control_out constructs a list with all necessary control parameters
for outcome model.
control_out(
epsilon = 1e-04,
maxit = 100,
trace = FALSE,
k = 5,
penalty = c("SCAD", "lasso", "MCP"),
a_SCAD = 3.7,
a_MCP = 3,
lambda_min = 0.001,
nlambda = 100,
nfolds = 10,
treetype = c("kd", "rp", "ball"),
searchtype = c("standard", "priority"),
pmm_match_type = 1,
pmm_weights = c("none", "dist"),
pmm_k_choice = c("none", "min_var"),
pmm_reg_engine = c("glm", "loess"),
npar_loess = stats::loess.control(surface = "direct", trace.hat = "approximate")
)List with selected parameters.
Tolerance for fitting algorithms. Default is 1e-6.
Maximum number of iterations.
logical value. If TRUE trace steps of the fitting algorithms. Default is FALSE.
The k parameter in the RANN::nn2() function. Default is 5.
penalty algorithm for variable selection. Default is SCAD
The tuning parameter of the SCAD penalty for outcome model. Default is 3.7.
The tuning parameter of the MCP penalty for outcome model. Default is 3.
The smallest value for lambda, as a fraction of lambda.max. Default is .001.
The number of lambda values. Default is 100.
The number of folds during cross-validation for variables selection model.
Type of tree for nearest neighbour imputation (for the NN and PMM estimator) passed to RANN::nn2() function.
Type of search for nearest neighbour imputation (for the NN and PMM estimator) passed to RANN::nn2() function.
(Only for the PMM Estimator)
Indicates how to select 'closest' unit from non-probability sample for each
unit in probability sample. Either 1 (default) or 2 where
2 is matching by minimizing distance between \(\hat{y}_{i}\) for
\(i \in S_{A}\) and \(y_{j}\) for \(j \in S_{B}\) and 1
is matching by minimizing distance between \(\hat{y}_{i}\) for
\(i \in S_{A}\) and \(\hat{y}_{i}\) for \(i \in S_{A}\).
(Only for the PMM Estimator)
Indicate how to weight k nearest neighbours in \(S_{B}\) to
create imputed value for units in \(S_{A}\). The default value
"none" indicates that mean of k nearest \(y\)'s from
\(S_{B}\) should be used whereas "prop_dist" results in
weighted mean of these k values where weights are inversely
proportional to distance between matched values.
(Only for the PMM Estimator) Character value indicating how k hyper-parameter
should be chosen, by default "none" meaning k provided in
control_outcome argument will be used. For now the only other
option "min_var" means that k will be chosen by minimizing
estimated variance of estimator for mean. Parameter k provided in
this control list will be chosen as starting point.
(Only for the PMM Estimator) whether to use parametric ("glm")
or non-parametric ("loess") regression model for the outcome. The default is "glm".
control parameters for the stats::loess via the stats::loess.control function.
nonprob() -- for fitting procedure with non-probability samples.