fitsaemodel.control function

Tuning Parameters of fitsaemodel

This function is used to define global settings and parameters that are used by fitsaemodel().

fitsaemodel.control(niter = 40, iter = c(200, 200), acc = 1e-05,
    dec = 0, decorr = 0, init = "default", k_Inf = 20000, ...)

Arguments

• niter: [integer] the maximum number of outer-loop iterations (default: niter = 40).

• iter: [integer] the maximum number of inner-loop iterations. It can be a vector of size 2. The first element of the vector refers to the estimation of the regression coefficients $beta$; the second element refers to the estimation of the variance of the unit-level errors, $v$; the maximum number of iterations used to compute the ratio of variances, $d$, cannot be modified (default: iter = c(200, 200)).

• acc: [numeric] numeric tolerance used in the termination rule of the iterative updating algorithms. It can be a vector of size 4. The positions 1:4 of the vector acc refer to 1: (overall) outer-loop, 2: regression coefficients, $beta$, 3: variance component, $v$, 4: ratio of variances $d$; default: acc = 1e-05.

• dec: [integer] type of matrix square root (decomposition); dec = 0 for eigenvalue decomposition (default) or dec = 1 for Cholesky decomposition.

• decorr: [integer] type of decorrelation of the residuals; decorr = 0: no robust decorrelation (default); decorr = 1: means are replaced by medians.

• init: [character] method by which the main algorithm is initialized. By default, init = "default"

  the algorithm is initialized by a robust fixed-effects estimator; alternatively, (and provided that the robustbase package is installed) one may choose one of the high-breakdown-point initial estimators: "lts" (fast least-trimmed squares, LTS, regression) or "s" (regression S-estimator). For more details on the initialization methods, see documentation of fitsaemodel().

• k_Inf: [numeric] tuning constant of the robust estimator that represents infinity (default: k_Inf = 20000).

• ...: additional arguments (not used).

Details

Changing the default values of the parameters may result in failure of convergence or loss of convergence speed.

Returns

A list with entries

• niter
• iter
• acc
• k_Inf
• init
• dec
• decorr
• add

See Also

fitsaemodel()

Examples

# use the landsat data
head(landsat)

# define the saemodel using the landsat data
model <- saemodel(formula = HACorn ~ PixelsCorn + PixelsSoybeans,
    area = ~CountyName,
    data = subset(landsat, subset = (outlier == FALSE)))

# summary of the model
summary(model)

# obtain the maximum likelihood estimates with, for instance, 'niter = 50'
# number of outer-loop iterations (by default: niter = 40). Here, we use
# 'niter = 50' for the sake of demonstration, not because it is needed.
fitsaemodel("ml", model, niter = 50)