The sample is no longer probabilistic. This creates all kinds of biases, which is a problem.
Let me give you a concrete example. In the area of housing, an organization has found some surprising data on Quebec City. The results suggest that the condition of housing in Quebec City improved between 2006 and 2011. However, landlords and tenant groups agree that the situation has deteriorated according to the Front d'action populaire en réaménagement urbain (FRAPRU) in Quebec City. Immigrant households appear to have better housing conditions, while the opposite is true. We had never seen this kind of distortion in the past. In terms of the National Household Survey, the non-response rate was 32% compared to 6% for the census conducted in 2006 with the long form.
Do you understand the gravity of the situation?