Eh, what the heck. I'll explain the difference for you.
For example, say there are 100 stores in france... and i have 2 stores. (Smaller numbers for benefit of ease, feel free to add zeros if you'd like.)
If I was completeling a survey i'd survey those 2 stores and say they sold 3 Wiis, 2 PS3s and 1 Xbox.
I'd assume that my survey was accurate and other people would take it to mean that there were 150 wii's sold, 100 PS3's and 50 Xbox 360s.
While projection statistics work like this.
I'd survey at first, find the exact same numbers as the survey above, but when I got other more reliable numbers like from chart track, or sold to retail numbers from the big three. I'd find that infact the Wii sold 300, the PS3 sold 100 and the 360 sold 100.
After a few weeks i can be sure that these stats keep going. These same stores seem to account for less coverage of the Wii and 360 then the PS3 by about double.
Therefore when I got the data i would multiply the Wii and 360 by two. (Not this is EXTREMELY simplfied as the actual projection formula you would run it through usually ends up way more complicated as different stores have different coverage and cover different bases and the like.)
Projection statistics get better each with as the formula changes. While surveys just except a direct 1 to 1 and unless the confidence interval is close (i think that's what it's called.) there is a good chance they can be way off.
As such, a lot of the time your formula is way more important than the number of stores you have coverage for. Also the kinds of stores. For examle, 1 Wal-mart store in the US would be way more valuable then say, 20 stores of a small chain of 20 all stationed in NY. That 1,000 callers likely all came from the same area of France for example.
The problem with payed sources is that often times they will keep an inferior formula on the off chance the way they change their complicated formula would cause giant errors... sometimes to offset this they'll have two or three groups of stats guys using different formulas on the same data to see if the new formulas will be more effective.
