NEWS

ssmodels 2.0.1 (2025-06-02)

Corrected the initialization of the start values in the HeckmanSK function. Previously, it was relying on a two-step method to generate starting values, which could lead to numerical instability in some cases. Now, a more robust initialization is implemented to ensure better convergence and numerical stability.
Fixed the display of the log-likelihood in the summary methods of all functions (e.g., summary.HeckmanSK, summary.HeckmanCL, summary.HeckmanBS, etc.). Previously, these were reporting the negative of the log-likelihood. They now correctly display the log-likelihood value as returned by the optimization procedure.

Complete overhaul of the package, improving organization, readability, and performance of all functions.
Rewritten log-likelihood and gradient functions (loglik_* and gradlik_*) for enhanced numerical stability and clarity.
Fixed discrepancies where analytical gradients did not match numerical gradients.
Comprehensive documentation updates for all functions, ensuring better understanding and usage.
Added two new helper functions:
- postprocess_theta(): streamlines parameter transformations for clear interpretation and improved consistency across models.
- extract_model_components(): extracts model.frame, model.matrix, and model.response objects in a robust and reusable way.
All functions now follow consistent coding style and best practices.
Significant performance improvements, making the package lighter and more efficient.

Fixed issues with incorrect gradient calculations for sigma and rho parameters.
Corrected numerical errors in several model functions.

Updated vignette and examples to reflect the new structure and improvements.
Switched pkgdown site to Bootstrap 5 for improved readability and responsiveness.

Initial implementation of the classic Heckman model (HeckmanCL()) and foundational sample selection models.
Basic infrastructure for selection bias correction in econometric models.