Multidimensional Scaling of Sorting Data: A Comparison of Three Procedures

Area of Science:

• Psychology
• Statistics
• Data Analysis

Background:

• Dissimilarity measures (D) from stimulus sortings can be analyzed using Multidimensional Scaling (MDS).
• An alternative involves transforming D to profile distances (Δ), a method previously criticized by Drasgow and Jones (1979).

Purpose of the Study:

• To compare the efficacy of MDS on direct dissimilarity measures (D) versus profile distances (Δ).
• To evaluate these methods against Multiple Correspondence Analysis (HOMALS) and MDS on similarity ratings.

Main Methods:

• Applied Multidimensional Scaling (MDS) to both direct dissimilarity measures (D) and profile distances (Δ).
• Utilized Multiple Correspondence Analysis (HOMALS) on raw sorting data.
• Conducted MDS on pairwise similarity ratings from the same subjects.

Main Results:

• Multidimensional Scaling (MDS) on direct dissimilarity measures (D) showed slight superiority over MDS on profile distances (Δ).
• Results from MDS on profile distances (Δ) were similar to those obtained from the more efficient HOMALS program.
• Observed average differences between the compared methods were very small.

Conclusions:

• MDS on direct dissimilarity measures (D) is a slightly preferable method for analyzing stimulus sorting data.
• MDS on profile distances (Δ) offers a viable alternative, yielding comparable results to HOMALS.
• The choice of method has minimal impact on the overall findings in analyzing stimulus dissimilarity data.