I lol at this

Weather At-a-Glance:

On Average:

Marietta, GA is warmer than Corvallis, OR by 6°F.   Marietta, GA is wetter than Corvallis, OR by 8.2 in.

Records:

Corvallis, OR recorded the highest temperature of 108°F in 1981.   Marietta, GA recorded the lowest temperature of -10°F in 1985.

"Marietta, GA is wetter than Corvallis...."

Weather At-a-Glance:

On Average:

The average temperatures are the same for San Francisco, CA and Marietta, GA.   Marietta, GA is wetter than San Francisco, CA by 29.5 in.

Records:

San Francisco, CA recorded the highest temperature of 103°F in 2000.   Marietta, GA recorded the lowest temperature of -10°F in 1985.

"Marietta, GA is wetter than San Francisco..."

Strange, I think. I would not have expected this. I suppose its looking at total precip, not precip frequency??

Weather At-a-Glance:

On Average:

San Francisco, CA is warmer than Corvallis, OR by 6°F.   Corvallis, OR is wetter than San Francisco, CA by 21.4 in.

Again, how interesting when you think of these charts. Corvallis is sort of a skewed bell curve with a really hot july and august and the rest about the same. San Francisco is a flat line above all the rest of the months. But that skew pulls our entire mean above the SFO mean. 

I have  learned today that statistics and real life don't really seem to agree. 

On a side note, regarding a particular issue with multivariate weighting statistics. I was thinking of this the other day, how to weight the clustering coefficient, boundary parameter and the other stuff. It occurred to me this: PCA is really a weighting based idea, right? The eigenvectors are the weights, and the eigenvalues are the associated variances. So the weight that should be assigned to each one should be the proportion of the variation that it explains.

I think the original PCA was called "multivariate scaling" actually and.... looks it up... some guy named W.S. Torgerson came up with it. I have also read here that there is a special set of algorithms for "multiscale multivariate analysis for given measurements of dissimilarity measures and variables in mulitvariate data" and that this is found in Meuman J and Verboon P, 1989.