PRINCIPAL COMPONENTS : PRINCOMP

expects as its input a matrix of correlations or covariances. It is included here to allow comparison with the dimensions identified by non-metric MDS procedures for the same data. An error is reported if the input matrix does not consist of correlations or covariances, i.e the product of one or more pairs of symmetric off-diagonal terms is greater than the product of the corresponding diagonal terms.

DATA: 2-way, 1-mode  scalar products                  
TRANSFORM: Linear                                

MODEL: SCALAR PRODUCTS

Principal components is a mathematical technique, with no underlying statistical model, which is frequently used to identify a limited number of linear combinations of the original variables that can be used to summarise the data, losing as little information as possible. Technically, it simply produces an orthogonal rotation of the input matrix to its principal axes, or eigenvectors. By default, PRINCOMP will list all n eigenvalues (latent roots) and principal components (eigenvectors) of a matrix of n variables.

In many sets of multivariate data the variables will be measured in different units and are standardised before analysis. This is equivalent to extracting the principal components as eigenvectors of the matrix of correlations, rather than of the covariance matrix. The eigenvalues and principal components of these matrices are not generally the same, and choosing to analyse correlations is equivalent to deciding to consider the variables to be equally important.

PRINCOMP automatically lists all principal components of the input matrix. The size of the matrix is given by N OF STIMULI and the matrix is read by the READ MATRIX command. The format of the input matrix is given by the parameter DATA TYPE in the PARAMETERS command.

PARAMETER
Keyword     Default          Description

INPUT COMMANDS
Keyword                                            Function
N OF STIMULI    [number]         Number of stimuli in the analysis

LABELS  [followed by a series    Optionally identify the stimuli,
            of labels (<= 65 chars)  followed by the subjects, as 
            each on a separate       required. Labels should identify
            line]                            all variables, without omission.

DATA TYPE      1      1: Lower triangular matrix without diagonal
                              2: Lower triangular matrix with diagonal
                              3: Upper triangular matrix without diagonal
                              4: Upper triangular matrix with diagonal
                              5: Full symmetric matrix.

READ MATRIX                            Start reading data for run

PLOT options (to main output file)
Option                            Description
COMPONENTS     Plots the principal components.
                         If a parameter is added, this specifies the number
                         of normalized principal components to be plotted.
                         (Plotting all components is liable to generate a
                         rather large output file.)
ROOTS               Produces a 'scree plot' of the latent roots
                         against the principal components.

NOTES
1. The READ MATRIX command is obligatory in PRINCOMP.
2. There are no PRINT options as such in PRINCOMP.
    By default, the eigenvalues (or latent roots) of the input matrix are
    listed in descending order, together with the corresponding
    eigenvectors or principal components, and the proportions of the
    total variance accounted for by each.
3. LABELS Allows you to add optional labels (following the command
    and then on successive lines) to identify variables.

PROGRAM LIMITS
Maximum no. of stimuli = 300
Maximum no. of dimensions = 8

See also

  • The NewMDSX commands in full