Tplyr Options

Default Layer Formats

Declaring string formats and summaries that need to be performed is one of the more verbose parts of ‘Tplyr’. Furthermore, this is something that will often be fairly consistent within a study, as you’ll likely want to look across a consistent set of descriptive statistics, or your count/shift tables will likely require the same sort of “n (%)” formatting.

Using the format options is very similar to setting a string format. The only difference is that you need to enter the string formats as a named list instead of as separate parameters to a function call.

options(
    # Count layer defaults
  tplyr.count_layer_default_formats =
    list(n_counts = f_str("xxx (xx%)", n, pct),
         riskdiff = f_str('xx.xxx', dif)
         ),

  # Desc layer defaults
  tplyr.desc_layer_default_formats =
    list("n"        = f_str("xxx", n),
         "Mean (SD)"= f_str("a.a+1 (a.a+2)", mean, sd),
         "Median"   = f_str("a.a+4", median)
         ),

  # Shift layer defaults
  tplyr.shift_layer_default_formats = list(f_str("xxx", n))
)
 

Here you can see that ‘Tplyr’ picks up these option changes. In the table below, we didn’t use set_format_strings() anywhere - instead we let ‘Tplyr’ pick up the default formats from the options.

tplyr_table(adsl, TRT01P) %>% 
  add_layer(
    group_desc(AGE, by = "Age (years)")
  ) %>% 
  add_layer(
    group_count(AGEGR1, by = "Categorical Age Groups")
  ) %>% 
  build() %>% 
  kable()

One important thing to understand about how these options work in particular is the scoping.

• ‘Tplyr’ options have the broadest scope, as they work across the entire session for any table.
• Setting formats at the tplyr_table() level will override the ‘Tplyr’ options and extends to any layer of the specified type in the current table.
• Setting formats at the tplyr_layer() level will always be prioritized over ‘Tplyr’ options and any tplyr_table() formats set. This has the narrowest scope and will always be used when specified.

To demonstrate, consider the following. The ‘Tplyr’ options remain set from the block above.

tplyr_table(adsl, TRT01P) %>% 
  set_count_layer_formats(n_counts = f_str("xx (xxx%)", n, pct)) %>% 
  set_desc_layer_formats("Mean (SD)" = f_str("a.a+1 (a.a+2)", mean, sd)) %>% 
  add_layer(
    group_desc(AGE, by = "Age (Years)")
  ) %>% 
  add_layer(
    group_count(AGEGR1, by = "Categorical Age Groups")
  ) %>% 
  add_layer(
    group_count(ETHNIC, by = "Ethnicity") %>% 
      set_format_strings(f_str("xxxxx (xx.xxx%)", n, pct))
  ) %>% 
  build() %>% 
  kable()

In the above output:

• The descriptive statistics layer for “Age (Years)” uses the specified table default using set_desc_layer_formats()
• The first count layer for “Categorical Age Groups” uses the specified table default using set_count_layer_formats()
• The second count layer for “Ethnicity” uses the layer level format specified using set_format_strings()

Each of the outputs ignores the ‘Tpylr’ option defaults.

Precision Cap

‘Tplyr’ defaults to avoiding capping precision. We do this because capping precision should be a conscious decision, where you as the user specifically set the limit of how many decimal places are relevant to a specific result. One way to cap precision is by using the cap parameter within set_format_strings(). But perhaps you have a specific limit to what you’d like to see on any output. Here, we offer the tplyr.precision_cap option to set whatever cap you wish.

options(tplyr.precision_cap = c('int'=2, 'dec'=2))

Similar to the layer defaults, setting a precision cap at the layer level will override the tplyr.precision_cap option.

tplyr_table(adsl, TRT01P) %>% 
  add_layer(
    group_desc(HEIGHTBL, by = "Height at Baseline") %>% 
      set_format_strings(
        'Mean (SD)' = f_str('a.a (a.a+1)', mean, sd)
      )
  ) %>% 
  add_layer(
    group_desc(HEIGHTBL, by = "Height at Baseline (Limited)") %>% 
      set_format_strings(
        'Mean (SD)' = f_str('a.a (a.a+1)', mean, sd),
        cap = c('int'= 1, 'dec'=0)
      )
  ) %>% 
  build() %>% 
  kable()

Both layers in the above example are summarizing the same data. The top layer is using the tplyr.precision_cap option set above, which limits to 2 integer place and 2 decimal places. To bottom layer applies its own cap to override the option. Recall a few things about auto precision:

• Integers will not be truncated - rather, the auto precision will default how much padding should be applied. If the number exceeds that number of spaces, then the length will be extended (i.e. if 2 integer places are allotted, 100 will still consume 3 places).
• The cap applies to the spaces allotted by the ‘a’. If the cap is 2, ‘a+1’ will not exceed 3 spaces.

The bottom layer overrides the ‘Tplyr’ option. Instead, integers are capped at 1 space, and decimals are capped at 0.

Custom Summaries

Custom summaries allow you to extend the capabilities of descriptive statistics layers in ‘Tplyr’. Maybe our defaults don’t work how you’d like them to, or maybe you have some custom functions within your organization that you commonly would like to use. Specifying the custom summaries you wish to use in every table would prove quite tedious - therefore, the tplyr.custom_summaries option is a better choice.

options(tplyr.custom_summaries = rlang::quos(
  geometric_mean = exp(sum(log(.var[.var > 0]),na.rm=TRUE) / length(.var))
))

Note that the table code used to produce the output is the same. Now ‘Tplyr’ used the custom summary function for geometric_mean as specified in the tplyr.custom_summaries option. Also note the use of rlang::quos(). We’ve done our best to mask this from the user everywhere possible and make the interfaces clean and intuitive, but a great deal of ‘Tplyr’ is built using ‘rlang’ and non-standard evaluation. Within this option is one of the very few instances where a user needs to concern themselves with the use of quosures. If you’d like to learn more about non-standard evaluation and quosures, we recommend Section IV in Advanced R.

Now that geometric mean is set within the ‘Tplyr’ options, you can use it within your descriptive statistics layers, just like it was one of the built-in summaries.

tplyr_table(adsl, TRT01P) %>% 
  add_layer(
    group_desc(AGE) %>% 
      set_format_strings('Geometric Mean' = f_str('xx.xx', geometric_mean))
  ) %>% 
  build() %>% 
  kable()

Scientific Notation

By default, R will switch to scientific notation for a number less than .001. This is controlled by the scipen option. The default value of scipen is 0. If you’d like to increase the decimal places required for scientific notation to be triggered, increase the value of scipen. The value of scipen is the number of orders of ten smaller (i.e. decimal places preceded by 0’s) required to switch to scientific notation. Decreasing the value of scipen will cause R to switch to scientific location for larger numbers.

This is easier to understand with an example.

options(scipen = 0) # This is the default
.0001
#> [1] 1e-04

options(scipen = 1) # Require 5 decimal places instead

.0001
#> [1] 0.0001
.00001
#> [1] 1e-05

options(scipen = -1) # Only require 3 decimal places
.001
#> [1] 1e-03

In ‘Tplyr’, we have the option tplyr.scipen. This is the scipen setting that will be used only while the ‘Tplyr’ table is being built. This allows you to use a different scipen setting within ‘Tplyr’ than your R session. The default value we use in ‘Tplyr’ is 9999, which is intended to totally prevent numbers from switching to scientific notation. We want this to be a conscious decision that you make in order to prevent any unexpected outputs.

options(tplyr.scipen = -3)
t <- tplyr_table(adae, TRTA) %>% 
  add_layer(
    group_count(AEDECOD) %>% 
      add_risk_diff(c('Xanomeline Low Dose', 'Placebo'))
  )

suppressWarnings(build(t)) %>% # Chi-squared warnings occur with small samples
  head() %>% 
  kable()

Note that the risk-difference variables above have mostly shifted to scientific notation. This is because the limit has been shifted to .1 within the ‘Tplyr’ build.

Quantile Algorithms

There are many algorithms available to compute quantile R has 9, controlled by the type parameter in quantile(). The descriptive statistics offer built-in summaries for Q1, Q3, and Interquartile range, all three of which use the quantile() in the underlying implementation. Given that we offer this default, we felt it was important to offer you the flexibility to change the algorithm. You can do this with tplyr.quantile_type.

The default we chose to use is the R default of Type 7:

\[ m = 1-p. p[k] = (k - 1) / (n - 1). \textrm{In this case, } p[k] = mode[F(x[k])]. \textrm{This is used by S.} \] The example below demonstrates using the default quantile algorithm in R

tplyr_table(adsl, TRT01P) %>% 
  add_layer(
    group_desc(CUMDOSE) %>% 
      set_format_strings("Q1, Q3" = f_str('xxxxx, xxxxx', q1, q3))
  ) %>% 
  build() %>%
  kable()

Within the clinical world, you may wish to match the way that SAS calculates quantiles. To match SAS’s definition, use Type 3:

\[ \textrm{Nearest even order statistic. γ = 0 if g = 0 and j is even, and 1 otherwise.} \]

options(tplyr.quantile_type = 3)

tplyr_table(adsl, TRT01P) %>% 
  add_layer(
    group_desc(CUMDOSE) %>% 
      set_format_strings("Q1, Q3" = f_str('xxxxx, xxxxx', q1, q3))
  ) %>% 
  build() %>% 
    select(-starts_with("ord")) %>% 
  kable()

IBM Rounding

In certain cases users may want to match tables produced by other languages that IBM rounding. Tplyr offers the option ‘tplyr.IBMRounding’ to change the default rounding behavior of Tplyr tables. Review var1_4 in the tables below.

Using the default R behavior

tplyr_table(mtcars, gear) %>%
  add_layer(
    group_desc(qsec) %>%
      set_format_strings(mean = f_str("xx.xx", mean))
  ) %>%
  build()
#> # A tibble: 1 × 6
#>   row_label1 var1_3 var1_4 var1_5 ord_layer_index ord_layer_1
#>   <chr>      <chr>  <chr>  <chr>            <int>       <int>
#> 1 mean       17.47  18.75  15.50                1           1

Using IBM rounding

withr::with_options(
  list(tplyr.IBMRounding = TRUE),
  {
    tplyr_table(mtcars, gear) %>%
      add_layer(
        group_desc(qsec) %>%
          set_format_strings(mean = f_str("xx.xx", mean))
      ) %>%
      build()
  }
)
#> Warning: You have enabled IBM Rounding. This is an experimental feature.
#> •  If you have feedback please get in touch with the maintainers!
#> This warning is displayed once every 8 hours.
#> # A tibble: 1 × 6
#>   row_label1 var1_3 var1_4 var1_5 ord_layer_index ord_layer_1
#>   <chr>      <chr>  <chr>  <chr>            <int>       <int>
#> 1 mean       17.47  18.75  15.50                1           1