The sampling methods they use for this kind of statistic are usually very accurate and representative of the country as a whole; the company I work for predicted the results (percentage votes per party) of the last general election to within 0.8%, just by using fair sampling and questioning techniques.
The inaccuracy with this type of figure is usually the wording itself getting skewed, instead of the number itself; like kizer said, it applies to those aged 25-30 only, which makes it sound a lot more reasonable, don't you think?