NB() gains the wild bootstrap: type = "wild" (default "pairs", unchanged), with weights ("rademacher", "mammen", "webb", "normal"), hc ("hc3", "hc0", "hc1", "hc2", "hc4", "hc4m", "hc5"; the same leverage adjustments as betaSandwich), and x ("random" resamples the regressors with their residuals, "fixed" keeps them). The effect size functions and intervals work on the wild draws unchanged. The sampler is in R for now.
New FSqNB() (Cohen’s f-squared and the local f-squared of each regressor) and SCorSqNB() (the image of a signed semipartial interval under squaring); equations and definitions in the help pages of the effect size functions; a package help page.
Faster bootstrap: the bootstrap and jackknife covariance matrices (closed-form leave-one-out) and the effect sizes are computed in C++ (Rcpp, RcppArmadillo), once per NB() call. The bootstrap samples are the same as before for the same seed.
Minor
DeltaRSqNB() gains adj = TRUE for differences in adjusted R-squared. confint() and summary() gain type = "mpc" for modified percentile intervals of Delta R-squared (sr2), unadjusted or adjusted. The sample F test uses the interval’s alpha and the unadjusted increment. Upper limits and ordinary percentile interpolation are unchanged. Existing defaults and "pc", "bc", "bca" intervals are unchanged.
Selective standardization: the arguments std_y and std_x standardize only y or only the named regressors (for example, leave a dummy variable in its 0-1 units). The defaults give the same results as before.
The thetahatstar and jackknife elements of nb and betanb objects are now matrices (one draw per row) instead of lists. nb objects also hold es and es_jackknife.
The internal copies from linearAlgebra 0.0.5 (a faster duplication matrix and its inverse), nBootstrap 0.1.0 (quantiles in C++), strRegression 0.0.2, and processLM 0.0.11 are current.
Documentation
The examples name every argument and call stats::lm(). The Algina et al. references of DeltaRSqNB() are in full APA form (2008a, 2008b).