RR(), OR(), and RD() now return effects in the conventional direction: risk/odds of the non-reference (intervention) arm over the reference arm, and risk(non-reference) - risk(reference) for risk difference. Previous versions returned the reciprocal for RR/OR and a sign-flipped value for RD. Numerical outputs from prior versions are not directly comparable. Confidence-interval endpoints (RRLL/RRUL, ORLL/ORUL, RDLL/RDUL) follow the same convention.
make_two_by_two_() now returns the non-reference treatment in row 1 and the reference treatment in row 2, matching the convention assumed by the effect estimators. Code reading the matrix by rownames is unaffected; code indexing by position needs to be reviewed.
Bug fixes
make_two_by_two_() now deduplicates by (subject, treatment) instead of by subject alone. This fixes incorrect 2×2 cells for designs where the same USUBJID appears under more than one treatment value (e.g. crossover); parallel-group analyses are unaffected.
chefStats 0.1.1
Centralized the Git-Actions to {ramnog}
chefStats 0.1.0
Initial release of chefstats
Contains an initial selection of statistical functions for use in the {ramnog} ecosystem.
