- First CRAN submission.
- Improve detection of interactions involving the exposure variable for
"margstd_delta"
.
tidy.risks()
: Increase defaultbootrepeats
to 1000, consistent withsummary()
.
- Breaking change: For consistency, the default option for model fitting
(
approach = "auto"
) now always uses marginal standardization after fitting a logistic model. The previous approach, which relied on different models fitted, is available asapproach = "legacy"
. - If requesting
approach = "margstd_delta"
in presence of interaction terms involving the exposure variable, a warning is displayed. Model fitting with"auto"
uses the bootstrap (i.e.,"margstd_boot"
) in that case. approach = "margstd_boot"
bug fix: Keep categorical exposures of type factor in the correct order.- Include
breastcancer
dataset in the package. - Internal changes:
- {addreg} and {logbin} are now soft dependencies (
Suggests:
instead ofImports:
) - Remove {lifecycle} dependency
- Compatibility with tidyselect 1.2.0 variable selection
- {addreg} and {logbin} are now soft dependencies (
- Breaking changes:
- Rename
approach = "glm_start"
to"glm_startp"
(for Poisson). - Rename
approach = "margstd"
to"margstd_boot"
. - For consistency with other approaches, no longer treat numeric variables
with only two levels (e.g.,
1
and2
) as categorical inapproach = "margstd_boot"
.
- Rename
- New estimators:
approach = "margstd_delta"
, marginal standardization after fitting a logistic model with standard errors via the delta method.approach = "margstd_boot"
now also implements average marginal effects to handle continuous exposures.approach = "duplicate"
, the case duplication method for risk ratios, proposed by Miettinen, with cluster-robust standard errors proposed by Schouten et al.approach = "glm_startd"
, using the case duplication-based coefficients as starting values for binomial models.rr_rd_mantel_haenszel()
: New function for comparison purposes.
- Changes to parameters:
approach = "auto"
, the default, now attempts model fitting in this order of priority:approach = "glm"
;approach = "glm_startp"
(for risk ratios only);approach = "margstd_delta"
. If all fail, the user is shown the error messages from a plain logistic model.- Bootstrap repeats (
bootrepeats
) forapproach = "margstd_boot"
now default to1000
.
- Bug fixes:
summary.robpoisson()
: Fix sandwich standard errors.tidy()
output was correct.
- Programming changes:
- Do not attach the logbin package to the namespace; export
logbin::conv.test()
on its behalf. Move MASS package (needed only for testthat) toSuggests
. - Remove usage of unexported functions from
stats
. - For
approach = "margstd_boot"
, avoid two rounds of bootstrap for standard error and confidence intervals separately. Rewrite internal fitting functionfit_and_predict()
, replacingeststd()
. Overall, bootstrapping is more than two times faster now.
- Do not attach the logbin package to the namespace; export
tidy(bootverbose = TRUE)
: For BCabootstrap confidence intervals, also returnjacksd.low
andjacksd.high
, the jackknife-based Monte-Carlo standard errors for the upper and lower confidence limits.riskdiff()
: Remove leftover "logistic" parameter.summary.risks()
,tidy.risk()
: fix error handling if no model converged.
- Fix bugs in
bootci = "normal"
and insummary.risks()
. - Return name of dataset.
- Expand test coverage.
- Expand bootstrapping options after marginal standardization:
- Parametric BCa bootstrap confidence intervals via the bcaboot package are the new default.
- Parametric normality-based intervals are a backup.
- Non-parametric bootstrapping with BCa intervals is retained as an option for completeness.
- Remove precision
weight
option. - Expand documentation.
- First release