I for example had old account once ...
If you had your account in May 2006 and you participated in the survey then, your data is only connected to May 2006. If in August 2006 you were asked again on the same account with the same machine, that other distinct date (data, datum) is only connected to August 2006. If in March 2008 you were asked again on your new account with your new machine and you participated, that new date (data, datum) is only connected to March 2008.
There is no data evaluated across months. The counters are reset to zero each time a new month begins. And another random (some number fe: ) 10000 points are sampled over the timespan of that month.
If you have been chosen randomly and you are not logged in, and and you don't for a few days, you're out; and the system randomly chooses another account. (I don't really think Valve holds any of that data, I would fetch it from client machines always, there's no point in storing it)
Even though Valve would have the technical possibility to include just about every account in the survey, it's not necessary for getting solid statistics; it only creates a great deal of infrastructure (to handle it) and bandwidth occupation as well as binds more people to verify the consistancy and correctness of the survey.
What's imaginable is that Valve run's two (or more) pools for the survey to form a control-mechanism. If the deviation between the groups is beyond some threshold some adjustments have to take place (for example continue running all pools untill deviation is below the threshold).
I for one am quite positive those 4% of machines with DX8 hardware aren't in active gaming use anymore
Based on feeling? Intuition? Logic?
I can say my neighbour here (3rd world) has probably a DX7 machine and uses Steam. Our government runs in part on W98 machines.
I'd tend to believe that the offer from steam is sufficient broad and good that even poor people without hardware-updates for years can take advantage of their services.