Changelog
Source:NEWS.md
nonprobsampling (development version)
Unified solver diagnostics in
print(),summary(), and verbose logs. These outputs distinguish the stopping reason from whether the final equation residual meetsftol. Fitted objects now record the solver’s residual and the tolerance used to assess it.Added guidance on harmonizing factor levels, contrast coding, and reference categories across samples before creating survey design objects. The help page and vignette clarify that contrast coding is not checked by the package.
Added standardization guidance to the
est_pw()documentation: use the same centering and scaling constants across all datasets before estimation, rather than standardizing each dataset separately or usingscale(x)directly inp_formula. Solver diagnostics now suggest standardizing continuous auxiliary variables with substantially different scales whennleqslvdoes not converge sufficiently.Fixed variance calculations when reference survey data contain columns with internal helper names such as
h_x. Calibration, ALP, CLW, and multi-reference calibration now avoid these name collisions, preventing unrelated columns from changing standard errors and confidence intervals.Fixed cumulative precalibration with multiple reference surveys to retain variables shared by only some surveys. Each survey now uses its own formula variables, allowing calibrated targets to pass through intermediate surveys even when no variable occurs in every survey.
Fixed missing-value handling for character domains in
pwmean(): empty and whitespace-only strings are now detected before applyingna.actionin the pseudo-weighted and unweighted estimation paths. Omitted row indices include these values, andna.failrejects them.Corrected the
sp1_bootstraphelp example to usecombined.weights = TRUE, because the bootstrap replicate weights already include the main survey weight.Added a
confint()method forpwmeanobjects, inherited bypwmean_factor, providing normal-approximation confidence intervals with configurable confidence levels and estimate group selection.Added a
coef()method forpwmeanobjects to extract pseudo-weighted means.pwmean_factorobjects inherit the method and return category prevalence, named by category and domain.Added a
coef()method forpw_fitobjects to extract participation model coefficients.