Skip to contents

Compare, choose and apply transformations

Source: R/transform.R
transform_compare.Rd

Workflow based on the shape of the marginal distribution:

  • transform_compare(): skewness, excess kurtosis, Shapiro-Wilk W and p of every variable under every transformation (a decision tool, to be read with the plots);

  • transform_choose(): one transformation per variable: the highest W, but the preferred method (log, interpretable as a multiplicative effect) is kept when its W is within tol of the best, and "none" is kept when the variable is already about symmetric (|skewness| < skew_ok);

  • transform_propagate(): copies the reference-visit choice (same method and same lambda) to the other visits;

  • transform_apply(): adds the transformed variables as new prefixed columns (LOG_, BC_, ...).

Usage

transform_compare(
  df,
  vars,
  methods = c("none", "log", "log1p", "sqrt", "boxcox")
)

transform_choose(cmp, prefer = c("log", "log1p"), tol = 0.01, skew_ok = 0.5)

transform_propagate(choice, from = "_V0$", to = "_V3")

transform_apply(
  df,
  choice,
  prefixes = c(log = "LOG", log10 = "LOG10", log1p = "LOG1P", sqrt = "SQRT", boxcox =
    "BC")
)

Arguments

df

A data frame.

vars

Numeric variables (names or tidyselect).

methods

Transformations compared, among "none", "log", "log10", "log1p", "sqrt", "boxcox" (the latter needs the car package).

cmp

Output of transform_compare().

prefer

Preferred methods, in order.

tol

Tolerance on W for the preferred method.

skew_ok

Absolute skewness below which no transformation is applied.

choice

Output of transform_choose() (columns Variable, Method, and lambda, shift for Box-Cox).

from

Regular expression of the reference-visit suffix.

to

Replacement suffix(es) of the other visits.

prefixes

Named vector method -> prefix of the new column.

Value

transform_compare(): one row per variable x method. transform_choose(): one row per variable with a Reason column. transform_propagate(): the choice with extra rows for the other visits. transform_apply(): list(data, map) with map = original variable, new column, method and lambda.

Examples

cmp <- transform_compare(cohort_sim, c(NFL_V0, CRP_V0), methods = c("none", "log", "sqrt"))
choice <- transform_choose(cmp)
choice <- transform_propagate(choice)
res <- transform_apply(cohort_sim, choice)
res$map
#> # A tibble: 4 × 4
#>   Variable New        Method lambda
#>   <chr>    <chr>      <chr>   <dbl>
#> 1 NFL_V0   LOG_NFL_V0 log        NA
#> 2 CRP_V0   LOG_CRP_V0 log        NA
#> 3 NFL_V3   LOG_NFL_V3 log        NA
#> 4 CRP_V3   LOG_CRP_V3 log        NA