A Multiple Sorting Procedure for Studying Conceptual Systems

Full text

1
A Multiple Sorting Procedure for
Studying Conceptual Systems
David Canter, Jennifer Brown & Linda Groat (1985)
in Brenner,M., Brown,J. & Canter,D. (eds) The Research Interview:
Uses and Approaches. London: Academic Press. 79-114,
Also in
D. Canter (ed) Psychologu in Action Aldershot: Dartmouth, 40-71
°We are grateful to Judith Sixsmith for her comments on this chapter
Constructs and categories
Many psychologists have emphasised that the ability to function in the world relates
closely to the ability to form categories and to construct systems of classification by
which nonidentical stimuli can be treated as equivalent (e.g., Miller 1956; Bruner et al,
1956; Rosch, 1977). As Smith and Medin (1981) have recently reiterated, if we had to
deal with objects, issues, behaviour, or feelings on the basis of each unique example, then
the effort involved would make intelligent existence virtually impossible. Thus, an
understanding of the categories people use and how they assign concepts to those
categories is one of the central clues to the understanding of human behaviour. As
consequence, one of the important questions for many investigations is the nature and
organisation of the concepts that people have, specific to the issues being explored.
In the present chapter, a procedure for exploring the categories and systems of
classification that people use in any given context will be described. It is known as the
multiple sorting procedure and it allows a flexible exploration of conceptual systems
either at the individual or the group level. The rationale for the procedure will first be
discussed and then examples of its use for answering a variety of different research
questions will be presented.
In this discussion of the nature and organisation of people's conceptual systems, an
important distinction must be made between the underlying categorisation processes and
the 'ordinary' explanations that people give for their actions. It is the former that is the
focus of this chapter; the latter will be discussed in another chapter by Brown and Canter.
As Brown and Canter argue, many research questions are best answered by reference to
'ordinary' explanations, especially when the expertise of the individual being questioned

2
and the unique understanding that he or she can bring to the situation are central.
Alteratively, in those studies where the research questions focus on the general conceptual
processes underlying the explanations people might give, it is frequently fruitful to
explore the categorical organisation of those conceptual processes.
For example, if the research were questioning the compromises involved in administering
a prison, then the explanations of the prison governor would becrucial to the study. On
the other hand, if the differences in the experiences of inmates of different prisons were
being explored, then it would be important to examine the classification schemes which
prisoners applied to their prison experience.
The study of personal systems of classification and of explanations are not inevitably
distinct. They are both part of the general psychological approach that places an emphasis
on understanding the individual's own framework for dealing with and making sense of
the world. They do, however, place an emphasis on different aspects of people's
conceptualisations, and are consequently of particular relevance for different types of
research question.
Thus, although the study of the personal categorisation processes people use in
thinking and acting can be recognised as being part of the general exploration of meaning,
it does focus especially on subjective or personal meaning. In the book they edited,
Personal Meanings, Shepherd and Watson (1982) show with many examples that in both
a clinical and a scientific mode of operation, practitioners need to construe the personal
meanings of others. This construal requires, they argue, the development of a framework
for describing the professional understanding of the meanings utilised by others. For such
a framework to be authentic, Shepherd and Watson insist, following Harre and Secord
(1972), that it must draw upon an intensive rather than extensive approach to data
collection. This involves working directly with individuals in their own terms, respecting
their ability to formulate ways of thinking about the world and their experience of it. This
contrasts with the use of standard questionnaires or structured interviewing procedures in
which the researcher has formulated views on what the respondent will wish to comment
upon, and so the researcher is, in effect, checking the extent to which the respondent will
endorse the experimenter's speculations.
The intensive study of personal meanings also has strong parallels in the studies of
subjective meaning carried out by Szalay and Deese (1978). They argue for a clear
distinction between 'lexical' and subjective meaning. The former being an attempt to
define in the public forum (as in a dictionary) the commonly held meanings of words, the
latter being an account of what is salient to an individual together with an indication of its
affectivity. They see the study of these meanings as being crucial to the understanding of
culture.
It is their focus on culture that leads Szalay and Deese to refer to subjective meaning,
and the client-oriented perspective of Shepherd and Watson which leads them to deal with
personal meaning. Yet they both have much in common. They both emphasise the need

3
to understand the conceptual system of the individuals being studied or helped. The
conceptual framework of constructs and the categories on which the respondent draws are
seen by both as the starting point for understanding the respondent's actions in the world.
In Britain, at least, the concern with understanding the personal conceptual systems of
individuals was spurred on by the writings of Kelly (1955) and helped along by the
prolific enthusiasm of Fransella and Bannister (e.g., Fransella and Bannister, 1977). Yet,
the view that each individual had a unique way of construing the world was not alien to
William James many years earlier (1890) andwas emphasised in some of Allport's
writings (1937), when he argued for the value of an idiographic approach.
Anthropologists and sociologists, especially those with a structuralist orientation, have
also emphasised throughout the present century the importance of understanding
individuals' systems of meaning (cf. Douglas, 1977). Furthermore, social psychologists,
in studying the role of situations in human behaviour, have established the importance of
the interpretations people make of those situations in which they find themselves (Argyle
et al, 1981).
Restrictive explorations
The brief review above reveals that there are two common themes in many disparate
writings on psychology. One is the need to explore the view of the world as understood
by the respondents in any enquiry. The second is the recognition that world view is built
around the categorisation schemes people employ in their daily lives. Yet, unfortunately,
psychologists have been influenced by a further consideration, which has tended to dilute
the impact of these two themes: the desire for quantitative, preferably computer
analysable, results. Most computing procedures have limitations that are so fundamental
that they are taken for granted and rarely challenged, thus influencing the data collection
procedures in ways so subtle that researchers are unaware of them. A self structuring
cycle is then set in motion. Data are collected in a form that fits known methods of
analysis. Standard analytical procedures gain in popularity and are easy to use because
they fit the usual data.Data are then commonly collected in the form appropriate to the
standard procedures. Thus the existing capabilities of readily available computing
procedures help to generate standard forms of data collection, even if those computing
procedures are inappropriate for the psychological issues being studied. Without going
into a lot of technical detail, a number of restrictions imposed by conventional, widely
used, statistical procedures can be summarized:
1. The most commonly used statistics tend to limit data to those having a strong,
clear, linear order. Categorical data are seen as being difficult to accommodate.
Thus, rating scales (e.g., 7-point) are much preferred to qualitative categories.
2. The procedures limit the structure of the set of variables, so that there are the
same number for each respondent. Furthermore, the number of divisions into
which each variable is coded is constrained, so that it is the same for all people.
Analysis is limited to the manipulation of arithmetic means and correlations over

4
large groups, but this requires that the actual organisation of the data for each
respondent is identical.
3. Because of their computational efficiency and mathematical elegance, statistical
models have tended to be restricted to those that are based on assumptions of
underlying linear dimensions and that consequently generate dimensional
explanatory models. Qualitative models, although increasing in popularity, are
still rare.
These constraints on the analysis of data have become more apparent with the
increasing availability of other computing procedures that do not have these limitations
and with the strengthening of the idiographic perspective. Indeed, it is being recognised
that the popularity of procedures such as the semantic differential (Osgood et al, 1957) are
due to the ease of data analysis rather than any conviction that they are measuring
important aspects of human experience. The semantic differential with its 7-point scales,
standard set of items, and factor analysis of results, has been shown to be insensitive to
differences between cultures (Osgood, 1962), and, although this may be of interest to
cross-cultural psychologists, it does not suggest itself as a technique that will reveal
important differences between individuals.
In effect, the semantic differential constrains the concepts people can reveal by
providing them with a set of terms to which to respond and by giving precise instructions
as to how that response can be structured. Procedures that allow some possibility for the
respondent to frame his/her own answers are essential if the essence of any given
individual's conceptual system is to be established. Thus, open-ended procedures,
especially those built around the interaction potentials provided by the one-to-one
interview, recommend themselves to the student of conceptual systems.
Many researchers (unaware of the range of analyses now available) are fearful of
embracing open-ended procedures because they are concerned that their results will be
difficult to interpret and the report or publication they seek will be difficult to structure.
Thus, even when they are interested in their respondents' understanding of the world, they
explore it through multiple-choice questions or very constrained rating procedures. Yet,
serious researchers will still insist on what is usually termed 'good pilot research'. This
does involve talking to people in a relaxed, open-ended way and learning from them about
the concepts they use in a particular context. It is often at this stage that the real
objectives, and in effect the major findings, of the research emerge. Subsequent research
frequently only clarifies a little, or provides numerical support for, the insights gained at
this 'pilot' stage. This is a curious state of affairs when data comes from one part of the
research activity and insights from another. Research would be more effective if
procedures allowed the interviewees to express their own view of the issues at hand, in
their own way, whilst still providing information that is structured enough for systematic
analysis and reporting.

5
Beyond the Repertory Grid
The interview, with its potential for subtle interactions and its concern with the
interviewee's understandings, is a fruitful context in which to explore people's concepts.
Over the past few years a number of procedures have emerged for generating and
examining people's conceptual systems within that context. One of the most popular is
Kelly's repertory grid (Kelly, 1955). As many authors have noted (e.g., Fransella and
Bannister, 1977; Adams-Weber, 1979; Bonarius et al, 1981), the repertory grid, deriving
as it does from a theory of people that puts emphasis on their conceptual systems, does
have much to recommend it; yet the Role Repertory Test, which has evolved from Kelly's
original proposals, is often used with less sympathy for Kelly's Personal Construct
Theory than might be expected. Furthermore, the forms of statistical analysis known to
Kelly limited the forms of development in grid analysis procedures, which has had direct
consequences for the forms of grid which he and his followers have developed.
Fransella and Bannister (1977) comment on many of these weaknesses of the grid as
used. They point out:
1. The grid 'has been turned into a technology which generates its own problems
and then solves these problems. Such problems do not necessarily relate to any
attempt to understand the meaning which the person attaches to his universe' (p.
113).
2. Grid use has been limited by the 'requirement that the subject present his
judgements in handy grid statistical format before we can analyse pattern'
(p. 116).
3. It is a fair guess that it is the mathematical ingenuity of the grid which has
attracted psychologists rather than its possibilities as a way of changing the
relationship between 'psychologist' and 'subject' (p. 117).
Recent developments in computing procedures have weakened some of these
criticisms, especially interactive on-line computing, which allows a much more flexible
exploration of construct systems (cf. Shaw, 1982), but the main point made by Fransella
and Bannister, that the grid technology as such has masked other possibilities for
exploring personal constructs, still remains.
The repertory grid technique is neither as unique in its contribution nor as definitively
special to personal construct theory as its users often claim. Kelly himself traces the
origins of the grid to the sorting procedures used by Vygotsky (1934) and others, and thus
puts his grid technique firmly in the realm of the exploration of categories and concepts.
He writes:
Methodologically the Repertory Test is an application of the familiar concept formation test procedure. It uses
as 'objects' those persons with whom the subject has had to deal in his daily living. Instead of sorting
Vygotsky blocks or BRL objects the subject sorts people. The technique bears some resemblance to the
sorting employed in the Horowitz Faces Test. It is also somewhat similar to Hartley's later procedure in which
he used pictures in a sorting test. Rotter and lessor have also experimented tentatively with the formation of
'social concepts' in the sorting of paper dolls of the Make a Picture Story (M.A.P.S.) Test. (Kelly, 1955, Vol. 1,

6
pp. 219-220)
He emphasises that his test differs from these other procedures in two ways. First, it is concerned with content, 'how the
items are dealt with', as well as the more usual concern for the level of abstraction involved. Second, it is 'aimed at 76
Psychology in Action
role constructs'. This latter emphasis was seen as being relevant to clinical practice, not an
inevitable emphasis for all studies of personal construct systems.
Instead of Q-sorts and Paired-comparisons
The Q'-sort' technique was, like the repertory grid, developed as a way of examining the
critical concepts people hold about role figures or events of significance to them
(Stephenson, 1953). But, while this method enables people to assign elements to
categories, the categories themselves are specified, usually as increments of an adjectival
scale. Moreover, the Q-sort is typically used in a form whereby the interviewee is
required to assign elements to the categories in a specified (almost always an
approximately normal) distribution (Pitt and Zube, 1979). The use of an enforced
distribution is defended, in part, on the grounds that the procedure provides data that is
more conveniently processed (Block, 1961), and eliminates the problem, inherent in
rating scale procedures, of different individuals calibrating the scale in different ways
(Palmer, 1980). These restrictions on the interviewee's sorting behaviour thus make the
Q-sort more akin to the semantic differential technique than to the intensive one-to-one
interview procedure we are advocating.
Other highly restrictive sorting procedures have recently been developed as an
alternative to paired-comparison judgements of similarity. For example, Ward (1977) and
Ward and Russell (1981) have used sorting procedures, in which both the sorting criteria
and the number of categories are specified, as a means of generating similarity matrices.
Although Ward argues that the process of sorting is probably more ’natural’ for the
interviewee than similarity judgements, the key argument for its use seems to be that it is
less time consuming than paired-comparisons while at the same time provides equivalent
similarity data that is suitable for multidimensional scaling procedures.
Indeed, the development of multidimensional scaling procedures grew out of the
analysis of similarity judgements of pairs of stimuli. Schiffman et al, (1981) see similarity
judgements as ’the primary means for recovering the underlying structure of relationships
among a group of stimuli' (p. 19). They go on to state that they think that similarity
judgements are to be preferred to verbal descriptors because such descriptors are 'highly
subjective and often conceptually incomplete' (p. 20). The view of the authors of the
present chapter is that, whilst there may be some validity to this contention in the
experimental study of perceptual stimuli, to which Schiffman and her colleagues
repeatedly make reference, such a view of all human conceptualisations is unnecessarily
restrictive and has not been defended with any theoretical strength.
It is our contention that perceived similarity is a more complex phenomena than can
accurately be described by a single rating. Perceived similarity may, in fact, be defined
by a set of multiple categorisations based on a wide variety of criteria. In many cases it

7
is the overall patterns that emerge as a result of the concepts people themselves
naturally apply to the objects or elements that is of psychological concern. Even when
people are unable to put words on their categorisation of elements, it is the structure
they impose on the world that should be the starting point for the psychologist, rather
than any general mathematical theory.
For, although interview-based sorting procedures do have a long history, it is only
recently that the full possibilities of this approach have become apparent. These
possibilities attempt to avoid the limitations of earlier procedures. The multiple sorting
procedure does not impose a view of the likely structure and content of an individual’s
conceptual system on the interviewee. It minimises the 'technique for its own sake'
syndrome by allowing the exploration of both the nature and the organisation of
concepts about any issue, maintaining the freedom and open-ended qualities considered
so essential by many researchers, yet still providing for systematic analysis of
individuals or groups. The use of the multiple sorting procedure and systematic
analysis of data from it is possible, in part because of developments in nonmetric
multidimensional scaling procedures, the use of which will also be illustrated later in
this chapter.
Sorting as a focus for an interview
As has been noted, many of the explorations of which interviews are a part are aimed at
coming to grips with the conceptualisations of the interviewee, whether it is a market
research study, such as that looking at the corporate image of banks (Frost and Canter,
1982), or a more theoretical exploration of architects' use of stylistic terms (Groat,
1982), or even research of a more pragmatic nature, looking at why people move house
(Brown and Sime, 1980). In all cases it is the particular categories and concepts people
use that is at issue, as well as the way in which they use them. The interview is
especially suited to these types of exploration, because the interviewer and the
interviewee can explore each other's understandings of the questions being asked and
because the one-to-one situation can accommodate a more intensive interaction.
Unfortunately though, the potentials of the interview are frequently its pitfall.
Asking open-ended questions in the relaxed way thought to increase rapport is the
formula for unanalysable material. What is needed is a way of providing a focus for the
interview to guide and structure the material produced without constraining the
interviewee unduly. Bruner et al (1956) were some of the first to show clearly the
possibilities for exploring the nature of the concepts people have by studying how they
assign elements to categories. Such a procedure provides a focus for the interview,
allowing other related material beyond that generated by the sorting to be noted. Yet
few have followed this lead out of the laboratory by using as elements material of
direct significance to the responding individuals.
Sorting procedures of various types have probably been used most frequently in the

8
environmental psychology field, perhaps because they enable researchers to use
illustrations and other visual material which are difficult to accommodate within other
procedures. Specific applications of sorting technique within environmental psychology
have ranged from those used simply to generate similarity matrices(Ward, 1977;
Horayangkura, 1978; Ward and Russell, 1981) to those seeking to integrate the sorting
process with the verbal descriptions and explanations inherent in a one-to-one interview
situation (Garling, 1976; Palmer, 1978; Groat, 1982). In the case of the latter, the
researchers have intentionally used the sorting technique precisely because it is free of the
limitations discussed earlier.
In the case of social psychology, one of the earliest approaches to the sorting stimuli is
found in the work by Thurstone and Chave (1929), who used the judgements people made
of questionnaire items as a basis for assigning weights to those items. It was the discovery
that the attitude of the judges influenced the pattern of judgement that lead Sherif and
Sherif (1969) to develop the 'own categories' procedure and direct measure of 'ego-
involvement' in attitudinal issues. In the 'own categories' procedure, judges assign
attitudinal items to categories in terms of how extreme the attitudes expressed are thought
to be. The distribution of the items in the categories is then used as a measure of the
intensity of the judge's own attitudes. This differs from the clinical object sorting
procedure, which Kelly discussed, in that the distribution of items to categories in a
predetermined sorting concept is the main concern.
Contemporary psychologists such as Eckman (1975) have also used free sorting
procedures in their work on normal verbal communication. In a related manner, Tajfel
(1981) developed a theory of social categorisation to explain 'in' and 'out' group
behaviour. Tajfel (1978) states: 'The role of categorisation in perceptual and other
cognitive activities has been for many years one of the central issues in psychological
theory' (p. 305). .
Tajfel's work involves organising information in certain ways, examining differences
and similarities between the content of categorisations. The chief function of this process
resides in its role as a tool in systematizing the environment for action. However, Tajfel
argues that assigning items to categories is influenced by the other categories in the
structure of a person's experience. His experimental work was aimed at unravelling the
complexities of prejudice through the process of category assignments.
Clearly then, in using the sorting procedures as an interview focus, the interviewer's
task is to identify the interviewee's salient categories and the pattern of assignments used
to relate categories to elements. The more freedom the interviewee can be given in
performing this task the more likely that the interviewer will learn something of the
interviewee's construct system rather than just clarifying his own. Such freedom should
extend to the range and structure of the categories, of which the constructs are composed,
as well as to constructs and elements sorted.

9
The multiple sorting procedure
The multiple sorting procedure advocated here asks little of the interviewees other than
that they assign elements to categories of their own devising; it differs from the other
previously discussed response formats in that no limitations are necessarily placed on how
the sorting is to be done. In fact, the respondent is
encouraged to sort the elements, using different criteria, a number of times. The rationale
for this less restrictive version of the sorting process is the belief that the meanings and
explanations associated with an individual's use of categories are as important as the
actual distribution of elements into the categories.
The actual act of sorting items is a common activity. For example, in choosing a house,
people will literally sort through the particulars sent to them by estate agents. In many
other areas of choice, whether it be clothing, books, partners, or political parties, there is
an explicit selection on the basis of a personal categorisation scheme. But even when a
selection is not overtly involved, such as in evaluating how successful a given setting is
likely to be for a given activity, or an essay in gaining a good mark, the judgement is
based on an implicit categorisation scheme. The multiple sorting procedure aims to bring
to light these personal schemes.
To carry out the multiple sorting, a person is presented with a set of elements and an
introduction and instructions as follows:
I am carrying out a study of what people think and feel about children [A] so I am
asking a number of people chosen at random [B] to look at the following
pictures [C] and sort them into groups in such a way that all the pictures in any
group are similar to each other in some important way and different from those in the
other groups. You can put the picture into as many groups as you like and put as
many pictures into each group as you like. It is your views that count.
When you have carried out a sorting, I would like you to tell me the reasons [D]
for your sorting and what it is that the pictures in each group have in common
[E].
When you have sorted the pictures once I will ask you to do it again [F], using
any different principles you can think of and we will carry on as many times as you
feel able to produce different sorts. Please feel free to tell me whatever occurs to you
as you are sorting the pictures.
The items underlined and indicated with letters in [] are those components of the
instructions that are likely to change for different procedures in relation to different
research questions. It must be emphasised, however, that these instructions are only a
general statement of what is possible. The flexibility of the procedure is such that many
different variations of the instructions are possible. Pilot work is always essential in order
to discover what particular instructions are appropriate for each study, although typically
all components [A] to [F] must be explicitly dealt with.

10
The elements to be sorted ([C] in the instructions), depending on the research question,
may be generated by the interviewee or the interviewer; they may be labels, concepts,
objects, pictures, or whatever, as will be illustrated. The person is usually asked to look
through the elements to familiarise him/herself with them; also, the purposes of the
research enterprise are explained (relating to instructions components [A] and [B]). In
particular, it is pointed out that the interviewer is interested in the interviewee's ways of
thinking about the elements presented. The interviewee is then asked to sort the elements
into groups so that all the elements in any given group have something important in
common, which distinguishes them ider a preliminary example, here drawn from a
multiple sort carried out with lbler we will call Ace. We were interested in Ace's views of
various casinos, irt of a larger project to study what it was that gamblers enjoyed about
ling. The particular purpose of the sorting procedure was to see the basis on 1 a gambler
selects which Casino to visit and to get some understanding of his of the Casinos
available. We wanted to know what sort of world a gambler s around in, what type of
choices he sees as being available to him.
Ace was asked to list on cards all the casinos he knew in any detail and to assign s for his
own convenience. For the researcher's convenience, each card had a letter on the back. On
his first sort, Ace chose to divide the cards into three lings. These groupings were
recorded as shown in Table 1, by the simple process of noting under each group the letter
for the card, this stage, the researcher has an indication, without any verbal labelling, of
category scheme for the respondent. Such information can be very valuable, especially
when working with groups of people who are not especially articulate. There are a number
of further developments of the procedure possible within the same framework. The verbal
concomitants of the category scheme can be red by asking the interviewees to indicate the
basis on which they have carried le sorting, as in the instructions [D] and [E].This
generates two levels of description. The first is a superordinate description of the principle
for the sorting, instructions [D], for example, 'whether the casinos have frills or not, or 'the
amount of money to play the lowest stake'. The second is a set of category labels each of
the groups (instructions [E]), for example, for the 'frills' sort, Ace's categories were 'places
with no frills', 'places with sedate dining', and 'vaudeville'; for the 'stakes’ sort, Ace's
categories were 'less than £5,' 'between £5 and £25', greater than £25'.
A useful way of recording this verbal information shown by reference to Ace's sorting of
casinos is also in Table 1. The categories are summarised with a description of the
category scheme for the sort as well as labels for each of the groups within this sort. Other
comments and points of clarification made by the indent can easily be accommodated
within this format, as well as any order might be given to the category groupings. Given
the value of the procedure for exploring a content domain, these comments may generate
material of value in their own right. Thus, the researcher need not reduce the responses to
bipolar scales, which are often ambiguous when considered at some time after the
interview.

11
Table 1 Record of Ace's Sorts
First Sort: 'Class of Casino'
1. 'Gaming Halls': G, H, D, A
2. 'Middle Class': B, C
3. 'High Class': E, F Second
Sort: 'Type of Frills'
1. 'Just Gambling': A
2. 'Baudeville': B, G, H
3. 'Sedate Dining': E, D, C, F
Third Sort: 'Size of the Stake'
1. 'Less than £5': A
2. 'Between £5 and £25’: G, H, B
3. 'Greater than £25': C, D, E, F
Fourth Sort: 'Most likely place for me to make money at'
1. 'Most likely’: A, G, H
2. 'Not so much': B
3. ’Too expensive': C, D, E, F
Fifth Sort: 'Preference'
1. 'Most preferred': A, G, H, E
2. 'Solid Casinos': C, D
3. 'Bit Quiet': F
4. 4. 'Did not like at all':B
Casinos: A-Golden Nugget; B-Playboy; C-Park Lane; D-Palm Beach; E-Hereford; F-Park
Tower; G and H-Las Vegas casinos.
Unlike the analyses discussed by Schiffman et al (1981), and used, for example, by Ward
and Russell (1981), the multiple sorting data need not be reduced to association matrixes,
typically aggregated across groups. Both the superordinate description and the category
labels can be subjected to content analysis and to multivariate statistical analyses, but it
should be noted that no structure or order to these descriptions is initially assumed or
implied. This is particularly important for the category labels. The bipolar dichotomies of
rating scales are not assumed, nor are the order of items from ranking or scaling. If the
interviewee specifies a particular order, as in the 'amount of the stake' example, then note
can be taken of that, but if any order might be more obscure, as in the 'frills' example,
then that can be utilised as well. Indeed, category schemes frequently emerge that are not
simply bipolar; and this raises important questions about the extent to which such
bipolarity, assumed in much research, is an actual feature of psychological processes or
an artefact of the structured measuring instruments used. Furthermore, in some cases an
interviewee may choose to sort some of the elements and leave others as irrelevant to the
overall sort taken into account. This irrelevant group is treated as forming a further
category and can be incorporated in the subsequent analysis without any loss of
information or imposition of a superordinate categorisation on all the elements.

12
Having produced one sort of the elements, it is of value in many projects for the
multiple sorting to continue by asking people to examine the elements again and try to
produce another category scheme with new descriptors (instructions [F] Table 1 gives a
summary of all five sortings produced by Ace in the interview conducted with him.
Analysis of this will be discussed later. It should also be note: that the number of
elements sorted here (eight casinos) is limited by the number of casinos to which anyone
has ready access in London, and might not give the richest picture possible. The process
can continue as many times as the interviewee feels able to sort the elements. In research
carried out to date, two or three sorts are common, but up to seven or eight are frequently
possible, with 15 or more occurring in some cases. The number of elements that it seems
fruitful to use is in the region of 15 to 25. Depending on the individual, of course, a
complete set of sorts may take anything from 10 minutes to well over an hour, which may
also be extended insofar as the sorting is used as a focus for other issues explored in the
interview.
Hypothesis Testing
The example used so far, from the casino sortings, is simple enough to illustrate the
procedure in use with one person, as a basis for getting to understand some particular
aspects of that individual's conceptual system. But the power of the multiple sorting task
as a means for testing hypotheses of conceptualisation; common across a number of
people, can also be readily illustrated. Let us consider for example, the work of Bishop
(1983).
Bishop had as a central concern the role the age of buildings played in people’s views
of their surroundings. However, he was aware that his own fascination with the age of
buildings might have given him a particular perspective and that this way of thinking
about buildings might not have been very important to most people However, since it is
clear that people can comment on a building's age, any dire: questioning about age or its
significance might have given a spuriously hip: weighting to the role of age. Bishop
therefore carried out a multiple sorting w a number of respondents. He did this by
preparing a set of photographs of buildings which differed in age and gave them to people
to carry out a set of free sorts, as described above.
Bishop's hypothesis was strongly supported. Thirty of the thirty-five people he asked used age as a
basis for sorting, although only eight used it as the basis of their first sort. Bishop went a step further
and classified his respondents in ten-; of the type of age sorting they made, showing quite
convincingly that the.: understanding of architectural age varied greatly, although they spontaneous:
used the concept. This differentiation of his groups laid the foundations for development of his
study.
To see the potential range of uses of, and variation on, the multiple sorting procedure
and ways of analysing data from it, we will now turn to other specific examples.

13
Variations in elements sorted
The types of elements that can be used for sorting are limited only by the imagination of
the investigator and the practicalities of what can be carried about and sorted on the
surfaces available. Indeed, the development of microcomputers offers some intriguing
possibilities for increasing the range and variety of elements that can be sorted; for
example, moving objects, even for monitoring the process of sorting by recording the
hesitations and false starts that might otherwise get lost in a paper and pencil record. From
the initial uses of sorting procedures, as noted earlier, a great number of different objects
have been sorted. But in the more recent explorations of the content and structure of free,
multiple sortings, a variety of representations of objects, or simulations, have also been
used. Groat (1982), for example, used photographs of buildings taken from architectural
magazines, books and slide collections, to explore how architects' ways of thinking about
famous buildings compared with the conceptual systems of accountants. Oakley (1980)
used labels of places to stay such as hotel, parents' home, or hospital, to examine the
views residents had of Salvation Army hostels in which they were living. Grainger (1980)
had architects and their clients sort the activities a proposed building might house, in
order to establish their different understandings of what the building's functions were to
be.
Focus of the elements
In general, the more concrete and specific the elements are and the more familiar the
respondent is with the elements, the more likely it is that they will be able to produce a
number of rich and varied sorts. Abstract labels of possible emotions, for example, are
likely to encourage relatively few sorts, whereas a set of detailed descriptions of actual
places a person has direct experience of is likely to lead to the generation of a great many
sorts from each respondent. The selection of elements will always need to be guided by an
awareness of what the respondents are normally used to considering and whether the
research is best served by a simulation, a representation of some entity, or by reference to
the actual phenomenon itself.
A further consideration in selecting the range of elements to use is how big a variation to
select. If general stereotypic sorts are of interest, then a very broad range across the
element domain is advisable. For example, a study of conceptualisations of medical
specialities among medical students would possibly be best studied using a list of all the
specialities as organised in a medical text book. On the other hand, if students' individual
choices of future careers were being explored,then a subset of specialities described in
relation to their working context and with reference to the students’ direct experience may
well generate more specific sortings, revealing the idiosyncrasies of particular individuals’
conceptual systems.
Generation of the elements
In considering how the elements should be generated, two matters need to be considered:
(1) whether the elements are to be generated by the investigator or not, and (2) whether the

14
elements will have a specified structure or be a sample of some identified population.
If the researcher is setting out to test some hypotheses about people's conceptual
systems, then it is likely that the elements will be identified by the researcher. For
example, Groat (1982) chose photographs of buildings to test her particular hypotheses
about architects’ conceptions of styles. On the other hand, in the example with Ace
described above, it was essential to elicit the casinos of which he had direct experience.
Similarly, Groat ensured that her set of photographs included three specified styles and
four building functions, whereas for the gambler, all the casinos he had actually visited
were used.
The generation of the elements thus has a direct bearing on whether the sorting
procedure is to be used for exploratory, heuristic, or descriptive purposes or hypotheses
testing. This procedure then has potential at many different stages and in many different
areas of research endeavour.
Construct elaboration
As has already been mentioned, the sorting procedure allows constructs to be elaborated in
many different ways, depending on the goals of the research activity and on the
capabilities of the respondents. If the research is aimed, for example, at identifying
whether residents of a hostel think of its function differently as a consequence of how long
they have been there, then a knowledge of which other places of residence they think are
similar to their own hostel may be of great value. For instance, Oakley (1980), in his study
of hostel residents, generated data from the sorting procedure without probity for labels of
the categories being used. His respondents did find verbalisation difficult, but the
groupings of the elements themselves provided him with some useful basic data, which
enabled the Salvation Army to clarify some of the principles on which to consider the
provision of new hostels.
On the other hand, if the aim of the research had been to look directly at the processes of individual rehabilitation, it would
have been necessary to uncover the concepts residents use for deciding where they are going to stay. In this case, labels
associated with each category, or group of elements, would have to be elicited. It is likely that a different set of elements
would have been of use in such a study, so that respondents with few verbal skills could be encouraged to express their
understanding of what is available to them.Foci of analysis
The reluctance of earlier researchers to use procedures as open ended as the multiple
sorting task, may to some extent be due to the difficulty they perceive in analysing the
data generated. However, besides the developments in content analysis, discussed by
Barbara Mostyn in an earlier chapter, all of which can be directly applied to the category
descriptions generated during the sorting, it is possible to use nonmetric MDS procedures.
These enable the analysis to be focused on different issues depending on the research
question.
What is meant here by 'focusing' is that the research procedure can be tuned to any of a
number of different aspects of the material potentially available. The researcher can, for
example, choose to deal with differences between groups or to concentrate on particular
sorting criteria within individuals. The multiple sorting procedure as such has no special

15
limitation as to the research enterprise for which it is appropriate. It is simply a data
generating procedure which can be harnessed to the goals of a wide range of projects.
First, we shall consider studies of group differences, looking at the relationships
between elements and then the relationships between categories. Second, we shall
consider studies of differences within individuals.
Group differences
Elements, concepts, and people. In any study of conceptual systems there exist three
broad ways in which the data can be examined: by considering differences between the
people, differences between the elements, or differences between the concepts and
categories to which the elements are assigned. The data matrix that is always possible can
be thought of as a cube, as shown in Figure 1.
The importance of the data cube is that it shows the variety of possibilities there are for
data analysis. In essence, each of the planes of the cube, A, B, or C, provides a different
analysis possibility by aggravating combinations of the dimensions. For example, in the
prison study described below only one aspect of the concepts was dealt with - similarity
to 'this prison'. Thus, it was plane A, elements across people, which was the focus of
analysis. A study centring on the structure of the concepts a group of people have across
one or many elements would be dealing with the data in plane C, because it would require
the differences between people to be ignored. A study comparing people in their
conceptualisations would be drawn from plane B.
Figure 1 can be used as a guide to help clarify the research question by showing which
'slice' through the cube is being used and what ’collapsing' of data from another
dimension is necessary. It is extremely complex to carry out analyses that combine all
three aspects of the data in one operation. It is usually more appropriate to proceed
through the analysis in stages, working with one plane of the data cube at a time.
One of the most obvious uses for the multiple sorting procedure is to compare the
conceptual systems of different groups. There is now a large literature showing how
different groups of individuals addressing the same topic may have quite different
conceptualisations about it, which in turn give rise to different evaluations of the issues at
hand and related actions. However, as Canter (1977) pointed out, studies using standard
response formats, such as the semantic differential, tend to underestimate the difference in
perspectives between individuals. Indeed, such procedures tend only to indicate small
differences in emphases rather than revealing the radical differences in conceptual systems
commonly present when different groups interact in relation to some common object.
The repertory grid is commonly used for group comparisons and frequently with some
success (Adams-Webber, 1979); but it does have severe practical limitations, both in the
number of elements that can be dealt with in any given study and in the overall time taken
to complete a grid (Canter et al, 1976). For comparisons of groups it is also frequently the
case that much of the detail generated by the grid is superfluous and not used in analysis.
An open, free sorting procedure often has the advantages of individual sensitivity without
the procedural disadvantages of the grid.
Ambrose showed the value of a sorting procedure for revealing group differences in a
study reported in Canter (1980). In a study of different prisons, inmates and members of
staff were asked to sort labels describing places in which people might live. One of the
labels to be included in this sort was ’this prison'. A matrix was derived for each
respondent showing which other card was put into the same sort as the cards specifying
'this prison'. This matrix, in effect, consists of a series of profiles for each individual
indicating whether or not they saw their particular prison as similar to all the other places.

16
A multidimensional scalogram analysis (MSAI; see Lingoes, 1973, and Zvulun, 1978 for
details) was carried out to see whether there were any similarities or differences between
the different respondents and their different institutions. Figures 2 and 3 reveal the
results.
Figure 2 shows the partitioning of the space for the prisons and Figure 3 the
partitioning of the space for the prison staff and the prisoners. Each point in this space
represents an individual. The closer together any two individuals are the more similar are
their profiles in the data matrix. The advantage of the MSA procedure is that it only deals
with each response as a categorical one comparing the categories with each other. No
order is assumed between the various categories, nor is any similarity of meaning
assigned to the categories for each of the variables. The variables in this case were created
by each of the cards used in the sort.
Looking at these MSA results, it is clear that there is no difference between staff and
inmates. No clear regions of the space can be identified for these two different groups. In
other words, there is not an effect of role on their perceptions of the particular prison. On
the other hand, there are clear regional partitions for the different institutions.
Furthermore, the order of the three regions through the space places the institutions in a
sequence, from those that are most strict in their regime to those that are least strict. This
shows that the strictness of the regime can be recaptured from the assignment of the
institutions to the place categories.
Figure 1 The data cube
ELEMENTS A

17
It is also interesting to see here that through individuals' free sorts, the perspective shared
by prisoners and staff on the nature of the institution is revealed. In this particular case the
role groups may well be assumed to be individuals who would not necessarily be expected
to work together. Nonetheless, the sorting demonstrates that they do share an
understanding of the nature of the institutions.
Figure 2 MSAI of card sort of residential setup by respondents in
three prison establishments. W, satellite design prison in a rural
location with a regional catchment area; R, block design prison in
an urban location with a regional catchment area; P, radial design
prison in an urban location with a local catchment.
1
Figure 3 MSAI card sort of residential settings showing staff and inmate respondents: S, staff;
I, inmates.

18
The significance of this finding is increased when it is realised that there is no
'ay in which the prisoners could have guessed what the prison officers would lone in a free
sort, especially across three different institutions. Yet, in a conventional questionnaire it
would have been very difficult to remove social desirability bias from such a situation.
Furthermore, the language requirements in of fluency and vocabulary that would have
been necessary to question people these subtleties would also have been demanding on
many prisoners. However, in the present circumstances a simple assigning of cards to
sorting tries appears to have been sufficient to reveal some intriguing differences. Of
course, a detailed understanding of the conceptual basis of these differences not be
achieved without a further analysis of the concepts used by staff and inmates.
Unfortunately, because of the constrained nature of what was possible in the prison
system, the details of the conceptualisations were not explored by Ambrose. We will
therefore turn to another example to illustrate these elaborated studies.
The structure of the elements. In some situations it is of particular interest to try and
identify the type of conceptual system groups are using. In other words, the structure and
content of elements shared by groups of individuals need to be established. With the
prison example this was less possible because only one sort was made by each individual,
and the particular analysis carried out (reported above) focused on the relationship of one
element to all the others. However, it is also possible to carry out analyses that look at the
comparison of every element with every other element. Hawkins (1983) did just such an
analysis using labels of a variety of possible places with residents of different psychiatric
day centres.
Hawkins asked each individual to sort a set of cards labelling places where they might
spend their day. Three different day centres were involved in this study, and Hawkins was
able to compare the structure of the elements for each of these. She did this by creating an
association matrix containing the frequency with which every element was assigned to the
same group as every other element, across all sorts and respondents (plane A in Figure 1).
A Smallest Space Analysis was carried out on each of the association matrixes created for
each of the day centres (cf. Lingoes, 1973; Shye, 1978 for details).
This analysis generates a plot showing that elements more similar to each other in the
pattern of sortings to which they are subjected are closer together. Figure 4 shows the
SSA plot for the three different hostels; to aid interpretation regions have been indicated
on these plots. It can be seen that the overall structures have a number of similarities.
They all show the existence of five groups of elements: work, leisure, service, therapy,
and residential. They also show that these groups are qualitatively sequenced, around a
circle, rather than having a simple, quantitative linear order to them. Yet, there are some
clear differences in the way the residents of the three day centres see the various places.
In other words, it is not solely their view of the location of their own particular hostel that
is different, but the residents of each day facility actually have a different system of
thinking about other possible locations. For example, to the residents of centre C the
therapeutic group, including 'this day centre' is seen as being between leisure and
residence, whereas for the people in centre A it is between work and residence. This

19
reflects a differing emphasis on rehabilitation to work in the various centres. For centre B,
'this centre' and the other therapy elements are confused with residential items coming
close to hotel and house. This relates to the fact that the residents of these day centres
have typically been using them for up to eight years and they are more chronic and indeed
settled into their daily use of these places as somewhere to go.
This is particularly important for both design and development of therapeutic
programs. If the whole regime of a particular psychiatric day centre relates to the way the
residents conceptualise the opportunities available to them, and this consequently differs
from one centre to another, then any generalised guidance suggested for use in all day
centres, which aimed to help people to move into the community, would be ineffective if
it did not take into account the conception of 'the community' particular to any given day
centre. The results indicate that the attempts to move individuals from their centre out
into the community require subtle understanding of how those individuals actually
conceive of the community itself.
The structure of constructs. Another focus for analysis is the establishment of the
underlying constructual processes the individual brings to bear on a pattern of elements.
This issue is particularly well illustrated by the work of Groat (1982) to which reference
has already been made. She was concerned with whether or not architects would
conceptualise works of architecture in different ways from accountants. She was able to
establish quite clearly, using procedures like those described in earlier sections, that the
actual sorting of the elements was different for the two groups. However, it was important
to Groat's work that she should establish what types of conceptual issues were actually
paramount in the judgements being made.

20
(a)
Figure 4 Position of each place on SSA-I plot looking at the frequencies from (a) centre A,
(b) centre B, and (c) centre C.
In order to examine the conceptual issues, she developed a matrix based on the
categories within similar sorts. In other words, she first identified, through context
analysis, the types of sort used by each of her groups of respondents. For example, both
groups contained a number of individuals who referred to both the function and the style
of the building. Two separate matrixes were produced, one for the function categories,
and one for style. In each of these matrixes, the categories were the columns and the 24
buildings sorted were the rows (plane C of Figure 4). The cells of this matrix were a
dichotomous score indicating whether that particular building was ever assigned to that
category. Smallest Space Analysis of this data showed that the structure of the 'function'
categories was very similar for both groups. They both divided the photographs into
domestic and non-domestic buildings and within each of these groups distinguished the
buildings in terms of scale. However, the style categories were quite distinct. The
accountants made a big distinction between what they saw as ’traditional’,'modern1, and
'futuristic' buildings, whereas architects used a classification scheme clearly drawn from
the literature of architectural criticism, distinguishing 'Expressionist', 'Brutalist', and 'Post-
Modern'.
Groat's study thus shows very well how a detailed analysis of the structure of the
conceptualisations of the two groups can reveal subtle differences and similarities in their
category schemes. Such differences would normally be hidden by structured
questionnaire and interviewing procedures and would be extremely difficult to establish
with repertory grids, unless separate grids were developed for each respondent with the

21
consequent time consuming analysis that would involve
Differences within individuals
In our initial example of a multiple sorting we referred to Ace, a gambler. As part of the
same study, a casino manager also went through a sorting procedure using the casino and
parts of casinos of which he had direct experience. The results of these sortings are given
in Table 2. The sortings from these two people, taken together, serve to illustrate the way
in which very specific foci can be developed for analysis, dealing directly with the
unique, idiosyncratic conceptualisations of particular key individuals.
When individuals carry out detailed sorts on elements that are special to themselves,
there is always a possibility that over a variety of sorts they repeat similar categories,
simply assigning different labels to each categorisation. Thus, an individual who is fluent
but not especially cognitively complex may generate a large number of apparently
different sorts, which on closer examination are found to have little in the way of
variation between the different sortings.
This is an especially important point if comparison is to be made of individuals, because it
is the key aspects of their conceptual systems that we need to understand, not simply how
many words they can string together. Thus, it is necessary to do an analysis for each
individual and to reveal the main conceptual structure within which the individual is
working. In regard to the gambler and the manager separate analyses for each was carried
out and a schematic representation prepared to facilitate a comparison of their two
conceptual systems.
The analysis here again used MSAI. In this case each of the sortings acted as a separate
variable and each individual had a separate matrix. The matrix consisted of the elements
as rows and the sortings as columns (a slice through plane B of Figure 1). The cells of the
matrix are numbers indicating the sorting categories to which the different elements were
assigned. Each matrix was put into a separate MSA analysis. The analysis, in this
instance, generates a configuration in which each element (in this case, casinos) was a
point in the space. The closer together any two casinos are in this spatial representation
the more similar they are in terms of the categories that are assigned to them over the
number of sorts carried out by each individual.

22
Table 2
Record of Casino Manager’s Sortinga
First Sort: 1 Staff Recruitment *
1. 'Career Staff: A, B, C, F
2. ’Recruit from outside’: E, D,
G Second Sort: ’Staff Training:
1. ’Little training’: A, B, C, F
2. ’More training’: E, D, G
Third Sort: ’Staff Benefits’
1. ’Mainly for senior Staff: A, B, C, F
2. ’Also for lower Staff: E, D,
G Fourth Sort: ’Sex of Staff
1. ’Male only’: A, B, C, E, D
2. ’Male and Female’: F, G
Fifth Sort: ’Staff Contact with Customers’
1. ’None’: A
2. ’Good with company support’: G, E, D
3. ’Good with no company support’: F
4. ’Unclear’: B, C
Sixth Sort: ’Staff Experience’
1. ’Trainee Staff: A, B, C
2. ’Mixed’: E, D
3. ’Inexperienced Staff’: F, G
Seventh Sort: ’Whether Takes Cheques or Cash’
1. ’Cash’: A, B, C
2. ’Mixed': E, D
3. ’Cheques’: F, G
Eighth Sort: ’Concern for Customer Quality’
1. ’Quantity only’: A, B, C, E
2. ’Quality and Quantity’: D
3. ___________________ ’Quality’: F, 6 _____________________________ ____
________________________ _________________________________________ ____
aCasinos: A-Golden Nugget; B-Palm Beach; C-International; D-Hereford; E-Park Lane;
F-Curzon House; G-Gladbroke.
Table 3 shows the data matrix derived from the sorts illustrated in Table 1. Figure 5a
shows the MSA for the gambler and Figure 5b shows the MSA for the manager.

23
The partitioning of these figures is derived from an examination of the way in which
each individual sort contributes to the spatial configuration. Thus, it is clear that the
manager divides casinos up on the basis of how they deal with the clientele and how the
overall casino management deals with their staff. This gives a two-way classification of
casinos: those that select their staff carefully but are not too selective of their clientele;
those selective about their clientele but not so careful of their staff; and those not
especially careful about how they chose their staff or their clientele. This reveals the
division the manager makes between the staff and the clientele and the way in which his
perspective relates to selectivity and overall standards. At first sight, the gambler's MSA
reveals a very different sorting.
Essentially, there is a two-way division between those casinos that are very up-market
and those casinos that are more general. The gambler makes a more precise distinction
within the more general casinos between those that have added frills like the famous
Playboy Clubs, or those that are just large gaming halls with little extra, and a group in
between. Clearly, the gambler makes much more refined judgements about the nature of
the action going on within the casino than does the manager. However, they both share
the superordinate categorisation of how selective the casinos are.
This selectivity of casinos throws an interesting light on the whole gambling experience.
It shows that an individual, in effect, is playing himself into some sort of exclusive club.
These casinos, then, unlike those in the United States, may gain some of their important
qualities from the way in which both the management and the gamblers draw lines
between who can afford to be in which places. Certainly, further discussion of these
Table 3
Data matrix derived from the sorting produced by agea
Elements
(Casinos)
First
Second
Third
Fourth
Fifth
A
1
1
1
1
1
B
2
2
2
2
4
C
2
3
3
3
2
D
1
3
3
3
2
E
3
3
3
3
1
F
3
3
3
3
3
G
1
2
2
1
1
H
1
2
2
1
1
aNumber in cells are categories derived from sortings as shown in
Table 1.

24
conclusions with the respondents here as well as with other management groups would be
necessary to test that hypothesis more fully.
Again, it would be difficult to see quite how such a result could be derived from a
conventional questionnaire procedure. Open-ended interviews could well have revealed
the same sort of material, but they might have hidden the underlying structures in people's
conceptualisation, while of course emphasising other aspects of casinos that may well be
important.

25

26
Figure 5 Summary MSAI of (a) Ace’s sortings (see Tables 1 and 3 for basis), and (b)
Casino Manager's Sorting (seeTable2 for basis).
Category salience. One aspect of the categories employed in sorting that is especially
amenable to exploration, but not as yet examined, is the salience or significance to the
individual of the categories used. The role of any categorisation scheme in the overall
sorting structure has been explored, as discussed above, but the importance to the
individual of one sort over another has not been scrutinised. Yet, the sorting procedure
does lend itself to such examinations by virtue of a number of properties that potentially
might reveal salience.
The range of sort convenience. One way in which the importance of different concepts
can be examined is in relation to the appropriateness of those categories for all the
elements involved. Kelly (1955) discusses the importance of establishing what he calls the
'range of convenience' of a construct, which he defines as covering 'all those things to
which the user found its application useful' (p. 137). In the sorting procedure, it is always
possible for the respondent to produce a sort which only covers a subset of the elements
and for the remaining elements to be assigned to a general class indicating their
irrelevance to the sorting criteria. Measures and content analyses of the range of items to
which different category schemes were applied would help to clarify the salience of
different sorting categories.
The significance of sort order. In a multiple sorting task it is clear that sorts follow each
other in a distinct order. The question is therefore raised as to what the significance might
be of the order in which sorts are elicited. In their study of different numbers of sorts
Rosenberg and Kim (1975) came to the conclusion that sort order carried no significance.
Unfortunately they only examined the first two sorts rather than looking at a larger
number. Informal discussion with respondents does suggest that there may be contexts in
which the order does carry significance and relates to the salience of the sorting categories
being used. Some instructions may heighten this possibility, particularly instructions
emphasising the use of sorts that the respondent believes reveal 'important aspects of the
elements being sorted’.
Studies asking the respondent directly what importance they attach to different sortings
are quite feasible as well as content analyses of sorts in terms of sort order. Such studies
would be of value not only because little is known about the significance of sort order, but
also it may reveal some interesting properties of conceptual systems not illuminated by
other procedures.
The relevance of category distribution. As discussed above, an early version of a sorting
task was used by Sherif and Sherif (1969) in the development of Thurstone attitude
scales. They had judges assign attitudinal statements to an ordered set of categories. They
argued that the distribution of items in a sorting revealed something of the intensity of the

27
sorter's judgements; the more skewed the distribution of categories the more extreme the
judges' views. In other words, if people assigned a similar number of elements to each
category, then they were likely to hold much less extreme views than someone who put
most of the elements in one or two outlying categories.
Although the Sherif and Sherif 'own categories' procedure, as they called it, is different
in a number of important respects from the multiple sorting task, it does point to the
possible value of studying the number of categories used and the number of items
assigned to each category. Simple indexes of the distribution of items per category could
easily be devised and used as a basis of this study. Such indexes could be directly related
to the literature on cognitive complexity (Bieri, 1971; Streufert and Streufert, 1978) and
thus provide an important link to the discussion of the role of cognitive structure in
attitude change.
Theoretical clarification
As discussed, the multiple sorting procedure has evolved out of a variety of origins in
clinical, experimental, environmental, and multivariate psychological research. It is
consequently inevitable that some of the differences of opinion between practitioners in
these areas has provided a basis for some confusion in the theoretical issues underlying
the use of the multiple sorting procedure. All of these issues require debate and open up
interesting areas of possible research.
Categories and constructs. Kelly (1955) makes it very clear that constructs are distinct
from category schemes, although the labels given to categories may usefully identify
constructs in certain circumstances. He even writes that he uses the term construct 'in a
manner which is somewhat parallel to the common usage of "concept" ' (p. 69). One
important assumption about constructs in Kelly's terms is that they are dichotomous.
They enshrine bipolarity of aspects of the similarity and differences between elements.
In this sense, categories to which elements in a sort are assigned are the constructs of
the user, although only one pole of the construct may be specified. However, in Kelly's
terms, the bipolarity of the constructs is an assumption of his theory, it is not open to test
within his theory, nor as a consequence can it be tested using a repertory grid. This is not
the case for multiple sorting. If a person assigns elements to categories that can be
ordered from least to most along a single bipolar concept, then that category scheme
reveals a construct (e.g., Ace's sorting according to size of minimum bet). If, however,
the sorting produces a set of categories that are multipolar, then it would be inappropriate
to regard this category scheme as consisting of constructs (e.g., Groat's architects'
classification of building styles as Brutalist, Post-Modern, Expressionist, etc).
Thus, the multiple sorting procedure does allow one of Kelly's fundamental
assumptions to be tested. Indeed, it is the emergence of classification schemes that are not
obviously constructs which is one of the starting points for considering the multiple
sorting procedure rather than the repertory grid. But this leads to the question of the

28
conditions under which people use constructs as opposed to multipolar category schemes
and the possibility of converting one system of classification into the other.
98 Psychology in Action
Individual differences in sorting competence. Anyone who uses the multiple sorting
procedure will come across respondents who will find the task difficult and challenging to
complete. This raises the question of how natural the task is to all people and of what
differences there might be between people in their ability to carry out a multiple sorting
task. Of course, the specific nature of the task itself does need to be considered. A sorting
of abstract concepts is likely to be more difficult than a sorting of places to go on holiday.
It is also necessary to distinguish between difficulties people may have in understanding
what they are to do and difficulties in actually doing it.
These individual differences are of interest because the procedure does have roots in
clinical psychological concerns with understanding the difficulties people have in coping
with the world. If there are circumstances in which respondents find it difficult to form
categories stable enough to describe, then the reasons for this should be searched out. The
comments people make when carrying out sorts can give a valuable clue here to the
difficulties they are facing and these can be related to measures such as the number of
sorts and the time taken to complete a sort. The exploration of who finds what type of
sorting difficult and why is likely to repay the effort involved.
Should sortings be reliable. The sorting task is likely to be a self-exploration for the
individuals doing it, a learning process in which they come to understand more about their
own conceptual system. As a consequence it is possible that an individual would not give
the same sortings twice. Certainly the order in which he/she carried out his/her sortings is
likely to vary from one session to the next. What is being studied is the overall conceptual
system a person uses. It is likely that using two parallel sets of elements, in analogy with
parallel forms of reliability testing, would not obviate the effects of the increase in self-
understanding associated with a sorting task.
No published studies can be found dealing directly with these issues, but advocates of
repertory grid procedure (e.g., Fransella and Bannister, 1977, Chapter 6) have gone to
some length to argue that reliability can easily be a measure of the insensitivity of a
procedure to changing circumstances rather than a valuable psychometric property. With
respect to the multiple sorting procedure, it is likely that stable individuals would generate
reliable responses over two or three sorting sessions, but only if the procedure itself did
not contribute to a fuller understanding of their conceptual systems or their personal
growth. Only direct tests of these important questions can help answer them. Here, as with
the other questions, the comments people make during the sorting procedure could
provide valuable clues.
Exploring face validity. The validity of a sorting must depend a lot on the conditions
under which it is carried out. The essence of a sorting task is to establish the individuals'
own understanding of their personal conceptual system. The extent to which individuals
will feel able and be able to generate that system will depend on how they understand the
instructions, the personal relevance of the elements,

29
A Multiple Sorting Procedure 99
and so on. Thus, as much as any other data collection procedure, the conditions under
which the data are collected need to be carefully reported and interpreted, in terms of how
the procedure was experienced by the respondent.
In this framework, face validity is given considerably more prominence than in other
psychometric procedures. This is because one very important meaning to the validity of a
sorting task is the extent to which the respondent and the investigator have a shared
understanding of what the procedure is measuring. In this respect the details of the
results, both in content and structure, can be examined to see what they reveal of the
respondent's understanding of the task as presented. For example, consider the situation
in which the sortings generated are all based on objective, physical aspects of the
elements, such as their weight and size, yet the investigator was apparently looking for
emotional significance of the elements. This suggests the procedure was not tapping what
the investigator thought it was. Careful scrutiny of the details of the procedure and its
context may reveal the basis of this lack of apparent validity. For instance, perhaps the
investigator introduced himself as a designer and thus set up expectations as to the type
of sorting that would be appropriate.
The consequences of subtle changes in the instructions or context of a multiple sort for
the results produced are directly amenable to empirical study. Here the vast literature on
interviewing procedure and threats to its validity would have many points of relevance.
The consequences of interviewer sex and experience are obvious examples from this
literature, but there are many others mentioned throughout the present volume.
Modifications and developments of the sorting procedure
The following are possible ways in which the sorting procedure can be developed and
modified in order to answer some research questions more directly:
1. The creation of element sets to sort can be taken a number of steps. The selection
of a carefully matched set of elements, possibly factorially designed is one step.
But by adding other descriptors, experimentally varied sets of elements can be
produced.
2. Sorting procedures in which one set of elements is sorted into another set offer a
number of prospects for exploring the relationship between different conceptual
domains. If both sets of elements are sorted independently as well as together,
there is the possibility of a very close analysis of the fit between domains.
3. Asking people to sort elements into provided category schemes as well as free
sorts generates links to studies using other methods of concept exploration.
4. Ranking and rating of sorts against different criteria such as importance provides
the opportunity for the development of classifications of sorts themselves -
higher-order sorting.
Group sorting procedures have been used from time to time and can add greatly
to the cost effectiveness of data collection. However, these procedures are likely
to be limited to groups that are quite sophisticated or to very simple aspects of

30
sorting, such as paired comparisons.
Values and contraindications
As mentioned throughout this chapter, the sorting procedure has the flexibility to be
applied to answer a wide variety of research questions, but there are some types of
questions for which it may be inappropriate. Only further use will help to clarify the
boundaries of advantageous and disadvantageous uses, but a few pointers can be given
now.
The multiple sorting procedure clearly has strength when the elaboration of the meaning
of a concept is central to the research question. Studies of how people use the word
'home', or whether 'post-modern' architecture has an identifiable public recognition, or the
conditions under which people will describe their actions as 'panic', all lend themselves to
exploration using a sorting procedure. If these concepts have a common but highly
ambiguous currency, then the demands of the sorting procedure may well help to
disentangle the different meanings. Also, when conceptual systems are being explored
with groups whose verbal fluency may be restricted such as children or psychiatric
patients, then a sorting procedure may be especially useful.
There are two types of research questions, however, for which multiple sorting may be
less appropriate; they -fall into two general classes. The first are those questions in which
a very personal, idiosyncratic perspective is what is being sought. The indepth
psychotherapeutic interview can never be replaced by sorting tasks. The second class of
research questions are those concerned with a direct understanding of processes,
especially sequences of action. For example, studies of how people make decisions under
stress or cope with bereavement are less likely to prove fruitful if built around a study of
personal category schemes than if they focus directly on the stages through which people
go and what moves them from one stage to the next.
Conclusions
This chapter has presented a detailed account of multiple sorting procedures, with
respect to both their theoretical origins and their numerous applications to open-
ended interview situations. As the first section of the chapter has demonstrated,
the multiple sorting procedure has roots in both the early clinical object sorting
techniques and the paired comparison procedures advocated recently by
multidimensional scaling enthusiasts. But more important, the multiple sorting
procedure derives from two parallel concerns in psychology: the significance of
the respondents' own view of the world, and the recognition that world view is
built around a pattern of categorisations. In this respect, the multiple sorting
procedure reveals theoretical links to Kelly’s work in the development of the
repertory grid and to other more recent research in social and clinical psychology.

31
With respect to its applications to the interview process, the second portion of this
chapter has provided examples of its adaptability, and ease of administration. However, it
is clear that one of its primary virtues may also be a burden to the researcher: it probably
makes even greater demands than the repertory grid on the intellectual stamina of the
investigator, forcing her or him to clarify exactly what it is that he/she is looking for and
why. In this respect it serves as an appropriate complement to other forms of the interview
procedure, such as the use of ordinary explanations. The multiple sorting task thus takes
its place amongst the family of interviewing procedures, but only future developments
and use will establish the role it is to play.
References
Adams-Weber, I.R. (1979), Personal Construct Theory: Concepts and Applications, Wiley,
Chichester.
Allport, G.W. (1937), Personality: A Psychological Interpretation, Holt
Saunders, New York.
Argyle, M., Furnham, A. & Graham, I.A. (1981), Social Situations, Cambridge
University Press, London.
Bieri, I. (1971), Cognitive Structures in Personality, in H.M. Schroder and P.
Suedfeld (eds), Personality Theory and Information Processing, Ronald Press,
New York.
Bishop, R. (1983), The Perception and Importance of Time in Architecture, Ph.D
thesis, University of Surrey.
Block, I. (1961), The Q-sort Method in Personality Assessment and Psychiatric
Research, Charles C. Thomas, Springfield, Illinois.
Bonarius, H., Holland, R. and Rosenberg, S. (eds) (1981), Personal Construct
Theory: Recent Advances in Theory and Practice, Macmillan, London.
Brown, I. & Sime, I. (1980), 'A methodology for accounts', in M. Brenner (ed.),
Social Methods and Social Life, Academic Press, London, pp. 157-188.
Bruner, I., Goodnow, I. & Austin, G. (1956), A Study of Thinking, Wiley, New
York.
Canter, D. (1977), The Psychology of Place, London Architectural Press.
Canter, D. (1983), The Potential of Facet Theory for Applied Social Psychology, Quality and Quantity,
Elsevier Scientific Publishing Co, Amsterdam.

32
Canter, D., Ambrose, I., Brown, I., Comber, M. & Hirsch, A. (1980), Prison
Design and Use, Final Report to the Home office, University of Surrey.
Canter, D., Brown, I. & Richardson, H. (1976), Constructs Without Tears: Is
There Life Beyond the Grid, paper presented to the British Psychological Society
Annual Conference, Exeter.
Douglas, M. (ed.) (1977), Rules and Meanings, Penguin, Harmondsworth.
Eckman, P. (1975), 'Unmasking the Face', Prentice Hall, New Jersey.
Fransella, F. & Bannister, D. (1977), A Manual for Repertory Grid Technique,
Academic Press, London.
Frost, A. & Canter, D. (1982), 'Consumer psychology', in S. Canter and D.
Canter (eds.), Psychology in Practice: Perspectives on Professional Psychology,
Wiley, Chichester.
Garling, T. (1976), 'The Structural Analysis of Environmental Perception and
Cognition, Environment and Behavior, Vol. 8(3), pp. 385-415.
Grainger, B. (1980), A Study of Concepts in the Building Design Process, Msc.
Dissertation, University of Surrey.
Groat, L. & Canter, D.V. (1979), 'Does Post-modernism Communicate',
Progressive Architecture, December, pp. 84-87.
Groat, L. (1982), 'Meaning in Post-modern Architecture: An Examination Using
the Multiple Sorting Task', Journal of Environmental Psychology, Vol. 2(3), pp.
3-22.
Hane, R. & Secord, P. (1972), The Explanation of Social Behaviour, Blackwell,
Oxford.
Hawkins, C. (1983), Differing Conceptualisations of Users of Three Psychiatric Day Centres, MSc
Dissertation, University of Surrey.
Horayangkura, V. (1978), 'Semantic differential structures', Environment and
Behaviour, Vol. 10(4), pp. 555-584.
James, W. (1890), Principles of Psychology, Holt Saunders, New York.
Kelly, G. (1955), The Psychology of Personal Constructs, Norton, New York.
Krampen, M. (1979), Meaning in the Urban Environment, Pion, London.

33
Lingoes, I. (1973), The Cuttman-Lingoes Nonmetric Program Series, Ann Arbor,
Mathesis Press, Michigan.
Miller, G. A. (1956), 'The Magical Number Seven, Plus or Minus Two,
Psychological Review, Vol. 43, pp. 81-114
Nasar, J. (1980), 'On Determining Dimensions of Environmental Perception',
Edra, Vol. 11, pp. 245-256, Environmental Design Research Association,
Washington.
Oakley, R. (1980), Profiles and Perspectives of Hostel Residents,
Msc. Dissertation, University of Surrey.
Oostendorp, A. & Berlyne, D.E. (1978), 'Dimensions in the Perception of
Architecture', Scandinavian Journal of Psychology, Vol. 19, pp. 145-150.
Osgood, G.E. (1962), 'Studies in the Generality of Affective Meaning Systems',
American Psychologist, Vol. 17, pp. 10-28.
Osgood, G.E., Suci, G.I. & Tanenbaum, P.M. (1957), The Measurement of
Meaning, University of Illinois Press, Urbana.
Palmer, I. (1978), 'Citizen Assessment of the Coastal Visual Resource', in
Coastal Zone 78, New York American Society of Civil Engineers.
Pitt, D.G. & Zube, E.H. (1979), 'The Q-sort Method: Use in Landscape
Assessment Research and Landscape Planning’, in G. H. Eisner (ed.), Our
National Landscape, Pacific Southwest Forest and Range Experiment Station,
Berkeley.
Rosch, E. (1977), 'Human Categorization', in N. Warren (ed.), Advances in
Cross-cultural Psychology (Vol I), Academic Press, London.
Rosenberg, S. & Kim, M. P. (1975), 'The Method of Sorting a Data Gathering
Procedure in Multivariate Research', Multivariate Behavioral Research, October,
pp. 489-502.
Schiffman, S.S., Reynolds, L.M. & Young, F.W. (1981), Introduction to
Multidimensional Scaling, Academic Press, London.
Shaw, M.L.G. (1982), 'The Extraction of Personal Meaning from a Repertory
Grid', in E. Shepherd and I.E. Watson (eds), Personal Meanings,Wiley,
Chichester.
Shepherd, E. & Watson, I.P. (1982) (eds), Personal Meanings, Wiley, Chichester.

34
Sherif, M. & Sherif, G. (1969), Social Psychology, Harper and Row/John
Weatherhill, New York/Tokyo.
Shye, S. (ed.) (1978), Theory Construction and Data Analysis in the Behavioral Sciences, Jossey-
Bass, San Fransisco.
Smith, A.E. & Medin, D.L. (1981), Categories and Concepts, Harvard Press,
London.
Stephenson, W. (1953), The Study of Behavior: Q-technique and its
Methodology, University of Chicago Press, Chicago.
Streufert, S. & Streufert, S. (1978), Behavior in the Complex Environment, Holt
Saunders, New York.
Szalay, L.B. & Deese, I. (1978), Subjective Meaning in Culture: An Assessment Through Word
Association, Hillside, NJ: Erlbaum.
Tajfel, H.G. (1978), The Structure of our Views about Society in H. Tajfel and G.
Fraser (eds), An Introduction to Social Psychology, pp. 302-321, Penguin,
Harmondsworth.
Tajfel, H. (1981), Human Groups and Social Categories, Cambridge University
Press, Cambridge.
Thurstone, L.L. & Chave, E.J. (1929), The Measurement of Attitudes, University
of Chicago Press, Chicago.
Vygotsky, L. (1934), Thought and Language, MA: MIT Press, Boston.
Ward, L.M. (1977), 'Multidimensional Scaling of the Molar Physical
Environment' ,The Journal of Multivariate Behavioral Research, Vol. 12, pp. 23-
42.
Ward, L.M. & Russell, I.A. (1981), 'Cognitive Set and the Perception of Place',
Environment and Behaviour, Vol. 13(5), pp. 610-632.
Zvulun, E. (1978), 'Multidimensional Scalogram Analysis: The Method and its
Application' in S. Shye (ed.), Theory construction and Data Analysis in the
Behavioral Sciences, pp. 237-264, Jossey-Bass, San Fransisco.