Triad Test

sf_triad image.png
Which one does not fit? Which two are most alike?























Definition


A type of structured method used often in cognitive anthropology, in which an informant is shown three stimuli at a time from the same cultural domain. The researcher records the answer to the questions: Which two fit together? Which one does not fit?

The choices the informant makes (and the reasons for those choices) can be compared across people or groups to identify similarities and differences in how people think. In other words, by studying which items are considered most similar, a researcher is able to identify salient components of the way people categorize and relate things (Bernard 2011:230-1).


Relevant Characteristics

This is a potentially powerful method that can:
  • Identify cognitions about a cultural domain
  • Identify individual cognitions
  • Identify systems of thinking


An Illustration of Triad Tests:

    • Researcher: Which two of these are most alike? Why?

      • 1. St. Germaine | 2. St. Theophilus | 3. St. Ides

    • Person A: 1 and 2. St. Germaine and St. Theophilus are attorneys in Catholic history
    • Person B: 1 and 2. They are Catholic saints
    • Person C: 1 and 2. They streets in town, but the ‘Street’ abbreviation is written in front for some reason.
    • Person D: 1 and 3. St. Germaine and St. Ides are both kinds of alcohol.
    • Person E: 1 and 2. We learned about them in Sunday school

The responses show how powerfully the triad tests illustrate respondent cognitions.
Deductions from informant responses:
  • Identify cognitions about a cultural domain: We see that the religious cultural domain (Catholicism in this sample) is central, and the people are well educated in finer details of its history.
  • Identify individual cognitions: Person C seems to be more familiar with the geographical layout of the town. He thought of this first, before he thought of Catholic history. Person D is more preoccupied with alcohol, as this was his initial thought.
  • Identify systems of thinking: We can deduce that this is a very Catholic cultural group, and this mentality may influence other factors (perhaps even the way the streets are named).

Problems in informant responses:
  • We don't know what truly represents the culture…alcohol, Catholicism or streets. Do we assume the responses are mutually exclusive?
  • IMPORTANT: Thus, when conducting triad tests, contain your stimuli to a singular cultural domain, to every extent possible.


A Real Example of Triad Tests
    • Mother | Sister | Daughter
    • Mother | Sister | Wife
    • Daughter | Wife | Niece

  • Looking at the top line, some might say that the daughter and sister are most alike because they seem to be closer in age. Others might see mother and daughter as more alike because they have a closer relationship.
  • Likewise, in the second line, one might think that mother and wife are closest because sister is in a different generation. Or, they might see sister and mother as being blood relatives, and therefore closer.

Triad tests identify which of these types of determinations are made in a particular sample. Determining why informants chose as they do often requires further ethnographic data, such as interviews.

Method Made Easy


1Of_These_Things.jpg

Triad Test Designs:
All possible combinations of 3 are used (thus, you must have a relatively small set of stimuli). Due to subject fatigue and resource limitations, it is necessary to limit the number of triads an informant is presented with. In order to do this, the researcher must either use a small set of stimuli, or adjust the design to limit the triads. This adjustment can be achieved by:

  1. Balanced incomplete block design (BIB): limits the number of times pairs of stimuli appear.
    • It helps control the number of combinations the informant is presented with
  2. Lamda-1 design: allows pairs of stimuli to appear one time throughout the test
  3. Lamda-2 design: allows pairs of stimuli to appear two times throughout the test

TIP: Randomly generate triad lists to maintain data integrity and control for the phenomenon of order effects (where the responses are biased by the order the stimuli appear in).


Step 1: Similarity Matrix- contains measurements of relations between items
  1. List each stimulus in the triad test on a horizontal and vertical axis
  2. Tally each time a pair is chosen as similar on the matrix. The greater the number in the cell, the more similar the two stimuli


Step 2: Multidimensional Scaling Analysis (MDS):
  1. Overview
    • Creates map of how variables relate
    • Pearson’s R sometimes used to determine the linear correlation between the X and Y variables (giving a value of -1 to 1 inclusive)
    • The number of dimensions is determined by n-1. However, this is not realistic to interpret once you reach the 3rd dimension, so 2-dimensional diagrams are used
    • Stress is a measure of how far the graph is from one that is theoretically perfectly proportional. The lower the stress, the better the solution.
    • When combining informant data to achieve a cultural consensus, determine the total number of times a pair is chosen as similar, and divide by the number of informants. This gives the percent of times the informant marked the items as similar and generates an aggregate proximity matrix. When MDS is run on these data, the diagram generated represents the mental map of the group, and illustrates a cultural consensus.
  2. Use Anthropac software to execute diagram


Illustrative Example:

Step 1: A DIS-Similarity Matrix: Distances between Nine U.S. Cities (in Miles)
sf_sim matrix.jpg

Step 2: A two-dimensional MDS diagram of the matrix:
sf_disorientation.jpg
sf_MDS us map.png
The result? A map of the US, upside-down and backwards















MDS: creates a relative map with no regard to the absolute spatial position of the entity. This makes sense when you remember that informant perception does not have an absolute spatial positioning.

Advantages


sf_Pros-of-File-Sharing.jpg
Aaaaeeeyyy!

  • • Helps establish shared views and ways of categorizing within a population
  • • Provides quantitative data from qualitative responses
  • • Inexpensive



Limitations


bad_edited.png

• Provides an examination of beliefs, but not behavior
• Requires a statistically significant sample size
• Makes assumption that informants share a common culture (that differences in answers are reflective of differences in individual cognition, rather than the presence of subculture)
• Often must be accompanied by additional qualitative methods to determine why informants make the choices the make






Analysis


  1. Compare results across the patterns of agreement/ disagreement, why they are chosen, and how similar/different they are perceived to be across stimuli.
  2. Multivariate Analysis (Causal Modeling): can be used to determine how the variables are related when there are many stimuli and the matrix becomes overwhelming.
    • MDS: create a map of the stimuli, with the distance between the stimuli representing how close/far apart the stimuli are in the cognitive landscape. See Methods Made Easy’ for an example of detailed analysis
    • Cluster Analysis: another descriptive tool for understanding how things group together. Clustering illustrated which things are bunched together and in what order.

Method in Context



sf_the-office.jpg
Boster et al (1987) used triad tests to determine perceptions of an office social/power network.

Researchers asked informants to identify which employee in the Office of Academic Affairs was most different from the other two. Student employees (both undergrad and grad) were included as well as professional staff and support staff, as was the Vice-Chancellor himself. The number of times a pair was chosen as most similar was summed up across all the informants. Additional structured methods such as rankings and pile-sorts were also utilized. The results of the triad test showed that the informants seemed to base their response on the relative employment hierarchical ranks or social status.

The MDS diagram showed all student employees grouped in a band in one section, and the professional and staff making another group. The distinction between undergraduates (M, I, N, J) and graduate students (H, D) is far less clear, as is the distinction between professional (A, O, G, P) and support staff (B, F, E, C,K) (Boster et al 1987).
sf_Office MDS.png

Online Resources



For an additional overview of cognitive anthropology please visit:
http://anthropology.ua.edu/cultures/cultures.php?culture=Cognitive%20Anthropology


Further Reading and References



sf_references.jpeg
Boster, J. S., J. C. Johnson, and S. C.Weller
1987. Social position and shared knowledge: Actors’perceptions of status, role, and social structure. Social Networks 9:375–87.

Bernard, Russel H
2011. Research Methods in Anthropology: Qualitative and Quantitative Approaches, Fourth Edition. AltaMira Press

Borgatti, S. P
1997 Consensus Analysis. <http://www.analytictech.com/borgatti/consensu.htm>

Dressler, William Mauro C. Balieiro, Rosane P. Ribeiro, Jose E. dos Santos.
2007. A prospective study of cultural consonance and depressive symptoms in urban Brazil, Social Science & Medicine, 65:10 2058-2069

Libertino, LD MML Osornio and G Hourgh.
2012. Analysis of data from a free-listing study of menus by different income-level populations. Food Qualiy and Preference

Quinlan, M.
2005 Considerations for Collecting Freelists in the Field: Examples from Ethobotany. Field Methods 17(3):219-234.

Weller, S. C.
2007 Cultural Consensus Theory: Applications and Frequently Asked Questions. Field Methods 19(4):339-368.