Synthetic Purposive Sampling: Site Selection for External Validity

Usage

sps(
  X,
  N_s,
  stratify = NULL,
  site_include = NULL,
  site_exclude = NULL,
  lambda = c(1, 1, 0),
  seed = 1234,
  max_iter = 10,
  solver = "ECOS_BB",
  num_iter = NULL
)

Arguments

X: Site-level variables for the target population of sites. Row names should be names of sites. X cannot contain missing data.
N_s: Number of study sites to be selected.
stratify: (Optional. Default = NULL) Output from function stratify_sps(). This argument helps users incorporate practical and logistical constraints. See examples on http://naokiegami.com/spsR/articles/stratify_sps.html
site_include: (Optional. Default = NULL) Names of sites users want to always include (or have already selected).
site_exclude: (Optional. Default = NULL) Names of sites users want to always exclude.
lambda: Values of the tuning parameters. If users want to change how to balance three parts of the objective function, they can change lambda. Default values are c(1, 1, 0). Users who want to fine-tune the tuning parameters, please see the methodological details in Egami and Lee (2023+) (https://naokiegami.com/paper/sps.pdf).
seed: Numeric. seed used internally. Default = 1234.
max_iter: Numeric. The number of iterations used in the optimization. Default = 10.
solver: Solver we use in the internal CVXR optimization. Default = ECOS_BB. See the CVXR website for information on other solvers.
num_iter: Numeric. The number of iterations used in the underlying optimization.

Value

sps returns an object of sps class.

selected_sites: Names of sites the SPS algorithm selected.
W: Estimated weights to approximate non-selected sites using selected sites. W will be used in the subsequent estimation of the average-site ATE.
obj: Estimated values of the objective function, separately for three parts.
internal: Objects useful for internal use of the function.

References

Egami and Lee. (2023+). Designing Multi-Context Studies for External Validity: Site Selection via Synthetic Purposive Sampling. Available at https://naokiegami.com/paper/sps.pdf.