R help - Mar 2011 - Avoiding choosing parameters with mix[mixdist]

Hi,

I am working on a population of an invasive clam. The data are the size of each
clam per station (2mm on average). Each station is found at a different distance
from a power nuclear station, so at different water temperatures. The fist step
I want to do is to identify cohort size at each station or (zone of water
temperature). The second step will be to see whether the size or number of
cohort differ based on temperature.

I use mixtdist package and more specifically function mix to id Cohort.

Here is an exemple :

mixp1<-mixparam(c(0.5,2.0,3.8,5.4,8.4,10.4),c(0.4,0.4,0.4,0.4,0.4,0.4))
fitcohgr60_1 <- mix(siz9gr60,mixp1,"norm")

The problem with mix is that I  had to impose the mixparam myself (both mean and
se) and than mix model modified the entered parameters so it fitted my data
best. Therefore suggested cohort strongly depend of what and how much parameters
I put first.  The parameter included are  supposed to be based on cohort size in
literature, but in this case cohort size strongly depends on temperature (and my
samples are smaller than in literature). Therefore, I tried to estimate mean
from what I saw graphically, but SE are particularly difficult to guesstimate
and I do not want to influence my results by choosing the parameters myself. In
addition I had to chose the distribution function, but again I would have like
the model to tell me which one was best.

My question is : Is there any way, with mixdist or other package, not to impose
parameters, but rather having the function to suggest you what was best by
analyzing your data. I read couple document on the subject but I haven't
seen that anywhere.


I hope this is clear...

Many thanks


Anouk Simard
Courriel :
anouk.simard@mrnf.gouv.qc.ca<mailto:anouk.simard@mrnf.gouv.qc.ca>




	[[alternative HTML version deleted]]