Scaling and Dimensional Analysis
- William Jacoby, Michigan State University
This workshop will focus on several strategies for producing geometric representations of structure in data. These methods tend to be used for three main reasons:
(1) Data reduction. Typically, multiple indicators are combined to improve the quality of measurement. For example, a researcher may want to combine individual responses across a set of survey questions which ask about a common topic. In this case, the objective may be to obtain more fine-grained resolution of respondents' attitudes on the topic than can be obtained from any single survey item.
(2) Evaluating dimensionality. How many distinct sources of variability underlie a set of empirical indicators? For example, a researcher may want to determine the evaluative criteria that respondents bring to bear on a given stimulus object. The objective may be to recover the "mental maps" that give rise to individual beliefs and attitudes.
(3) Measurement. Scaling methods are often used to assign numerical scores to aspects of empirical objects (which may be qualitative in nature). Here, the objective often is to obtain reliable interval-level variables that can be employed in subsequent statistical analyses.
Specific scaling methods to be covered in the course include summated rating scales, unfolding models, principal components analysis, factor analysis, multidimensional scaling, and correspondence analysis. In-class examples will rely primarily on the R statistical computing environment and Stata software. However, the scaling techniques covered in this course also are available in all of the other major statistical packages (e.g., SPSS and SAS). No prior exposure to, or experience with, scaling methods is necessary; however, course participants should be familiar (and comfortable) with multiple regression analysis.
Fees: Consult the fee structure.
Location: ICPSR -- Ann Arbor, MI
Date(s): July 18 - August 12
Time: 1:00 PM - 3:00 PM