Development and Psychopathology, 9 (1997), 193-229Copyright 0 1997 Cambridge University PressPrinted in the United States of America

The Hubble hypothesis and thedevelopmentalist’ s dilemma

JOHN E. RICHTERSNational Institute of Mental Health

AbstractDevelopmental psychopathology stands poised at the close of the 20th century on the horns of a major scientificdilemma. The essence of this dilemma lies in the contrast between its heuristically rich open system concepts on theone hand, and the closed system paradigm it adopted from mainstream psychology for investigating those models onthe other. Many of the research methods, assessment strategies, and data analytic models of psychology’s paradigmare predicated on closed system assumptions and explanatory models. Thus, they are fundamentally inadequate forstudying humans, who are unparalleled among open systems in their wide ranging capacities for equifinal andmultifinal functioning. Developmental psychopathology faces two challenges in successfully negotiating thedevelopmentalist’s dilemma. The first lies in recognizing how the current paradigm encourages research practicesthat are antithetical to developmental principles, yet continue to flourish. I argue that the developmentalist’sdilemma is sustained by long standing, mutually enabling weaknesses in the paradigm’s discovery methods andscientific standards. These interdependent weaknesses function like a distorted lens on the research process byvariously sustaining the illusion of theoretical progress, obscuring the need for fundamental reforms, and bothconstraining and misguiding reform efforts. An understanding of how these influences arise and take their tollprovides a foundation and rationale for engaging the second challenge. The essence of this challenge will be findingways to resolve the developmentalist’s dilemma outside the constraints of the existing paradigm by developingindigenous research strategies, methods, and standards with fidelity to the complexity of developmental phenomena.

Introduction

Within minutes of receiving their first datatransmission from the Hubble Space Tele-scope in 1990, earthbound astronomers knewthat something had gone terribly wrong. Thelong-awaited launching of the Hubble hadtaken place several months earlier followingdecades of planning, instrumentation design,and construction. This was the eagerly antici-pated, first, and flagship mission of the Na-tional Aeronautics and Space Administra-tion’s Great Observatories program, and

The views expressed in this paper are those of the authorand are not to be interpreted as official views of the U.S.Department of Health and Human Services, the NationalInstitutes of Health, or the National Institute of MentalHealth.

Address correspondence and reprint requests to: Dr.J. E. Richters, Child & Adolescent Disorder Branch,Room l8C-17, NIMH, 5600 Fishers Lane, Rockville, MD20857. E-mail: jrichter@NIH.gov.

astronomy’s expectations were high; the Hub-ble mission had been touted in advance as thegreatest stride in astronomy since Galileo’stelescope spied the moons of Jupiter (NewYork Times, 1990). Its scientific objective wasliterally to look backward in time and provideastronomers with high-resolution images ofthe universe never before seen-images ofgalaxies that ceased existing billions of yearsago. To capture these images, the Hubble hadbeen equipped with the most sophisticatedand expensive optical system ever developedand launched into orbit 200 miles above thedistorting influence of the earth’s atmosphere.Despite its technological sophistication theHubble was transmitting blurred images ofthe universe back to earth that were no clearerthan those produced by ground-based tele-scopes. As diagnostic tests would soon reveal,the source of the problem was a spherical ab-erration deep in the Hubble’s intricate optical

193

194 Richters

system. Human error had resulted in the Hub-ble’s 94-in. primary mirror being ground tothe wrong prescription, rendering it near-sighted. The Hubble, in effect, had beenlaunched into orbit on the most significantand expensive exploratory mission in the his-tory of astronomy with the wrong pair ofglasses.

The exacting details of the spherical aber-ration diagnosis provided the necessary foun-dation for an intensive program of recoveryefforts that culminated 42 months later in acomplete restoration of the Hubble’s vision.The Hubble recovery effort is best remem-bered for its televised images of the final dra-matic moments in which astronauts floating inspace lowered a corrective optics package intoHubble while circling the earth at 20,000miles/hr. The defining scientific moment ofthe Hubble rescue, however, took place farfrom public view when astronomers first rec-ognized that something was wrong. Althoughit passed routinely and without fanfare, themoment was decisive to the ultimate successof the Hubble rescue in the sense that a mis-judgment at that early juncture could have re-sulted in a radically different course of actionand outcome. A misinterpretation of the Hub-ble data as accurate, for example, could haveresulted in a protracted period of scientificactivity focused on the wrong phenomena.Astronomers, in effect, would have beenstudying illusory phenomena actually createdby the Hubble’s lens flaw in the mistaken be-lief that they were studying inherent complex-ities of the structures and motions of distantgalaxies. Alternatively, the Hubble missionmight have been compromised indefinitely ifthe distortion had been recognized but inade-quately understood, or adequately understoodbut deemed not correctable at its source. Anynumber of missteps in the process of recog-nizing and diagnosing the Hubble’s visionproblem could have resulted in a prolongeddiversion of scientific resources away fromthe main objectives of the Hubble exploratorymission.

vanced knowledge base and their scientificquestions were carefully anchored in and de-signed to extend that base. In addition, bene-fitting from advances in physics, telemetry,optics, and communications, they were rely-ing on sophisticated instruments capable ofyielding sufficiently precise and detailed ob-servations to address those questions. Thespecifications for Hubble’s optical system, forexample, called for its primary mirror to beground to a smoothness which, if scaled to thewidth of the continental United States, wouldresult in no mountain or valley varying fromthe mean surface by more than 21/2 in. It ishardly surprising that scientists working atthis level of detail would be exquisitely sensi-tive to the presence and meaning of discrep-ancies between their expectations and obser-vations.

Psychology’s Progress as a Science

It is difficult to imagine that astronomerswere ever at serious risk for these alternativescenarios. The Hubble scientists were operat-ing on the basis of a well established, ad-

It would be an understatement to say that thescience of psychology does not yet possessanything of comparable precision in its theo-ries, research instruments, or methods. Nor isit yet able to proceed with surefooted con-fidence in its ability to discriminate suc-cessfully between facts and artifacts, flawedtheory and flawed data, real and illusory phe-nomena. To be sure, psychology has accom-plished much during its first full century as ascience. Its research activities have uncovereda wealth of previously unknown facts aboutmyriad aspects of human sensation, percep-tion, learning, language, cognition, emotion,motivation, personality, and social function-ing and development (Gerstein, Luce,Smelser, & Sperlich, 1988). Moreover, thesefacts, in combination with the language, con-cepts, and theories of psychology have had animmeasurable transforming influence on howindividuals in Western society think aboutthemselves and others. Despite these gains, itis generally acknowledged that psychologyhas fallen far short of its own expectations aswell as society’s for scientific progress in the20th century. In commenting on widespreadpublic perceptions of psychology’s practicalscientific contributions to society, psycholo-gist-researcher Ronald Fox reluctantly of-

Hubble hypothesis 195

fered the following assessment in his 1994presidential address before the American Psy-chological Association (Fox, 1996, p. 779):

We have not succeeded in demonstrating the publicbenefit that is derived from investments in psycho-logical research. Social and behavioral sciences re-search funding has been under attack for severalyears now. From my discussions with legislatorsand government officials, it is clear that one of thereasons for the attacks is the belief among many ofthose who allocate the funds that considerablemoney has been invested in our research with littleto show for it in terms of practical help for soci-ety’s problems.

Fox and many other psychologists findsuch assessments unduly harsh and overstatedbecause they give insufficient weight to psy-chology’s genuine contributions. Nonetheless,most also admit that those contributions palein comparison to the expectations and enthusi-asm of the field following World War II (Sar-ason, 1981). Even more important, there havebeen growing concerns within psychology’sscientific community about its difficulties inestablishing sustained theoretical progress inmany areas of research. In a sobering assess-ment almost two decades ago, Meehl charac-terized the natural history of theories in so-called soft psychology as a sequence begin-ning with initial enthusiasm, followed by re-peated attempts to demonstrate a theory’smerits, disillusionment about negative find-ings, growing bafflement about inconsistentand unreplicated results, and multiple resortto ad hoc excuses. In the end. according toMeehl, researchers eventually lose interestand move on to other ideas as their theories“ . . . suffer the fate that General MacArthurascribed to old generals-They never die.they just slowly fade away” (Meehl, 1978, p.807). Indeed. ideas and theories that arewidely accepted and embraced by one genera-tion of psychologists are routinely discardedor ignored by the next. Even within a givengeneration, dramatic findings that promise tomove the knowledge base forward frequentlyhave a half-life of only a few years beforethey fade from memory. Despite its accom-plishments, psychology has not yet producedanything comparable to the established theo-

ries and laws of the natural sciences in termsof their explanatory power or capacity to yieldprecise and reliable predictions.

Observations about psychology’s compara-tively slow progress are often met with re-minders about its scientific youth and thecomplexities of its mission. Not only is psy-chology considerably younger as a formal dis-cipline than most established sciences, butmany scholars agree that psychological phe-nomena are inherently more complex andchallenging to study than most phenomena inthe physical and biological sciences. Psycho-logical functioning and behavior is a productof the most complex and sophisticated organknown to science; the human brain is actuallydesigned by nature to write and revise its soft-ware across the life span (Lykken, 1991).Moreover, for a variety of obvious practicaland ethical reasons psychology is disadvan-taged in comparison to many of the naturalsciences (astronomy not included) by not be-ing able to experimentally manipulate someof its most important phenomena. This is es-pecially true of research domains once classi-fied by Cronbach as “correlational” psychol-ogy, not necessarily in the statistical sense,but because they must rely on observing, or-ganizing, and analyzing phenomena as theyare found in nature (Cronbach, 1957). Fur-thermore, psychology has been disadvantagedby not being able to capitalize on the ad-vances and technological accomplishments ofphysical and biological sciences, and has thusbeen hampered for much of its history by lim-itations in instrumentation and measurementtechnology. Finally. humans are reactive inpowerful, complex, and not well understoodways to the knowledge or suspicion that theyare being studied. In contrast to the objectsof interest to the physical sciences, humanssometimes misunderstand, lie, forget, distort,confabulate, mischaracterize, misjudge, proj-ect. and deny in reaction to being questionedand/or observed by researchers.

Persistent paradigm problems

There have been other more troubling voicesover the years calling attention to fundamentalinadequacies in psychology’s scientific ap-

6196 Richters

proach to these challenges. In his recent at-tempt to answer the frequently asked question“What’s wrong with psychology anyway?”David Lykken concluded “ ... psychologyisn’t doing very well as a scientific discipline,and something seems to be wrong somewhere. . . due in part to a defect in our research tra-dition . . . ” (Lykken 1991. p. 4). Lykken isone of many distinguished scholars who haveargued that psychology’s basic problem formore than a century has been the absence ofa coherent, agreed upon paradigm’ for defin-ing the boundaries, objectives, phenomena,and methodology of its science. As philoso-pher of science Ernest Nagel noted almost 40years ago, the social and behavioral sciencesgenerally lack anything approaching the al-most complete unanimity commonly found inthe natural sciences about “ . . . what are mat-ters of established fact, what are the reason-ably satisfactory explanations (if any) for theassumed facts, and what are some of the validprocedures in sound inquiry” (Nagel, 1961,p. 448).

Thomas Kuhn, in his classic analysis ofscientific revolutions, described the naturalhistory of sciences in terms of a cyclical pat-tern of stages or periods (Kuhn, 1970). Eachscience begins in a preparadigm period char-acterized by the proliferation of competingschools of thought, a diversity of fact gather-ing activities, and minimal cumulative theo-retical progress. Eventually, when a particularviewpoint or school ascends to dominance, itserves as the shared paradigm for a period ofnormal science characterized by the relativeabsence (or silence) of competing schools, in-tensive coordinated research activity focusedon solving puzzles defined by the paradigm,and resulting theoretical progress. Initially, asanomalies emerge during periods of normalscience, they are either shelved for a periodor dismissed as unimportant as scientists con-centrate on puzzles that can be solved suc-cessfully within the paradigm. Over time,however, as the balance between successful

1. I use the term “paradigm” throughout this paper as ageneric reference to the scientific practices and proce-dures characteristic of psychology’s mainstream re-search community.

problem solving and important unexplainedanomalies shifts. insecurity eventually sets inabout the viability of the paradigm. This, inturn, gives rise to a loosening of the para-digm’s restrictions, an active consideration ofalternatives, and the reemergence of compet-ing schools of thought. When a replacementparadigm is finally adopted, the process be-gins again with a new period of normal sci-ence.

Although Kuhn’s thesis about the revolu-tionary nature of paradigm replacement wasvigorously challenged and eventually soft-ened by Kuhn himself (Kuhn, 1977), his basicdistinction between periods of normal and cri-sis science is a useful one. So also is his de-scription of the initial preparadigm period infledgling fields of science, such as chemistrybefore Boyle, astronomy before Copernicus,electricity before Franklin, and physical op-tics before Newton. Prior to the Copernicanrevolution, for example, astronomers found itnecessary to make so many ad hoc adjust-ments to the Ptolemaic system that its com-plexity eventually far exceeded its accuracy.Similarly, prior to Newton, there was no gen-erally accepted view about the nature of light.There were numerous competing schools ofthought based on one or another variant ofEpicurean, Aristotelian, or Platonic theory,each able to explain particular aspects of lightphenomena but not others. Each also claimedthat the phenomena it could best explain werepreeminently significant, offering ad hoc ex-planations for-or dismissing the theoreticalsignificance of-those it could not (Kuhn,1970).

The parallels between these early sciencesand late 20th century psychology have not es-caped the attention of scholars. Kuhn himselfwas drawn initially to his study of scientificparadigms by a curiosity about why, in con-trast to fields such as physics, astronomy, andbiology, disciplines like psychology seemedperpetually engaged in controversies about the“ . . . fundamentals of their science” (Kuhn,1970, p. viii). Nagel in a strikingly similarbut earlier observation about psychology,noted with puzzlement that in most fields ofscience disagreements over fundamentals tendto occur not in the mainstream but on the

Hubble hypothesis ’ 197

frontiers of scientific activity. Kuhn’s subse-quent explanation, of course, was that psy-chology is still in its initial preparadigm pe-riod of development. Now, almost 30 yearslater, psychology’s reputation as a prepara-digm science is so taken for granted by mostscholars that the claim is commonly assertedas self-evident without elaboration (Bevan &Kessel, 1994; Horgan, 1996; Koch & Leary,1992; Lakatos, 1977; McGuire, 1983; Robin-son, 1984).

One of the defining characteristics of pre-paradigm science is a proliferation of ideasabout exactly what is wrong and how to makeit right. Psychology has been no exception.Critics dating back to its founding have dif-fered considerably both in their explanationsfor its difficulties and in their proposed solu-tions, The common core complaint underlyingmany criticisms has been that psychology’sfounders erred strategically by modeling theirscience after the 19th century natural sci-ences. The narrow emphasis of that paradigmon passive physical objects, Humean notionsof causality, reductionistic explanations, andmechanistic laws, according to critics, ren-dered it unsuitable for a coherent science ofpsychology. In contrast to the objects of inter-est to classical physics, human beings are liv-ing, active, interactive, reactive, and adaptiveorganisms. Thus, a science of psychology re-quires a fundamentally different approach toscientific investigation based on what Sig-mund Koch would later call “indigenousmethodologies” capable of reflecting and illu-minating the richness of human phenomena(Bevan & Kessel, 1994; Koch, 1959, 1961).

Ironically, psychology’s embrace of thisparadigm took place at about the same timethat the physical sciences were abandoning itslimitations. The classical physics model isnow recognized by the natural sciences them-selves as narrowly useful only in the study ofclosed physical systems and under a limitedrange of circumstances (Brandt, 1973). Inmodern physics, for example, the physicalworld is no longer viewed as similar at alllevels; there is a widespread recognition ofthe need to employ different conceptual strat-egies and explanatory models at different lev-els of analysis; and the earlier, narrow empha-

sis on efficient causality has given way tomeson fields, gravitational fields, and mag-netic fields, none of which requires or is ex-pected to be accounted for by traditional mod-els of Humean causality.

The more immediate concern of psycholo-gy’s founders, though, was to establish its in-dependence from philosophy and secure itsstatus as a genuine field of science. And, inpursuing this objective, they were influencedstrongly by a sentiment expressed by JohnStuart Mill, who asserted that “ . . . the back-ward state of the Moral Sciences can only beremedied by applying to them the methods ofPhysical Science, duly extended and general-ized” (quoted in Leahey, 1987, p. 4). Thus,rather than risking an investment in the devel-opment of new methodologies, psychologyembraced explanatory models, discoverymethods, and research strategies that had al-ready proven successful in the established19th century natural sciences. The predictableand enduring consequence of this decision,according to critics, has been to limit psychol-ogy’s coherence, credibility, and potential asa science (Bevan, 1991; Bevan & Kessel,1994; Smith & Torrey, 1996).

Although many of psychology’s criticshave agreed in general about the inadequaciesof its mechanistic foundations, there has beenlittle consensus about a remedy. Within eachsucceeding generation psychologists have ini-tiated isolated efforts to champion particularresearch methods or strategies aimed at over-coming one or more of the paradigm’s per-ceived inadequacies. Occasionally, such ef-forts have resulted in the emergence ofspecialized subdisciplines and/or new scien-tific journals devoted to particular methods orsubstantive issues. In the main, however, suchefforts have not translated into anything ap-proximating the kinds of fundamental reformsthat would be required to replace psycholo-gy’s paradigm with one based on more indig-enous methodologies and research strategieswith greater fidelity to human psychologicalphenomena (Koch, 1959, 1961). On the con-trary, psychology has become increasinglyfragmented into narrowly defined subdisci-plines, each pursuing its objectives in relativeisolation from the others. As Copernicus de-

198 Richters

scribed the state of astronomy before the Cop- throughout the natural sciences. Those whoernican revolution, “ . . .. it is as though an art- align themselves with the developmental psy-ist were to gather the hands, feet, head, and chopathology perspective tend to share theother members for his images from diverse objective of gaining a process-level under-models, each part expertly drawn, but not re- standing of how, why, and in what ways indi-lated to a single body, and since they in no vidual differences in normal and abnormal so-way match each other, the result would be cial, emotional, cognitive, and behavioralmonster rather than man” (quoted in Kuhn, development emerge, interact, and develop1957, p. 83). Moreover, late 20th century psy- across the life span. Most also recognize thechology continues to be characterized by necessity of employing organizational princi-widespread dissatisfaction with the existing ples and systems concepts such as equifinal-paradigm yet little consensus about how to ity, multifinality, epigenesis, and transactionalproceed (Lykken, 1991). influences in the pursuit of these objectives.

Developmental Psychopathology

Some of the most integrative and progressivereform efforts in recent decades have arisennot with the field of psychology per se, butin the hybrid discipline of developmentalpsychopathology (Cicchetti, 1990, 1993; Cic-chetti & Cohen, 1995a, 1995b; Rutter & Gar-mezy, 1983; Sroufe & Rutter, 1984). Histori-cally, developmental scientists have played aprominent role in discussion of psychology’sparadigm (Baldwin, 1895; Bronfenbrenner,1979; Hinde, 1992; Kagan, 1992; Lewin,1931a, 1931b; Magnusson, 1985; Overton &Horowitz, 199 1; Overton, 1984a; Sameroff &Chandler, 1975). Developmental psychopath-ology evolved as a discipline, in fact, out of along history of progressive efforts to over-come the limited disciplinary boundaries andnarrowly focused paradigms of traditional so-cial, psychological, and biological sciences(Cicchetti, 1984). In terms of breadth anddepth, it stands in sharp contrast to the in-creasingly narrow foci of psychology’s spe-cialty areas and subdisciplines. Its broadly de-fined conceptual framework accommodates,benefits from, encourages, facilitates, andcontributes to integrative research across thefull array of social and biological sciences inthe service of a developmental understandingof human development and functioning.

Also reflecting its progressiveness, devel-opmental psychopathology’s emphases onopen system concepts and organismic, ho-listic, organizational, and transactional ap-proaches to research parallel the dramatic20th century ascendancy of these frameworks

The first challenge in any revolution or re-form movement lies in building consensusabout the need for change. The second lies infinding ways to translate that consensus intoaction. Viewed from this perspective, devel-opmental psychopathology has been remark-ably successful in establishing the necessaryfoundation for reform. The heuristic value ofits interdisciplinary, integrative framework isnow widely recognized throughout the humansciences. During the past two decades alone ithas fueled the establishment of new scientificjournals, special sections in numerous existingscientific journals, and a plethora of bookchapters, edited volumes, and books. In addi-tion, its basic principles were embraced by the1989 Institute of Medicine report Research onChildren and Adolescents with Mental, Be-havioral, and Developmental Disorders (In-stitute of Medicine, 1989), and in the morerecent National Institute of Mental Health’s1990 National Plan for Research on Childand Adolescent Mental Disorders (NationalAdvisory Mental Health Council, 1990). Forall of these reasons, developmental psycho-pathology seems poised to define a new cen-ter of gravity for many areas of psychologicalscience in the 21st century.

Developmentalist’s dilemma

The major dilemma facing developmentalpsychopathology in realizing its potential liesin the contrast between its heuristically richopen system concepts and developmentalmodels on the one hand, and the closed sys-tem paradigm on which it relies for investigat-ing those models on the other. At present,

Hubble hypothesis 199

most empirical research in developmentalpsychopathology is based largely on the para-digm it adopted from mainstream psychology.The research methods, assessment strategies,and data analytic models of this paradigm,however, remain wedded to the closed systemassumptions and explanatory models of 19thcentury science. It therefore emphasizes thestudy of variables over individuals, the mod-eling of complex interactions among vari-ables, and the search for invariant causalexplanations to account for phenotypicallysimilar patterns of overt functioning across in-dividuals. These strategies make perfect sensefor the study of within-class structurally ho-mogenous objects functioning in closed phys-ical systems. But they make little sense asstrategies for studying human functioning anddevelopment, given that humans are unparal-leled among open systems in their wide rang-ing capacities for equifinal and multifinalfunctioning (Allport, 1960; Cairns, 1986; Lewin,193 la; Lykken, 1991; Valsiner, 1986a; vonBertalanffy, 1968).

Kurt Lewin criticized mainstream psychol-ogy’s research paradigm on these groundsover 60 years ago, prior to the formal intro-duction of open system concepts and princi-ples such as equifinality and multifinality(Lewin, 193la, 1931b). Research in psychol-ogy was already characterized at the time bycommon practices of aggregating subjects onthe basis of phenotypic similarities, assessingpsychological and environmental variables inisolation from each other and from their con-texts, and submitting the data to group-levelstatistical analysis. Lewin warned that aver-ages obtained in this way on large samples ofchildren rendered virtually hopeless the possi-bility of a meaningful understanding of indi-vidual differences in human functioning: “Theconcepts of average child and average envi-ronment have no utility whatever for the in-vestigation of dynamics . . . An inferencefrom the average to the particular case is ...impossible” (Lewin, 1931b, p. 95). GordonAllport made similar arguments three decadeslater with specific reference to open systemsin his classic article entitled The Open Systemin Personality Theory (Allport, 1960). And,with increasing frequency since that time, se-

nior developmental scholars have raised simi-lar concerns, arguing for the need to developresearch methods and data analytic strategieswith greater fidelity to developmental phe-nomena (Cairns, 1986; Emde, 1994; Hinde,1992; Hinde & Dennis, 1986; Kagan, 1992;Magnusson, 1985; Radke-Yarrow & Sher-man, 1990; Sameroff & Chandler, 1975; Val-siner, 1986a).

Despite this long-standing recognition, analternative paradigm for studying develop-mental phenomena has yet to emerge and gainwidespread acceptance in the scientific com-munity. Consequently, developmental psy-chopathologists necessarily rely for their em-pirical research on the established researchmethods and strategies of psychology’s exist-ing paradigm. This confronts them with thechallenges of studying an organismic modelof development with research methods andstrategies based on mechanistic model. Inevi-tably, the incompatibilities between thesemodels translate into a major dilemma as re-searchers must choose between adhering todevelopmental principles and frameworks orconforming to research practices engenderedand reinforced by the existing paradigm.

A review of contemporary developmentalpsychopathology research reveals that the de-velopmentalist’s dilemma is often resolved infavor of existing methods. Developmentalprinciples embraced in the Introduction sec-tions of empirical manuscripts are routinelyignored and/or violated in Methods and Re-sults sections, only to resurface in Discussionsections. There is often clear evidence of de-velopmental sophistication in the thinkingleading to and following empirical research.But much of this sophistication is sacrificedin the research process itself as developmentalprinciples and frameworks are transduced intoclosed system research strategies. Thus, withthe exception of new theoretical constructs,greater emphasis on psychometric sophistica-tion, and more elaborate data analytic models,the contemporary empirical literature of de-velopmental psychopathology continues to re-flect many of the same practices lamented byLewin over 60 years ago. Subjects who arerecruited on the basis of background factorsand/or phenotypic similarities in behavior are

200 Richters

presumed a priori to be homogenous with re-spect to the causal factors underlying theirfunctioning; isolated parcels of disembodiedinformation are collected about each subject;and data analytic strategies that presupposewithin-sample structural homogeneity are em-ployed in the service of identifying a unitarycausal structure to account for individual dif-ferences in the phenomena of interest.

Although these widespread practices arecontraindicated by the impressive human ca-pacities for equifinal and multifinal function-ing, they are perpetuated by the illusion ofprogress created by the paradigm’s weak sci-entific standards. Those standards have beencriticized harshly by numerous statisticians,psychologists, and philosophers of science aspseudoscientific and misleading (Meehl, 1990).Although these flawed standards are seldomdiscussed in the developmental literature, theyplay a powerful role in blinding researchers tothe indispensable negative feedback abouttheir theories and ideas that is the self-correct-ing essence of scientific progress. Thoseflawed standards, in turn, are virtually re-quired by inadequacies in the paradigm’sclosed system methods and practices. I arguebelow that an examination of these interde-pendent weaknesses reveals much about whythe paradigm requires fundamental reformsand why those reforms have not been under-taken.

Overview

The remainder of this paper is devoted to anexamination of the developmentalist’s di-lemma and a consideration of its implicationsfor paradigm reform in developmental psy-chopathology research. My point of departureis the premise that mutually enabling weak-nesses in the discovery and justification pro-cesses (Reichenbach, 1938) function like adistorted lens on the research process by vari-ously sustaining the illusion of theoreticalprogress, obscuring the need for fundamentalreforms, and both constraining and misguid-ing reform efforts. These undermining influ-ences are certainly not unique to the disciplineof developmental psychopathology; they areevident across the broad array of social, be-

havioral, and psychological sciences that relyfor their empirical research on psychology’sparadigm. Unlike many other disciplines,however, developmental psychopathologyfaces the genuine dilemma of having explic-itly embraced open system principles that arefundamentally incompatible with the para-digm’s closed system assumptions. Develop-mental psychopathology is actually at oddswith itself in the empirical arena by attempt-ing to study a complex open system withstrategies and methods of scientific discoverybased on mechanistic assumptions.

Developmental psychopathology faces twochallenges in successfully negotiating the de-velopmentalist’s dilemma. The first lies inrecognizing how and under what conditionsthe current paradigm encourages researchpractices that are antithetical to sound scienceand the developmental agenda. The secondchallenge lies in developing and adoptingstrategies for overcoming these inadequaciesand for translating open system principles anddevelopmental frameworks into equally so-phisticated research designs, methods, andpractices.

I begin by focusing on some of the funda-mental inadequacies of the existing discoveryparadigm for studying human functioningfrom an open system perspective. I then pres-ent a basic critique of the justification para-digm, summarizing the interrelated statistical,logical, and conceptual weaknesses for whichit has been so harshly criticized. I make noattempt to cover the already extensive criticalliteratures concerning either set of paradigmweaknesses. My purpose rather is to establishenough of a foundation for examining howthey enable and obscure each other’s under-mining influences in the practical context ofday to day research. I conclude by consideringthe implications of this analysis for undertak-ing a program of constructive reforms.

Psychology’s Paradigmatic Lenses

The problem of infidelity between develop-mental psychopathology’s heuristics and tra-ditional research methods can be conceptual-ized within Reichenbach’s classic frameworkas arising in the scientific context of discovery

Hubble hypothesis

(Reichenbach, 1938). These are the activitiesof a science that occupy the initial phases ofthe research process, ranging from basic de-cisions about what kinds of ideas and phe-nomena are considered worth studying, to de-cisions about how they are studied andanalyzed. The context of discovery, in brief,concerns how scientists go about generatingtheir ideas and evidence. The context of justi-fication, in contrast, concerns how scientistsgo about evaluating those ideas. Thus, justifi-cation activities occur later in the researchprocess and include the procedures and deci-sion processes through which a science evalu-ates the merits of its ideas.

Historically, discovery and justification is-sues have been discussed as separate para-digm problems in psychology, by differentauthors, for different audiences, and often indifferent publication outlets. Discovery issuesalways have figured prominently in the main-stream literature of psychology because oftheir obvious direct relevance to substantivequestions about the conceptual frameworks,theories, and research methods in particularresearch domains. These issues of method arealmost never discussed, however, with refer-ence to the justification process. Typically,they are viewed as problems that emerge andcan be remedied in the context of discovery.

Justification discussions have tended to ac-cumulate outside the literatures of particularsubstantive domains in more general forums.Like discovery issues, however, psychology’sscientific standards tend to be discussed as ab-stract conceptual problems that emerge andare to be resolved in the context of justifica-tion. They are almost never discussed withreference to specific processes of discovery.Even those who have discussed both types ofissues have tended to discuss them as rela-tively independent problems. Meehl’s classic1978 critique of the slow progress of so-calledsoft psychology, for example, is half devotedto cogent summary discussions of 15 complexdiscovery issues and problems facing psy-chology and half to issues of justification.Neither set of issues, however, is discussedwith reference to or within the context of theother (Meehl, 1978). In a similarly thoughtful,more recent analysis of psychology’s para-

201

digm problems Lykken also discussed discov-ery and justification issues in relative isola-tion from one another (Lykken, 1991).

Although discovery and justification activ-ities are conceptually different and temporallyseparated in the research process, they are notindependent. They are interdependent in thesense that decisions and activities in each con-text inevitably anticipate those in the other.Discovery decisions about how to conceptual-ize phenomena, frame research questions, de-sign studies, choose variables, assess subjects,and analyze data, necessarily anticipate thelogic, standards, and procedures by whichthey will be evaluated later in the context ofjustification. Similarly, the coherence of thejustification process hinges decisively on itsfidelity and goodness of fit with the assump-tions, strategies, and procedures of the discov-ery process. While this interdependence istrue in all areas of science, it is an especiallypowerful and negative one in psychology be-cause it exists as a dynamic between weak-nesses in the discovery and justification pro-cesses.

In the latest in his series of penetrating cri-tiques of psychology’s justification practices,Paul Meehl commented, “This is my last at-tempt to call attention to a methodologicalproblem of our field that I insist is not minorbut of grave import” (Meehl, 1990, p. 108).Meehl’s lament is a common one among psy-chology’s discovery and justification critics,who have expressed varying degrees of frus-tration, disbelief, puzzlement, disappointment,and concern about why long-standing weak-nesses in the paradigm have not given rise toconcentrated reform efforts. I argue belowthat the answer lies in the mutually enablinginterdependence between weaknesses in theparadigm’s discovery and justification pro-cesses. Each obscures the inadequacies of theother and imposes constraints on attemptedsolutions. Combined, these weaknesses func-tion like a distorted lens in the research enter-prise by sustaining the illusion of progress,obscuring the need for fundamental paradigmreforms, and both constraining and misdirect-ing isolated reform efforts.standing pressing problempsychological sciences. But

This is a long-throughout theit constitutes a

202 Richters

genuine dilemma for the discipline of devel-opmental psychopathology because of its ex-plicit commitment to open system conceptsthat are incompatible with psychology’sclosed system paradigm.

Context of discovery

Mechanistic and organismic world views. Ev-ery scientific discovery process is framed andshaped by a series of initial decisions aboutwhich phenomena are worth studying, whattheir essential natures are, how they should beconceptualized, and what strategies, proce-dures, and methods should be used to studythem. Many of these decisions, however, arebased on background assumptions that are soimplicit that the researcher is often unawarethat he or she is making consequential choices.They seem more like obvious choices thanlike important decisions over which the in-vestigator has control. Pepper characterizedscholarly approaches to knowledge in termsof major world hypotheses, each of which isassociated with a different set of core philo-sophical assumptions about the nature of thephysical world and thus how it can and shouldbe studied (Pepper, 1942). The mechanismand organicism world views are particularlyrelevant for present purposes because of theircentral significance to most critiques of psy-chology’s discovery paradigm.

During the 19th century the dominant worldview in the physical sciences was mechanism.Newtonian physics viewed the universe asconsisting of passive physical objects that in-teracted with one another in clockwork fash-ion, with energy as the causal link betweenobjects. The task of science was therefore tostudy individual objects, identify regularitiesamong patterns of actions and interactions be-tween them, and determine the general lawsthat accounted for them. To understand thecomplex workings of any given whole or sys-tem or machine, according to this view, oneneed only break it down into its individual,uniform components, examine the characteris-tics of each in isolation from the other, andadd them together. The whole, from a mecha-nistic viewpoint, is nothing more than an ad-ditive sum of its individual parts.

Organicism, which has gained increasingdominance during the 20th century, takes asits root metaphor the integrated, living organ-ism. It emphasizes holism and emergence be-yond mere additive complexity; active, teleo-logical predilection beyond passivity andreactivity; dynamic beyond static systems;and structural differentiation and integrationover time. Although organicism allows forand can accommodate mechanistic phenom-ena, scholars have long debated whether thereverse is true-that is, whether a mechanisticmodel is inherently incapable of accommodat-ing or explaining many of the essential char-acteristics of human functioning such asgrowth, adaptiveness, development, inten-tionality, structural reorganization, and othercharacteristics of living organisms (Nagel,1961). It is not necessary to take a stand onthis issue one way or another, however, to ap-preciate the practical barriers to studying or-ganismic phenomena within a mechanisticframework.

Scientific hard cores. The importance ofworldviews lies in the powerful central rolethey play in guiding, facilitating, and/or con-straining the activities of a science. Overton,elaborating on the significance of worldviewassumptions and commitments, has arguedthat they function as an essential part of thehard core that serves as the foundation forscientific programs (Lakatos, 1977; Laudan,1977; Overton, 1984a). The defining charac-teristic of hard cores is that they are shieldedfrom potential falsification by the researchprogram’s positive heuristic and belt of auxil-iary hypotheses (Lakatos, 1977). The positiveheuristic includes suggestions and guidelinesfor examining, challenging, and changingonly the refutable (i.e., non-hard-core) fea-tures of the research program and for modify-ing and refining the protective belt of auxil-iary hypotheses. Thus, only the particularideas, theories, hypotheses, methods, and re-search strategies of a research program aresubject to scrutiny, challenge, and potentialfalsification. The hard core itself is taken forgranted and remains relatively insulated.Within the hard core of every science, for ex-ample, are the assumptions that its phenom-

Hubble hypothesis

ena are rule governed and that the relevantlaws are, in principle, discoverable. Thus, thefailures of a science may be attributed to er-rors in particular theories, methods, proce-dures, and strategies, or even to intrinsic com-plexities in its phenomena, but not toweaknesses in its hard-core assumptions. Thepower of the hard core was best expressed byLaudan’s characterization of research tradi-tions as “ . . . . a set of ontological and method-ological do’s and don’t’s . . .. To attempt whatis forbidden by the metaphysics and method-ology of a research tradition is to put oneselfoutside the tradition and to repudiate it” (Lau-dan, 1977, p. 80).

The basic conceptual and philosophicalcriticisms of a strict mechanistic approach tostudying human functioning have been dis-cussed extensively in the literature of psychol-ogy (e.g., Altman & Rogoff, 1987; Lewin,193la, 193lb; Nagel, 1961; Overton & Horo-witz, 1991; Pepper, 1942). So also have thecomparative merits of transactional, dialectic,and organizational variants of organismicapproaches (Lerner & Kaufman, 1985; Over-ton & Horowitz, 1991). I concentrate belowon some of the practical limitations of themechanistic model and advantages of the or-ganismic model for thinking about and con-ducting day to day research in psychologicalscience, My point of departure is a questionraised by Kagan et al. almost 20 years agoin reaction to the statement, “The interactionsbetween biological and environmental forcesdetermine the psychological growth of the or-ganism: What in heaven’s name does thatfourteen word sentence mean?” (Kagan et al.,1978, p. 44). Indeed, descriptions of this typeare often so general that they could mean al-most anything and therefore communicatevery little. I will attempt to elaborate on somespecific implications of an organismic ap-proach to human functioning.

Closed versus open systems. One of the essen-tial differences between mechanistic and or-ganismic models lies in the distinction be-tween closed and open physical systems (vonBertalanffy, 1968). All nonliving physicalmatter falls within the category of closed sys-tems, which are arrangements of physical

203

entities connected in ways that constitute aunity or whole, and are completely isolatedfrom their external environment. That is, theentities in a closed system interact only withone another and are not subject to outside in-teractions or influences of any kind (Popper,1991). Because they exchange neither energynor matter with the external environment,closed systems are subject to the second lawof thermodynamics. At rest, they tend towardthe state of highest probability (equilibrium),maximal disorder, and the disappearance ofexisting differentiations in structure. Strictlyspeaking, the only truly closed physical sys-tem is the universe itself, because it presum-ably has no external environment. As a practi-cal matter, however, nonorganic physicalsystems can be conceptualized along a contin-uum of closedness as a function of their rela-tive degrees of isolation from external factors(e.g., temperature, force). Although onlyquasi-closed in this sense, they are nonethe-less commonly referred to as closed systems.

The metaphor most closely associated withclosed systems is the machine, characterizedby fixed patterns of mechanical reactions andinteractions among its parts, governed by uni-versal laws, producing fixed end states. To theextent that a closed system is completely iso-lated from external factors, its end states aredetermined by initial conditions. The finalconcentrations resulting from chemical equi-librium in a closed system, for example, arecompletely determined by initial conditions,such that a change or modification (whethermanipulated or observed) in those conditionsor intervening processes will produce predict-able changes in the final concentrations (vonBertalanffy, 1968). Similarly, with adequateknowledge of a completely closed system, itis possible to infer backward from a givenstate to its initial conditions.

Equifinality in open systems. All living organ-isms fall into the category of open systems,which exchange energy and matter with theenvironment and are not therefore subject toentropy. Beyond thermodynamic openness,living organisms are distinguished fromclosed systems by unique functioning char-acteristics such as growth, adaptiveness,

204 Richters

Substance Dependency

Attention Seeking

Deviant Values

Revenge

Impulsivity

Figure 1. Equifinality in human functioning as an illustration of different personality struc-tures and processes that might give rise equifinally to persistent antisocial behavior in differ-ent individuals, depending on a matrix of other internal and external factors.

Athlete *

Performer

Professor

‘4

Clergy

Politician

Salesperson

Figure 2. Multifinality in human functioning as an illustration of different overt behaviorpatterns that might arise multifinally from a common characteristic such as extroversion,depending on a matrix of other factors internal and external to an individual.

regeneration, emergence, differentiation, dy-namic self-stabilization, and self-reorganiza-tion. Organisms are also distinguished fromclosed systems by the functioning capacitiesof equifinality and multifinality. Equifinalityis the ability to reach similar outcomes or endstates from different starting points andthrough different processes (Figure 1). Con-versely, multifinality (Figure 2) is the abilityto reach different outcomes from similar start-ing points and/or through similar processes(von Bertalanffy, 1968). Thus, in contrast toclosed systems, the initial conditions of opensystems imply nothing about their end statesand their end states imply nothing about theirinitial conditions.

Equifinality was observed and labeledoriginally in the biological sciences followingdiscoveries by Spemann and others that cellu-lar tissues can be transplanted successfullyfrom one area of presumptive growth to an-other (Spemann, 1938). During critical peri-ods of growth and maturation, for example,

tissues from the neural plate of amphibia canbe transplanted successfully to areas of limbgrowth, where they take on characteristics ofmuscle or skin tissue instead of brain tissue.Similarly, in embryology, it was demonstratedthat living organisms (end states) could de-velop from ova that were whole, divided, orfused (Waddington, 1957). Even in ontogeny,equifinality is evident in the fact that organ-isms often overcome-make up for-tempo-rary delays in physical and physiological mat-uration by nonetheless reaching similar endstates. The underlying principle of equifinalityin biological systems is teleological in thesense that the end states of the organism areunderstood to be genetically precoded. More-over, because the elements of a living organ-ism are reciprocally related and mutually reg-ulating, their complex organization allows forthe achievement of different end states fromdifferent starting points and through differentprocesses.

Equifinality is not restricted to teleological

Hubble hypothesis 205

.

biological processes. As von Bertalanffy noted,it is also found in psychological and sociolog-ical systems. In an early and elegant demon-stration of fail-safe behavioral systems. forexample, psychobiologists demonstrated thatthe ability of female rats to retrieve their pupsa-as not dependent on the stimuli of any singlesensory system (Beach & Jaynes, 1956; Her-renkohl & Rosenberg, 1977). Regardless ofwhether their tactile, olfactory, auditory, orvisual sensory channels were destroyed, andeven when rendered anosmic, their ability toretrieve their pups was preserved by the abil-ity to function equifinally.

It is in this broader sense that equifinalityis often used in developmental psychopathol-ogy to refer to different structural/causal pro-cesses underlying similar overt patterns orsyndromes of cognition, emotion, behavior,and/or psychopathology (Cicchetti & Rich-ters, 1993; Cicchetti & Rogosch, 1996; Rich-ters & Cicchetti, 1993a, 1993b). In thisbroader usage, equifinality refers to a definingcharacteristic of human functioning so ubiqui-tous and widely recognized that it is simplytaken for granted in social discourse and be-havior.

Individuals vary considerably in theirbeliefs, values, attitudes. preferences, temper-aments, intelligence(s), personality charac-teristics, experiential histories. current cir-cumstances, and anticipated futures. Thesedifferences, in turn, can give rise to very simi-lar patterns of overt behavior and functioningfor different reasons (equifinality) as well asvery different overt behaviors for similar rea-sons (multifinality). There are virtually unlim-ited different ways in which any given feel-ing, desire, or objective (e.g., anger. love.empathy, fear, greed, anxiety, sadness. happi-ness, survival) may be expressed or actedupon (multifinality). Conversely. very differ-ent feelings, desires, and objectives may beexpressed through similar overt patterns offunctioning (equifinality). For example. indi-viduals may contribute generously to charitiesprimarily for tax write-off purposes, to en-hance their reputations in the community (forgood purposes or bad), because they aredeeply committed to a charity’s objectives, ormore generally to the principle of helping oth-

ers (equifinality). Not everyone who is com-mitted to helping others, however, contributesgenerously to charities. Depending on a ma-trix of other internal and external factors, theymay instead participate in church activities,volunteer their time to charitable organiza-tions, run for elective office, join lobbying ef-forts, or merely act compassionately towardothers in their daily lives (multifinality).

Equifinality and multifinality also havesignificant implications for thinking about theso-called mental disorders of childhood. Con-sider, for example, the common childhoodsyndrome of attention deficit hyperactivitydisorder (ADHD), codified in the Diagnosticand Statistical Manual of Mental Disorders,4th edition (American Psychiatric Associa-tion, 1994). The diagnosis of ADHD is basedstrictly on reports about a child’s patterns ofinattention, hyperactivity, distractability, andso forth, not on a process-level assay of thecauses or structures responsible for those be-haviors. Thus, although children who meetADHD diagnostic criteria share in common ageneral constellation of overt functioning def-icits, there is no reason a priori to believe thatthey stem from a common etiology or under-lying structure. Moreover, there are soundconceptual, theoretical, and clinical reasonsfor believing that the population of childrenwho meet ADHD criteria is heterogeneouswith respect to the relevant causal factors.Different subsets of children diagnosed withADHD may have different kinds of inherited,biologically acquired, and/or environmentallyacquired diatheses giving rise to the overtsyndrome. Others may be responding primar-ily to environmental stressors in the absenceof an underlying diathesis. And yet for othersthe symptoms may be products of diathesis/circumstance interactions.

Equifinality and structure. N o t e t h a t t h eextension of equifinality beyond biologicalstructures per se to human psychological andsocial functioning introduces an additional el-ement of complexity. In biology, there areconsiderable limits to the extent that differentend states can be accomplished by differentstructures. The functions of the heart, for ex-ample, cannot be performed by the liver.

206 Richters

Moreover, hearts, livers, and kidneys performthe same functions in all humans. In psycho-logical and social functioning, however, thereare very few constraints on the ability of dif-ferent processes and structures to compensatefor and/or facilitate each other in the serviceof end states.

The digital computer provides a usefulanalogy for thinking about equifinality interms of the structures underlying overt hu-man behavior (Elasser, 1966; Lykken, 1991).All functions carried out by a digital computerstem from the same unchanging hardwareconfiguration. Each time a software programis loaded to perform a particular task such asbookkeeping, word processing, statisticalanalyses, or graphics, however, the computerundergoes elaborate structural modifications.It would be impossible therefore to under-stand the computer’s overt functioning in anyof these domains by studying the hardware it-self. Each program constitutes a different un-derlying structure which operates in differentways and according to different algorithms inthe service of accomplishing different tasks.Moreover, the same tasks may be accom-plished in different ways on the same com-puter (e.g., SPSS vs. SAS statistical pack-ages). Thus, understanding the computer’sovert functioning r quires a consideration ofthe software from which it arises. Humanbrain structures, of course, constitute the bio-logical equivalent of computer hardware.Moreover, unless we are Cartesian dualists,we must assume that all human psychologicalfunctioning and behavior is ultimately a prod-uct of the brain.

Nonetheless, for good reasons, most of ourtheories do not focus directly on biochemicalor electrophysiological processes of the brain.Rather, they concern higher level processesand structures of personality, temperament,intelligence, and so forth that constitute thestructural equivalents of software programs.These are the emergent characteristics of thehuman nervous system’s extraordinary plas-ticity and capacity for structural modifiabilityat every level of organization from molecularto molar. These higher order structures, inturn, emerge and develop through each indi-vidual’s unique combination of inheritance,

learning, and experience. It is precisely thismodifiability that allows individuals toachieve similar overt functions through differ-ent processes (equifinality) and different overtfunctions through similar processes (multifi-nality). Obviously, then, there may be struc-turally different variants of many of thesehigher order characteristics, as well as qualita-tively different configurations among them, toconsider in approaching the study of overtfunctioning.2

Structure in closed versus open systems. Thebasic business of every science is to describethe nature of its entities and to develop theo-ries about their structures from which theirfunctioning can be deduced. The coherence ofthese activities rests on the success of a sci-ence in organizing its entities into classes thatare sufficiently homogenous in theory-rele-vant structure that they can be considered in-terchangeable for theoretical purposes (Elas-ser, 1966). Quantum theory, for example,absolutely depends on the assumption ofwithin-class homogeneity. In fact, it is impos-sible to formulate the mathematics of quan-tum mechanics without the axiom that themembers of each class of particle (e.g., elec-trons, protons) are rigorously alike and inter-changeable (Elasser, 1966). Similarly, classi-cal physics requires that its objects aresufficiently homogenous in this respect thatwithin-class irregularities can be averaged outstatistically for purposes of prediction and ex-planation. It is also this structural homogene-ity that makes it coherent to pursue a singletheory to account for the actions and interac-tions among phenomena within a closedsystem.

The extraordinary human capacities forequifinal and multifinal functioning, however,

2. The limitations of the computer software analogy areapparent here. Although different word-processing pro-grams (e.g., Word and WordPerfect) may function sim-ilarly based on different underlying structures (e.g.,algorithms, subroutines, etc.), they are nonetheless de-signed to accomplish the same objectives. In the caseof human functioning. however, similar patterns of sur-face functioning may not only stem from different un-derlying structures (like the computer), but may be inthe service of different goals or purposes (unlike thecomputer).

Hubble hypothesis 207

render the structural homogeneity assumptionuntenable. Very similar patterns of overt func-tioning may be caused by qualitatively differ-ent underlying structures both within the sameindividual at different points in time, andacross different individuals at the same time(equifinality). Conversely, different patternsof overt functioning may stem from very sim-ilar processes within the same individual overtime, and across different individuals at thesame time (multifinality) (Cicchetti & Ro-gosch, 1996; von Bertalanffy, 1968). Thesestructural differences in turn, have tremen-dous implications for the psychological stimu-lus properties of stressors, events, and experi-ences, the causal relevance of which may varyqualitatively across individuals. In a closedsystem consisting of within-class structurallyhomogeneous objects there is warrant for as-suming that the influence of a given variablewill vary only quantitatively across membersof the class. In the realm of human function-ing, however, there is no basis for this as-sumption. The meaning, salience, and causalimpact of a given event or experience canvary qualitatively across individuals as a func-tion of their differences in genetic endow-ment, experiential history, resulting internalstructures, and current circumstances. As theStoic philosopher Epictetus observed “Menare disturbed not by things, but by the viewswhich they take of them.” From an open sys-tem standpoint, these different “views” are afunction of qualitative differences in internalstructures.

The distinction between closed and opensystems reveals the sense in which psycholo-gy’s approach to discovery can be character-ized and criticized as mechanistic. It is cer-tainly not the case that most psychologistsendorse a mechanistic model of human psy-chological functioning. Most, in fact, wouldreject it vigorously as an anachronism. None-theless, the mechanistic model is built intopsychology’s paradigm in the sense that eachmajor step in the discovery process makessense only as part of an overall strategy forstudying objects that are within-class structur-ally homogeneous in relatively closed physi-cal systems. That is, the discovery process it-self is framed as a search for universal laws

to account for individual differences in the de-velopment and functioning of individuals whoare presumed to be structurally homogenous.I argue below that the homogeneity assump-tion is evident throughout the discoveryprocess in the selection and recruitment ofsubjects, the selection and assessment of vari-ables, and in the analysis and interpretation ofdata (see Interdependence Problem section).Although this assumption is seldom justifiedand is often contraindicated by commonsense, it has been protected for decades by anoverly forgiving justification process.

Context of justification

Traditionally, scientific justification has beenthe chief concern of logicians, philosophers,and historians of science, notably because ofits significance to the so-called problem of de-marcation between science and pseudoscience(Lakatos, 1977). From a philosophy of sci-ence standpoint, it is irrelevant how scientistsarrive at their ideas or observations within thecontext of discovery. Standardized researchmethods and procedures are indispensable, ofcourse, for the coherence of a scientific com-munity’s activities and communications. Butthey are of little concern to the philosopher ofscience because the merits of all ideas, regard-less of how they are arrived at in the contextof discovery, must be evaluated finally in thecontext of justification. The scientific enter-prise is distinguished from other ways of fix-ing beliefs about the world, such as variousforms of dogmatism and pseudoscience, bythe rigor with which it questions and evalu-ates the ideas it discovers. Thus, criticismsabout the scientific standards of a disciplineare tantamount to questions about its legiti-macy as a field of science.

It is therefore especially noteworthy andtroubling that psychology stands alone amongrecognized sciences in being criticizedharshly for its justification practices, notablyfor weaknesses in the logic, methods, andstandards by which it evaluates theoreticalclaims. The distinguished Nobel laureatephysicist Richard Feynman was quite explicitin characterizing psychology as a “CargoCult” pseudoscience on these grounds by

208 Richters

comparing its scientific practices to the super-stitious rituals of South Sea islanders follow-ing the second World War. Having witnessedmilitary aircraft landing on their island withmaterials and supplies during the war, theyengaged in elaborate rituals using bars ofbamboo in place of antennas, and buildingwooden huts to look like control towers, allin a vain effort to attract the planes again.Like the South Sea islanders, according toFeynman, many psychologists and other so-cial scientists follow only the apparent pre-cepts and forms of scientific investigation.But, because they are missing something es-sential, they suffer the same fate: “ . . . theirplanes don’t land” (Feynman, 1986, p. 179).

As psychology’s justification processeshave come into question over the decades,they have been discussed most often by andfor those most familiar with the highly spe-cialized languages and concepts of logic, phi-losophy of science, and statistics. Althoughthe essential criticisms are rather straightfor-ward and compelling, their impact potentialhas been limited by three related factors. First,the issues often have been discussed in techni-cal terms that make them seem variously inac-cessible, esoteric, and/or of little practicalconsequence. It is not uncommon, for exam-ple, to encounter complex formulas of sym-bolic logic and difficult statistical notations injustification discussions (Meehl, 1990; Roze-boom, 1982). To the uninitiated, justificationissues often seem indistinguishable from tech-nical discussions about the relative merits ofapplying a Bonferroni versus Scheffe statisti-cal correction procedure under a given set ofcircumstances. Second, with some notableand welcome recent exceptions, justificationissues often have accumulated outside themainstream literature of psychology in spe-cialized books, chapters, and journals. Thisprobably has reinforced the misperceptionthat, whatever the technical merits of justifi-cation arguments, they have little practicalimport for the day to day activities of thepracticing scientist. As Meehl insisted almost20 years ago, however, “I am not makingsome nit-picking statistician’s correction. I amsaying that the whole business is so radicallydefective as to be scientifically pointless”

(Meehl, 1978, p. 26). Third, justificationproblems characteristically have been dis-cussed in relative isolation from specific is-sues of discovery, further obscuring theirpractical implications.

Scientific conventions of theory evaluation. Inmost developed sciences, theories are evalu-ated by their explanatory power and ability tomake predictions of phenomena that are bothprecise and novel. Precision typically is in-dexed by the degree of accuracy with whicha theory can make point predictions of phe-nomena within a very narrow range of speci-ficity. Relatedly, novelty is indexed by a theo-ry’s ability to anticipate and predict facts andobservations that would be unlikely absent thetheory. In Salmon’s classic phrase, a theoryis considered impressive to the extent that itsaccurately predicted observations would beotherwise explainable only as a damn strangecoincidence (Salmon, 1984). Following fromthis principle, when different theories offercompeting explanations for the same phenom-ena, the one that generates the most accuratepredictions and anticipates the most noveloutcomes generally is considered-at least forthe time-the better theory. Thus, in general,theories achieve their status by having sur-vived what Popper referred to as grave risksof refutation in the empirical arena (Popper,1959).

Theory testing in psychology. In most areas ofpsychology neither precision nor novelty arerequired to justify theoretical claims. Instead,they are evaluated through formalized proce-dures of hypothesis testing, the outcomes ofwhich are determined by a null-hypothesistest of statistical significance. Theories inmost nonexperimental areas of psychologyare of a sufficiently general nature that theypredict and purport to explain that theory-rele-vant variables will not be random with respectto each other and will be positively or nega-tively correlated in the population of interest.That is. they yield directional rather thanpoint predictions of group differences or cor-relation values. such that a theory positing acausal influence of variable A on variable Bwill be tested by determining whether the

Hubble hypothesis 209

variables are at all correlated in the direction(above or below 0) predicted. Thus, the ob-served correlation between A and B will betested against the null hypothesis (Ho) of norelationship-literally, a covariation value ofzero. By convention, if the computed correla-tion is greater than zero, Ho is rejected in fa-vor of the investigator’s theory. Conversely,if the correlation is not greater than zero theinvestigator failed to reject Ho and instead re-

jects the theory.Null-hypothesis significance tests. An ab-

solute comparison of the AB correlation with. zero would be straightforward if it were based

on all subjects in the population of interestand if the variables were assessed with perfectprecision. The fact that it is based on only asample drawn from that population, however,raises the possibility that it may be an artifactof sampling bias. That is, even in a populationin which two variables are randomly distrib-uted with respect to each other, any particularsample drawn from that population may none-theless reflect a correlation greater than zero.Thus, to guard against the risk of erroneouslyconcluding in favor of the theory on the basisof such an artifactual correlation, a test of sta-tistical significance is conducted using Ho asthe standard of comparison. The basis for thistest, simply stated, is the fact that repeatedsamples drawn from a population in whichvariables are truly random with respect toeach other will nonetheless yield a normaldistribution of correlation coefficients aboveand below zero. This hypothetical distributionprovides the basis for estimating the probabil-

. ity of detecting a nonzero correlation in asample drawn from a zero-correlation popula-tion. To protect against making such a Type

1 error-rejecting Ho when it is true-a rela-tively low probability value (usually p < .05or .01) is adopted. Thus, the sample basedcorrelation is considered statistically signifi-cant-and therefore supportive of the the-ory-if the probability of detecting a correla-tion of similar or greater magnitude from azero-correlation population would be very un-likely.

It has been estimated that over 90% of em-pirical articles published in psychology’sleading scientific journals rely on null-hy-

pothesis testing as the sole means for inferringconclusions from research data (Loftus,1993). Estes recently reported the proportionof empirical studies presenting significancetests in current issues of leading psychologyjournals as follows: Memory and Cognition(1 OO%), British Quarterly Journal of Experi-mental Psychology (100%) Journal of Abnor-mal Psychology (100%) and Journal of Per-sonality and Social Psychology (88%) (Estes,in press). Despite its ubiquity, the ways inwhich null-hypothesis testing is relied upon inpsychology has been criticized by numerousstatisticians, philosophers of science, and psy-chologists as the equivalent of no scientificstandard at all for evaluating substantive theo-ries. It warrants emphasis that null-hypothe-sis significance testing per se is recognized asa valuable, necessary, and elegant tool for an-swering particular types of questions undercertain specified conditions (Hagen, 1997).These virtues do not, however, extend to mostof the applications for which psychology re-lies on statistical significance testing. Specifi-cally, it is psychology’s widespread relianceon significance testing as a theoretical toolthat has been harshly criticized (Cohen, 1990,1994; Gigerenzer et al., 1989; Glymour, 1980;Loftus, 1991; Lykken, 1968, 1991; Meehl,1978, 1990; Morrison & Henkel, 1970).

Null hypothesis is quasi always false. Thefirst criticism is that in the social, behavioral,and biological sciences, the null hypothesis isalmost never true (Bakan, 1966; Edwards,1965)! Or, in Meehl’s terms, it is “ . . . quasi-always false . due to the empirical fact thateverything is correlated with everything else,more or less” (Meehl, 1990, p. 123). This phe-nomenon has been referred to both as the crudfactor in psychology (Lykken, 1991), and,less pejoratively, as psychology’s ambientcorrelational background (Meehl, 1990). Therather obvious general explanation is that itprobably reflects complex, multideterminedcausal relationships among genetic linkage as

3. These criticisms typically have not been extended totechnological uses of statistical significance testingsuch as clinical trials, which more closely approximatethe applications and conditions for which it was devel-oped (Meehl. 1990).

210 Richters

well as autocatalytic processes of affect, cog-nition, and behavior both within and betweenindividuals (Meehl, 1990). Whatever itssources, the troubling empirical fact is thatnonzero correlations (both positive and nega-tive) have been demonstrated routinely be-tween randomly chosen variables in large datasets (Evans & McConnell, 1941; Meehl,1967, 1990; Lykken, 1991; Thurstone, 1938).Thus, a theory’s directional predictions standa 50/50 chance of being confirmed even if thetheory has absolutely no merit, assuming suf-ficient statistical power to meet criteria forstatistical significance!

Meehl’s paradox. The combination of direc-tional predictions and the so-called crud fac-tor creates an additional problem, dubbedMeehl’s Paradox (Meehl, 1990; Leventhal,1992; Lykken, 1991). In developed sciences,which emphasize point predictions withinvery narrow boundaries of error, increases inmeasurement precision create more difficulthurdles for theories by increasing sensitivityto deviations between theoretically predictedand observed values. In psychology, wherethe predictions are only directional, increasesin precision actually do the reverse. That is,they lower the evidentiary hurdle by increas-ing the likelihood of falsely rejecting HO

(Type 1 error) and capitalizing on ambientbackground correlations that are irrelevant tothe theory. The assumption-supported byconsiderable data-that Ho is quasi alwaysfalse means that most variables will be corre-lated to some degree for reasons that havenothing to do with the theory being explored.In many areas of physical science this is arelatively minor consideration for three rea-sons. First, background correlations are muchless likely to exist. Second, the physical sci-ences are much better able to isolate many oftheir basic phenomena in laboratory settingsand manipulate them in ways that reasonablyensure the absence of artifactual correlations.Third, and most important, the physical scien-tist typically is making a rather precise pointprediction within a very narrow range of errorfrom his/her theory. Thus, to the extent thatmeasurement errors can be reduced through

greater precision, it becomes easier to detecttrue (nonartifactual) differences or discrepan-cies between predictions and observations.

Everything is reversed in psychological re-search. Background correlations are ubiqui-tous in psychology generally, and particularlyin uncontrolled nonlaboratory settings. More-over, in contrast to the physical scientist’spoint prediction, a directional predictionallows for any value (or group difference)other than absolute zero as confirmatory evi-dence as long as it is in the right direction.Therefore, an increase in measurement preci-sion will increase rather than decrease thelikelihood of capitalizing on theory-irrelevantassociations and therefore lower rather thanraise the evidentiary hurdle a theory mustjump in the justification process. Relatedly,the statistical power for detecting a correlationas statistically significant also increases withsample size. Thus, literally any deviationfrom Ho in the population will result in its re-jection if the sample is large enough; thelarger the sample the higher the risk of com-mitting a Type 1 error! As I argue below, thefrequently recommended replacement stan-dard of effect sizes is often an inadequate so-lution to this problem (see Justification incontext below).

Statistical versus theoretical hypotheses. Asomewhat separate problem is that relying ona test of significance to evaluate the merits ofa theory confuses the researcher’s statisticalhypothesis with his/her theoretical hypothesis,and relatedly conflates the very different con-cepts of statistical significance and theoreticalsupport. The H0 and alternative hypothesis ina test of statistical significance (e.g., aboveexample) are statistical hypotheses concerninga population parameter. That is, the purposeof the significance test is solely to aid the in-vestigator in drawing an inference about anunobserved population parameter from the ac-tual value(s) computed on an observed sam-ple. The test has nothing whatever to do withthe merits of any particular theoretical expla-nation for why the value exists. The theoreti-cal hypothesis, on the other hand, concerns arelation between the theory and patterns in the

Hubble hypothesis

sample data. Thus, whereas the statistical sig-nificance test is a way of using the sample todraw an inference about the population, thecorrelation is the investigator’s method ofdrawing an inference from the sample to thetheory. The significance test, in short, is not inany way a test of the merits of the substantivetheory. Meehl explains this difference suc-cinctly by telling students-“ . . assume youhad the parameter; what would you know, and-

. how confidently?’ (Meehl, 1990, p. 117).

Misinterpretations of p values. Even those. who recognize that the significance test refers

only a link between the sample and popula-tion nonetheless often misinterpret even itsstatistical meaning, usually in one of threeways. First, the p of a significance test is ofteninterpreted as the probability that the ob-served sample-based correlation was due tochance. Second, it is often referred to as theprobability of detecting a correlation of simi-lar or greater magnitude from another sampledrawn from the same population. The third,most egregious, and pervasive misinterpreta-tion is that 1 -p value (e.g., 1 - .05 = .95) isthe probability that the alternative (i.e., inves-tigator’s) hypothesis is correct. Oakes recentlyestimated that approximately 95% of aca-demic psychologists believe this latter inter-pretation. None of these three interpretationsis accurate! The p value is nothing more thanthe probability of detecting a correlation (orgroup difference) of the same or greater mag-nitude in a sample randomly drawn from azero-correlation (or zero group difference).population. It is hardly surprising that the sta-tistical significance test is so commonly mis-

. interpreted and misused in psychological re-search. Erroneous interpretations have beentaught in some of psychology’s classic, mostwidely used and respected textbooks on mea-surement and statistics (Anastasi, 1958, p. 11;Ferguson, 1959, p. 133; Guilford, 1942, pp.156-166; Nunnally, 1975, pp. 194-196).Moreover, for more than a decade, the 2ndedition of the official Publication Manual ofthe American Psychological Association mis-led psychologists by informing them that datatrends failing to meet the usual levels of sig-

211

n i f i c a n c e “ are best interpreted as causedby chance and are best reported as such”(American Psychological Association, 1974,p. 19).

Logical fallacy. Weak uses of significancetests to evaluate theories are further com-pounded by a conceptual and logical error inreasoning that has nothing to do with statis-tics. The error stems in the first place frompsychology’s emphasis on empirical corrobo-ration to judge the merits of theories. In thedeveloped sciences, theories are evaluated bysubmitting them to grave risks of refutation inthe empirical arena (Popper, 1959). The logicof refutation conforms to two forms of argu-ment that are deductively valid from a prem-ise. If, for example, theory T holds that vari-able X causes variable Y under specifiedconditions, the premise is that T entails or im-plies the existence of 0 under those condi-tions (T:. 0). Following deductively fromthis premise, modus ponens holds that if the-ory T is true, observation 0 must be true(T :. 0). Also following deductively from theoriginal premise, modus tollens holds that ifobservation 0 is not true, theory T is not true(T :. O). Psychology’s reliance on signifi-cance testing to evaluate theories relies in-stead on an invalid form of argument knownas affirming the consequent (0 :. T). That is,it relies on the fallacy of concluding from(T :. 0) that, if observation 0 is true, theoryT is true. Obviously, this interpretation restson the faulty assumption that the presence ofchildren’s antisocial behavior necessarily im-plies a history of exposure to marital distress.Since numerous plausible competing theoriescan be invoked to account for and success-fully predict the same directional association,

4. Although sufficient for present purposes, this is a grossoversimplification of theory testing conventions fortwo reasons. First, the core of a theory is challenged inthe face of disconfirming evidence only if the implicitceteris paribus (i.e.. all things being equal) clause canbe assumed to hold--which is virtually impossible toestablish in most domains of psychology citing Meehl(1990). Second, theories are not rejected on the basisof a single test of experiment. but on the basis of theirlong-term performance over a range of tests within thecontext of a research program (Lakatos, 1977).

212

the correlation itself implies nothing about themerits of a particular theory. Again, referringback to the distinction between statistical andtheoretical hypotheses, a rejection of H0 istechnically (but only when used properly) acorrect application of modus tollens concern-ing the statistical hypothesis of no difference.The conclusion that a confirmed prediction,directional or point, in any way corroboratesthe theory purporting to explain the associa-tion is a simple error in logic (Rorer, 1991).

Summary

Alone, any one of these weaknesses would besufficiently consequential to compromise theintegrity of psychology’s justification proccesses. Collectively, they have been character-ized variously as feckless, impoverished,weak, illogical, misleading, a terrible mistake,and even one of the worst things to happen inthe history of psychology (Meehl, 1978). Ja-cob Cohen, one of social and behavioralsciences’ leading statisticians, characterizedmainstream research psychologists as being“ . . mesmerized by a single all-purpose,mechanized ‘objective’ ritual in which weconvert numbers into other numbers . . Wehave come to neglect close scrutiny of wherethe numbers come from” (Cohen, 1990, p.1310). Echoing similar concerns, criminolo-gist Michael Maltz recently criticized thewidespread misuse of null-hypothesis testingin psychology and allied disciplines as being“ . . in all too many cases a meaningless ex-ercise” (Maltz, 1994). Geoffrey Loftus, uponbeing appointed as editor of the journal Mem-ory and Cognition, characterized the practicesas “ . . . at best severely limited, and at worsthighly misleading” (Loftus, 1993, p. l), aform of “tyranny” and a “virtually barrentechnique” that lies “ . . . at the heart of muchmalaise in psychological research” (Loftus,1991, p. 1). Imre Lakatos, one of the 20thcentury’s leading philosophers of science, waseven more stinging in his assessment, criticiz-ing psychology as a largely ad hoc enterpriseresulting in a research literature of disparateand often conflicting findings “cancerousgrowth’ amounting to “intellectual pollution”(Lakatos, 1978, p. 89). Surprisingly, there has

Richters

been little evidence in psychology’s contem-porary scientific journals or ongoing researchactivities that the majority of research psy-chologists are even aware of these criticismsof its standards.

The Interdependence Problem

The ways in which discovery and justificationweaknesses interact to undermine the coher-ence and potential of research are most easilyunderstood in the context of day to day re-search. In the discussion below 1 consider thecase of a prototypical study in which the in-vestigator is interested in testing a theory thatassigns a causal role to marital distress in thedevelopment of children’s antisocial behavior.Thus, the theory both predicts and purportsto explain an association between these twovariables. In the more common multivariatecase a theory might predict such an effectonly under certain mediating conditions. Al-though the issues and arguments below areeasier to illustrate with the more simple bivar-iate case, they apply with equal force to themultivariate case. The research begins in thecontext of discovery with decisions aboutstudy design, sample selection, data collec-tion, and data analysis. Following this, the in-terpretability and inferential relevance ofthese data to the investigator’s theory will beevaluated in the context of justification.

Study design

Note that the initial decision to conduct re-search in the service of testing a theory al-ready reflects an influence of the anticipatedjustification process. That process is narrowlydefined in terms of a hypotheticodeductivemodel of theory evaluation in which the in-vestigator begins with a theory, deduces atestable hypothesis, and collects theory-rele-vant data to be analyzed and summarized inthe service of testing the theory. The para-digm makes a clear distinction between thefunctions of discovery and justification, andvalues discovery activities only to the extentthat they are conducted in the service of test-ing a theory. An investigator interested in-stead in a more creative endeavor of explor-

Hubble hypothesis

atory, descriptive research regarding someaspect of marital distress, children’s antisocialbehavior, or their links, would suffer from theoutset. The likelihood of research fundingfrom most sources for such descriptive/ex-ploratory work is relatively low because thesame peer review justification standards usedto evaluate the empirical worth of a study forlater journal publication will be applied to theinitial funding proposal as well. It often will

. be evaluated as weak and/or nonscientific tothe extent that it is not designed in the serviceof a hypothetico-deductive theory test. Real-

. izing this, the investigator will embed-andsometimes mask-any exploratory/descrip-tive goals within the framework of a tradi-tional test of a theory. The sensitivity of thejustification process to statistical issues ofpsychometrics, variable to subject ratios andstatistical power, however, will impose con-siderable constraints on the nature and depthof whatever exploratory research is conductedwithin the theory-testing framework of thestudy design.

Sample selection

The investigator’s first practical task will beto recruit a sample of subjects from the popu-lation of interest. In the study of a closed sys-tem an investigator samples the objects of in-terest on the basis of their known or presumedwithin-class structural homogeneity. Mostsamples in psychological research, however,are composed either of self-selected individu-als of unknown heterogeneity in their underly-

ing structures or randomly selected individu-als recruited from a specified population ofinterest to the researcher. Ironically, although

.random samples are valued for purposes ofgeneralization, they are almost by definitionmore likely than a typical convenience sampleto be structurally heterogeneous with respectto a given phenomenon.

The prototypical risk factor study, how-ever, will proceed on the implicit assumptionthat, with respect to the theory-relevant struc-tures under study, individuals recruited intothe sample are homogeneous. That is, it willbe assumed a priori that they are interchange-able members of a class who differ quantifica-

213

tively but not qualitatively with respect to the-ory-relevant structures and variables. If theyare further subclassified at some point in thestudy, the reclassification typically will bebased again on their phenotypic similarities,not with reference to possible differences inthe presumed underlying structures responsi-ble for those similarities.

There may be specific empirical or theoret-ical reasons for distinguishing subtypes ofsubjects on the basis of socioeconomic status,age, sex, or background characteristics. But inthe absence of a specific rationale for suchdistinctions the default assumption will bethat members of the sample are structurallyinterchangeable. Most of the emphasis in re-cruitment will be placed on securing a samplewith sufficient phenotypic variability in thevariables of interest to warrant a reasonabletest of the study hypothesis. Once subjects arerecruited on the basis of phenotypic similari-ties, every subsequent step in the discoveryand justification processes will be based nec-essarily on the simplifying assumption ofstructural homogeneity across subjects. If thisassumption is rejected, there is no rationale orjustification for including them in the samesample for purposes of exploring a model ortheory about a unitary underlying structure.

Data collection

The structural homogeneity assumption alsosimplifies the data collection process. That is,it justifies (a) assessing the same variableswith respect to each subject, (b) assessingeach variable in isolation from the others, and(c) employing the same standardized data col-lection procedures across all subjects. For rea-sons explained below, it also requires prac-tices (a) and (c) but not (b). These strategiesare consistent with the study of a closed sys-tem because within-class structural homoge-neity implies that any influence of a variableor assessment procedure will be uniform, dif-fering only quantitatively across members ofthe class. To the extent that members of theclass/sample are structurally heterogeneous,however, this threefold standardization strat-egy may have opposite effects.

214 Richters

Variable selection. Structural heterogeneity,for example, entails the considerable possibil-ity that a variable such as marital distress maybe a causal factor for antisocial behavior insome children, an effect for others, a spuriouscorrelate for others, and unrelated for yet oth-ers. Variables such as peer rejection, attribu-tional bias, family disharmony, parentingpractices, parent symptomatology, and soforth, are all obvious examples of variablesthat may play any of these qualitatively differ-ent roles with respect to a given outcome vari-able of interest as a function of structural dif-ferences across members of a sample. Unlessthe study design includes a deliberate attemptto identify such differences and classify theminto different samples, the coherence of as-sessing exposure to marital distress across allsubjects absolutely requires the assumption ofstructural homogeneity (see also Inferencesubsection). Note that the justification processis again necessarily anticipated at this point inthe selection of a restricted set of variablesto assess. The statistical requirements of thejustification process will restrict this set to asufficiently low ratio of variables to subjectsto minimize the risk of committing a Type 1error. In light of the possibility of structuralheterogeneity within the sample, the risk ofcommitting Type 2 errors is at least as great.But these latter risks will not figure promi-nently in the investigator’s choice of variablesbecause structural homogeneity is assumed.The emphasis, therefore, will be on assessingonly those variables that are directly relevantto understanding the possible causal role ofmarital distress on antisocial behavior, whichis assumed to hold or not hold uniformlyacross the entire sample.

Standardization. Similarly, the principle ofusing standardized measurement and assess-ment procedures in research is necessary forobvious reasons. In a sample of structurallyhomogenous individuals this is easily accom-plished by adhering rigorously to the samemethods and procedures for each subject. But,to the extent that members of the sample dif-fer qualitatively in their measurement-relevantstructures, it may be necessary to administerdifferent methods and procedures to different

subjects in order to achieve true standardiza-tion. This may seem counterintuitive at first,but it is merely an extension of the common-sense principle that requires altering assess-ment procedures to accommodate individualdifferences across subjects in language, age,grade level, cultural context, and so forth. Ob-viously it would make no sense to require thestandardized administration of oral interviewsin English to subjects who are non-Englishspeaking, deaf, or too young to comprehendthe meaning of the interview. These factorsreflect important structural differences acrossindividuals that are easily recognized on thebasis of overt functioning and are thereforeroutinely accommodated in the standardiza-tion process. Equally significant structural dif-ferences across subjects that are not easilyrecognizable on the surface may have majorinfluences on their interpretations of ques-tions, response patterns, thresholds for re-sponding, attributions about the purpose ofquestions, etc. To the extent that they are notobvious on the basis of overt functioning,however, the standardization process will pro-ceed on the simplifying assumption that anysuch differences are essentially random andwill therefore cancel each other out across thesample.

Data analysis

When assessments are completed on all sub-jects, the resulting data typically will be ex-amined for patterns of covariation betweenand among variables across individuals in thesample. Those patterns in turn will be inter-preted with reference to the presumed com-mon underlying structure(s). Again, on theassumption of within-class theory-relevantstructural homogeneity, the attempt to ac-count for patterns of covariation with a singletheory makes perfect sense. To the extent thatmembers of the sample are structurally het-erogeneous, however, a sample-based analysisof covariation makes no sense for addressingtheoretical questions about causal structure.To understand why, it is necessary to distin-guish carefully between the descriptive andinferential meanings of a correlation coeffi-cient.

Hubble hypothesis 215

Description. Assume, for example, that the in-vestigator has computed a correlation of r=.34 across the sample between individual dif-ferences in marital distress and children’s lev-els of antisocial behavior. What could the cor-relation mean descriptively? One possibilityis that the variables covary slightly but rela-tively uniformly across all children in thesample. Another is that they covary consider-ably for a small subset of children and not atall for others. There are numerous other de-scriptive possibilities depending of course onhow the two variables are distributed acrossthe sample in both univariate and bivariate(and multivariate, in the more complex case)space, each of which would (or should) in oneway or another place absolute constraints onany inferential interpretation. Some of thesepatterns (e.g., outliers) can be detected by spe-cialized statistical techniques, whereas othersrequire visual inspection of the bivariate ormultivariate scatterplot. Let us assume, how-ever, a textbook situation in which the covari-ation pattern is fairly evenly distributed acrosscases. Descriptively, we would be correct insaying that increments in levels of marital dis-tress are associated with increments in levelsof child antisocial behavior. The crucial ques-tion, of course, is what might the correlationmean inferentially?

Inference. In a closed system the inferentialpossibilities for a .34 correlation betweenvariables A and B would be a small influenceof A on B, a small influence of B on A, oran influence of some unmeasured variable orvariables (including measurement artifacts)on both A and B (Figure 3).5

Assuming that the values of these variableswere observed rather than experimentally ma-nipulated, any one of these models might beoperating in a closed system. Importantly, onthe assumption of structural homogeneityamong the objects across which the variableswere assessed, these are competing models inthe sense that only one can be correct. In an

5. In principle, of course, evidence that B preceded Awould rule out the A + B model. In practice, however,this is exceedingly more difficult to establish in longi-tudinal, sample-based individual differences researchthan most empirical reports would suggest.

B

B

B

A - B2525

II

0’0 5 1010 15 2020 2525

A

A -B2525

LL

010 5 1010 15 2020 2525

A

Figure 3. Three competing models of explanationfor a correlation of .34 between factors A and B ina simple closed system.

open system, however, equifinality raises thepossibility that different structural modelsmight be operating for different children orsubsets of children in the sample. The relevantcausal structures for some children may in-clude a diathesis for impulsivity (inherited oracquired), poor self-esteem, a violent homeenvironment and/or community, and minimalsupport or nurturance from caregivers. Forothers, the relevant structures may be positivecharacteristics such as high intelligence andresourcefulness in conjunction with condi-tions of poverty, poor parent monitoring, and/or a criminogenic neighborhood. And yet oth-ers for whom none of these factors is relevantmay have been socialized into a deviant value

216 Richters

2525

II

55 1010 1515 2020 2525

B: Child Antisocial Behavior

Figure 4. A hypothetical data set in which the .34 correlation between marital distress andchild antisocial behavior arising from theory-relevant structural heterogeneity within thesamnle. The symbols reflect children for whom marital distress is a cause, effect, or spuriouscorrelate of antisocial behavior.

system, or may be coping defensively in ahostile, dangerous environment.

Obviously, any number of qualitativelydifferent causal structures may give rise tosimilar patterns of antisocial behavior in dif-ferent children (Figure 4). Moreover, one caneasily imagine how marital distress might bea cause, consequence, or spurious correlatewithin structurally different variants of thesesubtypes. Thus, in addition to the three possi-bilities (cause, consequence, spurious corre-late) within a closed system, there are poten-tially unlimited combinations of possibilitiesfor the role of marital distress in a structurallyheterogeneous sample of children. Maritaldistress may be a powerful influence on anti-social behavior for a subset of children, a con-sequence of antisocial behavior for another,and yet a spurious correlate for others. Arti-facts resulting from unrecognized heterogene-ity may work against the investigator in twoways. First, they may engulf and mask a theo-retically important causal link between mari-tal distress and antisocial behavior that holdsfor only a subset of children. Second, for thesame reasons, they may give rise to an illu-sory correlation across the sample even ifmarital distress plays no causal role for evena subset of children. Thus, inferences aboutthe theoretical meaning of correlations bothobserved and not observed in structurally het-erogeneous samples may be equally treach-erous!

Justification

Recall (see Statistical versus theoreticalhypotheses subsection) that the investigatorenters the justification process with twohypotheses, one statistical, the other theoreti-cal. From the investigator’s standpoint the im-portant statistical question is-what is theprobability that the sample-based correlationof .34 between marital distress and child anti-social behavior exists in the population of in-terest. Following convention, the observedcorrelation will be tested against the Ho of noassociation in the population by conducting astatistical significance test. Assuming a suffi-cient sample size to achieve statistical signifi-cance, the investigator will reject H0 and thusfail to reject the statistical hypothesis of .34.This outcome, however, does not address theinvestigator’s question about the populationvalue. The significance level is an index ofthe probability of detecting a correlation of.34 or greater in a sample randomly drawnfrom a population in which the variables areunrelated (H0). Although it implies nothingabout what the correlation value might be inthe population, this is most often how the in-vestigator will interpret a statistically signifi-cant correlation. For purposes of argument,however, let us assume following Meehl thatthe investigator has computed the correlationnot on a sample but on the entire population.Thus, the statistical question of generalizabil-

Hubble hypothesis 217

ity disappears and the only question remain-ing is what does the correlation mean withreference to the investigator’s theory?

The purpose of the study of course was toevaluate a theory-based hypothesis about theinfluence of marital distress on children’s an-tisocial behavior. There is no statistical testfor evaluating a substantive hypothesis per se.Thus, the correlation itself (with the require-ment of statistical significance if it is sample

based) typically will be accepted as supportfor the investigator’s theory. Again, althoughthis inference is based on a logical fallacy of

. affirming the consequent (i.e., T > 0 :. 0 >T), it will be accepted as long as the theory isintuitively plausible and does not contradicteither a widely accepted premise or an estab-lished empirical fact. If the existing knowl-edge base is sufficiently advanced the investi-gator might be required to demonstrate thatthe correlation still holds while controllingstatistically for other variables. Moreover, ina highly developed research domain the evi-dentiary burden may be increased such thatthe investigator also must demonstrate that hisor her theory accounts as well or better forthe observed correlation(s) than an alternativetheoretical model. Such a direct test of com-peting theoretical models on the same data setis quite consistent with conventions of theorytesting in the established sciences.

But the logic of covariance procedures andcompeting model justification strategies restson the assumption of within-sample structuralhomogeneity. Moreover, this is a substantivetheoretical requirement not to be confused

. with a statistical requirement concerning thedistributional properties of a variable. Manywidely used statistical procedures are known

. to be quite robust to violations of their distri-butional assumptions. Thus, statistical re-quirements are commonly viewed as idealiza-tions that can be ignored with impunity formost practical purposes, and researchers areable to proceed on the basis of simplifyingassumptions that may or may not be true.Theoretical interpretations of sample-baseddata, however, are not robust to violations ofthe structural homogeneity assumption. Theyabsolutely depend on the assumption that allsubjects share the same theory-relevant struc-

tures, and that theory-irrelevant differencesamong them are relatively small and nonsys-tematic (Valsiner, 1986b). Moreover, theoreti-cal inferences are not merely weakened orqualified but invalidated by violations of thehomogeneity assumption.

Thus, to justify interpretations of correla-tions, partial correlations, semipartial correla-tions, interaction terms, and the like, an inves-tigator must assume that a single explanatorymodel is sufficient ultimately to account forindividual differences in the outcome measureacross all members of the sample. To the ex-tent that subjects differ qualitatively and notmerely quantitatively with respect to thestructure(s) of theoretical interest, however,everything changes. Not only is there no justi-fication for employing any of these strategieson a structurally heterogeneous sample, but ifthey are applied the results are likely to bemisleading in unpredictable ways. By defini-tion, any deviation from structural homogene-ity means that a minimum of two explanatorymodels will be required to account for thedata. Moreover, artifactual data patterns intro-duced by undetected structural heterogeneitymay favor or disfavor one or both theories re-gardless of their merits! Thus, even if boththeories are absolutely wrong, a correlation of.34 between marital distress and antisocial be-havior may nonetheless arise from artifactual(e.g., spurious or reverse causal) associationsresulting from the undetected presence ofstructurally different subsets of subjects in thesample. Conversely, if one of the theories iscorrect, these same artifacts can as easilymask the causally relevant association thatholds for a subset of cases. Given the smallassociation indexed by a .34 correlation (i.e.,12% shared variance), it would take only afew structurally different cases in a typicalsample to produce either of these untowardand undetectable effects. Note that the twodata columns from the investigator’s bivariatescatterplot presented in Figure 5 each includecases for which marital distress is a cause,consequence, and spurious correlate. Theslope of the regression line passing throughthese columns reflects the optimal least-squares trajectory based on an average of theirscores on each variable.

218 Richters

25

00 5 1010 1515 20 25B: Child Antisocial Behavior

Figure 5. An enlarged illustration of structural heterogeneity in two data columns from thebivariate scatterplot Figure 4.

In computing this solution. however, thealgorithm obviously is sensitive only to thequantitative properties of these variables; allscores are treated as qualitatively the same.The logic of the investigator’s interpretationof this correlation, however, is supremely de-pendent on the assumption that the data pointsdo not represent qualitative differences. Obvi-ously, the least-squares solution based onchildren characterized by the marital distressas causal structure is far different from the so-lution based on all column scores. Thus, the.34 correlation is a terribly misleading indexof the association in which the investigator isreally interested. Suppose, however, that theinvestigator took seriously the possibility ofstructural heterogeneity as reflected in thescatterplot data in Figure 5. A moment’s re-flection will reveal that the likelihood of be-ing able to discriminate between structurallydifferent cases with such similar scores in themiddle of a distribution is exceedingly low.

Constrained Solutions

Historically, the tendency of critics has beento consider psychology’s discovery and justi-fication weaknesses in relative isolation fromone another. Consequently, proposed solu-tions to paradigm problems in one contextoften have suffered from either ignoring orbeing overly constrained by unrecognizedweaknesses in the other. An understanding ofwhy these approaches are inadequate provides

a foundation and rationale for reaching be-yond the constraints of the current paradigmfor solutions to the developmentalist’s di-lemma.

Justif ication in context

Isolated considerations of the justificationproblem have resulted in calls to abandonnull-hypothesis testing in favor of more rigor-ous standards for evaluating theoretical prop-ositions. The most commonly recommendedreplacement standard has been to requiremore precise and “risky” theoretical predic-tions, usually indexed by narrower confidenceintervals or larger effect sizes. These recom-mendations seem reasonable as solutions toan isolated problem of justification. Withinthe practical constraints of the current discov-ery paradigm, however, it is difficult to imag-ine how these standards might be imple-mented. The relatively primitive state ofmeasurement technology in combination withthe very general nature of most theoreticalpropositions pose formidable obstacles to im-plementing more rigorous standards. It isn’tobvious, for example. whether a true improve-ment in most theories would result in an in-crease or decrease in the expected effect sizesof particular factors. In addition, artifacts re-sulting from correlated methods variance andvarious sources of sampling and measurementbias may overwhelm any theoretically ex-pected correlation in many domains. Relat-

Hubble hypothesis 219

edly, the sheer number of variables that wouldbe necessary to compare and test competingtheoretical models on the same data set wouldimpose formidable sample size requirementsin order to maintain sufficient statisticalpower for minimizing Type 1 errors and max-imizing sensitivity to effects.

When these recommendations are recon-sidered outside the artificial constraints froma developmental perspective, however, they

l are not merely impractical but often mis-guided. Until there is a reasonable basis forthe assumption that subjects in a sample are

. homogenous with respect to their theory-rele-vant structures, there is no justification fordrawing any conclusions-however tentative-about a unitary underlying structure. Inter-pretations of regression coefficients, partialcorrelations, semipartial correlations, and testsof competing theoretical models are no morejustified and are potentially as misleading asinterpretations of simple correlations whenbased on structurally heterogeneous samples.

Discovery in context

Similarly, efforts to wrestle with discovery is-sues have been undermined by the mutuallyenabling interdependence between the para-digm’s justification weaknesses and its hard-core structural homogeneity assumption. Forexample, even though the questionablenessof the homogeneity assumption is easy to un-derstand in the abstract, its practical con-sequences are obscured by the illusion ofprogress sustained by flawed justificationstandards. The effect of this illusion is to pre-define the researcher’s task as one of choos-ing an overt syndrome of interest, identifying

the right combination of putative causal fac-tors, and modeling complex interactionsamong variables in the service of characteriz-ing the unitary underlying causal structure.Presumed homogeneity also justifies a focuson maximizing the proportion of variance ex-plained in an outcome of interest by a fixedset of theory-relevant predictors. Within thissimplified framework, problems of discoveryare seen as resolvable by identifying addi-tional predictor variables and interactionterms, and employing more sophisticated data

analytic strategies to model complex interac-tions.

From a developmental perspective, how-ever, the search for more variables and moresophisticated data modeling techniques maybe grasps in the wrong direction. The coher-ence of these strategies absolutely dependson the assumption of structural homogene-ity across individuals. From a developmentalstandpoint this assumption is often highlyquestionable, defies common sense, and iscontradicted by the human capacities for equi-final and multifinal functioning. It is sustainedonly by the tranquilizing effect of the para-digm’s flawed justification standards, whichin turn are sustained and virtually required bydeficiencies in the discovery.

Local solutions

It would be wrong to conclude that these in-terdependent paradigm weaknesses alwaysand necessarily lead to flawed conclusionsand misguided research efforts. To be sure,developmental psychopathology research hasachieved significant progress in many areasdespite its paradigmatic handicaps (Cic-chetti & Cohen, 1995a, 1995b). Thoughtfulresearchers have long recognized the inade-quacies of the paradigm and have struggled towrestle in meaningful ways with their phe-nomena within its considerable constraints.This is the sub-rosa reality of the research en-terprise not reflected in the sterilized, formalreports of research findings published in jour-nal articles. The details and machinations ofcreative exploratory data analyses, for exam-ple, are seldom reflected in the literature.When results are negative they are often sup-pressed and not reported for sound conceptualand theoretical reasons. When outcomes arepositive they are often reported sans detailsabout how variables were combined and re-combined, and data analyses cast and recast,until they yielded patterns that made sense tothe investigator.

Although many of these practices are tech-nical violations of the statistical standards ofthe justification process, they are often justi-fied as reasonable adaptations to unreasonablerequirements. Ultimately, however, they are

220 Richters

inadequate as solutions to the developmental-ist’s dilemma for two reasons. First, local cir-cumvention strategies only free researchers-and partially at that-from the arbitrarystandards of the flawed justification process.The trade-off is that this also leaves research-ers with no uniform scientific standard at all,which is the essential self-correcting engineof scientific activity. Second, as indicated ear-lier, major decisions about study design anddata collection are so constrained by the the-ory-testing and statistical requirements of theparadigm that resulting data are often of lim-ited usefulness for purposes beyond the hy-pothesis testing objectives for which theywere collected (see Study design subsection).

Resolving Developmentalist’s Dilemma

It is one thing to understand the inadequaciesof the existing paradigm. It is quite another toengage the considerable challenges of devel-oping a suitable alternative for guiding andinterpreting developmental psychopathologyresearch. A useful starting point is to considerbriefly what the human capacities for equifi-nal and multifinal functioning do not implyabout the challenges ahead. First, they do notimply that individuals will differ structurallyto the extent that each must be studied as his/her own unique class. It is certainly the casethat individuals are literally unique in manyimportant respects. But the coherence of de-velopmental science rests on a reasonable ex-pectation that we are also sufficiently alike inimportant respects to warrant the search forstructural similarities across individuals andfor the laws governing the emergence, devel-opment, functioning, and functions of thosestructures. Second, equifinality and multifi-nality do not imply that all overt functioningpatterns will be characterized by theory-rele-vant structural heterogeneity. There may bemany overt characteristics for which the the-ory-relevant structures are relatively homoge-nous across individuals, and many others forwhich most individuals will be characterizedby a relatively small number of qualitativelydifferent structures. Third, they do not implythat all structural differences between individ-uals are relevant to all research questions.

There may be many overt syndromes thatstem from structures and processes that differqualitatively across individuals at one level oforganization but converge onto a final com-mon pathway at others. Thus, for purposes ofresearch, the alternative to structural homoge-neity is neither complete nor unmanageableheterogeneity.

From individual differences to differentindividuals

The possibility of multiple causal structuresunderlying similar patterns of overt function-ing does not merely add to the complexity ofstudying causal processes and structures. Itradically redefines the discovery challenge inways that require a fundamental reconceptual-ization of developmental psychopathology re-search. The reason, simply, is that the entireindividual differences research tradition abso-lutely depends on the simplifying assumptionthat overt functioning characteristics (e.g.,shyness, sociability, depression, aggression,self-esteem, etc.) are relatively straightfor-ward reflections of unknown, complex, butnonetheless unitary causal structures that arecommon across individuals. If this assump-tion is invalid, so also are the assumptionsthat (a) individual differences in the psycho-logical characteristics of subjects reflect onlyquantitative differences in their underlyingstructures, (b) each variable (e.g., internalstructure or external factor) under consider-ation either is or is not causally relevant to theovert characteristic, and therefore (c) itscausal role, if any, in producing that charac-teristic, however direct or indirect, will al-ways be the same. In short, a rejection of thestructural homogeneity assumption perforcerequires a rejection of research strategies,methods, and assessment procedures that de-pend for their coherence and justification onits validity.

Inevitably, understanding individual differ-ences in overt functioning will require under-standing the qualitatively different structuresfrom which they arise in the population. This,in turn, requires an initial shift from researchconcerning individual differences to researchconcerning different individuals. The unit of

l

Hubble hypothesis 221

analysis in individual differences research, ofcourse. is not an individual or group of indi-viduals. but the sample itself. Data points con-tributed by individuals within a sample aremerely used to construct an average hypothet-ical individual characterized by the structureshared in common by sample members. Thisstrategy loses its coherence if individualswithin the sample do not share a common the-ory-relevant structure. If there is reason to be-lieve that a given characteristic of functioningmay stem from different structures and pro-cesses in different individuals, then the imme-diate challenge is to determine how manyqualitatively different structures there are andwhich individuals are characterized by whichstructures. The necessary starting point forthis search is an initial intensive focus on thecausal structures and functioning of particularindividuals, only then followed by a searchfor others characterized by similar theory-rel-evant structures.

From causal factors to causal structures

The shift in focus from samples to individualsalso requires a fundamentally different ap-proach to identifying causal structures andprocesses. In sample-based research, thesearch for causal structure is typically ap-proached with bottom-up strategies. That is,models of causal structure are constructed bycombining individual factors and processesthat have been identified as correlates of overtfunctioning patterns. But the coherence of thisapproach depends on the simplifying assump-tion that a given factor either is or is not caus-ally related to the phenomenon of interest. Tothe extent that different causal structures maygive rise to the same phenomenon across dif-ferent individuals, the logic underlying bot-tom-up strategies breaks down. Any causal re-lation(s) between a variable or set of variablesand the overt characteristic of interest may bevirtually undetectable in a general searchacross structurally heterogeneous cases. Thus,the challenge of identifying causal structureat the individual level will require top-downstrategies that begin at the highest levels ofbroadly defined explanatory constructs and

work systematically down the hierarchy tospecific processes and factors.

A useful framework for thinking about top-down discovery strategies is provided byMackie’s concept of causally sufficient condi-tions in the naturalistic world (Mackie, 1974;see also Russell, 1974, for an extension ofMackie’s work on development). In the studyof closed systems in physics, theories fre-quently can be refined to the point of beingable to specify conditions that are both neces-sary and sufficient for the occurrence of aphenomenon (Elasser, 1981). Many effects inthe natural world, however, can be producedby a variety of conditions that are neither nec-essary nor sufficient for their occurrence. Abuilding fire, for example, may be caused byany number of different conditions, such asan act of arson, an electrical short-circuit, acarelessly discarded cigarette, or a bolt oflightning, etc. Each of these conditions is un-necessary but sufficient for producing a build-ing fire under certain circumstances (e.g., thepresence of combustible material, oxygen,etc.). Moreover, each condition consists of adifferent constellation of elements, each ofwhich is causally insufficient but contributesnon-redundantly to the causal efficacy of thatcondition. The elements of the arson condi-tion, for example, might include a flammableliquid, a lighted match, and a person intendingto bum the building down. Alone, each ele-ment is causally insufficient for producing afire; it is only in combination that they consti-tute the arson condition. Thus, they are insuf-ficient, nonredundant components of an un-necessary but sufficient (inus) condition (i.e.,arson) for producing the building fire effect.

Different causal structures that are capableof producing similar characteristics of overt-functioning can be conceptualized usefully asinus conditions. The added complexity ofstudying inus conditions underlying humanfunctioning, of course, is that any given factormay play a qualitatively different causal roleacross different inus conditions. As in our ear-lier example, factors such as marital distress,parent symptomatology, poor peer relations,school failure and the like may be causal inone or more inus conditions for antisocial be-havior, consequential in others, and spurious

222 Richters

in yet others. Nonetheless, the inus concept isuseful for thinking initially about unnecessarybut sufficient causal conditions. Consider, forexample, the variety of potential inus condi-tions that might give rise to a pattern of per-sistent antisocial behavior in childhood. Onecondition might be a criminogenic neighbor-hood and/or family environment in which an-tisocial behavior is modeled, expected, and/orrewarded by major influences in a child’s life.This is certainly a plausible model to considerin late 20th century America, where in manymajor cities the allure of drug-related crime isever present, and where gangs virtually con-trol the social commerce and economic life ofmany neighborhoods (Richters, 1996). It isnot difficult to imagine that the causal struc-tures responsible for antisocial behavior inthese children will differ in important respectsfrom those giving rise to antisocial behavioramong children raised in a mainstream culturecharacterized by fundamentally different in-fluences, socialization pressures, and opportu-nities. Even within each of these differentsubcultures there may be children whose anti-social behavior was initiated and/or is sus-tained primarily by the need for money tosupport drug or alcohol dependence, an ex-pression of anger against parents or adult au-thority figures, or a method of attention seek-ing. Persistent antisocial behavior in otherchildren may stem primarily from inherited oracquired deficits in nervous system function-ing that translate into problems of impulsecontrol or difficulties learning from experi-ence either absent or irrespective of theseother conditions.

identified most easily. The value of inten-sively studying well characterized exemplarsof carefully defined structures has been amplydemonstrated in the developmental literature(Hinde, 1992; Hinde & Dennis, 1986; Kagan,1992; Kavale & Forness, 1987).

These are not, of course, mutually exclu-sive sources of influence, and there are nodoubt many children whose antisocial behav-ior arises from complex combinations of theseand other sources of influence. The necessarybasis for understanding these more complexcausal structures, however, may be an under-standing of the structures characteristic ofchildren for whom more primary sources ofinfluence can be identified. This does not im-ply the study of extreme cases: it is actuallythe opposite. It is a call for focusing initiallyon children for whom the primary reasons orexplanations for antisocial behavior can be

Referring back to our earlier example, theinitial emphasis would be on identifying themost common primary causes of buildingfires (i.e., Unnecessary but sufficient condi-tions), rather than their individual compo-nents (i.e., insufficient, nonredundant ele-ments). Thus, approaching the structuralheterogeneity challenge in the domain of chil-dren’s antisocial behavior would begin by de-veloping an initial taxonomy of putative inusconditions-primary reasons or explanations-for persistent antisocial behavior. The work-ing assumption would be that qualitativelydifferent inus conditions would be reflected inthe qualitative nature, history, and patterns ofantisocial behavior across time and contexts.A formulation of prototype descriptions forcandidate inus conditions would then providea foundation and rationale for the intensivestudy of individuals typified by those condi-tions. The assumptions underlying this boot-strapping strategy may be fundamentallywrong, and there may be obvious reasons whyit is unlikely to succeed. An overemphasis onthese possibilities, however, would miss thepoint. Overcoming the weaknesses of the ex-isting paradigm and developing suitable alter-natives will require a willingness to proceedfrom common sense, to accept the inevitabilityof false starts and mistakes, and to value andharness the corrective feedback they provide.

The next step, of course, would be to iden-tify individual children with antisocial behav-ior patterns consistent with a particular inuscondition. This would not initially requiremethods for reliably classifying antisocialchildren into a candidate inus condition.Rather, the initial goal literally would be toidentify a single child for whom there are con-crete, plausible reasons for believing that heor she is a likely prototype of a particular inuscondition. Note that the shift in focus fromsample to individual fundamentally redefinesthe challenges and opportunities of the dis-covery process.

l

Hubble hypothesis 223

Consider, for example, the contrast be-tween sample-based strategies for studyingthe causes of antisocial behavior and the morecommon individual-based strategy employedin practical nonresearch settings. In a sample-based study the researcher will probably nevermeet or interview the antisocial child and his/her parents. Moreover, on the basis of idio-syncratic, pragmatic considerations oftenloosely linked to a particular theoretical no-

* tion, the research design will focus on collect-ing a fixed, circumscribed set of informationthat may or may not be relevant to different

children in the sample, thereby necessarily ig-noring information that might be decisivelyimportant to understanding any particularchild. Furthermore, in the service of scientificrigor the researcher will require that all infor-mation be collected in a standardized wayfrom the child and his/her family, typically inthe form of standardized questionnaires and/or checklists and/or structured interviews. Bydefinition, the standardized process will ig-nore many characteristics of individuals thatmay have an important influence on how theyunderstand, interpret, and respond to ques-tions. Also, parcels of information concerningeach subject will be collected in isolationfrom one another with no effort to evaluatethem configurally in an effort to determinelikely and unlikely explanations for the child’santisocial behavior. In fact, the assumption ofstructural homogeneity renders this task un-necessary. Because the causes of antisocialbehavior are presumed to be qualitatively thesame across children in the sample, evidence

will be sought through data analytic proce-dures identifying patterns of covariation be-tween and among variables across the sample.

When the unit of analysis shifts from sam-ple to individual each of these restrictions isnecessarily relaxed. The burdens of discoverybecome not unlike those of a parent, clinician,social worker, or probation officer confrontedwith an antisocial child. In these settingsthe discovery process is directed at gainingenough of an understanding of the nature, his-tory, and current context of the child’s prob-lems and circumstances to formulate a hy-pothesis about what seems to be motivating,causing, and influencing the child’s antisocial

behavior and based on that hypothesis, an ini-tial strategy for intervening. Although the re-searcher’s objective is not to formulate an in-tervention strategy, the discovery challengeand objective is very much the same. The cli-nician would probably begin by collectingcertain common information about the child,parents, history, and current circumstanceswith respect to every child. But the clinicianwill also remain flexible about reaching be-yond these initial boundaries for additional in-formation depending on the specifics of eachparticular case. He or she will also make lib-eral use of alternative wording, rephrasing,asking questions in different ways, probing,mirroring, and a range of other interview tech-niques for ensuring that the parents and childunderstand the intended meaning and nuanceof questions asked and that the clinician is in-terpreting accurately the intended meaning oftheir answers. The clinician also may decideto seek information from additional infor-mants depending on the nature of the case,looking for points of convergence and diver-gence, sources of and motivations for bias,and so forth, in the process of trying to makesense of the different streams of information.The clinician also will pay very careful atten-tion to the nature of timing of various kindsof events and circumstances relative to thechild’s behavior patterns, because they can bedecisive in ruling in or out of contention cer-tain potential explanations for the child’s be-havior. Also, the clinician will want access toall relevant pieces of information about thechild, family history, and circumstances at thesame time in the service of making complexconfigural judgments about which factorsseem to be most important, deserve the mostexplanatory weight, and have the most clini-cal significance in understanding the child’sbehavior. This configural decision is decisivebecause the meaning and significance to theclinician of any particular piece of informa-tion can hinge exquisitely on the rest of thepattern of information in the matrix.

The nature of this undertaking immediatelyraises three major questions. The first iswhether it is necessary or desirable to requireconfigural, clinical, inferential judgmentsabout the meaning of the information con-

224 Richters

cerning a child. After all, there is an impres-sive body of data extending back decadesshowing that the clinical judgments are oftenfar inferior to fixed procedures and computeralgorithms when it comes to combining datato make predictions (Meehl, 1954). This liter-ature. however, concerns the very specificcase of the comparative accuracy of humansand computers in predicting or classifying aphenomenon according to a known formula.The problem posed by structural heterogene-ity is that we do not yet know what the for-mula is or should be. Paul Meehl, in a follow-up article to his classic book Clinical versusstatistical prediction was quite explicit aboutconditions under which human judgments areindispensable: “Sometimes there is no for-mula because the prediction problem is tooopen-ended . sometimes the very categoriz-ing of the raw observations involves gestatedstimulus equivalences for which the laws areunknown, and hence cannot be formulated(although the clinician himself exemplifiesthese laws and can therefore ‘utilize’ them);in still other cases there is no formula becausenobody has bothered to make one” (Meehl,1954; Meehl, 1973, p. 83). Meehl also identi-fied several decision-making conditions underwhich it is necessary to rely on clinical judg-ment because there is nothing else available.The first is when the decision task is openended in the sense that the content of thejudgment itself must be decided by the clini-cian. The second is when the decision taskrequires a recognition of unanalyzed stimulusequivalences, such as perceptual gestalts, psy-chological similarities in physically dissimilarevents. These are recognition tasks for whichthe clinician implicitly relies on various formsof analogical and primary-process thinking,the algorithms of which are beyond the levelof awareness. A third condition is when a pat-tern of information is relatively rare, or in anyevent not expected, but its appearance is suffi-ciently relevant to warrant contramanding amore typical interpretation. Fourth, there areconditions wherein judgments rely on hypo-thetical mental constructs of such a generalnature that they permit a wide range of con-crete manifestations that can be recognizedbut not anticipated in advance. These condi-

tions are likely to be prominent in any initialattempt to recognize behavior and informationpatterns that are plausible reflections of par-ticular inus conditions. Thus, configural hu-man judgments at the point of data assessmentmay be indispensable in the initial stages ofindividual-based strategies.

The second issue is whether there is a con-siderable potential for capitalizing on chancein making configural judgments at the indi-vidual level. Suppose, for example, that a fac- tor or experience that seems to the researcherto be central to understanding a particularchild’s antisocial behavior turns out to be so common among non-antisocial children that itis neither a risk factor nor correlate of antiso-cial behavior in the population. In sample-based research, on the assumption of struc-tural homogeneity, this would often be suffi-cient for questioning the causal relevance of afactor. A rejection of the homogeneity as-sumption, however, invalidates this logic.Again, it is quite possible for a factor to becausally relevant in one inus condition for anoutcome and not others, such that it may notbe related to antisocial behavior in a structur-ally heterogeneous population. This does notimply that the clinical judgment is correct,only that it is not invalidated by informationbased on the general population.

Rethinking justification

This leads naturally to the third and most dif-ficult question of how to guard against illu-sory explanations when focusing on an indi-vidual. It is obviously quite easy in a clinicalor research setting as it is in daily life to be-lieve passionately in explanations that mayhave no merit whatever. Thus, if a clinicaljudgment about the importance of a factorcannot be evaluated by reference to its overallrelation to antisocial behavior in the generalpopulation, by what criteria can it be evalu-ated? This is the most important of the threeissues because sound justification standardsprovide the indispensable self-correcting feed-back for shaping and insuring the integrity ofcreativity in the discovery process. In general,judgments made at the individual level mustbe evaluated by qualitative criteria such as co-

r

Hubble hypothesis 225

herence. explanatory power, and the ability toaccount for facts that are otherwise difficultto explain, especially when applied to individ-uals with putatively similar causal structures.

The crucial question of how to translatethese general principles into concrete criteriaand standards for evaluating individual-fo-cused research will be one of the major chal-lenges of emphasizing greater creativity andexploration in the discovery process. This willbe especially challenging in developmentalpsychopathology research because the com-plexities of its phenomena are such that ameaningful discovery process will require awide range of research strategies, methods,and procedures. One of the many lessons ofthe current paradigm is that no single methodor research strategy should be held out a priorias inherently superior or inferior to another.The strengths and weaknesses of a given dis-covery methodology can be evaluated onlywith specific reference to the phenomena andresearch questions under study. A guidingprinciple for thinking about discovery meth-ods and strategies should be methodologicalpluralism, a willingness to develop and em-brace whatever methods and strategies arenecessary for understanding the phenomenaof interest, including categorical, dimensional,idiographic, nomothetic, variable-based, indi-vidual-based, cross-sectional, longitudinal, ex-perimental, quasiexperimental, historical, andethnographic approaches (Bergman & Mag-nusson, in press; Bhatara, McMillin, & Krum-mer, 1995; Gould, 1986, 1987; Magnusson &Allen, 1983; Magnusson & Bergman, 1988;Maton, 1993; McCord, 1993; Shadish, 1995;Shames, 1990).

Methodological pluralism in the discoveryprocess cannot flourish within the constraintsof an arbitrarily restrictive and misguided(and misguiding) justification process. Be-yond the flaws in its logic and standards, thecurrent justification paradigm constrains thediscovery process to a narrowly defined rangeof activities in the service of ritualized hy-pothesis testing. The requirements imposed

tential and forces researchers into prematuretests of theories. It does not preclude moremeaningful approaches to discovery, but itdoes engender and reinforce research prac-tices that are often antithetical to sound dis-covery. In addition, the narrow emphasis on“explained variance” as the chief criterion forevaluating theories often relegates explana-tory power to a secondary or tertiary status byemphasizing prediction.

Although the current justification processlacks rigor for testing theories, it is not obvi-ous that a continued emphasis on formal the-ory testing is either practical or desirable in‘many areas of developmental psychopathol-ogy research. In most domains we are notdealing with formal theories in the conven-tional ideal sense, but with very general theo-retical notions about particular aspects ofcomplex phenomena. The emphasis of thecurrent paradigm on theory testing has led toa narrow focus on maximizing the “ex-plained” variance in outcome measures of in-terest with shifting sets of predictor variables.In many domains of children’s social andemotional development and psychopathologywe continually struggle against what seems tobe an upper limit of explained variance,which can often be reproduced with very dif-ferent constellations of variables. In the end,it is seldom clear even within the forgivingstandards of the justification paradigm whatinterpretations of multivariate effects are jus-tified or how they relate to the original phe-nomena of interest (Cairns, 1986; Kaplan,1996). A more reasonable and practical ap-proach to scientific rigor under these circum-stances would be to emphasize questions thatare answerable in straightforward ways andanswers that are interpretable and justifiableby reasonable standards. The most construc-tive direction for reform would be to replacethe current emphasis on ambitious claimsbased on weak evidence with a justificationprocess based on the time-honored principlesof strong inference (Platt, 1964).

by hypothesis testing, population generaliz- Conclusionability, minimizing Type 1 errors, and main-- __taining low variables-to-subjects ratios often Differences between mature and fledgling sci-rob the discovery process of its creative po- ences are often less apparent in the errors they

226 Richters

commit than in the speed with which they rec-ognize and respond to those errors. By mostaccounts, astronomy’s launching of the myo-pic Hubble Space Telescope was one of the20th century’s leading scientific blunders.Nonetheless, it took astronomers only mo-ments to recognize that Hubble was transmit-ting distorted images of the universe back toearth, only days to diagnose precisely, and amatter of months to correct. Psychologylaunched its equivalent of the myopic Hubblewhen it founded its science on the closed sys-tem assumptions and explanatory models of19th century classical physics. By itself, thiserror would have been only a short-termhandicap. Like all mistaken ideas and prem-ises in science it would have been remediedeventually by the rigors of scientific method.By relying on null-hypothesis significancetesting procedures to evaluate research find-ings, however, psychology deprived itself ofthe essential self-correcting engine of scien-tific method. As a consequence, psychology’sreaction time in recognizing and correcting itsinitial error has been measured so far in de-cades rather than months.

Psychology’s long-standing paradigm prob-lems are especially salient to the newly for-malized discipline of developmental psy-chopathology for three related reasons. First,developmental psychopathology stands poisedto define a new center of gravity for integra-tive, multidisciplinary research in many areasof psychological science because of the tre-mendous heuristic power of developmental,open system framework. Second, the researchstrategies, methods, and data analytic proce-dures of psychology’s paradigm on which de-velopmental psychopathology currently de-pends are fundamentally inadequate inimportant respects to the study of complexopen systems. As a consequence, much of therichness and heuristic power of the develop-mental framework is lost when empirical re-search is conceptualized and carried outwithin the constraints of the current paradigm.Resolving the developmentalist’s dilemmawill require more than a recognition of the in-adequacies of the existing paradigm. It willrequire intensive efforts to develop indige-nous research strategies, methods, and stan-

dards with fidelity to the complexity of devel-opmental phenomena.

It would be wrong to conclude that the cur-rent paradigm always leads to wrong conclu-sions and misguided research efforts. None-theless, a dispassionate assessment of itsmutually enabling discovery and justificationweaknesses provides a reasonable basis forconcern about the possibility that, in manyimportant areas of research, we may be un-knowingly (and, within the constraints of the paradigm, unknowably) studying artifactsrather than facts, flawed theory and flaweddata, and illusory phenomena of our own cre- ation. In short, we may be committing Type 3errors by answering the wrong questionswhen we should be answering the right ques-tions (Mitroff & Featheringham, 1974). It isgenerally true in science that as long as thetools of a paradigm prove capable of solvingthe problems it defines, science functions bestby relying on those tools. The reason, as Kuhnnoted, is clear: “ . . . in manufacture so in sci-ence-retooling is an extravagance to be re-served for the occasion that demands it”(Kuhn, 1970, p. 76). It is difficult to image amore compelling occasion for retooling thanthe emergence of an exciting new disciplinefor which the existing paradigm is inadequateand within which the necessary tools do notexist.

Ultimately, the responsibility for construc-tive reform rests with each individual investi-gator. The lesson of psychology’s history,however, is that individual efforts are unlikelyto flourish or succeed in the absence of amore broadly based commitment to reform.The responsibility and power for engenderingand maintaining such a commitment, in turn,rests with institutions of scientific leadershipand influence. Editorial boards of scientificjournals and public and private funding agen-cies are in unique positions to implement thenecessary incentive structures for encouragingefforts toward paradigm reform. Develop-mental psychopathology’s scientific successesin the 21st century will hinge decisively on itswillingness and ability to translate sophisti-cated developmental thinking into equally so-phisticated developmental research designs,methods, and practices. The heuristic value of

Hubble hypothesis 227

the developmental psychopathology frame-work is nowhere more evident than in theclarity with which it reveals the nature of psy-

References

Allport. G. W. (1960). The open system in personalitytheory. Journal of Abnormal and Social Psychology,61. 301-310.

Altman, I., & Rogoff, B. (1987). World views in psychol-l ogy: Trait, interactional, organismic, and transactional

perspectives. In D. Stokols & I. Altman (Eds.), Hand-

book of environmen!al psychology (pp. 740). NewYork: Wiley.

American Psychological Association. (1974). Publicationmanual of the American Psychological Association(2nd ed.). Baltimore: Garamond/Pridemark Press.

Anastasi, A. (1958). Differential psychology (3rd ed.).New York: Macmillan.

Bakan, D. (1966). The test of significance in psychologi-cal research. Psychological Bulletin, 66, 423-437.

Baldwin, J. M. (1895). Mental development of the childand the race: Methods and processes. New York:Macmillan.

Beach, F. A., & Jaynes (1956). Studies of maternal re-trieving in rats, III: Sensory cues involved in the lac-tating female’s response to her young. Behaviour, 10,104-125.

Bergman, L. R., & Magnusson, D. (in press). A person-oriented approach in research on developmental psy-chopathology. Development and Psychopathology.

Bevan, W. (1991). Contemporary psychology: A tour in-side the onion, American Psychologist, 46, 475-483.

Bevan, W., & Kessel, F. (1994). Plain truths and homecooking: Thoughts on the making and remaking ofpsychology. American Psychologist, 49, 505-509.

Bhatara, V. S., McMillin, .I M., & Krummer, M. (1995).Aggressive behavior and fire-setting in a 4-year-oldboy associated with ingestion of ground beef contami-nated with bovine thyroid tissue: A case report andreview of neuropsychiatric thyrotoxicosis. Journal ofChild and Adolescent Psychopharmacology 5, 255-271.

Brandt, L. W. (1973). The physics of the physicist andthe physics of the psychologist. International Journalof Psychology, 8, 61-72.

Bronfenbrenner, U. (1979). The ecology of human devel-opment: Experiments by nature and design. Cam-bridge, MA: Harvard University Press.

Cairns, R. B. 1986). Phenomena lost: Issues in the studyof development. In .I. Valsiner (Ed.), The individualsubject and scientific psychology (p p. 97-l12). NewYork: Plenum Press.

Cicchetti, D. (1984). The emergence of developmentalpsychopathology. Child Development, 55, 1-7.

Cicchetti, D. (1990). A historical perspective on the disci-pline of developmental psychopathology. In J. Rolf,A. Masten, D. Cicchetti, K. Nuechterlein, & S. Wein-traub (Eds.), Risk and protective factors in the devel-opment of psychopathology (pp. 2-28). New York:Cambridge University Press.

Cicchetti, D. (1993). Developmental psychopathology:Reactions, reflections, and projections. Develop-

Gigerenzer, G., Swijtink, Z., Porter, T., Daston, L.,Beatty, J., & Kruger, L. (1989). The empire ofchurzce: HOW probability changed science and every-day life. Cambridge: Cambridge University Press.

Glymour, C. (1980). Hypothetico-deductivism is hope-less. Philosophy of Science, 47, 322-325 .

Gould, S. J. (1986). Evolution and the triumph of homol-ogy, or why history matters. American Scientist, 74,60-69.

Gould, S. J. (1987). Darwinism defined: Sifting fact frommental Review, 13, 471-502. theory. Discover, 8, 64-70.

chology’s long-standing paradigm problemsand in the powerful prescription it providesfor a corrective lens.

Cicchetti, D., & Cohen, D. J. (Ed%). (1995a). Develop-mermd psychopathology: Vol. 1. Theory and methods.New York: John Wiley & Sons, Inc.

Cicchetti, D., & Cohen, D. J. (Eds). (1995b). Develop-mental psychopathology: Vol. 2. Risk, disorder, undadaptation. New York: John Wiley & Sons, Inc.

Cicchetti, D., & Richters, J. E. (1993). Developmentalpsychopathology considerations in the study conductdisorder. Development and Psychopathology, 5, 33 I-344.

Cicchetti, D., & Rogosch, F. A. (1996). Equifinality andmultifinality in developmental psychopathology. De-velopment and Psychopathology, 8, 597-600.

Cohen, J. (1990). Things I have learned so far. AmericanPsychologist, 45, 1304-I 3 12.

Cohen, J. (1994). The earth is round (p < .05). AmericanPsychologist, 49, 997-1003.

Cronbach, L. J. (1957). The two disciplines of scientificpsychology. American Psychologist, 12, 671-684.

Edwards, W. (1965). A tactical note on the relation be-tween scientific and statistical hypotheses. Psycholog-ical Bulletin, 63, 400-402.

Elasser, W. M. (1966). Atom and organism A new ap-proach to theoretical biology. Princeton, NJ: PrincetonUniversity Press.

Elasser, W. M. (1981). Principles of a new biological the-ory: A summary. Journal of Theoretical Biology, 89,131-150.

Emde, R. N. (1994). Individuality, context, and the searchfor meaning. Child Development, 65, 719-737.

Estes, W. K. (in press). Significance testing in psycholog-ical research: Some persisting issues. PsychologicalScience.

Evans, C., & McConnell, T. R. (1941). A new measureof introversion-extroversion. Journal of Psychology,12, 111-124.

Ferguson, L. (1959). Statistical analysis in psychologyand education. New York: McGraw-Hill.

Feynman, R. (1986). Surely you’re joking, Mr. Feynman!New York: Bantam Books.

Fox, R. E. (1996). Charlatanism, scientism, and psycholo-gy’s social contract. American Psychologist, 51, 777-784.

Gerstein, D. R., Luce, R. D., Smelser. N. J., & Sperlich,S. (1988). The behavioral and social sciences:Achievements and opportunities. Washington, DC:National Academy Press.

228 Richters

Guilford, J. P. (1942). Fundamental statistics in psychol- Loftus, G. R. (1991). On the tyranny of hypothesis testingogy and education (3rd ed.). New York: McGraw- in the social sciences. Contemporary Psychology, 36,Hill. 102-105.

Herrenkohl, L. R., & Rosenberg, P. A. (1972). Exterocep-tive stimulation of maternal behavior in the naive rat.Physiology & Behavior, 8, 595-598.

Hinde, R. A. (1992). Developmental psychology in thecontext of other behavioral sciences. DevelopmentalPsychology, 28, 1018-1029.

Loftus, G. R. (1993). Editorial comment. Memo ry & Cog-nition, 21, 1-3.

Hinde, R. A., & Dennis, A. (1986). Categorizing individ-uals: An alternative to linear analysis. InternationalJournal Behavioral Development, 9, 105-l19.

Horgan, J. (1996). Why Freud isn’t dead. American Sci-entist, December, 106-l11.

Institute of Medicine. (1989). Research on children andadolescents with mental, behavioral, and develop-mental disorders. Washington, DC: National Acad-emy Press.

Kagan, J. (1992). Yesterday’s premises, tomorrow’spromises. Developmental Psychology, 28, 990-997.

Kagan, J.. Kearsley, R. B., & Zelazo, P. R. (1978). In-fancy: Its place in haman development. Cambridge,MA: Harvard University Press.

Kaplan, H. B. (1996). Toward an understanding of resil-ience: A critical review of definitions and models. InM. D. Glantz, J. Johnson, & L. Huffman (Eds.), Resil-iency and development: Positive life adaptations.New York: Plenum Press.

Kavale, K. A., & Fomess, S. R. (1987). The far side ofheterogeneity: A critical analysis of empirical subtyp-ing research in learning disabilities. Journal of Learn-ing Disabilities, 20, 374-382.

Koch, S. (1959). Psychology: A study of science (Vol. 3).New York: McGraw-Hill.

Lykken, D. T. (1968). Statistical significance in psycho-logical research. Psychological Bulletin, 70, 151-159.

Lykken, D. T. (1991). What’s wrong with psychologyanyway? In D. Cicchetti & W. M. Grove (Eds.),Thinking clearly about psychology (Vol. 1, pp. 3-39).Minneapolis, MN: University of Minnesota Press.

Mackie, J. L. (1974). The cement of the universe: A studyof causation. Oxford: Oxford University Press.

Magnusson, D. (1985). Implications of an interaction-al paradigm for research on human development. In-temational Journal of Behavioral Development, 8,115-137.

Magnusson, D., & Allen, V. L. (1983). Implications andapplications of an interactional perspective for humandevelopment. In D. Magnusson & V. L. Allen (Eds.),Human development: An interactional perspective(pp. 369-387). New York: Academic Press.

Magnusson, D., & Bergman, L. (1988). Individual andvariable-based approaches to longitudinal research onearly risk factors. In M. Rutter (Ed.), Studies of psy-chosocial risk: The power of longitudinal data (pp.45-61). Cambridge: Cambridge University Press.

Maltz, M. D. (1994). Deviating from the mean: The de-clining significance of significance. Journal of Re-search in Crime and Delinquency, 31, 434-436.

Maton, K. I. (1993). A bridge between cultures: Linkedethnographic empirical methodology for culture-an-chored research. American Journal of CommunityPsychology, 21, 701-121.

Koch, S. (1961). Psychological science versus the science-humanism antinomy: Intimations of a significant sci-ence of man. American Psychologist, 16, 629-639.

Koch, S., & Leary, D. E. (Eds.). (1992). A century ofpsychology as science. Washington, DC: AmericanPsychological Association Press.

Kuhn, T. S. (1957). The Copernican revolution. Chicago:University of Chicago Press.

Kuhn, T. S. (1970). The structure of scientific revolutions(2nd ed.). Chicago: University of Chicago Press.

Kuhn, T. S. (1977). The essential tension. Chicago: Uni-versity of Chicago Press.

McCord, J. (1993). Descriptions and predictions: Threeproblems for the future of criminological research.Journal of Research in Crime and Delinquency, 30,412425.

McGuire, W. J. (1983). A contextualist theory of knowl-edge: Its implications for innovation and reform inpsychological research. Advances in Experimental .Social Psychology, 16, 241.

Lakatos, I. (1977). The methodology of scientific researchprogrammes :Philosophical papers (Vol. I). J. Wor-ral & E. G. Zahar (Eds.). Cambridge: Cambridge Uni-versity Press.

Meehl, P. E. (1954). Clinical versus statistical prediction:A theoretical analysis and review of the evidence.Minneapolis: University of Minnesota Press.

Meehl, P. E. (1967). Theory testing in psychology andphysics: A methodological paradox. Philosophy ofScience, 34, 103-I 15.

Laudan, L. (1977). Progress and its problems: Toward atheory of scientific growth. Berkeley: University ofCalifornia at Berkeley.

Leahey, T. H. (1987). A history of psychology: Main car-rents in psychological thought (2nd ed.). New York:Prentice-Hall.

Meehl, P. E. (1973). Psychodiagnosis: Selected papers.Minneapolis, MN: University of Minnesota Press.

Meehl, P. E. (1978). Theoretical risks and tabular aster-isks: Sir Karl, Sir Ronald, and the slow progress ofsoft psychology. Journal of Consulting and ClinicalPsychology, 46, l-42.

Lerner, R. M., & Kaufman, M. B. (1985). The conceptof development in contextualism. Developmental Re-view, 5, 309-333.

Leventhal, L. (1992). Nudging aside Meehl’s paradox.Canadian Psychology, 35, 283-298.

Lewin, K. (1931a). The conflict between Aristotelian andGalilean models of thought in contemporary psychol-ogy. Journal of General Psychology, 5, 147-117.

Lewin, K. (1931b). Environmental forces in child behav-ior and development. In C. Murchison (Ed.), A hand-book of child psychology (2nd ed., pp. 590-625).Worster, MA: Clark University Press.

Meehl, P. E. (1990). Appraising and amending theories:The strategy of Lakatosian defense and two principlesthat warrant it. Psychological Inquiry. I, 108-141.

Mitroff, I., & Featheringham, T. (1974). On systematicproblem solving and the error of the third kind. Be-havioral Science, 19, 383-393.

Morrison, D. E.. & Henkel. R. E. (Eds.). (1970). The sig-nificance test controversy. Chicago: Aldine.

Nagel, E. (1961). The structure of science. New York:Harcourt, Brace. and World.

National Advisory Mental Health Council. (1990). Na-tional Plan for Research on Child and AdolescentMental Disorders. Rockville, MD: National Instituteof Mental Health.

I

Hubble hypothesis 229

New York Times. (1990, July 1). Calamities in space. Sec-tion 1. Editorial desk, p. 16.

Nunnally, J. C. (1975). Introduction to statistics in psy-chology and education. New York: McCraw~Hill.

Overton. W. F. (1984a). World views and their influenceon psychological theory and research: Kuhn-Lakatos-Laudan. In H. W. Reese (Ed.), Advances in child Development and behavior (Vol. 18, pp. 19 l-225). NewYork: Academic Press.

Overton, W. F. (1984b). Comments on Beilin’s epistemol-ogy and Palermo’s defense of Kuhn. In H. W. Reese(Ed.). Advances in child development and behavior(Vol. 18, pp. 273-276) . New York: Academic Press.

Overton, W. F., & Horowitz, H. A. (1991). Develop-mental psychopathology: Integrations and integra-tions. In D. Cicchetti & S. Toth (Eds.), RochesterSymposium on Developmental Psychopathology: Vol.3. Models and integrations (pp. 142). Rochester,NY: University of Rochester Press.

Pepper, S. C. (1942). World hypotheses. Berkeley: Uni-versity of California.

Platt, J. R. (1964). Strong inference: Certain systematicmethods of scientific thinking may produce much morerapid progress than others. Science, 146, 347-362.

Popper, K. R. (1959). The logic of scientific discovery.London: Hutchinson.

Popper, K. R. (1991). Of clouds and clocks: An approachto the problem of rationality and the freedom of man.In D. Cicchetti & W. M. Grove (Eds.), Thinkingclearly about psychology.(Vol. 1, pp. 100-l39). Min-neapolis, MN: University of Minnesota Press.

Radke-Yarrow, M., & Sherman, T. (1990). Hard grow-ing: Children who survive. In J. Rolf, A. Masten, D.Cicchetti, K. Nuechterlein, & S. Weintraub (Eds.),Risk and protective factors in the development of psy-chopathology (pp. 97-l19). New York: CambridgeUniversity Press.

Reichenbach, H. (1938). Experience and prediction. Chi-cago: University of Chicago.

Richters, J. E. (1996). Disordered views of antisocialchildren: A late 20th century perspective. In C. Ferris(Ed.), Understanding aggressive behavior in children.New York: New York Academy of Sciences.

Richters, J. E.. & Cicchetti, D. (1993a). Editorial: Towarda developmental perspective on conduct disorder. De-velopment and Psvchopathology, 5, 1-4.

Richters, J. E., & Cicchetti, D. (1993b). Mark Twainmeets DSM-III-R: Conduct disorder, development.and the concept of harmful dysfunction. Developmentand Psychopathology, 5, 5-29.

Robinson, D. N. (1984). The new philosophy of science:A reply to Manicas and Secord. American Psyholo-gist, 39, 920-921.

Rorer, L. G. (1991). Some myths of science in psychol-ogy. In D. Cicchetti & W. M. Grove (Ed.). Thinkingclearly about psychology (Vol. 1. pp. 61-87). Minne-apolis, MN: University of Minnesota Press.

Rozeboom, W. W. (1982). Let’s dump hypothctico-de-ductivism for the right reasons. Philosophy of Sci-ence, 49, 637-647.

Russell, J. (1990). Causal explanations in cognitive de-velopment. In G. Butterworth & P. Bryant (Eds.),Causes of development: Interdisciplinary perspectives(pp. 111-134). London: Harvester, WheatsheafHemel Hemstead.

Rutter, M., & Garmery. N. (1983). Developmental psy-chopathology. In E. M. Hetherington (Ed.), Carmi-chaels manual of child psychology: Vol. 4. Socialand personality development (pp. 775-911). NewYork: Wiley.

Salmon, W. C. (1984). Scientific explanation and thecausal Structure of the world. Princeton, NJ:Princeton University Press.

Sameroff, A. (1995). Developmental systems: Contextsand evolution. In D. Cicchetti & D. Cohen (Eds.), De-velopmental psychopathology: Risk, disorder and ad-uptation (pp. 238-294). New York: Wiley-Intersci-encc.

Sameroff, A. J., & Chandler, M. J. (1975). Reproductiverisk and the continuum of caretaking casualty. In F.D. Horowitz. M. Hetherington, S. Starr-Salapatek, &G. Siegel (Eds.), Review of child development re-search (pp. 187-244). Chicago: University of Chi-cago Press.

Sarason, S. B. (1981). Psychology misdirected. NewYork: The Free Press.

Shadish, W. R. (1995). Philosophy of science and thequantitative-qualitative debates: Thirteen common er-rors. Education and Program Planning , I8, 63-75.

Shames, M. L. (1990). On data, methods, and theory: Anepistemological evaluation of psychology. CanadianPsychology, 31, 229-238 .

Smith. P. M., & Torrey. B. B. (1996). The future of thebehavioral and social sciences. Science, 271, 61, 1-612.

Spemann. H. (1938). Embryonic development and induc-tion. New Haven, CT: Yale University Press.

Sroufe. L. A., & Rutter, M. (1984). The domain of devel-opmental psychopathology. Child Development, 83,173-189.

Thurstone. L. L. (1938). Primary mental abilities. Chi-cago: University of Chicago Press.

Valsiner. J. (Ed.). (1986a). The individual subject and sci-entific psychology. New York: Plenum Press.

Valsiner. J. (1986b). Between groups and individuals:Psychologists’ and laypersons’ interpretations of cor-relational findings. In J. Valsiner (Ed.), The individualSubject and scientific psychology (pp. 113-l52). NewYork: Plenum Press.

von Bertalanffy. L. (1968). General system theory. NewYork: George Braziller. Inc.

Waddington. C. H. (1957). The strategy of genes . Lon-don: Allen & Unwin.