Presentation Transcript
Multivariate Grouping Procedures in Marketing Research: Multivariate Grouping Procedures in Marketing Research Factor Analysis, Multidimensional Scaling (MDS), Cluster Analysis
Purpose of Procedures: Purpose of Procedures Procedures involve several variables, with none of the variables being treated as dependent on the others.
Main purpose: to form groups of some type
Group variables into underlying factors (FA)
Group customers into market segments (CA)
Group products to generate product space maps that show how different products are positioned (MDS)
Factor Analysis: Factor Analysis Groups variables on the basis of correlations
Example: Handling factor is observed through 2 questions:
Rate the Mercedes E320 sedan on how well it can handle a curvy road (1-7)
Rate the Mercedes E320 sedan on its ability to maneuver to avoid unexpected objects on the road (1-7)
Factor Analysis: Factor Analysis General form of a factor:
F = w1x1 + w2x2 + … + wkxk
Descriptive Results:
Factor loadings: Correlation between observed variables and factor (.50+ is desirable)
Eigenvalue: Sum of squared loadings for all variables on a factor. Eigenvalues provide measure of % of variance in the contributing variables explained by the factor
Factor scores: Values for each observation (e.g., respondent) on the factor.
Factor Analysis: Factor Analysis Inferential results:
Is overall analysis effective in grouping variables?
OK if 70%+ of variance explained by factors
OK if factors are meaningfully interpreted
How many factors should we retain?
As many factors as eigenvalues > 1
Elbows on scree plot
Interpretive judgments
What variables belong to which factors?
Assign variables to factors according to loadings
Applications of Factor Analysis: Applications of Factor Analysis Interpret underlying constructs (insight)
Reduce variables to more manageable number
Pick one per factor
Build an index
Use factor scores
Simplify future data collection
Create perceptual maps (need scoring coefficients)
Segmentation
Factor Analysis: Factor Analysis Issues to consider:
Need interval or ratio scales
Assumes linear relationships
Principal component analysis most often used
FA relies on judgment rules
How to use information to reduce number of variables?
Use factor scores
Use average of individual variables