Overview (PRINCALS command)

PRINCALS (principal components analysis by means of alternating least squares) analyzes a set of variables for major dimensions of variation. The variables can be of mixed optimal scaling levels, and the relationships among observed variables are not assumed to be linear.

Options

Optimal Scaling Level. You can specify the optimal scaling level for each variable to be used in the analysis.

Number of Cases. You can restrict the analysis to the first n observations.

Number of Dimensions. You can specify how many dimensions PRINCALS should compute.

Iterations and Convergence. You can specify the maximum number of iterations and the value of a convergence criterion.

Display Output. The output can include all available statistics, only the default statistics, or only the specific statistics you request. You can also control whether some of these statistics are plotted.

Saving Scores. You can save object scores in the active dataset.

Writing Matrices. You can write a matrix data file containing category quantifications and loadings for use in further analyses.

Basic Specification

  • The basic specification is the PRINCALS command and the VARIABLES subcommand. PRINCALS performs the analysis assuming an ordinal level of optimal scaling for all variables and uses all cases to compute a two-dimensional solution. By default, marginal frequencies, eigenvalues, and summary measures of fit and loss are displayed, and quantifications and object scores are plotted.

Subcommand Order

  • The VARIABLES subcommand must precede all others.
  • Other subcommands can appear in any order.

Operations

  • If the ANALYSIS subcommand is specified more than once, PRINCALS is not executed. For all other subcommands, only the last occurrence of each subcommand is executed.
  • PRINCALS treats every value in the range of 1 to the maximum value specified on VARIABLES as a valid category. Use the AUTORECODE or RECODE command if you want to recode a categorical variable with nonsequential values or with a large number of categories to avoid unnecessary output. For variables treated as numeric, recoding is not recommended because the intervals between consecutive categories will not be maintained.

Limitations

  • String variables are not allowed; use AUTORECODE to recode nominal string variables into numeric ones before using PRINCALS.
  • The data must be positive integers. Zeros and negative values are treated as system-missing and are excluded from the analysis. Fractional values are truncated after the decimal and are included in the analysis. If one of the levels of a categorical variable has been coded 0 or a negative value and you want to treat it as a valid category, use the AUTORECODE or RECODE command to recode the values of that variable (see AUTORECODE and RECODE for more information).
  • PRINCALS ignores user-missing value specifications. Positive user-missing values less than the maximum value on the VARIABLES subcommand are treated as valid category values and are included in the analysis. If you do not want the category included, you can use COMPUTE or RECODE to change the value to something outside of the valid range. Values outside of the range (less than 1 or greater than the maximum value) are treated as system-missing.