How to Use PREFMAP - A Program Which Relates Preference Data to Multidimensional Scaling Solution (1)

Jih Jie Chang and J. Douglas Carroll
Bell Laboratories
Murray Hill, New Jersey


NON-TECHNICAL INTRODUCTION

The PREFMAP program relates a stimulus space to preference data (the stimulus space must be produced by another PC-MDS program such as KYST or INDSCAL). A hierarchy of vector and ideal point models are performed. PREFMAP starts with coordinate input describing a stimulus configuration in a specified number of dimensions. Also input is a set of preference rankings, in which each subject ranks the stimuli. In context of this stimulus input, a set of preference scales is evaluated.

PREFMAP finds for each subject, an ideal point that is positioned within the stimulus space. For each of these ideal points, the squared Euclidean distances from each stimulus to the ideal point are linearly (for metric model) or monotonically (for non-metric model) related to the preferences expressed by the subject. The ideal point is defined differently in each of the four models.

Each subject's data is evaluated using any of the four models (called phases). When a model is selected, it is tested for its ability to describe the distance between the subject's ideal points (or preference vector in model 4) and the stimuli.

The PREFMAP output consists of an analysis for each phase requested and a Summary Table showing correlations within phases, F ratios for each phase and F ratios analyzing differences between phases. It is the analysis of correlations that shows how much of the variance in a subject's preferences are explained by a given phase or model. The F ratios show the improvement attributable to movement to a more sophisticated model.

While the Summary Table is developed for, and is technically correct for evaluating only metric data, the table may be used as a directional indicator of relative improvement attributable to model shifts when non-metric data is analyzed.


TECHNICAL INTRODUCTION

PREFMAP relates preference data to a multidimensional solution via a hierarchy of models, ranging from a linear 'vector model' through the Coombs, Bennett and Hays 'multidimensional unfolding model', and finally including generalizations of the multidimensional unfolding model.  The method is described in Carroll's paper (1972).

Given a stimulus configuration of N points in K dimensional space and a set of preference scales in the form of a subject-by- stimulus matrix, the program finds for each individual an ideal point in the given stimulus space such that the squared Euclidean distances (defined differently in each of the four models) from each stimulus to the ideal point are linearly (in the metric case) or monotonically (in the nonmetric case) related to the preferences expressed by the subject.

PREFMAP consists of four phases corresponding to the four models.  Phase 1 is the most general model in which each individual is allowed their own orientation and weighting of dimensions.  Phase 2 allows each individual differential weighting of dimensions only.  In Phase 3 all subjects are assumed to have the same orientation and weighting of dimensions, while Phase 4 corresponds to the vector model.  The algorithm for the linear vector model is basically the same as that described in Chang and Carroll (1968), for the case in which linear correlation between the preference scales (data) and the projections of the stimulus points on the fitted vectors is maximized.

PREFMAP allows either metric or nonmetric analysis (or both) in all four phases.


Input And Limits

The input data is in the form of a subjects-by-stimuli (or objects) matrix.  Each row of the matrix represents a subject and the entries in the row are the preference judgments made by that subject on each of the stimuli (or objects).  The maximum number of subjects that can be analyzed in PREFMAP is 99 and the maximum number of stimulus points allowed is 150. The maximum number of dimensions allowed is five. The coordinates of N stimulus points in K dimensional space must be provided by the user in PREFMAP.

Printed Output

The results of the analysis are printed for each subject within each phase consecutively.  They will be listed separately for each phase.

A. Phase 1--Each subject is allowed their own orientation and pattern of weights of the rotated dimensions.  For each subject, the output consists of:


1. The direction cosines of the new (individual) axes with respect to the group stimulus space.
2. The composite transformation matrix.
3. The transformed coordinates of the individual space.
4. The coordinates of the ideal point in the old group stimulus space.
5. The coordinates of the ideal point in the new individual space.
6. The importances (or the weights), on each new or rotated dimension.  A negative weight indicates an anti-ideal or least preferred vis a vis that particular dimension.

B. Phase 2--Each subject is allowed differential weighting of dimensions only.


1. The coordinates of the ideal point in the group stimulus space.
2. The importances on each dimension.  Same explanation applies here for negative weights.

C. Phase 3--

Each subject has the same orientation and weighting of the axes.  The same kinds of results as in Phase 2 are printed.  The importance in this case are the same for all dimensions.  Though the weights differ from subject to subject, they reflect only the differences in scale factors.  The sign of the weights are to be interpreted in the same way as in Phases 1 and 2.

D. Phase 4--

This is the vector model.  The direction cosines of the fitted subject vectors are printed. PREFMAP is to be run in consecutive phases i.e., 2 to 4; 3 to 4 etc. If Phase 4 is run alone, the vector is reversed.

E.

At the end of the run, a composite table of correlations, F ratios and root mean squares are printed.  The correlations are between the obtained squared distances and the preference scale values or monotonically transformed preference scale values in the event the monotonic option is used for each subject (including the average subject) in each phase.  The F ratios are printed with degrees of freedom for each subject in each phase and for pairs of phases.  The root mean squares are for each phase.

Plotted Output

The program produces four kinds of plots, namely:  1) the individual ideal point plot which is only appropriate for Phase 1 where the individual orientation of the axes is allowed.  The plot shows the ideal point in its rotated stimulus space; 2) the composite ideal point plot is drawn respectively at the end of Phases 1, 2 and 3.  The plot shows all the ideal points in the group stimulus space; 3) the function plot of obtained squared distances versus the scale values which shows how linearly or monotonically the solution is related to the preference scales; 4) the vector plot which is appropriate only for Phase 4 and shows each subject as a vector in the group stimulus space.

Input File Arrangement

Each of the numbered lines listed below may refer to a set of images, e.g., line 3 is a set of images including the format and the matrix of the scale values.

Line 1. Parameter Line (Must Contain ALL 15 Parameters)

Line 2, 3.  Stimulus Space Coordinates.

Line 2. A format line (use floating point format).

Line 3. X--a matrix of the coordinates of N stimuli in K dimensional space.  X must be entered as indicated by IRX.

Lines 4, 5.  These lines are present only when IRWT = 1.

Line 4. A format line (using floating point format).

Line 5. Weights--one value for each dimension.  There should be K values.

Lines 6, 7.  Scale values.

Line 6. Format line for reading (use floating point format).

Line 7. S--scale values.  S is a matrix of (NSUB) subjects by stimulus. Each row represents a subject.

Line 8. Line 8. Labels for each of the n stimuli, followed by labels for each of the nsub subjects.


THE PARAMETER LINE: MANDATORY PARAMETERS:

Line 1.  Parameters (Free Format).  There are 15 parameters.  The first six parameters are defined as the 'data parameters' which are mandatory.  The remaining 9 are 'option parameters' which will be discussed in detail in the "Parameter Options" section.  A summary of these parameters and their default values are given in Table 1.  N, K, NSUB, ISV, NORS, IRX, IPS, IPE, IRWT, LFITSW, IAV, MAXIT, ISHAT, IPLOT, CRIT.

N number of stimuli. (Usually # Rows in 1st matrix)

K number of dimensions. (Usually # Columns in 1st matrix)

NSUB number of subjects. (# Rows in 2nd matrix)

ISV 0, small scale value represents greater preference.
        1, large scale value represents greater preference.

NORS 1, normalize scale values, for each subject make length = 1.
        0, do not normalize scale values.

IRX 0, the coordinates of the stimulus points are entered as N by K. (Stimuli x Dimensions)
        1, the coordinates are entered as K by N. (Dim. x Stim.)

OPTIONAL PARAMETERS (See Next Section:):

IPS, IPE, IRWT

Phase Parameters

LFITSW, IAV, MAXIT, ISHAT

Nonmetric Parameters

IPLOT

Plotting Parameters

CRIT

present only if doing nonmetric analysis

OPTIONAL PARAMETERS:

Phase Parameter Options--IPS, IPE, IRWT

The program computes solutions for Phase 1 to Phase 4 sequentially.  However, the starting and the ending phases can be changed simply by setting IPS equal to the phase where computation starts and IPE to the phase it stops, for example, if IPS = 3 and IPE = 4, the program will compute solutions for Phases 3 and 4 only.  In case the computation starts on Phase 3, one has the option of applying different weights to the dimensions of the group stimulus space.  This can be accomplished by setting IRWT = 1 and the weights are read, as described in section IV.

Nonmetric Parameter Options--LFITSW, IAV, MAXIT, ISHAT

The nonmetric analysis employs an iterative procedure for monotone regression.  On each iteration the program finds for each subject the best monotone fit of d2 (the squared Euclidean distances between the stimuli and the ideal point) to the data matrix S (the preference scale values).  (See Kruskal 1964 a,b)

LFITSW: option of how d2 is assumed to be related to S.

--0 linearly (or metrically).

--1 monotonically (plan monotone regression, appropriate when there are no ties in the data).

--2 monotonically (block monotone regression using the primary approach of treating ties among values for a group of equal data values, no restriction is placed on ordering within blocks for the fitted values).

--3 monotonically (block monotone regression using the secondary approach of treating ties among data values for a group of equal data values, the fitted values are required to be equal).

IAV--option on how to define the preference scale values of the average subject in each phase.  In the metric analysis this option is irrelevant because once the average subject's scale values are computed in the starting phase, they remain the same.  However, in the nonmetric case each subject's scale values are replaced in each phase by the best monotone fit values.  Therefore, the average subject's scale values are different in each phase.

--0 average subject's scale values are computed in the starting phase and remain the same.

--1 average subject's scale values are computed for each new phase.

MAXIT--number of iterations on monotone fit.  The default value for MAXIT is 15.

ISHAT--parameter deciding which set of scale values are used in each phase.

--1 at the beginning of a new phase, use the last monotone fit of the scale values from the previous phase.

--0 at the beginning of a new phase, use the original scale values.

Plotting Parameters

With respect to the four kinds of plot earlier described, option parameter IPLOT designates the kinds of plot to be produced.  As default the program generates the average subject's ideal point plot and the composite ideal point plot for Phases 1, 2 and 3,. for Phase 4, the vector plot.

Symbols on Plot

The stimulus points are labeled by sequential numbers generated in the program.  The ideal points may be labeled by either numbers or letters.

Parameters

IPLOT--option on kinds of plot for Phases 1 and 2.

--0 draw ideal point plot for the average subject only.

--1 draw ideal point plot for the average subject only and function plots for each subject.

--2 draw ideal point plot for the average subject and both function plots and ideal point plots for each subject.

Iterative Procedure Stop

CRIT is the criterion for stopping iterative procedure on monotone fit.  If (SI-SI-1)2 <_ CRIT, iteration is terminated.  (Where I denotes Ith iteration.) (CRIT=.001 seems to be a reasonable value) 


REFERENCES

Carroll, J.D.  "Individual Differences and Multidimensional Scaling."  In R.N. Shepard, A.K. Romney, and S.B. Nerlove (Eds.), Multidimensional Scaling:  Theory and Applications in the Behavioral Sciences.  Vol. 1, 1972.

Carroll, J.D., and Chang, J.J.  "An Alternate Solution To The 'Metric Unfolding Problem.'" Paper presented at the Psychometric Society meeting, April 1971, St. Louis, Missouri.

Chang, J.J., and Carroll, J.D.  "How To Use PROFIT, A Computer Program For Property Fitting By Optimizing Nonlinear And Linear Correlation."  Unpublished paper, Bell Laboratories, 1968.

Kruskal, J.B.  "Multidimensional Scaling By Optimizing Goodness-of-fit To a Nonmetric Hypothesis."  Psychometrika, 1964, 29, 1-27.  (a)

Kruskal, J.B.  "Nonmetric Multidimensional Scaling:  A Numerical Method."  Psychometrika, 1964, 29, 115-129.  (b)



NOTES
1.  The original title of this document was "How to Use PREFMAP and PREFMAP2 - Programs Which Relate Preference Data to Multidimensional Scaling Solution." The original paper has been modified for the PC-MDS version by Scott M. Smith.


Sample Data File
  11   2   5   0   1   0   2   4   0   1   1  15   0   0  1
 (2X,10F7.3) 
 1  -.504  -.054             MUSTANG 
 2   .332  -.156             CADILLAC SEVILLE 
 3   .174  -.018             LINCOLN CONTINENTAL 
 4 -1.429   .083             FORD ESCORT 
 5   .323   .506             CORVETTE 
 6 -1.402  -.512             CHEV. CHEVETTE 
 7   .493  -.565             NISSAN 300 ZX 
 8 -1.142   .498             RENAULT ALLIANCE 
 9   .712   .381             PORSCHE 944 
10  1.154  -.367             JAGUAR XJ6 
11  1.290   .205             MERCEDES 500 SEL 
(5X,11F2.0) 
001  0210071103120408010506         
002  0103040502070911100809         
003  0609081007110405030201 
004  0301020805100611040709 
005  0805070904110210060301 
MUSTANG
CADILLAC
LINCOLN
ESCORT
CORVETTE
CHEVETTE
300ZS
ALLINACE
944
XJ6
500SEL
SUBJ1
SUBJ2
SUBJ3
SUBJ4
SUBJ5



SAMPLE PREFMAP OUTPUT

                                P R E F M A P                           | 11 2 5 0 1 0 2 4 0 1 1 15 0 0 
            MDSCALING VIA A GENERALIZATION OF COOMBS UNFOLDING MODEL    | (2X,10F7.3) 
                    BY DR. J. D. CARROLL AND JIH JIE CHANG              | 1  -.504  -.054    MUSTANG 
                               PC - MDS VERSION                         | 2   .332  -.156    CADILLAC SEVILLE 
 ANALYSIS TITLE:                                                        | 3   .174  -.018    LINCOLN CONTINENTAL
 DATA IS READ FROM FILE: NEWCARPR.DAT                                   | 4 -1.429   .083    FORD ESCORT 
 OUTPUT FILE IS: NEWCARPR.PRN                                           | 5   .323   .506    CORVETTE 
                                                                        | 6 -1.402  -.512    CHEV. CHEVETTE 
 ****************************************************************-*******| 7   .493  -.565    NISSAN 300 ZX 
  N        NO. OF STIMULI                                      11       | 8 -1.142   .498    RENAULT ALLIANCE 
  K        NO. OF DIMENSIONS                                    2       | 9   .712   .381    PORSCHE 944 
  NSUB     NO. OF SUBJECTS                                      5       |10  1.154  -.367    JAGUAR XJ6 
  ISV    0=SMALL SCALE VALUE REPRESENTS GREATER PREF.           0       |11  1.290   .205    MERCEDES 500 SEL 
  NORS   1=NORMALIZE SCALE VALUES                               1       |(5X,11F2.0) 
  IRX    0=STIMULUS COORDINATES N BY K, OR 1 = K BY N           0       |001  0210071103120408010506 BRAND RANKS
  IPS      STARTING PHASE                                       2       |002  0103040502070911100809 FOR FIVE
  IPE      ENDING PHASE                                         4       |003  0609081007110405030201 SUBJECTS
  IRWT   1=READ IN WEIGHTS,  0=NO WEIGHTS READ IN               0       |004  0301020805100611040709 
  LFITSW   HOW D**2 IS RELATED TO SCALE VALUES                  1       |005  0805070904110210060301 
         0=LINEARLY,                                                    | 
         1=MONOTONE WITH NO TIES,                                       | 
         2=BLOCK MONOTONE WITH ORDERING IN BLOCKS                       |INPUT PARAMETERS TO PREFMAP 
         3=BLOCK MONOTONE WITH EQUALITY IN BLOCKS                       | 
  IAV    0=AVERAGE SUBJECTS COMPUTED ONCE FOR ALL PHASES,       1       | 
         1=CALCULATE EACH PHASE                                         | 
  MAXIT    MAXIMUM ITERATIONS, WHEN 0 IT IS SET TO 15          15       | 
  ISHAT  0=USE SCALE VALUES FROM PREVIOUS PHASE,                0       | 
         1=USE ORIG VALUES                                              | 
  IPLOT  0=AVERAGE SUBJECTS,                                    0       | 
         1=AVERAGE SUBJECTS & SUBJECT FUNCTIONS,                        | 
         2=ALL PLOTS                                                    | 
  CRIT     CRITERIA FOR STOPPING MONOTONE FIT                   .0010   | 
 ***********************************************************************| 
    *****IDENTIFICATION KEY FOR PLOTS WITH IDENTIFIED POINTS*****       |PLOT IDENTIFICATIONS FOR FIRST 
 PT #   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15       |50 STIMULI 
 CHAR   1   2   3   4   5   6   7   8   9   A   B   C   D   E   F       | 
 PT #  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30       | 
 CHAR   G   H   I   J   K   L   M   N   O   P   Q   R   S   T   U       | 
 PT #  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45       | 
 CHAR   V   W   X   Y   Z   +   /   =   *   &   $   @   %   ?   <       | 
 PT #  46  47  48  49  50                                               | 
 CHAR   (   )   "   #   @                                               | 
                                                                        | 
 POINT NUMBERS ABOVE 50 IDENTIFIED AS  >, MULTIPLE POINTS IDENTIFIED AS | 
                                                                        | 
 POINTS 1 TO 11 ARE STIMULI AND POINTS  12 TO   16 ARE IDEAL POINTS     | 
                                                                        | 
 VARIABLE FORMAT (STIMULUS COORDINATES)  =  (2X,10F7.3)                 | 
                                                                        | 

 ORIGINAL CONFIGURATION  (X MATRIX)                                     |PRINTOUT OF ORIGINAL INPUT 
                                                                        |FOR THIS EXAMPLE, THE CONFIGURATION
     1   -.50400   -.05400                                              |MATRIX WAS PRODUCED BY A KYST ANALYSIS 
     2    .33200   -.15600                                              |OF A LOWER HALF MATRIX.
     3    .17400   -.01800                                              | 
     4  -1.42900    .08300                                              | 
     5    .32300    .50600                                              | 
     6  -1.40200   -.51200                                              | 
     7    .49300   -.56500                                              | 
     8  -1.14200    .49800                                              | 
     9    .71200    .38100                                              | 
    10   1.15400   -.36700                                              | 
    11   1.29000    .20500                                              | 
                                                                        | 
 VARIABLE FORMAT  (SCALE VALUES) = (5X,11F2.0)                          | 
 PHASE  2                                                               |PHASE 2 MODEL:  ELLIPTICAL IDEAL POINT
                                                                        |EACH SUBJECT REFLECTS DIFFERENTIAL
X MATRIX, (INPUT CONFIGURATION AFTER NORMALIZATION)                     |WEIGHTING OF THE DIMENSIONS ONLY.
   1      -.5040       .3320       .1740     -1.4290       .3230     -1.|    
           .4930     -1.1420        .7120      1.1540      1.2900       |X MATRIX=ORIGINAL CONFIGURATION MATRIX
   2      -.0540      -.1560      -.0180       .0830       .5060      -.| AFTER NORMALIZATION                 
          -.5650       .4980        .3810      -.3670       .2050       |ITERATION PROCESS BEGINS, SOLVING FOR
                                                                        |IDEAL POINT COORD. FOR EACH SUBJECT 
 PHASE  2                                                               | 
 SUBJECT   1                                                            | 
 SCALE VALUES BEFORE NORMALIZATION FOR SUBJECT    1                     | 
        2.00000     10.00000      7.00000     11.00000      3.00000     | 
        4.00000      8.00000      1.00000      5.00000      6.00000     | 
                                                                        |S = NORMALIZED SUBJECT PREFERENCE DATA:
 S (VECTOR OF SCALE VALUES, E.G. PREFERENCES)                           |EXAMPLE FOR SUBJECT 1: 
        -.36614       .31940       .06232       .40509      -.28045     |NORMALIZED      ORIGINAL RANKS 
        -.19475       .14801      -.45183      -.10906      -.02337     |-.45183     =         1 
 BEGIN ITERATION ON MONOTONE FIT                                        |-.36614     =         2 
                                                                        |   .        =         . 
 ITERATION  1                                                           |   .        =         . 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE|+.31940     =        10 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               |+.40509     =        11 
        -.06728      -.14567      -.22842       .14894      -.47362     | 
                                                                        |ITERATION TO DETERMINE SUBJECT 1'S 
 (CORRELATION)=       .75768                                            | DISTANCE FROM STIMULI
                                                                        | 
 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            |BETA VALUES ARE IMPORTANCES (WEIGHTS)
          .13028       .00024      -.00880       .49816      -.26028    |ON EACH NEW OR ROTATED DIMENSION.
         -.04967       .13746      -.17592       .05835       .00128    |NEGATIVE WEIGHTS INDICATE AN ANTI-
                                                                        |IDEAL OR LEAST PREFERRED ITEM IN TERMS
 ITERATION  2                                                           |OF THAT PARTICULAR DIMENSION. 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE|
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               |BETA VALUES ARE THE UNSTANDARDIZED
        -.12995      -.17341      -.23621       .20145      -.34357     |REGRESSION COEFFICIENTS FOR THE SCALE
                                                                        |VALUES ON THE STIMULUS COORDINATES.
                                                                        |THE FIRST BETA VALUE (-.13541) IS THE
                                                                        |INTERCEPT .  THE SIGN OF THE LAST BETA
                                                                        |VALUE (-.27486) SHOWS WHETHER THE IDEAL
                                                                        |POINT IS POSITIVE OR NEGATIVE.  THE 
                                                                        |INTERMEDIATE AND LAST BETA VALUES SHOW
                                                                        |THE IDEAL POINT COORDINATES.

 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            |
         .02038      -.13682      -.14988       .50725      -.37243     | 
        -.14269       .12797      -.29115      -.02137      -.08127     |SIGNED DSQ = SQUARES OF DISTANCE FROM
                                                                        |THE IDEAL POINT TO EACH STIMULUS
 ITERATION  3                                                           | 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE| 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               | 
        -.13541      -.17683      -.24129       .19738      -.27486     | 
                                                                        | 
 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            | 
         .01608      -.14141      -.15595       .49841      -.36440     | 
        -.12603       .13561      -.29309      -.02510      -.09609     | 
 END OF ITERATION, REACHED CRITERION                                    | 
                                                                        | 
 (CORRELATION)=       .92762                                            | 
                                                                        | 
 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            | 
         .13814      -.01935      -.03389       .62048      -.24234     | 
        -.00397       .25768      -.17103       .09697       .02597     | 
                                                                        | 
 ***********************************************************************| 
                                                                        |COORDINATES OF THE IDEAL POINT IN
 SUBJECT     1                                                          |THE OLD GROUP STIMULUS SPACE. 
                                                                        | 
 COORDINATES OF IDEAL POINT WITH RESPECT TO OLD AXES                    |COORDINATES OF IDEAL POINT IN THE NEW 
              .44796        -.43893                                     |INDIVIDUAL SPACE.  THESE ARE THE SAME 
                                                                        |AS THE LAST BETA.
 IMPORTANCES OF NEW AXES                                                |IN PHASE 2 ANALYSIS, ONLY THE SIGN 
              .19738        -.27486                                     |IS IMPORTANT.  IF +, THE IDEAL IS 
                                                                        |POSITIVE, IF -, THE IDEAL IS NEGATIVE.
 ***********************************************************************|
 SUBJECT   2                                                            |THE SIZE OF THE NUMBER DOES NOT  
   .       .                                                            |INFLUENCE THE DIMENSIONAL WEIGHTS IN 
   .       .                                                            |THE PHASE 2 ANALYSIS.
   .       .                                                            | 
 SUBJECT   5                                                            | 
 ***********************************************************************| 
 AVERAGE SUBJECT                                                        |S = NORMALIZED PREFERENCES FOR THE
                                                                        |AVERAGE SUBJECT.  FOR THIS SAMPLE,
 S (VECTOR OF SCALE VALUES, E.G. PREFERENCES)                           |    11 9 10 2 7 1 4 3 5 6 8 
        -.17416      -.14835      -.15716       .26763      -.10830     |IS THE AVERAGE SUBJECT'S PREFERENCE 
        -.01852       .25334      -.09050      -.10212      -.12932     |ORDER. 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE| 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               | 
        -.18166      -.12423      -.02234       .12500       .55753     | 
                                                                        | 
 (CORRELATION)=       .98120                                            | 
                                                                        | 
 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            | 
         .12829       .02068       .01384       .46587       .13545     | 
         .19083       .46313       .07842       .13749       .09770     | 
                                                                        | 

************************************************************************| 
 SUBJECT     6                                                          | 
 COORDINATES OF IDEAL POINT WITH RESPECT TO OLD AXES                    | 
              .49692         .02004                                     | 
                                                                        | 
 IMPORTANCES OF NEW AXES                                                | 
              .12500         .55753                                     | 
 ***********************************************************************| 
                                                                        | 
 PHASE  3                                                               | 
                                                                        | 
 X MATRIX, (INPUT CONFIGURATION AFTER NORMALIZATION)                    | 
   1      -.1782       .1174       .0615      -.5052       .1142      -.| 
           .1743      -.4038       .2517       .4080       .4561        | 
   2      -.0403      -.1165      -.0134       .0620       .3778      -.| 
          -.4219       .3718       .2845      -.2740       .1531        | 
                                                                        | 
 PHASE  3                                                               | 
 SUBJECT   1                                                            | 
                                                                        | 
 S (VECTOR OF SCALE VALUES, E.G. PREFERENCES)                           | 
        -.20230      -.00337      -.10730       .57895      -.20230     | 
        -.14639      -.00337      -.34045      -.10730      -.10730     | 
 BEGIN ITERATION ON MONOTONE FIT                                        | 
                                                                        | 
 ITERATION  1                                                           | 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE| 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               | 
        -.15549      -.56457      -.32391       .85609                  | 
                                                                        | 
 (CORRELATION)=       .86617                                            | 
                                                                        | 
 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            | 
         .26596       .11859       .09673       .61070       .07023     | 
         .34033       .48916       .01298       .18893       .01478     | 
                                                                        | 
 ITERATION  2                                                           | 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE| 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               | 
        -.15368      -.59209      -.34717       .84614                  | 
                                                                        | 
 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            | 
        -.00593      -.15960      -.18208       .34319      -.22064     | 
         .06588       .21123      -.27939      -.09572      -.28103     | 
 END OF ITERATION, REACHED CRITERION                                    | 
                                                                        | 
 (CORRELATION)=       .89734                                            | 
                                                                        | 
 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            | 
         .28694       .13327       .11079       .63606       .07222     | 
         .35875       .50410       .01348       .19715       .01184     | 
                                                                        | 

 ***********************************************************************| 
 SUBJECT     1                                                          | 
 COORDINATES OF IDEAL POINT WITH RESPECT TO OLD AXES                    | 
              .34988         .20515                                     | 
                                                                        | 
 IMPORTANCES OF NEW AXES                                                | 
              .84614         .84614                                     | 
 ***********************************************************************| 
 SUBJECT   2                                                            | 
   .       .                                                            | 
   .       .                                                            | 
   .       .                                                            | 
 SUBJECT   5                                                            | 
 ***********************************************************************| 
 AVERAGE SUBJECT                                                        |THE BETA VALUES ARE HERE DIMENSION
                                                                        |WEIGHTS.  THE LAST BETA WEIGHT IS
 S (VECTOR OF SCALE VALUES, E.G. PREFERENCES)                           |REPEATED BELOW AS IMPORTANCES OF 
        -.17419      -.14698      -.17851       .28420      -.09283     |THE AXES.
        -.07514       .25897      -.09655      -.11103      -.11926     | 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE| 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               | 
        -.19387      -.37799      -.00314      1.06721                  | 
                                                                        | 
 (CORRELATION)=       .98137                                            | 
                                                                        | 
 SIGNED DSQ, (SIGNED DISTANCE SQUARED FROM STIMULI TO IDEAL)            | 
         .13657       .01865       .01449       .50077       .15538     | 
         .19127       .50646       .09142       .13791       .10760     | 
                                                                        | 
 ***********************************************************************| 
 SUBJECT     6                                                          | 
                                                                        | 
 COORDINATES OF IDEAL POINT WITH RESPECT TO OLD AXES                    | 
              .17709         .00147                                     | 
                                                                        | 
 IMPORTANCES OF NEW AXES                                                | 
             1.06721        1.06721                                     | 
 ***********************************************************************| 
 STIMULI COORDINATES                                                    | 
 DIMENSION        1              2                                      | 
 STIMULI                                                                | 
   1            -.17819        -.04032                                  | 
   2             .11738        -.11648                                  | 
   3             .06152        -.01344                                  | 
   4            -.50524         .06197                                  | 
   5             .11420         .37782                                  | 
   6            -.49569        -.38230                                  | 
   7             .17430        -.42187                                  | 
   8            -.40376         .37185                                  | 
   9             .25173         .28448                                  | 
  10             .40801        -.27403                                  | 
  11             .45609         .15307                                  | 


 COORDINATES OF IDEAL POINTS                                            | 
 DIMENSION        1              2                                      | 
 SUBJECTS                                                               | 
   1             .34988         .20515                                  | 
   2            -.08728         .00457                                  | 
   3            -.49957        -.01849                                  | 
   4            -.00849        -.04644                                  | 
   5            1.31268        -.21980                                  | 
   6             .17709         .00147                                  | 
 SUBJECT   6 IS THE AVERAGE SUBJECT                                     |WEIGHTS OF AXES SHOW THE IMPORTANCE
 WEIGHTS OF AXES                                                        |OF NEW AXES AND IF THE IDEAL POINT
 DIMENSION        1              2                                      |IS + OR - . THIS IS A REPEAT OF THE
   1             .84614         .84614                                  |FINAL BETA LISTED FOR EACH SUBJECT.
   2            2.42537        2.42537                                  |THE SIZE OF THE BETA SHOWS IMPORTANCE
   3            -.92105        -.92105                                  |OF THE DIMENSION AS A COMPONENT OF 
   4            2.62741        2.62741                                  |PREFERENCE.  A SMALL WEIGHT MEANS
   5             .32343         .32343                                  |THAT LARGER CHANGES IN THE DIMENSION
   6            1.06721        1.06721                                  |CAN BE MADE WITHOUT INFLUENCING
 SUBJECT   6 IS THE AVERAGE SUBJECT                                     |PREFERENCE.
 STIMULI AND IDEAL POINTS:                                              |
          .*....*....*....*....*....*....*....*....*....*....*....*....*|THE PREFERENCE CONTOURS AROUND
     1.50**                              |                             *|THE POINTS ARE HYPERBOLIC AND NOT
     1.38**                              |                             *|ELLIPTICAL.  THEREFORE, THE MIXED
     1.27**                              |                             *|+ - IDEAL POSITIONS ARE DIFFICULT 
     1.15**                              |                             *|TO INTERPRET. 
     1.04**                              |                             *| 
      .92**                              |                             *| 
      .81**                              |                             *| 
      .69**                              |                             *| 
      .58**                              |                             *| 
      .46**                              |                             *| 
      .35**                        8     | 5                           *| 
      .23**                              |   9C                        *| 
      .12**                      4       |      B                      *| NOTE THAT POINT E IS A NEGATIVE
      .00**-----------------------E---1-DF3-H--------------------------*| IDEAL POINT
     -.12**                              | 2                           *| 
     -.23**                              |     A             G         *| 
     -.35**                       6      |                             *| 
     -.46**                              |  7                          *| 
     -.58**                              |                             *| 
     -.69**                              |                             *| 
     -.81**                              |                             *| 
     -.92**                              |                             *| 
    -1.04**                              |                             *| 
    -1.15**                              |                             *| 
    -1.27**                              |                             *| 
    -1.38**                              |                             *| 
    -1.50**                              |                             *| 
          .*....*....*....*....*....*....*....*....*....*....*....*....*| 
          .   -1.6667.  -1.0000.   -.3333.    .3333.   1.0000.   1.6667.| 
         -2.0000   -1.3333    -.6667     .0000     .6667    1.3333    2.| 


 PHASE  4                                                               |PHASE 4 IS THE MOST GENERAL MODEL OF
                                                                        |PREFMAP, PROVIDING NEGATIVE IDEAL
 X MATRIX, (INPUT CONFIGURATION AFTER NORMALIZATION)                    |AND DIFFERENTIAL WEIGHTING OF 
   1      -.1782       .1174       .0615      -.5052       .1142      -.|DIMENSIONS.  ELLIPTICAL PREFERENCE 
           .1743      -.4038       .2517       .4080       .4561        |CONTOURS AROUND THE IDEAL POINTS ARE
   2      -.0403      -.1165      -.0134       .0620       .3778      -.|CREATED. ROTATION IS PRESENT
          -.4219       .3718       .2845      -.2740       .1531        | 
                                                                        | 
 PHASE  4                                                               | 
 SUBJECT   1                                                            | 
 S (VECTOR OF SCALE VALUES, E.G. PREFERENCES)                           | 
         .14893      -.03206       .14893      -.42620       .14893     | 
         .14893      -.03206       .34697       .14893       .14893     | 
 BEGIN ITERATION ON MONOTONE FIT                                        | 
                                                                        | 
 ITERATION  1                                                           | 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE| 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               | 
        -.00005       .70151       .39606                               | 
                                                                        | 
 (CORRELATION)=       .84496                                            | 
                                                                        | 
 PROJECTIONS ON THE FITTED VECTOR                                       | 
        -.17499       .04495       .04696      -.40949       .28519     | 
        -.05562      -.16879       .35907       .22057       .47242     | 
                                                                        | 
 ITERATION  2                                                           | 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE| 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               | 
        -.00005       .72036       .37021                               | 
                                                                        | 
 PROJECTIONS ON THE FITTED VECTOR                                       | 
        -.14334       .04139       .03929      -.34106       .22209     | 
        -.03067      -.15324       .28661       .19242       .38517     | 
 END OF ITERATION, REACHED CRITERION                                    | 
                                                                        | 
 (CORRELATION)=       .85443                                            | 
                                                                        | 
 PROJECTIONS ON THE FITTED VECTOR                                       | 
        -.17692       .05116       .04857      -.42104       .27427     | 
        -.03780      -.18915       .35393       .23763       .47562     | 
                                                                        | 
 SUBJECT   2                                                            | 
   .       .                                                            | 
   .       .                                                            | 
   .       .                                                            | 
 SUBJECT   5                                                            | 
                                                                        | 
 END OF ITERATION, REACHED CRITERION                                    | 
                                                                        | 
 (CORRELATION)=       .98461                                            | 
                                                                        | 

 PROJECTIONS ON THE FITTED VECTOR                                       | 
        -.17078       .13255       .06280      -.50893       .06012     | 
         .23172      -.45190       .20937       .44239       .43013     | 
 AVERAGE SUBJECT                                                        | 
                                                                        | 
 S (VECTOR OF SCALE VALUES, E.G. PREFERENCES)                           | 
         .10573       .09787       .11067      -.22826       .14746     | 
         .06973      -.29308       .13753       .13116       .13495     | 
 BETA VALUES  (IN THE MOST GENERAL CASE THERE ARE (2K + K(K-1)/2 + 1) TE| 
 QUADRATIC, LINEAR, THEN A CONSTANT TERM)                               | 
        -.00002       .51639       .08465                               | 
                                                                        | 
 (CORRELATION)=       .87549                                            | 
                                                                        | 
 PROJECTIONS ON THE FITTED VECTOR                                       | 
        -.18237       .09699       .05853      -.48856       .17382     | 
         .10376      -.33829       .29444       .35830       .47485     | 
                                                                        | 
 STIMULI COORDINATES                                                    |2 DIMENSIONAL COORDINATES OF 11
 DIMENSION        1              2                                      | STIMULI
 STIMULI                                                                | 
   1            -.17819        -.04032                                  | 
   2             .11738        -.11648                                  | 
   3             .06152        -.01344                                  | 
   4            -.50524         .06197                                  | 
   5             .11420         .37782                                  | 
   6            -.49569        -.38230                                  | 
   7             .17430        -.42187                                  | 
   8            -.40376         .37185                                  | 
   9             .25173         .28448                                  | 
  10             .40801        -.27403                                  | 
  11             .45609         .15307                                  | 
                                                                        |  


     STIMULI AND IDEAL POINTS:                                          |POINTS 1 - 9, A, B = STIMULI 
          .*....*....*....*....*....*....*....*....*....*....*....*....*|POINTS C - H = IDEAL POINTS 
     1.50**                              |                             *| 
     1.38**                              |                             *| 
     1.27**                              |                             *| 
     1.15**                              |                             *| 
     1.04**                              |                             *| 
      .92**                              |                             *| 
      .81**                              |                             *| 
      .69**                              |                             *| 
      .58**                              |                             *| 
      .46**                              |            C                *| 
      .35**                        8     | 5           E               *| 
      .23**                              |   9                         *| 
      .12**                      4       |      B       H              *| 
      .00**---------------------------1--03----------------------------*| 
     -.12**                              | 2            ;              *| 
     -.23**               D              |     A                       *| 
     -.35**                       6      |                             *| 
     -.46**                              |  7                          *| 
     -.58**                              |                             *| 
     -.69**                              |                             *| 
     -.81**                              |                             *| 
     -.92**                              |                             *| 
    -1.04**                              |                             *| 
    -1.15**                              |                             *| 
    -1.27**                              |                             *| 
    -1.38**                              |                             *| 
    -1.50**                              |                             *| 
          .*....*....*....*....*....*....*....*....*....*....*....*....*| 
          .   -1.6667.  -1.0000.   -.3333.    .3333.   1.0000.   1.6667.| 
         -2.0000   -1.3333    -.6667     .0000     .6667    1.3333    2.| 
                                                                        | 
 DIRECTION COSINES OF FITTED SUBJECT VECTORS                            |DIRECTION COSINES: 
                      DIMENSION                                         |IF THE VALUE IS > .9, LITTLE ROTATION
 SUBJECT       1        2                                               |WILL OCCUR BECAUSE POINTS ARE ALREADY
       1       .8894    .4571                                           |HIGHLY CORRELATED WITH THE DIMENSIONS.
       2      -.9818   -.1899                                           |IF THIS OCCURS, THE PHASE 3 ANALYSIS 
       3       .9270    .3750                                           |IS AN APPROPRIATE PLACE TO START 
       4       .9915   -.1302                                           |(COMPLETE ANALYSIS FOR PHASES  1-3)
       5       .9901   -.1402                                           | 
      AVG R   .9868    .1618                                            | 
         CORRELATION (PHASE)                  F RATIO (PHASE)           |WITHIN PHASE ANALYSIS: 
         R1       R2       R3       R4       F1       F2       F3       | 
 DF                                          5 5      4 6      3 7      | 
 SUBJ                                                                   | 
  1       .000     .928     .897     .854     .000    9.252    9.646   1| 
  2       .000     .888     .888     .498     .000    5.583    8.675    | 
  3       .000     .979     .936     .928     .000   35.316   16.362   2|MODELS 2,4,3 FIT DATA FOR SUBJECT 3
  4       .000     .997     .998     .670     .000  249.496  653.733    |MODELS 3,2 FIT SUBJECT 4  
  5       .000     .991     .987     .985     .000   81.634   88.161  12|=> MODELS 4,3,2 FIT SUBJECT 5 
 AVG      .000     .981     .981     .875     .000   38.781   60.866   1|=> AVERAGE SUBJECT FOR PHASE 3 MODEL IS SIGNIFICANT. 

         F RATIO (BETWEEN PHASE)                                        | 
         F12      F13      F14      F23      F24      F34               | 
 DF      1 5      2 5      3 5      1 6      2 6      1 7               |BETWEEN PHASE ANALYSIS: 
 SUBJ                                                                   | 
  1       .000     .000     .000    2.377    2.805    2.702             |=> NO DIFFERENCE FOR SUBJECT 1 
  2       .000     .000     .000     .006    7.652   17.832             |MODEL 3 IS BETTER THAN 4 FOR SUB 2
  3       .000     .000     .000   12.380    7.204     .772             |MODEL 2 IS BETTER THAN 3 FOR SUB 3
  4       .000     .000     .000   -2.429  273.937 1078.809             |MODEL 3 IS BETTER THAN 4 FOR SUB 4
  5       .000     .000     .000    2.574    2.079    1.293             |NO MODEL PREDICTS WELL FOR SUB 5
  AVG     .000     .000     .000    -.051   15.813   37.276             |AVERAGE SUB MOVING FROM PH. 3 TO 4
                                                                        |IS NEARLY SIGNIFICANT.  2 TO 4 IS
 ROOT MEAN SQUARE                                                       |SIGNIFICANT ALSO.
 PHASE                                                                  | 
  1       .000                                                          | 
  2       .957                                                          | 
  3       .942                                                          | 
  4       .807                                                          | 
                                                                        | 
                                                                        | 
     AN F - VALUE OF 1000.0 IN THE ABOVE TABLE INDICATES                | 
     A POSSIBLE DIVISION BY ZERO. I.E. R IS VERY CLOSE TO 1.00          |