This is a method for discovering the latent structure in
an unsorted list of statements or ideas. The investigator
writes each statement on a small index card and requests six
or more informants to sort these cards into groups or clusters,
working on their own. The results of the individual sorts
are then combined and if necessary analysed statistically.
If the informants are representative of the user population
for whom the application is being designed, then the result
will reflect the structure in which the users expect the ideas
or concepts should be presented.
Collect the statements you wish to analyse. Write each statement
on a separate card. If your statements already exist in computer
format, print them out on labels and stick the labels on index
cards. Number the cards uniquely on the back.
If you have pre-defined categories into which you want the
statements sorted, prepare 'place mats' on which your informants
may place the cards. Otherwise ensure there is a large empty
table for the informant to place their piles of cards on.
Recruit informants who will be typical of the user population
for whose benefit you are preparing the analysis. It is useful
to have at least six such informants.
Shuffle the card deck so that all the informants don't get
the same sequence of cards or worse, that each informant gets
the previous informant's sequence. The informants receive
the stack of cards, and then sort them into piles on the table
in front of them, using place mats if desired. You may give
a rough indication of the number of piles of cards you will
expect to see, to give the informants a common understanding
of the expected grain of analysis.
Explain that there may well be an 'unsortable' pile, but
that they should attempt to place as many cards together into
piles as they reasonably can.
Informants may attempt to show the relationship between piles
of cards by spatial proximity on the table: ensure you note
this information down: in the classic analytic technique,
this information may get lost.
At the end, note which cards have been put together by the
respondent by noting the numbers on the back of the cards.
If you have not supplied place mats, invite the users to give
a name to each pile of cards ("how would you describe
the cards in this pile?")
If your clusters are relatively clear and straightforward,
you may simply summarise the cards which are usually placed
in each pile and give an overall name for the cluster. However,
the result of card sorting is not always so straightforward.
The most common method of analysis for complex data from
card sorting is a statistical method called cluster analysis.
There are two main approaches to cluster analysis for this
kind of material: linkage and hierarchical. See [...] for
further details. In order to prepare for both types of analysis,
construct a similarity matrix. If there are n cards,
the matrix is n x n symmetric. Most methods
require the upper and lower quadrants to be filled redundantly,
and for the diagonals to be filled with 1.00. In each of the
cells, compute the probability p of the cards denoted
by the two common co-ordinates being together: that is, if
two cards x and y were sorted into the same
pile q times and there are m respondents altogether,
p = q / m
Some cluster analysis methods may require an index of dissimilarity,
so compute p' = 1 - p and put 0.00 in the diagonals.
A worksheet and an MS-DOS program for elementary linkage
analysis may be found at
Computer supported methods for card sorting may be found
The nearest alternative is the affinity
diagram (concept wall) technique. This latter has the
merit that a hierarchical arrangement can be deduced without
recourse to mathematics.
After card sorting you should proceed to
design activities such as creating a (paper) prototype
of the structure of the material you have been investigating.
However, you may also wish to run a focus
group or a brain storming
session to further refine and to expand or enlarge on the
results you have so far obtained.
Lorr, Cluster Analysis for the Social Sciences