Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Statscraft 0.15.0, release v0.15.0

CATPCA

Categorical principal components: quantify each variable by optimal scaling (spline, ordinal, nominal, multiple nominal or numeric) and reduce them to a few components.

Syntax

CATPCA [/VARIABLES=] varlist
   /ANALYSIS=varlist [([WEIGHT=n] [LEVEL=SPORD | SPNOM | ORDI | NOMI | MNOM | NUME]
                      [DEGREE=2 | n] [INKNOT=2 | n])] [varlist [(...)] ...]
   [/DISCRETIZATION=[varlist [([GROUPING] [NCAT=7 | n] [DISTR=NORMAL | UNIFORM] | [EQINTV=width]
                               | RANKING | MULTIPLYING)] ...]]
   [/MISSING=[varlist | ALL [([PASSIVE] [MODEIMPU | EXTRACAT | RANDIMPU]
                               | ACTIVE [MODEIMPU | EXTRACAT | RANDIMPU] | LISTWISE)] ...]]
   [/SUPPLEMENTARY=[OBJECT(case numbers)] [VARIABLE(varlist)]]
   [/CONFIGURATION=[INITIAL | FIXED]('file' | dataset)] [/DIMENSION=2 | n]
   [/NORMALIZATION=VPRINCIPAL | OPRINCIPAL | SYMMETRICAL | INDEPENDENT | q]
   [/MAXITER=100 | n] [/CRITITER=.00001 | value]
   [/ROTATION=[NOROTATE | VARIMAX | EQUAMAX | QUARTIMAX | PROMAX[(4 | kappa)] | OBLIMIN[(0 | delta)]]
              [KAISER | NOKAISER]]
   [/RESAMPLE=NONE | BOOTSTRAP[(samples, percent, BALANCED | UNBALANCED, PROCRU | REFLEC)]]
   [/PRINT=[DESCRIP[(varlist)]] [LOADING[(SORT | NOSORT)]] [CORR] [VAF] [OCORR] [QUANT[(varlist)]]
          [HISTORY] [OBJECT[([(labelvar)] varlist)]] [NONE]]
   [/PLOT=[OBJECT[(varlist)][(len)]] [LOADING[(varlist [(CENTR[(varlist)])])][(len)]]
         [CATEGORY(varlist)[(len)]] [JOINTCAT[(varlist)][(len)]] [TRANS(var[(n)] ...)[(len)]]
         [RESID(var[(n)] ...)[(len)]] [BIPLOT[(LOADING | CENTR[(varlist)])][(varlist)][(len)]]
         [TRIPLOT[(varlist[(varlist)])][(len)]] [PROJCENTR(var, varlist)[(len)]] [VAF]
         [NDIM(first, last)] [NONE]]
   [/SAVE=[TRDATA[(root[(n)])]] [OBJECT[(root[(n)])]] [APPROX[(root)]]]
   [/OUTFILE=[DISCRDATA('file' | dataset)] [TRDATA('file' | dataset)] [OBJECT('file' | dataset)]
            [APPROX('file' | dataset)]].
Variables are coded as for MULTIPLE CORRESPONDENCE: whole codes from 1, lower codes and
missing values missing, strings coded in sorted order, fractions cut into 7 groups unless
DISCRETIZATION says otherwise; DISCRETIZATION, MISSING, SUPPLEMENTARY, CONFIGURATION, DIMENSION,
MAXITER and CRITITER work as there. Each variable gets a value for each category, chosen so
that the components explain the quantified variables as well as they can. LEVEL says what the
values must respect: NUME equal steps between codes (plain principal components when every
variable is numeric), ORDI the order of the codes, NOMI nothing but the grouping, SPORD (the
default) and SPNOM a smooth curve through the codes, a spline of DEGREE pieces joined at INKNOT
interior knots placed at quantiles of the cases' codes (SPORD keeps it rising), and MNOM a
separate value per component, the centroid of the category's cases. WEIGHT gives a variable
more say. The fit starts from the exact numeric solution (iteration 0), then alternates
quantifications and component scores until the variance accounted for gains less than
CRITITER. An extra category from ACTIVE EXTRACAT is left out of the level's restriction after
iteration 0. The components are then turned to principal axes, and a component is reflected
when its negative loadings have the larger mean square. Supplementary variables are fitted to
the final scores. NORMALIZATION q gives the scores the power (1 + q)/2 of the root of each
eigenvalue and the loadings and multiple nominal centroids the power (1 - q)/2 less one, so
that scores times loadings stay the same: VPRINCIPAL -1 (standard scores, loadings are
correlations), OPRINCIPAL 1, SYMMETRICAL 0; INDEPENDENT scales the scores only. ROTATION turns
the loadings of the variables that have loadings, with Kaiser normalization unless NOKAISER,
and the scores and centroids with them. RESAMPLE=BOOTSTRAP is not supported yet (an error);
a rotation with BOOTSTRAP is an error too. SPLIT FILE does not apply.
Tables: Case Processing Summary, Iteration History (iteration 0 and the last, or all with
HISTORY: variance accounted for, its increase, and the loss split into the centroid part and
the part from holding single variables to a line) and Model Summary (Cronbach's alpha,
eigenvalue and percent of variance per dimension and in total) always; DESCRIP frequencies,
missing count and mode; QUANT per variable the quantification, centroid and vector
coordinates; VAF the variance each variable accounts for, by centroids and by vectors; OCORR
and CORR the correlations of the original and of the quantified variables (one matrix per
dimension when a variable is MNOM), with eigenvalues; OBJECT the component scores; LOADING the
component loadings (SORT groups them by their largest loading), with the rotated loadings and
transformation matrix, or pattern, structure and component correlations after an oblique
rotation. Without PRINT, or with PRINT alone: DESCRIP LOADING CORR.
Charts (OBJECT and LOADING by default): component scores, colored by the categories of each
named variable; loadings, with MNOM centroids after CENTR; category centroids and vectors of
each variable; all centroids together (JOINTCAT); quantification by code (TRANS); quantified
value against its approximation (RESID); scores with loadings or centroids (BIPLOT) or with
both (TRIPLOT), scores and centroids shrunk to the loadings, neither with INDEPENDENT; the
centroids of one variable projected on the loading direction of others (PROJCENTR, also as a
table); bars of each variable's variance accounted for per dimension and in total (VAF). Two
dimensions give one scatter, more a scatterplot matrix, NDIM(a, b) one scatter of a against
each of a+1 to b. A label length of 0 shows values and names instead of labels.
SAVE adds the quantified variables rootK_S (TRA, K the place on ANALYSIS, S the first free set
number; an MNOM variable gives rootK_D_S per dimension D with the root cut to 3 characters),
the scores rootD_S (OBSCO) and the approximations rootK_S (APP: scores times loadings, the
first score for an MNOM variable). Excluded cases and passive missing values are
system-missing. OUTFILE writes the codes (DISCRDATA), TRAK or TRAK_D, OBSCOD and APPK; a file
with no keyword means TRDATA.

Example

CATPCA pace crowd parking shade /ANALYSIS=pace crowd(LEVEL=ORDI) parking(LEVEL=NOMI) shade(LEVEL=SPNOM INKNOT=1) /MISSING=shade(ACTIVE EXTRACAT) /ROTATION=VARIMAX /PRINT=LOADING(SORT) VAF QUANT(parking) /PLOT=BIPLOT TRANS(shade) /SAVE=OBJECT.

statscraft help CATPCA prints the same text in the terminal. All commands