; see also Cicchetti &Dawson, 2002). Second, rating scales are knowntrics procedures designed specifically to deal

with factor scores (Waller & Meehl, 1998). Un- to elicit systematic response tendencies, in-cluding both halo effects (e.g., Saal, Dow-fortunately, many taxometric investigations con-

ducted to date have used individual items from ney, & Lahey, 1980) and positive/negative re-sponse biases (e.g., Macmillan & Creelman,factor analytically derived scales. Although

this strategy boosts the number of indicators 1990; Rajendar, 1996), either of which mightproduce spurious taxonic structure. Third, hu-available for analysis, it may hamstring ef-

forts to identify latent taxa due to low mea- man beings are prone toward categorical think-ing (see Malt, 1993; Rosch & Lloyd, 1978;surement precision.

A second strategy for reducing measure- Smith, 1995), which includes a natural ten-dency to classify based on past experiencesment error is to assess growth in the same

construct across time. In doing so, each addi- and preexisting beliefs (see Cantor & Genero,1986; Cantor & Mischel, 1979; Flanagan &tional assessment point provides enhanced pre-

cision (Rogosa, Brandt, & Zimowski, 1982). Blashfield, 2002; Semin & Rosch, 1981). Whenmaking categorical decisions, raters also be-For example, gender differences in the growth

trajectories of depressive symptoms emerge come more confident with increasing experi-ence (Dawes, Faust, & Meehl, 1989), and theirbefore mean differences in depression can be

detected (Cole, Tram, Martin, Hoffman, Ruiz, beliefs become more divergent over time, asassessed by Likert measures (Simon, Pham,Jacquez, & Maschman, 2002). This increased

precision suggests that longitudinal growth Le, & Holyoak, 2001). Indeed, as much as50% of the variance in Likert scales may betrajectories may be useful in taxometrics in-

vestigations. To date, however, only one lon- attributable to rater bias (Hoyt & Kerns, 1999).Thus, an exclusive reliance on rating scalegitudinal taxometric study has been conducted

(Tyrka, Haslam, et al., 1995), and growth tra- data may be problematic when performingtaxometrics research, particularly when evi-jectories were not used as indicators. This point

should be of particular interest to develop- dence suggests that raters hold categorical be-liefs about the nature of a construct. Attach-mental psychopathologists, given their con-

cern with symptom emergence over time. ment classifications, for example, are assignedbased on expert ratings of children’s behaviorFinally, precision can be increased by us-

ing indicators that carry inherently less mea- during separation from and reunion with theirmothers (Ainsworth, Blehar, Waters, & Wall,surement error than self- and other-report. Such

data sources might include behavior–observa- 1978; Sroufe & Waters, 1977). Because manyresearchers endorse categorical models of at-tional measures, biological or physiological

markers, or etiological information such as tachment (see Fraley & Waller, 1998), subtlebiases could impact their ratings, producingage of onset, among others. When collected

carefully, these indicators are more precise by latent distributions that appear to be discrete.In fact, we have demonstrated that manipulat-nature than data derived from Likert scales.ing the cognitive sets of raters as categoricalversus dimensional can produce spurious taxa

Using indicators from multiple levelswhen only Likert scale measures are used (Beau-

of analysischaine & Waters, 2003). Because no single in-dicator can drive taxonic findings, this doesThere are several additional reasons to include

variables other than those collected via rating not preclude the use of rating scale data intaxometrics investigations. However, it doesscales in taxometrics research. First, a much

stronger case for the validity of a disorder as suggest that variables from other levels ofanalysis should be included.a discrete class can be made when indicators

are drawn from multiple levels of analysis, in-Proving the nullcluding behavioral symptoms, biological signs,

and etiological markers (Feighner, Robins, In reporting the results of taxometrics analy-ses, it has become common to interpret nega-Guze, Woodruff, Winokur, & Munoz, 1972;


tive findings as supporting a continuous model sults from multiple MAXCOV runs will beinconsistent and existing latent taxa will go un-of the construct being assessed. Because CCK

procedures are structure seeking, however, there detected (e.g., Beauchaine & Beauchaine, 2002;Meehl, 1995).is a logical inconsistency in this practice. The

null hypothesis in taxometrics research is that Some elaboration on this two-stage proce-dure is warranted, given that it has been criti-the analyzed trait or disorder is distributed along

a continuum, and the alternative hypothesis is cized as a means of stacking the deck in favorof taxonic outcomes (Widiger, 2001). Providedthat it is distributed categorically. To con-

clude that a construct is continuously distrib- that one uses variables from several levels ofanalysis and examines multiple consistency testsuted based on negative findings is therefore

tantamount to proving the null. Thus, it is not (Meehl, 1995; Waller & Meehl, 1998), this isunlikely to be the case. Consistency tests referthe case that taxometrics analyses can indicate

the presence of a continuous distribution, as to comparisons of CCK-derived values of ahost of parameters describing the latent distri-some have suggested (e.g., Widiger, 2001). It

should be emphasized that this is not an ar- bution of a taxon. Suppose, for example, thatI begin with 15 putative indicators of schizo-cane point born of adherence to statistical

dogma. As outlined above, taxometrics analy- typy, which are winnowed down to 7 throughinitial MAMBAC analyses. The first consis-ses require high measurement precision and

indicator validity to detect fairly large effects tency test is to examine the base rate estimatesof the taxon in each of the MAMBAC runs.by psychological standards. Thus, CCKs can-

not distinguish types from continua for traitsSeven indicators yield 2�k2� or 42 combinationsor disorders with no associated indicators of

large effect size. Although there are a numberof variables for analysis (each indicator withinof reasons to suspect that many if not mosta variable pair can be used as both input andtraits are distributed continuously (Klein &output). If there is a true latent taxon present,Riso, 1993), negative results from taxometricsthen valid indicators should produce consis-analyses do not provide strong support for thetent estimates of the taxon base rate, regard-continuity of a trait any more than positiveless of the variable pairings. Thus, only whenresults from cluster analyses provide stronga preponderance of base rates fall within a smallsupport for typologies.range of values is the taxon hypothesis tenta-tively supported. If such consistency is found

Reducing candidate indicators across MAMBAC runs, then the indicatorsand examining multiple consistency tests are subjected to MAXCOV analyses. For the

7 indicator case, there areProvided that the construct being assessed isdiscrete, some markers are typically more pre-cise than others, and therefore do a better job i × (i − 1)!

(i − 3)!2!,

of differentiating taxon group members fromnontaxon group members, even if great careis taken in selecting indicators. Because of this, or 105 trivariate combinations available for

analysis. Each of these combinations producesMeehl has suggested that candidate variablesbe screened first using MAMBAC, and that estimates of the taxon base rate, the taxon

group mean, the HITMAX value, and the trueonly those that appear to mark a latent taxonbe subjected to subsequent MAXCOV analy- and false positive rates. Again, only when a

preponderance of these parameters cluster aroundses (Meehl, 1995, 2001). Typically, this is thestrategy used to filter an initial group of can- similar values is a taxonic interpretation sup-

ported. In addition, recent work suggests thatdidate indicators into a smaller number thatare more efficient at differentiating between a goodness of fit index (Joreskog & Sorbom,

2001) exceeding .90 is highly suggestive ofgroups (e.g., Waller et al., 1996). This processis necessary because if even a minority of in- taxonicity (Waller & Meehl, 1998).

The likelihood of several indicators drawndicators are inefficient discriminators, the re-

T. P. Beauchaine514

from biological, psychological, and etiologi- tological. Although this may seem obvious,evidence of taxonic structure in eating disor-cal levels of analysis converging on single

base rates, HITMAX values, and taxon group der symptoms may derive from such a sam-pling bias. Williamson, Womble, Smeets, Nete-means across 105 separate MAXCOV runs is

exceedingly low if there is no taxon present. meyer, Thaw, Kutlesic, and Gleaves (2002)compared DSM-IV symptoms from 201 womenIn fact, Monte Carlo simulations of 100s of

1000s of MAXCOV runs suggest that false diagnosed as having an eating disorder withthose from 116 normal-weight controls. Mem-positive taxonic findings occur at a negligible

rate when valid indicators are used (Beau- bers of the former group were all receivingtreatment for an eating disorder, whereas mem-chaine & Beauchaine, 2002). Thus, although

Widiger (2001) may be correct in suggesting bers of the latter group were recruited frompsychology classes. Based on taxometric anal-that taxonic distributions of attitudes are com-

mon, the same cannot be said for basic behav- yses, the authors concluded that bulimia andbinge eating disorder are discrete syndromes.ioral traits that are measured at several levels

of analysis. The appropriate sampling procedure for test-ing taxonic hypotheses is to recruit across awide range of symptomatology for a given

Samplingdisorder, with a representative proportion ofparticipants at each symptom level. AlthoughWhen conducting taxometrics research, both

sample size and sampling procedures must be some authors have suggested that the ultimateutility of CCKs depends upon their ability toconsidered. Monte Carlo simulations suggest

that the absolute minimum sample size for re- differentiate real taxa from pseudotaxa (e.g.,Waldman & Lilienfeld, 2001), no statisticalliable resolution of taxonic structure is around

200 with valid indicators, but sample sizes of procedure is immune to sampling biases andholding CCKs to a higher standard than other300 or more are preferred (Beauchaine & Beau-

chaine, 2002; Meehl, 1995). Unfortunately, this inferential statistics is probably unjustified.Thus, we need to think carefully about the na-precludes taxometric searches with many data

sets, aside from those derived from epidemio- ture of the data we subject to taxometric anal-yses or any other method.logical samples. Moreover, there are no short-

cuts around the sample size issue; the moreparticipants the better when conducting taxo-

Taxometrics andmetrics. Nevertheless, if variables are selected

Developmental Psychopathologycarefully across studies and equivalent sam-pling procedures are used, data sets may be Putting the requirements and limitations of

taxometric methods aside, what specific areascombined toward addressing taxonic questions.In some laboratories, this may mean combin- of interest to developmental psychopathologists

can be addressed by using the techniques? Thising data collected across several years of study.Finally, samples must be selected that do question is best answered by first considering

that developmental psychopathology is an in-not generate taxonic structure artificially. Theterm pseudotaxonicity has been used to refer tegrative discipline concerned with complex

processes through which multiple causal agentsto false positive taxonic findings resulting frombiases in sample selection (e.g., Beauchaine & interact to produce diverse adjustment outcomes

(Cicchetti, 1993; Cicchetti & Rogosch, 2002;Waters, 2003; Brown, 2001; Meehl, 1996).If, for example, one recruits a group of psy- Cicchetti & Toth, 1998). Because this frame-

work places an emphasis on individual copingchiatrically impaired individuals who exceedan extreme threshold of symptomatology and in response to intricate combinations of psy-

chological, biological, and environmental risk,compares them with a normative control group,then it should be no surprise if CCKs reveal person-centered approaches to studying psy-

chopathology have received increasing atten-a latent taxon. Confirmation of taxonicity in asample recruited for bimodality is simply tau- tion in the field (e.g., Caspi, 1998; Cicchetti


& Rogosch, 1999; Richters, 1997). Given that tify at-risk children, particularly at such ayoung age, may have important implicationstaxometric methods represent a person-centered

approach, it is all the more surprising that they for prevention and intervention.have gone largely unnoticed by developmentalpsychopathologists. Moreover, a number of im-

Identifying subtypes of disordersportant issues emerge from the developmentalpsychopathology perspective that can be ad- Since its inception about two decades ago, a

core theme in developmental psychopathol-dressed in part through well-conceived taxo-metrics studies. Several of these are presented ogy has been that observable syndromes arise

from many developmental pathways and etio-below.logical causes and that our current diagnosticsystem fails to capture heterogeneity among

Identifying children who are at riskindividuals within psychiatric classes (e.g., Cic-

for future psychopathologychetti & Rogosch, 1996). Some cases of de-pression, for example, appear to be influencedAs noted above, at least two discretely distrib-

uted traits that indicate elevated risk for psy- more by environmental risk, whereas othersappear to be influenced more by biologicalchopathology have been discovered in children

through taxometrics investigations. Erlenmeyer– risk (see Cicchetti & Rogosch, 2002; Harring-ton, Rutter, & Fombonne, 1996). Moreover,Kimling et al. (1989) demonstrated that mea-

sures of cognitive functioning and neuromotor these depression subtypes cannot be differen-tiated based solely on the behavioral criteriaperformance can identify schizotypy in 7- to

12-year-olds. Moreover, nearly 50% of chil- outlined in the DSM-IV. Given this, it is curi-ous that potential biological markers of de-dren with at least one parent with schizophre-

nia belonged to the schizotypy taxon group, pression and other psychiatric syndromes areoften eschewed in favor of strictly behavioralcompared with only 4% of age-matched con-

trols. Thus, among children known to be at criteria. The dexamethasone suppression test(DST), for example, an indicator of hypothal-heightened risk for developing schizophrenia

because one or both parents are afflicted, tax- amic–pituitary–adrenal axis (HPA) reactivity,was once considered a promising biologicalometric methods can pinpoint those individu-

als of particular vulnerability. Similar results marker of endogenous depression. Due to mod-erate specificity, however, some authors con-have been reported in adolescent samples (Tyrka

et al., 1995). I have already noted the poten- cluded that clinical symptoms are better indi-cators of the endogenous depression constructtial implications that these findings have for

prevention, given the role of familial and (e.g., Casat & Powell, 1988; Lu, Ho, Huang,& Lin, 1988). Thus, rather than entertainingother environmental influences on the onset

and expression of schizophrenia. the possibility that behavioral symptoms lackspecificity in identifying biologically basedThe second high-risk trait that has been

identified in children is behavioral reactivity. depressions, the utility of the marker wasquestioned and a potential opportunity to re-In a taxometric investigation of the responses

of 4-month-old infants to a series of visual, fine our diagnostic system was lost. More re-cent evidence suggests that depressed patientsauditory, and olfactory stimuli, Woodward et

al. (2000) identified an extremely reactive who exhibit dexamethasone nonsupression areat nearly 10 times the risk for future suicidetaxon group, who engaged in more arching,

crying, hyperextension, and leg movements than those who exhibit normal DST results(Coryell & Schlesser, 2001). Thus, abnormalduring stimulus presentations than other in-

fants. Furthermore, the 10% of infants who HPA reactivity may indeed mark a more viru-lent subtype of depression.belonged to the taxon group were behavior-

ally inhibited at age 4.5, marking a potential It should be emphasized that treating bio-logical markers as subordinate to behavioralpredisposition to later anxiety disorders (see

Kagan, 1994). Here again, the ability to iden- symptoms makes little sense scientifically. Be-

T. P. Beauchaine516

cause behavioral symptoms and biological mark- Asperger’s disorder reflect discrete behavioralsyndromes or fall at different points along aers are both manifest indicators of latent psy-

chopathology constructs, there is no a priori continuum of an autistic spectrum. This ques-tion has received considerable attention in thereason to suspect that the former are more

valid or reliable than the latter. Nevertheless, child psychopathology literature since the ap-pearance of Asperger’s disorder in the DSM-behavioral syndromes are often used as the

gold standard against which the validities of IV and has yet to be resolved (Volkmar, Klin,Schultz, Rubin, & Bronen, 2000). Both disor-biological markers are judged. In taxometrics

research, no such preference exists. Rather, ders are characterized by impaired social in-teraction, stereotyped behaviors, and restrictedany available indicator of adequate validity,

either behavioral or biological, can and should interests and activities. Autism is diagnosedwhen these symptoms are accompanied by de-be used in efforts to identify latent subgroups

within current diagnostic categories. In the case lays in verbal communication and/or cogni-tive development. When delays are not ob-of depression, promising candidates for use as

indicators in taxometric analyses include HPA served in either of these domains, Asperger’sdisorder is diagnosed.reactivity (e.g., Coryell & Schlesser, 2001)

and longitudinal symptom course (e.g., Cole Based on different concordance rates acrossdisorders in the first degree relatives of pro-et al., 2002), both of which are more precise

than cross-sectional symptom patterns (see bands (Volkmar, Klin, & Pauls, 1998), differ-ent patterns of comorbidity across diagnos-above). Symptoms of CD might also serve as

an indicator in differentiating between depres- tic groups (Ghaziuddin, Weidmer–Mikhail, &Ghaziuddin, 1998), and the aforementionedsion subgroups. This follows from the conten-

tion that depressions with and without CD differences in patterns of verbal and nonver-bal skills (e.g., Klin, Volkmar, Sparrow, Cic-symptoms are etiologically distinct, with the

latter deriving more from biological determi- chetti, & Roarke, 1995; Volkmar et al., 1994),some authors have concluded that autistic dis-nants (e.g., Harrington, Fudge, Rutter, Pick-

les, & Hill, 1991; Meller & Borchard, 1996; order and Asperger’s disorder are indeed dis-crete (see also Klin, 1994; Ozonoff, Rogers,Panak & Garber, 1992). If taxometric analy-

ses indicate discrete depression subtypes, then & Pennington, 1991). Others, however, haveargued for a continuous autism spectrum modela modification of our current diagnostic sys-

tem might be warranted. Preliminary evidence based on similarities in clinical features (e.g.,Eisenmajer, Prior, Leekam, Wing, Gould, Wel-suggests that endogenous depression may rep-

resent a discrete class among both adolescents ham, & Ong, 1996), neuropsychological testperformance (Miller & Ozonoff, 2000), andand adults (Ambrosini et al., 2002; Beach &

Amir, 2003; Haslam & Beck, 1994), although brain stem abnormities (Bauman, 1996).As noted by Meehl (1995), taxonomicanalyses to date have been conducted exclu-

sively with self-report indicators. If discrete questions such as these are irresolvable if nottested empirically by subjecting phenotypicendogenous and exogenous depression sub-

types are confirmed, it does not imply that indicators to formal taxometric analyses. Au-tism spectrum disorders provide an especiallyenvironmentally induced or influenced de-

pressions are unimportant or undeserving of ripe opportunity for testing taxonic conjec-tures because the autism phenotype has beentreatment. We are simply in a better position

to formulate effective interventions when we specified at several levels of analysis, with anumber of precise measures (see Akshoomoff,know as much as possible about the etiologi-

cal origins of a disorder (Cicchetti & Hin- Pierce, & Courchesne, 2002; Dawson, Webb,Schellenberg, Dager, Friedman, Aylward, &shaw, 2002), and different treatment approaches

may be required for different subtypes of de- Richards, 2002; Klin, Jones, Schultz, Volk-mar, & Cohen, 2002). At the psychophysio-pression.

A second diagnostic question that might be logical level, for example, children with au-tism spectrum disorders exhibit attenuatedaddressed through the use of formal taxome-

tric methods is whether autistic disorder and event-related potentials when presented with


familiar faces (Dawson, Carver, Meltzoff, Pa- addition, few studies have included informa-tion about biological markers or symptom course.nagiotides, McPartland, & Webb, 2002). At

the social level, infants with autism direct their However, the question of discreteness couldbe addressed by subjecting carefully selectedgaze toward others less frequently and orient

to their names less often than controls (Oster- variables to taxometric analyses. Moreover,given the size and comprehensiveness of ex-ling, Dawson, & Munson, 2002). Moreover,

adults with autism attend less to the eyes of isting data sets, taxometric analyses of ADHDmight be possible sooner rather than later.others when observing social interactions (Klin

et al., 2002). Finally, accelerated trajectories Similar to the case of depression, putativeindicators of inattention should extend beyondin brain growth are observed in autism, with

normal brain volume at birth but larger than DSM-IV criteria. One potential indicator is slug-gish cognitive tempo (SCT), a behavioral con-normal brain volume emerging between ages

2 and 4 years (Courchesne et al., 2001). As struct comprising symptoms of lethargy, day-dreaming, drowsiness, and hypoactivity (Carlsonnoted earlier, developmental trajectories such

as these provide increased precision over cross- & Mann, 2002; Lahey, Carlson, & Frick, 1997).Despite correlating specifically with ADHD/I insectional measures, and may therefore be par-

ticularly useful in taxometric studies. Note the DSM-IV field trials, SCT items were droppedfrom the final criterion list because the work-that these are but three examples from a num-

ber of potential indicators of the autism phe- group sought a single set of inattentivity itemsfor both ADHD subtypes (Carlson & Mann,notype (see Dawson, Webb, et al., 2002). If

these traits are distributed discretely, thereby 2002; Frick et al., 1994). Nevertheless, mea-sures of SCT are elevated among ADHD/I chil-indicating a typological distinction between

autistic disorder and Asperger’s disorder, fol- dren compared with other ADHD groups(McBurnett, Pfiffner, & Frick, 2001; Milichlow-up studies aimed at elucidating differen-

tial etiological mechanisms should be pur- et al., 2001). Although this does not guaranteethat ADHD/I marks a discrete disorder, it doessued. As noted above, such differences are

likely to have implications for both treatment suggest that markers of SCT might be usefulin taxometrics investigations.and long-term course.

There are additional examples of diagnostic It should be noted, however, that SCTand inattention are not fully overlapping andcategories that might be refined based on taxo-

metrics investigations. One clear candidate is therefore cannot be used as proxies for oneanother. Rather, SCT appears to mark a subsetattention-deficit/hyperactivity disorder (ADHD),

where a debate exists in the literature regard- of ADHD/I children who are also differenti-ated from ADHD/C children on other symp-ing the distinctiveness of the combined type

(ADHD/C), which is characterized by hyper- toms, including lower levels of externalizingbehaviors and higher levels of anxiety, depres-activity/impulsivity and inattention, and the

inattentive type (ADHD/I), which is charac- sion, and social withdrawal (Carlson & Mann,2002). Indeed, Carlson and Mann concludedterized primarily by the latter. Some authors

have argued that the disorders are likely to that ADHD/I children high on SCT representa distinct diagnostic group, whereas ADHD/Ibe discrete (e.g., Milich, Balentine, & Lynam,

2001), whereas others have argued that cur- children low on SCT are more similar to chil-dren with ADHD/C. If taxometrics investiga-rent data are inconclusive (e.g., Barkley, 2001;

Hinshaw, 2001; Lahey, 2001). Evidence of- tions sort children along these lines, a newdiagnostic category might be indicated (seefered for discrete subtypes has come from fac-

tor analytic studies that differentiate among also Milich et al., 2001).Research on the biology of ADHD also sug-symptoms and from cluster analytic studies

that differentiate among children based on symp- gests some potential indicators for taxometricinvestigations. For example, several studiestoms. As noted earlier, these methods are not

suitable for distinguishing types from con- have demonstrated reduced urinary 3-methoxy-4-hydroxyphenylglycol (MHPG), a norepineph-tinua, and results have not been fully consis-

tent across studies (see Milich et al., 2001). In rine metabolite, among children with ADHD

T. P. Beauchaine518

(e.g., Shekim, Dekirmenjian, Chapel, & as discrete latent classes among adults, includ-ing schizotypy (e.g., Blanchard et al., 2000;Davis, 1982; Shekim, Sinclair, Glaser, Hor-

witz, Javaid, & Bylund, 1987). Moreover, Korfine & Lenzenweger, 1995; Lenzenweger,1999; Tyrka, Cannon, et al., 1995), dissocia-MHPG deficiencies have been tied specifi-

cally to the biobehavioral substrates of impul- tive experiences (Waller et al., 1996; Waller& Ross, 1997), and psychopathy (Harris et al.,sivity (see Beauchaine, 2001), and may there-

fore be associated more strongly with the 1994). What is not clear is when in develop-ment these traits emerge as discrete entities.ADHD/C subtype. Consistent with this inter-

pretation, anxious children exhibit increased The case of psychopathy provides a particu-larly good example, as there has been muchurinary MHPG (Kagan, Reznick, & Snidman,

1987). Given the elevated symptoms of anxi- speculation about the developmental precur-sors of the construct and whether fledglingety observed in ADHD/I children (see Hin-

shaw, 2002a), particularly those with SCT psychopaths can be identified premorbidly.One school of thought is that comorbid con-(Carlson & Mann, 2002), urinary MHPG may

differentiate between ADHD subgroups in duct problems and hyperactivity/impulsivitymark future psychopathy (e.g., Lynam, 1996,taxometric analyses.

Finally, several studies have suggested that 1998). Because the prevalence rate of ADHDamong children and adolescents with CD ap-the ADHD/C subtype is characterized by ear-

lier ages of onset and referral than the ADHD/ proaches 70%, however (e.g., Klein et al., 1997),it is unlikely that comorbid ADHD marks aI subtype (e.g., Lahey et al., 1997; Faraone,

Biederman, Weber, & Russell, 1998). One po- predisposition for psychopathy with enoughspecificity to be used as an indicator in taxome-tential explanation for this is that hyperactive/

impulsive behaviors become unmanageable trics research. As noted earlier, variables thathold more promise include callous unemotionalfor parents and teachers before symptoms of

inattention are noticed (Milich et al., 2001). If traits and physiological markers of underarousal(e.g., Barry et al., 2000; Beauchaine, 2001;this explanation is ruled out in future studies,

age of onset might also be used as an indica- Beauchaine et al., 2001; Frick et al., 2000).Determining when the trajectory towardtor in taxometric analyses of ADHD subtypes.

Other markers are unlikely to serve as suit- psychopathy emerges as a discrete class willrequire carefully designed taxometrics studiesable indicators in taxometric analyses, either

because they do not differentiate between with progressively younger samples4 and mayhave important implications for treatment. ItADHD/C and ADHD/I subgroups, or because

they are of small effect size. For example, is well established that interventions for seri-ous conduct problems are of limited effect byboth the DAT1 and DRD4 dopamine genes

have been identified as candidates in contrib- the time probands reach adolescence (e.g., Di-shion & Patterson, 1992; Ruma, Burke, &uting to the ADHD phenotype. In theory, al-

lelic variants of these genes should contribute Thompson, 1996), but are more successfulwith younger children (e.g., Webster–Strat-specifically to hyperactivity/impulsivity and be

related more strongly to ADHD/C than to ton & Hammond, 1997; Webster–Stratton,Reid, & Hammond, 2002). Moreover, biologi-ADHD/I. However, both population- and fam-

ily-based studies suggest relatively low risk cal markers that were once considered to bestable indices of a diathesis for severe conductvalues, indicating modest genetic associations

between DAT1 and DRD4 allelic variants and problems now appear to be malleable in veryyoung children. For example, Raine, Venables,symptoms of hyperactivity/impulsivity (e.g.,

LaHoste et al., 1996; Smalley et al., 1998). Dalais, Mellingen, Reynolds, and Mednick(2001) reported that an enriched preschool en-Thus, effect sizes may be inadequate for use

in taxometric investigations. vironment, including child social skills train-

Locating bifurcation points4. Preliminary evidence suggests that a predisposition to-in the development of discrete traits

ward psychopathy may be distributed as a discrete la-As noted above, there are several psycho- tent class among children in Grades 4–8 (Skilling et

al., 2001).pathological traits that appear to be distributed


ing, extensive parental involvement, thorough come. In the context of intervention research,a moderator is any variable present at baselineinstruction of teachers in behavioral manage-

ment, and weekly counseling sessions for par- that discriminates among subgroups of indi-viduals who respond differentially to treatmentents, conferred a 61% increase in electroder-

mal activity on children 6–8 years later, (e.g., Kraemer, Stice, Kazdin, Offord, & Kup-fer, 2001). Because CCKs identify subpopula-compared with controls who were assigned

randomly to a no treatment condition. These tions who are different in kind, they can pro-vide us with non-arbitrary groups for whomdata suggest that long-term changes in the

functioning of biological systems implicated differential treatment response might be as-sessed. This in turn allows us to zero in onin aggression can be effected through multi-

faceted interventions in the preschool years. causal processes that enhance or diminish theimpact of an intervention in different groupsAfter such systems consolidate into discretely

distributed patterns of functioning, however, (see Hinshaw, 2002b). Indeed, in their recentreport on effective interventions for conductinterventions may simply be too late. Identify-

ing the point at which bifurcation into discrete problems, Brestan and Eyberg (1998) notedthat a considerable challenge facing the fieldtrajectories occurs may therefore inform ef-

forts to prevent and/or alter psychopathic in formulating the next generation of treat-ments is to identify those children who are nottraits. Although some might assume that genet-

ically influenced disorders are characterized served by current interventions and to deter-mine what additional resources they require.by clear, immutable biological markers very

early in life, this is typically not the case. It is known, for example, that one-third of chil-dren do not benefit from the most successfulThere is increasing recognition, for example,

that environmental factors contribute to the interventions for conduct problems (Webster–Stratton & Hammond, 1997; Webster–Strattonexpression of genes and that genetic effects

on behavior increase across the life span, in et al., 2002). Yet little is known about child-specific predictors of treatment response. Be-part due to exposure to accumulated environ-

mental risk (see, e.g., Goldsmith, Gottesman, cause well-conducted taxometrics research in-cludes specification of individual characteristics& Lemery, 1997; Rutter, Dunn, Plomin, Si-

monoff, Pickles, Maughan, Ormel, Meyer, & at behavioral, biological, psychological, andetiological levels, explorations of moderationEaves, 1997). For example, children who are

impulsive, a trait that is highly heritable, are will necessarily be enriched compared with thecurrent strategy of subgrouping children basedmore likely to develop conduct problems when

placed in high risk environments (Burt, Krueger, on often arbitrary distinctions in behavioralsymptoms alone.McGue, & Iacono, 2001; Lynam, Caspi, Mof-

fitt, Wikstrom, Loeber, & Novak, 2000; Pat- In addition, treatment response itself mightbecome an indicator in taxometric studies. Interson, DeGarmo, & Knutson, 2000). Inter-

ventions that occur prior to the behavioral the case of psychopathy, for example, treat-ment resistance may be an additional markerexpression of a genetically influenced trait

may therefore hold considerable promise in of an underlying trait that results in emotionalunresponsiveness, physiological underarousal,altering trajectories toward psychopathology. In

contrast, interventions delivered after a thresh- and a deficiency in learning from correctiveexperiences. All of these characteristics makeold of accumulated risk has been reached may

not prevent or alter the expression of a genetic it difficult to establish therapeutic leverageand may result in treatment failure. By track-liability. To the extent that discretely distrib-

uted traits mark genetic liabilities for psycho- ing a number of theoretically important indi-cators throughout the intervention process, apathology (e.g., Meehl, 1995), identifying bi-

furcation points may help us converge on ages more refined basic understanding of the char-acteristics of treatment nonresponders will befor optimal intervention effects.attained. This knowledge can in turn be used

Identifying moderators of treatment outcome to identify treatment-resistant individuals pro-spectively and to channel them away from in-Taxometric investigations can also help to iden-

tify person-specific moderators of treatment out- terventions that are unlikely to benefit them

T. P. Beauchaine520

and toward new interventions targeting their the taxon and nontaxon groups? These andother questions take on considerably more mean-specific behavioral traits.ing when subgroups of children are defined bynonarbitrary cutoffs. Answering each question

Elucidating mechanisms of equifinalitywill require carefully designed studies evalu-

and multifinalityating multiple risk models for child psycho-pathology, an approach that has already provenAs noted above, equifinality suggests that a

given disorder can be the end state of numer- fruitful with normative samples (e.g., Lengua,2002).ous developmental pathways, and multifinal-

ity suggests that children in similar high-risksituations can diverge toward quite disparate

Summary and Recommendationsend states, only some of which will be disor-

for Future Researchdered (Cicchetti & Rogosch, 1996). This frame-work implies that a greater understanding of To date, taxometric methods have been used

quite sparingly by developmental psychopa-the diversity of individual outcomes will beattained through longitudinal analyses of child- thologists. One probable reason for this is a

well-founded wariness of categorical classifi-specific variables and how these interact withenvironmental experiences to produce disor- cation systems, which fail to account for within-

group heterogeneity in biological, develop-dered and nondisordered outcomes. Thus, simi-lar to the case of identifying moderators of treat- mental, etiological, and cultural influences on

behavior. I have argued, however, that thement, elucidating mechanisms of equifinalityand multifinality requires that homogeneous limitations of categorizing have very little to

do with the ontological status of specific traitssubgroups of individuals who are at differen-tial risk for psychopathology be identified and or disorders as dimensional versus discrete. I

have argued further that developmental psy-followed throughout the natural course of de-velopment (see Hinshaw, 2002b; Richters & chopathologists should be concerned with iden-

tifying discrete behavioral syndromes, becauseCicchetti, 1993). Recall that Woodward et al.(2000) reported evidence that behavioral reac- doing so could result in a number of research

advances, both basic and applied. These includetivity is distributed discretely in childhood,with roughly 10% of 4-month-old infants fall- (a) identifying children who are at increased

risk for developing psychopathology, (b) iden-ing into an extremely reactive taxon group whoscored high on measures of behavioral inhibi- tifying discrete subtypes of disorders within

current diagnostic classes, (c) locating sensi-tion 4 years later. Given that the taxon can beidentified at a very young age, these findings tive periods in the development of discrete

pathological traits, (d) discovering moderatorsprovide a unique opportunity to explore longi-tudinal outcomes among qualitatively distinct of treatment-outcome, and (e) elucidating mech-

anisms of equifinality and multifinality. Finally,groups of children who are at differential bio-logical risk for later psychopathology (see Ka- I have provided several examples of diagnos-

tic classes for which our understanding mightgan, 1997).A number of questions emerge from the be refined through formal taxometrics investi-

gations, including anxiety disorders, ADHD,Woodward et al. (2000) findings. What setsof environmental experiences (e.g., family, autism spectrum disorders, conduct problems,

depression, and schizophrenia.school, neighborhood) result in taxon groupmembers developing later anxiety disorders or However, despite the potential utility of

taxometric methods in clarifying some of thedepression? What sets of environmental expe-riences protect taxon group members from de- most fundamental diagnostic questions in psy-

chopathology, they are unlikely to be used inveloping later psychopathology? What differ-ences in environmental experiences result in the future by developmental psychopatholo-

gists if two barriers are not overcome. First,equifinal outcomes for taxon group memberscompared with nontaxon group members? What there has been a lack of communication across

interdisciplinary boundaries, with almost alldifferences in environmental experiences re-sult in multifinal outcomes for members within taxometric investigations appearing in the adult


psychopathology literature. Thus, many de- will be in the greatest position to select appro-priately precise indicators from behavioral, bi-velopmental psychopathologists may simply

be unaware of the techniques. Although this ap- ological, developmental, physiological, and eti-ological levels of analysis.pears to be changing based on a small set of

recent taxometrics studies with juvenile sam- A second barrier to conducting taxometricsresearch has been a lack of readily availableples (Fraley & Spieker, in press; Skilling et al.,

2001; Woodward et al., 2000), increased in- software for performing analyses. In fact, thereare currently no statistics packages that offerterdisciplinary communication is essential if

the pitfalls of taxometrics research identified a taxometrics module. Thus, interested re-searchers have been forced to write their ownin the adult psychopathology literature are to

be avoided. As reviewed in previous sections, software, a somewhat daunting task given thecomplexity of CCK algorithms (see, e.g., Meehlthese include problems with invalid variable

selection, inadequate measurement precision, & Yonce, 1994, 1996; Waller & Meehl, 1998).However, several taxometrics programs havemisinterpretation of null findings, and misguided

sampling procedures, among others. Develop- been written by the Waller, Meehl, and Yoncegroup, and are now available on their taxome-mental psychopathologists are in a unique po-

sition to benefit from the work of adult psy- trics home page (http://peabody.vanderbilt.edu/depts/psych_and_hd/faculty/wallern/tx.html).chopathologists, and it would be unfortunate

indeed if lessons learned in their research ef- These programs run in the R language, whichis both free and relatively easy to learn. Ourforts were ignored.

Perhaps the most critical issue identified to group has also posted free taxometrics pro-grams on our website, which are written indate is the need for taxometric hypotheses to

be tested in studies that are carefully and pro- Statistica (1998) BASIC (http://tbeauchaine.psych.washington.edu/tb/cbs/taxometrics.htm).spectively planned. As outlined in earlier sec-

tions, much of the extant taxometrics research Moreover, both the Waller group and our groupare currently developing additional programshas been conducted with datasets of conve-

nience, using primarily Likert-type scales as that should make taxometric methods accessi-ble to a broadened range of researchers.indicators. Although understandable given the

large sample sizes required for taxometric Taxometric methods are now 30 years old.Their advantages, limitations, and operatinganalyses, such ad hoc approaches are unlikely

to provide variables from multiple levels of characteristics have been outlined in a numberof theoretical, empirical, and simulation stud-analysis that are of adequate precision for test-

ing taxonic conjectures. Rather, putative mark- ies. It is hoped that developmental psychopa-thologists will draw from the lessons learneders of a hypothesized latent taxon must be se-

lected in advance based on strong theory, and from these studies in designing methodologi-cally rigorous and theoretically informed tax-must be measured with minimal error. As such,

taxometrics studies are likely to be informed ometrics research that will advance the fieldin the emergent 21st century.best by interdisciplinary research teams, who

References

Achenbach, T. M. (1991). Manual for the Child Behavior lam, N. (2002). Taxonicity of adolescent melancholia:A categorical or dimensional construct? Journal ofChecklist/4-18 and 1991 profile. Burlington, VT:

University of Vermont, Department of Psychiatry. Psychiatric Research, 36, 247–256.American Psychiatric Association. (1994). DiagnosticAchenbach, T. M. (1993). Empirically based taxonomy.

Burlington, VT: University of Vermont Department and Statistical Manual of Mental Disorders (4th ed.).Washington, DC: Author.of Psychiatry.

Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, Atlas, R. S., & Overall, J. E. (1994). Comparative evalua-tion of two superior stopping rules for hierarchicalS. (1978). Patterns of attachment: A psychological

study of the Strange Situation. Hillsdale, NJ: Erlbaum. cluster analysis. Psychometrika, 59, 581–591.Avila, M. T., McMahon, R. P., Elliott, A. R., & Thaker,Akshoomoff, N., Pierce, K., & Courchesne, E. (2002).

The neurobiological basis of autism from a develop- G. K. (2002). Neuropsychological markers of vulner-ability to schizophrenia: Sensitivity and specificity ofmental perspective. Development and Psychopathol-

ogy, 14, 613–634. specific quantitative eye movement measures. Jour-nal of Abnormal Psychology, 111, 259–267.Ambrosini, P. J., Bennett, D. S., Cleland, C. M., & Has-

T. P. Beauchaine522

Barkley, R. A. (2001). The inattentive type of ADHD as Milton & G. Klerman (Eds.), Contemporary direc-a distinct disorder: What remains to be done. Clinical tions in psychopathology: Toward the DSM-IV (pp.Psychology Science and Practice, 8, 489–493. 233–256). New York: Guilford Press.

Bauman, M. L. (1996). Neuroanotomic observations of Cantor, N., & Mischel, W. (1979). Prototypes in personthe brain in pervasive developmental disorders. Jour- perception. In L. Berkowitz (Ed.), Advances in exper-nal of Autism and Developmental Disorders, 26, 199– imental social psychology (Vol. 12). Orlando, FL:203. Academic Press.

Barry, C. T., Frick, P. J., DeShazo, T. M., McCoy, M. Cantwell, D. P. (1996). Classification of child and adoles-Ellis, M., & Loney, B. R. (2000). The importance of cent psychopathology. Journal of Child Psychologycallous–unemotional traits for extending the concept and Psychiatry, 37, 3–12.of psychopathy to children. Journal of Abnormal Psy- Carlson, C. L., & Mann, M. (2002). Sluggish cognitivechology, 109, 335–340. tempo predicts a different pattern of impairment in

Beach, S. R., & Amir, N. (2003). Is depression taxonic, the attention deficit hyperactivity disorder, predomi-dimensional, or both? Journal of Abnormal Psychol- nantly inattentive type. Journal of Clinical Child andogy, 112, 228–236. Adolescent Psychology, 31, 123–129.

Beauchaine, T. P. (1999). Definitions and levels of analy- Casat, C. D., & Powell, K. (1988). The dexamethasonesis. American Psychologist, 54, 439–440. suppression test in children and adolescents with ma-

Beauchaine, T. P. (2001). Vagal tone, development, and jor depressive disorder: A review. Journal of ClinicalGray’s motivational theory: Toward an integrated Psychiatry, 10, 390–393.model of autonomic nervous system functioning in Caspi, A. (1998). Personality development across the lifepsychopathology. Development and Psychopathology, course. In W. Damon (Series Ed.) & N. Eisenberg13, 183–214. (Ed.). Handbook of child psychology: Vol. 3. Social,

Beauchaine, T. P., & Beauchaine III, R. J. (2002). A emotional, and personality development (5th ed., pp.comparison of maximum covariance and k-means 312–371). New York: Wiley.cluster analysis in classifying cases into known taxon Cicchetti, D. (1993). Developmental psychopathology:groups. Psychological Methods, 7, 245–261. Reactions, reflections, projections. Developmental

Beauchaine, T. P., Katkin, E. S., Strassberg, Z., & Snarr, Review, 13, 471–502.J. (2001). Disinhibitory psychopathology in male ado- Cicchetti, D., & Dawson, G. (Eds.). (2002). Multiple lev-lescents: Discriminating conduct disorder from atten- els of analysis [Special issue]. Development and Psy-tion-deficit/hyperactivity disorder through concurrent chopathology, 14(3).assessment of multiple autonomic states. Journal of Cicchetti, D., & Hinshaw, S. P. (Eds.). (2002). PreventionAbnormal Psychology, 110, 610–624. and intervention science: Implications for develop-

Beauchaine, T. P., & Waters, E. (2003). Pseudotaxonicity mental theory [Special Issue]. Development and Psy-in MAMBAC and MAXCOV analyses of rating scale chopathology, 14(4).data: Turning continua into classes by manipulating Cicchetti, D., & Rogosch, F. A. (1996). Equifinality andobserver’s expectations. Psychological Methods, 8, multifinality in developmental psychopathology. De-3–15. velopment and Psychopathology, 8, 597–600.

Blanchard, J. J., Gangestad, S. W., Brown, S. A., & Cicchetti, D., & Rogosch, F. A. (1999). Conceptual andHoran, W. P. (2000). Hedonic capacity and schizo- methodological issues in developmental psychopath-typy revisited: A taxometric analysis of social anhe- ology research. In P. C. Kendall, J. N. Butcher, &donia. Journal of Abnormal Psychology, 109, 87–95. G. N. Holmbeck (Eds.), Handbook of research meth-

Blashfield, R. K., & Aldenderfer, M. S. (1988). The ods in clinical psychology (2nd ed., pp. 433–465).methods and problems of cluster analysis. In J. Nessel- New York: Wiley.roade & R. B. Cattell (Eds.), Handbook of multivari- Cicchetti, D., & Rogosch, F. A. (2002). A developmentalate experimental psychology (2nd ed., pp. 447–474).

psychopathology perspective on adolescence. JournalNew York: Plenum Press.of Consulting and Clinical Psychology, 70, 6–20.Bremner, J. D., & Vermetten, E. (2001). Stress and devel-

Cicchetti, D., & Toth, S. L. (1998). Perspectives on re-opment: Behavioral and biological consequences. De-search and practice in developmental psychopathol-velopment and Psychopathology, 13, 473–489.ogy. In W. Damon (Series Ed.), I. E. Sigel, & K. A.Brestan, E. V., & Eyberg, S. M. (1998). Effective psycho-Renninger (Vol. Eds.), Handbook of child psycho-social treatments of conduct-disordered children andpathology: Vol 4. Child psychology in practice (5thadolescents: 29 years, 82 studies, and 5,272 kids.ed., pp. 479–583). New York: Wiley.Journal of Clinical Child Psychology, 27, 180–189.

Clark, L. A., Watson, D., & Reynolds, S. (1995). Diagno-Brooks, D. S., Murphy, D., Janota, I., & Lishman, W. A.sis and classification in psychopathology: Challenges(1987). Early-onset Huntington’s chorea. British Jour-to the current system and future directions. Annualnal of Psychiatry, 151, 850–852.Review of Psychology, 46, 121–153.Brown, T. A. (2001). Taxometric methods and the classi-

Cohen, J. (1988). Statistical power analysis for the be-fication and comorbidity of mental disorders: Meth-havioral sciences (2nd ed.). New York: Academicodological and conceptual consideration. ClinicalPress.Psychology Science and Practice, 8, 534–541.

Cole, D. A., Tram, J. M., Martin, J. M., Hoffman, K. B.,Burt, S. A., Krueger, R. F., McGue, M., & Iacono, W. G.Ruiz, M. D., Jacquez, F. M., & Maschman, T. L.(2001). Sources of covariation among attention-defi-(2002). Individual differences in the emergence of de-cit/hyperactivity disorder, oppositional defiant disor-pressive symptoms in children and adolescents: Ader, and conduct disorder: The importance of sharedlongitudinal investigation of parent and child reports.environment. Journal of Abnormal Psychology, 110,Journal of Abnormal Psychology, 111, 156–165.516–525.

Cornblatt, B., Obuchowski, M., Roberts, S., Pollack,Cantor, N., & Genero, N. (1986). Psychiatric diagnosisand natural categorization: A close analogy. In T. S., & Erlenmeyer–Kimling, L. (1999). Cognitive and


behavioral precursors of schizophrenia. Development L. E. Beutler & M. L. Malik (Eds.), Rethinking theDSM (pp. 121–145). Washington, DC: Americanand Psychopathology, 11, 487–508.

Coryell, W., & Schlesser, M. (2001). The dexamethasone Psychological Association.Fleiss, J. L. (1972). Classification of the depressive disor-suppression test and suicide prediction. American

Journal of Psychiatry, 158, 748–753. ders by numerical taxonomy. Journal of PsychiatricResearch, 9, 141–153.Courchesne, E., Karns, C., Davis, H. R., Ziccardi, R.,

Carper, R., Tigue, Z., Chisum, H. J., Moses, P., Fraley, R. C., & Spieker, S. J. (2003). Are infant attach-ment patterns continuously or categorically distrib-Pierce, K., Lord, C., Lincoln, A. J., Pizzo, S., Schreib-

man, L., Haas, R. H., Akshoomoff, N., & Courchesne, uted? A taxometric analysis of Strange Situation be-havior. Developmental Psychology, 39, 387–404.R. Y. (2001). Unusual brain growth patterns in early

life in patients with autistic disorder. Neurology, 57, Fraley, R. C., & Waller, N. G. (1998). Adult attachmentpatterns: A test of the typological model. In J. A. Sim-245–254.

Cummings, E. M., Davies, P. T., & Campbell, S. B. pson & W. S. Rholes (Eds.), Attachment theory andclose relationships (pp. 77–114). New York: Guilford(2000). Developmental psychopathology and family

process. New York: Guilford Press. Press.Frick, P. J., Bodin, S. D., & Barry, C. T. (2000). Psycho-Curtis, C. E., Calkins, M. E., Grove, W. M., Feil, K. J., &

Iacono, W. G. (2001). Saccadic disinhibition in pa- pathic traits and conduct problems in community andclinic-referred samples of children: Further develop-tients with acute and remitted schizophrenia and their

first-degree biological relatives. American Journal of ment of the Psychopathy Screening Device. Psycho-logical Assessment, 12, 382–393.Psychiatry, 158, 100–106.

Dawes, R. M., Faust, D., & Meehl, P. E. (1989). Clinical Frick, P. J., Lahey, B. B., Applegate, B., Kerdyck, L.,Ollendick, T., Hynd, G. W., Garfinkel, B., Greenhill,versus actuarial judgment. Science, 243, 1668–1674.

Dawson, G., Carver, L., Meltzoff, A. N., Panagiotides, L., Biederman, J., Barkley, R., McBurnett, K., New-corn, J., & Waldman, I. (1994). DSM-IV field trialsH., McPartland, J., & Webb, S. J. (2002). Neural cor-

relates of face and object recognition in young chil- for the disruptive behavior disorders: Symptom utilityestimates. Journal of the American Academy of Childdren with autism spectrum disorder, developmental

delay, and typical development. Child Development, and Adolescent Psychiatry, 33, 529–539.Ghaziuddin, M., Weidmer–Mikhail, E., & Ghaziuddin,73, 700–717.

Dawson, G., Webb, S., Schellenberg, G. D., Dager, S., N. (1998). Comorbidity of Asperger syndrome: Apreliminary report. Journal of Intellectual DisabilitiesFriedman, S., Aylward, E., & Richards, T. (2002).

Defining the broader phenotype of autism: Genetic, Research, 42, 279–283.Golden, R. R., & Meehl, P. E. (1979). Detection of thebrain, and behavioral perspectives. Development and

Psychopathology, 14, 581–611. schizoid taxon with MMPI indicators. Journal of Ab-normal Psychology, 88, 212–233.Dishion, T. J., & Patterson, G. R. (1992). Age effects in

parent training outcome. Behavior Therapy, 23, 719– Goldsmith, H. H., Gottesman, I. I., & Lemery, K. S.(1997). Epigenetic approaches to developmental psy-729.

Dulmus, C. N., & Smyth, N. J. (2000). Early-onset chopathology. Development and Psychopathology, 9,365–387.schizophrenia: A literature review of empirically based

interventions. Child and Adolescent Social Work Jour- Grayson, D. A. (1987). Can categorical and dimensionalviews of psychiatric illness be distinguished? Britishnal, 17, 55–69.

Eisenmajer, R., Prior, M., Leekam, S., Wing, L., Gould, Journal of Psychiatry, 151, 355–361.Grove, W. M., & Meehl, P. E. (1993). Simple regression-J., Welham, M., & Ong, B. (1996). Comparison of

clinical symptoms in autism and Asperger’s disorder. based procedures for taxometric investigation. Psy-chological Reports, 73, 707–737.Journal of the American Academy of Child and Ado-

lescent Psychiatry, 35, 1523–1531. Hamilton, E. B., Asarnow, J. R., & Tompson, M. C.(1999). Family interaction styles of children with de-Erlenmeyer–Kimling, L., Golden, R. R., & Cornblatt,

B. A. (1989). A taxometric analysis of cognitive and pressive disorders, schizophrenia-spectrum disorders,and normal controls. Family Process, 38, 463–476.neuromotor variables in children at risk for schizo-

phrenia. Journal of Abnormal Psychology, 98, 203– Harrington, R., Fudge, H., Rutter, M., Pickles, A., & Hill,J. (1991). Adult outcome of childhood and adolescent208.

Falloon, I. R. H., Boyd, J. L., McGill, C. W., Williamson, depression: II. Links with antisocial disorders. Jour-nal of the American Academy of Child and AdolescentM., Razani, J., Moss, H. B., Gilderman, A. M., &

Simpson, G. M. (1985). Family management in the Psychiatry, 30, 434–439.Harrington, R., Rutter, M., & Fombonne, E. (1996). De-prevention of morbidity in schizophrenia. Archives of

General Psychiatry, 42, 887–896. velopmental pathways in depression: Multiple mean-ings, antecedents, and endpoints. Development andFaraone, S. V., Biederman, J., Weber, W., & Russell,

R. L. (1998). Psychiatric, neuropsychological, and Psychopathology, 8, 601–616.Harris, G. T., Rice, M. E., & Quinsey, V. L. (1994). Psy-psychosocial features of DSM-IV subtypes of atten-

tion-deficit/hyperactivity disorder: Results from a chopathy as a taxon: Evidence that psychopaths area discrete class. Journal of Consulting and Clinicalclinically-referred sample. Journal of the American

Academy of Child and Adolescent Psychiatry, 37, Psychology, 62, 387–397.Haslam, N., & Beck, A. T. (1994). Subtyping major de-185–193.

Feighner, J. P., Robins, E., Guze, S. B., Woodruff, R. A., pression: A taxometric analysis. Journal of AbnormalPsychology, 103, 686–692.Winokur, G., & Munoz, R. (1972). Diagnostic criteria

for use in psychiatric research. Archives of General Hinshaw, S. P. (2001). Is the inattentive subtype ofADHD a separate disorder? Clinical Psychology Sci-Psychiatry, 26, 57–63.

Flanagan, E. H., & Blashfield, R. K. (2002). Psychiatric ence and Practice, 8, 498–501.Hinshaw, S. P. (2002a). Is ADHD an impairing conditionclassification through the lens of ethnobiology. In

T. P. Beauchaine524

in childhood and adolescence? In P. S. Jensen & J. R. logical characterization of Asperger syndrome: Con-vergence with nonverbal learning disabilities syndrome.Cooper (Eds.). Attention deficit hyperactivity disorder

(pp. 5.1–5.21). Kingston, NJ: Civic Research Insti- Journal of Child Psychology and Psychiatry, 36,1127–1140tute.

Hinshaw, S. P. (2002b). Prevention/intervention trials and Korfine, L., & Lenzenweger, M. F. (1995). The taxonic-ity of schizotypy: A replication. Journal of Abnormaldevelopmental theory: Commentary on the Fast Track

special section. Journal of Abnormal Child Psychol- Psychology, 104, 26–31.Kovacs, M. (1992). Children’s Depression Inventory.ogy, 30, 53–59.

Hinshaw, S. P., Lahey, B. B., & Hart, E. L. (1993). Issues North Tonawanda, NY: Multi-Health Systems.Kraemer, H. C., Stice, E., Kazdin, A., Offord, D., &of taxonomy and comorbidity in the development of

conduct disorder. Development and Psychopathology, Kupfer, D. (2001). How do risk factors work to-gether? Mediators, moderators, and independent,5, 31–49.

Hinshaw, S. P., & Lee, S. S. (2003). Conduct and opposi- overlapping, and proxy risk factors. American Jour-nal of Psychiatry, 158, 848–856.tional defiant disorders. In E. J. Mash & R. A. Bark-

ley (Eds.), Child psychopathology (2nd ed., pp. 144– Krieger, A. M., & Green, P. E. (1999). A cautionary noteon using internal cross validation to select the number198). New York: Guilford Press.

Hinshaw, S. P., & Park, T. (1999). Research problems of clusters. Psychometrika, 64, 341–353.Lahey, B. B. (2001). Should the combined and predomi-and issues: Toward a more definitive science of dis-

ruptive behavior disorders. In H. C. Quay & A. E. nantly inattentive types of ADHD be considered dis-tinct and unrelated disorders? Not now, at least. Clini-Hogan (Eds.), Handbook of disruptive behavior disor-

ders (pp. 593–620). New York: Plenum Press. cal Psychology Science and Practice, 8, 494–497.Lahey, B. B., Carlson, C. L., & Frick, P. J. (1997). Atten-Hogarty, G. E., Anderson, C. M., Teiss, D. J., Kornblith,

S. J., Greenwald, D. P., Ulrich, R. F., & Carter, M. tion-deficit disorder without hyperactivity. In T. A.Widiger, A. J. Frances, H. A. Pincus, R. Ross., M. B.(1991). Family psychoeducation, social skills train-

ing, and maintenance chemotherapy in the aftercare First, & W. Davis (Eds.), DSM-IV sourcebook (Vol.3, pp. 163–188). Washington, DC: American Psychi-treatment of schizophrenia. Archives of General Psy-

chiatry, 48, 340–347. atric Association.LaHoste, G. J., Swanson, J. M., Wigal, S. B., Glabe, C.,Houts, A. C. (2002). Discovery, invention, and the expan-

sion of the modern Diagnostic and Statistical Manual Wigal, T., King, N., & Kennedy, J. L. (1996). Dopa-mine D4 receptor polymorphism is associated with at-of Mental Disorders. In L. E. Beutler & M. L. Malik

(Eds.), Rethinking the DSM (pp. 17–65). Washington, tention deficit hyperactivity disorder. Molecular Psy-chiatry, 1, 121–124.DC: American Psychological Association.

Hoyt, W. T., & Kerns, M. D. (1999). Magnitude and Lazarsfeld, P. F., & Henry, N. W. (1968). Latent struc-ture analysis. Boston: Houghton Mifflin.moderators of bias in observer ratings: A meta-analy-

sis. Psychological Methods, 4, 403–424. Lengua, L. J. (2002). The contribution of emotionalityand self-regulation to the understanding of children’sJensen, P. S., & Hoagwood, K. (1997). The book of

names: DSM-IV in context. Development and Psycho- response to multiple risk. Child Development, 73,144–161.pathology, 9, 231–249.

Joreskog, K. G., & Sorbom, D. (2001). LISREL 8: User’s Lenzenweger, M. F. (1999). Deeper into the schizotypytaxon: On the robust nature of maximum covariancereference guide. Lincolnwood, IL: Scientific Software

International. analysis. Journal of Abnormal Psychology, 108, 182–187.Kagan, J. (1994). Galen’s prophecy. New York: Basic

Books. Lenzenweger, M. F., & Korfine, L. (1992). Confirmingthe latent structure and base rate of schizotypy: A tax-Kagan, J. (1997). Conceptualizing psychopathology: The

importance of developmental profiles. Development ometric analysis. Journal of Abnormal Psychology,101, 567–571.and Psychopathology, 9, 321–334.

Kagan, J., Reznick, J. S., & Snidman, N. (1987). The Levy, D. L., Holzman, P. S., Matthysse, S., & Mendell,N. R. (1993). Eye tracking dysfunction and schizo-physiology and psychology of behavioral inhibition in

children. Child Development, 58, 1459–1473. phrenia. Schizophrenia Bulletin, 19, 461–536.Lilienfeld, S. O., & Marino, L. (1995). Mental disorderKendell, R. E. (1989). Clinical validity. Psychological

Medicine, 19, 45–55. as a Roschian concept: A critique of Wakefield’s“harmful dysfunction” analysis. Journal of AbnormalKlein, D. N., & Riso, L. P. (1993). Psychiatric disorders:

Problems of boundaries and comorbidity. In C. G. Psychology, 104, 411–420.Lilienfeld, S. O., & Marino, L. (1999). Essentialism re-Costello (Ed.), Basic issues in psychopathology (pp.

19–66). New York: Guilford Press. visited: Evolutionary theory and concept of mentaldisorder. Journal of Abnormal Psychology, 108, 400–Klein, R. G., Abikoff, H., Klass, E., Ganeles, D., Seese,

L. M., & Pollack, S. (1997). Clinical efficacy of 411.Lu, R., Ho, S., Huang, H., & Lin, Y. (1988). The speci-methylphenidate in conduct disorder with and without

attention deficit hyperactivity disorder. Archives of ficity of the dexamethasone suppression test in endog-enous depressive patients. Neuropsychopharmacol-General Psychiatry, 54, 1073–1080.

Klin, A. (1994). Asperger syndrome. Psychiatric Clinics ogy, 1, 157–162.Lykken, D. T. (1995). The antisocial personalities. Mah-of North America, 3, 131–148.

Klin, A., Jones, W., Schultz, R., Volkmar, F., & Cohen, wah, NJ: Erlbaum.Lynam, D. R. (1996). The early identification of chronicD. (2002). Defining and quantifying the social pheno-

type of autism. American Journal of Psychiatry, 159, offenders: Who is the fledgling psychopath? Psycho-logical Bulletin, 120, 209–234.895–908.

Klin, A., Volkmar, F. R., Sparrow, S. S., Cicchetti, D. V., Lynam, D. R. (1998). Early identification of the fledglingpsychopath: Locating the psychopathic child in the& Roarke, B. P. (1995). Validity and neuropsycho-


current nomenclature. Journal of Abnormal Psychol- of major depression and conduct disorder. Journal ofAffective Disorders, 39, 123–126.ogy, 107, 566–575.

Lynam, D. R., Caspi, A., Moffitt, T. E., Wikstrom, P. H., Milich, R., Balentine, A. C., & Lynam, D. R. (2001).ADHD combined type and ADHD predominantly in-Loeber, R., & Novak, S. (2000). The interaction be-

tween impulsivity and neighborhood context on of- attentive type are distinct and unrelated disorders.Clinical Psychology Science and Practice, 8, 463–fending: The effects of impulsivity are stronger in

poorer neighborhoods. Journal of Abnormal Psychol- 488.Miller, J. N., & Ozonoff, S. (2000). The external validityogy, 109, 563–574.

MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, of Asperger disorder: Lack of evidence from the do-main of neuropsychology. Journal of Abnormal Psy-D. D. (2002). On the practice of dichotomization

of quantitative variables. Psychological Methods, 7, chology, 109, 227–238.Murphy, E. A. (1964). One cause? Many causes? The19–40.

Macmillan, N. A., & Creelman, C. D. (1990). Response argument from the bimodal distribution. Journal ofChronic Disease, 17, 301–324.bias: Characteristics of detection theory, threshold

theory and “nonparametric” indices. Psychological Nagin, D. S. (1999). Analyzing developmental trajecto-ries: A semiparametric, group-based approach. Psy-Bulletin, 107, 401–413.

Malt, B. (1993). Concept structure and category bound- chological Methods, 4, 139–157.Nagin, D. S., & Tremblay, R. E. (2001). Analyzing de-aries. In G. V. Nakamura, D. L. Medin, & R. Taraban

(Eds.), Categorization by humans and machines. The velopmental trajectories of distinct but related behav-iors: A group-based method. Psychological Methods,psychology of learning and motivation: Advances in

research and theory (Vol. 29, pp. 363–390). San 6, 18–34.Nunnally, J. C., & Bernstein, I. H. (1994). PsychometricDiego, CA: Academic Press.

Mayr, E. (1942). Systematics and the Origin of Species. theory (3rd ed.). New York: McGraw–Hill.Osterling, J. A., Dawson, G., & Munson, J. A. (2002).New York: Columbia University Press.

McBurnett, K., Pfiffner, L. J., & Frick, P. J. (2001). Early recognition of 1-year-old infants with autismspectrum disorders versus mental retardation. Devel-Symptom properties as a function of ADHD type: An

argument for continued study of sluggish cognitive opment and Psychopathology, 14, 239–251.Ozonoff, S., Rogers, S. J., & Pennington, B. F. (1991).tempo. Journal of Abnormal Child Psychology, 29,

207–213. Asperger’s syndrome: Evidence of an empirical dis-tinction from high-functioning autism. Journal ofMeehl, P. E. (1962). Schizotaxia, schizotypy, schizophre-

nia. American Psychologist, 17, 827–838. Child Psychology and Psychiatry, 32, 1107–1122.Panak, W. F., & Garber, J. (1992). Role of aggression,Meehl, P. E. (1973). MAXCOV-HITMAX: A taxonomic

search method for loose genetic syndromes. In Psy- rejection, and attributions in the prediction of depres-sion in children. Development and Psychopathology,chodiagnosis: Selected papers (pp. 200–224). Minne-

apolis, MN: University of Minnesota Press. 4, 145–165.Patterson, G. R., DeGarmo, D. S., & Knutson, N. (2000).Meehl, P. E. (1975). Hedonic capacity: Some conjectures.

Bulletin of the Menninger Clinic, 39, 295–307. Hyperactive and antisocial behaviors: Comorbid ortwo points in the same process? Development andMeehl, P. E. (1979). A funny thing happened to us on

the way to the latent entities. Journal of Personality Psychopathology, 12, 91–106.Raine, A., Venables, P. H., Dalais, C., Mellingen, K.,Assessment, 43, 563–581.

Meehl, P. E. (1989). Schizotaxia revisited. Archives of Reynolds, C., & Mednick, S. A. (2001). Early educa-tional and health enrichment at age 3–5 years is asso-General Psychiatry, 46, 935–944.

Meehl, P. E. (1990). Toward an integrated theory of ciated with increased autonomic and central nervoussystem arousal and orienting at age 11 years: Evi-schizotaxia, schizotypy, and schizophrenia. Journal of

Personality Disorders, 4, 1–99. dence from the Mauritius Child Health Project. Psy-chophysiology, 38, 254–266.Meehl, P. E. (1992). Factors and taxa, traits and types,

differences of degree and differences in kind. Journal Rajendar, G. K. (1996). The influence of positive andnegative wording and issue involvement on responsesof Personality, 60, 117–174.

Meehl, P. E. (1995). Bootstraps taxometrics: Solving the to Likert scales in marketing research. Journal of theMarket Research Society, 38, 235–246.classification problem in psychopathology. American

Psychologist, 50, 266–275. Regier, D. A., Narrow, W. E., Rae, D.S., Manderscheid,R. W., Locke, B. Z., & Goodwin, F. K. (1993). TheMeehl, P. E. (1996). MOXCOV pseudotaxonicity. Ameri-

can Psychologist, 51, 1184–1186. de facto mental and addictive disorders service sys-tem. Epidemiologic Catchment Area prospective 1-Meehl, P. E. (1999). Clarifications about taxometric

method. Applied and Preventive Psychology, 8, 165– year prevalence rates of disorders and services. Ar-chives of General Psychiatry, 50, 85–94.174.

Meehl, P. E. (2001). Comorbidity and taxometrics. Clini- Richters, J. E. (1997). The Hubble hypothesis and the de-velopmentalist’s dilemma. Development and Psycho-cal Psychology Science and Practice, 8, 507–519.

Meehl, P. E., & Yonce, L. J. (1994). Taxometric analysis: pathology, 9, 193–229.Richters, J. E., & Cicchetti, D. (1993). Mark Twain meetsI. Detecting taxonicity with two quantitative indica-

tors using means above and below a sliding cut DSM-III-R: Conduct disorder, development, and theconcept of harmful dysfunction. Development and(MAMBAC procedure). Psychological Reports, 74,

1059–1274. Psychopathology, 5, 5–29.Richters, J. E., & Hinshaw, S. P. (1999). The abductionMeehl, P. E., & Yonce, L. J. (1996). Taxometric analyses

II. Detecting taxonicity using covariance of two quan- of disorder in psychiatry. Journal of Abnormal Psy-chology, 108, 438–445.titative indicators in successive intervals of a third in-

dicator. Psychological Reports, 78, 1091–1227. Robins, E., & Guze, S. B. (1970). Establishment of diag-nostic validity in psychiatric illness: Its application toMeller, W. H., & Borchardt, C. M. (1996). Comorbidity

T. P. Beauchaine526

schizophrenia. American Journal of Psychiatry, 126, Statistica [Computer software]. (1998). Tulsa, OK: Stat-soft.983–987.

Stevens, S. S. (1951). Handbook of experimental psychol-Rogosa, D., Brandt, D., & Zimowski, M. (1982). Aogy. New York: Wiley.growth curve approach to the measurement of change.

Strube, M. J. (1989). Evidence for the type in Type APsychological Bulletin, 92, 726–748.behavior: A taxometric analysis. Journal of Personal-Rosch, E., & Lloyd, B. (1978). Cognition and categoriza-ity and Social Psychology, 56, 972–987.tion. New York: Wiley.

Szasz, T. (2000). Second commentary on “Aristotle’sRuma, P. R., Burke, R. V., & Thompson, R. W. (1996).function argument.” Philosophy, Psychiatry, and Psy-Group parent training: Is it effective for children ofchology, 7, 3–16.all ages? Behavior Therapy, 27, 159–169.

Tyrka, A., Haslam, N., & Cannon, T. D. (1995). Detec-Rutter, M., Dunn, J., Plomin, R., Simonoff, E., Pickles,tion of a longitudinally-stable taxon of individuals atA., Maughan, B., Ormel, J., Meyer, J., & Eaves, L.risk for schizophrenia spectrum disorders. In A.(1997). Integrating nature and nurture: ImplicationsRaine, T. Lencz, & S. A. Mednick (Eds.) Schizotypalof person–environment correlations and interactionspersonality disorder (pp. 168–191). New York: Cam-for developmental psychopathology. Developmentbridge University Press.and Psychopathology, 9, 335–364.

Tyrka, A. R., Cannon, T. D., Haslam, N., Mednick, S. A.,Rutter, M., & Sroufe, L. A. (2000). Developmental psy-Schulsinger, F., Schulsinger, H., & Parnas, J. (1995).chopathology: Concepts and challenges. DevelopmentThe latent structure of schizotypy: I. Premorbid indi-and Psychopathology, 12, 265–296.cators of a taxon in individuals at risk for schizophre-Saal, F. E., Downey, R. G., & Lahey, M. A. (1980). Rat-nia-spectrum disorders. Journal of Abnormal Psychol-ing the ratings: Assessing the psychometric quality ofogy, 104, 173–183.rating data. Psychological Bulletin, 88, 413–428.

Volkmar, F. R., Klin, A., & Pauls, D. (1998). NosologicalSemin, G. R., & Rosch, E. (1981). Activation of bipolarand genetic aspects of Asperger syndrome. Journal ofprototypes in attribute inferences. Journal of Experi-Autism and Developmental Disorders, 28, 457–463.mental Social Psychology, 17, 472–484.

Volkmar, F. R., Klin, A., Schultz, R. T., Rubin, E., &Shekim, W. O., Dekirmenjian, H., Chapel, J. L., & Davis,Bronen, R. (2000). Asperger’s disorder. AmericanJ. M. (1982). Effects of d-amphetamine on urinaryJournal of Psychiatry, 157, 262–267.metabolites of dopamine and norepinephrine in hy-

Volkmar, F. R., Klin, A., Siegel, B., Szatmari, P., Lord,peractive boys. American Journal of Psychiatry, 139,C., Campbell, M., Freeman, B. J., Cicchetti, D. V.,485–488.Rutter, M., Kline, W., Buitelaar, J., Hattab, Y., Fom-Shekim, W. O., Sinclair, E., Glaser, R., Horwitz, E., Ja-bonne, E., Fuentes, J., Werry, J., Stone, W., Kerbesh-vaid, J., & Bylund, D. B. (1987). Norepinephrine andian, J., Hoshino, Y., Bregman, J., Loveland, K., Szy-dopamine metabolites and educational variables inmanski, L., & Towbin, K. (1994). Field trial forboys with attention deficit disorder and hyperactivity.autistic disorder in DSM-IV. American Journal ofJournal of Child Neurology, 2, 50–56.Psychiatry, 151, 1361–1367.Simon, D., Pham, L. B., Le, Q. A., & Holyoak, K. J.

Wakefield, J. C. (1992). The concept of mental disorder:(2001). The emergence of coherence over the courseOn the boundary between biological facts and socialof decision making. Journal of Experimental Psychol-values. American Psychologist, 47, 373–388.ogy: Learning, Memory, and Cognition, 27, 1250–

Wakefield, J. C. (1993). Limits of operationalism: A cri-1260.tique of Spitzer and Endicott’s (1978) proposed oper-Skilling, T. A., Quinsey, V. L., & Craig, W. M. (2001).ation criteria for mental disorder. Journal of Abnor-Evidence of a taxon underlying serious antisocial be-mal Psychology, 102, 160–172.havior in boys. Criminal Justice and Behavior, 28,

Wakefield, J. C. (1997). When is development disor-450–470.dered? Developmental psychopathology and the

Smalley, S. L., Bailey, J. N., Palmer, C. G., Cantwell, harmful dysfunction analysis of mental disorder. De-D. P., McGough, J. J., Del-Homme, M. A., Asarnow, velopment and Psychopathology, 9, 269–290.J. R., Woodward, J. A., Ramsey, C., & Nelson, S. F. Wakefield, J. C. (1999). Evolutionary versus prototype(1998). Evidence that the dopamine D4 receptor is analysis of the concept of disorder. Journal of Abnor-a susceptibility gene in attention deficit hyperactivity mal Psychology, 108, 374–399.disorder. Molecular Psychiatry, 3, 427–430. Wakefield, J. C., Pottick, K. J., & Kirk, S. A. (2002).

Smith, E. E. (1995). Concepts and categorization. In E. Should the DSM-IV diagnostic criteria for conductSmith & D. Osherson (Eds.), Thinking: An invitation disorder consider social context? American Journal ofto cognitive science. (2nd ed., Vol. 3, pp. 3–33). Psychiatry, 159, 380–386.Cambridge, MA: MIT Press. Waldman, I. D., & Lilienfeld, S. O. (2001). Applications

Sonuga–Barke, E. J. S. (1998). Categorical models of of taxometric methods to problems of comorbidity:childhood disorder: A conceptual and empirical anal- Perspectives and challenges. Clinical Psychology Sci-ysis. Journal of Child Psychology and Psychiatry, 39, ence and Practice, 8, 520–527.115–133. Waller, N. G., & Meehl, P. E. (1998). Multivariate taxo-

Sroufe, L. A. (1997). Psychopathology as an outcome of metric procedures: Distinguishing types from continua.development. Development and Psychopathology, 9, Newbury Park, CA: Sage.251–268. Waller, N. G., Putnam, F. W., & Carlson, E. B. (1996).

Sroufe, L. A., & Rutter, M. (1984). The domain of devel- Types of dissociation and dissociative types: A taxo-opmental psychopathology. Child Development, 55, metric analysis of dissociative experiences. Psycho-17–29. logical Methods, 1, 300–321.

Sroufe, L. A., & Waters, E. (1977). Attachment as an Waller, N. G., & Ross, C. A. (1997). The prevalence andorganizational construct. Child Development, 48, the biometric structure of pathological dissociation in

the general population: Taxometric and behavior ge-1184–1199.


netic findings. Journal of Abnormal Psychology, 106, Widiger, T. A. (2001). What can be learned from taxome-tric analyses? Clinical Psychology Science and Prac-499–510.

Waters, E., & Beauchaine, T. P. (2003). Are there really tice, 8, 528–533.Williamson, D. A., Womble, L. G., Smeets, M. A. M.,patterns of attachment? Comment on Fraley and

Spieker. Developmental Psychology, 39, 417–422. Netemeyer, R. G., Thaw, J. M., Kutlesic, V., &Gleaves, D. H. (2002). Latent structure of eating dis-Webster–Stratton, C., & Hammond, M. (1997). Treating

children with early-onset conduct problems: A compar- orders symptoms: A factor analytic and taxometric in-vestigation. American Journal of Psychiatry, 159,ison of child and parent training interventions. Journal

of Consulting and Clinical Psychology, 65, 93–109. 412–418.Woodward, S. A., Lenzenweger, M. F., Kagan, J., Snid-Webster–Stratton, C., Reid, M. J., & Hammond, M.

(2002). Social skills and problem solving training for man, N., & Arcus, D. (2000). Taxonic structure ofinfant reactivity: Evidence from a taxometric perspec-children with early-onset conduct problems: Who

benefits? Journal of Child Psychology and Psychia- tive. Psychological Science, 11, 296–301.try, 42, 943–952.

Taxometrics and developmental psychopathology 2003.pdf · 2003. 8. 3. · Taxometrics and developmental psychopathology 503 cally more interested in the processes through Western

Documents