It's been proven to work. The main trick is making sure you've selected the right 1400 people.
Yeah this is what I was getting at. People go, oh 31%, that's a sound number, must be fact.
In reality though, you can skew whatever result you want by selecting a CERTAIN 31%.
Say if you polled all 1,000 members of a group. 30% sounds like a good enough sample size to get whatever number you are looking for. But say those results of all 1000 break down to 50% red, 30% white and 20% blue. If you wanted to show that Blue was the winning opinion, you could just choose to poll all the people you estimate to poll blue, and then 10% of white. If your results from a 30% study then show 66% percent blue, 34% white and 0% red, you would be very wrong if you just extrapolated those numbers and thought 30% polled was significant enough to draw that conclusion.
Studied these a lot in psych class, it was almost a philisophical debate within a psychology class of how you know you got an accurate sampling group.