Workshops in Methods
Indiana University:

MULTIDIMENSIONAL SCALING

"Rotating through m-space since 1984! "

12:00 Noon-3:00 p.m. Friday, December 7, 2012 Room 120 Woodburn Hall Instructor: William Jacoby Michigan State University Department of Political Science E-mail: jacoby@msu.edu

Would you like to draw pictures of your data, in ways that reveal structures which are not obvious from inspection of the data values, alone? Multidimensional scaling (MDS) tries to accomplish exactly that objective. To be more precise, MDS produces a “map” of stimuli, based upon information about the “proximities” among those stimuli.

Multidimensional scaling methods have many potential applications in empirical research. They can be used to: simplify the contents of large, complex datasets; model similarities among sets of objects; estimate the cognitive structures underlying survey responses; and optimize the measurement characteristics of qualitative observations. MDS can be generalized to show individual differences across distinct data sources (e..g, subsets of survey respondents or data collected at different time points). It also can be adapted to represent respondent preferences among a set of stimuli (so-called “ideal points” models).

This workshop will provide an introduction to multidimensional scaling. Specific topics to be covered include: The basic idea of MDS; types of data that might be input to MDS; the general estimation procedure; interpretation of results; different varieties of MDS; and software options for performing MDS analyses. The workshop is geared toward a general audience. It does not assume any prior experience with MDS or familiarity with advanced statistical methods (i.e., beyond basic regression analysis).

Workshop Outline:

• Click here to obtain a copy of the outline for the material to be covered in the workshop (PDF).

Presentation Slides:

• Click here for presentation slides (PDF).

References:

• Click here for a list of references pertaining to multidimensional scaling.

MDS of Intercity Driving Distances:

• Click here for a Stata DO file that carries out an MDS analysis of the data on driving distances between ten US cities.
• Click here for an ASCII text file containing the driving distances between ten US cities.
• Click here for another text file containing the driving distances between ten US cities. This file contains a header record with variable names.

MDS of Socioeconomic Characteristics of Ten US Cities

• Click here for a Stata DO file that carries out a metric MDS analysis of dissimilarities among the the socioeconomic characteristics of ten US cities.
• Click here for a Stata DO file that carries out a nonmetric MDS analysis of dissimilarities among the the socioeconomic characteristics of ten US cities.
• Click here for an ASCII text file containing data on socioeconomic characteristics of ten US cities.

MDS of 1992 American Presidential Candidates

• Click here for a Stata DO file that carries out a nonmetric MDS analysis of dissimilarities among 1992 American presidential candidates and other political figures.
• Click here for an ASCII text file containing the matrix of perceptual dissimilarities among 1992 American presidential candidates (required for preceding Stata DO file).
• Click here for another ASCII text file containing the matrix of perceptual dissimilarities among 1992 American presidential candidates. This file contains a header record with variable names.

MDS of 2004 American Presidential Candidates

• Click here for a Stata DO file that carries out a nonmetric MDS analysis of 2004 American presidential candidates and other political figures.
• Click here for the Stata data file, "therms.dta", required for the preceding DO file.

Interpreting Multidimensional Scaling Solutions

• Click here for a Stata DO file that relates external criteria to the MDS configuration obtained from the preceding analysis of 2004 presidential candidates.
• Click here for an ASCII text file containing the MDS configuration and the information on external variables related to the 2004 American presidential candidates (required for preceding Stata DO file).
• Click here for a Stata DO file that carries out the metric MDS of the 2004 presidential candidates and then performs a cluster analysis on the resultant scaling configuration (requires data file, "therms.dta").

Multidimensional Scaling in R

• Click here for an ASCII text file containing perceptual dissimilarities among 2004 American presidential candidates and political figures. This file contains a header record with variable names.
• Click here for an R script that carries out a nonmetric MDS of the 2004 perceptual dissimilarities using the R function, "smacofSym", from the "smacof" package. This R session also embeds external criteria into the scaling solution and performs a cluster analysis of the scaling solution.
• Click here for an ASCII text file containing external information about 2004 American presidential candidates and political figures (required for preceding R script). This file contains header record with variable names.

ICPSR Summer Program | William Jacoby Web Site