Why the -1/2 power law?

Juan is coming by in a bit to finish setting up the bootloader on the Acer after installing the XP partition.

I realized after my way out on Wednesday that my interpretation of the power-law scaling of the OCXO's Allan deviation was completely wrong - it would have predicted a -2 scaling, rather than the -1/2 that we see!

I discussed this with Ray this morning, and he thinks it is due to a sample-size effect. The standard deviation sigma_mean of a measurement of a mean from a sample is 1/sqrt(N) times the standard deviation sigma of the underlying distribution, where N is the number of elements of the sample. So, if we think of Allan deviation as being a measurement of the mean frequency displacement between adjacent averaging windows, where the displacement is a difference of frequencies, and the mean frequencies are both measured over windows of size N containing N elements, and the instantaneous frequency fluctuates with a standard deviation of sigma, then one would expect the mean frequency measurements to have fluctuations of magnitude 1/sqrt(N) lower than the magnitude of the real underlying frequency fluctuations.