mds presentation natalia angelina

Upload: robinvarshney

Post on 03-Jun-2018

230 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/11/2019 MDS Presentation Natalia Angelina

    1/35

    MultidimensionalScaling (MDS)

    Angelina Anastasova

    Natalia Jaworska

    PSY5121 March 18/2008

  • 8/11/2019 MDS Presentation Natalia Angelina

    2/35

    Multidimensional Scaling (MDS):

    What Is It?

    Generally regarded as exploratory data analysis(Ding, 2006).

    Reduces large amounts of datainto easy-to-visualizestructures.

    Attempts to find structure(visual representation) in a set ofdistance measures, e.g. dis/similarities, between objects/cases.

    Shows how variables/objects are related perceptually.

    How? By assigning cases to specific locations in space.

    Distances between points in space match dis/similarities asclosely as possible:

    Similar objects: Close pointsDissimilar objects: Far apart points

  • 8/11/2019 MDS Presentation Natalia Angelina

    3/35

    MDS Example: City Distances

    Distances

    Matrix:

    Symmetric

    Spatial Map

    Dimensions

    1: North/South

    2: East/West

    Cluster

  • 8/11/2019 MDS Presentation Natalia Angelina

    4/35

  • 8/11/2019 MDS Presentation Natalia Angelina

    5/35

    Data Collection for MDS (1)

    Direct/raw data: Proximities values directly obtained

    from empirical, subjective scaling. E.g. Rating or ranking dis/similarities (Likert scales).

    Indirect/derived data: Computed from other measurements:correlations or confusion data (based on mistakes) (Davidson, 1983).

    E.g. Letters of alphabet presented briefly and must be identified. Rarelyconfused letters given high dissimilarity values, those that are confusedget low values.

    Data collection: Pairwise comparison, grouping/sorting tasks,direct ranking, objective method (e.g. city distances).

    Pairwise comparisons: All object pairs randomly presented:# of pairs = n(n-1)/2, n= # of objects/cases

    Can be tedious and inefficient process.

  • 8/11/2019 MDS Presentation Natalia Angelina

    6/35

  • 8/11/2019 MDS Presentation Natalia Angelina

    7/35

  • 8/11/2019 MDS Presentation Natalia Angelina

    8/35

    Types of MDS (2)

    More typical in Social Sciences is the classification of

    MDS based on nature of responses:

    1) DecompositionalMDS: Subjects rate objects on an

    overall basis, an impression, without reference toobjective attributes.

    Production of a spatial configuration for an individual and

    a composite map for group.

    2) CompositionalMDS: Subjects rate objectson a variety of specific, pre-specified attributes(e.g. size).

    No maps for individuals, only composite maps.

  • 8/11/2019 MDS Presentation Natalia Angelina

    9/35

    Classical MDS uses Euclidean principles to model

    data proximities in geometrical space, where distance(dij) between points iandjis defined as:

    xiand xjspecify coordinates of points i

    and j on dimension a, respectively.

    The modeled Euclidean distances are related to the observed

    proximities, ij, by some transformation/function (f).

    Most MDS models assume that the data have the form:

    ij =f(dij) All MDS algorithms are a variation of the above (Davidson,

    1983).

    The MDS Model

  • 8/11/2019 MDS Presentation Natalia Angelina

    10/35

    Output of MDS

    MDS Map/Perceptual Map/Spatial Representation:

    1) Clusters: Groupings in a MDS spatialrepresentation.

    These may represent a domain/subdomain.2) Dimensions: Hidden structures in data. Orderedgroupings that explain similarity between items.

    Axes are meaningless and orientation is arbitrary.

    In theory, there is no limit to the number ofdimensions.

    In reality, the number of dimensions that can be

    perceived and interpreted is limited.

  • 8/11/2019 MDS Presentation Natalia Angelina

    11/35

    Diagnostics of MDS (1)

    MDS attempts to find a spatial configuration Xsuchthat the following is true:f(ij) dij(X)

    Stress(Kruskals) function: Measures degree ofcorrespondence between distances among points on the

    MDS map and the matrix input.Proportion of variance of disparities

    notaccounted for by the model:

    Range 0-1: Smaller stress = better representation.

    None-zero stress: Some/all distances in the map aredistortions of the input data.

    Rule of thumb: 0.1 is excellent; 0.15 not tolerable.

  • 8/11/2019 MDS Presentation Natalia Angelina

    12/35

    R2(RSQ): Proportion of variance of the disparitiesaccounted for by the MDS procedure.

    R20.6 is an acceptable fit.

    Weirdness Index: Correspondence of subjects map and the

    aggregate map outlier identification. Range 0-1: 0 indicates that subjects weights are proportional to the

    average subjects weights; as the subjects score becomes moreextreme, index approaches 1.

    Shepard Diagram: Scatterplot of input proximities (X-axis)against output distances (Y-axis) for every pair of items.

    Step-line produced. If map distances fall on the step-line thisindicates that input proximities are perfectly reproduced by the MDSmodel (dimensional solution).

    Diagnostics of MDS (2)

  • 8/11/2019 MDS Presentation Natalia Angelina

    13/35

    Interpretation of Dimensions

    Squeezing data into 2-D enables readability but maynot be appropriate: Poor, distorted representation of thedata (high stress).

    Scree plot: Stress vs.

    number of dimensions.E.g. cities distance

    Primary objective in dimension interpretation: Obtain

    best fit with the smallest number of possibledimensions.

    How does one assign meaning to dimensions?

  • 8/11/2019 MDS Presentation Natalia Angelina

    14/35

    Meaning of Dimensions

    Subjective Procedures:Labelling the dimensions by visual

    inspection, subjective

    interpretation, and informationfrom respondents.

    Experts evaluate and identify thedimensions.

  • 8/11/2019 MDS Presentation Natalia Angelina

    15/35

    Validating MDS Results

    Split-sample comparison:

    Original sample is divided and a correlationbetween the variables is conducted.

    Multi-sample comparison: New sample is collected and a correlation is

    conducted between the old and new data.

    Comparisons are done visually or with a simple

    correlation of coordinates or variables.Assessing whether MDS solution(dimensionality extraction) changes in asubstantial way.

  • 8/11/2019 MDS Presentation Natalia Angelina

    16/35

    MDS Caveats

    Respondents probably perceive stimulidifferently. In non-aggregate data, differentdimensions may emerge.

    Respondents may attach different levels ofimportance to a dimension.

    Importance of a dimension may change over time.

    Interpretation of dimensions is subjective.

    Generally, more than four times as many objectsas dimensions should be compared for the MDSmodel to be stable.

  • 8/11/2019 MDS Presentation Natalia Angelina

    17/35

    Advantages of MDS An alternative to the GLM.

    Does not require assumptions of linearity,metricity, or multivariate normality.

    Can be used to model nonlinear relationships.

    Dimensionality solution can be obtained fromindividuals; gives insight into how individualsdiffer from aggregate data.

    Reveals dimensions without the need for definedattributes.

    Dimensions that emerge from MDS can beincorporated into regression analysis to assess

    their relationship with other variables.

  • 8/11/2019 MDS Presentation Natalia Angelina

    18/35

    Disadvantages of MDS

    Provides a global measure of dis/similarity but

    does not provide much insight into subtleties (Streetet al., 2001).

    Increased dimensionality: Difficult to representand decreases intuitive understanding of the data.

    As such, the model of the data becomes as

    complicated as the data itself.

    Determination of meanings of dimensions is

    subjective.

  • 8/11/2019 MDS Presentation Natalia Angelina

    19/35

    A Tiny Break . . .

  • 8/11/2019 MDS Presentation Natalia Angelina

    20/35

    SPSSing MDS

    In the SPSS Data Editor window, click: Analyze>

    Scale> Multidimensional Scaling

  • 8/11/2019 MDS Presentation Natalia Angelina

    21/35

    Select four or more Variablesthat you want to test.

    You may select a single variable for the Individual

    Matrices forwindow (depending on the distances optionselected).

  • 8/11/2019 MDS Presentation Natalia Angelina

    22/35

    If Data are distances(e.g. cities distances) option is

    selected, click on the Shapebutton to define

    characteristic of the dissimilarities/proximity matrices.

    If Create distance from data

    isselected, click on the

    Measurebutton to control the

    computation of dissimilarities,

    to transform values, and to

    compute distances.

  • 8/11/2019 MDS Presentation Natalia Angelina

    23/35

    In the Multidimensional Scaling dialog box, click on the

    Modelbutton to control the level of measurement,

    conditionality, dimensions, and the scaling model.

    Click on the Optionsbutton to control the

    display options, iteration criteria, and

    treatment of missing values.

  • 8/11/2019 MDS Presentation Natalia Angelina

    24/35

    MDS: A Psychological Example

    Multidimensional scaling modelling approach to latentprofile analysis in psychological research (Ding, 2006)

    Basic premise: Utilize MDS to investigate types orprofiles of people.

    Profile: From applied psych where test batteries areused to extract and construct distinctivefeatures/characteristics in people.

    MDS method was used to:

    Derive profiles (dimensions) that could provide informationregarding psychosocial adjustment patterns in adolescents.

    Assess if individuals could follow different profile patternsthan those extracted from group data, i.e. deviations fromthe derived normative profiles.

  • 8/11/2019 MDS Presentation Natalia Angelina

    25/35

  • 8/11/2019 MDS Presentation Natalia Angelina

    26/35

    Data for MDS

    Scored data for MDS profile analysis Sample data for 14 individuals:

    BI=body image, PR=peer relations, FR=family relations, MC=mastery & coping,VE=vocational & educational goal, SA=superior adjustment, PMI-1=profile matchindex for Profile 1, PMI-2=profile match index for Profile 2, LS=life satisfaction,

    Dep=depression, PL=psychological loneliness

  • 8/11/2019 MDS Presentation Natalia Angelina

    27/35

    MDS map

    Euclidean distance model

    Profile 1

    3210-1-2

    2.0

    1.5

    1.0

    .5

    0.0

    -.5

    -1.0

    -1.5

    save mc

    fr

    pr

    bi

    The Analysis: Step by Step

    Step 1: Estimate the number of profiles(dimensions) from the latent variables.

    Kruskal's stress = 0.00478Excellent stress value.

    RSQ = 0.9998

    Configuration derived in 2

    dimensions.

  • 8/11/2019 MDS Presentation Natalia Angelina

    28/35

    MDS map

    Euclidean distance model

    Profile 1

    3210-1-2

    2.0

    1.5

    1.0

    .5

    0.0

    -.5

    -1.0

    -1.5

    save mc

    fr

    pr

    bi

    Scale values of two MDS profiles (dimensions) in

    psychosocial adjustment.

    Normative profiles of

    psychosocial adjustments

    in young adults.

    Each profile represents

    prototypical individual.

  • 8/11/2019 MDS Presentation Natalia Angelina

    29/35

    Step 2: Using the estimated scale values as

    independent variables and observed variables

    as dependent variables estimate: Individual profile match index (PMI):

    The extent of individual variability along a profile.

    Intra-individual variability across profiles.

    PMI-1=profile match index for Profile 1, PMI-2=profile

    match index for Profile 2, LS=life satisfaction,

    Dep=depression, PL=psychological loneliness

    Fit index:

    The proportion of variance

    in the individuals observed

    data that can be accounted

    for by the profiles.

  • 8/11/2019 MDS Presentation Natalia Angelina

    30/35

    Individual Profiles vs. Aggregate

    PMI-1 PMI-2 FIT

    Subject 1 -0.73 0.29 0.94

    Subject 2 -0.38 0.23 0.99

    Subject 4 -0.16 0.24 0.32

    Profile 1 Profile 2 Subject 1 Subject 2 Subject 4

    Body Image (BI) 2.28 -0.5 2.82 3.82 5.09

    Peer Relations (PR) 0.23 1.49 5.1 5 5.3

    Family Relations (FR) 0.7 -1.2 5 4.71 4.69

    Mastery & Coping (MC) -0.25 0.14 4.6 4.9 6

    Voc-Ed Goals (VE) -1.49 0 5.7 5.4 6

    Superior Adjust. (SA) -0.08 0.08 4.3 4.9 5.5

    -2

    -1

    0

    1

    2

    3

    4

    5

    6

    7

    1 2 3 4 5 6

    Profile 1

    Profile 2

    Subject 1

    Subject 2

    Subject 4

  • 8/11/2019 MDS Presentation Natalia Angelina

    31/35

    Step 3: Assess the association between profiles

    and other factors by regression.

    Profile 1: -High scores on Body Image - higher degree of life satisfaction.

    -High scores on the Vocational-Educational Goal - higher degree of depression.Profile 2: -Higher scores on the family relationships profile - higher degree of psychological

    loneliness.

    Level: -Average scores of individuals psychosocial adjustment.

    -Overall positive psychosocial adjustment scores suggest less depression or

    psychological loneliness and higher degree of life satisfaction.

  • 8/11/2019 MDS Presentation Natalia Angelina

    32/35

    Commentary on MDS Profile

    Analysis

    Strength of MDS profile analysis:

    Provides representation of what typical

    configurations or profiles of variables exist in the

    population and how individuals differ with respect

    to these profiles.

    Enables identification/analysis of:

    Individuals who develop in an idiographic (specific

    and subjective) manner; not consistent with

    aggregate profiles.

  • 8/11/2019 MDS Presentation Natalia Angelina

    33/35

  • 8/11/2019 MDS Presentation Natalia Angelina

    34/35

    Thank You!

    Questions?

  • 8/11/2019 MDS Presentation Natalia Angelina

    35/35

    References

    Davidson, M. L. (1983).Multidimensional scaling. New York: J. Wileyand Sons.

    Ding, C. S. (2006). Multidimensional scaling modelling approach to latentprofile analysis in psychological research.International Journal ofPsychology41 (3), 226-238.

    Kruskal, J.B. & Wish M.1978.Multidimensional Scaling. Sage.

    Street, H., Sheeran, P., & Orbell, S. (2001). Exploring the relationshipbetween different psychosocial determinants of depression: amultidimensional scaling analysis.Journal of Affective Disorders 64,5367.

    Takane, Y., Young, F.W., & de Leeuw, J. (1977). Nonmetric individualdifferences multidimensional scaling: An alternating least squares methodwith optimal scaling features,Psychometrika42 (1), 767.

    Young, F.W., Takane, Y., & Lewyckyj, R. (1978). Three notes onALSCAL,Psychometrika43 (3), 433435.

    http://www.analytictech.com/borgatti/profit.htm

    http://www2.chass.ncsu.edu/garson/pa765/mds.htm

    http://www.terry.uga.edu/~pholmes/MARK9650/Classnotes4.pdf

    http://www.analytictech.com/borgatti/profit.htmhttp://www2.chass.ncsu.edu/garson/pa765/mds.htmhttp://www.terry.uga.edu/~pholmes/MARK9650/Classnotes4.pdfhttp://www.terry.uga.edu/~pholmes/MARK9650/Classnotes4.pdfhttp://www2.chass.ncsu.edu/garson/pa765/mds.htmhttp://www.analytictech.com/borgatti/profit.htm