beezdemand

broom Integration

• New augment() methods for all model classes provide fitted values and residuals in a tidy tibble:
• augment.beezdemand_fixed(): Returns .fitted, .resid
• augment.beezdemand_hurdle(): Returns .fitted, .fitted_link, .fitted_prob, .resid, .resid_response
• augment.beezdemand_nlme(): Returns .fitted, .resid, .fixed

Model Comparison Framework

• New compare_models() function for unified model comparison across all beezdemand model classes. Reports AIC, BIC, delta_AIC, delta_BIC, and performs likelihood ratio tests when models are from the same backend and nested.
• New anova() S3 methods for comparing nested models:
• anova.beezdemand_hurdle(): LRT for nested hurdle models
• anova.beezdemand_nlme(): Delegates to nlme::anova.lme()

Model Diagnostics Suite

• New check_demand_model() generic with methods for all model classes. Performs comprehensive diagnostics including convergence checks, boundary condition detection, random effect variance assessment, and residual outlier detection. Returns structured diagnostics object with issues and recommendations. (Named check_demand_model() to avoid conflict with performance::check_model().)
• New plot_residuals() function creates diagnostic plots: residuals vs fitted, histogram of residuals, and Q-Q plots. Works with all model classes via the augment() infrastructure.
• New plot_qq() function creates Q-Q plots for random effects to assess normality assumptions in hurdle and NLME models.

Normalized Alpha (Alpha Star)

• All model classes now compute alpha_star (normalized alpha, Strategy B; Rzeszutek et al., 2025), which makes the elasticity parameter comparable across different values of k. Available in FitCurves() output (columns alpha_star and alpha_star_se), tidy() on beezdemand_fixed objects, and tidy() on beezdemand_hurdle objects. Standard errors are obtained via the delta method. See ?param-registry for details.

Modern Wrappers for Legacy Functions

• New get_empirical_measures() as a modern replacement for GetEmpirical(). Returns a beezdemand_empirical S3 object; access the results via $measures.
• New get_descriptive_summary() as a modern replacement for GetDescriptives(). Returns a beezdemand_descriptive S3 object; access the results via $statistics.
• New get_k() as a modern replacement for GetK(). Returns a single numeric k value with optional verbose output.

Other New Features

• New confint() methods for extracting confidence intervals from all model classes: beezdemand_fixed, beezdemand_hurdle, beezdemand_nlme, and cp_model_nls.
• New migration guide vignette (vignette("migration-guide")) documenting the transition from FitCurves() to fit_demand_fixed().

Breaking Changes

• summary() methods for beezdemand_hurdle and beezdemand_nlme now return structured summary objects instead of printing directly. Use print(summary(fit)) for console output. Programmatic access is now possible: s <- summary(fit); s$coefficients.
• fit_demand_hurdle() now fits demand parameters in natural-log space (log_q0, log_alpha, log_k) and reports back-transformed values; the param_space argument has been removed.
• fit_cp_nls() now uses log10-parameterized optimizer coefficients (log10_qalone, I, log10_beta) across equation forms; the "exponential" form fits on the log10(y) response scale and filters y <= 0 with a warning. predict.cp_model_nls() now always returns y_pred on the natural y scale; for "exponential" it additionally returns y_pred_log10 (and no longer returns y_pred_natural).

Additional New Features

• New fit_demand_fixed() function provides a modern interface for individual demand curve fitting. Returns a structured S3 object with summary(), tidy(), and glance() methods. This wrapper offers the same functionality as FitCurves() but with a standardized API.
• New systematicity wrappers with unified output vocabulary:
• check_systematic_demand() for purchase task data (wraps CheckUnsystematic())
• check_systematic_cp() for cross-price data (wraps check_unsystematic_cp())

Both return beezdemand_systematicity objects with identical column

schemas (differing only in NA values for domain-specific fields).

• First-class tidy() and glance() support is now guaranteed across all beezdemand model classes. All methods return tibbles with standardized columns including model_class and backend.
• All summary objects now inherit from beezdemand_summary base class, enabling shared fallback behavior and consistent field availability.

API Standardization

This release introduces **Stability Contracts** for all model classes:

• **summary() objects** now return structured S3 objects with class c("summary.<class>", "beezdemand_summary"). Required fields include: call, model_class, backend, nobs, n_subjects, converged, logLik, AIC, BIC, coefficients (tibble), notes.
• **tidy() methods** return tibbles with columns: term, estimate, std.error, statistic, p.value. Multi-part models include a component column (e.g., "fixed", "variance", "derived").
• **glance() methods** return 1-row tibbles with columns: model_class, backend, nobs, n_subjects, converged, logLik, AIC, BIC.

API Changes

• fit_cp_nls() and fit_cp_linear() now accept x_var/y_var to map non-standard column names to canonical ones ("x", "y"). fit_cp_linear() additionally accepts id_var, group_var, and target_var. Default behavior is unchanged when these arguments are omitted.
• fit_cp_linear() gains explicit filter_target and target_level top-level arguments (previously these were handled implicitly via validate_cp_data()). Existing calls without these arguments are unaffected.
• fit_cp_nls(start_vals=) is deprecated in favor of start_values=. The old argument still works but emits a deprecation warning.

---