A practical example of this is when they release raw census data here in Australia, anything with a very small sample size (Eg: less than 100 individuals) may be purposely changed so it's anywhere between 0-1% off.