runm {unmconf}

Generate synthetic data

Description

runm() generates synthetic data for use of modeling with unmeasured confounders. Defaults to the case of one unmeasured confounder present and fixed parameter values. Can be customized. Currently set up to have at most two unmeasured confounders to pair with unm_glm().

Usage

runm(
  n,
  type = "int",
  missing_prop = 0.8,
  response = "bin",
  response_param = NULL,
  response_model_coefs = c(int = -1, z1 = 0.5, z2 = 0.5, z3 = 0.5, u1 = 0.5, x = 0.5),
  treatment_model_coefs = c(int = -1, z1 = 0.5, z2 = 0.5, z3 = 0.5, u1 = 0.5),
  covariate_fam_list = list("norm", "bin", "norm"),
  covariate_param_list = list(c(mean = 0, sd = 1), prob = 0.3, c(0, 2)),
  unmeasured_fam_list = list("norm"),
  unmeasured_param_list = list(c(mean = 0, sd = 1))
)

Arguments

n

Number of observations. When type = "int", n is a vector of length 1. When type = "ext", n can either be a vector of length 1 or 2. For the case when n is of length 2, ⁠n = (n_main, n_external)⁠, where n_main corresponds to the main study sample size and n_external corresponds to the external validation sample size. For the case when n is of length 1, n will be split evenly between main study and external validation observations, with the main study getting the additional observation when n is odd.

type

Type of validation source. Can be "int" for internal validation or "ext" for external validation. Defaults to "int".

missing_prop

Proportion of missing values for internal validation scenario (i.e., when type = "int").

response

"norm", "bin", "pois", or "gam". Defaults to "bin".

response_param

Nuisance parameters for response type. For "norm", the default standard deviation is 1. For "gam", the default shape parameter is 2. For "pois", an offset variable is added to the dataset that is uniformly distributed from 1 to 10.

response_model_coefs

A named vector of coefficients to generate data from the response model. This must include an intercept (⁠"int" = ⁠), a coefficient for each covariate specified, a coefficient for each unmeasured confounder, and a treatment coefficient (⁠"x" = ⁠). The coefficients for the covariates and treatment will be denoted with "beta[.]" and the unmeasured confounders with "lambda[.]".

treatment_model_coefs

A named vector of coefficients to generate data from the treatment model. This must include an intercept (⁠"int" = ⁠), a coefficient for each covariate specified, and a coefficient for each unmeasured confounder. The coefficients for the covariates and unmeasured confounders will be denoted with "eta[.]".

covariate_fam_list

A list of either "norm" or "bin", where the length of the list matches the number of covariates in the model.

covariate_param_list

A list of parameters for the respective distributions in covariate_fam_list, where the length of the list matches the length of covariate_fam_list.

unmeasured_fam_list

A list of either "norm" or "bin", where the length of the list matches the number of unmeasured confounders in the model. This can be at most a length of 2 to pair with unm_glm().

unmeasured_param_list

A list of parameters for the respective distributions in unmeasured_fam_list, where the length of the list matches the length of unmeasured_fam_list.

Value

A tibble

Examples

runm(100)
runm(n = 100, type = "int", missing_prop = .75)
runm(n = 100, type = "int", missing_prop = .75) |> attr("params")
runm(100, type = "int", response = "norm")
runm(100, type = "int", response = "norm") |> attr("params")
runm(100, type = "int", response = "norm", response_param = 3) |> attr("params")
runm(100, type = "int", response = "gam")
runm(100, type = "int", response = "gam", response_param = 5) |> attr("params")
runm(100, type = "int", missing_prop = .5, response = "pois")

runm(n = 100, type = "ext")
runm(n = 100, type = "ext") |> attr("params")
runm(n = c(10, 10), type = "ext")
runm(100, type = "ext", response = "norm")
runm(100, type = "int", response = "norm", response_param = 3) |> attr("params")
runm(100, type = "ext", response = "gam")
runm(100, type = "ext", response = "pois")

runm(
  n = 100,
  type = "int",
  missing_prop = .80,
  response = "norm",
  response_param = c("si_y" = 2),
  response_model_coefs = c("int" = -1, "z" = .4,
                           "u1" = .75, "u2" = .75, "x" = .75),
  treatment_model_coefs = c("int" = -1, "z" = .4,
                            "u1" = .75, "u2" = .75),
  covariate_fam_list = list("norm"),
  covariate_param_list = list(c(mean = 0, sd = 1)),
  unmeasured_fam_list = list("norm", "bin"),
  unmeasured_param_list = list(c(mean = 0, sd = 1), c(.3))
)

runm(
  n = c(20, 30),
  type = "ext",
  response = "norm",
  response_param = c("si_y" = 2),
  response_model_coefs = c("int" = -1, "z1" = .4, "z2" = .5, "z3" = .4,
                           "u1" = .75, "u2" = .75, "x" = .75),
  treatment_model_coefs = c("int" = -1, "z1" = .4, "z2" = .5, "z3" = .4,
                            "u1" = .75, "u2" = .75),
  covariate_fam_list = list("norm", "bin", "norm"),
  covariate_param_list = list(c(mean = 0, sd = 1), c(.3), c(0, 2)),
  unmeasured_fam_list = list("norm", "bin"),
  unmeasured_param_list = list(c(mean = 0, sd = 1), c(.3))
)