Algorithmic Refinements in Candidate Search
The release notes for Stat-Ease version 25.0.7 focus primarily on numerical stability inside the coordinate exchange algorithm. In previous builds, generating D-optimal or I-optimal designs across non-convex formulation spaces occasionally hit infinite iteration traps whenever three or more component bounds intersected at extreme corners. Build 25.0.7 introduces dynamic epsilon stepping, allowing the matrix explorer to resolve boundary points cleanly without precision decay.
Beyond formula space exploration, response surface modeling routines now exhibit tighter standard error bands around split-plot hard-to-change variables. The underlying linear regression routines take advantage of multi-threaded matrix operations, speeding up variance inflation factor checks during model reduction phases.
“When experimental runs cost thousands of dollars per batch, algorithmic rounding errors in optimal point selection are unacceptable. Version 25.0.7 guarantees solver convergence even inside congested boundary corridors.”
Interactive diagnostics also received quality-of-life adjustments. Externally studentized residual plots and Cook distance bars calculate instantly without memory leaks during multi-window sessions, ensuring smoother navigation across large screening trials with dozens of continuous responses.