depower 2026.11.11

Remove trailing slash on void elements to pass CRAN vnu HTML check.
Fix the asymptotic variance of the ratio in wald_test_bnb(). The function used Eq. (10) of Rettiganti and Nagaraja (2012), which omits a factor of the mean from one term of the denominator. The function now uses the inverse of the Fisher information matrix in the appendix of that paper, which gives ratio * (1 + ratio) / (n * mean1).
- The test statistic, p-value, and confidence interval change. The size of the change depends on the mean and the dispersion parameter, and it grows with the dispersion parameter.
- The results of power() also change, because wald_test_bnb() is its default test for BNB data.
- The variance can no longer be negative, which removes that cause of missing p-values.
- The default test now has type I error near the nominal level for small samples. For example, the asymptotic test at n = 4, mean1 = 9.75, and dispersion = 2 has type I error 0.05, compared with 0.07 before this fix.
Count near-equal test statistics as ties in randomization and simulated p-values. Numerical optimization puts small errors in the test statistics, and exact randomization p-values could change by one count because of these errors. The comparison now uses a relative tolerance of 1e-3, which can only increase the p-value.

depower 2026.1.30

Add functions add_power_ci() and add_power_pi() for confidence intervals and prediction intervals of the power estimate, respectively.
- They add uncertainty intervals for power to the object return by power().
- Function plot.depower() will add shaded regions for the intervals.
Add functions eval_power_ci() and eval_power_pi() for confidence intervals and prediction intervals of the power estimate, respectively.
- In contrast to the add_power_*() variants, these are for planning power simulation studies.
- Assess uncertainty in hypothetical simulation-based power estimates before committing computational time to a simulation run.

Add functions asymptotic() and simulated() for specification of type of test.
Add argument distribution to functions:
- wald_test_nb()
- wald_test_bnb()
- lrt_nb()
- lrt_bnb()
for control of running an asymptotic, randomization, or parametric simulation test.

Fix bug where equal_dispersion argument in function lrt_nb() was not passed downstream and instead fixed to FALSE.
Fix calculation of confidence limits for mean2 and dispersion2 in functions glm_nb(), glmm_bnb(), and glmm_poisson().

First release to CRAN.

Power
- power()
Plot
- plot.depower()
Simulate
- sim_log_lognormal()
- sim_nb()
- sim_bnb()
Test
- t_test_welch()
- t_test_paired()
- wald_test_nb()
- wald_test_bnb()
- lrt_nb()
- lrt_bnb()
- glm_nb()
- glmm_bnb()
- glmm_poisson()
MLE
- mle_nb_null() and mle_nb_alt()
- mle_bnb_null() and mle_bnb_alt()
Log-likelihood
- nll_nb_null() and nll_nb_alt()
- nll_bnb_null() and nll_bnb_alt()

Proof of concept.