TESTING FOR UNIDIMENSIONALITY WITH FACTOR ANALYSIS

Factor analysis is a technique for data reduction. If you have 30 items in a pool of potential scale items, and responses from a sample of people to those pool items, factor analysis lets you reduce the 30 items to a smaller set—say, 5 or 6. Each item is given a score, called its factor loading. This tells you how much each item ‘‘belongs’’ to each of the underlying factors. (See chapter 22 for a brief introduction to factor analysis and Comrey [1992] for more coverage.)

If a scale is unidimensional, there will be a single, overwhelming factor that underlies all the variables (items) and all the items will ‘‘load high’’ on that single factor. If a scale is multidimensional, then there will be a series of factors that underlie sets of variables. Scale developers get a large pool of potential scale items (at least 40) and ask a lot of people (at least 200) to respond to the items. Then they run the factor analysis and select those items that load high on the factor or factors (the underlying concept or concepts) they are trying to understand (box 11.2).

BOX 11.2

HOW SCALES ARE DEVELOPED

If you want to see what professional scale developers do, consult any of the following: Klonoff and Landrine (2000) (a scale for measuring acculturation among African Americans), Staats et al. (1996) (a scale measuring commitment to pets), Sin and Yau (2004) (a scale for measuring female role orientation in China), and Simpson and Gangstad (1991) (a scale that measures willingness to engage in uncommitted sexual relations).

You may not develop major scales for others to use but what you should do is test the unidimensionality of any composite measure you develop for your own field data, using factor analysis—once you understand the principles of scale development that I've laid out here. Examples of scales developed by anthropologists include one to measure depression in old people (Gatz and Hur- wicz 1990) in the United States and one to measure domestic cooperation in Barbados (Handwerker 1996a). I'll walkyou through Handwerker's scale in chapter 22 when we get to factor analysis.

 
Source
< Prev   CONTENTS   Source   Next >