update -- We've made some changes to the MozCast systems - they won't be noticeable to most, but here are the details for the hard-core data folks...
The original MozCast temperature was built around a fixed, 1,000 query set (which we chose as a sort of keyword "lab", with consistency being a key requirement). Put simply, we wanted the sample to be tightly controlled and consistent, day after day.
A while back, when the Feature Graph was launched, it was based on a 10,000 query set, with half of those queries being localized to 5 different cities. Essentially, MozCast is now 11 different 1K weather "stations".
While the Feature Graph runs on the 10K set, the temperature and "top-view" metrics still run on the 1K set. This happened for a few reasons - historical consistency being the chief one. Interestingly, the 10K set didn't reduce the noise that much (parsing the signal from the noise is the biggest challenge of tracking SERP flux by far).
In addition, some of the top-view metrics are very dependent on the sample size. For example, as your keyword set grows, your domain diversity decreases. Basically, the top domains keep occurring, whereas new long-tail domains pop up. So, the more queries you consider, the less diverse the "world" appears.
We've finally decided to bite the bullet and switch everything to the 10K set (as running two systems was getting silly, and the old system really is outdated). This means that the top-view metrics have changed and we've put a new history in place. Don't obsess about the absolute numbers for those - it's really about trends over time.
The historical temperatures have been preserved, but as of today, we're calculating temperatures over a new data set. Instead of a single station, MozCast now takes the 5 stations that are delocalized (generic US results, essentially), calculates the temperature for each one, and then selects the median of those 5 stations. This should help insure that we're not picking up flux due to large-scale tests or data center differences.
Hopefully, all of these changes provide better data and more stability going forward. If you have any specific questions, please hit up +Pete Meyers
, either on G+ or Twitter (@dr_pete).