Assess variation of continuous variables across groups

Describe the spread of continuous numeric variables within groups. This is a diagnostic companion to assess_distribution() and is intended to make variation visible before a group comparison is interpreted.

Usage

assess_variance(
  data,
  vars = NULL,
  by,
  digits = 2,
  test = c("levene", "none", "bartlett"),
  format = c("table", "tibble")
)

Arguments

data: A data frame.
vars: Continuous numeric variables to assess. Bare names or a character vector are accepted. When omitted, all detected continuous variables are assessed. Categorical, ordinal, logical, date-time, and binary variables are rejected when explicitly selected.
by: Grouping variable, supplied as a bare name or character string. It must be categorical, binary, logical, or ordinal and contain at least two observed groups.
digits: Number of decimal places.
test: Variance hypothesis test to display: "levene" (default), "none", or "bartlett". Levene's test is median-centred (the robust Brown-Forsythe form). Both are supporting diagnostics, not gatekeepers for ANOVA or Welch methods.
format: Output format: "table" (default) or "tibble".

Value

With format = "table", a gt_variance object that prints as one readable row per variable: each group's n and SD, the observed SD ratio, the requested test p-value, and a plain-language interpretation. $summary contains the full group-level values and $diagnostics retains technical test metadata. With format = "tibble", the detailed summary tibble is returned.

Details

assess_variance() reports group sample sizes, standard deviations, variances, and the ratio of the largest to the smallest group SD and variance. These ratios are descriptive diagnostics, not pass/fail tests. The function deliberately does not run a variance hypothesis test by default, and it does not choose an inferential test. Welch t-tests and Welch ANOVA do not require equal variances for independent groups; this function does not assess pairing, repeated-measures sphericity, or select a repeated-measures method. The default is the median-centred Levene test (often called the Brown-Forsythe modification) as supporting information. It is less sensitive to non-normality than Bartlett's test. Set test = "none" for descriptive spread only, or test = "bartlett" when the normal-distribution assumption is justified. Neither test is used to select a test in compare_groups().

Examples

assess_variance(mtcars, vars = c(mpg, wt), by = am)
assess_variance(mtcars, vars = "mpg", by = am, digits = 1)
assess_variance(mtcars, vars = "mpg", by = am, test = "bartlett")
assess_variance(mtcars, vars = "mpg", by = am, test = "levene")