NEWS

evbsreg 1.2.1

devbs() documentation stated that the extreme-value index of the response is always 2*gamma. That holds only for gamma > 0; for gamma <= 0 the index is gamma itself (Ferreira, Gomes and Leiva, 2012, Theorems 4.1-4.2).
evbs_return_level() documentation now states that the level is conditional on the covariate value, that period is in units of one observation (months, for monthly maxima), and what reading it as an annual return level assumes.
evbs_monitor() documentation now states that the endpoint is evaluated at the covariate value of the running maximum, as the code does.
inst/scripts/: the simulation scripts used gamma in {-0.2, -0.1, 0.1, 0.2} whereas the paper's tables use {-0.2, 0, 0.2}; Scenario 3 used alpha = 0.5, which with the contamination alpha - 0.5 gave a degenerate alpha = 0. Both now match the paper (Scenario 3: beta = (1, 2), alpha = 1.2).

cnc_diagnostics_gev() also returns Bbar, the mean aggregate contribution computed exactly as colMeans(cnc_diagnostics(fit)$Bj) is for the EVBS model, so that influence rankings can be compared across the two families. The existing Bj (normalised diagonal of the influence matrix) is unchanged.
gev_monitor() computes the prospective identifiability index of evbs_monitor() under the GEV regression.

plot_aggregate_contributions() labelled flagged observations with a fixed offset above each point, so labels for observations close together on the index axis overlapped and became unreadable. Labels are now staggered vertically for points that fall within a closeness threshold of each other, leaving isolated points unaffected.
envelope_qq() saved and restored the caller's entire par() state (par(no.readonly = TRUE)), which includes the multi-panel layout position (mfg). When called as one step of a multi-panel figure set up by the caller (par(mfrow = c(1, 2)) followed by envelope_qq() then a second plot()), restoring the saved state on exit rewound the layout position, so the second plot overwrote the first panel instead of advancing to the second, leaving one panel blank. The function now saves and restores only pty, the one parameter it actually changes.

evbs_monitor() computes the prospective endpoint-identifiability index $F_t$, combining the conformal normal curvature with the finite upper endpoint, and plot.evbs_monitor() draws the corresponding control chart. Values below one indicate that the endpoint is determined by a single observation and should not be quoted as a design value.

logEVBS() provides a gamlss.family implementation of the log-EVBS distribution, allowing every parameter, including the tail-shape parameter, to depend on covariates or smooth terms. With a constant predictor it reproduces the fixed-parameter fit of evbsreg.fit(). Companion functions dlogEVBS(), plogEVBS() and qlogEVBS() are also provided.
devbs(), pevbs() and qevbs() complete the distribution family. The package previously exported only revbs(), so users had no way to compute quantiles or return levels.
evbs_endpoint() returns the finite upper endpoint of the fitted model when gama < 0 (Weibull max-domain of attraction).
evbs_return_level() computes return levels and expected shortfall from the exact EVBS quantile function.
gevreg.fit(), gev_scores() and cnc_diagnostics_gev() extend the local influence framework to the generalized extreme-value regression model.
evbs_block_boot() provides moving-block bootstrap standard errors for series with residual serial dependence.

The extreme-value index of the EVBS response is 2 * gama, not gama. The transformation preserves the max-domain of attraction but doubles the tail index. Documentation updated accordingly.
The GEV return-level formula does not apply to the EVBS response. Use qevbs() or evbs_return_level().
gevreg.fit() centres are strongly recommended: the GEV likelihood is poorly conditioned for uncentred covariates and may silently fail to converge.

Initial CRAN Release - Complete implementation of local influence diagnostics for Extreme-Value Birnbaum-Saunders (EVBS) regression models
Estimation - evbsreg.fit() function for joint maximum likelihood estimation of EVBS regression models with flexible parameter specification
Diagnostics - Conformal normal curvature-based local influence diagnostics under three perturbation schemes:
- Case-weight perturbation
- Response variable perturbation
- Explanatory variable perturbation
Residuals - Randomized quantile residuals (rcoxsnell(), rqrandomized()) with simulation envelopes for model validation
Visualization - Publication-quality diagnostic and density plots:
- plot_cnc() for local influence plots
- envelope_qq() for quantile-quantile plots with envelopes
- plot_evbs_alpha() and plot_evbs_gama() for parameter density visualization
- plot_aggregate_contributions() for influence aggregation
- plot_normalized_eigenvalues() for eigenvalue analysis
Monte Carlo Utilities - generate_evbs_data() and generate_logevbs_data() for simulation studies
Random Number Generation - revbs() for generating random variates from EVBS distributions with flexible GEV parent distributions

The methods implemented in this package are described in:

Application to real-world data:

Monthly maximum wind gust data from Itajai, Brazil (included in itajai dataset)

For more information, visit: https://raydonal.github.io/evbsreg/