NEWS

---

mlmodels 0.2.0

New Features

• Clarke test (clarketest()): Full implementation of Clarke's (2007) distribution-free test for non-nested models. Includes both binomial/sign and signed-rank (Wilcoxon) variants, three penalty options (none, akaike, schwarz), model-based bootstrap, and an informative print method with expected values under the null and robustness notes.

• Vuong test (vuongtest()): Now supports the same three penalty options (none, akaike, schwarz) for consistency with Clarke's test and Vuong (1989). Bootstrap functionality has been added and aligned with the rest of the test suite.

• Wald test (waldtest()): Major upgrade — new constraints argument (replaces coef_names) that supports arbitrary linear combinations of coefficients.

Improvements

• Much better handling of weights across all tests and estimators. Statistics are now properly scaled to ensure correct asymptotic behavior.
• Enhanced logLik(), AIC(), and BIC() methods to return effective or scaled statistics depending on user request.
• Improved interaction with maxLik optimizer, leading to more reliable convergence in many models.
• Enhanced print methods for hypothesis tests (clearer conclusions, reference values under the null, bootstrap robustness warnings).
• Better print methods for summaries of weighted models, showing both effective and scaled (to sample size) measures where appropriate.
• Updated documentation and examples for non-nested model comparison.

Bug fixes & minor changes

• Fixed bootstrap storage (numeric vectors instead of character).
• Various internal cleanups and consistency improvements across the test suite.

mlmodels 0.1.2 (2026-05-08)

• Added return values in the documentation of exported functions that were missing them.
• Added references to implemented methods in the description.

mlmodels 0.1.1

• Fixed weighted log-likelihood calculation in ml_logit() (both homoskedastic and heteroskedastic versions). This bug previously caused incorrect log-likelihood, AIC, BIC, and convergence issues in weighted logit models.
• All other models were already handling weights correctly.

mlmodels 0.1.0

• First public release - Initial CRAN submission
• Provides maximum likelihood estimation for Gaussian (linear and log-normal), logit, probit, Poisson, negative binomial (NB1 and NB2), gamma, and beta models.
• Consistent S3 interface with support for modeling scale parameters.
• Multiple variance-covariance estimators (OIM, OPG, robust, cluster-robust, bootstrap, jackknife).
• Full suite of post-estimation tools and hypothesis tests.
• Compatible with marginaleffects for marginal effects and predictions.
• Comprehensive vignettes covering main model families and diagnostics.