Hi
I use WinBUGS occasionally for statistical analysis. I would like some
generic advice
on the preferred method of analysing grouped continuous data. As a trivial
example,
I am appending a .odc file reporting the modelling and analysis of some
simulated
grouped Gaussian data.
In this file the data have been analysed two ways. The first method
calculates the
probability of the data falling in each group and uses a multinomial
distribution.
However, this type of approach is not used much in the WinBUGS Examples,
where the
models are nearly always specified in terms of random variables, rather than
probabilities. The method seems quite slow even on this simple problem.
Indeed,
the probability/multinomial method does not work well (it's incredibly slow)
on more
complex real problems that I am faced with. Also, the Update Tool ticks the
`adapting'
box.
The second method attempts to generate an equivalent ungrouped data set with
a string
of NAs, and upper and lower censoring limits. Unfortunately it does not
compile.
WinBUGS reports a `type guard' trap, which I do not understand. I would like
to
know the problem here. In any case, it is somewhat tedious to have to
generate and manipulate
the ungrouped data. What if I had 50000 grouped data values?
More generally, I would like to know if there is a better way (than either
of these)
to use WinBUGS to analyse such data.
Many thanks
Geoff Laslett
----------------------------------------------------------------------------
--
Geoff Laslett
CSIRO Mathematical and Information Sciences
Private Bag 10
Clayton South MDC
Clayton
Victoria 3169
Australia
Tel: +61 3 9545 8018
Fax: +61 3 9545 8080
email: [log in to unmask]
----------------------------------------------------------------------------
--
|