211 lines (162 sloc)
- Ensure compatibility with broom 0.7.0 (#32)
- Update infrastructure using
usethis::use_tidy_ci(), bumps R dependency to
z.test() to fill the niche in
stats. Useful to verify hand-made calculations done in educational contexts, as I assume nobody else will ever need a z-test function.
tadaa_z.test() while we're at it.
- Remove internal Levene-test from
tadaa_t.test(), defaulting to
var.equal = FALSE as
stats::t.test() does. Turns out this was a pretty bad idea due to the multiplicity problem, and the Welch-test is good enough.
- Remove the post-hoc / empirical
tadaa_normtest(). It never should have been a thing in the first place.
Fixes & Features
inv() for inverting Likert-scales
pval_string() conform to APA guidelines (no more leading zeros)
- More accurate R dependency (
R >= 2.10) (i.e. more inclusive)
- Add a bunch more tests
ord_pairs to retrieve $N_c$, $N_d$, ties and total number of pairs for contingency tables. Internals for this function are straight up copied from this gist.
etasq in case you want to show students what $\eta^2$ is without having to explain ANOVA.
ord_somers_d not returning the correct value for
symmetric = TRUE
sjlabelled::zap_labels is a thing
ryouready dependency, in favor of the more versatile
vcd dependency, also in favor of
DescTools functions are very similar to what the wrappers do, so the wrappers might be removed in the future.)
lazyeval dependency in
tadaa_int by being better at
dplyr dependency by being better at R.
haven dependency by not re-exporting
as_factor anymore ¯\_(ツ)_/¯
sjmisc depencency because why did we depend on that again?
sjlabelled dependency (only used for re-exports).
lsr dependency in favor of, you guessed it,
DescTools for eta in
- Fix error in SEM calculation in
tadaa_one_sample for t-tests
- Make sure
tadaa_-test functions use
tadaatoolbox::pval_string instead of the
- Remove recoded
leist var from
ngo, as it should be computed from
leistung by students.
Suggests because we only need it in one function, sometimes.
- More compact table output in
[tadaa_]likertize is removed. Use
labels_to_factor is removed because various
- Silence warnings in functions using
ord_tau to calculate all the Taus
Under the Hood
tadaa_chisq for a $\chi^2$-Test with OR and effect size.
- Switched from
sjmisc to the new
sjlabelled (Thanks, @strengejacke!)
- Also re-export
magrittr::%$% because it's really handy sometimes.
- Fix #30: Undocumented arguments in
Tweaks, Patches & Bug Fixes
tadaa_pairwise_t as an extension of
stats::pairwise.t.test that works with two grouping factors and thereby can test interactions.
- Also knows the Sidak method for p-adjustment, both regular and step-down procedures.
- See this and that
tadaa_pairwise_gh for the Games Howell post-hoc procedure as an alternative to
tadaa_pairwise_tukey while we're at it. Just a thin wrapper for
stats::TukeyHSD but with tidied output and usage consistent with the previous
tadaa_plot_tukey to plot Tukey HSD results as error bars because boy do I like error bars.
tadaa_balance as a replacement for
tadaa_heatmap to check equality of group sizes.
magrittr as all the cool kids to these days.
magrittr because I happen to really like it.
sjmisc, as they're handy.
haven as a replacement for the deprecated
- Is now an alias for
theme_readthedown, will probably become the new canonical version.
- Now finally adds vertical space to the
x axis title via proper margining.
type is now
3, for generally safer results and consistency with SPSS.
check_contrasts option for
type = 3 sums of squares, which ensures each non-ordered factor has
- Now auto-factorizes independent variables by default, fixes #24.
- Now imports
methods, which should fix an issue during
rmarkdown processing where the function
is couldn't be found. If not, manually
library(methods) as a workaround.
- Fix wrong
sprinkle labelling causing
eta.sq to be formatted like a p-value.
show_power argument to calculate power via
- Requires more testing against software like G*power to ensure accuracy.
- Internal Levene test now uses
center = "median" for more robust results, as it should.
- Now also uses $\alpha = 0.05$ instead of $\alpha = 0.1$.
- Use new argument
var.equal to override internal Levene test.
- Power should now be properly reported for
alternative = "less" or
conf.level argument used for CI and power calculations
- Now doesn't return the absolute effect size by default.
paired argument so effects for paired tests are now a thing.
- Also fix
direction argument not being honored.
print (logical) argument to suppress printing if so desired. The output will still be returned invisibly.
tadaa_one_sample: Should make sense now.
na.rm argument from
tadaa_wilcoxon because it's problematic, and in case of
paired = TRUE it would have produced flat out wrong results.
- Documentation improvements
print = markdown output of
print = "console" now has headers with unparsed $\LaTeX$-expressions, but who uses that anyway.
labels_to_factor: Was a wrapper around
haven::as_factor and is obsolete by now, as
as_factor can do the same thing this function was built for.
tadaa_likertize is renamed to
likertize, deprecated since
sjmisc::split_var is probably better anyway.
Tweaks, Patches & Bug Fixes
tadaa_aov now knows about types, uses type 1 by default and can do types 2 and 3.
- Method for effect size calculation now uses
lsr::etaSquared, which also takes a
tadaa_mean_ci: Plots means with 95% confidence intervals as errorbars (thanks Christoph for the suggestion).
tadaa_one_sample: For one-sample t-tests and finally an easy z-test.
confint_norm: Helper to get CIs, similar to
tadaa_wilcoxon: For when
tadaa_t.test isn't non-parametric enough. Same usage.
- Additionally displays medians of each group.
tadaa_kruskal: For when
tadaa_aov isn't non-parametric enough, too.
mean_ci_sem because it's more
show_n option to
tadaa_int: Optionally display N in subtitle.
- Fix documentation inconsistencies.
- Turns out
< 0.05. Well. That was embarrassing.
- Minor tweaks to
theme_readthedown regarding text placement.
- Remove superfluous variables from
- New function:
tadaa_normtest lets you do tests for normality (4 methods) over multiple variables.
- New function:
tadaa_heatmap generates a heatmap. Mhhh, heatmaps.
- New function:
tadaa_sem shows the standard error of the mean and it's confidence interval
- New function:
pval_string as a modification of
pixiedust::pvalString that includes
p < .05.
- Added a ggplot2 theme for the rmdformats::readthedown Rmd template.
tadaa_t.test and the new
tadaa_normtest now return a
data.frame by default, allowing further shenanigans
- New options in
grid = TRUE for the two interaction plots to be printed in a grid via
- Choose the plot labels via the
tadaa_int plot output now also is a little tidier and optimized for smaller widths.
tadaa_aov now also shows Cohen's f for easier power calculations based on f (
- Add option
modus, so multiple results will be concatenated to a character by default.
- Add additional option
modus because guessing about return value classes is no joke.
- Fix issues with
tadaa_ord as ordinal equivalent of
- Make table output of both of the former functions smaller in width by abbreviating column names
brewer_palette option to
- Dependencies declared in
DESCRIPTION are still experimental because of uncertainty regarding failing travis builds. I don't know what's going on there.
family tag to
tadaa_ functions so their documentation is linked
- Fix return values of
_nom which did not work as I expected
- Fix typo in
- Bump dependencies
pixiedust depends on R (>= 3.2.1), so we might as well depend on that version, too
- Specifiy minimal versions for
broom, just to be safe
- Add vignette
README to be less redundant compared to the usage vignette
- Added a
NEWS.md file to track changes to the package.
- Working on CRAN compatibility
- Submitted to CRAN
- It's on CRAN, yay!