MultiSpline vs silx

MultiSpline fits spline-based nonlinear models to multilevel and longitudinal data in R. The package reached CRAN in February 2026 with five functions covering fitting, summary, prediction, plotting and intraclass correlations. Version 0.2.0, seven weeks later, adds cross-classified and nested random-effect structures, automatic knot selection, a multilevel R-squared variance partition, derivative-based interpretation with turning points, model comparison against polynomials, and cluster heterogeneity analysis, while keeping every 0.1.0 call valid.

The arc is a research package being built out into a workflow at speed: 0.1.0 could fit a curve, 0.2.0 can tell you where the curve turns, how much variance each level explains, and whether a spline beats a polynomial at all. Backward compatibility was preserved across that expansion, which suggests the author is building for outside users rather than a single paper. The JOSS submission referenced in 0.1.1 points at academic distribution as the intended channel.

With the interpretation and diagnostics layers now in place, the next release will most likely extend the supported model families beyond the current lmer and glmer backends.