New Cluster Analysis Tools in LHS 9.17.3

Sawtooth has recently added a full suite of segmentation methods to Lighthouse Studio 9.17.3.

To introduce this suite, this session covers the basics of successful segmentation, focusing on the two most critical elements, variable measurement scales and variable selection.

• Cautious variable measurement can prevent a researcher from hobbling their segmentation from the start, so we’ll discuss measurement and scaling methods that promote successful segmentation.
• Often overlooked but equally important is selecting which candidate variables to include in your analysis as basis variables. We’ll share some R&D work we’ve done on this topic that identifies the most successful variable selection methods. We plan to include the better methods in a future release of Lighthouse Studio, 9.18.

After these preliminaries we’ll describe and compare different segmentation algorithms – how they work, what are their strengths and weaknesses. We’ll also share the results of some R&D studies comparing the methods under a variety of adverse data conditions.

After those preliminaries, the second half of the presentation will feature demonstrations of how Lighthouse Studio has implemented each of several families of segmentation algorithms for your use, including:

• K-means cluster analyses
• Hierarchical cluster analyses
• Cluster ensembling
• Model-based clustering (AKA latent class analysis, finite mixture modeling)

‍