Removed the dependency on ufs, which is
scheduled for archival on CRAN. mg_rel_table() no longer
calls ufs::scaleStructure(); the internal consistency
measures are now computed inside ccpsyc using lavaan
and psych, both of which were already
imported.
clearing_fa() no longer calls
xlsx::read.xlsx(). The xlsx package was
not declared in Imports, which caused an
undeclared-dependency check failure. Excel key files are now read with
readxl (moved to Suggests), which also
removes the Java requirement. Supplying an .xlsx key file
without readxl installed now raises an informative error.
calculate_dif_indices() is now actually exported and
documented; it was tagged for export but missing from
NAMESPACE.
Declared stats in Imports and dropped
the unused rlang dependency.
Removed the redundant data/example.csv, which
duplicated data/example.rda and defined the same
example object.
release_bonferroni() passed the rlang
.data pronoun as a data argument to
dplyr::filter(), causing the function to error.
splitgroup() used an invalid column selection that
could not work as written; it now drops the grouping column and returns
the remaining ones.
pancultural() signalled an error with
print() rather than stop(), and coerced
nfactors to character before passing it to
psych::principal().
gamma_hat_scaled() no longer computes an unused
quantity.
The glb column of mg_rel_table() now
reports the maximum split-half reliability (Guttman’s lambda 4) via
psych::splitHalf(). The previous route,
psych::glb(), calls psych::ICLUST(), which
draws a diagram even when plot = FALSE and therefore fails
whenever no graphics device has been initialised. Values are very close
to, but not always identical to, those reported previously.
Alpha and omega confidence intervals are now computed
analytically rather than through the previous wrapper. Alpha reproduces
the normal-theory interval exactly; omega is estimated from a
single-factor congeneric model with a delta-method interval. The
seed argument of mg_rel_table() is retained
for backward compatibility but is no longer used, and a
conf.level argument has been added.
Measures that cannot be estimated for a group (fewer than three
items, or a model that fails to converge) now return NA
instead of erroring.
dMACS(),
mg_rel_table() and
calculate_dif_indices().MNCI() and
gamma_hat_scaled() are no longer given .Rd
pages, which lacked \value sections.mg_rel_table() and added tolerances to
the existing numerical tests so they are robust across BLAS
implementations.