Attitudinal segmentation with large numbers of diverse
statements is problematic; it’s often difficult to find focused, cohesive,
reduction techniques are used (typically Factor Analysis) to create a
smaller number of composite variables which are combinations of all the
statements which are then used.
However unless all
the statements in a sub set are strongly correlated it is possible
for consumers to have similar factor scores despite having very different
statement scores. This means people can be put in the same segments
despite having quite different attitudinal profiles. The result is a set of
less cohesive, less homogenous segments.
To avoid this
problem we used Bayesian network analysis to create a smaller set of
factors for a leading snack producer. This machine learning technique produces a probability based
graphical network model detailing the relationship between all variables in
the analysis. We applied this to derive a robust
consumer segmentation based on attitudes to health, nutrition and
How does it work?
- The width of the line denotes the strength of the relationship
- The colour denotes the nature of the correlations: Red = Negative, Blue = Positive
- 32 statements uncover 10 factors: the key input into our segmentation which resulted in five clear consumer typologies
Bayesian Network Analysis poster