MultiSpline 0.2.0

MultiSpline 0.2.0

Now on CRAN — install.packages("MultiSpline")

This is a major update adding a complete estimation, interpretation, and
diagnostic framework for nonlinear multilevel spline modeling in R.
All v0.1.0 function calls remain valid without modification.

---

What's new

Cross-classified and nested multilevel structures

Supply two cluster variables and choose cross-classified or nested
random effects:

# Cross-classified
nl_fit(data = df, y = "score", x = "age",
       cluster = c("school_id", "student_id"), nested = FALSE)

# Nested  (higher-level grouping first)
nl_fit(data = df, y = "score", x = "age",
       cluster = c("school_id", "student_id"), nested = TRUE)

Automatic knot / df selection

# Standalone diagnostic
nl_knots(data = df, y = "score", x = "age", df_range = 2:8)

# Or inside nl_fit
nl_fit(data = df, y = "score", x = "age", df = "auto")

Multilevel R-squared decomposition

nl_r2(fit)
# Returns marginal R2m, conditional R2c, and a full variance
# partition table (r2_mlm style) across all random-effect levels

Derivative-based interpretation

deriv <- nl_derivatives(pred, x = "age")   # d1, d2 with CI bands
tp    <- nl_turning_points(deriv, x = "age") # maxima, minima, inflections

nl_plot(deriv_df = deriv, x = "age", type = "slope")
nl_plot(deriv_df = deriv, x = "age", type = "curvature")
nl_plot(pred_df = pred, deriv_df = deriv, x = "age", type = "combo")

Model comparison workflow

nl_compare(fit, polynomial_degrees = 2:3)
# AIC, BIC, log-likelihood, LRT vs linear for all candidate models

Cluster heterogeneity analysis

nl_het(fit, n_clusters_plot = 30)
# Plots subject BLUPs vs population mean
# LRT: random slopes vs random intercepts

Additional features

• B-spline basis: method = "bs", bs_degree argument
• Random spline slopes: random_slope = TRUE
• CI for glmerMod via delta method (default) or bootstrap
  (glmer_ci = "boot")

---

Bug fixes

• nl_r2() variance partition no longer produces NA rows in nested models
• nl_predict() CI now correct when controls are character variables
• nl_plot() no longer errors when no time column is present

---

Installation

# From CRAN (recommended)
install.packages("MultiSpline")

# Development version from GitHub
remotes::install_github("causalfragility-lab/MultiSpline")

---

Citation

citation("MultiSpline")

Hait, S. (2026). MultiSpline: Spline-Based Nonlinear Modeling for
Multilevel and Longitudinal Data. R package version 0.2.0.
https://CRAN.R-project.org/package=MultiSpline

---

Full changelog

See [NEWS.md](https://github.com/causalfragility-lab/MultiSpline/blob/main/NEWS.md)

MultiSpline v0.1.1

MultiSpline v0.1.1

• Added references to DESCRIPTION
• Replaced dontrun with donttest in examples
• Updated README with version and citation
• Updated JOSS paper and bibliography

MultiSpline v0.1.0 - Initial Release

Initial CRAN submission of MultiSpline v0.1.0

• nl_fit() for nonlinear multilevel modeling
• nl_summary() for coefficient output
• nl_predict() for predictions
• nl_plot() for visualization
• nl_icc() for intraclass correlations