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.
logLik(), AIC(), and BIC() methods to return effective or
scaled statistics depending on user request.maxLik optimizer, leading to more reliable
convergence in many models.ml_logit() (both homoskedastic and heteroskedastic versions).
This bug previously caused incorrect log-likelihood, AIC, BIC, and convergence issues in weighted logit models.marginaleffects for marginal effects and predictions.