The short-form census goes to 100% of the population. It has about 10 questions on it. Most of them are basic demographics such as age, sex, and relationships between people, whether they are married or not. There are also language questions about official languages, mother tongue, and so on. We ask 100% of the population. Because it is a census, the estimates that we generate from those questions have no sampling variability. There is no “accurate to within between 0% and 1%, 95 times out of 100”. The numbers are considered to be the numbers. They are considered to be absolutely accurate, not subject to sampling variability.
The questions that we ask on the long-form census are a much larger set of questions. We get into education, ethnic origin, visible minority status, and so on. We won't be getting into income this time. The long form is fairly onerous on the population. Since 1971, instead of asking those questions of 100% of the population, we ask them of.... The ratio has varied over time. It has been one in five, one in three. It is now one in four. For this census round, we are going to do one in four.
Those estimates are very solid for larger areas and larger populations. As you get into very, very small units, there is more sampling variability. There is more statistical variability as a result of sampling.
The view is that the trade-off, the reduction in burden and the savings in terms of costs, justifies that decrease in the reliability of the estimates.