[time-nuts] I think I've become a Volt nut too.

[Magnus Danielson]

Mike Naruta AA8K wrote:
> 
> Sure Chuck.  What I was talking about was a part of statistics that we 
> in our gnat-hair-splitting compulsive group may forget about.
> 
> Let's assume that our 100,000 standards were carefully calibrated 
> against THE standard.  There is a small amount of error in the 
> calibration process.  Let us even assume that the error in the 
> calibration process is normally distributed.
> 
> It is not impossible that for a sample of 100,000 secondary standards, 
> that the errors would be all be off in the same direction, compared to 
> the standard's value.
> 
> Now, granted, this would be a small probability indeed.  But it is 
> possible to toss a coin fifty times and have fifty "heads". The smart 
> bet is that it won't.

Well, if the distribution of these is only random and of benign 
randomness like gaussian noise.

If you have an aging mechanism for instance, over time this huge set 
would drift in that direction and that would produce a moving average 
value...

The rate of calibration to a primary standard would be one of the 
parameters needed to set the limit of drift.

So, systematic drift is not canceled by large number statistics. It just 
doest not obey the underlying assumption. Long-term noise of clocks 
obery the f^-3 noise which does not converge nicely and statistical 
measures needs to be adapted to provide reasnoble measures. This is why 
we have ADEV and friends.

Again, this is why you need to separate stability with reproducability 
aspects.

Cheers,
Magnus