Randomisations - randomised cluster outputs should seed PRNG from init state of original output

[GoogleCodeExporter]

Cluster analyses currently use a random selection as a tie breaker (see issue 
#116).  

The issue here is that the system will always seed from parameter 
PRNG_LAST_STATE in preference to PRNG_INIT_STATE.  In most cases such behaviour 
is sensible as it guarantees a continuing sequence of random values.  

However, the issue with the randomisations is that it guarantees that all 
randomisation cluster analyses will always start from the last PRNG state in 
the original cluster analysis.  It is more sensible to use the same start point 
so the same sequence of values is used across the original and randomised, 
rather than one for the original and a second for the randomised.  

Possibly there should be an option to use a continuing sequence of values, but 
this makes parallelisation difficult.  


The main impact of this change will be on the rand_no_change randomisation, but 
any effect is currently illusory as the randomised clusters will always be the 
same and yet different from the original.  Possibly rand_no_change should have 
an option to override the PRNG_INIT_STATE to generate a more random sequence.

Original issue reported on code.google.com by shawnlaffan on 10 Oct 2011 at 4:28