Cultural Anthropology Methods

Volume 5 , Issue 3 October 1993

Welcome to CAM Using A N T H O P A C 3.5 and a

CAM is where researchers in the social
sciences discuss how t o use and teach
Spreadsheet to Compute a
qualitative and q u a n t i t a t i v e research Free'List Salience Index 1 :
methods.
CAM costs $ 15 U S C per year, including ]. Jerome Smith
postage, for three issues sent to libraries or to Department of Anthropology,
individuals anywhere in the United States University of South Florida
(including Puerto Rico, the U.S. Virgin
Islands, and Guam). For Canada and Mexico,
Introduction The problem with expressing a mean rank of
add $2.00 postage (all orders to Canada and
mention in free lists is that free listing is an
Mexico are sent airmail). For all other
Free listing is a valuable method for exploring open-ended task. Not all items will show up
countries, add $4.50 for surface mail, $7.50
the content of cultural domains. Informants in all lists. Assigning a value to "no mention"
for air mail. Purchase orders are accepted.
are asked questions like "What are all the (0, Iist-length+1, etc.) for purposes of
Florida residents, please add 6% (90 cents)
kinds of (steps in, parts of, etc.) X?" Their establishing a mean score across all lists is an
sales tax. Send your check to ECS-CAM,
responses create lists that can then be arbitrary solution of little value. Treating
2815 N W 38th Dr., Gainesville, FL 32605.
examined for various patterns of cultural "no mention" as missing data, essentially
Submitting articles: Send all copy to knowledge. In my own ongoing research the net mean procedure noted above, can
E C S - C A M , 2815 N W 3 8 t h Drive, into English color terminology, for instance, easily yield distorted results, as when an
Gainesville, FL 32605. Send hard copy and I have been asking groups of college students item is mentioned in only one list but as the
disk. (We prefer MS-DOS, if it's convenient to list "all the colors you can think of in two third item in that list (thus, mean rank=3.
for you.) For more information, or for minutes." See Bolton, Curtis, and Thomas 1980:316,
enquiries about submitting an article, contact Table 3, Figure 1, below). This may account
any of the editors of CAM: for the relatively infrequent use of net mean
At least two characteristics of free lists are
worthy of careful consideration: (1) the ranks in free-list salience research.
frequency of mention of items across the
lists and (2) the order of mention of items in Rather than abandon the useful notion that
Editors lists. T h e frequency measure is easily order of mention is a measure of salience,
H. Russell Bernard (Florida) computed and often relied on as the single the two measures of frequency and order can
[email protected] measure of the salience of items in a free list be combined into a single index of free-list
Fax:904-376-8617 (Weller a n d Romney 1988:9-20).
Perrri Pelro (Connecticut) salience for a given item by calculating a
PJPELTO@UCONNVM A N T H R O P A C , a package authored by gross mean percentile rank for that item
Stephen Borgatti (South Carolina) Stephen Borgatti and familiar to many across all lists:
N040016@UN1VSCVM readers of CAM (Borgatti 1989, 1991),
Oswald Werner (Northwestern)
Fax:708-467-1778
provides a n easy-to-use procedure for free-list sum of the item's percentile ranks
creating and cleaning up free-list data sets salience =
Associate Editors and for computing item frequencies. The of an item total number of lists
Thomas Schweizer (Cologne) most frequently mentioned items can then
[email protected] be used in other procedures for exploring a
Hartmut Lang (Cologne) Items mentioned early and often would have
[email protected] cultural domain. a relatively high mean percentile rank. An
Michael Huberman (Harvard) item mentioned first - and first in every list
HUBERMM@HUGSE1
Joseph Maxwell (Harvard) Information about the order of mention of - would have an index of 100.1 terns occurring
MAXWELJO@HUGSEl items in free lists is also readily available, butlater in lists and/or not at all in some lists
Matthew Miles (Ctr. for Policy Research) not so conveniently calibrated for the lists as would have indexes that declined toward 0
MATOD@CUVMB
Barry Wellman (Toronto)
a set. The usual approach is to calculate the (the extreme case being a phantom item
[email protected] mean rank of items across the lists in which that was never mentioned in any list at all).
they occur (Romney and D'Andrade 1964;
H Managing Editor
B o l t o n , C u r t i s , and T h o m a s 1 9 8 0 ) . Use of percentile measures is especially
Carole M. Bernard (ECS) Interestingly, Weller and Romney (1988) helpful in rendering both the order and
CAM is a producr of Edirorial Consulting Services of do not mention this tabulation, nor does frequency of mention comparable across
Gainesville, FL. CAM is issued three times a year, in A N T H R O P A C 3 . 5 p r o v i d e for its
February, June, and October. ISSN 1064-8631.
calculation. 2 Continued on page 2
Continued from page I
Table I: Co/or term data in a spreadsheet format for calculating free-list salience indices.
open-ended lists of unequal length and Complete data for computations: informants = 51, items = 139, longest list = 31
composition. I believe this index nicely
captures what we generally have in mind Lookup Table Index Marrix Measures
when we speak of an item's salience in a
cultural domain: T h e higher the index, the Fred Joan Bill Sequence Irems Fred Joan Bill Index Freq Rank
RED BLUE RED 0 RED 100 95 100 85 49 2.94
earlier and more frequently the item has
YELLOW RED BLUE 1 BLUE 89 100 93 80 47 2.96
been mentioned - on average - in free lists BLUE YELLOW BLACK 2 GREEN 26 85 80 74 47 4.14
generated by informants. BLACK GREEN GREEN 3 YELLOW 95 90 67 71 46 4.24
WHITE BROWN ORANGE 4 BLACK 81
84 75 87 58 45 6.18
TAUPE BLACK YELLOW 5 ORANGE 32 30 73 57 45 6.39
A N T H R O P A C ' s Freelist Procedure MAUVE WHITE PINK 6 WHITE 79 70 00 53 42 6.00
MAROON TURQUOISE PURPLE 7 PURPLE 00 50 53 45 40 6.63
PINK SCARLET NAVY 8 BROWN 00 80 27 33 37 7.73
Salience indexes for free-list data can be GREY MAROON MAROON 9 PINK 58 35 60 32 39 8.57
computed in a spreadsheet program using SKY BLUE PURPLE VIOLET 10 VIOLET 00 45 33 32 29 4.78
files created by A N T H R O P A C ' s FREELIST AQUA VIOLET BROWN 11 GREY 53 05 20 27 32 7.37
procedure. The first task is the construction AQUAMARINE NAVY GREY 12 TURQUOISE 00 65 13 14 16 3.35
of an ASCII file for use in A N T H R O P A C ' s ORANGE PINK TURQUOISE 13 MAGENTA 00 00 00 13 13 2.55
GREEN ORANGE LAVENDER 1114 TAN 11 00 00 11 22 6.47
FREELIST procedure. T h e file consists ofa
NAVY-BLUE AQUA 15 MAROON 63 55 40 11 16 4.18
single column of lists - one informant after CHARTREUSE FUCHSIA 16 INDIGO 00 00 00 11 9 1.31
another. Although not necessary to the TAN MAUVE 17 BEIGE 00 00 00 10 15 3.84
FREELIST procedure, the order of mention OFF-WHITE BLUE-GREEN 18 FUCHSIA 00 20 00 10 15 3.98
of items must be retained in the data lists for GREY 19 TEAL 00 00 00 10 15 4.08
further computation of a free-list salience
index that takes order of mention into
31
account.
139 irems
FREELIST has a subprocedure for clean-
ing up the original data file so that all Counr19 20 15 Distance 51 50 49
items repeated across informant lists are
spelled the same. T h e o u t p u t file is
CLEANED.DAT. FREELIST also produces
zero so that for any item in any list the Consider, for instance, the value in the
FREQ.OUT, which contains a single list of
sequence numbers state how many items index matrix cell for Fred's mention of
all items with their frequencies, sorted in
were mentioned before that item in that list. GREEN. In this spreadsheet format, the
descending frequency of mention across
general formula required for a percentile
informant lists.
For a given item in a given list, the d ifference ranking here is:
between its sequence number and its count
A Spreadsheet Format can be divided by its count and multiplied count - sequence
by 100 to express the item's percentile - x 100
ANTHROPAC's FREELIST ASCI 1 output ranking in that list. T h e average of the count
files are easily brought into a spreadsheet, item's p e r c e n t i l e rankings across all
where their further manipulation is relatively informants' lists is its gross mean percentile
ranking - its free-list salience index. The where the sequence number is returned by
straightforward. The spreadsheet file will
required calculations are carried out and the spreadsheet's lookup function. (This
have three parts (Figure 1): (1) a lookup
reported for each item in the cells of the formula may need to be embedded in an if-
table, (2) an index matrix, and (3) measures
index matrix by means of the spreadsheet's then error test in order to return a zero in
of salience. I use data from from my work on
color terms (see Table 1) to illustrate how lookup function. case the item in question is not in the
the salience indexes can be computed. informant's list at all.)
The index matrix is built from the imported
The lookup table is essentially FREELlST's F R E Q . O U T file, also p r o d u c e d by The general form of a lookup function is
CLEANED.DAT output file, broken out A N T H R O P A C ' s FREELIST procedure. LOOKUP(x,y,z), where x is the item in
into separate columns for each informant's Matrix rows are all the items mentioned in question (GREEN), y is the range in the
list, with columns labeled accordingly, and the lists. Matrix columns are inserted and lookup table from Fred's column through
the order of mention of items in each list labeled for each informant - in the same the sequence column, and z is the distance
proceeding downward row by row. (Names left-to-right order as they appear in the from Fred's column to the sequence column
have been substituted for ID codes here to lookup table - keep ing the ex isting frequency (taken from the distance row below the
make discussion easy.) A n t i c i p a t i n g column off to the right. A distance row of index matrix). Basically, the lookup function
subsequent spreadsheet computations, there descending numbers beneath the matrix searches in Fred's lookup table column for
is a "size" row at the bottom of the table allows for more efficient use of the GREEN and returns the corresponding
showing the item count for each list. The spreadsheet's lookup function in calculating number from the table's sequence column.
rightmost column of the lookup table is a percentile rankings for each informant's
sequence column denoting the rows' items' mention ofa color terra, that is, each ITEM For Fred's mention of GREEN, then, the
sequential order of mention, beginning at x INFORMANT cell. formula becomes

Page 2 CAM October 1993

 19-14 - a simple direct measure of salience good Notes
-x 100 = 26 for initial domain explorations.
1
19 ANTHROPAC 4.0 has been released.
However, the free-list salience index takes ANTHROPAC 3.5 is now available as "freeware,"
as shown in Table 1. both measures into account by recognizing according to Stephen Borgatti, the author ofthe
the open-ended nature of free-list exercises program. The new version of the program is
T h e three rightmost columns of t h e and by combining frequency and sequence available from Analytic Technologies, 306 S.
spreadsheet in Table 1 are the free-list into a single expression ofthe overall notion Walker St., Columbia, SC 29205. The regular
salience index (gross mean percentile of domain item salience: how often and how price is $125, plus shipping, and includes a 200-
ranking), the frequency of mention, and the early culturally competent informants name page reference manual and methods guide.
net mean rank of mention - which for Student price: $39, plus shipping. For more
the items that comprise a domain's most
information on ANTHROPAC, phone 803-771-
GREEN are 74, 47, and 4.14, respectively. important elements. 7643. Ed.
Discussion 2
References The free-list salience index is now much easier
This spreadsheet format may seem a bit to produce in ANTHROPAC 4.0 and higher,
complex, especially for people who don't Bolton, R., A. T. Curtis, and L. L. Thomas 1980. because the DATA>IMPORT>FREELIST
use spreadsheets regularly. However, Nepali Color Terms: Salience on a Listing Task. procedure creates a respondent-by-item output
efficiencies abound. For instance, if attention Journal of the Steward Anthropological Society matrix (called FREEMAT). This matrix contains
J2(l):309-321. the rank order data that have to be calculated by
is paid to the layout and to judicious use of
hand from ANTHROPAC 3.5's FREEL1ST
relative and absolute addressing of cells and
cell ranges, the entire body of the index Borgatti, S. B. 1989. Using ANTHROPAC to output file. Simply prepare FREEMAT for export
matrix can be created by copying a single Investigate a Cultural Domain. Cultural to your spreadsheet by recoding missing values
Anthropology Methods Newsletter 1(2): 11. (1E38 in A N T H R O P A C 3.5) to 0 in
well-defined formula cell into all the other ANTHROPAC 4.0. Then convert the recoded
cells of the matrix. And, once established, FREEMAT matrix to an ASCII file. Bring the
the spreadsheet may easily serve as a template Borgatti, S. B. 1991. ANTHROPAC 3.5.
ASCII file into your spreadsheet and create a
for all subsequent salience indexing of Columbia, SC: Analytic Technologies.
new spreadsheet matrix of percentile rankings
A N T H R O P A C ' s free-list output data sets. from the raw rank-order data. The formula for
Borgatti, S.B. 1992. ANTHROPAC 4.0.
the cells in the new matrix is, generically,
Columbia, SC: Analytic Technologies.
As I said earlier, net mean rank of mention
is not a very useful measure of salience in Romney, A. K. and R. G. DAndrade 1964. ((row.maximum - reference.cell) + 1 /row.maximum)
and of itself, given the distortions that come Cognitive Aspects of English Kin Terms. *100
from ignoring missing items. O n the other American Anthropologist 66(3) Part 2:146-170.
hand, frequency numbers are easy to obtain Borgatti (personal communication) hopes to
through A N T H R O P A C and, except for the Weller, S. C. and A. K. Romney 1988. Systematic include the free-list salience index soon as an
many tied ranks, they deliver - as advertised Data Collection. Newbury Park, CA: Sage. option within ANTHROPAC.

Creating Historical Time Series specifically interested for historical periods

of up to 30 or 40 years from survey data - if

with Survey Data you plan ahead and incorporate important

survey design features. Here is how you can
do it.
(First in a series)
W. Penn Handwerker (1) Plan to estimate parameters for age
cohorts, so stratify your sample by age.
Department of Anthropology, Randomly sample 5-year cohorts, for
California State University, Humboldt example. I have found that a total sample of
around 400 cases can yield acceptably good
Social and cultural explanations often appeal level variables like Gross National Product, estimates for many variables for a time-span
to macro-level historical processes. Tests of i n f a n t m o r t a l i t y , and real i n c o m e . of up to 40 years, including retrospective
these explanations require macro-level Conventional data bases don't include estimates of period age-specific fertility.
historical data. variables like the proportion of women who
complete secondary school, the proportion But samples of as few as 200 total cases may
Historical times series can provide excellent of people who grow up in lower-class homes, suffice, or samples of more than 1000 may be
data for these tests, as well as for exploratory or the number of sexual partners people necessary, depending on the rarity of the
data analyses. Anthropologists rarely use have during adolescence, which may be of event that you want to estimate (e.g., large
time series. O n e reason is that time-series far more theoretical and policy interest. samples will be necessary for reasonable
data available through official sources rarely estimates of infant mortality or the incidence
include variables that most interest us. W h e n you can't find what you want in of sexual abuse).
conventional data bases, it is often possible
Government statistical offices regularly to create your own. You can generate time- (2) Identify your age cohorts clearly. In the
make available time-series data for macro- series data for variables in which you are best of all possible worlds, be able to
Continued on page 7
CAM October 1993 Page 3