Cluster
Data is frequently cross tabulated by pre-defined variables. An example would be breaking down a total sample by respondent categorised by,say, an age range, the gender of respondent or region where a respondent resides and so on.
Such cross tabulations are usually predicted prior to the fieldwork stage of a research project and specifically built into the questionnaire design and analysis planning.
A cluster analysis aims to identify (unexpected) groupings of respondents within a sample that manifest themselves after the survey is complete.
Cluster analysis is an exercise to arrange observed data into groups that have strong commonalities and/or also have distinct differences with other groups.
For instance, a group of supporters in a sporting stadium are a cluster of people that may have seemingly little similarities in certain categorisations (e.g. age, gender, and ethnicity) but have strong similarities elsewhere – travelling time to the stadium or a desire to have an experience or witness an event/participant in person.
Example
KGS ltd conducted a study with installers of gas central heating boilers. Demographics to categorise respondents were built in to the survey:
- Region
- Number of employees
- New build vs Repair/Maintenance sectors
There were few differences between these pre-supposed demographics.
KGS Ltd performed a cluster analysis on responses that rated manufacturers of boilers in regards to spare part availability, brand name and customer services.
The resulting analysis revealed distinct groups with highly clustered traits regardless of the pre-supposed demographics.
This enabled the study sponsor to redirect their marketing effort on radically different criteria than previously done with a consequent increase in profitability.