Skip to contents

Check that _s (Glass's delta) variables are correctly standardized

Source: R/check_s_scaling.R
check_s_scaling.Rd

For each _s-suffixed outcome variable, reports the control-group SD (which should equal 1.0 by construction), the treatment-group SD, and the ratio of treatment SD to control SD. A treatment-to-control SD ratio far from 1 indicates that the treatment changed the outcome variance, which is exactly the situation where Glass's delta is preferable to Cohen's d.

Usage

check_s_scaling(
  data,
  treatment,
  control_value = 0,
  study_id = NULL,
  outcomes = NULL,
  prefixes = c("D_", "Y_")
)

Arguments

data

A data frame, after scale_by_control() has been applied.

treatment

Character scalar. Name of the treatment column.

control_value

Scalar. Value of treatment identifying the control group (default: 0). An error is raised when no row matches, rather than returning a table of NA. All arms other than this one are pooled into treatment_sd, so with three or more arms that column is a pooled figure and not one arm's.

study_id

Optional character scalar. If provided, a study_id column is appended to the returned tibble.

outcomes

Character vector of _s column names to check, or NULL (default) to auto-select by prefixes and "_s" suffix.

prefixes

Character vector of column-name prefixes used for auto-selection when outcomes is NULL (default: c("D_", "Y_")).

Value

A tibble with columns variable, control_sd, treatment_sd, sd_ratio, control_sd_ok, and optionally study_id. control_sd_ok is TRUE when control_sd rounds to 1.000 (within floating-point tolerance), confirming the standardization is correct.

Details

Auto-selects columns whose names start with D_ or Y_ and end with _s. Supply outcomes to override this selection.

See also

scale_by_control

Other outcome scaling: scale_by_control()

Examples

dat <- data.frame(
  Z = c(0L, 0L, 0L, 1L, 1L, 1L),
  D_belief_01 = c(0.2, 0.4, 0.3, 0.6, 0.8, 0.7),
  Y_attitude_01 = c(0.3, 0.5, 0.4, 0.4, 0.6, 0.5)
)
dat <- scale_by_control(dat, treatment = "Z")
check_s_scaling(dat, treatment = "Z")
#> # A tibble: 2 × 5
#>   variable     control_sd treatment_sd sd_ratio control_sd_ok
#>   <chr>             <dbl>        <dbl>    <dbl> <lgl>        
#> 1 D_belief_s            1            1    1     TRUE         
#> 2 Y_attitude_s          1            1    1.000 TRUE