How to Use INDSCAL, a Computer Program for Canonical Decomposition of N-way Tables and Individual Differences in Multidimensional Scaling (1)

J. Douglas Carroll and Jih Jie Chang
Bell Telephone Laboratories


NON-TECHNICAL INTRODUCTION

INDSCAL is a program designed for the analysis of individual differences for two or more subjects. INDSCAL is generally not used to analyze data for a single subject. INDSCAL requires as input, a separate data matrix for each subject or group of subjects. Broadly defined, a subject may be an actual individual, or more often, a group of individuals aggregated together because of some common characteristic.

INDSCAL performs two types of analysis: an analysis of individual differences and a canonical decomposition analysis. The analysis of individual differences identifies weights that each subject uses to evaluate the stimuli. The stimuli are identified in terms of a set of underlying dimensions that are common to all subjects. In the individual differences analysis, the canonical decomposition analysis is used to identify the perceptual dimensions underlying the stimulus space.

It should be noted that a separate canonical decomposition analysis may be performed. In this case, rectangular data matrices may be analyzed, identifying the underlying dimensions of each of the n dimensional input matrices.

When performing individual differences analysis, INDSCAL input data is generally of some type of similarity or dissimilarity data. The analysis may be conducted on lower half or full symmetric matrices. The lower half matrices may be either similarity, dissimilarity, Euclidean distance, correlation, or covariance measures. The full matrices may be read in as either similarity or dissimilarity measures.

INDSCAL performs an n way analysis. The n=three way analysis is by far the most common. In this situation, n2 and n3 define the n2 by n3 matrix of similarity judgments, and n1 defines the number of slices in this 3 dimensional matrix. Slices are here defined as respondents or respondent groups.

The INDSCAL solution identifies the underlying dimensions common to the stimuli. Importances of these dimensions are reported. Plots of the importance weights used by the individuals in evaluating the stimuli are produced. Also produced are plots of the stimuli in the underlying dimensional space. Plots for the individuals are not produced, but the plot coordinates for the individuals are easy to compute and could be easily graphed in a program such as Lotus 1-2-3 or Harvard Graphics.


TECHNICAL INTRODUCTION

INDSCAL is a computer program capable of performing two kinds of analyses, namely, CANDECOMP analysis and INDIFF analysis.

Before going into a more detailed description, it is essential to clarify the relationship of these two analyses.  CANDECOMP is used for the more general analysis called 'Canonical decomposition of N-way tables,' while INDIFF performs the analysis of individual differences in multidimensional scaling described by Carroll and Chang (1968).  Nevertheless, CANDECOMP actually forms the core of INDIFF analysis.  By using the appropriate options provided in the program, the user is able to use INDSCAL to do either CANDECOMP or INDIFF analysis.

In the ensuing paragraphs, there will be no mention of their specific names when the description applies to both analyses.

CANDECOMP is a method for decomposing arbitrary N-way tables into a kind of product of N matrices of appropriate dimensions.  The N-way tables are assumed to have n1, n2,...nN components respectively, where N is less than or equal to 7 (as presently programmed).  The N matrices, into which this N-way table is decomposed, will be of the order r x n1, r x n2,..., r x nN, respectively, where r is the common dimensionality of the several spaces defined by the matrices CANDECOMP takes as input rectangular data matrices.

INDIFF uses the CANDECOMP method to do individual differences analysis.  In this case the input may be the square matrices of similarities, dissimilarities, Euclidean distances or scalar product matrices.  If any one of the first three serves as input, an early step is to convert the data to scalar products.

INDIFF treats the first of the N matrices into which the N-way table is decomposed as the subjects' weight matrix and matrices 2 and 3 as stimulus matrices.  Matrices 2 and 3 should asymptotically be equal.  In practice matrix 2 is set equal to matrix 3 as a last step and the other matrices (1, 4,...N) are recomputed accordingly.  The output of INDSCAL consists of the N matrices resulting from the analysis.  In addition, plots for each pair of the dimensions in each matrix are generated showing the ni points (where i=1,...N) in r dimensions.


INPUT DATA STRUCTURE

1. In the case of CANDECOMP analysis

Let n1, n2,...nN be respectively the number of components in the N-way tables, where N is less than or equal to 7.  The data can be viewed as many rectangular matrices (slices) each of order n2 (rows) by n3 (columns).  A 3-way table would correspond to a stack of n1 such matrices (slices).  A 4-way table would then be composed of n4 such 3-way tables as units.  In the same manner we can construct data for higher way tables.

The n2 by n3 slices described above are read in as follows:  The first data line (or lines) has n3 values corresponding to the first row of the first rectangular matrix (slice) in the first n1 unit,. this is followed by the second row and down to the n2th row.  Each matrix is followed by the next without interruption.



2. In the case of INDIFF analysis

The data structure for INDIFF is basically the same as in CANDECOMP.  However, in the INDIFF case n2 and n3 are ordinarily assumed to be identical, that is every two-way matrix derived by holding n1 and n4...nN constant is symmetric.

The n2 by n3 matrix may initially be similarity, dissimilarity or Euclidean distance matrix.  They can be entered in the form of a full symmetric matrix or a lower half matrix without diagonal.  If so there are options for converting them to scalar product matrices, by equations given in Torgerson (1958) as referred to in Carroll and Chang (1968).

In the most common situation, say a 3-way analysis, the n1 matrices (each of n2 rows and n3 columns) will correspond to individuals, and these matrices will be read in one at a time.


INDSCAL Options

Initial Configuration

The analysis starts with (N-1) sets of matrices.  That is the program requires that the initial matrices 2 through N be either supplied by the user or generated in the program.  If to be read in, the Ith matrix must be entered as an r x nI matrix, where r is the common dimensionality specified by the user and I=2, 3,...N.

Option on Method of Analysis

Instead of solving for several dimensions simultaneously, the user may choose the option of doing separate one dimensional analyses, then combining them to form a single r dimensional analysis.  While this results in a less general solution than does the more usual solution, it assures certain orthogonality properties for successive components.

Solution for Weight Matrices for Fixed Stimulus Matrices (in INDIFF analysis)

This option is used when the coordinates of N stimulus points in K dimensions have been obtained, say from some multidimensional scaling procedure and the user would like to find the subjects' weights for this fixed stimulus space.

The program reads in the initial matrices, proceeds with the iterative procedure solving for the remaining matrices while keeping matrices 2 and 3 unchanged throughout the analysis.  In the case of a 3-way analysis, this procedure will 'converge' in one iteration,. in higher way cases more iterations will be required.



Equating The Stimulus Matrices (in INDIFF analysis)

At the end of the iterative procedure, the program provides the option of equating the 2 stimulus matrices, matrices 2 and 3.  It is essential that in the solution space matrix 2 should be identical to 3 if the input data is in the form of a lower half or full symmetric matrix.

After equating matrices 2 and 3, the iterative procedure continues, but matrices 2 and 3 are kept fixed while the remaining tables are estimated by the iterative procedure (this is similar to the option in 3 for estimating) weight matrices with a fixed stimulus matrix, except that the fixed matrix is an internally computed one rather than one provided as input by the user).

Separate Solutions in Spaces of Different Dimensions.

On the parameter line (see Section IV) if MAXDIM is set to 4 and MINDIM to 1, the program will compute successively in spaces of dimensions 4, 3, 2 and 1.  The initial configuration for each successive dimensions is taken from the solution of the previous computation.


INDSCAL Parameters

PARAMETERS:

N, MAXDIM, MINDIM, IRDATA, ITMAX, ISET, IOY, IDR, ISAM, IPUNSP, IRN, IVEC, IP, IA, IS, CRIT

The parameter variables are read as FREE FORMAT .

N - number of ways (N <_ 7)

MAXDIM - Maximum number of dimensions (Max. = 10)

MINDIM - Minimum number of dimensions (Min. = 1)

(MAXDIM must be <_ 10 or (MAXDIM * Max. N1) <_ 4000)

IRDATA -


0, read in an arbitrary N-way table as described
1, read in lower half similarity matrices without diagonals.
2, read in lower half dissimilarity matrices without diagonals.
3, read in lower half Euclidean distance matrices without diagonals.
4, read in lower half correlation matrices without diagonals.
5, read in lower half covariance matrices with diagonals.
6, read in full symmetric similarity matrices.  The diagonals will be ignored by the program.
7, read in full symmetric dissimilarity matrices.  The diagonals will be ignored.

IRDATA causes scalar products are computed except for options 4 and 5. Additive Constants are estimated except for options 3, 4 and 5.

ITMAX - Maximum number of iterations.  (Max. = 50). Usually from 15 to 20 iterations is sufficient.

ISET

1, at the end of the iterative procedure, set matrix 2 equal to matrix 3, proceed to iterate again but keeping matrices 2 and 3 constant.
0, do not set matrix 2 equal to matrix 3.

IOY

1, compute all dimensions simultaneously.
0, do separate one dimensional solutions.

IDR

1, compute correlations between data and solution for each n2 by n3 matrix.
0, do not compute correlations

ISAM

1, keep matrices 2 and 3 unchanged and solve for the remaining matrices.
0, solve for all matrices.

IPUNSP

1, output to a file the scalar product matrices.
0, do not output the scalar product matrices.

IRN

0, read in the initial matrices 2 to N.
An eight digit odd integer for generating the random initial matrices. This number must be enclosed in apostrophes '13577539' . Note that the results are sensitive to changes in this parameter.

IVEC

0, a standard data matrix is to be read in (preferred option).
1, the program will read Tricon type output.

IP

0, do not output normalized A matrices to a disk file.
1, yes, output normalized A matrices for a disk file.

IA

0, do not output original data to the output file.
1, yes, output original data to the output file.

IS

0, do not print intermediate matrices (preferred).
1, yes, print intermediate matrices.

CRIT

The decimal value criterion for terminating iterations. If [y-y(I-1)]2 - [y-y(I)]2 < CRIT iteration stops, where y(I) stands for estimated y's on Ith iteration.

Line 2. NWT(1), NWT(2)...NWT(N) are read in FREE format.

NWT is number of components or objects within modes.

NWTN(I) is the name used in the program for ni, i=1,...N ni < = 64000 where n1 must be < = 100, n2 and n3 must be < = 45.

Line 3. The format for reading in data.

Line 4. Data set, in the form specified in option IRDATA.

Line 5. This set of commands is present only if IRN=0.

Required are:

a. The format for reading in the initial matrices.

b. Initial matrices - matrices 2 to N.  Each matrix must be entered as a dimension by ni matrix where i=2,3,...N. 

Read in matrix 2 first then followed by 3 and up to matrix N.  Note that if matrix 2 is identical to matrix 3, read in matrix 2 twice followed by matrix 4...N.


REFERENCE

1. Carroll, J.D. and Chang, J.J.  "A New Method for Dealing with Individual Difference in Multidimensional Scaling."  Unpublished report, Bell Telephone Laboratories, 1968.


NOTES

1. This paper was modified for the PC-MDS version of INDSCAL by Scott M. Smith


                   INDSCAL SAMPLE DATA

 3 3 2 2 25 1 0 1 0 0 '12345677' 0 0 1 0 .001
  10  10  10           
(2X,9F2.0)             
 0116                  
018147                 
01563271               
0187684471             
016035219834           
01849498579999         
0150877973199245       
019925539852179984     
01169290837944241898   
0209                   
029070                 
02876506               
0287778383             
023379258939           
02868699229040         
0281305788693997       
027420947805819288     
02232672940276812005   
0349                   
039696                 
03979294               
0368129093             
037744889026           
03979394259349         
0354769294202493       
034748929435189423     
03214790926867875515   
0423                   
049951                 
04992378               
0490162249             
047455509913           
04148877755070         
0425954899997999       
046036692421539999     
04008972817771745171   
0562                   
057716                 
05981455               
0576224047             
058416168107           
05178036936090         
0576938680943619       
057420163805180671     
05107278929286160299   
0685 
068215 
06972856 
0651313643 
067927078207 
06138438877682 
0682997368804020 
066924302716122880 
06158078907266170595 
0710 
075375 
07999999 
0787276599 
076066729999 
07969990109075 
0798999198883499 
077315909909569575 
07546284999553859149 
0814 
086147 
08799677 
0872211273 
086612288113 
08666475417182 
0851673293496686 
080720677115567669 
08195106882581500883 
0911 
099069 
09722690 
0993176924 
093934369880 
09268277855399 
0980747599938713 
097308913517179991 
09246290768564772465 
1069 
106358 
10768579 
1052145181 
106139358336 
10809093067885 
1028878394644490 
108020929851238033 
10782840993671826213
INDSCAL SAMPLE OUTPUT


                                  I N D S C A L                         |
                         INDIVIDUAL  DIFFERENCES  SCALING               |3 3 2 2 25 1 0 1 0 0 '12345677' 0 0 1 0 .001
                      BY DR. J. D. CARROLL AND JIH JIE CHANG            |  10  10  10 
                                  PC-MDS VERSION                        |(2X,9F2.0) 
                                                                        |0116 
 ANALYSIS TITLE: SCHIFFMAN COLA DATA PP. 33-34                          |018147 
 DATA IS READ FROM FILE: SRY.DAT                                        |01563271 
 OUTPUT FILE IS: TEST2.PRN                                              |0187684471 
                                                                        |016035219834 
 INDIFF- INDIVIDUAL DIFFERENCES ANALYSIS USING CANONICAL DECOMPOSITION  |01849498579999 
 OF   3 WAY TABLE IN   3 DIMENSIONS                                     |0150877973199245 
                                                                        |019925539852179984 
 TITLE: SCHIFFMAN COLA DATA PP. 33-34                                   |01169290837944241898 
 PARAMETERS                                                             |0209 
 N        NO. OF STIMULI                                       3        |029070 
 NF       DIMENSION OF SOLUTION                                3        |02876506 
 MAXDIM   MAXIMUM NO. OF DIMENSIONS                            3        |0287778383 
 MINDIM   MINIMUM NO. OF DIMENSIONS                            2        |023379258939 
 IRDATA   TYPE OF DATA INPUT                                   2        |02868699229040 
 ITMAX    MAXIMUM NO. OF ITERATIONS                           25        |0281305788693997 
 ISET     OPTION TO SET MATRIX 2 EQUAL TO MATRIX 3             1        |027420947805819288 
 IOY      SELECT SIMULTANEOUS SOLUTION                         0        |02232672940276812005 
 IDR      CORRELATIONS FOR EACH SUBJECT                        1        |0349 
 ISAM     SOLVE FOR ALL MATRICES                               0        |039696 
 IPUNSP   PRINT SCALAR PRODUCT MATRICES                        0        |03979294 
 IRN      RANDOM NUMBER GENERATOR START SET             12345677        |0368129093 
 CRIT     CRITERIA FOR QUITTING ITERATION                      .001     |037744889026 
 IVEC     MATRIX OR VECTOR FORM FOR DATA                       0        |03979394259349 
 IP       OUTPUT NORMALIZED A-MATRIX                           0        |0354769294202493 
 IA       PRINT ORIGINAL DATA MATRICES                         1        |034748929435189423 
 IS       PRINT INTERMEDIATE ITERATIVE MATRICES                0        |03214790926867875515 
                                                                        |0423 
 MATRIX SIZES    10  10  10                                             |049951 
 ***********************************************************************|04992378 
    *****IDENTIFICATION KEY FOR PLOTS WITH IDENTIFIED POINTS*****       |0490162249 
 PT #   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15       |047455509913 
 CHAR   1   2   3   4   5   6   7   8   9   A   B   C   D   E   F       |04148877755070 
 PT #  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30       |0425954899997999 
 CHAR   G   H   I   J   K   L   M   N   O   P   Q   R   S   T   U       |046036692421539999 
 PT #  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45       |04008972817771745171 
 CHAR   V   W   X   Y   Z   +   /   =   *   &   $   @   [   ?   <       |0562 
 PT #  46  47  48  49  50                                               |057716 
 CHAR   (   )   "   ;   ]                                               |05981455 
 POINT NUMBERS ABOVE 50 IDENTIFIED AS  > MULTIPLE POINTS IDENTIFIED AS  |0576224047 
                                                                        |058416168107 
 INITIAL A MATRICES                                                     |05178036936090 
 MATRIX 1                                                               |0576938680943619 
  1      1.0000      1.0000      1.0000      1.0000      1.0000         |057420163805180671 
         1.0000      1.0000      1.0000      1.0000      1.0000         |05107278929286160299 
  2      1.0000      1.0000      1.0000      1.0000      1.0000         |0685 
         1.0000      1.0000      1.0000      1.0000      1.0000         |068215 
  3      1.0000      1.0000      1.0000      1.0000      1.0000         |06972856 
         1.0000      1.0000      1.0000      1.0000      1.0000         |0651313643 
                                                                        |067927078207 
 MATRIX 2                                                               |06138438877682 
  1       .4257      -.0724      -.1040       .4653      -.1853         |0682997368804020 
         -.3849      -.0541       .3826      -.0469      -.3351         |066924302716122880 
  2       .3026       .1942      -.3516      -.2383       .2954         |06158078907266170595 
          .3221       .3436      -.4229       .1126      -.3603         |0710 
  3       .4448       .3780       .4900       .0394      -.4308         |075375 
         -.2456      -.2815      -.4792      -.4867       .2676         |07999999 
                                                                        |0787276599 
  MATRIX 3                                                               |076066729999 
  1      -.2278      -.4010      -.2592      -.1818       .3562         |07969990109075 
         -.1681       .1906      -.4663      -.3248       .2688         |0798999198883499 
  2       .2047      -.0229      -.2792      -.3818       .3765         |077315909909569575 
          .4745      -.1201      -.0642       .1184       .3631         |07546284999553859149 
  3      -.3375      -.3675      -.2500      -.0406       .3499         |0814 
         -.3976      -.0936       .0471       .3627       .2707         |086147 
                                                                        |08799677 
 HISTORY OF COMPUTATION                                                 |0872211273 
 ITERATION        CORRELATIONS BETWEEN                                  |086612288113 
                   Y(DATA) AND YHAT         (R**2)            (1-R**2)  |08666475417182 
      0               -.049669             .002467             .997533  |0851673293496686 
      1                .454765             .206811             .793189  |080720677115567669 
      2                .568041             .322671             .677329  |08195106882581500883 
      3                .596882             .356268             .643732  |0911 
      4                .646604             .418097             .581903  |099069 
      5                .731775             .535495             .464505  |09722690 
      6                .775493             .601389             .398611  |0993176924 
      7                .787444             .620068             .379932  |093934369880 
      8                .791152             .625921             .374079  |09268277855399 
      9                .792406             .627908             .372092  |0980747599938713 
     10                .792987             .628828             .371172  |097308913517179991 
     11                .793351             .629405             .370595  |09246290768564772465 
     12                .793609             .629815             .370185  |1069 
     13                .793797             .630114             .369886  |106358 
     14                .793933             .630330             .369670  |10768579 
     15                .794032             .630486             .369514  |1052145181 
     16                .794102             .630598             .369402  |106139358336 
     17                .794152             .630678             .369322  |10809093067885 
 ***********************************************************************|1028878394644490 
 EQUATE MATRIX 2 AND MATRIX 3, ITERATE AGAIN                            |108020929851238033 
 INITIAL A MATRICES                                                     |10782840993671826213 
 MATRIX 1                                                               | 
  1      -.0664      -.1137      -.0796      -.1329      -.0729         |VALUES FOR MATRIX 1: CORRELATION BETWEEN 
         -.0829      -.1045      -.1228      -.1263      -.0706         |ORIGINAL (Y DATA) AND COMPUTED 
  2       .0816       .1411       .1785       .0592       .0164         |DISTANCES (Y HAT). 
          .0192       .1566       .1335       .0483       .1652         |VAF = SQUARE OF THE CORRELATION, OR 1-TRV = 
  3      -.1752      -.0518      -.0564      -.1774      -.2176         |      1 - TOTAL RESIDUAL VARIANCE 
         -.2144      -.0534      -.1159      -.1657      -.0566         | 
                                                                        |RESIDUAL VARIANCE = 
 MATRIX 2                                                               | 1 - CORRELATION BETWEEN Y AND Y-HAT 
  1       .4911       .5188      -.8744       .5474       .0105         | 
         -.6039       .1831      -.7128       .4543      -.0141         | 
  2       .2175       .2887      -.1341      -.7295       .1839         | 
          .1180      -.6643       .1534       .3295       .2370         | 
  3      -.5473       .4163       .3463       .2937       .4616         | 
          .3534      -.4958      -.5589       .4388      -.7081         | 
                                                                        |VALUES FOR MATRIX 2:  (SAME AS MATRIX 1) 
 MATRIX 3                                                               | 
  1       .4911       .5188      -.8744       .5474       .0105         | 
         -.6039       .1831      -.7128       .4543      -.0141         | 
  2       .2175       .2887      -.1341      -.7295       .1839         | 
          .1180      -.6643       .1534       .3295       .2370         | 
  3      -.5473       .4163       .3463       .2937       .4616         | 
          .3534      -.4958      -.5589       .4388      -.7081         | 
                                                                        | 
 HISTORY OF COMPUTATION                                                 | 
 ITERATION        CORRELATIONS BETWEEN                                  | 
                   Y(DATA) AND YHAT         (R**2)            (1-R**2)  | 
      0               -.420259             .176618             .823382  | 
      1                .794166             .630700             .369300  | 
                                                                        | 
 
 NORMALIZED A MATRICES                                                  | 
 MATRIX   1                                                             |SUBJECT WEIGHTS FOR EACH DIMENSION 
  1         .64756      .33860      .22328                              |(SUBJECT 1 AND 2).  IF WEIGHTS ARE 
  2         .19180      .58402      .37798                              |NEGATIVE, THEY ARE TREATED AS 0. 
  3         .20780      .73905      .26733                              |IF WEIGHTS ARE LARGE, DIMENSIONALITY 
  4         .65787      .24576      .43888                              |MAY BE TOO LARGE. 
  5         .80606      .06846      .24051                              | 
  6         .79444      .07978      .27314                              |THE SUM OF SQUARES OF WGTS FOR 
  7         .19913      .64866      .34416                              |DIM 1 + DIM 2 + ...+DIM K 
  8         .42905      .55362      .40794                              |SHOWS HOW WELL THE WEIGHTED STIMULUS 
  9         .61510      .20031      .41530                              |FITS THE DATA FOR EACH SUBJECT. 
 10         .20900      .68505      .23409                              |SUM OF SQUARES ? R2 
                                                                        |MATRIX 2:  GIVES THE NORMALIZED STIMULUS 
 MATRIX   2                                                             |COORDINATES FOR EACH DIMENSION. 
  1        -.36331      .18661      .29956                              |THE SUM OF THE COORDINATES = 0 AND THE 
  2         .27636      .24774      .31646                              |SUM OF THE SQUARED COORDINATES = 1. 
  3         .22991     -.11510     -.53337                              | 
  4         .19499     -.62599      .33392                              | 
  5         .30641      .15777      .00643                              |MATRIX 3:  SAME AS MATRIX 2 
  6         .23464      .10129     -.36838                              | 
  7        -.32917     -.57003      .11166                              | 
  8        -.37103      .13160     -.43481                              | 
  9         .29127      .28272      .27710                              | 
 10        -.47006      .20339     -.00859                              | 
                                                                        |
 MATRIX   3                                                             | 
  1        -.36331      .18661      .29956                              | 
  2         .27636      .24774      .31646                              | 
  3         .22991     -.11510     -.53337                              |MATRIX 1: 
  4         .19499     -.62599      .33392                              |DIAGONAL = SUM OF SQUARES OF SUBJECT WGTS. 
  5         .30641      .15777      .00643                              |FOR EACH DIM. IF EACH ELEMENT IS DIVIDED 
  6         .23464      .10129     -.36838                              |BY NWT(1), THE NUMBER OF SUBJECTS, THIS 
  7        -.32917     -.57003      .11166                              |SHOWS THE RELATIVE IMPORTANCE OF EACH 
  8        -.37103      .13160     -.43481                              |DIMENSION. 
  9         .29127      .28272      .27710                              | 
 10        -.47006      .20339     -.00859                              |RELATIVE IMPORTANCES DECREASE DOWN THE 
                                                                        |DIAGONAL. OFF DIAGONAL ENTRIES ARE IGNORED. 
 MATRIX   1                                                             |MATRIX 1:
 SUMS OF PRODUCTS                                                       |SUMS OF SQUARES:  SUM OF MAIN DIAGONAL OF 
  1        2.85870     1.39817     1.52015                              |MATRIX 1: SUM OF SQUARES/NWT(1) = 
  2        1.39817     2.31003     1.33268                              |R-SQUARED (HISTORY OF COMPUTATION).
  3        1.52015     1.33268     1.10139                              |DIAGONALS SHOW IMPORTANCE OF DIMENSIONS. 
 SUM OF SQUARES =       6.27013                                         |MATRIX 2: 
                                                                        |OFF DIAGONALS SHOW CORRELATIONS BETWEEN 
 MATRIX   2                                                             |DIMENSIONS.  IF OFF DIAGONALS ARE CLOSE 
 SUMS OF PRODUCTS                                                       |TO 0, THEN UNCORRELATED. 
  1        1.00000      .04980      .04597                              |IF OFF DIAGONALS ARE LARGE, EITHER THE 
  2         .04980     1.00000     -.09392                              |DIMENSIONS ARE CORRELATED, OR THE 
  3         .04597     -.09392     1.00000                              |SOLUTION HAS NOT PROPERLY CONVERGED. 
 SUM OF SQUARES =       3.00000                                         |IF LARGE OFF DIAGONAL VALUES ARE FOUND, 
                                                                        |RERUN ANALYSIS IN A LOWER DIMENSIONALITY. 
 MATRIX   3                                                             | 
 SUMS OF PRODUCTS                                                       | 
  1        1.00000      .04980      .04597                              |SUM OF SQUARES = # OF DIMENSIONS 
  2         .04980     1.00000     -.09392                              | 
  3         .04597     -.09392     1.00000                              |MATRIX 3:  SAME AS MATRIX 2 
 SUM OF SQUARES =       3.00000                                         | 
           THIS IS PLOT OF DIMENSION  1 VS.DIMENSION  2 FOR TABLE NO.  1 |PLOT OF SUBJECT WEIGHTS FOR ALL PAIRS 
          +....+....+....+....+....+....+....+....+....+....+....+....+ |OF DIMENSIONS OF STIMULUS SPACES 
     1.20+                              |                              +|FOR ALL PAIRS OF DIMENSIONS. 
         .                              |                              .| 
         .                              |                              .|(TABLE 1 IS PLOT OF SUBJECT WGTS) 
      .92+                              |                              +| 
         .                              |                              .|
         .                              |    3                         .| 
      .65+                              |    #                         +| 
         .                              |    2     8                   .| 
         .                              |                              .| 
      .37+                              |               1              +| 
         .                              |               4              .| 
         .                              |              9               .| 
      .09+                              |                   #          +| 
         .------------------------------0------------------------------.| 
         .                              |                              .| 
     -.18+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
     -.46+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
     -.74+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
    -1.02+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2| 
          THIS IS PLOT OF DIMENSION  1 VS.DIMENSION  3 FOR TABLE NO.  1 | 
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
     1.20+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
      .92+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
      .65+                              |                              +| 
         .                              |                              .| 
         .                              |               4              .| 
      .37+                              |    #     8   9               +| 
         .                              |    #              #          .| 
         .                              |               1              .| 
      .09+                              |                              +| 
         .------------------------------0------------------------------.| 
         .                              |                              .| 
     -.18+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
     -.46+                              |                              +| 
         .                              |                              .| 
         .                              |                              .|
     -.74+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
    -1.02+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2|  
 
          THIS IS PLOT OF DIMENSION  2 VS.DIMENSION  3 FOR TABLE NO.  1 | 
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
     1.20+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
      .92+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
      .65+                              |                              +| 
         .                              |                              .| 
         .                              |     4                        .| 
      .37+                              |    9        827              +| 
         .                              | #              A3            .| 
         .                              |       1                      .| 
      .09+                              |                              +| 
         .------------------------------0------------------------------.| 
         .                              |                              .| 
     -.18+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
     -.46+                              |                              +| 
         .                              |                              .| 
         .                              |                              .|
     -.74+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
    -1.02+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2|  


          THIS IS PLOT OF DIMENSION  1 VS.DIMENSION  2 FOR TABLE NO.  2 | TABLE 2 GRAPHS SHOW THE GROUP
          +....+....+....+....+....+....+....+....+....+....+....+....+ | STIMULUS SPACE FOR THE RESPECTIVE 
     1.20+                              |                              +| DIMENSIONS.  THE STIMULI (OBJECTS)
         .                              |                              .| ARE HERE PLOTTED FOR THE AGGREGATE
         .                              |                              .| OF ALL SUBJECTS.
      .92+                              |                              +| 
         .                              |                              .| 
         .                              |                              .| 
      .65+                              |                              +| INDIVIDUAL PERCEPTUAL SPACES ARE NOT
         .                              |                              .| PLOTTED.  THE COORDINATES CAN, HOWEVER
         .                              |                              .| BE CALCULATED BY MULTIPLYING THE 
      .37+                              |      #                       +| STIMULUS COORDINATES (MATRIX 2 PLOTTED 
         .                              |       5                      .| IN TABLE 2) OF EACH DIMENSION BY THE 
         .                              |     6                        .| SQUARE ROOT OF THE SUBJECT'S WEIGHTS
      .09+                              |                              +| FOR THAT DIMENSION.  THIS MULTIPLICATION
         .------------------------------0------------------------------.| WILL GIVE THE COORDINATE OF THE STIMULUS 
         .                              |     3                        .| FOR THE INDIVIDUAL ON THE DIMENSION.
     -.18+                              |                              +| THESE PERCEPTUAL SPACES ALLOW FOR THE
         .                              |                              .| COMPARISON OF INDIVIDUAL SUBJECTS
         .                              |                              .| OR SUBJECT GROUPS IF AGGREGATE 
     -.46+                              |                              +| SIMILARITIES DATA IS USED.
         .                      7       |                              .|  
         .                              |    4                         .|  
     -.74+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
    -1.02+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2|  
           THIS IS PLOT OF DIMENSION  1 VS.DIMENSION  3 FOR TABLE NO.  2 | 
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
     1.20+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .92+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .65+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .37+                              |    4                         +|  
         .                     1        |      #                       .|  
         .                              |                              .|  
      .09+                      7       |                              +|  
         .------------------A-----------0-------5----------------------.|  
         .                              |                              .|  
     -.18+                              |                              +|  
         .                              |                              .|  
         .                              |     6                        .|  
     -.46+                     8        |                              +|  
         .                              |     3                        .|  
         .                              |                              .|  
     -.74+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
    -1.02+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2|  

          THIS IS PLOT OF DIMENSION  2 VS.DIMENSION  3 FOR TABLE NO.  2 | 
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
     1.20+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .92+                              |                              +|  
         .                              |                              .|  
         .                              |                              .| 
      .65+                              |                              +| 
         .                              |                              .|  
         .                              |                              .|  
      .37+              4               |                              +|  
         .                              |    129                       .|  
         .                              |                              .|  
      .09+                7             |                              +|  
         .------------------------------0---5A-------------------------.|  
         .                              |                              .|  
     -.18+                              |                              +|  
         .                              |                              .|  
         .                              |  6                           .|  
     -.46+                              |  8                           +|  
         .                           3  |                              .|  
         .                              |                              .|  
     -.74+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
    -1.02+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2|  
 
          THIS IS PLOT OF DIMENSION  1 VS.DIMENSION  2 FOR TABLE NO.  3 | TABLE 3 IS A REPEAT OF TABLE 2.
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
     1.20+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .92+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .65+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .37+                              |                              +|  
         .                              |      #                       .|  
         .                  A  1        |       5                      .|  
      .09+                     8        |     6                        +|  
         .------------------------------0------------------------------.|  
         .                              |     3                        .|  
     -.18+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
     -.46+                              |                              +|  
         .                      7       |                              .|  
         .                              |    4                         .|  
     -.74+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
    -1.02+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2|  

          THIS IS PLOT OF DIMENSION  1 VS.DIMENSION  3 FOR TABLE NO.  3 | 
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
     1.20+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .92+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .65+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .37+                              |    4                         +|  
         .                     1        |      #                       .|  
         .                              |                              .|  
      .09+                      7       |                              +|  
         .------------------A-----------0-------5----------------------.|  
         .                              |                              .|  
     -.18+                              |                              +|  
         .                              |                              .|  
         .                              |     6                        .|  
     -.46+                     8        |                              +|  
         .                              |     3                        .|  
         .                              |                              .|  
     -.74+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
    -1.02+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2|  
           THIS IS PLOT OF DIMENSION  2 VS.DIMENSION  3 FOR TABLE NO.  3 | 
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
     1.20+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .92+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .65+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
      .37+              4               |                              +|  
         .                              |    129                       .|  
         .                              |                              .|  
      .09+                7             |                              +|  
         .------------------------------0---5A-------------------------.|  
         .                              |                              .|  
     -.18+                              |                              +|  
         .                              |                              .|  
         .                              |  6                           .|  
     -.46+                              |  8                           +|  
         .                           3  |                              .|  
         .                              |                              .|  
     -.74+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
    -1.02+                              |                              +|  
         .                              |                              .|  
         .                              |                              .|  
          +....+....+....+....+....+....+....+....+....+....+....+....+ | 
        -1.2 -1.0  -.8  -.6  -.4  -.2   .0   .2   .4   .6   .8  1.0  1.2|  
                                                                        | 
 CORRELATION BETWEEN COMPUTED SCORES AND ORIGINAL DATA FOR SUBJECTS     | 
   1      .766075                                                       | 
   2      .724902                                                       | 
   3      .815670                                                       |  
   4      .830497                                                       | 
   5      .844774                                                       |  
   6      .844815                                                       | 
   7      .764019                                                       | 
   8      .814188                                                       | 
   9      .770781                                                       | 
  10      .755986                                                       | 
                                                                        | 
          AVERAGE SUBJECT CORR. COEFF. =   .79317                       | 
          MEAN SQUARE CORR. COEFF. =       .63070                       |