These functions are all methods for class aster or
summary.aster objects.
# S3 method for aster
summary(object, info = c("expected", "observed"),
info.tol = sqrt(.Machine$double.eps), show.graph = FALSE, ...)# S3 method for summary.aster
print(x, digits = max(3, getOption("digits") - 3),
signif.stars = getOption("show.signif.stars"), ...)
summary.aster returns an object of class "summary.aster"
list with the same components as object, which is of class
"aster".
an object of class "aster", usually, a result of a
call to aster.
the type of Fisher information use to compute standard errors.
tolerance for eigenvalues of Fisher information.
If eval is the vector of eigenvalues of the information matrix,
then eval < cond.tol * max(eval) are considered zero. Hence the
corresponding eigenvectors are directions of constancy or recession of
the log likelihood.
if TRUE, show the graphical model.
an object of class "summary.aster", usually, a result of a
call to summary.aster.
the number of significant digits to use when printing.
logical. If TRUE, “significance stars”
are printed for each coefficient.
further arguments passed to or from other methods.
This function may give an error message
"cannot compute standard errors, apparent directions of recession".
There are two reasons why this can happen.
There may actually be a direction of recession (DOR). Then the maximum likelihood estimate does not exist; increasing the likelihood drives (some of) the coefficients to infinity or minus infinity.
This function's guess at the DOR can be extracted
from the error object obtained by wrapping this function
in try and then extracting the dor component
of the condition attribute of the error object.
An example of this is on the help page for the foobar
data set.
This function's guessed DOR are apparent null eigenvector(s) of the Fisher
information matrix. Due to inaccuracy of computer arithmetic, this is
only a guess. What are deemed null eigenvectors is controlled by the
info.tol argument of this function. Reducing info.tol
to perhaps 1e-9 or 1e-10 or even a little lower
may make the putative DOR go away. In this case they were probably
bogus (see next item). Reducing info.tol to near or below the
machine epsilon .Machine$double.eps (.Machine)
instructs this function to feed you garbage with no error or warning.
Putative DOR are probably true DOR if they are highly patterned with many zero or nearly zero components and other components that are nearly (small) integer multiples of each other. Putative DOR are probably bogus if they look like random noise.
DOR, if true, cannot simply be ignored. For more information,
including how to do more rigorous investigation of whether putative
DOR are true or bogus,
see the example on the help page for the foobar
data set and the reference cited on that help page.
All of the putative directions of recession may be bogus. Due to inaccuracy of computer arithmetic, ill-conditioning of predictor variables, or ill-conditioning of the aster model itself (large graphs cause more inaccurate computation), what appear to be null eigenvectors of the Fisher information matrix need not be true null eigenvectors.
In this case, the problem will go away when info.tol is decreased
slightly. Only when one has proved that there is no DOR, should one
use info.tol = 1e-20 which says to ignore the problem altogether
(whether putative DOR are true or bogus).
aster, summary.