LaMo: 21 Feb 2008.
[Q]: The package appears not to work, when I try e.g. demo(“csda-article”) I get an error message like:
...
R> Error in .jnew("sir/sim/SimSellke") :
R> Failed to create object of class `sir/sim/SimSellke'
[A]: In order for RLadyBug to
work properly one needs read/write access rights to the current
directory (getwd()). Furthermore, the R routines in many cases
interface an underlying Java program, which only works with the help of
the rJava package, which requires a working java virtual machine to
work. To check if rJava is working properly try the following:
R> library("rJava")
R> .jinit()
R> f <- .jnew("java/awt/Frame","Hello")
R> .jcall(f,,"setVisible",TRUE)
A small window with the name "Hello" should appear. If these steps go
through then rJava is working and the problem is not in rJava.
[Q]: The program takes forever and I have no idea of why it is taking so long.
[A]: The underlying Java
program writes a debug file ladybug.system.out into the current working
directory (getwd()). There you also find a ladybug.system.err file,
which contains any error messages, which are written to the screen
after the program ends.
[Q]: So “asis” for the incubation time means "its already the correct value, no estimation is necessary"?
[A]: Yes, one states that the
observed incubation time (the time between the E and I event) is the
actual length of the incubation time. The other option would be to
specify the incubation time to be a fixed constant, but in some cases
one wants more flexibility as one fixed number does not work. If one
knows that the incubation is subject to variations then it is necessary
to specify an appropriate distribution.
[Q]: The mathematical
developments on page 357 of the Höhle et al 2005 paper are rather
complicated. Do you confirm that beta indeed follows a gamma
distribution?
[A]: It is only the prior
distribution of beta which is assumed to follow a gamma distribution -
this is a typical approach in Bayesian inference for this topic (see
e.g. the quoted O'Neill papers). The bigger the variance of the
posterior the less information you put on beta a priori - i.e. the data
are allowed to speak. The posterior distribution of beta can have a
shape entirely different from a gamma distribution
[Q]: Why is a start value needed between parentheses, when values are unknown?
[A]: Because in the MCMC
sampling scheme one has to begin with a valid configuration of the
data, this means e.g. that there cannot come new exposed after the
number of infectious dropped to zero, etc. Automatically finding such a
configuration for a big set of missing data is not so easy, but
checking if a configuration is valid is straightforward. Therefore I
decided to leave this problem to the user - he/she has to pecify an
appropriate start value.
[Q]: Fortunately, in human
epidemiology, removal occurs once an individual is cured, in the vast
majority of infections. Removal times are therefore rarely (if ever)
recorded (but you are aware of this fact, owing to your interest in
surveillance methods). But unfortunately, recovery times seem
compulsory in RLadyBug. If so, it considerably narrows its scope in
human epidemiology.
[A]: This is indeed a problem, which was not handled in previous version 0.4-2 of RLadyBug.
Starting from version 0.4-3 unknown recovery times are also allowed and
are indicated by putting the recovery time in parenthesis, i.e. (start
value). Furthermore, one has to indicate an end time T until which the
epidemic was observed:
...
T 100
//x y E
I R D ignore
1 1 0 2 (18) 100
However, the software currently does not allow the recovery times to be censored, thus they are all forced to occur before T.