A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods

T.H.A. Bijmolt, M. Wedel

Research output: Book/ReportReportProfessional

287 Downloads (Pure)

Abstract

We compare three alternative Maximum Likelihood Multidimensional Scaling methods for pairwise dissimilarity ratings, namely MULTISCALE, MAXSCAL, and PROSCAL in a Monte Carlo study.The three MLMDS methods recover the true con gurations very well.The recovery of the true dimensionality depends on the test criterion (likelihood ratio test, AIC, or CAIC), as well as on the MLMDS method. The three MLMDS methods t the dissimilarity data equally well.The methods are relatively robust against violations of their distributional assumptions. MULTISCALE outperforms PROSCAL and MAXSCAL with respect to computation time.In a separate Monte Carlo study, it is shown that the MLMDS methods frequently converge to local optima, especially if a random start is used.Rational starts, however, turn out to provide a satisfactory solution for the local optima problem.Implications for researchers intending to apply MLMDS are provided.
Original languageEnglish
Place of PublicationTilburg
PublisherMarketing
Number of pages49
Volume725
Publication statusPublished - 1996

Publication series

NameFEW Research Memorandum
Volume725

Fingerprint

Maximum likelihood
Evaluation
Multidimensional scaling
Monte Carlo study
Dissimilarity
Violations
Likelihood ratio test
Rating
Dimensionality

Keywords

  • marketing
  • monte carlo technique
  • scaling

Cite this

Bijmolt, T. H. A., & Wedel, M. (1996). A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods. (FEW Research Memorandum; Vol. 725). Tilburg: Marketing.
Bijmolt, T.H.A. ; Wedel, M. / A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods. Tilburg : Marketing, 1996. 49 p. (FEW Research Memorandum).
@book{f72cc9d8f37043aaa22448c1302a48b8,
title = "A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods",
abstract = "We compare three alternative Maximum Likelihood Multidimensional Scaling methods for pairwise dissimilarity ratings, namely MULTISCALE, MAXSCAL, and PROSCAL in a Monte Carlo study.The three MLMDS methods recover the true con gurations very well.The recovery of the true dimensionality depends on the test criterion (likelihood ratio test, AIC, or CAIC), as well as on the MLMDS method. The three MLMDS methods t the dissimilarity data equally well.The methods are relatively robust against violations of their distributional assumptions. MULTISCALE outperforms PROSCAL and MAXSCAL with respect to computation time.In a separate Monte Carlo study, it is shown that the MLMDS methods frequently converge to local optima, especially if a random start is used.Rational starts, however, turn out to provide a satisfactory solution for the local optima problem.Implications for researchers intending to apply MLMDS are provided.",
keywords = "marketing, monte carlo technique, scaling",
author = "T.H.A. Bijmolt and M. Wedel",
note = "Pagination: 49",
year = "1996",
language = "English",
volume = "725",
series = "FEW Research Memorandum",
publisher = "Marketing",

}

Bijmolt, THA & Wedel, M 1996, A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods. FEW Research Memorandum, vol. 725, vol. 725, Marketing, Tilburg.

A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods. / Bijmolt, T.H.A.; Wedel, M.

Tilburg : Marketing, 1996. 49 p. (FEW Research Memorandum; Vol. 725).

Research output: Book/ReportReportProfessional

TY - BOOK

T1 - A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods

AU - Bijmolt, T.H.A.

AU - Wedel, M.

N1 - Pagination: 49

PY - 1996

Y1 - 1996

N2 - We compare three alternative Maximum Likelihood Multidimensional Scaling methods for pairwise dissimilarity ratings, namely MULTISCALE, MAXSCAL, and PROSCAL in a Monte Carlo study.The three MLMDS methods recover the true con gurations very well.The recovery of the true dimensionality depends on the test criterion (likelihood ratio test, AIC, or CAIC), as well as on the MLMDS method. The three MLMDS methods t the dissimilarity data equally well.The methods are relatively robust against violations of their distributional assumptions. MULTISCALE outperforms PROSCAL and MAXSCAL with respect to computation time.In a separate Monte Carlo study, it is shown that the MLMDS methods frequently converge to local optima, especially if a random start is used.Rational starts, however, turn out to provide a satisfactory solution for the local optima problem.Implications for researchers intending to apply MLMDS are provided.

AB - We compare three alternative Maximum Likelihood Multidimensional Scaling methods for pairwise dissimilarity ratings, namely MULTISCALE, MAXSCAL, and PROSCAL in a Monte Carlo study.The three MLMDS methods recover the true con gurations very well.The recovery of the true dimensionality depends on the test criterion (likelihood ratio test, AIC, or CAIC), as well as on the MLMDS method. The three MLMDS methods t the dissimilarity data equally well.The methods are relatively robust against violations of their distributional assumptions. MULTISCALE outperforms PROSCAL and MAXSCAL with respect to computation time.In a separate Monte Carlo study, it is shown that the MLMDS methods frequently converge to local optima, especially if a random start is used.Rational starts, however, turn out to provide a satisfactory solution for the local optima problem.Implications for researchers intending to apply MLMDS are provided.

KW - marketing

KW - monte carlo technique

KW - scaling

M3 - Report

VL - 725

T3 - FEW Research Memorandum

BT - A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods

PB - Marketing

CY - Tilburg

ER -

Bijmolt THA, Wedel M. A Monte Carlo Evaluation of Maximum Likelihood Multidimensional Scaling Methods. Tilburg: Marketing, 1996. 49 p. (FEW Research Memorandum).