APPLIED SURVEY RESEARCH TOOLS
Analysis of Covariance
Objective
To adjust data for the effects of other variables and thus permit analysis of variance to be used effectively.
Examples/Applications

Sample matching

Postexperimental adjustments
Assumptions

Same as analysis of variance

Regressions for treatment groups have parallel slopes
Mechanics

Uses regression to statistically match groups by canceling out the effects of initial score differences

Applies analysis of variance to adjusted scores
Seminal Articles/Texts

Barrow, Lionel; Journal of Advertising Research, May 1964

Green, Paul and Tull, Donald; Journal of Advertising Research, June 1964
Analysis of Variance
Objective
To decompose the total variance of an experiment into variance attributable to the effects of specific variables.
Examples/Applications

Instore experiments

Market tests

Product tests/concept tests

Survey order effects

Picking discriminators in segmentation studies

Conjoint measurement
Assumptions

Samples are drawn from normal population

All samples have the same variance

Additivity (all means are a function of marginal means and the general mean)
Mechanics

Total sum of squares

Between groups sum of squares

Within groups sum of squares

Degrees of freedom

Mean square

F ratio
Seminal Articles/Texts

Winer, B.J., "Statistical Principles in Experimental Design", McGrawHill 1962
Cluster Analysis
Objective
To form homogeneous groups of objects of some kind (e.g., people, brands, markets).
Examples/Applications

Market segmentation

Clarification of perceptual maps

Market typologies

Store typologies
Assumptions

The dimensions on which clustering is done are the important ones, and usually equally important

Variables are (relatively) uncorrelated
Mechanics

Derivation of similarities

Correlations

Distances

Hits

Crossproducts


Types

Hierarchical buildup (Johnson)

Hierarchical builddown (AID, CHAID)


Bump and push (SingletonKatz; HowardHarris)

Clumping
Seminal Articles/Texts

P. H. A. Sneath and R. R. Sokal; Numerical Taxonomy; W.H. Freeman, 1973

John A. Hartigan; Clustering Algorithms; Wiley & Sons, 1975
Canonical Correlations
Objective
To measure in precise quantitative terms the relationship between two sets of variables.
Examples/Applications

Advertising effectiveness

Relating measures of awareness, recall, attitudes, and/or purchase intention to variables such as advertising weight, media mix, or distribution level.


Sales performance

Relating measures such as new accounts opened to variables such as personality characteristics


Green's relating of two measures of risktaking to personailty characteristics

Segmentation solution testing
Assumptions

Same as regression analysis
Mechanics

Means and standard deviations

Correlation matrix

Canonical correlations

Coefficients for the first set of variables

Coefficients for the second set of variables
Seminal Articles/Texts

Stewart & Love; A General Canonical Correlation Index; Journal of Marketing Research; p.90, February, 1972

Psychological Bulletin; Vol. 70; pp 160163, 1968

Journal of Advetising Research; p.19, June 1969
Regression Analysis
Objective
To measure in precise quantitative terms the relationship between one dependent variable and one or more independent variables.
Examples/Applications

Prediction

Analysis

Developing performance standards

Screening variables

Media/trade program impact

Matching samples
Assumptions

Causal, or at least meaningful, relationships

Linearity

Independent variables are independent (low collinearity)

Heteroscedasticity (uncorrelated error terms)

Relationship can be explained in terms of variance around a line estimate
Mechanics

Estimating equation

Total variance

Explained and unexplained variance

Coefficient of determination and correlation

beta (slope) and b (intercept)

Partial correlations

Dummy variables

Recursive regressions (stepwise/best fit)
Seminal Articles/Texts

A.R. Baagaley, Intermediate Correlational Methods, John Wiley & Sons, 1964
Conjoint (TradeOff) Analysis
Objective
To measure the relative importance of benefits and/or product attributes, overall and for individual respondents.
Examples/Applications

Product/service design

Choice of buying incentives

Segmentation
Assumptions

All important variables are included in the list

All variables are salient for all respondents

All respondents use a compensatory decision process

The way in which respondents "play the game" mirrors real world decisions

People can analyze the importance of emotional or point of sale benefits, and will accurately report it (or, alternatively, that these benefits are unimportant)

The stimuli accurately and identically to all respondents

The ranges of intensity are equivalent for each benefit
Mechanics

Ratings, or alternatively rankings, choice or pairwise tradeoffs

Fractional factorial or orthogonal arrays (Latin squares)

"Build your own" design (may not be balanced, however)

Monanova or isotonic regression

Personal utilities (partworths)
Seminal Articles/Texts

Green, Paul; Consumer Research, Vol. 1, No. 2, p. 61; September 1974
Discriminant Analysis
Objective
To classify people or objects into predefined groups.
Examples/Applications

"Good" vs. "bad" markets

"Good" vs. "bad" performers (brands, people)

Concept acceptors vs. rejecters

Users vs. nonusers (brand, or product category)

Users vs. prospects

Responders to an offer vs. nonresponders
Assumptions

Equal covariance matrices for the groups being discriminated, others are the same as for regression analysis
Mechanics

Means and standard deviations

Within group matrices

F ratios

Discriminant functions

Classification matrix

Probabilities of group membership (hitmiss matrix)

Eigenvalues
Seminal Articles/Texts

Journal of Marketing Research, p.156, May 1969
Factor Analysis
Objective
To reduce a set of intercorrelated items to a smaller set of independent items; to obtain insight into the underlying structure.
Examples/Applications

Identifying underlying structures

Brand image

Benefits

Brand competition


Data reduction

Screening long lists of variables (i.e., benefit screening)

Phase II of segmentation studies

Simplification for presentation purposes


Categorizing people into groups
Assumptions

An underlying structure (i.e., may not always exist)

Additivity (the original matrix can be reproduced by linear combinations)

Orthogonality of factors (usually)
Mechanics

The original (unrotated) matrix

The extraction process

Rotation (i.e., varimax, quartimax)

Analysis (factor naming, meaning)
Seminal Articles/Texts

Harman, Harry; Modern Factor Analysis; 2nd Ed, University of Chicago Press, 1967

Fruchter, Benjamin; Introduction to Factor Analysis, D. Van Nostrand Co. 1954
Perceptual Mapping
Objective
To obtain a simplified picture of customers' perceptions of brands and brand images.
Examples/Applications

Hypothesis development

Business opportunity identification

Defining the competitive environment

Positioning (brands or categories)

Segmentation
Assumptions

Equal covariance matrices for the groups being discriminated, others are the same as for regression analysis
Mechanics

Obtaining the dissimilarities matrix

Direct (via sorting, paired comparisons)

Indirect (correlations, crossproducts, hits)


Types of maps

Simple space

Joint space

Vector

Discriminant function

Correspondence mapping


Interpreting maps

Number of dimensions

Clustering (distances)

Maximum r / explained variance

Ideal points (explicit, implied)

Seminal Articles/Texts

Green, Paul; Journal of Marketing, p.24, January 1975