Update statistics with full scan or sample?

Should I choose full scan while I update the statistics?

In an ideal situation, if data is 100% random and 100% even distributed, 10% sampling should be as accurate as full scan. But any uneven distribution of data can cause inaccuracy both with sampling and even with full scan.
In short, with skewed data, update stats with sampling can be less accurate than full scan ,but important fact is full scan will also not be accurate.