tabxplor

If R makes complex things simple, it can sometimes make simple things difficult. This is why tabxplor tries to make it easy to deal with multiple cross-tables: to create and manipulate them, but also to read them, using color helpers to highlight important informations (differences from totals, comparisons between lines or columns, contributions to variance, margins of error, etc.). It would love to enhance your data exploration experience with simple yet powerful tools. All functions are propelled by tidyverse, pipe-friendly, and render tibble data frames which can be easily manipulated with dplyr. In the same time, time-taking operations are done with data.table to go faster with big dataframes. Tables can be exported to Excel and in html with formats and colors.

Installation

You can install tabxplor from CRAN with:

install.packages("tabxplor")

Or from github with :

# install.packages(devtools)
devtools::install_github("BriceNocenti/tabxplor")

Base usage: cross-tables with color helpers

The main functions are made to be user-friendly and time-saving in data analysis workflows.

tab makes a simple cross-table:

library(tabxplor)
tab(forcats::gss_cat, marital, race)
#> # A tabxplor tab: 7 × 5
#>   marital       Other Black  White  Total
#>   <fct>           <n>   <n>    <n>    <n>
#> 1 No answer         2     2     13     17
#> 2 Never married   633 1 305  3 478  5 416
#> 3 Separated       110   196    437    743
#> 4 Divorced        212   495  2 676  3 383
#> 5 Widowed          70   262  1 475  1 807
#> 6 Married         932   869  8 316 10 117
#> 7 Total         1 959 3 129 16 395 21 483

When one of the row or column variables is numeric, tab calculates means by category of the other variable.

tab comes with options to weight the table, print percentages, manage totals, digits and missing values, add legends, gather rare categories in a “Others” level.

tab(forcats::gss_cat, marital, race, pct = "row", na = "drop", 
other_if_less_than = 1000, other_level = "Custom_other_level_name")

When a third variable is provided, tab makes a table with as many subtables as it has levels. With several tab_vars, it makes a subtable for each combination of their levels. The result is grouped: in dplyr, operations like sum() or all() are done within each subtable, and not for the whole dataframe.

Colors may be added to highlight over-represented and under-represented cells, and therefore help the user read the table. By default, with color = "diff", colors are based on the differences between a cell and it’s related total (which only works with means and row or col pct). When a percentage is superior to the average percentage of the line or column, it appears with shades of green (or blue). When it’s inferior, it appears with shades of red/orange. A color legend is added below the table. In RStudio colors are adapted to the theme, light or dark.

data <- forcats::gss_cat |> 
  dplyr::filter(year %in% c(2000, 2006, 2012), !marital %in% c("No answer", "Widowed"))
gss  <- "Source: General social survey 2000-2014"
gss2 <- "Source: General social survey 2000, 2006 and 2012"
tab(data, race, marital, year, subtext = gss2, pct = "row", color = "diff")

The sup_cols argument adds supplementary column variables to the table. With numeric variables, it calculates the mean for each category or the row variable. With text variables, only the first level is kept (you can choose which one to use by placing it first with forcats::fct_relevel). Use tab_many to keep all levels.

tab(dplyr::storms, category, status, sup_cols = c("pressure", "wind"))
#> # A tabxplor tab: 7 × 13
#>   category disturbance extratropical hurricane `other low`
#>   <fct>            <n>           <n>       <n>         <n>
#> 1 1                  0             0     2 548           0
#> 2 2                  0             0       993           0
#> 3 3                  0             0       593           0
#> 4 4                  0             0       553           0
#> 5 5                  0             0       116           0
#> 6 NA               171         2 151         0       1 453
#> 7 Total            171         2 151     4 803       1 453
#> # ℹ 8 more variables: `subtropical depression` <n>, `subtropical storm` <n>,
#> #   `tropical depression` <n>, `tropical storm` <n>, `tropical wave` <n>,
#> #   Total <n>, pressure <mean>, wind <mean>

References and comparison levels for colors

By default, to calculate colors, each cell is compared to the subtable’s related total.

When a third variable or more are provided, it’s possible to compare with the general total line instead, by setting comp = "all". Here, only the last total row is highlighted (TOTAL ENSEMBLE appears in white but other total rows in grey).

tab(data, race, marital, year, subtext = gss2, pct = "row", color = "diff", comp = "all")

With ref = "first", each row (or column) is compared to the first row (or column), which is particularly helpful to highlight historical evolutions. The first rows then appears in white (while rows totals are themselves colored like normal lines).

data <- data |> dplyr::mutate(year = as.factor(year))
tab(data, year, marital, race, pct = "row", color = "diff", ref = "first", tot = "col",
totaltab = "table")

When ref is a number, the nth row (or column) is used for comparison.

tab(data, year, marital, race, pct = "row", color = "diff", ref = 3)

Finally, when ref is a string, it it used as a regular expression, to match with the names of the rows (or columns).

tab(data, year, marital, race, pct = "col", tot = "row", color = "diff", ref = "Married")

Confidence intervals

It it possible to print confidence intervals for each cell:

tab(forcats::gss_cat, race, marital, pct = "row", ci = "cell")
#> # A tabxplor tab: 4 × 9
#>   race   `No answer` `Never married` Separated Divorced Widowed  Married  Total
#>   <fct>       <row%>          <row%>    <row%>   <row%>  <row%>   <row%> <row%>
#> 1 Other           0%        [30;34]%    [5;7]%  [9;12]%  [3;4]% [45;50]%   100%
#> 2 Black           0%        [40;43]%    [5;7]% [14;17]%  [7;9]% [26;29]%   100%
#> 3 White           0%        [21;22]%    [2;3]% [16;17]%      9% [50;51]%   100%
#> 4 Total           0%             25%        3%      16%      8%      47%   100%
#> # ℹ 1 more variable: n <n>

It is also possible to use confidence intervals to enhance colors helpers. With color = "diff_ci", the cells are only colored if the confidence interval of the difference between them and their reference cell (in total or first row/col) is superior to the difference itself. Otherwise, it means the cell is not significantly different from it’s reference in the total (or first) row: it turns grey, and the reader is not anymore tempted to over-interpret the difference.

tab(forcats::gss_cat, race, marital, pct = "row", color = "diff_ci")

Finally, another calculation appears helpful: the difference between the cell and the total, minus the confidence interval of this difference (or in other word, what remains of the difference after having subtracted the confidence interval). ci = "after_ci" highligths all the cells whose value is significantly different from the relative total (or first cell). This is particularly useful when working on small samples : we can see at a glance which numbers we have right to read and interpret.

tab(forcats::gss_cat, race, marital, subtext = gss, pct = "row", color = "after_ci")

Chi2 stats and contributions of cells to variance

chi2 = TRUE add summary statistics made in the chi2 metric: degrees of freedom (df), unweighted count, pvalue and (sub)table’s variance. Chi2 pvalue is colored in green when inferior to 5%, and in red when superior or equal to 5%, meaning that the table is not significantly different from the independent hypothesis (the two variables may be independent).

tab(forcats::gss_cat, race, marital, chi2 = TRUE)
#> chi2 stats      marital
#> df                   10
#> variance         0.0464
#> chi2                997
#> pvalue           <0.01%
#> count            21 483
#> 
#> # A tabxplor tab: 4 × 8
#>   race   `No answer` `Never married` Separated Divorced Widowed Married  Total
#>   <fct>          <n>             <n>       <n>      <n>     <n>     <n>    <n>
#> 1 Other            2             633       110      212      70     932  1 959
#> 2 Black            2           1 305       196      495     262     869  3 129
#> 3 White           13           3 478       437    2 676   1 475   8 316 16 395
#> 4 Total           17           5 416       743    3 383   1 807  10 117 21 483

Chi2 stats can also be used to color cells based on their contributions to the variance of the (sub)table, with color = "contrib". By default, only the cells whose contribution is superior to the mean contribution are colored. It highlights the cells which would stand out in a correspondence analysis (the two related categories would be located at the edges of the first axes ; here, being black is associated with never married and being separated).

tab(forcats::gss_cat, race, marital, color = "contrib")

Combine tabxplor and dplyr

The result of tab is a tibble::tibble data frame with class tab. It gets it’s own printing methods but, in the same time, can be transformed using most dplyr verbs, like a normal tibble.

library(dplyr)
tab(storms, category, status, sup_cols = c("pressure", "wind")) |>
filter(category != "-1") |>
dplyr::select(-`tropical depression`)
arrange(is_totrow(.), desc(category)) # use is_totrow to keep total rows order

With dplyr::arrange, don’t forget to keep the order of tab variables and total rows:

tab(data, race, marital, year, pct = "row") |>
arrange(year, is_totrow(.), desc(Married))

Draw more complex tables with tab_many

tab is a wrapper around the more powerful function tab_many, which can be used to customize your tables.

It’s possible, for example, to make a summary table of as many columns variables as you want (showing all levels, or showing only one specific level like here):

first_lvs <- c("Married", "$25000 or more", "Strong republican", "Protestant")
data <- forcats::gss_cat |> mutate(across(
where(is.factor),
~ forcats::fct_relevel(., first_lvs[first_lvs %in% levels(.)])
))
tab_many(data, race, c(marital, rincome, partyid, relig, age, tvhours),
levels = "first", pct = "row", chi2 = TRUE, color = "auto")

Using tab or tab_many with purrr::map and tibble::tribble, you can program several tables with different parameters all at once, in a readable way:

tabs <-
purrr::pmap(
tibble::tribble(
~row_var, ~col_vars       , ~pct , ~filter              , ~subtext               ,
"race"  , "marital"       , "no" , NULL                 , "Source: GSS 2000-2014",
"race"  , "marital"       , "row", NULL                 , "Source: GSS 2000-2014",
"race"  , "marital"       , "col", NULL                 , "Source: GSS 2000-2014",
"relig" , c("race", "age"), "row", "year %in% 2000:2010", "Source: GSS 2000-2010",
"relig" , c("race", "age"), "row", "year %in% 2010:2014", "Source: GSS 2010-2014",
NA_character_, "race"     , "no" , NULL                 , "Source: GSS 2000-2014",
),
.f = tab_many,
data = forcats::gss_cat, color = "auto", chi2 = TRUE)

Export to html, Excel or plot

To export a table to html with colors, tabxplor uses knitr::kable and kableExtra. In this format differences from totals, confidence intervals, contribution to variance, and unweighted counts, are available in a tooltip at cells hover.

tabs <- tab(forcats::gss_cat, race, marital, subtext = "Source: GSS 2000-2014", 
pct = "row", color = "diff")
tabs |> tab_kable()

To print an html table by default (for example, in RStudio viewer), use tabxplor options:

options(tabxplor.print = "kable") # default to options(tabxplor.print = "console")

tab_xl exports any table or list of tables to Excel, with all colors, chi2 stats and formatting. On Excel, it is still possible to do calculations on raw numbers (display is rounded but, below, decimals are kept).

tabs |> tab_xl(replace = TRUE, sheets = "unique")

tab_plot exports any table as a plot image.

tabs |> tab_plot()

Programming with tabxplor

When not doing data analysis but writing functions, you can use the sub-functions of tab_many step by step to attain more flexibility or speed. That way, it’s possible to write new functions to customize your tables even more.

data <- dplyr::starwars |>
tab_prepare(sex, hair_color, gender, other_if_less_than = 5,
na_drop_all = sex)

data |>
tab_plain(sex, hair_color, gender, tot = c("row", "col"), pct = "row", comp = "all") |>
tab_ci("diff", color = "after_ci")  |>
tab_chi2(calc = "p")

The whole architecture of tabxplor is powered by a special vector class, named tabxplor_fmt for formatted numbers. As a vctrs::record, it stores behind the scenes all the data necessary to calculate printed results, formats and colors. A set of functions are available to access or transform this data. ?fmt to get more information.

The simple way to recover the underlying numbers as numeric vectors is get_num, which extract the currently displayed field whatever it is :

tabs <- tab(forcats::gss_cat, race, marital, pct = "row")
tabs |> dplyr::mutate(across(where(is_fmt), get_num))
#> # A tabxplor tab: 4 × 9
#>   race   `No answer` `Never married` Separated Divorced Widowed Married Total
#>   <fct>        <dbl>           <dbl>     <dbl>    <dbl>   <dbl>   <dbl> <dbl>
#> 1 Other     0.00102            0.323    0.0562    0.108  0.0357   0.476     1
#> 2 Black     0.000639           0.417    0.0626    0.158  0.0837   0.278     1
#> 3 White     0.000793           0.212    0.0267    0.163  0.0900   0.507     1
#> 4 Total     0.000791           0.252    0.0346    0.157  0.0841   0.471     1
#> # ℹ 1 more variable: n <dbl>

To render character vectors (without colors), use format:

tabs |> mutate(across(where(is_fmt), format))

The following fields compose any fmt column (though many can be NA if not calculated) :

• display : name of the field to display, customisable for each cell (character)

• n : raw count (integer)

• wn : weighted count

• pct : percentages

• diff : differences from totals or reference cells

• digits : digits to display, customisable for each cell (integer)

• ctr : contributions of cells to variance (with color = "contrib")

• mean : means (for numeric column variables)

• var : variance (for numeric column variables ; Chi2 variance with pct)

• ci : confidence intervals

• rr : relative risks, needed to calculate odds ratio

• or : odds ratios (or relative risks ratios)

• in_totrow : TRUE if the cell is part of a total row, FALSE otherwise (logical)

• in_tottab : TRUE if the cell is part of a total table, FALSE otherwise (logical)

• in_refrow : TRUE if the cell is part of a reference row, FALSE otherwise (logical)

vctrs::vec_data(tabs$Married)
#>       n display digits wn       pct      mean         diff ctr var ci rr or
#> 1   932     pct      0 NA 0.4757529 1.0102402  0.004822432  NA  NA NA NA NA
#> 2   869     pct      0 NA 0.2777245 0.5897357 -0.193205991  NA  NA NA NA NA
#> 3  8316     pct      0 NA 0.5072278 1.0770757  0.036297310  NA  NA NA NA NA
#> 4 10117     pct      0 NA 0.4709305 1.0000000  0.000000000  NA  NA NA NA NA
#>   in_totrow in_tottab in_refrow
#> 1     FALSE     FALSE     FALSE
#> 2     FALSE     FALSE     FALSE
#> 3     FALSE     FALSE     FALSE
#> 4      TRUE     FALSE     FALSE

To get those underlying fields you can either use vctrs::fields or, more simply, $ :

tabs |> mutate(across(where(is_fmt), ~ vctrs::field(., "pct") ))

tabs$Married$pct
tabs$Married$n
tabs |> mutate(across(where(is_fmt), ~ .$n))

To change the field currently displayed, for the whole table or a single vector, you can use set_display():

tabs |> set_display("diff")
tabs |> mutate(across(where(is_fmt), ~ set_display(., "diff")))

To modify a field, you can use vctrs field<-. For example, to change the number of digits :

tab(forcats::gss_cat, race, marital, pct = "row") |> 
    mutate(across(where(is_fmt), ~ vctrs::`field<-`(., "digits", rep(2L, length(.)))))

Faster to write and easier to read, you can also use dplyr::mutate() on an fmt vector. For example, to create a new column with standards deviations and display it with decimals :

tab_num(data, race, c(age, tvhours), marital, digits = 1L, comp = "all") |>
  dplyr::mutate(dplyr::across( #Mutate over the whole table.
    c(age, tvhours),
    ~ dplyr::mutate(., #Mutate over each fmt vector's underlying data.frame.
                    var     = sqrt(var), 
                    display = "var", 
                    digits  = 2L) |> 
      set_color("no"),
    .names = "{.col}_sd"
  ))

Some helper functions exists for total rows, total tables and reference rows (is_totrow() / as_totrow(), is_tottab() / as_tottab(), is_refrow() / as_refrow()) :

tab(data, race, marital, year, pct = "row") |>
  dplyr::mutate(across( 
    where(is_fmt),
    ~ dplyr::if_else(is_totrow(.), 
                true  = mutate(., digits = 1L), 
                false = mutate(., digits = 2L))
  ))

Each fmt column have attributes, which you can access or modify with get_ and set_ functions :

• type / get_type() / set_type() : the type of the fmt vector, among c("n", "mean", "row", "col", "all", "all_tabs") ; it determines which calculations are done within tab_ functions.

• totcol / is_totcol() / as_totcol() : TRUE if the column is a total column, FALSE otherwise (logical)

• refcol / is_refcol() / as_refcol() : TRUE if the column is a reference column for comparison, FALSE otherwise (logical)

• color / get_color() / set_color() : the calculation to make to print colors ; among c("", "no", "diff", "diff_ci", "after_ci", "contrib")

• col_var / get_col_var() / set_col_var() : the name of the column variable (there can be many in one single table)

• comp_all / get_comp_all / set_comp_all() : when there are tab_vars, is the reference for comparison the subtable (FALSE), or the total table (TRUE) ?

• ref / get_ref_type() / set_diff_type() : the type of difference calculated, either "no", "tot" for totals, an index, or a regular expression.

• ci_type / get_ci_type() / set_ci_type() : the type of confidence interval, either "cell" or "diff"

For example, to print the number of observations of the total column :

tab(data, race, marital, year, pct = "row") |>
  mutate(across(where(is_totcol), ~ mutate(., display = "n") ))

Note that, if tab_vars are provided, the table is grouped and all operations are made within groups. To remove grouping (for example when it gives errors), use dplyr::ungroup().

If you only need the simplest table, with only numeric counts (no fmt), or even a base data.frame (not a tibble) :

tab_plain(data, race, marital, num = TRUE) # counts as a numeric vector
tab_plain(data, race, marital, df = TRUE)  # same, with unique class = "data.frame"