Transforms of a random effect eta with an estimated shape parameter, used
as cl <- tcl*exp(etaBoxCox(eta.cl, lambda)) (Petersson et al. 2009). Each
keeps eta = 0 at 0 and the rank order of eta, and has a shape giving the
identity, so the shape can be tested with a likelihood ratio test.
etaBoxCox() is (exp(eta)^lambda - 1)/lambda, which is eta at
lambda = 0 and skews the distribution (roughly lambda in (-3, 3)).
etaJohnDraper() is the modulus transform
sign(eta)*((abs(eta) + 1)^lambda - 1)/lambda, the identity at
lambda = 1, which changes the tails symmetrically. etaYeoJohnson() is
((eta + 1)^lambda - 1)/lambda for eta >= 0 and
-((1 - eta)^(2 - lambda) - 1)/(2 - lambda) below, the identity at
lambda = 1, which skews. etaHeavyTail() is eta*abs(eta)^theta
(Petersson's heavy-tailed transform), the identity at theta = 0, with
heavier tails for theta > 0 and a symmetric bimodal shape for theta in
(-1, 0).
etaLogit() is Petersson's logit transform
(expit(logit(theta1) + eta) - theta1)*theta2 (theta1 in (0, 1) sets the
skewness, theta2 > 0 the width; it is near the identity for theta1 = 0.5,
theta2 = 4). etaSinhAsinh() is the sinh-arcsinh transform of Jones and
Pewsey, centered so eta = 0 stays at 0,
sinh(delta*asinh(eta) - eps) + sinh(eps); eps sets the skewness and
delta > 0 the tails (heavier above 1), the identity at eps = 0,
delta = 1. etaTukeyGH() is Tukey's g-and-h transform
(exp(g*eta) - 1)/g*exp(h*eta^2/2); g sets the skewness and h >= 0 the
tails, the identity at g = 0, h = 0.
exprelD(z, n) is the n-th derivative of (exp(z) - 1)/z; it keeps every
derivative exact through a shape of 0. absPowLog(x, p, k) is
abs(x)^p*log(abs(x))^k with its limit at x = 0, the derivative helper of
etaHeavyTail(). In a model, named arguments are matched and expanded into the
positional call, eg etaBoxCox(lambda = lcl, eta = eta.cl).
Usage
etaBoxCox(eta, lambda)
etaJohnDraper(eta, lambda)
exprelD(z, n)
etaYeoJohnson(eta, lambda)
etaHeavyTail(eta, theta)
etaLogit(eta, theta1, theta2)
etaSinhAsinh(eta, eps = 0, delta = 1)
etaTukeyGH(eta, g = 0, h = 0)
absPowLog(x, p, k)Arguments
- eta
random effect
- lambda
shape parameter
- z
value of
z- n
derivative order (non-negative integer)
- theta
heavy-tail shape parameter
- theta1
logit skewness parameter, in
(0, 1)- theta2
logit width parameter, positive
- eps
sinh-arcsinh skewness parameter
- delta
sinh-arcsinh tail parameter, positive
- g
Tukey g (skewness) parameter
- h
Tukey h (tail) parameter, non-negative
- x
value of
x- p
power
- k
power of the log
References
Petersson KJ, Hanze E, Savic RM, Karlsson MO. Semiparametric distributions with estimated shape parameters. Pharm Res. 2009; 26(9):2174-85.
Jones MC, Pewsey A. Sinh-arcsinh distributions. Biometrika. 2009; 96(4):761-80.
See also
Other Special functions:
boxCox(),
delay(),
erf(),
gammap(),
gammapDer(),
gammapInv(),
gammaq(),
gammaqInv(),
logit(),
lowergamma(),
mexpit(),
mlogit(),
phi(),
phiU(),
probit(),
uppergamma()
Examples
etaBoxCox(c(-1, 0, 1), 0.5)
#> [1] -0.7869387 0.0000000 1.2974425
etaBoxCox(c(-1, 0, 1), 0) ## identity
#> [1] -1 0 1
etaJohnDraper(c(-1, 0, 1), 1) ## identity
#> [1] -1 0 1
etaYeoJohnson(c(-1, 0, 1), 1) ## identity
#> [1] -1 0 1
etaHeavyTail(c(-1, 0, 1), 0.5)
#> [1] -1 0 1
etaLogit(c(-1, 0, 1), 0.5, 4)
#> [1] -0.9242343 0.0000000 0.9242343
etaSinhAsinh(c(-1, 0, 1), 0, 1) ## identity
#> [1] -1 0 1
etaTukeyGH(c(-1, 0, 1), 0, 0) ## identity
#> [1] -1 0 1
exprelD(0, 0:3)
#> [1] 1.0000000 0.5000000 0.3333333 0.2500000
