vows (version 0.3.1)

semipar.mix.mp: Massively parallel semiparametric mixed models

Description

Fits a set of semiparametric mixed models, with a common design matrix, by repeated calls to gamm4. Only a single smooth term is permitted.

Usage

semipar.mix.mp(Y, x, param = NULL, random, data.ran, k = 10, norder = 4,
  pen.order = 2, knots = "quantile", store.gamm4 = FALSE)

Arguments

Y

$n \times V$ response matrix.

x

a vector giving the predictor upon which each column of Y is regressed.

param

a matrix or vector for the parametric terms in the model.

random

a formula, passed to gamm4, specifying the random effects structure in lmer style. See the example.

data.ran

a required data frame containing the factors used for random effects.

k

number of knots.

norder

order of B-splines: the default, 4, gives cubic B-splines.

pen.order

order of the derivative penalty.

knots

knot placement for the B-spline bases. The default, "quantile", gives knots at equally spaced quantiles of the data. The alternative, "equispaced", gives equally spaced knots.

store.gamm4

logical: should the gamm4 objects to be stored in the output? FALSE by default.

Value

• coef
  matrix of the coefficients obtained from gamm4 looping (including both parametric and nonparametric parts).
• bsplinecoef
  matrix of B-spline coefficients.
• pwdf
  vector of pointwise effective degrees of freedom.
• pwlsp
  vector of pointwise log smoothing parameters: grid values maximizing the restricted likelihood at each point.
• B
  matrix of basis function values.
• C
  the constraint matrix.
• Z
  transformation matrix to impose constraints.
• basis
  B-spline basis object, of the type created by the fda package; the coefficient estimates are with respect to this basis.

Details

Unlike semipar.mp, this function does not use large matrix multiplications to avoid looping through model fits. Instead it performs a separate call to gamm4 to fit a semiparametric mixed model for each column of Y.

Examples

Y = matrix(rnorm(3000),,3)
x1 = rnorm(1000)
x2 = matrix(rnorm(2000),,2)
family.fac <- factor(rep(1:20,rep(50,20)))
person.fac <- factor(rep(rep(1:25,rep(2,25)),rep(20,50)))
semimix = semipar.mix.mp(Y = Y, x = x1, param = x2, random = ~ (1|a/b),
          data.ran = data.frame(a = family.fac, b = person.fac))