actually, a representative sample is exactly when people are selected randomly.

If we could get 500 random people to answer the questions we'd get results with very high accuracy. Since it's not easy to sample people at random, the surveyors try to correct for that by trying to make sure that the responders represent proportionally segments of the population.