New function alphaN_power() computes the power of the calibrated test
against a standardized effect of size d: the design-time companion of
alphaN() (noncentral t for single coefficients, noncentral F for joint
tests, at the residual degrees of freedom). For model-specific effect
parameterizations (odds ratios, rate ratios), the calibrated alpha plugs
directly into the calculators of the pwrss package via their alpha
argument; see ?alphaN_power.
New function alphaN_power_plot() draws power against sample size, one
panel per effect size, with every calibration method evaluated at its own
alpha(n) and a fixed reference level as a dashed curve.
alphaN_report() gains a power_at argument: the settings report can now
document the power of the calibrated test against the effect sizes the
researcher cares about, alongside the alpha itself.
alphaN_report() now also states the scope of the "ES" and "moment"
calibrations: exact for the normal linear model when p is supplied
(evaluated at the effective sample size n - p), the conservative
large-sample form otherwise, and asymptotic for other generalized linear
models.
New function alphaN_report() writes a preregistration-ready Markdown
settings report: every input behind a calibrated alpha (sample size,
evidence target, method, prior settings), the resulting alpha, the
decision rule, and the references to cite. It mirrors the downloadable
report of the companion Shiny application, can write straight to a file,
and hard-wraps its lines (default 72 characters, see width) so the
report reads well in consoles, files, and rendered documents.
JAB() called without a covariate now returns a named vector with the
Bayes factor of every coefficient except the intercept (which remains
available on explicit request).
alphaN_plot() and JAB_plot() have been restyled: colorblind-safe
Okabe-Ito palette, light grid lines, open axes, and horizontal axis
labels. alphaN_plot() additionally gains a log argument ("x", "y",
"xy") which helps when the fast-falling "moment" curve is drawn next to
the others, and its tick labels always use plain notation ("0.0001" and
"10,000", never "1e-04").
New function klauerBF() exports the effect-size and moment Bayes factors
of Klauer, Meyer-Grant & Kellen (2025) that alphaN() inverts: from a
t statistic (one-sample test or single regression coefficient) or from an
F statistic for a joint test of q coefficients.
alphaN() (methods "ES" and "moment") gains arguments q and p.
With q > 1 the alpha level is calibrated for the joint F test of q
coefficients through the exact regression-case Bayes factors of Klauer et
al. (2025, Table 4), implemented natively including their Gaussian
hypergeometric term; p sets the number of retained model parameters so
that small-sample calibrations can use the effective sample size n - p
(residual degrees of freedom). The defaults (q = 1, p = 0) reproduce
the previous behavior exactly. The moment-prior default nu is now
5 + (q - 1) and the ES-prior scale recommendation generalizes to
r = sqrt((nu - 2)/(nu * q)) * de, both following the paper (unchanged
at q = 1).
New function n_effective() computes the effective sample size
n * (se/se_robust)^2 that Wulff & Taylor (2024) recommend as a
sensitivity check when calibrating alpha with clustered (panel) data.
alphaN_plot() gains a methods argument and can now draw the "ES"
and "moment" curves alongside the prior-fraction methods.
The regression-case implementation is validated against all printed Bayes factors in Table 8 of Klauer et al. (2025), in addition to the existing Table 7 anchors, and the q = 1 F form agrees with the validated t form to near machine precision.
The quadrature and inversion machinery is additionally stress-tested
against an independent oracle that integrates on the original effect
scale over an infinite range with a plain density ratio (none of the
package's substitution, windowing, or log-clamping choices); the
normal-limit switch at n - p = 50,000 is measured directly by running
both branches at the same effective sample size, and monotonicity of
alpha in n and BF, and of the Bayes factors in the test statistic,
is checked over grids that cross the switch, including joint tests and
small residual degrees of freedom.
?alphaN states the model scope of the "ES" and "moment" methods:
exact under the normal linear model, asymptotic (like the prior-fraction
methods) for other generalized linear models.citation("alphaN") now also lists Klauer et al. (2025) for users of the
"ES" and "moment" methods.alphaN() gains two methods based on Klauer, Meyer-Grant & Kellen (2024,
Psychonomic Bulletin & Review, doi:10.3758/s13423-024-02612-2):
method = "ES" calibrates alpha to their effect-size Bayes factor, whose
prior centers the alternative hypothesis on a prespecified effect size, and
method = "moment" calibrates alpha to their moment Bayes factor, under
which effects near zero are a priori implausible. New arguments de
(targeted effect size, default 0.5), nu, and r control the priors, with
defaults following the paper's recommendations. Because the moment prior
rules out near-zero effects, the alpha level it implies falls much faster
with n than under JAB.method = "ES", nu = 1, de = 0 with an explicit r
calibrates alpha to the default (Jeffreys-Zellner-Siow type) Bayes factor of
Rouder et al. (2009).alphaN() and JABt() now return correct results when n is a vector and
method = "robust" or method = "balanced". Previously, "robust" silently
applied the smallest sample size to every element and "balanced" failed
with an unrelated error.method, a missing df in JABp(..., z = FALSE),
a p outside (0, 1], a non-positive n or BF, and an unknown covariate
in JAB() (which now lists the coefficients available in the model).JAB_plot() gained an upper argument, passed on to the underlying
computations for method = "balanced".alphaN_plot() gained a ylim argument. The default now covers all four
curves; previously the y-axis was fixed to (0, 0.05), which silently clipped
the "balanced" curve for small Bayes factors.JAB() now determines the sample size via nobs().?JABp no longer has a placeholder title.JABp() expects a
two-sided p-value.NEWS.md file to track changes to the package.