Seven types of ambiguity

Theory and Society

Volumn 26, Issue 2-3, 1997, Pages 357-399

  Abbott, Andrew ^a  

^a UNIVERSITY OF CHICAGO   (United States)

Author keywords

[No Author keywords available]

Indexed keywords

EID: 26844525970     PISSN: 03042421     EISSN: None     Source Type: Journal    
DOI: 10.1023/a:1006870409021     Document Type: Article

References (8)

1. R. Barthes, S/Z (New York: Hill and Wang, 1974).
2. W. Empson,Seven Types of Ambiguity (New York: Meridian, 1957). My title is chosen in homage to Empson, whose book I very much enjoyed when I read it in the 1960s.
3. It should be noted that categorical measure is by some writers not considered measurement at all. See O. D. Duncan, Notes on Social Measurement (New York: Russell Sage, 1984).
4. The best-known expositor of this argument is David Freedman; see D. A. Freedman, "As Others See Us," Journal of Educational Statistics 12 (1987): 101-128. The Freedman critique is actually stronger than the critique mentioned here, for he has pointed out that indicator-based estimates of the effect parameters (the causes at the conceptual level) are in fact conditional on the pattern of causes expected and thus cannot even disconfirm it.
5. I have written about this semantic relationship extensively elsewhere, e.g., A. Abbott, "What Do Cases Do?" C. Ragin and H. Becker, editors,What is a Case? (Cambridge: Cambridge University Press, 1990), 53-82.
6. I have used machine-readable ASCII copies of the Index to the GSS bibliography, kindly provided by Tom W. Smith, Director of the GSS. For a general discussion of the GSS, see J. A. Davis and T. W. Smith. General Social Surveys, 1972-1984: Cumulative Codebook. (Chicago: NORC, 1994). There are 2,982 scholarly pieces in the bibliography, of which 181 use the RELITEN variable. I omit 26 items that are sociology or statistics texts or simply general reviews of trends in American society. In the 155 remaining, there are about 100 articles and chapters. (The rest are dissertations, theses, books, and conference papers.) For the detailed readings to be used here, I have thus read about one-third of this literature in detail. Note that my approach presupposes that people answered the surveyors "fairly and honestly." That much of my analysis is "positivist." However, a positivist would regard the ambiguities in their responses as "errors," on the basis of which to construct probabilistic arguments and inferences. But I wish to treat them as links that help bind this one variable into a larger and complex network of meaning. Finally, all programming (and programming errors) and analysis are my responsibility.
7. I chose to analyze 100 variables because 100 is a practical number for investigation. The exact number is arbitrary, but one can't analyze too few variables without possibly losing the connections between this literature and diverse others, and one can't analyze too many variables without coming rapidly to analyze the whole data set. Thus, 100 seemed a practical medium.
8. Hierarchical cluster analysis is a means of placing objects into hierarchically organized groupings - represented by a tree or dendrogram - using as input the triangular matrix of distances between each pair of cases. There are many algorithms for accomplishing this grouping, which often produce surprisingly different answers. I have done all clustering in this article using two such algorithms - single and average linkage - and have listed clusters that are strongly consistent across the algorithms. The two algorithms have quite different properties and emphasize different kinds of similarities in the data. Where they give similar answers, the clusterings can be taken as reasonably strong. "Cluster diameter" refers to the closeness of objects within a cluster relative to objects outside. Small cluster diameters indicate strong clusters. All of the cluster analyses given here have been interpreted with the aid of multidimensional scaling. The scalings are omitted from the article to save space, but inform the discussion at some points.