Convert experimental data to a schedule of potential outcomes

Takes a data frame in the long format typical of experimental data — one row per unit, with a single response column and a treatment assignment column — and pivots it into a wide schedule of potential outcomes before attaching the "schedule" class and metadata. The observed response for each unit fills its assigned treatment column; all other potential outcome columns are NA (unobserved). Use impute_unobserved to fill them in.

Usage

experiment_to_schedule(data, treatment, response, outcome_prefix = "Y")

Arguments

data

A data frame with one row per experimental unit.

treatment

The unquoted name of the column in data that contains the treatment assignments, which must be a factor.

response

The unquoted name of the column in data that contains the observed response variable.

outcome_prefix

A character string prepended to each treatment level name to form the potential outcome column names. Defaults to "Y".

Value

A data frame with class "schedule", containing the original columns plus one new column per treatment level (named <outcome_prefix><level>), and metadata attributes: treatment, treatment_levels, treatment_counts, potential_outcome_cols, and response.

See also

as_schedule, impute_unobserved

Examples

data <- data.frame(
  id = 1:6,
  treatment = factor(c("control", "control", "treatment",
                       "treatment", "control", "treatment")),
  response = c(10, 12, 15, 18, 11, 17)
)
experiment_to_schedule(data, treatment, response)
#>   id treatment response Ycontrol Ytreatment
#> 1  1   control       10       10         NA
#> 2  2   control       12       12         NA
#> 3  3 treatment       15       NA         15
#> 4  4 treatment       18       NA         18
#> 5  5   control       11       11         NA
#> 6  6 treatment       17       NA         17