SPSS Statistics

 View Only
  • 1.  Multidimensional scaling and creating a perceptual map for unpaierd data

    Posted Mon March 04, 2024 12:23 PM
    Edited by Frank Mon March 04, 2024 01:35 PM

    Hello Everyone,

    I am trying to run Multidimensional scaling to create a perceptual map for a 15 product brands (e.g., n= 10-20 observations per products) with 10 variables ratings of these products. The products/brands are not paired. Is there a way to do this in SPSS? I know this can be done in cluster analysis, but I like to try it in MDS.

    Thank you



    ------------------------------
    Dr. Frank
    Professor
    UMass
    ------------------------------



  • 2.  RE: Multidimensional scaling and creating a perceptual map for unpaierd data

    Posted Mon March 04, 2024 01:24 PM

    Please can you say more on the design of your data or display it? And what do you mean by "paired" in your case?



    ------------------------------
    Kirill Orlov
    ------------------------------



  • 3.  RE: Multidimensional scaling and creating a perceptual map for unpaierd data

    Posted Mon March 04, 2024 01:33 PM

    Hi Kirill

    I have 11 variables one of which is brand (categorical). The others (10) have been rated (likert scale). The brands have not been subjected to paired comparison as one would usually do in MDS. Thank you.

    Frank



    ------------------------------
    Dr. Frank
    Professor
    UMass
    ------------------------------



  • 4.  RE: Multidimensional scaling and creating a perceptual map for unpaierd data

    Posted Mon March 04, 2024 02:20 PM

    You could compute and output the brands x brands square symmetric matrix of some dis(similarity) measure based on your likert features (say, euclidean distance, but there are many more) by the Distances procedure (= command PROXIMITIES). Then input it to the MDS. In SPSS, MDS menu allows to request computation of distances right from itself, for user convenience.

    It sounded like you have multiple people having rated each brand. If you are considering to do a weighted-model MDS (such as INDSCAL model or Repeated MDS model) what needs multiple matrices to input, you could compute a separate matrix for each individual rater, rather than the single matrix based on all them.



    ------------------------------
    Kirill Orlov
    ------------------------------



  • 5.  RE: Multidimensional scaling and creating a perceptual map for unpaierd data

    Posted Mon March 04, 2024 02:30 PM

    P S. To to save a matrix from the PROXIMITIES, paste syntax it generates and add /MATRIX=OUT(*) subcommand.

    To save likewise multiple matrices, do the same run under split-file regime by "rater" variable.



    ------------------------------
    Kirill Orlov
    ------------------------------



  • 6.  RE: Multidimensional scaling and creating a perceptual map for unpaierd data

    Posted Mon March 04, 2024 08:32 PM

    Thank you. I had tried the distances option, but ended up with a matrix showing each observation as a data point in the output as correlation matrix. I will try your suggestion, but not really sure what goes into the weighted box?

    Best regards, 



    ------------------------------
    Dr. Frank
    Professor
    UMass
    ------------------------------



  • 7.  RE: Multidimensional scaling and creating a perceptual map for unpaierd data

    Posted Tue March 05, 2024 01:03 AM

    To compute distaces between brands you must have 10 Brands by 15 Features dataset. But you have also 20 Raters, so 200 cases total. If you want individual matrices for each rater, split file by rater (after sorting cases by Rater and Brand).

    If you want one combined matrix based on all the Raters, first compute their mean rate for each Feature, by Aggregate, saving out a new, 10 by 15 dataset with the means.

    Also, don't forget to consider whether any standardizations and what specifically are due to do as pre-processing of your initial data.



    ------------------------------
    Kirill Orlov
    ------------------------------



  • 8.  RE: Multidimensional scaling and creating a perceptual map for unpaierd data

    Posted Tue March 05, 2024 02:29 PM

    Thank you again,
    I need to have a combined Aggregate matrix based on all the Raters. However, having difficulty saving an out file as data set to be used. 



    ------------------------------
    Dr. Frank
    Professor
    UMass
    ------------------------------