Skip to contents

Box-Cox decision with car's documented rule

Source: R/transform.R
transform_boxcox.Rd

Every decision is made by an existing function with its documented default: lambda by maximum likelihood with its Wald 95% CI (car::powerTransform()), car's rounded power (the first of 1, 0, -1, 0.5, 0.33, -0.5, -0.33, 2, -2 inside the CI) and likelihood-ratio tests of lambda = 0 (log) and lambda = 1 (no transformation) (car::testTransform()).

  • transform_boxcox(): marginal distribution of strictly positive variables (values <= 0 or fewer than 10 values: kept raw);

  • transform_boxcox_resid(): for a response, the normality assumption concerns the residuals: lambda is estimated for <root><visit> in the linear model score ~ covariates. Scores with zeros use the Yeo-Johnson family (car::yjPower(), Yeo & Johnson 2000), equal to log(1 + y) for lambda = 0;

  • transform_apply_resid(): adds the transformed scores at every visit (prefixes LOG_, SQRT_, BC_, LOG1P_, YJ_) with the reference decision.

Usage

transform_boxcox(df, vars)

transform_boxcox_resid(df, roots, covariates, visit = "_V0")

transform_apply_resid(
  df,
  decision,
  visits = c("V0", "V3"),
  prefixes = c(log = "LOG", sqrt = "SQRT", boxcox = "BC", log1p = "LOG1P", yj = "YJ")
)

Arguments

df

A data frame.

vars

Strictly positive variables (names or tidyselect).

roots

Root names of the scores.

covariates

Covariates of the linear model.

visit

Suffix of the reference visit (with the underscore).

decision

Output of transform_boxcox_resid().

visits

Visit suffixes where the transformation is applied.

prefixes

Named vector method -> prefix.

Value

transform_boxcox(): one row per variable with lambda, CI, rounded power, LR p-values, Method / lambda / shift (input of transform_apply()), skewness and W before / after. transform_boxcox_resid(): one row per root with the skewness of the residuals before / after. transform_apply_resid(): list(data, map).

References

Box GEP, Cox DR (1964). An analysis of transformations. J R Stat Soc B 26:211-252. Fox J, Weisberg S (2019). An R Companion to Applied Regression, 3rd ed. Yeo IK, Johnson RA (2000). A new family of power transformations to improve normality or symmetry. Biometrika 87:954-959.

Examples

transform_boxcox(cohort_sim, c(NFL_V0, CRP_V0))
#> # A tibble: 2 × 16
#>   Variable     N lambda_ml ci_low ci_high rounded p_log   p_none Method lambda
#>   <chr>    <int>     <dbl>  <dbl>   <dbl>   <dbl> <dbl>    <dbl> <chr>   <dbl>
#> 1 NFL_V0     120   -0.0641 -0.337   0.209       0 0.644 3.61e-16 log        NA
#> 2 CRP_V0     115   -0.0199 -0.160   0.120       0 0.781 1.46e-40 log        NA
#> # ℹ 6 more variables: shift <dbl>, skew_before <dbl>, W_before <dbl>,
#> #   Reason <chr>, skew_after <dbl>, W_after <dbl>
dec <- transform_boxcox_resid(cohort_sim, "UPDRS", covariates = c("AGE", "SEX"))
transform_apply_resid(cohort_sim, dec)$map
#> # A tibble: 1 × 4
#>   Root  New   Method lambda
#>   <chr> <chr> <chr>   <dbl>
#> 1 UPDRS UPDRS none       NA