Another thing, I do the modelling for the Dutch and international colony loss data (the COLOSS questionnaires) and you can make models as fancy as you want but there generally three problems:  a) the data are "crappy", they contains all sorts of gaps and noise. For modeling this means: garbage in, garbage out.  b) the coverage is generally very low, 20% of the beekeeper population ending up in the data is considered a really good result. The loss and characteristics of the other 80% are unknown.  c) the data are peculair, they have specific shapes of frequency distributions and all kinds of clustering (aka non-independence) of observations is present. This means that in my opinion you can only "mine" the loss data for associations, introducing some crude categorical factors and trying to involve clustering of data into the models. Models generally point to strong location and beekeeper effects on loss numbers.  The data are generally unfit for prediction but not always, when "enough" good quality data is present you see that "bad locations" in one year tend to be "bad locations" in later years.  atb,Lennard  
  		 	   		  
             ***********************************************
The BEE-L mailing list is powered by L-Soft's renowned
LISTSERV(R) list management software.  For more information, go to:
http://www.lsoft.com/LISTSERV-powered.html

Guidelines for posting to BEE-L can be found at:
http://honeybeeworld.com/bee-l/guidelines.htm