Key Driver Analysis (Regression-Based)

Description

Finds which aspects of an experience most strongly predict an overall outcome such as satisfaction, NPS or SUS.

Ask users to rate an overall outcome and a set of specific attributes (ease of finding documents, speed, price for value). Regress the outcome on the attributes; standardised coefficients estimate how much each attribute drives the outcome. Plotting importance (coefficient) against performance (mean rating) produces a quadrant chart that shows where to invest: important attributes rated poorly. Sauro and Lewis's training highlights that correlated attributes make coefficients unstable, so factor analysis to group attributes, relative weights or Shapley-style decomposition give more trustworthy importance estimates.

Attributes

Loss type
Active
Effort
Hours

Steps

  1. Survey users on an overall outcome and 8 to 20 specific attributes.
  2. Check correlations among attributes; consider grouping them with factor analysis.
  3. Run multiple regression of the outcome on the attributes (or factor scores).
  4. Use standardised coefficients or relative-weight analysis to rank drivers.
  5. Plot importance against performance and target high-importance, low-performance areas.

Sources

  • Factor and Discriminant Analysis (Sauro & Lewis, MeasuringU training)
  • Quantifying the User Experience (Sauro & Lewis)
  • Practical Statistics for Data Scientists (Bruce & Bruce)
  • Feature Prioritization Methods (slide deck)

Also known as

  • Driver analysis
  • Importance-performance via regression

Categories

  • Analytics & Behaviour
  • Product Analytics & Strategy