Breaking change:
modelsummary() has
changed from “Model 1” to “(1)”. The benefits are: labels are no longer
in English by default; use less horizontal space; eliminate the “Model”
redundancy. Unfortunately, this could break code in some edge cases
where users rely on column names to manipulate tables. The old behavior
can be restored by calling:
options(modelsummary_model_labels="model")New features:
shape="rbind" to stack multiple regression tables and
create “panels” with labelled groups of models.fmt: new helper functions for different formatting
styles
fmt = fmt_decimal(2): decimal digitsfmt = fmt_decimal(digits = 2, pdigits = 4): decimal
digits with p value-specific settingfmt = fmt_sprintf("%.3f"): sprintf()
decimalfmt = fmt_sprintf("%.3e"): sprintf()
scientificfmt = fmt_significant(3): significant digitsfmt = fmt_statistic("estimate" = 2, "std.error" = 3):
statistic-specific formattingfmt = fmt_term("(Intercept)" = 2, "hp" = 3):
term-specific formattingfmt = fmt_identity(): raw valuesmodelsummary, such as Roman Numerals or letters in
parentheses.
options(modelsummary_model_labels = "(arabic)")options(modelsummary_panel_labels = "letters")modelplot(draw = FALSE) now returns a
p.value column. This allows conditional aesthetics (see the
modelplot vignette).marginaleffects
package.Bugs:
fixest models returns useless “group.x” and
“group.y” columns. Isse #591. Thanks to Adam Altmejd for the
report.Breaking change:
shape and output="dataframe"
arguments, there always used to be a group column. Now,
this column has the same name as the variable in the shape
formula (“response”, “component”, etc.).New features:
shape can include multiple groups.coef_rename can be an unnamed vector of length equal to
the number of terms in the final table, obtained after
coef_map and coef_omit are applied and models
are merged.coef_omit accepts numeric indices. Positive values:
coefficients to omit. Negative values: coefficients to keep.datasummary_skim: Increased maximum number of variables
to 250.Bug fixes:
Breaking change:
siunitx version 3.0.25 LaTeX package.title argument now respects the escape
argument for all kableExtra output formats. This can break
tables in which users manually escaped titles.New features:
align
argument for all output formats.
modelsummary(mod, align = "ld")update_modelsummary() function makes it easy to
install the dev versions of modelsummary and its
dependencies (mostly useful for Vincent and people who report
bugs).haven labels in
modelsummary(), datasummary(),
datasummary_skim()output = "filename.csv"output = "filename.xlsx"add_columns argument supported in
modelsummary()datasummary_balance supports the stars
argument.align = "d"
column.Bug fixes:
shape argument accepts interactions with the colon “:”
character. This combines two columns into one, which can be useful to
display terms and group names in a single column.parallel::mclapply. See
?modelsummarymodelsummary no longer computes confidence intervals
when not necessary, which can save some time. Also see:
conf_level=NULLThis first major release accompanies the publication of an article in the Journal of Statistical Software:
Arel-Bundock, Vincent (2022). “modelsummary: Data and Model Summaries in R.” Journal of Statistical Software, 103(1), 1-23. doi:10.18637/jss.v103.i01 https://doi.org/10.18637/jss.v103.i01.’
If you like modelsummary, please cite the JSS article
and tell your friends about it.
Minor changes:
gof_map="all" includes all available statistics.
gof_map="none" excludes all statistics.diagnostic=NULL and test=NULL by default for
speed.Breaking changes:
modelsummary_wide is no longer available. Use the
shape argument of modelsummary instead.modelsummary now uses the easystats
packages (performance and parameters) to
extract estimates and goodness-of-fit statistics instead of
broom. This can be reverted by setting a global option:
options(modelsummary_get="broom"). This change aims to (1)
increase consistency across models, (2) improve the developers’ ability
to push bug fixes upstream when necessary, and (3) improve support for
mixed effects, bayesian, and GAM models. The two main drawbacks are: (a)
The set of printed statistics may be slightly different from previous
versions of modelsummary (b) The group identifiers used in
the shape formula will also be different for certain models
(e.g., in nnet::multinom, y.level becomes
response).New features:
shape argument accepts a formula and can reshape
information in myriad ways. Deprecates the group argument.
Examples:
~ statistic: statistics are shown horizontally in
distinct columns.model ~ term: models in rows and terms in columns.term + y.level + statistic ~ model: grouped
coefficients for multivariate outcome in
nnet::multinomy.level ~ model: partial match is the same as the
previous formulafmt:
modelsummary(mod, fmt = list(estimate = 2, std.error = 1, rmse = 4))glue to apply functions to numeric values by
setting fmt = NULL. Example:
modelsummary(model, fmt = NULL, estimate = "{log(estimate)}")Bug fixes:
group_map rename issuelm models.datasummary_skim output to jpg should now worksescape fixesexponentiate argument for
modelsummary() and modelplot()gof_map accepts a vector such as
c("rmse", "nobs", "r.squared")rlang dependencydatasummary_balance:
~ 1 as a formula to summarize all data.Misc:
lmmodelsummary:
vcov strings like HC1 and
Robust are now case-insensitivegof_map now accepts a data.frame or tibble with a
fmt list-column which includes functions (see Examples in
docs)R2 is no longer computed by default for bayesian and
mixed effects models. An informative one-time warning is printed about
the metrics argument.datasummary_skim:
kableExtra factory:
col.names argument can now be passed to
kableExtra::kbl through the … ellipsis.Misc:
output = "github_document" is now supportedmodelsummary:
vcov argumentdatasummary_balance:
Misc:
modelsummary:
vcov argument handling for fixest
models (#357 by @grantmcdermott)fixest::i() variables and interactions
(#361 by @grantmcdermott)datasummary_correlation:
add_rows and add_columns arguments are now
available here.Misc:
modelsummary_factory_default,
modelsummary_factory_html, etc.intersectBug fixes:
datasummary_balance: escape variable names when
escape=TRUEBreaking change:
dcolumn for dot-aligned columns is
deprecated. Use “d” in the align argument instead.Other changes:
\num{}
function from the siunitx package by default. This produces
much nicer formatting. This can be disabled with a global option. See
?modelsummaryalign argument accepts a “d” column for
dot-alignment using the siunitx LaTeX package:
align="ldd".escape argument in most table-building
functions.threeparttable=TRUE argument
through ...tidyrmodelsummary:
group: The order of terms in the formula determines the
order of rows/columnsmodelsummary_wide:
modelplot:
datasummary_crosstab:
statistic=NULL produces a very basic crosstabdatasummary_crosstab:
modelsummary:
options("modelsummary_stars_note" = FALSE)longtable=TRUE works for LaTeX outputcoef_map or coef_rename are used.group = model ~ term + group is now supported.datasummary_skim:
datasummary_skim("categorical") keeps NA
by default. Users can convert variables to factors before calling
datasummary_skim to exclude NA.Other:
modelsummary(model1, model2)gt titles use the new caption argument in
the gt 0.3.0 functionBreaking change:
stars=TRUE have been
updated to be consistent with the default output from base R (e.g., in
summary.lm). The new significance thresholds are: “+” p < 0.1, “”
p < 0.05, ”” p < 0.01, ”” p < 0.001datasummary_crosstab:
datasummary:
N is smart enough to return either the number of
elements in a subset or the number of non-missing observations in a
variabledatasummary_balance:
NAs in factor variables by default. Users can
convert their variables with the factor() function to omit
NAs automatically.modelsummary:
get_gof (glance) is now
optional.sandwich, which allows things
like:
modelsummary(model, vcov = "bootstrap", R = 1000, cluster = "firm")Other:
output="jupyter"modelsummary:
modelsummary: group and
group_map for grouped parameters (e.g., outcome levels in
multinomial logit or components of gamlss model).dvnames() makes it easy to get dependent variable
column titles (thanks to @NickCH-K)output="modelsummary_list" to save a lightweight
list-based representation of the table which can be saved and fed to
modelsummary once more to get a full table.vcov adds a row to note the type of standard
errors.modelsummary accepts a single model with multiple
vcovs.get_gof forwards … to
model_performancecoef_map accepts unnamed vectors for easy
subsettingfixest::fixest_multi supportoptions(modelsummary_get) to set the order of
extraction functions to use under the hood (broom vs. easystats
vs. all)metrics argument of
performance::model_performance is available via
modelsummary’s … ellipsis to limit the GOF statistics in
Bayesian models.estimate="{estimate}{stars}"gt by setting
options(modelsummary_factory_html="gt")datasummary_correlation:
... forwarddatasummary_correlation_formatdatasummary_correlation’s method argument
accepts functions and “pearspear” (thanks to @joachim-gassen)datasummary:
datasummary functions and rounding accept
…, big.mark, etc.datasummary_skim:
Bug fixes and lints
tidy_custom allows partial term matchesmodelsummary(coef_rename) accepts functionscoef_rename for use in
modelsummary(coef_rename=coef_rename)modelplot accepts add_rows to add
reference categoriesstatistic_override becomes vcovestimate and
statisticextract is no longer
documented.threeparttable, colors, and many other
LaTeX optionspkgdown website and doc improvementsmitools tidiers