If the 1000 subjects are chosen well, it's mathematically legitimate -- and the authors give appropriate error bounds based on the ratio of the sample to the population. More detail here:

http://en.wikipedia.org/wiki/Sampling_(statistics)