Doing Q methodology: theory, method and interpretationsites.nd.edu/lapseylab/files/2014/10/Watts_2005_Doing-Q...that constitute obstacles to the use of one of the very first ‘alternative’

Doing Q methodology: theory,method and interpretation

Simon Watts1 and Paul Stenner2

1Nottingham Trent University, Burton Street, Nottingham NG1 4BU, UK; 2University College London,Gower Street, London, WC1E 6BT, UK

This paper has a marked practical aspect. We wish to encourage andfacilitate the use of Q methodology amongst psychologists interested inqualitative research. The paper duly answers a number of pertinent ‘how todo Q’ questions. Yet our primary intention is not to produce an exhaustive‘how to do Q’ guide. In discussing issues of theory, method andinterpretation in Q methodology, the main aim of the paper is ratherto address some of the more common misunderstandings andmisrepresentations that constitute obstacles to the use of one of the veryfirst ‘alternative’ methods to have been developed in the context ofpsychology. In addressing such obstacles, the paper hopes to bring ‘Whydo Q?’ questions to the fore. In so doing, Q methodology will alsobe ‘positioned’ in relation to a number of other qualitative researchmethods, each of which currently enjoys a degree of prominence withinthe psychological discipline. Qualitative Research in Psychology 2005; 2:67!/91

Key words: eigenvalues; factor analysis; factor extraction; factor inter-pretation; factor rotation; measurement; Q methodology; Q set; Q sorting;subjectivity; William Stephenson

Introduction

This paper has a marked practical aspect.We wish to encourage and facilitate the useof Q methodology amongst psychologistsinterested in qualitative research. This willinevitably involve our paying attention to

the ‘nuts and bolts’ of the method. Yet ourprimary intention is not to produce anexhaustive ‘how to do Q’ guide. In discuss-ing issues of theory, method and interpreta-tion in Q methodology, our main aim israther to address some of the commonmisunderstandings and misrepresentations

Correspondence: Simon Watts, Psychology Division, School of Social Sciences, Nottingham Trent University,Burton Street, Nottingham NG1 4BU, UK.E-mail: [email protected]

# 2005 Edward Arnold (Publishers) Ltd 10.1191/1478088705qp022oa

Qualitative Research in Psychology 2005; 2: 67!/91www.QualResearchPsych.com

that constitute obstacles to the use of one ofthe very first ‘alternative’ methods to havebeen developed in the context of psychol-ogy (Stephenson, 1935). Whilst, therefore,the paper will provide a deal of practicalinformation about the conduct of Q metho-dological research (such that a number ofpertinent ‘how to do Q’ questions will beanswered), it also aims to clarify how Qmethodology can be understood and posi-tioned in relation to some of the otherqualitative research methods that are cur-rently available to the psychologist. In sodoing, we hope to bring ‘why do Q?’ ques-tions to the fore.

To illustrate some typical misunderstand-ings of Q methodology, we turn first to arecent volume on psychological researchmethods by Haslam and McGarty (2003).We wish to stress that these misunderstand-ings are exceedingly common, and hence nospecial criticism should be directed at theseparticular authors. Indeed, it is to theircredit that they discuss Q methodology atall, given that many texts on qualitativepsychology do not (for example, Smith,2003). Nonetheless, their discussion illus-trates two of the most commonly committederrors. The first involves a misattribution oforigins: ‘In its original form participantsplaced 100 cards with statements aboutpersonal characteristics into piles rangingfrom ‘‘not characteristic of me’’ to ‘‘verycharacteristic of me’’’ (Smith, 2003: 364).But this is not Q methodology in its ‘origi-nal’ form. In fact, Haslam and McGarty areinstead referring to an unorthodox andsomewhat problematic application of Qmethodology, which is typically associatedwith Carl Rogers (1954), and which emergednearly 20 years after William Stephensonhad originally devised the methodology.

The second common error (which wasalso committed by Rogers) involves a

separation of the two fundamental aspectsof Q methodology: the Q sorting procedure(which is an original means of collectingdata) and the Q pattern analysis (which iseffected by means of a by-person factoranalysis). As we shall see, Stephensondesigned the former precisely in order toenable the legitimate application of thelatter. Indeed, it was the effective combina-tion of the two aspects !/ the Q sort and theQ (as opposed to R) technique factor analy-sis !/ that allowed Stephenson to makesubjectivity his principle research focus(the distinction between Q and R techni-ques will be clarified below). Rogers gath-ered data in the form of Q sorts, but did notsubject those sorts to a by-person factoranalysis. Cyril Burt conducted by-personfactor analyses on alternative forms of data(see Note 1 and Burt, 1937). Yet neither wasdoing Q methodology, because each em-ployed only one aspect of Stephenson’sprocedure.

These misunderstandings alone areenough to ‘muddy the waters’ of Q metho-dology. Sadly, however, its data analyticand data collection procedures are alsofrequently misrepresented . There is, forexample, a tendency for Q methodologicalfactor analysis to be erroneously (mis)iden-tified with its more familiar R methodolo-gical incarnation, and hence to be viewed asa ‘statistical method of data reduction thatidentifies and combines sets of dependentvariables that are measuring similar things’(McGarty and Haslam, 2003: 387). As weshall see, however, Q methodology makesno such psychometric claims. The methodemploys a by-person factor analysis in orderto identify groups of participants who makesense of (and who hence Q ‘sort’) a pool ofitems in comparable ways. Nothing morecomplicated is at issue.

68 S Watts and P Stenner

The Q sort itself also comes to be mis-represented as a passive ‘response dimen-sion’ (McGarty and Haslam, 2003: 364)upon which ‘statements’ are simply rankedinto piles, rather than as a dynamic mediumthrough which subjectivity can be activelyexpressed (Stephenson, 1953). This limitedand overly passive conceptualization of theQ sort once again emerges from Carl Rogers’application. Rogers used a Q sort composedof statements about personal characteristicsas a means of providing a standard measureof self, of ideal self, and of the distancebetween them. Yet, for the dedicated Qmethodologist, a completed Q sort indicatesonly that a set of items have been differ-entially valued by a specific participantaccording to some face valid and subjectivecriterion. This is not a measurement (forreasons which will become apparent).Rogers’ approach may have its own merits,therefore, but the image it creates is none-theless unhelpful for Q methodology, inas-much as it thoroughly confuses the Rmethodological ambition for researcher-ledobjective measurements (or psychometrics)with the Q methodological pursuit of parti-cipant-led subjective expressions and view-points .

Such confusions could easily be avoided(and we hope this paper will make asignificant contribution in this regard). Weadmit, however, that they result in partfrom the fact that Q methodology’s quanti-tative features render it a highly unusualqualitative research method (Curt, 1994;Watts and Stenner, 2003a). The method isactually qualiquantological (see Stennerand Stainton Rogers, 2004). This fact alonemay discourage some qualitative research-ers from engaging with Q methodology, asmany will have turned to qualitative meth-ods precisely in order to escape the parti-cular strand of quantitative logic and the

associated hypothetico-deductive methodsthat have long since claimed the name‘science’ in psychology (and in many otherrelated disciplines). It is also easy to avoidQ methodology in the belief that it offersnothing that one of a number of formsof textual analysis cannot do better andin a more straightforwardly ‘qualitative’fashion (Willig, 2001). Discourse analysis(Potter and Wetherell, 1987), narrative ana-lysis (Crossley, 2000) and interpretativephenomenological analysis (Smith, 1996),for example, may seem to offer more pala-table alternatives in this regard.

It would nonetheless be unfortunate wereany qualitative researcher to reject Q meth-odology for either of these reasons. It makeslittle sense, for example, as Harre (2004) hasrecently suggested within these pages, toavoid all forms of mathematical and quanti-tative representation (or to develop a knee-jerk aversion to science) simply becausethey have often been poorly employed bypsychologists. Qualitative methodologistshave rightly challenged the currently extant‘scientific’ or hypothetico-deductive ap-proach in psychology, but (and despite itsquantitative content) it is important torecognize that Stephenson’s Q methodologywas actually performing a similar functionlong before any significant qualitative tradi-tion had been established in the discipline(Stephenson, 1935).

In fact, Q methodology was designed forthe very purpose of challenging the dated,Newtonian logic of ‘testing’ that continuesto predominate in psychology. It also of-fered an early critique of the cognitiveassertion that people can properly bedivided into a series of psychological‘parts’. This same critique has, of course,subsequently become a typical feature of‘constructionist’ approaches in psychology(see, for example, Harre, 1999). In these

Q methodology 69

ways at least, Q methodology is a typicallyqualitative and a very critical method. Thesurprise, perhaps, is that it achieves itscritical stance through the embrace (ratherthan the rejection) of many natural scienti-fic assumptions. Indeed, having gaineddoctorates in both physics and psychology,Stephenson actually based his ‘challenge’ tothe Newtonian tradition in psychology onepistemological and ontological presump-tions (as well as mathematics) that weremore familiarly associated with thequantum mechanics of physics (see Wattsand Stenner, 2003a; Stephenson, 1936a,b,1988/89 for a full discussion of theseissues). Q methodology duly lends supportto Harre’s (2004: 4) recent assertion that it is‘the qualitative techniques and the meta-physical presumptions which back themthat come much closer to meeting the idealsof the natural sciences’.

Neither should one mistakenly believethat Q methodology does the same job asany of the other textual methods availableto the psychologist. Of course, one mustimmediately acknowledge that this is trueof all the textual methods, which havingdifferent emphases necessarily perform sub-tly different tasks. It is apparent, for exam-ple, that some of these methods prefer a‘paradigmatic’ (or thematic) form of analy-sis, whilst others take a more holistic or‘narrative’ approach (see Polkinghorne,1995). In the vast majority of cases, how-ever, it is the machinations of everyday talk,the specific situation or person, and thegeneration (and explication) of individualviewpoints which constitute the primaryresearch targets. None of this is ordinarilytrue of Q methodology.

This is not to deny that Q method-ology can be, and has been, effectivelyemployed as a powerful technique for singlecase studies of various kinds (Smith, 2001;

Stephenson, 1953). Yet many qualitativemethods are equally capable of such ana-lyses and repertory grid studies providefurther serious competition in the domainof the personal construct (Smith, 1995). Thecase-study approach is, however, not atypical application of Q methodology, espe-cially as it has come to be employed within(primarily British and social) psychology(see Stainton Rogers et al ., 1995). In thisguise, Q methodology ordinarily adopts amultiple-participant format and is mostoften deployed in order to explore (and tomake sense of) highly complex and sociallycontested concepts and subject mattersfrom the point of view of the group ofparticipants involved (Stainton Rogers,1995; Watts and Stenner, 2003a). It doesnot do this in a thematic fashion, nor does itfocus on the viewpoints of specific indivi-duals. It should be no surprise, therefore, tofind that this typical form of Q methodologydisappoints when themes and/or indivi-duals are the primary research targets.

To properly appreciate Q methodology,we need instead to recognize that it isessentially a gestalt procedure (Good,2000). This gestalt emphasis means it cannever ‘break-up’ its subject matter into aseries of constituent themes (which imme-diately distinguishes Q from various formsof discursive or interpretative phenomeno-logical analyses). What it can do, however,is show us the primary ways in which thesethemes are being interconnected or other-wise related by a group of participants. Inother words, it can show us the particularcombinations or configurations of themeswhich are preferred by the participantgroup. This openly holistic approachsuggests that Q methodology is most closelyrelated to Crossley’s (2000) ‘narrativeanalysis’.


Nevertheless, it differs from narrativeanalysis in at least three ways. First, andmost obviously, it does not deal with parti-cipants’ own discourse as such, but invitesparticipants to engage in the unusual task ofrelating (in a complex and in-depth way)with a set of prepared items. This ‘unusualtask’ evidently violates the cherished prin-ciple of ‘naturalism’. One can counter thisargument, however, inasmuch as the veryidea of ‘naturally occurring discourse’ ishighly problematic (Potter, 1997). There isalso no a priori reason to assume thatunfamiliar activities cannot yield usefulinsights. Of course, qualitative researchersshould never underestimate the impact ofresearch context on findings. Neithershould they ‘leap’ from findings to unwar-ranted knowledge claims. Yet it would beequally wrong were they to misrecognize‘naturalism’ (or any method or methodolo-gical rule) as some kind of absolute standard(Watts, 2000).

Secondly, Q methodology is not wellsuited to dealing with the unfolding tem-porality of narratives. Narrative analysisactively pursues this temporality and thenexamines the resultant stories in terms oftheir temporal structure (e.g., beginning,middle, end) and function. Q methodologyforsakes this important form of analysis inorder to pursue a ‘snap shot’ or temporallyfrozen image of a connected series of subjectpositions (or ‘view-points’). It then exam-ines these (methodologically frozen) posi-tions in terms of their overall structure,function and implications. And thirdly,whereas a narrative analysis focuses on thenarrative of specific individuals , Q metho-dology typically focuses on the range ofviewpoints that are favoured (or which areotherwise ‘shared’) by specific groups ofparticipants. In other words, the typical Qmethodological study very deliberately pur-

sues constructions and representations of asocial kind (Moscovici, 1981).

These differences allow Q methodologyto offer the psychologist a quite unique formof qualitative analysis. Indeed, in accentingthe group and their shared viewpoints, thisform of analysis provides an ideal (andnoticeably more macroscopic) complementto qualitative approaches which highlightthe ‘theme’ and/or ‘the individual’ in psy-chology. In the remainder of this paper, wewant to demonstrate this complementarityat work. In so doing, we hope also to allowthe interested reader to carry out effective Qmethodological studies.

History, theory and statistics:‘inverting’ the tradition

Q methodology was originally establishedvia a simple adaptation of the quantitativetechnique known as factor analysis. Thisadaptation was first introduced by WilliamStephenson in 1935. In this introductionand subsequent papers, Stephenson indi-cated that the conventional factor analyticprocedure could also be ‘inverted’:

Factor analysis . . . is concerned with a selectedpopulation of n individuals each of whom hasbeen measured in m tests. The (m)(m"/1)/2intercorrelations for these m variables are sub-jected to . . . factor analysis. The technique, how-ever, can also be inverted. We begin with apopulation of n different tests (or essays, pic-tures, traits or other measurable material), eachof which is . . . scaled by m individuals. The(m)(m"/1)/2 intercorrelations are then factorisedin the usual way (Stephenson, 1936b: 344!/45).

Such an inversion may appear mundane,but it actually initiates a series of quitedramatic methodological departures fromthe psychological tradition (Stephenson,1936b). First, it is the ‘n different tests ormeasurable materials’, not the participant

Q methodology 71

group, that become the study sample. Sec-ondly, the ‘variables’ are no longer testsor hypothesized traits, but the various per-sons who take part in the study. In otherwords, persons become the variables ofinterest in an inverted (or ‘Q’) study. Suchstudies actively explore ‘correlations be-tween persons or whole aspects of persons’(Stephenson, 1936b: 345). As a consequenceof these changes, it is also persons (not tests,traits or other types of variables) that loadonto the emergent factors of an invertedfactor analytic study.

These methodological changes are highlysignificant, then, in the sense that theydistinguish Q methodology from almostevery other popular psychometric techni-que used in psychology (these are knowncollectively as ‘R’ methodologies). They arealso indicative of possible theoretical de-partures from the psychological tradition.1

Stephenson was indeed very concernedabout the unremitting dominance of hy-pothetico-deductive methods within psy-chology and the concomitant neglect ofwhat he considered to be its proper subjectmatter: subjectivity. The following quote isone of a number that might have been usedfor illustrative purposes:

The hallmark of sound scientific procedurenowadays, it seems, is to assert hypotheses andto confirm predictions . . .There is need, how-ever, for care and discernment in thesematters . . . Psychology . . .has by no meansachieved a sophisticated theoretical status, withideal constructs such as physics has fashionedfor itself. The situations in psychology, therefore,call for an attitude of curiosity, as well as one ofhypothetico-deductive logic . . .We should bemaking discoveries rather than testing our rea-soning (Stephenson, 1953: 151).

Sadly, Stephenson’s warnings went largelyunheeded. Despite initial optimism and hisbelief that the inverted factor techniquewould allow the development of a new

psychology based upon ‘individual differ-ences in type ’ (Stephenson, 1936a: 205),Stephenson never truly managed to estab-lish his Q technique within British psychol-ogy. Meeting strong resistance to theoreticalchange and having failed to claim a Chair atOxford, Stephenson departed for the USA(and posts outside psychology) in 1947.

In recent years, however, similar concernshave again been expressed within psychol-ogy by a growing band of qualitative andcritical researchers (see, for example, Harreand Gillett, 1994). Like Stephenson, propo-nents of a qualitative psychology tend tovalue curiosity and to promote discoveryand understanding in preference to the logicof testing. They have also questioned thevery existence of the unobservable entitiesthat psychological tests invariably set out tomeasure and have duly rejected the subse-quent employment of these entities as ex-planatory devices (see Potter and Wetherell,1987). The ideal constructs of psycho-logy are, indeed, less than ideal. Couplethese conclusions with the centralitythat Stephenson’s inverted techniqueaffords to persons (in their ‘whole aspect’),as well as its stated intention to pursue‘empirical discoveries of a qualitative kind’(Stephenson, 1936a: 205), and one is onceagain reminded of the close relationshipthat connects Stephenson’s method with themodern qualitative tradition in psychology(Harre, 1999; Watts and Stenner, 2003a,b).

It should be no surprise, therefore, to findthat Q methodology has somewhat belat-edly found a ‘psychological’ home for itselfwithin that qualitative tradition. It hasinformed social constructionist studies(see, for example, Kitzinger, 1987; Stennerand Marshall, 1995; 1999), been adopted byfeminist researchers (Kitzinger, 1986; Senn,1996), and has even spawned its owntheoretical or ‘critical polytextualist’ take


on social psychology (Curt, 1994; Stennerand Eccleston, 1994). Q methodologyhas now been successfully applied tomany diverse psychological topics like, forexample, child abuse (Stainton Rogers andStainton Rogers, 1992), jealousy (Stennerand Stainton Rogers, 1998), I.B.S. (Stenneret al ., 2000), attitudes to environmentalissues (Capdevila and Stainton Rogers,2000) and love (Stenner and Watts, 1998).In so doing, it has more than demonstratedits ‘sense-making’ capacity and ability tofind qualitative ‘order’ even in domainswhere variability and disparity seem initi-ally to have prevailed (see, for example,Watts and Stenner, forthcoming).

Of course, Stephenson had no such trackrecord or qualitative tradition on which torely as he developed Q methodology.Neither would the discipline allow himthe luxury !/ now afforded to many qualita-tive psychologists !/ of turning to variouscritical philosophers as a means of securingtheoretical support (Brown and Stenner,2001; Stenner, 1993; 1996; 1998). It isperhaps understandable in such circum-stances, to find that Stephenson insteadturned to Freud for theoretical sustenance,especially given that psychoanalytic no-tions still permeate and influence a deal ofqualitative work in psychology (for exam-ple, Billig, 1997; Parker, 1997a,b). Indeed,the basic Freudian premise of pleasure/unpleasure lies at the very heart of the Qmethodological procedure (see Stephenson,1983) !/ a procedure which, perhaps equallyunderstandably, actually seems to resemblemany quantitative methods more closelythan it does its qualitative counterparts.

This procedure involves a heterogeneousset or sample of items (ordinarily a set ofstatements about a particular subject matter,although pictures, objects, and so on, mightalso be employed) being ranked or scaled

(along a standardized ranking distributionor continuum) by a group of participants.They were to do this according to their ownlikes and dislikes and hence as a function ofthe personal value they assigned to eachitem. The items would thus be rankedaccording to (what Stephenson called) their‘psychological significance’ for any givenparticipant (see Burt and Stephenson,1939). Those items of great psychologicalsignificance would evidently be ‘ranked orscored highly, whilst those of little relativesignificance . . . [would be] ranked or scoredlowly’ (Stephenson, 1936a: 347). The gen-eral idea was to render a formerly hetero-geneous set of items ‘homogeneous withrespect to . . . [a particular] individual’(p. 346), such that a single and gestaltconfiguration of items emerged.

As we have already implied, this sort ofranking procedure has some drawbacks fora method which now claims itself as aqualitative ‘alternative’, insofar as it canmake Q methodology look too much likethe tests, scales and questionnaires it pur-ports to challenge. Yet the role of theparticipants in Q methodology, the natureof the gathered data, and the way in whichthose data are subsequently interpreted,could not be more different from contextsin which testing and measurement areprioritized. It is well documented, for ex-ample, that test procedures necessitate acertain a priori imposition of meaning.Such meaning must effectively be ‘builtinto’ the measurement instrument itself(see Maraun, 1998). Each separate item ina conventional attitude scale is assigned anexact predefined meaning and is henceforthconsidered to represent a partial measure ofthe otherwise imperceptible psychologicalconstruct that the scale sets out to measure.To complete such a test is to be subjected tomeasurement . The resultant data will be

Q methodology 73

treated as an independent series of absolutemeasurements of you, and will be inter-preted only in relation to the operationaldefinition (of the psychological construct)that the researcher has chosen to employ.The possibility that differences of kind, ofmeaning, interpretation or ‘quality’ may (beat issue or) have informed responses to thetest is rarely considered. Indeed, qualitativeresearchers rightly chastise these methodsfor this lack of consideration (Potter andWetherell, 1987).

Q methodology, on the other hand,neither tests its participants nor imposesmeanings a priori . Instead, it asks its parti-cipants to decide what is ‘meaningful’ andhence what does (and what does not) havevalue and significance from their perspec-tive . A series of absolute measurementscannot result from this process. Instead, asingle set of essentially relative evaluations(and hence a gestalt configuration of items)is produced. These gestalt configurationshave been made by the participants, on thebasis of criteria which are personal to them(i.e., that which they consider to be ‘psy-chologically significant’), and it is thesegestalt configurations which constitute theresearch target of Q methodology.

It would, given the rationale presentedabove, be quite pointless (and a violation ofthe procedure) for the Q methodologist tostrictly predefine the meaning of the itemspresented to a participant, for any givenitem can evidently take on its ‘psychologi-cal significance’ (and hence its meaning fora participant) only in the context of anoverall configuration. Being the intendedresearch target, it is these overall configura-tions (not test results or measures) whichare then intercorrelated and factor analysedin a Q study (see Stephenson, 1936a). Thisproduces a set of factors (onto which parti-cipants load on the basis of the configura-

tions that they produce) which areexemplified and represented, not by differ-ent subsets of the presented items (as wouldbe the case in conventional factor analysis),but by all the presented items configured indifferent but characteristic ways. The mean-ing and significance of these configurationsmust then be ‘attributed a posteriorithrough interpretation rather than througha priori postulation’ (Brown, 1980: 54). Thisis the basis of the Q ‘method’. In thesections which follow, we shall discussthe mechanics of this procedure.

The generation of a Q set or ‘itemsampling’

The Q set is the collection of ‘heterogeneousitems’ which the participants will sort.There are many possibilities in this context.Stephenson (1936a), for example, per-formed early studies looking at people’spredilection for vases (using a Q set ofvases) and the hedonic value of certainodours (using a Q set of bottled fragrances).He also studied issues of personality byasking participants to value a population ofmoods (e.g., cheerful, elated, affectionate,etc.) as descriptors of their own personality.Such single word stimuli work very well(see Watts, 2002). It is more usual, however,in a qualitative and psychological context,for a Q set to be constituted of statements ,each making a different (but nonethelessrecognizable) assertion about the appropri-ate subject matter.

In keeping with its rejection of hypothe-tico-deductive logic, it is not surprising tofind that the formulation of specific hypoth-eses is typically inappropriate in Q metho-dology. As is true of many other qualitativemethods (and in response to its factoranalytic heritage), Q methodology is primar-


ily an exploratory technique. It cannotprove hypotheses. It can, however, bringa sense of coherence to research questionsthat have many, potentially complexand socially contested answers (StaintonRogers, 1995).

The research question plays a veryimportant part in any Q methodologicalstudy. This is because it dictates the natureand structure of the Q set to be generated.It will also act as a ‘condition of instruction’for the participants, and will hence guidethe actual sorting process. It is important,therefore, that the research question (or aclose approximation to it) is clearly definedbefore the study proper commences. Theresearch question must be straightforwardand clearly stated (containing a single pro-position only), and the Q set should enableparticipants to respond to the question inan effective fashion. If, for example, theresearch question is ‘What is partnershiplove?’, all the statements of the Q set mustrepresent possible responses to that ques-tion. If participants are asked to describetheir current love relationship, then all thestatements must represent possible descrip-tions of love relationships, and so on.

Whatever the research question, the Q setmust always be broadly representative ofthe opinion domain at issue. The generationof potential statements need not be theory-driven, however, as would be customary inthe design of a test or questionnaire. In-stead, it is carried out as a sampling task. AsCurt (1994: 128!/9) suggests, this is ‘oneplace where Q-method is noticeably a craft’.In common with other crafts, the Q metho-dologist must carry out this task skilfully,patiently and with an appropriate applica-tion of rigour. As a result, the generation ofthe final Q set can often ‘take up the bulk ofthe time and the effort involved (oftenseveral months of work in contrast to the

few hours or at most weeks involved inadministering the Q-sort)’ (Curt, 1994: 120).

In practice, the final Q set can be elicitedfrom any number of sources: by extensivereference to the academic literature (whichgenerally identifies and breaks down asubject matter into its key ‘themes’), fromboth literary and popular texts (magazines,television programmes, etc.) from formalinterviews, informal discussions and oftenvia pilot studies. A complete set of test orscale items can even be used to create aready-made Q set (see, for example, Watts,2002). In the end, the exact nature of thesampling task is of little consequence pro-vided that the final Q set can justifiablyclaim to be ‘broadly representative’ of therelevant opinion domain, and this aimmight clearly be satisfied in a number ofdifferent ways.

The exact size of the final Q set will, to agreat extent be dictated by the subjectmatter itself. Generally speaking, however,a Q set of somewhere between 40 and 80statements is considered satisfactory (cf.Curt, 1994; Stainton Rogers, 1995). Anyless than this and issues of adequate cover-age may be a problem. Any more and thesorting process can become unnecessarilyunwieldy. It is always best, however, toinitially generate an overly large number ofstatements, which can then be refined andreduced through processes of piloting.2

Even with effective piloting, however,there is a sense in which a Q set can neverreally be complete (as there is always ‘some-thing else’ that might potentially be said).Yet this is actually of little import, for theprocedural detail of Q methodology ensuresthat a Q set only needs to contain arepresentative condensation of information .This is because the main concern in a Qmethodological context is not the Q setitself (which is, in any event, not consid-

Q methodology 75

ered to possess any specific meaning priorto the sorting process), but the relative likesand dislikes, meanings, interpretations andoverall understandings which inform theparticipants’ engagement with the Q set. Inpractice, the qualitative detail of a Q meth-odological study actually gets filled out asthe study proceeds (rather than being ‘can-celled out’ as can often happen in thecontext of a test), with the subjective view-points of the participant group being centralto this process.

An example of this principle may beuseful. Watts and Stenner (forthcoming)used a Q set statement which suggestedthat ‘love is an intense commitment to apartner ’. The accuracy of this ‘hypothesis’is confirmed by the academic literature (see,for example, Fehr, 1988; Sternberg, 1986).Yet such confirmation says little about themeaning of commitment, nor does it tell ushow (or indeed ‘where’) the notion ofcommitment fits into people’s wider expec-tations and understandings of love.

In the study which followed, participantsgenerally considered this ‘commitment’statement to be a very good descriptor ofpartnership love (such that it was frequently‘ranked highly’). It was possible to observe,however, that it was ranked highly for avariety of quite different reasons : (a) be-cause some participants viewed partnershiplove as a long-term, practical style of com-mitment, such that ‘intensity’ implied athoroughgoing and enduring provision ofsupport; (b) because partnership love wasviewed as a short-term sexual and emo-tional commitment, such that ‘intensity’came to imply a total immersion in passion;and even (c) because the commitment oflove was feared, such that ‘intensity’ wasexperienced in terms of the placing of one’sself, emotions and trust in the hands ofanother person, and so on.

It becomes very obvious, therefore, notonly that the statement ‘love is an intensecommitment ’ contains a condensation ofpotential information, but also that thiscondensation was unfolded, unpacked andotherwise expressed in very different waysas the participant group engaged with thepresented items (and notice that we havecompletely ignored the participants whogave this statement a more lowly ranking).This unfolding of meaning can be observedacross every statement of the Q set. One canappreciate, therefore, how qualitative detailproliferates and ‘fills out’ as a Q studyproceeds.

In taking advantage of these processes ofparticipant engagement, Q methodology ex-ploits what Harvey (1997: 146!/47) hascalled ‘one of psychology’s most basic andwell established principles’, namely, ourdesire to structure and to ascribe meaningto all ‘impinging stimuli and events’. In-deed, it is these very desires which ensurethe robustness of Q methodology, as a groupof participants will ultimately make vigor-ous attempts to impose their viewpointsonto any set of statements they are given.In other words, even a ‘less than ideal . . .[Q set], because it invites active configura-tion by participants (‘effort after meaning’),may still produce useful results’ (StaintonRogers, 1995: 183). If a Q set is at least‘broadly representative’ of its subject matter,therefore, the engagement of the participantgroup with that Q set (and the resultantconfigurations) will afford a general over-view of relevant viewpoints ‘on the subject’(which is all that is required for the pur-poses of Q methodology). We will provide apractical demonstration of this process laterin the paper using the example Q set (about‘young offenders and their punishments’)outlined in Table 1.


The ranking procedure or ‘Q sorting’

The Q methodological ranking procedure !/

Q sorting !/ is what Brown (1980: 17) calls‘the technical means whereby data areobtained for factoring’. In practice, it issimply a convenient means of facilitatingthe (evaluations and) rankings of the parti-cipants. For reasons of simplicity and prag-matism, participants are not ordinarilyrequired to carry out a complete (1 to n )rank ordering of the Q set items. Instead,they simply assign each item a rankingposition in a fixed quasi-normal distribu-tion (illustrated in Figure 1 below) and‘along a simple, face-valid dimension, forexample most agree to most disagree, mostcharacteristic to most uncharacteristic, mostattractive to most unattractive’ (StaintonRogers, 1995: 180). An 11 or a 13 pointscale is generally employed, possible rank-ing values ranging from #/6 or #/5 for itemsthat are, say, ‘most agreeable’ in the viewof a particular participant, through ‘zero’, to"/5 or "/6 for items that are considered‘most disagreeable’.

Figure 1 demonstrates that the distribu-tion also dictates the number of items thatcan be assigned to each ranking position (in

the example, two items at the #/6 position,three at #/5, and so on). For this reason, it isalso known as a ‘forced’ distribution. Thisapparent forcing (or restriction) of partici-pants may alarm some qualitative research-ers. Yet such concerns are largelymisplaced. It is, in fact, quite possible toemploy different forms of distribution in thecontext of a Q methodological study, in-cluding completely ‘free’ distributions,which (as the name suggests) allow partici-pants to assign any number of items to anyof the available ranking positions.3

Yet Brown (1980: 288!/89) has presentedan array of statistical comparisons whichdemonstrate that ‘distribution effects arevirtually nil’. This means that the chosendistribution actually makes no noticeablecontribution to the factors which emergefrom a particular study (and this is the mainreason why a complete rank ordering of theitems is also unnecessary). Contradictory asit may seem, therefore, a forced distributionis actually no more restrictive than a ‘free’distribution. If Q methodologists generallyprefer the forced distribution, therefore, itis because it delimits unnecessary workand because it is convenient for theirparticipants.

–6 –5 –4 –3 –2 –1 0 +1 +2 +3 +4 +5 +6

[2] [2][3] [3]

[4] [4][5] [5]

[6] [6][6] [6]

[8]

–6 –5 –4 –3 –2 –1 0 +1 +2 +3 +4 +5 +6

M ost disagree Most agree

Figure 1 Example Fixed Quasi-Normal Distribution. Ranking values range from#/6, through ‘zero’, to"/6. Numbers in [brackets] indicate the number of items that can be assigned to any particular rank. Atotal of 60 items can be accommodated in the distribution illustrated.

Q methodology 77

We appreciate, however, that such argu-ments will not persuade those who considerany type of ranking procedure or distribu-tion to be overly restrictive. Concerns arealso frequently expressed about further re-strictions imposed by the Q methodologicalprocedure, particularly the provision ofa pre-designed Q set. There is, after all,always a finite number of items in the Q setand it is the researcher who has providedthose items (albeit on the basis of carefulsampling processes). These and associatedarguments often seem to culminate in thebelief that ‘you can only ever get back whatyou put in’ in a Q methodological context.

Once again, however, we would arguethat such objections are misplaced, forthey necessarily overlook the basic aimsand premises of the method. First, they (atleast partially) rely on the premise that astatement like ‘love is an intense commit-ment ’ must have a single, predefined mean-ing and hence that it could only ever beinterpreted in a single way. Yet we havealready shown that this (essentially R meth-odological) assumption is not relevant toQ methodology. The singular ‘putting in’ ofthis statement does not prevent our ‘gettingback’ multiple and qualitatively diverseresponses. The logic of this objection isfaulty. The same interview question neednot automatically elicit the same responseand the same is true of items in a Q set.

Secondly, these objections fail to noticethat it is overall item configurations whichrepresent the research target of Q metho-dology. With this in mind, Brown (1980)presents a series of analyses which demon-strate (because of the factorial nature of thesorting procedure) that a study which em-ployed a Q set of only 33 statements and aquite limited (#/4 to "/4) ranking distribu-tion would make available to its partici-pants ‘roughly 11 000 times as many

[sorting] options as there are people in theworld’. In other words, Q methodologypursues overall item configurations and itsprocedure renders a hyperastronomicalnumber of such configurations available toits participants. It is difficult to see how thismight sensibly be construed as restrictive.As Brown (1980: 267) rightly concludes,Q methodology leaves more than ‘sufficientroom for individuality [to be expressed]’.

In expressing their individuality via theQ sorting procedure, a participant willultimately be required to allocate all the Qset items an appropriate ranking position inthe distribution provided. A clear and‘gestalt’ configuration of items will dulyemerge. If the participant is happy withthis configuration, the various item num-bers (and hence the ‘form’ of the overallconfiguration) should be recorded (as theyare in Figure 2).

The next task for the researcher involvesthe gathering of supporting informationfrom the participant. Open-ended com-ments are duly requested. This can bedone via a brief post-sorting interview(which can then be transcribed and sub-jected to analysis), or simply via some formof ‘response booklet’ or post-sorting ques-tionnaire. Such post hoc analyses ordina-rily investigate: (a) how the participant hasinterpreted the items given especially highor low rankings in their Q sort, and whatimplications those items have in the contextof their overall viewpoint; (b) if there areany additional items they might have in-cluded in their own Q set (what they are,why they are important, and so on); and (c)if there are any further items about whichthe participant would like to pass comment,which they have not understood, or whichthey simply found confusing. Such open-ended comments are a vital part of the Qmethodological procedure, for they will aid


the later interpretation of the sorting con-figurations (and viewpoints) captured byeach of the emergent factors.

The participant group

Large numbers of participants are not re-quired for a Q methodological study.Q methodology aims to reveal (and toexplicate) some of the main viewpointsthat are favoured by a particular group ofparticipants. It probably does this mosteffectively when the participant groupcontains between 40 and 60 individuals(Stainton Rogers, 1995). This is only a ‘rule-of-thumb’, however, for highly effectiveQ studies can be carried out with far fewerparticipants. A salient viewpoint might, ofcourse, be revealed by reference to a singleparticipant. If, on the other hand, we wantto demonstrate that this viewpoint is sharedby several persons in ‘the group’, and henceto make sense of our subject matter (ratherthan a specific individual) on the basis ofsuch a consistency, we must evidently movebeyond the single case.

Yet the employment of very large num-bers of participants in a Q methodologicalcontext can itself be problematic. Indeed,such an approach can easily negate many ofthe subtle nuances, complexities, and hencemany of the essential qualities contained inthe data. This is obviously counterproduc-tive in the context of a qualitative techni-que. Those seeking to publish inmainstream psychological journals willalso find that statistical arguments stand inthe way of studies which try to use ‘toomany’ participants.4 In keeping to smallernumbers, therefore, an emphasis on qualityis maintained, pattern and consistency canstill be detected within the data, and theprospects of publication are also increased.

As is so often true, however, size is not allimportant. The exact constitution of theparticipant group must also be considered.In some contexts it may be sensible tostrategically sample participants, especiallyif they seem likely to express a particularlyinteresting or pivotal viewpoint. One can,for example, carry out Q methodologicalstudies which investigate particular events!/ a specific football match perhaps !/ in

[2][3]

[4][5] 27

[6] 07

–6 –5 –4 –3 –2 –1 0 +1 +2 +3 +4 +5 +641 01 39 53 23 02 34 10 35 57 05 50 2944 31 18 43 12 30 56 03 59 09 28 40 47

16 14 60 15 45 46 11 58 22 25 5251 21 32 36 55 13 42 04 33 [2]

17 24 48 49 51 26 37 [3]06 38 08 20 19 [4]

[5][6]

[6] [6][8]

–6 –5 –4 –3 –2 –1 0 +1 +2 +3 +4 +5 +6M ost disagree Most agree

Figure 2 Example Completed ‘Q Sort’ Configuration. Items 29 and 47 have been allocated rankings of#/6 which suggests that they represent the ‘most agreeable’ items in the view of this participant.Moving from right to left in the configuration, the items are becoming less and less agreeable until wearrive at items 41 and 44 which (having been ranked at "/6) are evidently considered to be the ‘mostdisagreeable’ items in the Q set.

Q methodology 79

which case one ought to ensure that Q sortsare gathered from as many of the obviouslypertinent demographic groups as possible.Both sets of fans, for example, both sets ofplayers, the ball boys perhaps, neutralspectators, the referee and his assistants,members of the press, and so on.

Where Q studies aim to investigate parti-cular concepts , however, participants maynot divide so obviously along lines pre-scribed by demographic characteristics. Insuch cases, it is better to avoid too manyassumptions a priori , particularly wherethese assumptions are based on precon-ceived demographic notions. The wholepoint of Q methodology is to allow indivi-duals to categorize themselves on the basisof the item configurations they produce(and hence via the viewpoints they ex-press). Its function is exploratory. In suchcircumstances, it is probably sensible to relyon more opportunistic sampling techni-ques, at least in the first instance, or untila series of Q methodological explorations(and their emergent factors) provide empiri-cal justification for the belief that certainviewpoints ‘belong’ exclusively to specificdemographic groups.

Statistical analysis (factor extraction,rotation and estimation)

Q methodology employs a by-person corre-lation and factor analytic procedure. Hence,it is the overall configurations produced bythe participants that are intercorrelated andfactor analysed. The initial correlation ma-trix duly reflects the relationship of each (Qsort) configuration with every other (Q sort)configuration (not the relationship of eachitem with every other item). To subject thismatrix to factor analysis is to produce a setof factors onto which the participants load

on the basis of the item configurations theyhave created. Hence, two participants thatload onto the same factor will have createdvery similar item configurations. Eachfactor duly captures a different item config-uration which is nonetheless shared by (andwhich is characteristic of) the participantswho load onto that factor.

There are now several dedicated Q meth-odological packages available which allowthe appropriate analyses to be conducted.PCQ for Windows is probably the bestcommercial product available (Stricklinand Almeida, 2001). PQ Method (Schmolck,2002), on the other hand, does the jobeffectively and is available as a free down-load from the internet.5 Such dedicatedpackages are to be recommended. Theyfacilitate data input, automatically generatethe initial by-person correlation matrix, andmake processes of factor extraction, rotationand estimation very straightforward.

Despite this automation, however, theresearcher still has to make some importantdecisions as the analysis proceeds. Differenttypes of factor analysis (and hence factorextraction) exist, as do different methods offactor rotation, and there often seems littlereason for preferring one system overanother. To further complicate matters, tra-ditional factor techniques do not automati-cally resolve themselves into a single,correct (by which we imply ‘mathematicallysuperior’) solution, as would the morerecently developed principal components(PCA) or cluster analytic techniques. This isparticularly pertinent in a Q methodologicalcontext, as the very oldest of the factortechniques (known as the centroid orsimple summation method) is generallypreferred.

For this reason, dedicated Q methodolo-gical packages tend to offer the centroidmethod as their default choice (and, indeed,


PCQ for Windows offers no alternative).This method offers a potentially infinitenumber of rotated solutions. Indeed, it isexactly this openness and indeterminacywhich appeals to the Q methodologist, asit leaves them free to consider any data setfrom a variety of perspectives, before select-ing the rotated solution which they considerto be the most appropriate and theoreticallyinformative. Having said that, both authorshave also analysed Q data using PCAand have found the results to be equallysatisfying.

The accent on theoretical discretion dis-cussed above has nonetheless led manyQ methodologists to be critical of modernfactor rotation techniques (such as varimax)which are perceived to reveal only the mostmathematically (not necessarily the mosttheoretically) informative solution. Suchpractitioners retain Stephenson’s preferencefor theoretical (judgemental or ‘by-hand’)rotation techniques (see, for example,Brown and Robyn, 2003). Why, argue theenthusiasts for hand-rotation, let a compu-ter decide which point of view to adopt onone’s data when an infinite number arepossible? Whilst this argument !/ whichhas been known to cause vertigo in thosewho hope to find certainty in statistics !/ isundoubtedly valid, in practice many Qmethodologists (like many more conven-tional factor analysts) prefer the simplicityand reliability of the varimax procedure.

The varimax procedure is also consonantwith one of our typical aims in using Qmethodology: namely, to reveal the range ofviewpoints that are favoured by our partici-pant group. Given this aim, it makes theo-retical sense for us to pursue a rotatedsolution which maximizes the amount ofvariance explained by the extracted factorsand as the varimax procedure automaticallyseeks this mathematically superior solution,

it also makes sense for us to prefer thistechnique. As an additional advantage froma qualitative perspective, the varimax pro-cedure is also highly reliant ‘on the topo-graphical features of the correlation matrix’(Brown, 1980: 238), which affords a wel-come priority to the input of the participantgroup on the emergent factor structure. Thisprioritization of participant input is ofcourse a shared feature of many qualitativeresearch techniques in psychology (Willig,2001). In the end, however, the technique ofrotation employed is always going to bedependent ‘on the nature of the data [gath-ered] and upon the aims of the investigator ’(Brown, 1980: 238) and there is ultimatelyno substitute for careful consideration inthe context of a particular study.

The next step is then to decide whichfactors should be selected for interpretation.A standard requirement is to select onlythose factors with an eigenvalue in excess of1.00.6 This is undoubtedly a somewhatarbitrary criterion in the context of Q meth-odology. It is well known, for example, thatseveral factors with eigenvalues in excess of1.00 might be extracted even from randomdata, as random patterns will always ariseand be detected. On the other hand, it is agenerally accepted means of safeguardingfactor reliabilities and factors which gobelow this minimum will ultimately serveno data-reductive purpose as they explainless of the overall study variance thanwould any single Q sort.7 A second stan-dard requirement is that an interpretableQ methodological factor must ordinarilyhave at least two Q sorts that load signifi-cantly upon it alone.8 Such significantlyloading Q sorts are called ‘factor exemplars’as they exemplify the shared item pattern orconfiguration that is characteristic of thatfactor. There might, in certain circum-stances, be a theoretical justification for

Q methodology 81

accepting and interpreting factors with onlyone exemplar, but the interest in sharedorientations nonetheless leads most Q re-searchers to adopt the more stringent criter-ion outlined above.

When, as is obviously typical, a factor isloaded by more than one exemplar, a factorestimate is generated through a procedure ofweighted averaging (this occurs automati-cally in programmes such as PCQ and PQMethod). In effect, the Q sorts of all parti-cipants that load significantly on a givenfactor are merged together to yield a single(factor exemplifying) Q sort which serves asan interpretable ‘best-estimate’ of the pat-tern or item configuration which charac-terizes that factor. Confounded Q sorts(which load significantly on two or morefactors) are excluded from this weightedaveraging procedure.9 The endpoint of thestatistical analysis is duly reached wheneach of the selected factors is represented byits own best-estimate Q sort or ‘factor array’(naturally, factors with a single exemplarwill be forced to use that individual’s ownQ sort as a best-estimate). These factorarrays or best-estimate Q sorts can then besubjected to interpretation.

Factor interpretation

The study files provided by dedicated Qmethodological packages can be confusingfor an uninitiated reader. They ordinarilyprovide a table which illustrates the load-ings of all participant Q sorts on all ex-tracted and rotated factors. Significantloadings will be marked with a star (*).10

The eigenvalues and the percentage of thestudy variance explained by each factor willalso be provided, as will a table (normallyentitled ‘Factor Correlations’) which indi-cates the intercorrelations of the various

factor arrays (giving a basic indication ofthe relationships between the factors). Awealth of other information may also beincluded. The most important aspect of thestudy file will nonetheless be the factorarrays themselves. These will be found ina table, ‘Item or Factor Scores’. Table 1 ispresented in a typical format.

The data summarized in Table 1 are takenfrom a Q study about young offenders andtheir punishment. It was carried out by anundergraduate student at University CollegeNorthampton and is reproduced here withblemishes intact. It nonetheless provides aclear example of Q methodology at workand is illustrative of the method’s efficacyand effectiveness even in the hands of aninexperienced user. We shall demonstratethis effectiveness by producing a brief inter-pretation of the study’s first factor (A).

The interpretative task in Q methodologyinvolves the production of a series ofsummarizing accounts, each of which ex-plicates the viewpoint being expressed by aparticular factor. These accounts are con-structed by careful reference to the position-ing and overall configuration of the items inthe relevant ‘best-estimate’ factor arrays.The obvious place to begin such delibera-tions is at the two ‘poles’ of the Factor Aconfiguration. Here one finds the itemsFactor A (or, to be exact, those participantswhose Q sorts exemplify factor A) feelsparticularly strongly about. There is a strongbelief, for example, that ‘young offendershave to be punished’ (as item 10 is rankedhighly at the#/4 position) and that they willnot ‘learn to behave properly’ unless theyare (item 22 ranked at#/3). Yet such punish-ment must ultimately be tailored to reflectboth the seriousness of the crime (item 18 isranked at #/4) and the age of the offender(item 06: #/3). In keeping with this stand onage, Factor A also rejects the idea that young


offenders should be tried as adults orimprisoned with adults (items 04 and 12are ranked at "/3 and "/4 respectively).Young offenders should not have to carryout manual work to repay their victims, norshould the death penalty be consideredfor underage murderers (item 17: "/3, item03: "/4).

One can observe that a relatively clearand unitary viewpoint is already beginningto emerge from these item rankings. It is amistake to assume, however, that all ‘theaction’ is taking place at the poles of thedistribution. Consider item 07, for example,which suggests that the parents of young

offenders should be punished for theirchild’s crimes. This item is ranked at ‘zero’by Factor A. Such a display of indifferenceand seeming neutrality surely indicates thatthis item is of little importance to theviewpoint being expressed? Yet when con-sideration is given to the ranking of thisitem by the other factors in the study ("/4,"/4, "/3 and "/3 respectively), the zeroawarded by Factor A suddenly becomes‘information rich’. Factor A does not thinkparents should be punished where minorcrimes have been committed (we know thisbecause item 24 is ranked at "/2), but theranking of item 07 suggests that ‘parental’

Table 1 Example item scores table for a study about the punishment of young offenders

Statements factors A B C D E

01 Current punishments are not severe enough to motivate change #/2 #/3 0 "/2 #/202 Young offenders attend counselling for rewards not for rehabilitation "/1 #/1 "/2 "/1 "/103 The death penalty is fair for young offenders who commit murder "/4 0 "/2 "/4 "/404 It is legitimate for under 18’s to be tried as adults "/3 "/2 "/2 0 005 Young offenders should be punished in other ways not prison #/2 0 #/2 #/2 006 Age is an important factor in the choice of a punishment type #/3 "/1 0 "/1 007 It is legitimate to punish parents for young offenders crimes 0 "/4 "/4 "/3 "/308 10 years old is the right age for criminal responsibility to begin 0 "/1 "/1 #/3 "/209 The age of criminal responsibility should be lower than 10 years "/2 "/3 "/3 "/2 010 Young offenders have to be punished #/4 #/4 #/2 #/4 #/411 Young offenders should be treated less severely than adults 0 "/2 0 0 "/212 Young offenders should be imprisoned with adults "/4 "/1 "/4 "/2 "/413 Young offenders should receive attention and support #/2 "/3 #/3 "/2 014 Community service is a good way of punishing young offenders #/1 0 #/1 "/1 #/115 Young offenders who commit murder should get life imprisonment "/1 #/4 #/4 #/1 "/216 Electronic tagging is an effective way of punishing young offenders #/1 #/3 #/1 #/1 #/117 Young offenders should repay their victim with manual work "/3 0 #/1 #/1 018 Punishments should reflect the seriousness of the crime #/4 #/1 #/2 #/4 #/219 Convicted young offenders should be made to pay fines "/1 "/4 "/1 "/3 "/320 Cautioning has no impact on the behaviour of young offenders #/1 #/1 0 0 #/321 Persistent young offenders should be taught a trade 0 "/2 #/3 #/2 #/122 If young offenders are not punished they won’t learn to behave #/3 #/2 #/4 #/3 #/323 Punishments are not severe enough to deter young offenders 0 2 0 0 #/424 Parents should be punished for minor crimes that their child commits "/2 0 "/3 "/4 "/125 Young offenders should not have their IDs protected when they commit serious

crimes"/2 #/2 "/1 #/2 #/2

This table indicates the rankings assigned to each item within each of the factor exemplifying Q sort(or item) configurations. Reading the table by column reveals the configuration (or comparativeranking) of items which characterizes a particular factor. In column A, for example, we can see thatFactor A ranked item 01 at #/2, item 02 at "/1, and so on. Reading the table by row reveals thecomparative ranking of a particular item across all the factors. In row 02, for example, we can see thatitem 01 was ranked at #/2 by Factor A, at #/3 by Factor B, and so on.

Q methodology 83

punishment has not been completely dis-missed by this factor.

This example demonstrates that much ofsignificance can occur in the supposedly‘neutral’ area of the configuration. Indeed,any interpretation which disregards theitem rankings in this area will almostcertainly fail to capture the subtleties ofthe viewpoint being expressed. A concen-tration on too few of the items in the arraywill also prevent the holistic or ‘gestalt’nature of the viewpoint being communi-cated. This represents a serious interpreta-tive failure given the stated research targetof Q methodology. One can safeguardagainst these problems, however, by check-ing the qualitative comments gathered fromparticipants who have loaded significantlyon the factor being interpreted. These com-ments can help to verify initial interpreta-tions of specific item rankings. Participant17, for example, tells us that without sui-table punishments young offenders ‘will notlearn what is and isn’t acceptable beha-viour ’. This reinforces the ranking of item22 at #/3. Perhaps more importantly, how-ever, participant 26 confirms the salience ofthis factor’s relatively positive ranking ofitem 07 when he suggests that ‘parents areaccountable . . . (for they have been) . . . in thebest position to exert influence ’ over theirchild.

If this process is carried to its conclusion,such that various item rankings and parti-cipant comments are effectively combined,a clear interpretation of the factor willemerge. An example interpretation of FactorA is shown. Notice that this interpretativeaccount has been given a title and that somepertinent demographic information hasbeen provided. The rankings which in-formed the interpretation at each stage arealso included in the text (as a pointof reference for the reader). The citing of

Factor A: A crime-related and age sensitiveapproach to the punishment of young offenders

Demographic information: Factor A has sixsignificantly loading participants and it explains17% of the study variance. It has an eigenvalue of5.40. Four of the loading participants are men,two are women. They have an average age of 23.84years.

Factor interpretation: Young offenders mustdefinitely be punished for their crimes (10: #/4).Cautioning has little impact and punishmentshanded out are generally not severe enough(20: #/1; 01: #/2). As participant 17 confirms,without suitable punishments young offenders‘will not learn what is and [what] isn’t acceptablebehaviour ’ (22: #/3). It is nonetheless importantthat the chosen punishment reflects both the ageof the offender and the type of crime they havecommitted (06: #/3; 18: #/4). Under 18’s shouldcertainly not be tried as adults, nor should they beimprisoned with adults (04: "/3; 12: "/4). Theyshould have their identities protected in thecontext of serious crimes and the death penaltyshould not be considered !/ even in cases ofmurder (25: "/2; 03: "/4). After all, youngoffenders may well have committed their crimewithout having a full ‘understanding of the law ’(participant 26 comment). Life imprisonmentis questionable for the same reason (15: "/1).Given this situation, it would be quite wrong tofurther lower the age of criminal responsibility(09: "/2). Prison sentences are probably not theright way to punish young offenders (05: #/2), norshould they be made to do manual work as ameans of repaying their victims (17: "/3). We needto take a more constructive approach topunishment, offering attention and support whereappropriate (13: #/2). Electronic tagging devicesand community service open up a range ofpossibilities in this regard (16: #/1; 14: #/1).Neither should we completely dismiss the idea ofpunishing the parents of young offenders if aserious crime has been committed (07: 0), althoughthis would probably be inappropriate in thecontext of more minor crimes (24: "/2). Indeed,there is a real sense in which the ‘parents areaccountable. . . [for they have been]. . .in the bestposition to exert [an] influence ’ over their errantchild (participant 26 comment).

Factor A: example factor interpretationThis interpretation uses the relative itemrankings and participant comments to arriveat a clear, ‘gestalt’ account of theviewpoint being expressed by Factor A.


‘14: #/1’, for example, indicates that item 14was ranked in the #/1 position by Factor Aand that this item is relevant to the currentsection of the interpretation:

In principle, of course, the process ofinterpretation is potentially never-ending,there always being different shades ofmeaning and emphases that could be drawnfrom the data, and hence different readingsapplied to it. Yet Q methodological inter-pretation is necessarily constrained by (andcan always be checked against) the subjec-tive input of the participant group, as, incontrast to many other qualitative methods,that input is actually reflected and frozen inthe relevant ‘factor exemplifying’ item con-figurations. In Q methodology, subjectiveinput produces objective structures. Nodoubt the procedure which makes thispossible also necessitates a certain loss ofecological validity in relation to other qua-litative methods. In partial compensation,however, the presence of these objectivestructures allows Q methodology to de-mand a great deal of the interpretingresearcher, for any failure to respondfully and faithfully to the subjective inputof the participant group can be easilydetected.

The accuracy and efficacy of the pro-duced interpretations can also be verifiedvia a second source: simply by askingrelevant (significantly loading) participantsto comment upon them. This can be a usefulsource of further insight, although oneshould bear in mind that all such interpre-tations are designed to communicate a‘shared’ viewpoint, and hence that theyneed not provide a veridical representationof a participant’s own opinion. Whilst,therefore, alternative readings of our exam-ple factor are a definite possibility, thesevarious checks and balances !/ thoroughattention to the item rankings, to participant

comment and to the verifying remarks ofloading participants !/ make us very con-fident that the main thrust of all competentreadings would be very similar to our own.Put another way, in common with mosthermeneutic endeavours, Q data make itrelatively simple to reject incompetent read-ings, whilst allowing scope for numeroussubtly different competent readings to co-exist.

We are thus confident that this interpre-tative process has produced a coherentexpression of the Factor A viewpoint !/

captured in the item rankings of our parti-cipants !/ about the ways in which youngoffenders should be punished. This view-point is ‘shared’ by members of the partici-pant group. In contrast to many otherqualitative methods, it has also been cap-tured in a genuinely gestalt fashion. Addthis viewpoint to the others revealed in thesame study (see the item rankings forFactors B, C, D and E in Table 1) and Qmethodology has allowed us to gain acomprehensive snapshot of the major view-points being expressed by our participantgroup.

Conclusion

We close with a caveat. Q methodologymakes no claim to have identified view-points that are consistent within individualsacross time. Understanding this is crucial tounderstanding the role and place of thismethodology. Factor A need not, and wethink should not, be thought of as (norreduced to) the expression of stable intra-psychic features of the factor exemplars(such as ‘attitudes’ or ‘personality traits’).To think otherwise is to impose a priori theessentialist and somewhat counterintuitiveassumption that a given participant is

Q methodology 85

capable of expressing only one coherentviewpoint on an issue. All that we can sayis that these participants did express theseviewpoints via these item configurations.Whilst this leaves individual factor exem-plars free to ‘change their minds’, we mightnonetheless expect the emergent manifoldof shared viewpoints to show a degree ofconsistency over time. In short, what isdecisive from a Q methodological perspec-tive is less ‘who said what about X?’ than‘what is currently being said about X?’

It is in this respect that Q methodologyprovides results that are consistent withother broadly constructionist endeavours,such as discourse analysis, social represen-tations theory, autopoietic systems theoryand ethogenics. In enabling a cartography ofsocial semantics, Q methodology is in arelation of complementarity with more the-matic or ‘individual’ approaches in qualita-tive psychology, and also with methods,like conversation analysis, that focus onthe micro processes of situated interaction.Using Q methodology, to conclude, doesnot entail a denial of microlevel commu-nicative pragmatics, nor of macrolevel so-cial structures. Rather it entails a focus onsubjectively expressed, socially organizedsemantic patterns, whether these are subse-quently grasped as ‘discourses’, ‘representa-tions’ or as ‘fields of the sayable and theseeable’ (Curt, 1994).

Acknowledgements

The authors would like to thank MichelleBotley for her hard work and for allowingthem to take advantage of the illustrativedata used in this paper. We are also gratefulto Rom Harre and the journal editors fortheir helpful comments.

Notes

1. Yet, in contrast to the methodologicalchanges mentioned, this theoretical departure isnot guaranteed by the act of inversion. One can,for example, choose to conduct a Q techniquefactor analysis on a collection of R methodologi-cal ‘test’ results. In other words, the analysis issimply carried out on ‘the transpose of the Rmatrix !/ i.e., as the correlation and factorizationby rows of the same matrix of data that in R isfactored by columns’ (Brown, 1980: 12!/13,emphasis added). This method evidently con-tinues to prioritize psychological traits (hypothe-sized cognitive ‘parts’ of persons) and theirmeasurement. No theoretical change occurshere, merely a statistical transposition of a setof data. This approach to Q technique factoranalysis was employed and promoted by CyrilBurt. Indeed, it is Burt’s ‘transposed matrixmodel’ which has invariably found its way intomainstream textbooks under the heading ‘Qtechnique’ factor analysis (see, for example,Maxwell, 1977). The main problem with thismodel (if one ignores possible theoretical objec-tions) is that a single matrix of data can properlybe transposed (and thus factor analysed by bothrow and column) only where a single measuringunit is employed throughout the matrix, ‘i.e.,when the same measuring unit (e.g. the inch) iscommon to both rows and columns’ (Brown,1980: 13). This condition is rarely satisfied,however, simply because the measurement ofdiffering traits (be they psychological or physi-cal) is hardly ever undertaken using the samemeasuring unit. As a result, the intercorrelation(and hence the direct comparison) of such traitsis highly problematic. Stephenson recognizedthese problems very quickly (for a summary, seeBurt and Stephenson, 1939). Indeed, he criti-cised Burt’s transposition procedure quite vocif-erously, for both methodological and theoreticalreasons (see, for example, Stephenson, 1936a). Ifthe Q technique was to invert the theoreticaltradition, he realized, a distinct change ofmethod and a distinct change in the nature ofthe gathered data would be required.


2. Piloting can be done in a number ofdifferent ways. As we implied in the main text,however, asking pilot participants to physicallysort a very large number of items is probablyunwise. It may be more sensible to ask a group ofpilot participants (or even one or two partici-pants with subject-specific expertise) to providegeneral comments on the construction of the Qset, remembering that the primary functions ofthe piloting process are to ensure: (a) thatsemantic duplication within the statements isavoided; (b) that statements are clearly expressed(usually in everyday rather than technical termi-nology); (c) that statements express a singleproposition only (as multiple propositions canmake sorting and interpretation highly proble-matic); (d) that the Q set is properly ‘balanced’(in the context of politically charged or particu-larly contentious research questions, it may bevery important to ensure that relatively equalnumbers of ‘pro’ and ‘anti’ responses are avail-able to your participants); and finally, andperhaps most importantly, (e) that the Q setprovides adequate coverage of relevant issuesand that it is, therefore, broadly representative ofthe appropriate opinion domain.3. The only exception to this rule is that at

least one item must be assigned to every rankingposition. If this is not done, the linear assump-tions of correlation would effectively be broken,rendering the intercorrelation of the resultant Qsorts highly problematic (Kline, 1994).4. The rules of conventional (R methodologi-

cal) factor analysis demand that your study bedesigned such that it has at least twice as manyparticipants as it does variables (Kline, 1994).Once the matrix is inverted, however, as it is in aQ methodological context, this same assumptionsuggests that you should have at least twice asmany items in the Q set as you do participants. Intruth, however, the two situations are hardlyequivalent and objections on these grounds areunusual. This may be because the initial assump-tion is itself somewhat arbitrary (even in an Rmethodological context). Nonetheless, if main-stream journal publications are a goal, it isprobably sensible to maintain at least a 1:1 ratioof items to participants (and certainly to avoid

studies which include many more participantsthan items).

5. The relevant internet address is: www.rz.unibw-muenchen.de/!/p41bsmk/qmethod

6. A factor’s eigenvalue (the word just means‘characteristic value’) is simply the sum ofsquared factor loadings for that factor. Dividethe eigenvalue by n (where n equals the numberof participants in your study) and multiply theresult by 100, and you will have calculated thevariance accounted for by that factor. This givesus the following equation (Eq. 1): V$/100(EV/n ).It follows, therefore, that one can also calculate afactor’s eigenvalue by multiplying the variance itaccounts for by n and then dividing the result by100. The following equation (Eq. 2) results: EV$/

V%/n/100. More detailed (and sometimes contra-dictory) observations about eigenvalues, theircalculation, and their use in a Q methodologicalcontext can be found in Brown (1980). Thispublication is available at the following webaddress by kind permission of Professor S.R.Brown: http://reserves.library.kent.edu/course-page. asp?cid$/203&page$/01.

7. Let us assume, for example, that ourhypothetical factor has an eigenvalue of 0.86and that our study includes 50 participants.Insert these figures into Eq. 1 of Note 6 and wesee that V$/100(0.86/50). In other words, ourfactor accounts for 1.72% of the study variance.A single Q sort in the same study will, however,account for 2% of the study variance (as 100%of the study variance divided amongst 50 parti-cipant Q sorts$/2%). In other words, the singleQ sort has more explanatory power than thefactor.

8. Arguably, decisions about the ‘signifi-cance’ of a loading should properly be made ontheoretical rather than statistical grounds. On theother hand, one will undoubtedly have to justifysuch decisions and statistical arguments arelikely to hold considerably more weight in apsychological context. A significant factor load-ing (at PB/0.05) can be calculated using thefollowing equation: 1.96(1/!n ) where n equalsthe number of items in your Q set. If, therefore,our Q set contains 60 items, a statisticallysignificant loading will be equal to or greater

Q methodology 87

than 1.96(1/!60)$/9/0.26. Of course, this willstill mean that 1/5 of your significant factorloadings are likely to have occurred by chance.It may be better, therefore, to employ the morestringent (PB/0.01) significance level. This canbe calculated using the equation: 2.58(1/!n ).Using the same example, this would mean that astatistically significant loading would have to beequal to or greater than 2.58(1/!60)$/9/0.34.Again, Brown (1980) provides a more detailedcommentary.9. As we shall see in the section on ‘Factor

Interpretation’, the confounding of a participantQ sort also tends to mean that their qualitativecomments are lost from the final analysis (as onegenerally only uses the comments of significantlyloading participants to support a particular factorinterpretation). This is a shame. We tend, there-fore, to employ a strategy which minimisesconfounding and which duly maximises thenumber of significantly loading participants ina particular study. Let us suppose you havefollowed the instructions laid out in Note 8above and have calculated that a significantloading in your study (at PB/0.01) must reach9/0.34 or above. You then find that (in using thissignificance level) 35 of 50 participant Q sortsload on a single factor, 4 have no significantloading and the remaining 11 are confounded.This means that 15 participant Q sorts (30% ofthe data) are not being used to construct thevarious factor exemplifying item configurations.In such circumstances, it may be sensible toconsider raising the level at which a loading issaid to be significant (which will only make yourcriterion more stringent from a statistical view-point). In the example, we might look again at thesolution with a significance level of 9/0.35 orabove, then 9/0.36 or above, and so on. At 9/0.40we find that 42 of the 50 participants loadsignificantly on a single factor, five now haveno significant loading and only three are con-founded. In other words, only eight participant Qsorts (16% of the data) are not now being used.We also find that raising the level still further (to9/0.41 or above) only serves to increase thenumber of Q sorts with no significant loadingand hence to reduce the overall numbers of

participants with significant (and usable) load-ings. At 9/0.40, therefore, we have maximizedthe number of participants with significant load-ings and have duly achieved our most satisfac-tory solution.

10. It is perfectly possible for a single factorto have both positive and negative significantloadings. This is known as a ‘bipolar’ factor. Thisbipolarity implies that two diametrically op-posed viewpoints are being expressed by theparticipants who load on such a factor, eachviewpoint having a factor exemplifying itemconfiguration that it is the ‘mirror-image’ of theother. In other words, the positive loaders willagree with the item rankings and overall itemconfiguration for that factor (as they are con-tained in the Item or Factor Score table in yourprintout !/ see Table 1 in the main text), whereasthe negative loaders are agreeing with an entirelyreversed configuration (and hence they are ad-vocating an opposed viewpoint). If the positiveloaders have ranked an item at #/4, therefore, thenegative loaders have effectively ranked thesame item at "/4. If the positive loaders haveranked an item at #/2, the negative loaders haveranked it at "/2, and so on. Such factorsnecessarily require you to produce two distinctfactor interpretations, the first interpreting thefactor exemplifying configuration as it appears inyour Item Score Table (which will reveal thepositive viewpoint), the second interpreting thesame configuration with all the item rankingsreversed (which will reveal the negative ordiametrically opposed viewpoint).

References

Billig, M. 1997: The dialogic unconscious: psy-choanalysis, discursive psychology and thenature of repression. British Journal of SocialPsychology 36, 139!/59.

Brown, S. and Stenner, P. 2001: Being affected:Spinoza and the psychology of emotion.International Journal of Group Tensions 30,81!/105.

Brown, S.R. 1980: Political subjectivity: applica-tions of Q methodology in political science .New Haven, CT: Yale University Press.


Brown, S.R. and Robyn, R. 2003: Reserving a keyplace for reality: philosophical foundationsof theoretical rotation. Paper presented atthe 19th annual conference of the Interna-tional Society for the Scientific Studyof Subjectivity, Kent State University, OH,2!/4 Oct.

Burt, C. 1937: Correlations between persons.British Journal of Psychology 28, 59!/96.

Burt, C. and Stephenson, W. 1939: Alternativeviews of correlations between persons. Psy-chometrika 4, 269!/81.

Capdevila, R. and Stainton Rogers, R. 2000: Ifyou go down to the woods today: narrativesof Newbury. In Addams, H. and Proops, J.,editors, Social discourse and environmentalpolicy. Cheltenham: Edward Elgar.

Crossley, M.L. 2000: Introducing narrative psy-chology: self, trauma, and the constructionof meaning . Buckingham: Open UniversityPress.

Curt, B. 1994: Textuality and tectonics: troublingsocial and psychological science . Bucking-ham: Open University Press.

Fehr, B. 1988: Prototype analysis of the conceptsof love and commitment. Journal of Person-ality and Social Psychology 55, 557!/79.

Good, J.M.M. 2000: William Stephenson, CyrilBurt and the demise of the London school.Paper presented at the History and Philoso-phy of Psychology section of the BritishPsychological Society, Annual conference,University of Ripon and York St John, UKApril.

Harre, R. 1999: The singular self: an introductionto the psychology of personhood . London:Sage.

______ 2004: Staking our claim for qualitativepsychology as science. Qualitative Researchin Psychology 1, 3!/14.

Harre, R. and Gillett, G. 1994: The discursivemind. London: Sage.

Harvey, O.J. 1997: Beliefs, knowledge, and mean-ing from the perspective of the receiver. InMcGarty, C. and Haslam, S.A., editors, The

message of social psychology. Oxford: Black-well.

Haslam, S.A. and McGarty, C. 2003: Researchmethods and statistics in psychology. Lon-don: Sage.

Kitzinger, C. 1986: Introducing and developing Qas a feminist methodology: a study of ac-counts of lesbianism. In Wilkinson, S., edi-tor, Feminist social psychology: developingtheory and practice . Milton Keynes: OpenUniversity Press.

______ 1987: The social construction of lesbian-ism . London: Sage.

Kline, P. 1994: An easy guide to factor analysis.London: Routledge.

Maraun, M.D. 1998: Measurement as a normativepractice: implications of Wittgenstein’s phi-losophy for measurement in psychology.Theory and Psychology 8, 435!/61.

Maxwell, A.E. 1977: Multivariate analysis inbehavioural research . London: Chapman &Hall.

Moscovici, S. 1981: On social representations. InForgas, J.P., editor, Social cognition: perspec-tives on everyday life. London: AcademicPress.

Parker, I. 1997a: Psychoanalytic culture: psycho-analytic discourse in Western society. Lon-don: Sage.

______ 1997b: Discourse analysis and psycho-analysis. British Journal of Social Psychol-ogy 36, 479!/95.

Polkinghorne, D.E. 1995: Narrative configurationin qualitative analysis. International Journalof Qualitative Studies in Education 8, 8!/25.

Potter, J. 1997: Discourse analysis as a way ofanalysing naturally occurring talk. In Silver-man, D., editor, Qualitative analysis: issuesof theory and method . London: Sage.

Potter, J. and Wetherell, M. 1987: Discourse andsocial psychology: beyond attitudes andbehaviour. London: Sage.

Rogers, C.R. 1954: Psychotherapy and personal-ity change . Chicago, IL: University ofChicago Press.

Q methodology 89

Schmolck, P. 2002: PQMethod (version 2.11).[Computer software]. Available at www.rz.unibw-muenchen.de/!/p41bsmk/qmethod[last accessed: 28 October 2004].

Senn, C.Y. 1996: Q methodology as feministmethodology: women’s views and experi-ences of pornography. In Wilkinson, S.,editor, Feminist social psychologies: inter-national perspectives. Buckingham: OpenUniversity Press.

______ 1995: Repertory grids: an interactive,case-study perspective. In Smith, J.A., Harre,R. and Van Langenhove, L., editors, Rethink-ing methods in psychology. London: Sage.

______ 1996: Beyond the divide between cogni-tion and discourse: using interpretativephenomenological analysis in health psy-chology. Psychology and Health 11, 261!/71.

______ editor, 2003: Qualitative psychology: apractical guide to research methods . Lon-don: Sage.

Smith, N.W. 2001: Current systems in psychol-ogy: history, theory, research, and applica-tions . London: Wadsworth.

Stainton Rogers, R. 1995: Q methodology. InSmith, J.A., Harre, R. and Van Langenhove,L., editors, Rethinking methods in psychol-ogy. London: Sage.

Stainton Rogers, R. and Stainton Rogers, W.1992: Stories of childhood: shifting agendasof child concern . Hemel Hempstead: Har-vester-Wheatsheaf.

Stainton Rogers, R., Stenner, P., Gleeson, K.and Stainton Rogers, W. 1995: Social psy-chology: a critical agenda. Cambridge: PolityPress.

Stenner, P. 1993: Wittgenstein and the textualityof emotion. Practice 9, 29!/35.

______ 1996: Shared pathways of Heidegger andFoucault. Manifold 3, 12!/17.

______ 1998: Heidegger and the subject: ques-tioning concerning psychology. Theory andPsychology 8, 59!/77.

Stenner, P., Dancey, C.P. and Watts, S. 2000: Theunderstanding of their illness amongstpeople with Irritable Bowel Syndrome: aQ methodological study. Social Scienceand Medicine 51, 439!/52.

Stenner, P. and Eccleston, C. 1994: On thetextuality of being: towards an invigoratedsocial constructionism. Theory and Psychol-ogy 4, 85!/103.

Stenner, P. and Marshall, H. 1995: A Q metho-dological study of rebelliousness. EuropeanJournal of Social Psychology 25, 621!/36.

______ 1999: On developmentality: researchingthe varied meanings of ‘independence’ and‘maturity’ extant amongst a sample of youngpeople in East London. Journal of YouthStudies 2, 297!/316.

Stenner, P. and Stainton Rogers, R. 1998: Jealousyas a manifold of divergent understandings: aQ methodological investigation. EuropeanJournal of Social Psychology 28, 71!/94.

______ 2004: Q methodology and qualiquan-tology: the example of discriminatingbetween emotions. In Tod, Z., Nerlich,B., Mckeown, S. and Clark, D., editors,Mixing methods in psychology. London:Routledge.

Stenner, P. and Watts, S. 1998: [Re]searching forlove: subjectivity and the ontology of theQ factor. Operant Subjectivity 21, 27!/48.

Stephenson, W. 1935: Correlating persons in-stead of tests. Character and Personality 4,17!/24.

______ 1936a: The foundations of psycho-metry: four factor systems. Psychometrika1, 195!/209.

______ 1936b: The inverted factor technique.British Journal of Psychology 26, 344!/61.

______ 1953: The study of behaviour: Q techni-que and its methodology. Chicago, IL: Uni-versity of Chicago Press.

______ 1983: Quantum theory and Q methodol-ogy: fictionalistic and probabilistic theories


conjoined. The Psychological Record 33,213!/30.

______ 1988/89: The quantumization of psycho-logical events. Operant Subjectivity 12,1!/23.

Sternberg, R.J. 1986: A triangular theory of love.Psychological Review 93, 119!/35.

Stricklin, M. and Almeida, J. 2001: PCQ: analysissoftware for Q-technique (Academic Edition,revised) [Computer software]. Available athttp://www.pcqsoft.com/ [last accessed: 28October 2004].

Watts, S. 2000: Critical psychology in the newmillenium. International Journal of CriticalPsychology 11, 169!/72.

______ 2002: Stories of partnership love: Qmethodological investigations. UnpublishedPh.D Thesis, University of East London.

Watts, S. and Stenner, P. 2003a: Q methodology,quantum theory and psychology. OperantSubjectivity 26, 155!/73.

______ 2003b: On normativity, meaninglessness,and the centrality of the self: a reply to thecomments of Good & Brown. Operant Sub-jectivity 26, 182!/89.

______ 2005: The subjective experience of part-nership love: a Q methodological study.British Journal of Social Psychology 44.

Willig, C. 2001: Introducing qualitative researchin psychology: adventures in theory andmethod. Buckingham: Open UniversityPress.

About the authorsDR SIMON WATTS is a Senior Lecturer in the Division of Psychology at Nottingham TrentUniversity. Simon teaches qualitative research methods and personal relationships andpublishes in the same areas. He completed a PhD in 2002 which examined the theoryand potential of Q methodology as a qualitative method, with particular reference to itsapplication in the context of partnership love. Simon has also taught in the Departmentof Human Relations at the University of East London, in the Departments of Psychologyat University College Northampton and University College London, and he currentlysupervises Masters level qualitative dissertations for the Open University.DR PAUL STENNER is a Lecturer in Psychology at University College London. Paul hasbeen publishing Q methodological research for over a decade and in relation to a varietyof subject matters, including jealousy, rebelliousness, love and I.B.S. Paul is a familiarname in qualitative psychology and is the coauthor of two related books. The first ofthese !/ ‘Textuality and Tectonics’ (written with Rex and Wendy Stainton Rogers andothers, under the collective pseudonym ‘Beryl Curt’) !/ was acclaimed by reviewers as aninnovative and groundbreaking text. The second !/ ‘Social Psychology: A CriticalAgenda’ (coauthored with Rex and Wendy Stainton Rogers and Kate Gleeson) !/ wasdescribed by Kenneth Gergen as ‘historically speaking, an enormously importantpublication’. Paul is currently researching in the area of ethics and the emotions. He haspreviously held posts at the Universities of East London and Bath.

Q methodology 91

Reproducedwith permission of the copyright owner. Further reproduction prohibitedwithout permission.

Documents

Doing Q methodology: theory, method and interpretationsites.nd.edu/lapseylab/files/2014/10/Watts_2005_Doing-Q...that constitute obstacles to the use of one of the very first ‘alternative’