This article appeared in CAM, the Cultural Anthropology Methods Journal, Vol 6(2) 1994.

The Successive Pile Sort1

James Boster
Department of Anthropology
University of California-Irvine

Introduction

This article is about the successive pile sort and how it elicits informants' judgments of similarities among
items in a cultural domain.

We often investigate what people know about specific cultural domains (e.g., birds, plants, diseases, types
of litigation, etc.). An important starting point for these investigations is finding out how people judge the
similarities and differences among the items in the domain. The two most common methods used to elicit
informants' similarity judgments are the triad test and the free pile sort (Weller and Romney 1988; Bernard
1994).

Triad Tests and Free Pile Sorts

In a triad test, informants are presented with three items at a time and asked to judge which of the three is the
most different from the other two. For example, an informant might be asked to judge which was most
different: a dog, a cat, or a gopher.

The triad test has two key advantages. First, it is simple to explain to informants; they can take the test
themselves without supervision. Second, informants' responses are comparable to one another since the same
information is extracted from each informant.

However, the task can be quite tedious and confusing for informants, because the number of triads goes up
proportional to the cube of the number of items. The formula for calculating the number of triples in a set of
items is ((N*(N-1)*(N-2))/6). Thus, there are 120 possible triples of 10 items and 1140 possible triples of 20
items.2

The free pile sort is a far easier method for eliciting similarity judgments from informants. In this task,
informants are asked to sort a collection of item into piles according to which items they think are most
similar to one another.

Almost anything that can be physically sorted can be used as stimuli: names of wines on cards, stuffed bird
specimens, pictures of fish, . . . The free pile sort has two principal advantages. First, informants find it
relatively easy (even fun) to do. Second, informants can easily judge the similarities among very large
numbers of items. Researchers commonly do pile sorts of 60-80 items. This would be out of the question with
a triad test.

The principal disadvantage of the free pile sort technique is that it is difficult to compare the responses of
different informants because informants may choose to divide the stimuli into wildly different numbers of
initial piles. This is known as the "lumper-splitter" problem; it has been described in depth by Boorman and
Arabie (1972; Arabie and Boorman 1973).
The successive pile sort was designed to overcome the tedium of the triad test while still allowing comparison
among informants' responses (Boster 1986, 1987). In this task, informants begin by sorting the items as they
would in an ordinary free pile sort. They are then asked to merge their piles until all piles are joined. Then
they are asked to split their original piles until all items are separated.

Successive Pile Sorts

To do a successive pile sort you need a collection of stimuli and a set of number cards with the numbers 1
through the total number of items minus one. If there are 25 items, then you need 24 number cards. Each
stimulus item should have associated with it a capital letter of the alphabet as an identifier. If there are more
than 26 items, lower-case letters may be used as identifiers of the next 26 items. I illustrate the method with
an example of 25 bird names on cards as stimuli.

Begin by laying the stimuli in random order in front of the informant and saying something like the following:

Here is a set of 25 cards, each with the name of a different kind of bird [or whatever domain you are
studying]. Please sort them into piles according to which birds you think are most similar to one
another. You can sort them according to whatever characteristics of the birds you like and into as
many piles as you like.

After informants have finished the sort, leave the stimuli as they are and record the number of piles. This
number of initial piles is the number "N" throughout these instructions. Next say:

Now please tell me which of these piles of birds [or whatever] are the most similar to one another?
That is, if you could only make [N-1] piles, which piles would you join together? [For example, if
the informant made seven piles, ask them which of the seven they would join if they had to make
only six piles.]

Take the piles that the informant indicates as being most similar and move them next to each other, if
necessary. Place the number card [N-1] between the pair of piles the informant says that he or she would join.
Now say:

Treating the piles you have just joined as a single pile, now which piles of birds would you join?

Place the number card [N-2] between this pair of piles. Again, move the piles around if necessary. Continue
the process until there are only two piles. Place the number card 1 between them and say:

Returning to your original piles, split the pile that contains the birds that are most different from one
another. That is, if you had to make [N+1] piles, which pile would you split and how?

Informants have almost as much freedom in choosing how to split the piles as they have in the initial sort.
For example, although informants can only split a pile into two, they may choose to split a pile with six items
into piles of one and five, two and four, or three and three. Place the number card [N] between the piles that
have just been split apart and say:

Again, split the pile that contains the birds that are most different from one another. That is, if you
had to make [N+2] piles, which pile would you split?

Place the number card [N+1] between the pile that is split. Continue this process until there is only one pile
remaining to be split and place the number card 24 (25 items - 1) between the remaining two items.

You should now have a sequence of items (stimuli) and number cards that alternates: item, number card, item,
number card, . . . item. Record the informant's successive sorting of the items by writing down the sequence
of item identifiers (letters of the alphabet) and the numbers on the number cards (e.g., B 2 A 4 D 1 F 3 E 5
C).

The principal advantage of the successive pile sort is that it allows study of intracultural variation in responses
to the similarity judgment task. Since the same amount of information is extracted from each informant, one
can reasonably construct representations of the responses of each informant and compare them. The principal
disadvantage is that it is more complicated to administer than either a free pile sort or a triad test and may be
particularly difficult for novice field assistants.

Notes

1. The successive pile sort data can be analyzed with the latest version of ANTHROPAC. Contact Analytic
Technologies, 306 S. Walker St., Columbia, SC 29025. Tel: 803-771-7643.

1. The tedium of the triad test can be substantially reduced by using a balanced incomplete block design
(Burton and Nerlove 1976). This design presents to informants a subset of the complete number of possible
triads such that each pair of items in the domain occurs the same number of times in the triad test. The term
"lambda" refers to the number of times each pair of items occurs in the design: in a lambda-1 design, each
pair occurs once in the set of triads; in a lambda-2 design, each pair occurs twice, and so on. For most
purposes, a lambda-2 design is both the smallest useful design and the largest necessary one. The time
savings can be enormous: a complete design for 25 items has 2300 triads, while a lambda-2 design has only
200 triads.

References

Arabie, P. and S. A. Boorman 1973. Multidimensional Scaling of Measures of Distances Between Partitions.
Journal of Mathematical Psychology 10:148-203.

Bernard, H. R. 1994. Research Methods in Anthropology, 2d edition. Newbury Park, CA: Sage.

Boorman, S. A. and P. Arabie 1972. Structural Measures and the Method of Sorting. In R. Shepard, A. K.
Romney, and S. B. Nerlove, eds., Multidimensional Scaling: Theory and Applications in the Behavioral
Sciences, Vol. I:225-249. New York: Seminar Press.

Boster, J. 1986. Can Individuals Recapitulate the Evolutionary Development of Color Lexicons? Ethnology
25(1):61-74.

Boster, J. 1987. Agreement Between Biological Classification Systems Is Not Dependent on Cultural
Transmission. American Anthropologist 89(4):48-54.

Burton, M. and S. Nerlove 1976. Balanced Designs for Triad Tests: Two Examples from English. Social
Science Research 5:247-267

Weller, S. and A. K. Romney 1988. Systematic Data Collection. Newbury Park, CA: Sage.