doudpackage 2.2.0
Bug fixes
- A statistical test that cannot be computed no longer aborts the
whole table. The p value is
NA, a warning names the
variable, and the rest of the table is produced. This used to fail with
non-numeric argument to mathematical function for a
constant variable, a factor with a single or an empty level, an
all-NA column, a quantitative variable with a single
observation, or a group with a single level.
- Unused levels of the group variable — typically left behind by a
subset of the data — are now dropped with a warning instead of producing
a whole column of
NaN (NA).
group_rows_labels now reorders the rows so that the
variables of a label are contiguous and the labels follow the order they
were given in. Row spans no longer overlap.group_rows_labels now matches whole variable names: a
label on age no longer drags in the rows of
age_group.group_rows_labels referring to a variable that has no
row in the table used to silently empty the whole table; it is now an
error naming the variables available.parseClassFun() failed with levels_to_keep
because filter() resolved to
stats::filter.digits.ql was ignored for the Total column of the
missing values of a quantitative variable.- A
tibble is accepted, as the documentation already
claimed.
["table"] is always a data.frame; without
a group it used to be a tibble.group = NULL now means the same as no group.quali = FALSE together with quanti = FALSE
is reported as such instead of failing with
object 'lst_VarGroup.quanti' not found.col.order no longer fails when the table has no
group.- Variables of an unhandled type (character,
Date,
logical…) raise a warning and are ignored instead of throwing an
error.
- Ordered factors are handled as factors.
- Rows with a missing value in the group variable are removed, with a
warning.
- Bug correction when parsing the table if only the group variable had
two levels.
- The row names of
["table"] are no longer the leftovers
of intermediate filtering.
- Argument checking gives an explicit message for an unknown group
variable, a group that is not a factor, a group with fewer than two non
empty levels, an empty dataset, and duplicated columns in
col.order.
New features
normality = "assess" is implemented: a Shapiro-Wilk
test decides variable by variable between mean (SD) with a parametric
test, and median (IQR) with a non parametric one.anaBiv() on a data.frame works; it accepts
normality and digits.p. It previously failed
for every input.[ on a VarGroup (the elements returned by
anaBiv()) also gives access to "name",
"type" and "normal".parseClassFun() takes a font argument,
instead of always using arial.- Parallel processing added as an option (
parallel,
mc.cores).
Breaking changes
normality = "manual" is no longer accepted. It has
never done anything but raise an error; use "assess",
"normal" or "non normal".
Other
- A
testthat suite covers the analyses, the rendering and
every bug above.
- The package no longer imports
stringi.
doudpackage 2.0.1
Major changes
- Added more than 2 sub-groups
- Choice between normal and non normal variables to use either t
test/Wilcoxon or ANOVA/Kruskal-Wallis
Minor changes
- Faster implementation (no loop used)