Data Fit (Stress) vs. Model Fit (Recovery) in Multidimensional Scaling

Authors

  • Ingwer Borg WWU, Münster

DOI:

https://doi.org/10.17713/ajs.v49i2.918

Abstract

The paper studies the relation of data fit (Stress) and model fit (recovery of a true latent configuration superimposed with random error) depending on the number of points and the error level in the data when running ordinal and interval MDS. Using 2-dimensional MDS with 6 to 100 points and ten levels of error, it is found that adding more points (given that n>10) is driving up the Stress of the MDS solutions but also leads to better recovery, in general. The recovery exceeds any reasonable statistical benchmarks, i.e. more points enable MDS to smooth out the error in the data so that the true configuration can surface more clearly in the MDS solution.

Published

2020-02-20

How to Cite

Borg, I. (2020). Data Fit (Stress) vs. Model Fit (Recovery) in Multidimensional Scaling. Austrian Journal of Statistics, 49(2), 43-52. https://doi.org/10.17713/ajs.v49i2.918