> "Sample" of 2700 and then calling it "National IT Survey" in a 20.7m population is a joke/travesty. For well designed, stratified and representative surveys

> "Sample" of 2700 and then calling it "National IT Survey" in a 20.7m population is a joke/travesty. For well designed, stratified and representative surveys 2700 is actually pretty good sample size and will get you to +/- couple of percent points accuracy. There's a whole well studied sciencee behind "design of experiments" that goes into pretty much any R&D or market research effort. Nothing will save you however if you don't design and administer the research well. If you don't believe this and know a bit of programming, simply whip up a Monte Carlo program to repeatedly randomly sample 2700 without replacement from a dataset of 10 million a few thousand times, measure and histogram your estimtes and see the range where you end up most of the time. I actually do this when teaching introductory stats to new analysts and market researchers because seeing is the first step to understanding :-)