filter_lambda: Optimal lambda and corresponding final GLMM LASSO model

In SOCR/DataSifterII: Data Sifter II Algorithm to Create Privacy-preserving Time-varying Data

Description

Select the optimal lambda that minimizes BIC. The starting values of the parameters play an essential part in GLMM LASSO model fitting. We first consider glmmPQL fitted values. Then, we consider 0s for sparse signal scenario.

Usage

filter_lambda(
  data,
  outcome,
  ID,
  lambdaseq,
  range = 0,
  family = gaussian(link = "identity"),
  startPQL = TRUE
)

Arguments

data	A long format complete data frame.
outcome	Outcome variable of interest.
ID	Name of the ID variable in the dataset.
lambdaseq	A vector of lambdas to consider.
range	The next lambda sequence to consider, given the best lambda from current lambdaseq.
family	Link function for the outcome.
startPQL	Logical. If TRUE initiate parameter estimates are generated from glmmPQL. IF FALSE, initiate parameter estimates are 0s.

Value

• glm1 - The optimal GLMM under LASSO regularization with smallest BIC.

• seqs - The next Lambda sequence to be considered, when further search is needed.

SOCR/DataSifterII documentation built on Dec. 15, 2021, 10:29 a.m.