Ow Mun Heng wrote:
> The results are valid (verified with actual data) but I don't
> understand
> the logic. All the Statistical books I've read marked stdev as sqrt
> (sum(x - ave(x))^2 / (n - 1). The formula is very different, hence the
> confusion.
A formula is not an algorithm. In particular, the naive way of
calculating variance or standard deviation has massive numerical
instability problems - anything involving sums of squares does.
There are a variety of alternate algorithms for stddev/variance, I
presume your other algorithm is similarly trying to avoid these same
issues (but I have not looked closely at it). You can also see
Wikipedia for one of the most well known, due to Knuth/Wellford:
http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
- John D. Burger
MITRE
