rControl: Control of Latent Factor Model Number Estimation Constructs...

In hheiling/glmmPen: High Dimensional Penalized Generalized Linear Mixed Models (pGLMM)

Control of Latent Factor Model Number Estimation Constructs the control structure for the estimation of the number of latent factors (r) for use within the glmmPen_FA and glmm_FA estimation procedures.

Description

Control of Latent Factor Model Number Estimation

Constructs the control structure for the estimation of the number of latent factors (r) for use within the glmmPen_FA and glmm_FA estimation procedures.

Usage

rControl(
  r = NULL,
  r_max = NULL,
  r_est_method = "GR",
  size = 25,
  sample = FALSE
)

Arguments

r	positive integer specifying number of latent common factors to assume in the model. If NULL (default), this value estimated from the data. See r_est_method for available estimation procedures, and the Details section for further details on the general estimation procedure. If r is specified, the no estimation procedure is performed and the algorithm uses the input value or r. All other parameters for this function are relevant for the estimation procedure.
r_max	positive integer specifying maximum number of latent factors to consider. If NULL (default), this value is automatically calculated.
r_est_method	character string indicating method used to estimate number of latent factors r. Default "GR" uses the Growth Ratio method of Ahn and Horenstein (2013) (<doi:10.3982/ECTA8968>). Other available options include "ER" for the Eigenvalue Ratio method of Ahn and Horenstein (2013) (<doi:10.3982/ECTA8968>) and "BN1" or "BN2", the Bai and Ng (2002) method (<dio:10.1111/1468-0262.00273>) using one of two penalties: (1) (d + p) / (d p) log(d p/(d+p)) or (2) (d + p) / (d p) log(min(d,p)) where d is the number of groups in the data and p is the number of total random effect covariates (including the intercept)
size	positive integer specifying the total number of pseudo random effect estimates to use in the estimation procedure for the number of latent factors r, which is restricted to be no less than 25. If this size is greater than the number of groups in the data (i.e.~the number of levels of the grouping variable), then a sampling procedure is used to increase the number of pseudo estimates to the value of size if the value of sample is TRUE.
sample	logical value indicating if the total number of pseudo random effect estimates to use in the estimation procedure for the number of latent common factors r should be larger than the number of unique groups in the data, where the number of pseudo estimates are increased to the value of size using a sampling procedure. Default is FALSE. If TRUE, the sampling procedure is only performed if the value of size is greater than the number of groups in the data.

Details

Estimation of r procedure: For each level of the group variable separately, we identify the observations within that group and fit a regular penalized generalized linear model where the penalty value is the minimum fixed effect penalty. These group-specific estimates, which we label as 'pseudo random effects', are placed into a matrix G (rows = number of levels of the grouping variable, columns = number of random effect covariates), and this pseudo random effects matrix is treated as the observed outcome matrix used in the "GR", "ER", and "BN" estimation procedures described above in the description of r_est_method.

hheiling/glmmPen documentation built on Jan. 15, 2024, 11:47 p.m.