Enhancing {ggplot2} plots with statistical analysis 📊📣
pairwise_comparisons() function for carrying out one-way pairwise
comparisons has now moved in {ggstatsplot} from {pairwiseComparisons}
package.A number of effect size estimates and their confidence intervals have
changed due to respective changes made in {effectsize} package version
0.5 release. For full details of these changes, see:
https://easystats.github.io/effectsize/news/index.html
For the same reason, the effect size for one-way contingency table has changed from Cramer's V to Pearson's C.
For plotting marginal distributions in ggscatterstats, {ggstatsplot} now
relies on ggside package instead of ggExtra. This was done to remove a
glaring inconsistency in the API. All functions in {ggstatsplot} produced
ggplot objects and could be further modified with ggplot2 functions,
except ggscatterstats, which led to a lot of confusion among users (e.g.
#28). This change gets rid of this inconsistency. But it comes at a cost:
there is no more marginal.type argument that lets you change the type of
marginal distribution graphic and histogram is the only possible option.
Note that this is not a breaking change. Your past code will continue to
work but it will now always produce a histogram instead of other marginal
graphic you might have chosen.
Minimum needed R version is now 4.0.
Online vignette about combine_plots has been removed. In case you want to
create a grid of plots, it is highly recommended that you use patchwork
package directly and not this wrapper around it which is mostly useful with
{ggstatsplot} plots.
ggscatterstats labeling arguments accept only unquoted inputs now, and not
quoted or string inputs. Allowing this was a bad design choice in the past
since most functions in {ggstatsplot}, inspired by tidyverse, expect
unquoted (x) - and not quoted ("x") - arguments. So this function was
the odd one out.
Gets rid of ipmisc dependency.
Removes movies_wide dataset, which was virtually identical to
movies_long dataset and was not used anywhere in the package. Also removes
the unused VR_dilemma dataset.
extract_stats function to extract dataframes containing statistical
details.There is finally a publication for {ggstatsplot} package!
https://joss.theoj.org/papers/10.21105/joss.03167
The ggcoefstats function defaults to NULL for xlab and ylab
arguments, which lets users change these labels if they wish to do so.
Additionally, the x-axis label, if not specified, now defaults to
"estimate". Whether this estimate corresponds to regression coefficient or
effect size like partial eta-squared should be clear from the label itself.
To reduce the dependency load, ggcorrplot moves from Imports to
Suggests.
The bar.fill argument in gghistostats is retired in favor of the new
bin.args argument that can be used to pass aesthetic arguments to
ggplot2::stat_bin.
ggstatsplot.layer argument has been retired. If the user chooses a
certain ggplot2 theme, it means they want that theme, and not
{ggstatsplot}'s varnish on it. So the previous behavior was undesirable.
This is a backward compatible change, so the plots should not look
different.
The pch size for ggcorrmat has been increased to 14 (#579) to increase
its visibility compared to the correlation value text.
ggwithinstats gains point.args to change geom_point.
Minor change to ggcorrmat legend title - content in parentheses is now
shown outside of it.
ggcoefstats didn't work when statistic for the given model was
chi-squared. This has been fixed.To reduce the dependency load, ggExtra moves from Imports to
Suggests.
All functions are more robust in the sense that when statistical analysis
fails, they will return only the plots with no subtitles/captions. This
helps avoid difficult-to-diagnose edge case failures when the primary
functions are used in grouped_ functions (e.g., #559). The ggpiestats
and ggbarstats functions always behaved this way, but the rest of the
functions now also mimic this behavior.
ggcoefstats labels do not contain degrees of freedom when they are not
available instead of displaying Inf.