In the GitHub version, estimator_AS_tidy() returns
standard errors, p-values, and 95 percent confidence intervals.
interference::estimates() computes effect variances only
when told which condition is control, and rdss had never passed
control_condition, so the variances were always empty. The
Horvitz-Thompson rows use Aronow and Samii’s conservative variance, the
larger of the Horvitz-Thompson and constant-effects variances, since the
plain Horvitz-Thompson variance of a difference can be negative.
Intervals for an effect whose exposure condition is rare can undercover;
the documentation gives the simulated rates.
On CRAN, estimator_AS_tidy() is a stub that says how
to install the GitHub version, as it was in 1.0.14, because the
‘interference’ package it wraps is not in any repository CRAN can reach.
The GitHub version carries the working function. The stub’s message now
points at DeclareDesign/rdss itself; 1.0.14’s pointed at a
remotes branch that no longer exists.
post_stratification_helper() passes
re.form = NA to marginaleffects::predictions()
only when the fit is a mixed model. With an lm_robust() or
other single-level fit, marginaleffects warned that re.form
is not a known argument on every call.
did_multiplegt_tidy() tidies
did_multiplegt(mode = "dyn"), returning each effect with
its standard error and confidence interval. The book’s
mode = "old" call returns NaN under
DIDmultiplegt 2.1.0, which first differences with
stats::lag() and so never sees a treatment switch; the
function now warns when that happens. mode = "dyn" needs
the ‘polars’ package, from https://rpolars.r-universe.dev.
estimator_AS_tidy() computes exposure probabilities
from permutatation_matrix again, so the book’s chapter 18
declaration produces estimates rather than nothing. Its argument list
returns to the one 1.0.14 shipped; the version on the main branch had
renamed the argument and moved the computation out to the caller, which
no published code passed.
estimator_AS_tidy() takes optional precomputed
exposure probabilities, obs_prob_exposure. They do not
change across simulations, so computing them once and passing them in
avoids recomputing them on every draw. Without it the function computes
them from permutatation_matrix as before.
estimator_AS_tidy() returns its explanatory message
instead of erroring when ‘interference’ is absent.
estimator_AS_tidy()’s documented argument now
matches its signature.
Declare the R >= 4.1.0 dependency the code already has, through its use of the native pipe. CRAN has noted this on all 13 flavors.
hex_add_alpha(col, 1) returns a valid eight-digit
color; it produced nine digits because floor(1 * 256) is
256.
Every help page is revised for a first-time reader: each says
what the function does and what it returns, names the columns it needs
(Y, Z, tau, subject,
task, profile), and links to related
functions. The dd_palette() page lists the palettes that
exist, fairfax has 238 rows (not 236), and the datasets
list their columns.
Remove a second, unused definition of tidy_stan().
The deprecated one, which calls broom.mixed::tidy(), was
already the one in effect.
A Get started article on the package website lists the rdss functions each section of the book uses, and works through the conjoint declaration. The datasets cite their sources, and the package has a hex logo.
Point URL and BugReports at the GitHub
repository.
Drop a stray zero-byte _pkgdown 2.yml, and stop
shipping README.Rmd, from the source tarball.
tidy_stan() in favor of
broom.mixed::tidy().get_rdss_file() to download the book’s
declarations.lapop_brazil dataset, resampled from the LAPOP
AmericasBarometer 2018/19 survey of Brazil. Used in the RDSS
exercises.