Scale outcome variables by control-group standard deviation (Glass's delta)

For each outcome variable, divides by its standard deviation in the control group to produce a standardized version. The standardized variable is added to the data frame with a _s suffix.

Usage

scale_by_control(
  data,
  treatment,
  control_value = 0,
  outcomes = NULL,
  prefixes = c("D_", "Y_"),
  strip_suffix = "_01"
)

Arguments

data: A data frame.
treatment: Character scalar. Name of the treatment column.
control_value: Scalar. Value of treatment identifying the control group (default: 0).
outcomes: Character vector of column names to standardize, or NULL (default) to auto-select by prefixes.
prefixes: Character vector of column-name prefixes used for auto-selection when outcomes is NULL (default: c("D_", "Y_")).
strip_suffix: Suffix removed from a column's name before "_s" is appended, or NULL to append to the name unchanged. The default "_01" suits a convention where a rescaled-to-unit-interval outcome carries that marker, so D_belief_01 becomes D_belief_s rather than D_belief_01_s: the value is no longer on the unit interval once divided by an SD, so keeping the marker would be a lie. Pass NULL if your names carry no such marker.

Value

The input data frame with the standardized columns appended. Each is named for its source with strip_suffix removed and "_s" added, so Y_turnout yields Y_turnout_s and Y_turnout_01 yields Y_turnout_s.

Details

Auto-selects columns whose names start with any prefix in prefixes (default: c("D_", "Y_")), excluding columns ending with "_missing" or "_s". Supply outcomes to override this selection.

The resulting _s variables are on Glass's delta scale: a one-unit difference equals one control-group SD. Using the control-group SD as the standardizer is Glass's delta (Glass, 1976), which is preferred when the treatment may change the variance of the outcome. Dividing by the pooled SD (Cohen's d) would conflate effect-size estimation with variance changes induced by the treatment. The _s variables are intentionally excluded from check_y_bounds() (which skips "_s" columns), since standardized values are not bounded to [0, 1]. Use check_s_scaling() to verify the standardization and inspect treatment-arm variance.

References

Glass, G. V. (1976). Primary, secondary, and meta-analysis of research. Educational Researcher, 5(10), 11–17. doi:10.3102/0013189X005010003

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)
)

# _01 is stripped, so the new columns are D_belief_s and Y_attitude_s
names(scale_by_control(dat, treatment = "Z"))
#> [1] "Z"             "D_belief_01"   "Y_attitude_01" "D_belief_s"   
#> [5] "Y_attitude_s" 

# Keep the full source name instead
names(scale_by_control(dat, treatment = "Z", strip_suffix = NULL))
#> [1] "Z"               "D_belief_01"     "Y_attitude_01"   "D_belief_01_s"  
#> [5] "Y_attitude_01_s"