PC-MDS PROGRAM DESCRIPTIONS AND SIZE LIMITATIONS

MultiDimensional PREFerence Scaling, performs an analysis of any type of preference data for up to 100 stimuli points (objects) and 100 subject vectors (attributes). The program develops vector directions for preferences and the configuration of stimuli in a common space.

Offers Kruskal Young Shepard Torgerson multidimensional scaling. KYST combines MDSCAL and TORSCA to include the powerful initial configuration of TORSCA along with the ability to rotate solutions to principal components. Proximity (similarity) data is used for up to 100 stimuli in up to 12 dimensions. KYST offers a variety of data input formats, distance measures, and monotone, polynomial and multivariate regression models.

Produces PREFerence MAPping analysis based on a generalization of the Coombsian unfolding model of preference. These models use a known configuration of stimuli to portray individual preference data. Up to 150 stimuli in 10 dimensions may be evaluated for 99 subject groups.

INdividual Differences SCALing produces an analysis of proximity (similarity) data. Program options include INDIFF, for scaling individual differences and CANDECOMP for performing the more general canonical decomposition of N-way tables. INDSCAL produces up to a 7 way solution for ten dimensions. The product of the total number of levels must be less than 64,000.

PROperty FITting by Optimizing Nonlinear or Linear Correlation is a technique for fitting outside property vectors into stimulus spaces. Two types of property fitting: Max R, and non linear correlation, are performed. PROFIT performs linear and nonlinear regression for a set of 60 properties, 500 stimuli in 10 dimensions.

MONotone ANalysis Of VAriance performs additive conjoint analysis of rank order data. This model handles full factorial designs and up to 12 factors can be handled with 100 levels per factor and a maximum of 5000 data points.

Performs a non-metric pairwise conjoint analysis. Utilities are produced for an attribute matrix of up to 10 levels by 10 levels. Up to 200 Attributes may be defined. Additive or multiplicative utilities may be specified. Importance values are attached to the utilities.

Performs a conjoint analysis of fractional factorial data. The program is compatible with design files from Bretton Clark's Conjoint Designer. This two stage program generates utilities, which are used in a simulator to generate choice share estimates using the Bradley-Terry-Luce, Logit and First Choice Models. 50 variables and 10,000 subjects.

Howard Harris CLUSTER analysis of a subject by variable matrix. The K-means method minimizes within-group variance at each clustering level. 2000 subjects, 40 Variables.

HIerarchical CLUSTering performs hierarchical cluster analysis on a 200 x 200 similarity or dissimilarity matrix.

Ward's method is a hierarchical agglomerative method of clustering. Ward's method attempts to optimize the minimum variance within clusters using the within groups sum of squares. 300 subjects and 300 variables.

Thurstone CASE 5 analysis computes scale scores from (1) raw paired comparisons data, (2) a frequency matrix of the number of times one stimulus was preferred to another stimulus, or (3) ranked data. A 60 stimuli by 60 stimuli matrix may be analyzed.

Computes a matrix of interpoint distances from a stimuli by variables data matrix. (100 stimuli and 100 attributes).

Cliff Factor MATCHing performs an orthogonal rotation of two matrices (such as multidimensional scaling or factor analysis solutions) to achieve congruence. FMATCH will fit a 50 by 50 target matrix to any number of data matrices.

CORrespondence ANalysis consists of techniques for mapping two way contingency tables for large data sets. Correspondence analysis is analogous to a principal components analysis of rows and columns of contingency tables (crosstabulation data). Correspondence analysis finds the best simultaneous representation of the categorical data. (Lebart, Morineau, Warwick algorithm)

CORRESPondence analysis uses the Carroll, Green and Schaffer algorithm for mapping two way contingency tables for large data sets. This algorithm double centers the data producing more accurate representations of structure. Up to 100 rows and columns of contingency table (crosstabulation) data may be analyzed.

Interactive stepwise DISCRIMinant analysis. 50 variables and 32,000 subjects.

(BMD ALGORITHM) (DISCRIM uses a SPSS type control file)

Interactive Stepwise Multiple regression analysis. Up to 50 variables and 32,000 subjects. (BMD ALGORITHM) (REGRESS uses a SPSS type control file)

Interactive FACTOR Analysis. Orthogonal and oblique rotation and output of factor scores are supported. Up to 50 variables may be analyzed for 32,000 subjects. (BMD ALGORITHM) (FACTOR uses a SPSS type control file)

Frequency analysis and descriptive statistics. Up to 250 variables may be analyzed for 32,000 subjects. (BMD ALGORITHM) (FREQ uses a SPSS type control file)

A full screen ASCII editor with full search and replace, blocking and file management capabilities. Multiple files may be opened and worked on simultaneously.