An Empirical Comparison of Variable Standardization Methods in Cluster Analysis

Area of Science:

• Marketing Research
• Data Science
• Statistical Analysis

Background:

• Standardizing data columns (mean zero, unit standard deviation) is common in marketing research before clustering entities.
• This practice persists even when variables use similar units, like 7-point scales.

Purpose of the Study:

• To examine six data column standardization methods against no standardization.
• To compare these methods using derived importances from conjoint analysis data.

Main Methods:

• Comparative analysis of six standardization techniques and a null case.
• Replication across ten large-scale datasets of conjoint-derived attribute importances.

Main Results:

• Prevailing column standardization practices may negatively impact segmentation results for certain marketing data.
• Data profiling outcomes demonstrate reasonable robustness across different column standardization methods.

Conclusions:

• The common practice of column standardization in marketing research requires careful consideration, as it can affect segmentation accuracy.
• While segmentation may be sensitive, background data profiling appears unaffected by the choice of standardization method.