Q Research Software

Users often get confused with the sample sizes in a Latent Class segment tree and don’t realize these are just estimations.

A customer wants to be able to run something like Sawtooth's Convergent Cluster & Ensemble Analysis (CCEA) in Q. More info about the approach is on Sawtooth's site here: [ https://sawtoothsoftware.com/advanced-analytical-tools/convergent-cluster-ensemble-analysis|https://sawtoothsoftware.com/advanced-analytical-tools/convergent-cluster-ensemble-analysis|smart-link ] Based on that info, it sounds like it's a cluster analysis approach that repeats the analysis multiple times from random starting points: {quote}Our Convergent Cluster & Ensemble Analysis (CCEA) System addresses this problem. CCEA automatically replicates each analysis 30 times from different random starting points. Each replication is compared to every other to assess its reproducibility. The most reproducible solution is chosen automatically, and its level of reproducibility is reported.{quote} The customer is currently using Latent Class Analysis for their segmentation, then exporting the data from Q and running this additional analysis externally.

Customer says, “Giving me more in-depth output for the statistics (main thing I’m missing is Wald statistic for the parameter estimates – I’m using a work around in Q calculating R2 values with MANOVA as suggested in a webinar, but I’d really love to see these wald statistics).”

Customer says an improvement for LCA in Q is “Being able to deal with ordinal data (Q only let’s me choose between categorical and numeric, but that sometimes leads to issues).