Computes some square root--log mixture link transformations, including their inverse and the first few derivatives.
sloglink(theta, bvalue = NULL, taumix.log = 1, tol = 1e-13,
nmax = 99, inverse = FALSE, deriv = 0, short = TRUE,
tag = FALSE, c10 = c(2, -2))
lcsloglink(theta, bvalue = NULL, pmix.log = 0.01, tol = 1e-13,
nmax = 99, inverse = FALSE, deriv = 0, short = TRUE,
tag = FALSE, c10 = c(2, -2))The following holds for the
linear combination (LC)
variant.
For deriv = 0,
(1 - pmix.log) * sqrtlink(mu, c10 = c10)
+ pmix.log * loglink(mu)
when inverse = FALSE,
and if inverse = TRUE then a nonlinear
equation is solved for mu,
given
eta passed in as theta.
For deriv = 1, then the function
returns d
eta / d
theta as a function of theta if
inverse = FALSE, else if inverse
= TRUE then it returns the reciprocal.
Numeric or character. See below for further details.
See Links.
Numeric, of length 1.
Mixing parameter directed at
loglink. Then
1 - exp(-taumix.log * theta) is
used to weight
sqrtlink.
Thus a 0 value will result in
loglink,
and a very large numeric such as 1e4
should be roughly equivalent to
sqrtlink over almost
all of the parameter space.
Numeric, of length 1.
Mixing probability assigned
to loglink. Then
1 - pmix.log is used to weight
sqrtlink.
Thus a 0 value will result in
sqrtlink
and 1 is equivalent to loglink.
Arguments fed into a function implementing a vectorized bisection method.
Details at Links.
See sqrtlink
and
loglink.
Thomas W. Yee
The default values for taumix.log
and pmix.log
may change in the future.
The name and order of the arguments
may change too.
For general information see
alogitlink.
alogitlink,
sqrtlink,
loglink,
Links,
poissonff,
hdeff.
mu <- seq(0.01, 3, length = 10)
sloglink(mu)
max(abs(sloglink(sloglink(mu), inv = TRUE) - mu)) # 0?
Run the code above in your browser using DataLab