I think Jim + Jtp explained the methods use to calculate uncountable objects by taking an average, pretty well.
Surely we mustn't see it as hard cold fact, more as a means to allow the mind to attach physical properties to a concept involving such great numbers, that our statistically retarded brains can't draw on other life examples where we've encountered 70,000,000,000,000,000,000,000 of anything!!
The human brain uses a mapping technique to assign an object it see's with defining characteristics and properties. It takes the sensory responses from the interaction with the object and lists:
Sand - Browny/Yellow, tiny, hard, inedible, plentiful; Application - Sandcastles, cement, hourglass; Location - beaches, desert, sea floor etc.
By mapping an object's qualities, the brain can use it's known characteristics to draw a comparison with any other mapped object in it's memory. This is an effective method of cognition for the brain's environment on Earth in our scale, however, it's flaw is the inability to then add any sort of meaning to huge numbers resulting from research data. No comparison.
The 'more stars than sand' example gives the brain a well known (mapped) object (grain of sand), a scale we are comfortable with (the Earth) and the natural assumption we arrive at to conceive how many grains there must be, allowing us to then compare this vast (but workable) number with cosmologists best guess for stars in the Universe [scale + comparison].
p.s. I've heard the 'number of grains of sand on Earth' used as a different comparison on the other scale. Prof. Brian Cox i know used it and said "There are more molecules in this glass of water (16oz glass approx) than grains of sand on Earth." Maybe the multi-verse theorists will start using it too!!
Good question though,
IHI