Version 1.9.12

Version 1.9.10

Version 1.9.8

Version 1.9.7

Version 1.9.5

Version 1.9.4

Cleaned up the suggests and imports (removed tidyverse and a few others that were not necessary for the package to function). The message printed upon attachment is no longer displayed.

Version 1.9.3

Added rowmeans.n() and rowsums.n() for row means and sums allowing n missing values across the row.

Version 1.9.2

table1() now can do the non-parametric Kruskal-Wallis Rank Sum test (when param = FALSE).

Also fixed a small issue with mutate_rowmeans() and mutate_rowsums().

Version 1.9.0

Added the option to have a total column in table1() even when using a grouping variable (total = TRUE will activate this option). The resulting tests (if test = TRUE) will give results of just the bivariate comparisons (grouping variable with the specified variable).

Version 1.8.8

Small bug where a warning message was printed too often was fixed.

Version 1.8.7

A minor update to fix an error that occurred when only a single variable was being summarized. Some code formatting under the hood was adjusted and some minor adjustments to the documentations were also made. Also, a warning is produced if any of the variables have no variability since this causes problems with tests and other formatting. Finally, a message is printed for test = TRUE when the variances are assumed to be unequal (based on a test). This is communicated in the documentation.

Version 1.8.0

Version 1.7.9

Mainly bug fixes, including the removal of any empty rows. This allows for more accurate counts for the n’s in really messy data. We also now re-export the pipe from magrittr.

Version 1.7.6

Updates to the "latex2" output option for table1() and tableC() where it now produces a character vector of the output and then uses a print method to produce the Latex ready code. This means you can access the object and make changes to the table before having it print. It also outputs \emph{Missing} when na.rm=FALSE. Thanks to Joshua Pritikin for the ideas and the pull requests leading to these improvements.

Other minor bug fixes.

%xt% is now deprecated. tableF() and tableX() effectively do what %xt% was designed to do, but with better and more thorough output.

Version 1.7.3

Minor changes and two new functions:

  1. rowmeans() provides a tidyverse friendly rowMeans()
  2. rowsums() provides a tidyverse friendly rowSums()

Version 1.7.2

Several new features were added:

  1. table1() can take a grouped_df from the tidyverse’s group_by(). This also means table1() can take on any number of stratifying variables in a clean way.
  2. table1() also has a new ‘latex2’ output option. This is not dependent on knitr::kable() and is better suited for this situation. Should provide better latex tables than before.
  3. tableF() is a new function that provides frequency tables (potentially by a grouping variable).
  4. General formatting was improved.
  5. Other internal improvements that should help with error-catching, syntax simplicity, and flexibility for the user.

Version 1.6.0

A new function was added: tableC(), which, like table1() provides a nicely formatted table. However, tableC() provides correlations instead of the descriptive statistics that table1() focuses on.

We are also deprecating the var_names argument in table1() since it is now possible to name the variable from within the function itself without relying on a separate argument.

Other small bug fixes.

Version 1.5.4

There are three notable changes in this update:

  1. Small bug fix when using piping and ask for one of the kable output options. This has now been fixed. Additionally, better error catching and improved function tests.
  2. The long() function has been added to the package. It is a wrapper of the stats::reshape() function with a direction set to “long”, but has added benefits, including a bit cleaner syntax, the ability to handle unbalanced data, and works well with the tidyverse.
  3. The wide() function has been added as well. It, like long(), is a wrapper of reshape() but with a set direction of “wide” and has added benefits as well.

Version 1.5.0

table1() now has the ability to apply any function to numeric variables that the user supplies (all functions that work with tapply() should work with table1(). Other than that, the changes are mainly internal changes in table1() that have bearing on a few arguments.

  1. If no variables are named, then all variables are summarized in the data.frame (the all argument is no longer used).
  2. Piping is automatically detected so piping is not longer used as an argument.
  3. rounding is no longer used given a user can define their own function with their own rounding limits.
  4. test_type is no longer an argument. The functionality of the "or" option was just not being used and was too cumbersome.
  5. format_output, condense and simple are combined into one type argument (defaults are still the same).
  6. output_type is now just output.

Finally, tableM() was removed from the package. It is easily adopted by table1() with small modifications to the splitby variable. I apologize for any inconvenience these changes cause. However, I believe it is best in the long run and will make using and upgrading the package much easier.

As an aside, most of these changes are due to a manuscript in review about the package. Several beneficial suggestions were made and so we made those changes at the cost of a small headache at first.

Thanks!

Version 1.4.1

Three big changes:

  1. simple_table1 is no longer with us. The functionality has been integrated into table1.
  2. table1 has a “condensed” version that leaves less white space and is more in line with some academic journals.
  3. The “simple” option has been added to table1 which allows only the percentages (instead of counts and percentages) for categorical variables.
  4. tableM is a new function that can analyze a missing data by putting a variable with missingness in the “missing_var” argument.

Enjoy!

Version 1.3.0

Three notable changes:

  1. A new function–tableM–has been added to the package. It produces a table much like table1 but analyzes a variable with missingness.
  2. table1 (and its related functions such as simple_table1 and tableM) now has better formatting and has an additional option output_type = "text2" which provides lines to separate the header from the rest of the table.
  3. table1 can give percentages based on the row using row_wise = TRUE. The default is giving percentages by column.

Also, a bug fix:

  1. table1 was rounding the percentages at the ten’s place instead of the tenth’s place. It now is fixed.

Thanks!

Version 1.2.3

A bug that rounded the percentages to the nearest tenth place was found and fixed for both table1 and simple_table1.

Version 1.2.2

An error was found when using table1 with a splitby factor with more than 2 levels. This has been fixed in version 1.2.2.

There were also two notable inclusions:

  1. In table1 you can now format the numbers in the output with a comma for large numbers (e.g., 20,000 vs. 20000).
  2. In table1 you can also calculate percentages of factor variables across rows instead of within groups.
  3. A version of table1 known as simple_table1 that reports the percentages of factors instead of counts and percentages.
  4. A new vignette all about table1 is included.

None of these changes will break programs already in place using this package.

Thanks.

Version 1.2.0

We are excited to announce that we are updating our furniture package from 1.0.1 to 1.2.0. We included a new operator (%xt%) for simple cross tabulations that includes a chi-square test and some data from NHANES 2005-2010. Although these additions are nice, most of the updates were to table1. These included:

  1. Bug fixes (the function got confused if there was an object in the global environment that had the same name as a variable in the data frame you were analyzing)
  2. Fewer dependencies (this will allow the package to be more stable across platforms with different package versions loaded without having to worry about packrat)
  3. Additional functionality in the splitby argument (the splitby argument can now have a formula [e.g. ~var] or can be a string [e.g. “var”])
  4. Additional functionality in the output_type argument (can now do any type of output that knitr::kable can do plus the regular “text” option)

None of these changes will break programs already in place using this package. Instead, there are additional uses of the functions as shown in points 3 and 4.

Thanks.